doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.pgen.1002534 | The Retrohoming of Linear Group II Intron RNAs in Drosophila melanogaster Occurs by Both DNA Ligase 4–Dependent and –Independent Mechanisms | Mobile group II introns are bacterial retrotransposons that are thought to have invaded early eukaryotes and evolved into introns and retroelements in higher organisms. In bacteria, group II introns typically retrohome via full reverse splicing of an excised intron lariat RNA into a DNA site, where it is reverse transcribed by the intron-encoded protein. Recently, we showed that linear group II intron RNAs, which can result from hydrolytic splicing or debranching of lariat RNAs, can retrohome in eukaryotes by performing only the first step of reverse splicing, ligating their 3′ end to the downstream DNA exon. Reverse transcription then yields an intron cDNA, whose free end is linked to the upstream DNA exon by an error-prone process that yields junctions similar to those formed by non-homologous end joining (NHEJ). Here, by using Drosophila melanogaster NHEJ mutants, we show that linear intron RNA retrohoming occurs by major Lig4-dependent and minor Lig4-independent mechanisms, which appear to be related to classical and alternate NHEJ, respectively. The DNA repair polymerase θ plays a crucial role in both pathways. Surprisingly, however, mutations in Ku70, which functions in capping chromosome ends during NHEJ, have only moderate, possibly indirect effects, suggesting that both Lig4 and the alternate end-joining ligase act in some retrohoming events independently of Ku. Another potential Lig4-independent mechanism, reverse transcriptase template switching from the intron RNA to the upstream exon DNA, occurs in vitro, but gives junctions differing from the majority in vivo. Our results show that group II introns can utilize cellular NHEJ enzymes for retromobility in higher organisms, possibly exploiting mechanisms that contribute to retrotransposition and mitigate DNA damage by resident retrotransposons. Additionally, our results reveal novel activities of group II intron reverse transcriptases, with implications for retrohoming mechanisms and potential biotechnological applications.
| Group II introns are bacterial mobile elements thought to be ancestors of introns and retrotransposons in higher organisms. They consist of a catalytically active intron RNA and an intron-encoded reverse transcriptase, which function together to promote intron integration into new DNA sites in a process called “retrohoming.” In bacteria, retrohoming occurs by the excised intron lariat RNA fully reverse splicing into a DNA site, where it is reverse transcribed, yielding an intron cDNA that is copied directly into the host genome. However, little is known about how group II introns behave in higher organisms. Here, we find that linear group II intron RNAs, which cannot fully reverse splice, retrohome in Drosophila melanogaster by attaching themselves to only one end of a DNA site. Reverse transcription then yields an intron cDNA, which is integrated into the recipient DNA by host enzymes that function in non-homologous end joining, a critical cellular DNA–repair pathway. Biochemical experiments exploring alternate mechanisms show that group II intron reverse transcriptases can also template switch efficiently from one RNA template to a second RNA or DNA template, thereby directly linking the two template sequences. Our findings have implications for retotransposition and DNA repair mechanisms and potential biotechnological applications.
| Mobile group II introns are site-specific retrotransposons that consist of a catalytically active intron RNA (ribozyme) and an intron-encoded protein (IEP), with reverse transcriptase (RT) activity [1]. Although they are found mainly in bacterial and organellar genomes, group II introns are thought to have played a major role in eukaryotic genome evolution as evolutionary ancestors of nuclear spliceosomal introns and retrotransposons in higher organisms [2]–[4]. Group II intron RNAs catalyze their own splicing via two sequential transesterification reactions that are the same as those for spliceosomal introns and yield an excised intron lariat with a branched 2′-5′ phosphodiester linkage [5], [6]. For mobile group II introns, the splicing reactions are assisted by the IEP, which binds specifically to the intron RNA and stabilizes the catalytically active RNA structure [7], [8]. The IEP then remains bound to the excised intron lariat RNA in a ribonucleoprotein particle (RNP) that promotes intron integration into new DNA sites [9], [10]. Intron integration is targeted to the ligated-exon junction in an intronless alleles in a process called “retrohoming”, but can also occur at lower frequency into ectopic sites that resemble the homing site in a process called “retrotransposition” or “ectopic retrohoming”. In both cases, the intron inserts into the new DNA site by a novel mechanism in which the excised intron lariat RNA fully reverse splices into a DNA strand and is reverse transcribed by the IEP, yielding an intron cDNA that is integrated into the genome by host enzymes [1], [10]–[14]. Retrohoming leads to the expansion of intron-containing alleles in a population, while ectopic retrohoming provides a means of intron dispersal to new sites.
Group II intron RNAs can also splice without branching by an alternate pathway, termed “hydrolytic splicing” [1]. In this pathway, the first transesterification, 5′-splice site cleavage, occurs by hydrolysis rather than branching, and the second transesterification yields ligated exons and an excised linear intron RNA. Hydrolytic splicing was first observed as a side reaction of group II intron self-splicing under non-physiological conditions [15], [16] and was demonstrated to occur in vivo by using a mutant yeast mitochondrial intron that was deleted for the branch-point A residue [17]. Some subclasses of group II introns lack the branch-point A residue and rely entirely on the hydrolytic mechanism for splicing in vivo [18], [19]. Linear group II intron RNAs can also be generated from excised intron lariat RNAs by debranching, which is believed to accelerate RNA turnover [20]. However, the physiological and evolutionary significance of hydrolytic splicing and linear group II intron RNAs have remained largely unclear.
The Lactococcus lactis Ll.LtrB intron, which has been used extensively as a model system for studying group II intron mobility mechanisms, has a broad host range and is actively mobile in Escherichia coli, making it possible to use E. coli genetic approaches to dissect mobility pathways [11], [21]. The major retrohoming pathway used by the Ll.LtrB intron in E. coli is shown in Figure 1A. After promoting splicing, the Ll.LtrB IEP, denoted LtrA protein, remains bound to the excised intron lariat RNA in RNPs that recognize a DNA target site at the ligated-exon junction (denoted E1–E2) of an intronless allele. The intron lariat RNA initiates retrohoming by fully reverse splicing into the top DNA strand, leading to insertion of the intron RNA between the two DNA exons. The IEP then uses a DNA endonuclease activity to cleave the bottom strand and uses the 3′ DNA end at the cleavage site as a primer for reverse transcription of the inserted intron RNA. In E. coli, the resulting intron cDNA is integrated into the host genome by a mechanism that involves degradation of the intron RNA template strand by a host RNase H and second-strand DNA synthesis by a host DNA polymerase [11], [21]. In variations of this mechanism, Ll.LtrB and other group II introns can also retrohome without bottom-strand DNA cleavage by using a nascent strand at a DNA replication fork to prime reverse transcription of the intron RNA [14], [22], [23], and yeast mitochondrial group II introns retrohome by using recombination rather than DNA repair for cDNA integration [24].
Recently, while carrying out experiments to test whether microinjected group II intron RNPs could be used for gene targeting in Xenopus laevis oocyte nuclei and Drosophila melanogaster embryos, we found that linear as well as lariat group II intron RNAs can retrohome in vivo [25]. This finding was surprising because, unlike lariat RNAs, linear group II intron RNAs can carry out only the first reverse-splicing step, ligation of the 3′ end of the intron RNA to the 5′ end of the 3′- exon DNA [26], [27]. While reverse transcription of fully reverse-spliced intron RNA yields an intron cDNA that can be extended directly by continued DNA synthesis into the upstream exon (Figure 1A), reverse transcription of a partially reverse-spliced intron RNA yields an intron cDNA with an unattached 3′ end that must be linked to the upstream exon DNA in a separate step (Figure 1B). Sequencing of 5′-integration junctions showed that this step occurs by an error-prone process. Although some events result in the precise insertion of the intron between the two DNA exons, most give 5′-integration junctions with 5′-exon deletions, intron 5′-end truncations, insertion of extra nucleotides at the intron-exon junction, or indications of DNA repair via base pairing of microhomologies on opposite sides of the break [25], similar to ligation junctions resulting from double-strand break repair by non-homologous end joining (NHEJ) [28]–[31]. NHEJ activities have been found to contribute to the retrotransposition of LINE elements and other retrotransposons in eukaryotes [32], [33] and may be exploited preferentially by retrotransposons to gain advantage in genetic conflict with their hosts, which rely on these enzymes for survival [34]. Thus, although group II introns are alien to Xenopus and Drosophila, they could be utilizing mechanisms that contribute to the retrotransposition of resident retroelements in eukaryotes and could be subject to host defenses that evolved to counter or mitigate such retrotransposition.
Although NHEJ seemed the most likely mechanism for attachment of the free cDNA to the 5′ exon in linear intron RNA retrohoming, an alternate possibility was that the RT template switches to the 5′-exon DNA, either directly or following incorporation of extra nucleotide residues at the end of the cDNA, resulting in synthesis of a continuous DNA bottom strand containing intron and 5′-exon sequences. Both template switching and incorporation of extra nucleotide residues at the ends of cDNA have been found for other non-LTR-retroelement RTs [35]–[37]. Although we thought this possibility unlikely because group II intron RTs appeared to have low DNA-dependent DNA polymerase activity in vitro [21], template switching and non-templated nucleotide addition by group II intron RTs have not been investigated previously.
Here, we used Drosophila melanogaster mutants to investigate the contribution of NHEJ activities to linear intron RNA retrohoming and assessed the involvement of template switching by comparing junctions formed by this mechanism in vitro with those formed during linear intron RNA retrohoming in vivo. Our results indicate that linear intron RNA retrohoming occurs primarily by a novel variation of NHEJ that uses host enzymes, including DNA ligase 4 (Lig4) and DNA repair polymerase θ (PolQ), but is minimally dependent upon Ku.
To investigate the involvement of NHEJ factors, we compared lariat and linear group II intron retrohoming in D. melanogaster embryos with mutations in the genes encoding DNA ligase 4 (Lig4), Ku70, and the DNA repair polymerase θ (PolQ) [38], [39]. For these experiments, we used a plasmid-based retrohoming assay in which an Ll.LtrB-ΔORF intron with a phage T7 promoter sequence inserted near its 3′ end integrates into a target site cloned in an AmpR-recipient plasmid upstream of a promoterless tetR reporter gene, thereby activating that gene (Figure 2A) [25], [26], [40]. The recipient plasmid was injected into the posterior of precellular blastoderm stage embryos, followed within 5 min by injection of lariat or linear RNPs, which were reconstituted in vitro from the purified IEP and intron RNA (see Materials and Methods). After incubating the embryos for 1 h at 30°C, nucleic acids were extracted and transformed into an E. coli strain (HMS174(DE3)), which expresses T7 RNA polymerase. The transformed bacteria were then plated on medium containing ampicillin or ampicillin and tetracycline, and mobility efficiencies were quantified as the ratio of (TetR+AmpR)/AmpR colonies.
Figure 2B compares the retrohoming efficiencies of lariat and linear intron RNAs in wild-type and mutant embryos, based on parallel assays in ten separate experiments (summarized in Table S1). For each strain in each experiment, 80 injected embryos were pooled prior to extracting nucleic acids and transforming them into E. coli. The results for the lig4 mutant show the retrohoming efficiency of the lariat intron was unchanged, whereas the retrohoming efficiency of the linear intron was decreased strongly (18% wild type), but could be restored to wild-type levels by ectopic expression of Lig4 from an integrated P-element (lig4−; P{lig4+} embryos). In the polQ mutant, the retrohoming efficiency of the linear intron was decreased to ≤0.5% of wild type, compared to 27% wild type for lariat RNPs. Finally, the ku70 mutant, a trans-heterozygote of two putative null alleles (see Materials and Methods), showed only moderately decreased retrohoming efficiencies for both lariat and linear intron retrohoming (67% and 46% wild type, respectively). The latter result was surprising because Ku and Lig4 ordinarily function together in the same NHEJ pathway [41], [42].
The strong differential inhibition of linear compared to lariat RNA retrohoming in the lig4 and polQ mutants supports models in which these enzymes function directly at unique steps in this process, presumably by providing the DNA ligase and repair DNA polymerase activities needed to link the intron cDNA to the upstream exon. The similar moderate decreases in lariat and linear intron retrohoming efficiency in the ku70 mutant could reflect that Ku functions at a common step in both pathways or could be an indirect effect (see Discussion).
D. melanogaster uses at least two NHEJ pathways to repair double-strand breaks: classical NHEJ (C-NHEJ), which is dependent upon Lig4 and Ku70, and alternate end-joining (alt-EJ), which operates without either factor and could be a mixture of different pathways [31], [38], [39], [43]. In a genetic assay for repair of double-strand breaks induced in the germline by the meganuclease I-SceI, lig4 and ku70 mutants inhibited NHEJ activity by 76–78%, leaving 22–24% residual activity that was attributed to alt-EJ [43]. Our finding above that the lig4 mutant shows similar degrees of inhibition and residual activity for linear intron retrohoming (82% and 18%, respectively), most simply suggests that Lig4-independent retrohoming occurs by using the alt-EJ pathway or components thereof.
Previous studies showed that the DNA repair junctions resulting from alt-EJ in Drosophila ku70 and lig4 mutants are generally similar to those for C-NHEJ [31], although in some assays, the lig4 mutant gave somewhat increased frequencies of junctions with extra nucleotide additions (55–63% compared to 30–36% for wild type [38], [39]). To further investigate whether Lig4-independent retrohoming occurs via alt-EJ, we compared 5′- and 3′-intron integration junctions resulting from linear intron retrohoming in the mutant and two commonly used wild-type strains (w1118 and Or-R) by PCR using primers flanking the junctions, followed by cloning and sequencing of the PCR products (Figure 3 and Figure 4).
The 5′- and 3′-integration junctions from lariat intron retrohoming and the 3′-integration junctions from linear intron retrohoming result from reverse-splicing reactions (see Figure 1), and as expected, the PCRs for these junctions gave single prominent products, with no differences between the wild-type and mutant strains (Figure 3, bottom gels; in each case, the expected precise junction sequence was confirmed by sequencing; see legend for details). By contrast, the 5′-integration junctions resulting from linear intron RNA retrohoming were heterogeneous in all strains, with a major band of the size expected for full-length intron insertion and smaller bands, which appeared most prominent in the lig4−; P{lig4+} and polQ− embryos (Figure 3, top gels).
DNA sequences of the 5′-integration junctions resulting from linear intron retrohoming in wild-type w1118 and lig4−, ku70−, and lig4−; P{lig4+} embryos are summarized in Figure 4A–4D, and their characteristics are compared by the bar graphs in Figure 5. As found previously [25], the 5′-integration junctions for linear intron RNA retrohoming in the wild-type embryos were heterogeneous with different combinations of 5′-exon deletions, 5′-intron truncations, and extra nucleotide additions (Figure 4A). Some of the junctions show evidence of DNA repair at regions of microhomology (parentheses), and in some cases, the extra nucleotides inserted at the junctions match and were presumably copied from neighboring sequences in the 5′ exon or intron (underlined).
The 5′ junctions resulting from linear intron retrohoming in the lig4−, ku70−, and lig4−; P{lig4+} mutants were generally similar to those in the wild type w1118, the more closely related wild-type strain, with no large differences in the percentage of junctions with 5′-exon deletions, 5′-intron truncations, extra nucleotide additions, or microhomologies (Figure 4 and Figure 5). Compared to the other strains assayed in parallel, the proportion of long 5′-intron truncations appears to be somewhat lower in the ku70− embryos and higher in the lig4− and lig4−; P{lig4+} embryos, but the significance of these findings is unclear, as the differences were not large and the proportions of full-length and shorter 5′-junction products in each stock were somewhat variable in different experiments. The similarity of the junction sequences resulting from linear intron RNA retrohoming in the lig4 and ku70 mutants to those resulting from double-strand break repair in these mutants [31], [39] supports the hypothesis that Lig4-independent linear intron retrohoming occurs predominantly by using components of the alt-EJ pathway.
The DNA repair polymerase θ (PolQ) has been shown to function in DNA end-joining repair in Drosophila, including a role for extra nucleotide addition at the repaired junctions [38]. To investigate the function of PolQ in linear intron RNA retrohoming, we compared the sequences of 5′-intregration junctions from parallel assays of this process in wild-type Or-R and polQ− embryos (Figure 4E and 4F, Figure 5). Because the number of unique junction sequences recovered from the polQ mutant was lower than those for the other strains and some of these junctions were represented multiple times, we calculated the proportion of junctions with different characteristics in Figure 5 relative to both the total number of junctions (left bars) and the total number of unique junctions sequences (right bars, asterisks). Both comparisons show that that the polQ mutation decreases the proportion of junctions containing extra nucleotide residues (4% of total junctions and 20% of unique junctions compared to >64% of junctions in wild-type Or-R and >48% in all other strains analyzed), as expected from the known function of PolQ. Further, the 5′ junctions from the polQ− embryos have a higher frequency of long (≥15 bp) 5′-exon deletions (72% of total junctions and 80% of unique junctions compared to 29% in wild-type Or-R) and a dramatically increased frequency of long (≥5 nt) microhomologies between exon and intron sequences (56% of total junctions and 40% of unique junctions compared to none among 170 total junctions from all the other strains analyzed). This increased frequency of long microhomologies may reflect that they are more stringently required for annealing of the 3′ end of the cDNA to the upstream exon in the absence of PolQ. We note that among the unique junction sequences from the polQ mutant, two with large deletions were recovered ≥10 times each. Although we cannot exclude that the repeated recovery of these junctions reflects differential amplification by PCR, both have ≥5 nt microhomologies that could have been used preferentially for annealing in multiple events, and indeed one of these junctions (Figure 4F bottom sequence) comprised 6 of 12 recovered junctions in an additional, separate experiment (data not included in Figure 4F). Considered together, the junction sequences indicate that PolQ functions in extra nucleotide addition to the 3′ end of the cDNA during linear intron RNA retrohoming and that this extra nucleotide addition may be critical for generating microhomologies that enable annealing between the 3′ end of the cDNA and the upstream exon DNA. Further, the strongly decreased frequency of linear intron RNA retrohoming in the polQ mutant indicates that PolQ functions in both the Lig4-dependent and Lig4-independent retrohoming pathways.
Although the residual linear intron RNA retrohoming events in the lig4 mutant can be accounted for by Lig4-independent (alt-EJ) NHEJ, it remained possible that template switching of the RT from the 5′ end of the intron RNA directly to the 3′ end of the 5′-exon DNA contributes to this process. Previous studies have shown that other non-LTR retroelement RTs are proficient at template switching directly to the 3′ end of a template strand with little or no complementarity to the cDNA end and that these events can be accompanied by extra nucleotide addition at the junctions, as found for NHEJ [35], [36], [37], [44].
To determine if a template-switching mechanism could be responsible for the manner of 5′ junctions observed during linear intron retrohoming, we carried out biochemical assays using small artificial substrates that simulate the situation at the 5′-integration junction just prior to completion of intron cDNA synthesis (Figure 6). The primary substrate consists of a 60-nt RNA template (Ll.LtrB RNA), whose 5′ end corresponds to that of the Ll.LtrB intron, with a 45-nt DNA primer representing the nascent cDNA (primer c) annealed to its 3′ end. The Ll.LtrB RT (LtrA) initiates reverse transcription of the intron RNA template from the annealed DNA primer and extends it to the 5′ end of the Ll.LtrB RNA template, where it can then switch to a second 40-nt DNA or RNA template with the nucleotide sequence of exon 1 (E1 RNA or DNA, red and black, lanes 5 and 6, respectively). The 3′ end of the Ll.LtrB RNA has an aminoblock to impede the RT from switching to a second molecule of the initial template.
Figure 6, lanes 5 and 6 show that the Ll.LtrB RT efficiently extends the annealed primer c (Pri c) to the end of the intron RNA template, yielding major labeled products of ∼60-nt, which were resolved as a doublet, along with smaller amounts of larger products of the size expected for template switching to the exon 1 (E1) DNA or RNA (∼100 nt) or to a second molecule of Ll.LtrB RNA despite the presence of the aminoblock (∼120 nt). Controls show that no labeled products were detected after incubating the RT with primer c in the presence or absence of the exon 1 RNA or DNA (lanes 2–4).
Cloning and sequencing of the gel bands confirmed that the major ∼60-nt products (bands a and b in lane 5 and h and i in lanes 6) correspond to cDNAs extending to or near the 5′ end of the intron RNA, with the doublet reflecting the addition of extra nucleotide residues, mostly A-residues, to the 3′ end of the cDNA upon reaching the end of the Ll.LtrB RNA template (Figure 7A and 7B). Such non-templated nucleotide addition is a common property of DNA polymerases and RTs [35], [36], [45]–[48].
The first set of larger gel bands (90–110 nts; band c–e in lane 5 and j–l in lane 6) corresponds to products resulting from template switching from the 5′ end of the intron to the 3′ end of exon 1 DNA or RNA (Figure 7A and 7B), as well as products resulting from template switching to the 3′ end or internal regions of the Ll.LtrB intron (Figure S1). Many of the template switches to exon 1 DNA occurred seamlessly, but small numbers of extra nucleotide residues, mostly A residues, were found at some junctions, as well as at the 3′ end of the cDNAs (Figure 7A; bands c–e). The template switches to exon 1 RNA showed similar characteristics, but with a higher proportion of junctions containing extra nucleotide residues (61% compared to 33% for exon 1 DNA; Figure 7B; bands j–l).
The second set of larger bands (120–140 nts; bands f and g in lane 5 and m and n in lane 6) contains products resulting from two sequential template switches to exon 1 DNA or RNA (Figure 7A and 7B, respectively) and/or the Ll.LtrB RNA (Figure S1). These products of multiple template switches have characteristics similar to those resulting from a single template switch, including addition of extra nucleotide residues, mostly A residues, at some template-switching junctions and at the 3′ ends of the cDNAs.
The above results were obtained under reaction optimized for reverse transcription by the Ll.LtrB RT in vitro (450 mM NaCl, 5 mM Mg2+), the high salt concentration helping to stabilize free protein and minimize aggregation of this RT [9]. However, similar results were obtained for template-switching reactions under near-physiological salt conditions (100 or 200 mM KCl, 5 mM Mg2+). Although the RT activity of the protein was lower under these conditions, the gel profiles show roughly equal levels of template switching to exon 1 RNA and DNA (Figure S2), and sequencing of the products showed similar template-switching junctions and patterns of non-templated nucleotide addition (Figure S3).
Finally, we tested whether the Ll.LtrB RT could template switch to double-stranded exon 1 DNA or RNA with an annealed bottom-strand DNA leaving either a blunt end or a 5′ bottom-strand overhang identical to that generated in vivo by the staggered double-strand break accompanying group II intron insertion (Figure 1; complete annealing confirmed by native gel analysis; Figure S4). Neither of these configurations significantly decreased the formation of the 100-nt product resulting from template switching to exon 1 DNA or RNA (Figure 6, lanes 7–9). DNA sequencing confirmed the template switch to double-stranded E1 DNA with a 5′-bottom-strand overhang and showed that this template switch was seamless in most cases (Figure 7C). The sequencing also showed several instances in which the template switch occurred to the penultimate rather than the 3′ terminal residue of exon 1 (Figure 7C), as well as template switches to Ll.LtrB RNA and the bottom-strand overhang oligonucleotide (Figure S5). Template switching to the penultimate nucleotide residue was not seen for single-strand acceptor DNA templates and could be related to the presence of the complementary DNA strand.
Together, the biochemical assays show that the Ll.LtrB RT can template switch from the 5′ end of the intron RNA to exon 1 and surprisingly that template switching is similarly efficient regardless of whether the exon 1 template is RNA or DNA or single- or double-stranded. However, the junctions resulting from template switching differ from those generated during retrohoming of linear intron RNA in vivo in that extra nucleotide additions are uniformly short and mostly A-residues.
Considered together, our results lead to the model shown in Figure 8 for the key steps in linear intron RNA retrohoming. The finding of strong differential inhibition of linear relative to lariat intron retrohoming in D. melanogaster mutants indicates that the NHEJ factor Lig4 is the predominant enzyme involved in ligating the intron cDNA to the upstream exon and that extra nucleotide addition by the DNA repair polymerase θ (PolQ) also plays a crucial role. Although Lig4 and PolQ appear to be the major enzymes playing these roles in D. melanogaster, residual linear intron RNA retrohoming with extra nucleotide addition occurs in both the lig4 and polQ mutants, indicating that other DNA ligases and polymerases can serve as backups that perform the same functions at lower efficiency. Biochemical experiments show that another possible Lig4-independent mechanism, template switching by the group II intron RT from the 5′ end of the intron RNA directly to the upstream exon DNA, is possible but gives junctions differing from the majority of those in vivo. It seems likely that the mechanism elucidated here involving host DNA ligases and repair polymerases is also used for linear intron RNA retrohoming in Xenopus laevis, where we observed similar 5′-integration junctions [25], and more generally, in other eukaryotes, including mammalian cells, where it could have implications for group II intron-based gene targeting. This mechanism also provides a possible means for proliferation of non-branching group II introns in prokaryotes, some of which encode a Ku homolog and ATP-dependent DNA ligases along with DNA repair polymerases and use them in NHEJ pathways related to those of higher organisms [49]. Additionally, features of this mechanism, including the use of both Lig4-dependent and alt-EJ and the requirement for extra nucleotide addition to the cDNA end by a DNA repair polymerase, may be used to promote retrotransposition and mitigate DNA damage caused by LINE elements and other retrotransposons [33], [50]–[52].
The involvement of Lig4 in linear intron RNA retrohoming in Drosophila is indicated by the findings that a lig4 mutation decreases the retrohoming efficiency of linear intron RNA by ∼80% while having no effect on the retrohoming of lariat RNA, and that the decreased retrohoming efficiency of the linear intron in the mutant could be restored to wild-type levels by ectopic expression of Lig4 from a P-element insertion. Most if not all of the residual linear intron RNA retrohoming in the lig4 mutant appears to occur by using components of the alt-EJ pathway, as judged both by similar levels of activity and characteristics of the cDNA ligation junction, particularly patterns of extra nucleotide and the use of microhomologies (see Results). In Drosophila, C-NHEJ and alt-EJ give generally similar double-strand break repair junctions, albeit with quantitative differences in the frequency of extra nucleotide addition in some assays [31], [38], [39], whereas in yeast or mammalian cells, alt-EJ junctions show increased deletion lengths and use of microhomologies [53]–[55].
The involvement of PolQ in linear intron RNA retrohoming is indicated by the findings that a PolQ mutation decreases the retrohoming efficiency by >99% and substantially decreases the frequency of 5′-integration junctions having extra nucleotide residues (4–20% of junctions compared to 67% for wild-type Or-R assayed in parallel and >48% in all other strains; Figure 4 and Figure 5). The mutation also increases the frequency of junctions with long (≥15 bp) 5′-exon deletions and long (≥5 nt) microhomologies. The latter increase is particularly striking, as such long microhomologies were found at 56% of the total and 40% of the unique junction sequences from the polQ mutant, but were not found at junctions (170 total) from any of the other strains analyzed (Figure 4 and Figure 5). The residual extra nucleotide addition in the polQ mutant, which was also seen at 15–20% of junctions in end-joining assays [38], [39], could be due to small amounts of the enzyme remaining in the mutant, which has an unidentified expression defect, or to an alternate DNA polymerase. The very strong decrease in linear intron RNA retrohoming efficiency in the polQ mutant (>99%) indicates that PolQ functions in both the Lig4-dependent and Lig4-independent retrohoming pathways.
PolQ could potentially play at least two roles in linear intron RNA retrohoming. First, extra nucleotide addition to the 3′ end of the cDNA by PolQ may be critical for generating microhomologies that can base pair with the upstream exon to facilitate DNA ligation. Second, PolQ contains a putative DNA helicase domain that could also contribute to retrohoming by promoting base pairing between microhomologies at the cDNA end and the upstream exon, either by annealing the cDNA end to complementary exon sequences or by unwinding the exon DNA strands, making the top strand more accessible to base pairing [38]. The increased frequency of long 5′-exon deletions in the polQ mutant may reflect a delay in cDNA attachment due to lack of suitable microhomologies and/or impaired annealing of complementary cDNA ends to the top strand. The striking increase in the frequency of long microhomologies at the 5′ junctions in the polQ mutant (see above) indicates that an alternate annealing mechanism exists in the polQ mutants, but that it is more dependent upon longer microhomologies between exon and intron sequences than the PolQ-assisted mechanism.
The function, if any, of Ku in linear intron RNA retrohoming is unclear. The finding that ku70 mutations moderately inhibit retrohoming of both linear and lariat intron RNA (54 and 33% inhibition, respectively) could reflect either that Ku contributes to both pathways or that Ku mutations affect one or both pathways indirectly. The Ku protein interacts with a stem-loop region of the RNA component of yeast and human telomerase [56]–[58], and it is possible that Ku may similarly bind to linear or lariat group II intron RNAs to protect them from degradation and/or recruit other DNA repair enzymes to the site. An alternate possibility is that Ku affects retrohoming efficiency indirectly by contributing to the repair of double-strand breaks induced by the intron RNP in the recipient plasmids. In yeast mitochondria, double-strand breaks resulting from abortive retrohoming events are substantially more frequent than completed integrations [59]. If not repaired correctly, such double-strand breaks could lead to loss of functional recipient plasmid target sites, which would appear as decreased retrohoming efficiencies in our assay. A similar indirect effect, involving repair of double-strand breaks in the recipient plasmid could also account for the moderate inhibitory effect of the polQ mutation on lariat intron RNA retrohoming.
Lig4 is ordinarily recruited to DNA breaks by Ku, and D. melanogaster mutations in either lig4 or ku70 give similar decreases in NHEJ efficiency, suggesting that Lig4 acts exclusively in Ku-dependent NHEJ [41]–[43]. By contrast, we find that linear intron RNA retrohoming is more strongly inhibited by a lig4 mutation than by putative null mutations in ku70 (82 and 54% inhibition, respectively). Even assuming that the inhibition by the ku70 mutations is a direct effect, these findings most simply suggest that a substantial proportion of linear intron RNA retrohoming events are promoted by Lig4 in the absence of Ku.
Unlike a conventional double-strand break, the double-strand break formed during linear intron RNA retrohoming has an RNA attached to one of the DNA ends, and this difference could potentially affect the recruitment and use of NHEJ activities. We noted previously that group II intron RNPs bind to both the 5′- and 3′-exons during retrohoming, and such bridging of the ligation junction could impede access and decrease the need for Ku to cap the broken DNA ends [25]. Additionally, the attached intron RNA could contribute directly to the recruitment of NHEJ activities. The interaction of Ku with telomerase RNA noted above is thought to help recruit telomerase to double-strand breaks [57], and it is possible that a similar interaction between Ku and the attached group II intron RNA contributes to the recruitment of Lig4 for some linear intron RNA retrohoming events. More generally, such a mechanism involving the interaction of Ku with RNA could also be used by LINE elements and other retrotransposons to recruit Lig4 and other NHEJ activities for cDNA integration and repair of DNA breaks.
Finally, our biochemical experiments demonstrate that template switching by the group II intron RT from the 5′ end of the intron RNA directly to the 3′ end of the upstream DNA exon is a potential alternate mechanism for cDNA attachment during linear intron RNA retrohoming. Although we found previously that the Ll.LtrB RT has low DNA-dependent DNA polymerase activity in vitro [21], we find here that it template switches and copies 5′-exon DNA templates as well as 5′-exon RNA templates (Figure 6), possibly reflecting that reverse transcription favors a conformation of the enzyme that can initiate more efficiently on DNA templates. In many cases, the template-switching junctions to DNA or RNA templates are seamless, but some have a small number of extra nucleotide residues, predominantly A-residues (corresponding to T-residues in the top strand) that were added by the RT to the 3′ end of the cDNA prior to the template switch. This pattern of extra nucleotide addition, which we found under both enzyme optimal and near-physiological conditions, differs from the majority of 5′-integration junctions resulting from linear intron retrohoming in vivo, where the extra nucleotide residues do not show a similar bias and sometimes correspond to copies of neighboring DNA sequences. It remains possible, however, that template switching by the Ll.LtrB RT could give different junctions in vivo, and that template switching and extra nucleotide addition by this enzyme contributes to some retrohoming events. We note that the ability of the group II intron RT's template switching activity to efficiently link sequences in two different templates could potentially be used to directly attach linker sequences containing primer-binding sites to the ends of cDNAs for cDNA cloning and sequencing applications.
Flies were raised in standard fly media at 22°C. The lig4169 mutant, obtained from Mitch McVey (Tufts University, Medford, MA), has a deletion that removes the start codon and most of the region encoding the ATPase and adenylation domains [31]. The ku707B2 and ku70Ex8 mutants were obtained from William Engels (University of Wisconsin, Madison, WI). The ku707B2 allele lacks 1,359 bp at the 3′ end of the 2,393-bp gene, including most of the DNA and Ku80-interaction domains [43]. The ku70Ex8 allele lacks at least 1 kb, including all of exon 1 and the start codon [43]. The ku707B2/ku70Ex8 genotype, generated by crosses between trans-heterozygous ku707B2/ku70Ex8 parents, is the same as that used previously to study double-strand break repair pathways [39], [43].
The mus308D2 stock [60] was obtained from the Drosophila Stock Center (Bloomington, IA). The mutation lies outside of the coding region and results in undetectable levels of PolQ protein expression [38].
To obtain transgenic flies harboring a lig4 rescue fragment, a 6-kb DNA segment containing the lig4 gene was amplified from w1118 genomic DNA by using the Expand High Fidelity PCR System (Roche Applied Science, Indianapolis, IN), with primers Lig4 F1 BamHI (5′-AAGAGGATCCAGTAGCTGTAGAAGCAGCCAAC) and Lig4 R1 XhoI 5′-AAGACTCGAGCAGCAGTTCCTCCGACATGAAG). This PCR product was inserted between BamHI and XhoI sites of the P-element transformation vector pCaSpeR4 [61], and transgenic flies were produced by GenetiVision (Houston, TX). A transgene insertion on chromosome 2 was recombined with lig4169 to generate the fly stock used in P{lig4+} rescue experiments.
Except for the trans-heterozygous ku707B2/ku70Ex8 embryos (see above), embryos used for microinjection were obtained from crosses between isogenic wild-type or homozygous mutant parents. For all stocks, precellular blastoderm embryos were collected in egg laying chambers in under 40 min, microinjected with recipient plasmids and group II intron RNPs, and incubated for 1 h at 30°C prior to DNA extraction.
pACD5C, which was used for synthesis of lariat and linear intron Ll.LtrB intron RNAs, is a derivative intron-donor plasmid pACD4C with a T7 promoter sequence inserted in the sense orientation at the SalI site in intron DIV [25], [62].
pBRR3-ltrB, the target plasmid for intron-integration assays, contains the Ll.LtrB intron homing site (ligated exon 1 and 2 sequences of the ltrB gene from positions −178 upstream to +91 downstream of the intron-insertion site) cloned upstream of a promoterless tetR gene in an AmpR pBR322-based vector [63].
pIMP-1P, used for expression of the LtrA protein for RNP reconstitution, contains the LtrA ORF cloned downstream of a tac promoter and Φ10 Shine-Dalgarno sequence in the expression vector pCYB2 (New England BioLabs, Ipswich, MA) [9]. LtrA is expressed from this plasmid as a fusion protein with a C-terminal tag containing an intein-linked chitin-binding domain, enabling LtrA purification via a chitin-affinity column, followed by intein-cleavage [9].
pMAL-LtrA, used for expression of the LtrA protein for biochemical assays, contains the LtrA ORF [64] cloned downstream of a tac promoter and Φ10 Shine-Dalgarno sequence between BamHI and HindIII of the protein-expression vector pMAL-c2t. The latter is a derivative of pMal-c2x (New England BioLabs, Ipswich MA) with a TEV protease-cleavage site in place of the factor Xa site [65]. LtrA is expressed from this plasmid with an N-terminal fusion to maltose-binding protein (MalE), enabling its purification via an amylose-affinity column, followed by TEV-protease cleavage to remove the tag (see below).
Ll.LtrB-ΔORF intron RNAs were transcribed from DNA templates generated by PCR of plasmid pACD5C with primers that append a phage T3 promoter sequence (underlined in sequences below) [26]. For the lariat precursor RNA, the PCR primers were pACD-T3 (5′-GGAGTCTAGAAATTAACCCTCACTAAAGGGAATTGTGAGCG) and NheIR (5′-CTAGCAGCACGCCATAGTGACTGGCG), and for linear intron RNA, the PCR primers were T3LIS-1G (5′-AATTAACCCTCACTAAAGTGCGCCCAGATAGGGTGTTAAGTCAAG) and HPLC-purified LtrB940a (5′-GTGAAGTAGGGAGGTACCGCCTTGTTC). The PCR products were purified by using the Wizard SV Gel and PCR Clean-up System (Promega), extracted with phenol-chloroform-isoamyl alcohol (phenol-CIA; 25∶24∶1 by volume), ethanol precipitated, and dissolved in nuclease-free water. In vitro transcription and the preparation of lariat and linear intron RNAs were as described [26].
The LtrA protein used for RNP reconstitution was expressed in E. coli BL21(DE3) from the intein-based expression vector pImp-1P and purified via a chitin-affinity column and intein cleavage, as described [9], except that the column buffer contained 50 mM Tris-HCl, pH 8.0, 0.1 mM EDTA, and 0.1% NP-40. Ll.LtrB RNPs were reconstituted with the purified LtrA protein and in vitro-synthesized lariat or linear Ll.LtrB-ΔORF intron RNA, as described [26], except that the final RNP pellet was dissolved in 10 mM KCl, 5 mM MgCl2, and 40 mM HEPES, pH 8.0.
The LtrA protein used in biochemical assays was expressed in E. coli BL21(DE3) from the plasmid pMAL-LtrA. Cells were grown in starter cultures of LB medium overnight at 37°C, inoculated into 0.5-l LB medium in ultra-yield flasks, and autoinduced by growing at 37°C for 3 h, followed by 18°C for 24 h [66]. Cells were harvested by centrifugation (Beckman JLA-8.1000; 4,000× g, 15 min, 4°C), resuspended in 1 M NaCl, 20 mM Tris-HCl, pH 7.5, 20% glycerol, and 0.1 mg/ml lysozyme (Sigma-Aldrich, St. Louis, MO), kept on ice for 15 min, and lysed by 3 freeze-thaw cycles on dry ice followed by sonication (Branson 450 Sonifier, Branson Ultrasonics, Danbury, CT; three or four 10 sec bursts on ice at an amplitude of 60%, with 10 sec between bursts). After pelleting cell debris (Beckman JA-14 rotor, 10,000 rpm, 30 min, 4°C), nucleic acids were precipitated from the supernatant with 0.4% polyethylenimine (PEI) and constant stirring for 20 min at 4°C, followed by centrifugation (Beckman JA-14 rotor, 14,000 rpm, 30 min, 4°C). Proteins were precipitated from the supernatant by adding ammonium sulfate to 50% saturation with constant stirring for 1 h at 4°C. The precipitated proteins were pelleted (Beckman JA-14 rotor, 14,000 rpm, 30 min, 4°C), dissolved in 500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol, and run through a 10-ml amylose column (FPLC; Amylose High-Flow resin; New England BioLabs, Ipswich, MA), which was washed with 3 column volumes of 500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol and eluted with 500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol containing 10 mM maltose. Fractions containing the MalE-LtrA fusion were incubated with TEV protease (80 µg/ml, 18 h, at 4°C), and imidazole was added to a final concentration of 40 mM. LtrA freed of the MalE tag was then purified by FPLC through a Ni-NTA equilibrated with 500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol, 40 mM imidazole. The Ni-NTA column, which takes advantage of endogenous histidine residues in LtrA's C-terminal domain, was washed with 3 column volumes of 500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol, 40 mM imidazole, and eluted in 500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol, 300 mM imidazole. Finally, the peak LtrA fractions from the Ni-NTA column were further purified through two tandem 1-ml heparin Sepharose columns (New England BioLabs). The columns were equilibrated with 500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol, loaded directly with LtrA protein from the Ni-NTA column, washed with 5-column volumes of 500 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol, and eluted with a 20-column volume gradient of 0.5 to 1 M NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol. The protein elutes approximately midway through the gradient at ∼750 mM NaCl. The purified protein was concentrated to 30 µM, exchanged into 100 mM NaCl, 20 mM Tris-HCl, pH 7.5, 10% glycerol by dialysis, flash-frozen in liquid nitrogen, and stored at −80°C.
Drosophila embryos were microinjected with ∼300 pl recipient plasmid pBRR3-ltrB at 1.4 mg/ml in solution with 500 mM MgCl2 and 17 mM dNTPs, followed within 5 min by ∼300 pl of Ll.LtrB lariat or linear RNPs at 2.6 mg/ml in 10 mM KCl, 5 mM MgCl2, 40 mM HEPES, pH 8.0. The RNPs consist of LtrA protein bound to Ll.LtrB lariat or linear intron RNA with a phage T7 promoter sequence inserted in intron domain IV, and the recipient plasmid contains the Ll.LtrB intron target site (ligated exon 1 and 2 sequences of the ltrB gene; E1 and E2) cloned upstream of a promoterless tetR gene in a pBR322-based vector carrying an AmpR marker. Site-specific integration of the intron into the target site introduces the T7 promoter upstream of the promoterless tetR gene, thereby activating that gene. Eighty embryos were injected and incubated at 30°C for 1 h for each assay. The pooled embryos were incubated in lysis buffer (20 mM Tris-HCl, pH 8.0, 5 mM EDTA, 400 mM NaCl, 1% SDS (w/v)) with 400 µg/ml proteinase K (Molecular Biology Grade; Sigma-Aldrich) for 1 h at 55°C, and then extracted with phenol-CIA. Nucleic acids were ethanol precipitated and dissolved in 12 µl of distilled water.
For assays of retrohoming efficiency, a 4-µl portion of the nucleic acid preparation was electroporated into electrocompetent E. coli HMS174(DE3) F−, hsdR, recA, rifr (Novagen, EMD Chemicals, Gibbstown, NJ), which expresses T7 RNA polymerase. Cells were plated at different dilutions on 2% agar containing LB medium with ampicillin (50 µg/ml) plus tetracycline (25 µg/ml) or the same concentration of ampicillin alone. Colonies were counted after overnight incubation at 37°C, and the integration efficiency was calculated as the ratio of (AmpR+TetR)/AmpR colonies.
For analysis of intron-integration junctions, a 1-µl portion of the nucleic acid preparation was used as template for PCR using Phusion High Fidelity PCR Master Mix with HF buffer (New England BioLabs). The 5′-junction PCRs were done with primers P1 (5′-CTGATCGATAGCTGAAACGC) and LtrB933a (5′-AGGGAGGTACCGCCTTGTTCACATTAC), and the 3′ junction PCRs were done with primers P3 (5′-CAGTGAATTTTTACGAACGAACAATAAC) and P4 (5′-AATGGACGATATCCCGCA). The PCR was done for 25 cycles for all strains, except for parallel assays of wild-type Or-R and the polQ mutant, which required 35 PCR cycles to obtain sufficient PCR product from the mutant. The PCR products were purified using a MinElute PCR purification Kit (Qiagen), cloned into a TOPO TA cloning vector (pCRII-TOPO; Invitrogen, Carlsbad, CA), and transformed into chemically competent E. coli (One Shot TOP10; Invitrogen). The cloned PCR products were then amplified from randomly picked colonies by colony PCR using Phusion High Fidelity PCR Master Mix with HF buffer and primers M13 F(-20) (5′- GTAAAACGACGGCCAGT) and M13 R(-26) (5′-CAGGAAACAGCTATGAC) for 25 cycles, and sequenced using primers M13 R(-24) (5′-GGAAACAGCTATGACCATG) or M13 F(-20) [67].
Biochemical assays were done by incubating purified LtrA protein (40 nM) with synthetic oligonucleotide substrates that correspond to the 5′ end of the Ll.LtrB intron (60-nt Ll.LtrB RNA; 40 nM) with an annealed 5′-32P-labeled DNA primer corresponding to nascent cDNA (45-nt Pri c; 44 nM) in the presence of exon 1 RNA or DNA (40-nt E1; 40 nM) in 20 µl of reaction medium containing 450 mM NaCl, 5 mM MgCl2, 20 mM Tris-HCl, pH 7.5, 1 mM dithiothreitol (DTT) and 200 µM dNTPs. The reaction components were assembled on ice with substrate added last and then incubated at 30°C for 30 min. Reactions were terminated by phenol-CIA extraction. Portions of the reaction product (3 µl) were added to an equal volume of gel loading buffer II (95% formamide, 18 mM EDTA and 0.025% each of SDS, xylene cyanol, and bromophenol blue; Ambion, Austin, TX), denatured at 98°C for 7 min, and analyzed by electrophoresis in a denaturing 10 or 15% polyacrylamide gel, which was visualized by scanning with a PhosphorImager (Typhoon Trio, GE Healthcare, Piscataway, NJ). 32P-labeled DNA products were excised from the gel, amplified by PCR, as described (Sabine Mohr, Scott Kuersten, and A.M.L., manuscript in preparation), and cloned into the TOPO-TA pCR2.1 vector (Invitrogen), according to the manufacturer's protocol. Random colonies were picked and the cloned PCR products were amplified by colony PCR using Phusion High Fidelity PCR Master Mix/HF buffer with primers M13 F(-20) and M13 R(-26), and sequenced using the M13 R(-24) primer (see above).
The oligonucleotides used in the biochemical assays were Ll.LtrB RNA [LtrB5'S20Anchor6,5 RNA] ((5′- GUGCGCCCAGAUAGGGUGUUCUCGUUGGCAAUGGUGUCCAACUUGUGCUGCCAGUGCUCG), with an aminoblock on its 3′ end); annealed primer c (5′- CGAGCACTGGCAGCACAAG-deoxyuridine-TGGACACCATTGCCAACGAGAACAC); and exon 1 DNA (5′-TGTGATTGCAACCCACGTCGATCGTGAACACATCCATAAC) or RNA (5′-UGUGAUUGCAACCCACGUCGAUCGUGAACACAUCCAUAAC). Oligonucleotides complementary to exon 1 DNA or RNA were: exon 1 AS (5′-GTTATGGATGTGTTCACGATCGACGTGGGTTGCAATCACA) and exon 1 AS+9 (5′-AATGATATGGTTATGGATGTGTTCACGATCGACGTGGGTTGCAATCACA).
DNA and RNA oligonucleotides used in the assays were obtained from Integrated DNA Technologies (IDT; Coralville, IA) and purified in a denaturing 10% (w/v) polyacrylamide gel. DNA primers were 5′-end labeled with [γ-32P]-ATP (10 Ci/mmol; Perkin-Elmer, Waltham, MA) by using phage T4 polynucleotide kinase (New England BioLabs) according to the manufacturer's protocol. Complementary oligonucleotides were annealed at ratios of 1∶1 (E1 oligonucleotides) or 1∶1.1 Ll.LtrB/primer c by mixing at 20 times the final concentration in annealing buffer (100 mM Tris-HCl, pH 7.5, and 5 mM EDTA), heating to 82°C, and slowly cooling to 25°C for 45 min. The efficiency of annealing was assessed by electrophoresis in a non-denaturing 6% polyacrylamide gel containing Tris-borate-EDTA (90 mM Tris, 90 mM boric acid, 2 mM EDTA) at 30°C [67].
|
10.1371/journal.pgen.1000156 | Three SRA-Domain Methylcytosine-Binding Proteins Cooperate to Maintain Global CpG Methylation and Epigenetic Silencing in Arabidopsis | Methylcytosine-binding proteins decipher the epigenetic information encoded by DNA methylation and provide a link between DNA methylation, modification of chromatin structure, and gene silencing. VARIANT IN METHYLATION 1 (VIM1) encodes an SRA (SET- and RING-associated) domain methylcytosine-binding protein in Arabidopsis thaliana, and loss of VIM1 function causes centromere DNA hypomethylation and centromeric heterochromatin decondensation in interphase. In the Arabidopsis genome, there are five VIM genes that share very high sequence similarity and encode proteins containing a PHD domain, two RING domains, and an SRA domain. To gain further insight into the function and potential redundancy among the VIM proteins, we investigated strains combining different vim mutations and transgenic vim knock-down lines that down-regulate multiple VIM family genes. The vim1 vim3 double mutant and the transgenic vim knock-down lines showed decreased DNA methylation primarily at CpG sites in genic regions, as well as repeated sequences in heterochromatic regions. In addition, transcriptional silencing was released in these plants at most heterochromatin regions examined. Interestingly, the vim1 vim3 mutant and vim knock-down lines gained ectopic CpHpH methylation in the 5S rRNA genes against a background of CpG hypomethylation. The vim1 vim2 vim3 triple mutant displayed abnormal morphological phenotypes including late flowering, which is associated with DNA hypomethylation of the 5′ region of FWA and release of FWA gene silencing. Our findings demonstrate that VIM1, VIM2, and VIM3 have overlapping functions in maintenance of global CpG methylation and epigenetic transcriptional silencing.
| Methylation of cytosine bases provides one layer of epigenetic information that is superimposed on the nucleotide sequence of a genome. Proteins that bind methylated cytosines and also help maintain that DNA modification are important linchpins in a self-propagating system mediating memory of epigenetic states. We previously demonstrated that the VIM1 (VARIANT IN METHYLATION 1) protein from the flowering plant Arabidopsis thaliana binds DNA that contains methylated cytosine and is required for complete methylation and compaction of centromeric DNA. In this study, we show that VIM1 works in concert with two related proteins, VIM2 and VIM3, to maintain cytosine methylation not only at centromeres but throughout the genome. VIM proteins act specifically in the DNA methylation pathway that targets CpG dinucleotides, which plants share with animals, rather than the plant-specific non-CpG methylation pathways. Loss of VIM1, VIM2, and VIM3 function also causes a reduction in transcriptional gene silencing at a variety of sequences, and leads to abnormal developmental phenotypes, including late flowering associated with loss of FWA gene silencing. Our results demonstrate that these three related VIM family proteins have overlapping functions in the MET1-mediated CpG methylation pathway.
| DNA cytosine methylation is an epigenetic mark important for many processes including parental imprinting, X chromosome inactivation, and the silencing of transposable elements [1]–[3]. In mammals, methylated cytosines are found almost exclusively in symmetrical CpG sequence contexts, and CpG methylation patterns are propagated after DNA replication by “maintenance” DNA methyltransferases (DNMT1-type) [2]. However, DNA cytosine methylation in plants can occur at CpHpG and CpHpH (where H = A, T, or C) as well as CpG sites. In Arabidopsis, these three categories of cytosine methylation are carried out by distinct activities [4]: CpG methylation is maintained primarily by the DNMT1-type methyltransferase, MET1; CHROMOMETHYLASE3 (CMT3) is responsible for CpHpG methylation; and methylation at CpHpH sites is accomplished by a de novo methyltransferase, DOMAINS REARRANGED METHYLASE (DRM). Cytosine methylation patterns in Arabidopsis are also trimmed by the action of DNA demethylases in the DEMETER (DME) family, which include REPRESSOR OF SILENCING 1 (ROS1) [5],[6].
Complex interactions between DNA methylation and histone modification articulate epigenetic gene expression states, and methylcytosine-binding proteins play an important role in interpreting the epigenetic information encoded by DNA methylation [3]. The best understood class of methylcytosine-binding proteins contains a conserved methylcytosine-binding domain (MBD) [7]. Several MBD proteins in mammals, including methyl-CpG-binding protein 1 (MeCP1), MeCP2, MBD2, and MBD4 have been identified with high affinity for methylated DNA, and the biological importance of mammalian methylcytosine-binding proteins has been shown in the wide range of severe phenotypes by mutations of these genes [8]–[10]. In contrast to mammalian MBD proteins, knowledge of plant MBD proteins remains relatively limited. Among 13 MBD proteins in the Arabidopsis proteome, three (AtMBD5, AtMBD6, and AtMBD7) have been shown to bind symmetrically methylated CpG sites in vitro [11]–[13]. Although developmental defects have been observed in lines carrying a loss-of-function mutation of AtMBD9 or a transgene directing RNAi against AtMBD11 transcripts, the role of any AtMBD proteins in epigenetic regulation remains to be defined [14],[15].
Recently a novel class of methylcytosine-binding proteins have been defined that interact with the modified base through an SRA (SET- and RING-Associated) domain [16],[17]. We previously reported that mutations in the Arabidopsis VIM1 gene, which encodes an SRA domain methylcytosine-binding protein, causes DNA hypomethylation and decondensation of centromeres in interphase [18]. The SRA domain of VIM1 shares amino acid similarity with mammalian UHRF1 (also known as mouse Np95 and human ICBP90), which has been implicated in regulation of chromatin modification [19],[20], transcription [21], and the cell cycle [22]. Recent reports demonstrate that UHRF1 is required for maintenance of CpG DNA methylation [23],[24]. UHRF1 physically interacts with DNMT1 and has been postulated to mediate the loading of DNMT1 on to replicating heterochromatin [23]–[25].
In the Arabidopsis genome, there are four genes that share high sequence similarity with VIM1. To obtain further insight into the function and potential redundancy among the VIM proteins, we have investigated vim double and triple mutants, as well as transgenic vim knock-down lines that down-regulate multiple VIM family genes. Our results indicate that VIM1, VIM2 and VIM3 have overlapping functions in maintenance of cytosine methylation at CpG dinucleotides distributed throughout the genome. Moreover, VIM proteins are required for transcriptional silencing of a variety of sequences, including centromeric repeats, transposons, and the parentally imprinted FWA locus.
VIM1 is a member of a small gene family, which was originally identified by a naturally-occurring null mutation in the Arabidopsis thaliana accession, Borky-4 (Bor-4) [18]. The Arabidopsis genome contains five VIM genes, each of which encodes a protein containing a PHD domain, two RING domains, and an SRA domain (Figure S1A). Interestingly, four of the VIM genes (VIM1 [At1g57820], VIM2 [At1g66050], VIM4 [At1g66040], and VIM5 [At1g57800]) are located within a 3 Mb region on the lower arm of chromosome 1. A reverse transcriptase pseudogene (At1g57810) is located between VIM1 and VIM5, and an unrelated putative pseudogene (At1g66045) is located between VIM2 and VIM4. The VIM proteins share high amino acid sequence identity (>68%) throughout their entire length, including the four previously recognized domains (Figure S1A).
We measured the steady state levels of transcripts using reverse transcriptase (RT)-PCR from all five VIM genes to obtain insight into their function and potential redundancy (Figure S1B and S1C). VIM1 was highly expressed in inflorescence tissue and to a lesser extent in two-week-old leaves of wild-type Columbia (Col) plants (Figure S1C). VIM3 [At5g39550] transcripts were found in both inflorescences and two-week-old leaves, while VIM2 was abundantly expressed in inflorescences. In contrast, VIM4 and VIM5 were absent from leaves or inflorescence tissue at this level of detection, suggesting that the steady-state levels of VIM4 and VIM5 are very low or that these may be pseudogenes. Based on our expression data, we concentrated on the functional analysis of VIM1, VIM2, and VIM3.
We first investigated the subnuclear localization of VIM1, VIM2, and VIM3 in interphase. Trangenes containing VIM1, VIM2, or VIM3 cDNAs fused with YFP at the N-terminus under the control of the strong, constitutive cauliflower mosaic virus (CaMV) 35S promoter were transformed into Col plants. Fixed nuclei from transgenic Col root cells expressing YFP-VIM1, YFP-VIM2, or YFP-VIM3 are shown in Figure 1A. YFP-VIM1, -VIM2, and -VIM3 fusion proteins were broadly distributed in the nucleus and were enriched in the heterochromatic chromocenters. The similar subnuclear localization of all three expressed VIM proteins suggests a possible functional redundancy among these VIM proteins.
We characterized T-DNA insertional mutations that disrupt the coding sequences of VIM1, VIM2, and VIM3 in the Col background. Using RT-PCR and primers flanking the T-DNA insertion sites, we did not detect expression of VIM1, VIM2, and VIM3 in vim1-2, vim2-2, and vim3-1 mutant plants, respectively, confirming that the T-DNA insertions likely destroy gene function (Figure S2 and data not shown). While the vim1-2 mutation caused a slight decrease in DNA methylation of centromeric 180-bp repeat arrays, neither the vim2-2 nor the vim3-1 allele led to a centromere repeat hypomethylation phenotype (Figure 1B). Therefore, VIM2 and/or VIM3 are not required for centromere DNA methylation, or alternatively, one or both of these genes function redundantly with VIM1 to maintain centromere DNA methylation.
To test these alternative hypotheses, we individually introduced cDNA copies of Col VIM1, VIM2 or VIM3 genes into Bor-4 plants, which carry the vim1-1 loss-of-function allele, and investigated the effect on centromere DNA methylation. VIM cDNAs fused with YFP at the N-terminus were expressed under the control of the 35S promoter (Figure S3). Expression of a wild-type Col VIM1 cDNA in Bor-4 plants fully restored DNA methylation at the centromere (Figure 1C). Over-expression of Col VIM2 or VIM3 cDNAs can also fully suppress the hypomethylation of centromeric repeats in Bor-4 vim1-1 plants, demonstrating that VIM2 and VIM3 proteins can function redundantly with VIM1.
To determine the extent of functional redundancy within the VIM gene family, we generated and characterized a set of double mutants combining two loss-of-function mutations among VIM1, VIM2, and VIM3. As an alternative approach, we also isolated transgenic plants with a coordinate decrease in the expression of VIM1, VIM2 and VIM3 genes. We took advantage of the variation in transgene expression among the Col transgenic lines expressing an YFP-VIM1 cDNA used in our nuclear localization analysis (Figure 1A). A significant proportion of the transgenic plants did not show any YFP expression and had lower steady-state transcript levels of VIM2 and VIM3 as well as VIM1 (Figure S2) presumably due to RNA interference. We chose two of these transgenic VIM family “knock-down” lines (vimKD-A and vimKD-B) for further analysis.
We tested whether DNA methylation at highly repetitive loci is dependent on the function of multiple VIM gene family members. Genomic DNA samples prepared from plants of the different vim genotypes were analyzed by DNA gel blot hybridization after digestion with methylation-sensitive restriction endonucleases. Cleavage by HpaII (5′-CCGG-3′) is inhibited by either CpG or CpHpG methylation; MspI (5′-CCGG-3′) digestion is blocked by CpHpG methylation; and NlaIII (5′-CATG-3′) activity is inhibited by CpHpH methylation. The vim1 vim2 and vim2 vim3 double mutants did not display any change in centromere or 45S rRNA gene cytosine methylation compared to vim1 and Col wild-type plants, respectively (Figure S4 and data not shown). In contrast, the vim1 vim3 double mutant and the two vim knock-down lines displayed decreased CpG and CpHpG methylation at HpaII and MspI sites in both the centromere and 45S rRNA genes relative to the vim1 single mutant (Figure 2A and 2B). There was no difference among the genotypes tested after NlaIII digestion at the 180-bp centromeric repeats and 45S rRNA genes (Figure S5), which might reflect a low level of CpHpH methylation. Our DNA blot data demonstrate that VIM proteins, particularly VIM1 and VIM3, function redundantly to maintain CpG and CpHpG methylation at the 180-bp centromere repeats, as well as in repetitive sequences outside the centromere.
To evaluate the potential role of the VIM genes in shaping DNA methylation patterns in other heterochromatic regions, the DNA methylation status of two transposable elements, the MULE DNA transposon AtMU1 and the gypsy-class LTR retroelement AtGP1, was compared among the different vim genotypes using bisulfite sequencing. We found that CpG sites in these elements were heavily methylated in Col wild-type, vim1, and vim3 plants, but CpG methylation was significantly decreased in the vim1 vim3 mutant and the two transgenic vim knock-down lines (Figure 2C and 2D; Table S1). In contrast, no substantial changes in CpHpG and CpHpH methylation were observed. These results further indicate that VIM proteins function redundantly to maintain cytosine methylation of heterochromatic sequences outside of the centromere.
Next, we assessed the DNA methylation pattern of 5S rRNA genes using DNA gel blot hybridization analysis and bisulfite sequencing. The vim1 vim3 mutant and the two transgenic vim knock-down lines had strongly decreased CpG methylation at HpaII sites and reduced CpHpG methylation at MspI sites in the 5S rRNA genes (Figure 3A). This hypomethylation was accompanied by CpHpH hypermethylation evidenced by higher molecular weight hybridization signals after HaeIII (5′-GGCC-3′) digestion (Figure 3A) and NlaIII digestion (Figure S5). Bisulfite sequencing of the 5S rRNA genes confirmed that CpG methylation was significantly decreased in the vim1 vim3 mutant and the two transgenic vim knock-down lines relative to Col wild-type plants (Figure 3B; Table S1). Most of the CpG sites are affected by the vim1 vim3 mutation combination and the coordinate knock-down of VIM gene expression in the transgenic lines, but the degree of hypomethylation at different CpG sites varied widely (5%–70% decrease in vimKD-B) (Figure 3C and data not shown). Our bisulfite sequencing analysis also demonstrated an increase in CpHpH methylation throughout the 5S rRNA genes in the vim1 vim3 mutant and the transgenic vim knock-down lines, especially vimKD-B.
Increased CpHpH methylation levels observed in the 5S rRNA genes could result from activation of DNA methylation activity and/or inhibition of DNA demethylation activity. To address these possibilities, we first monitored the levels of siRNA species that could target de novo methylation [26] to the 5S rRNA repeats, but found no significant changes in the abundance of siRNA from 5S rRNA (Figure S6). We also examined the steady-state levels of transcripts from the three major DNA methyltransferase genes (CMT3, DRM2, and MET1) and two DNA demethylase genes (DME and ROS1) by RT-PCR. Although no significant changes in steady-state transcript levels for the three DNA methyltransferase genes or DME were observed in any vim mutant (data not shown), ROS1 transcript accumulation was reduced in the vim1 vim3 mutant and the transgenic vim knock-down lines (Figure 3D). This result raises the possibility that ROS1 transcriptional repression in plants deficient in activity of the VIM proteins might contribute to CpHpH DNA hypermethylation in the 5S rRNA genes.
We analyzed genic cytosine methylation at three different loci [27], At4g00500 (lipase class 3 family protein), At4g13610 (MEE57, maternal effect embryo arrest 57), and At4g31150 (endonuclease V family protein), to determine whether VIM proteins are involved in DNA methylation of low-copy, expressed sequences not associated with heterochromatin. First, a PCR-based method was used to assay DNA methylation of these loci in the vim mutant lines. After HpaII treatment, unmethylated DNA will be digested and therefore not amplified by PCR. In wild-type Col, vim1, and vim3 samples, all three tested genic regions were easily amplified after HpaII digestion. In contrast, the abundance of PCR product for At4g00500 and At4g31150 (and to a lesser extent At4g13610) amplified from the vim1 vim3 and transgenic vim knock-down samples was lower than that from the wild-type Col sample, indicating a reduction in CpG methylation (Figure 4A).
To verify and extend the results obtained by the PCR-based method, bisulfite sequencing was carried out for the At4g31150 gene (Figure 4B; Table S1). DNA methylation at At4g31150 in wild-type Col plants was predominantly localized in CpG dinucleotides. The vim1 vim3 mutant and the transgenic vim knock-down lines showed a two- to three-fold reduction in CpG methylation at the locus. In wild-type Col, vim1, and vim3 samples, 73–91% of the CpG dinucleotide in the HpaII site of At4g31150 monitored in our PCR-based assay was methylated. Notably, less than 17% of those sites were methylated in vimKD-A, and the CpG in the HpaII site was completely unmethylated in the vim1 vim3 mutant and the vimKD-B line. In contrast to CpG methylation, wild-type Col samples had very low levels (<4%) of CpHpG and CpHpH methylation at At4g31150, and none of the tested genotypes displayed an effect on CpHpG and CpHpH methylation. These data confirm that VIM deficiency primarily affects CpG methylation, and demonstrate that the targets of VIM activity are broadly distributed in the genome to include genic regions.
RT-PCR was carried out to determine whether VIM deficiency affects gene expression. First, we investigated the level of transcripts generated from the 180-bp centromeric repeats in different vim genotypes. Elevated levels of 180-bp repeat transcripts from both strands were detected in the vim1 vim3 mutant and the two transgenic vim knock-down lines, suggesting that DNA hypomethylation of the 180-bp centromeric repeats caused by VIM deficiency is associated with a loss of transcriptional silencing of these repeats (Figure 5A). Next, we focused on the 5S rRNA genes, where loss of VIM function leads to dramatic changes in cytosine methylation. 5S rRNA genes are organized into pericentromeric tandem repeat arrays, and only a subset of the repeats is transcribed with the remainder being epigenetically silenced. To detect the release of 5S rRNA repeat silencing, we used primer pairs that detect two 5S transcript variants, resulting in amplification of 140 and 210 nucleotide products from cDNA templates prepared from wild-type Col plants. 5S-140 transcripts accumulated to a higher level in the vim1 vim3 mutant and the transgenic vim knock-down lines compared to the wild-type Col genotype, but no significant increase in 5S-210 was observed (Figure 5B). This result demonstrates that a decrease of 5S rRNA gene methylation at CpG sites in the vim1 vim3 mutant and the transgenic vim knock-down lines is associated with loss of 5S rRNA gene silencing of the smaller variant. We also examined the expression of the two loci representing methylated genes: At4g00500 and At4g31150. We detected equivalent levels of transcription for At4g00500 and At4g31150 in each genotype (Figure S7). These data suggest that the genic CpG methylation found in these genes does not lead to a general repression of gene expression and that VIM proteins are not playing an active role in regulating expression of these genes.
We also examined dispersed transposable elements for loss of transcriptional silencing. Specifically, we investigated the transcription status of the non-LTR retroelement AtLINE1-4, the LTR retrotransposon AtGP1, and the class II element AtMU1 using an RT-PCR assay on RNA samples derived from vim mutants (Figure 5C). No AtGP1 transcripts were detected in any of the vim mutant lines despite the loss of CpG methylation. However, we found that transcription of AtMU1 and AtLINE1-4 elements were activated in the vim1 vim3 mutant and the transgenic vim knock-down lines. These findings indicate that VIM proteins function redundantly to silence different types of transposable elements.
Despite the changes in DNA methylation and the release of gene silencing, the vim1 single mutant, the vim1 vim3 double mutant, and the transgenic vim knock-down lines did not display abnormal developmental phenotypes under standard growth conditions (data not shown). The lack of morphological phenotypes may be due to incomplete ablation of the expressed VIM family genes. We tested this hypothesis by combining vim1, vim2, and vim3 mutations in a single background and found that the triple mutant has distinct morphological phenotypes, including late flowering (Figure 6A and Figure S8) and production of aerial rosettes on the flowering stem (data not shown). We investigated the basis of the late flowering phenotype by examining the imprinted locus FWA, which is demethylated and reactivated in DNA hypomethylation mutants (e.g., met1) leading to delayed flowering [28],[29]. As shown in Figure 6 (panels B, C and D), homozygotes carrying the hypomorphic met1-1 allele show ectopic FWA expression in vegetative tissues and hypomethylation of the transposon-related repeat sequences that comprise the promoter region. A low level of ectopic FWA expression, apparently insufficient to affect flowering, was observed in the vim1 vim3 mutant sample, correlated with partial hypomethylation of the upstream region of the FWA gene (Figure 6C and 6D). The more extensive hypomethylation of the FWA upstream region in the vim1 vim2 vim3 triple mutant was associated with stronger FWA expression, which is expected to cause late flowering.
As seen in Figure 6, the vim1 vim2 vim3 triple mutant showed a stronger phenotype than that observed in the vim1 vim3 double mutant, indicating VIM2 is important in repressing FWA expression and hypermethylation of the upstream region of the FWA gene in vegetative tissues. To investigate whether VIM2 has a redundant function with VIM1 and VIM3 outside of the FWA locus, we examined DNA methylation in other genic regions. The abundance of PCR product for At4g00500 and At4g13610 amplified from the HpaII-digested vim1 vim2 vim3 mutant sample was significantly less than that from the vim1 vim3 mutant sample, indicating that the vim triple mutant displayed a more severe reduction in genic methylation compared to the vim1 vim3 mutant (Figure 7A).
As the vim triple mutant phenocopied met1 mutants in several aspects, including preference for CpG hypomethylation, ectopic CpHpH hypermethylation, reduced ROS1 expression, and late flowering associated with ectopic expression of FWA, we further explored the severity of phenotypes displayed by vim mutants relative to met1 mutants. Specifically, we compared transcriptional activation of transposons and DNA hypomethylation of highly repetitive heterochromatic regions in vim1 vim2 vim3 and met1 mutants. First, we re-examined the transcription status of AtLINE1-4 and AtMU1 using an RT-PCR assay. Loss of transcriptional silencing of AtMU1 and AtLINE1-4 elements was more severe in the vim1 vim2 vim3 mutant compared to the vim1 vim3 mutant (Figure 7B). Transcription of both transposons in the vim1 vim2 vim3 mutant was similar or higher than that observed in the hypomorphic met1-1 mutant. We next assessed the DNA methylation pattern of the 180-bp centromere repeats and the 5S rRNA genes by DNA gel blot hybridization analysis after digestion with HpaII. The vim1 vim2 vim3 triple mutant displayed stronger hypomethylation at HpaII sites relative to the vim1 vim3 double mutant for both repeat families (Figure 7C). Interestingly, the DNA hypomethylation phenotypes in the vim1 vim2 vim3 triple mutant were significantly stronger than those observed in the hypomorphic met1-1 mutant, but similar to the phenotypes exhibited by the met1-3 null mutant. Taken together, these findings indicate that VIM1, VIM2 and VIM3 function redundantly to silence different types of transposable elements and to maintain CpG methylation in the heterochromatic as well as genic regions. Further, our results indicate that simultaneous loss of VIM1, VIM2 and VIM3 function almost completely blocks CpG methylation.
The Arabidopsis genome contains five VIM genes, each of which encodes a protein containing a PHD domain, two RING domains, and an SRA domain. Previously, we reported that VIM1 is an unconventional methylcytosine-binding protein and that loss-of-function vim1 mutations cause cytosine hypomethylation and decondensation of centromeres [18]. Here we describe a broader analysis of the VIM gene family in Arabidopsis. Our results indicate that the expressed VIM proteins cooperate to maintain global CpG methylation and epigenetic transcriptional silencing.
We previously reported that VIM1 binds methylated CpG and CpHpG in vitro and similarly Johnson et al. [16] showed that ORTH1/VIM3 and ORTH2/VIM1 can bind methylated CpG, CpHpG, or CpHpH substrates. Our results provide additional support for the hypothesis that VIM1, VIM2, and VIM3 function redundantly. First, all three expressed VIM proteins showed a similar subnuclear localization. When ectopically expressed in transgenic plants under control of the 35S promoter, YFP-VIM1, VIM2, and VIM3 protein fusions were broadly distributed in the nucleus and were enriched in the heterochromatic chromocenters (Figure 1A). Second, VIM1, VIM2, and VIM3 were abundantly expressed in leaves and inflorescence tissue – an overlapping expression pattern that supports the possibility of functional redundancy (Figure S1C). Third, the heterologous overexpression of wild-type Col VIM2 or VIM3 coding sequences compensated for the loss of VIM1 with regard to centromere DNA methylation when transformed into Bor-4 vim1-1 plants (Figure 1C). Fourth, combining vim mutations or suppressing the expression of multiple VIM genes in transgenic lines resulted in stronger DNA hypomethylation than those exhibited by vim single mutants (Figure 2, 3, 4, 6, and 7). A release of gene silencing also occurred at the 180-bp centromeric repeats, 5S rRNA repeats, and some transposable elements in the vim1 vim3 double mutants and the transgenic vim knock-down lines (Figure 5). Simultaneous disruption of VIM1, VIM2 and VIM3 led to a loss of FWA gene silencing in vegetative tissues, as well as a more severe reduction in genic and tandem repeat methylation compared to the vim1 vim3 double mutant (Figure 6 and 7). These results indicate that VIM proteins play important, overlapping roles in maintenance of cytosine methylation and transcriptional silencing throughout the Arabidopsis genome.
Although VIM proteins have overlapping functions, a hierarchy exists among VIM proteins. The vim1 vim3 mutant displayed a strong synergistic effect on cytosine methylation compared to either single mutant, whereas the vim1 vim2 mutation combination showed no significant enhancement of DNA hypomethylation compared to the vim1 single mutant (Figure S4 and Figure 7A). The importance of VIM2 is demonstrated by the more severe cytosine hypomethylation and loss of transcriptional silencing displayed by the vim1 vim2 vim3 triple mutant compared to the vim1 vim3 double mutant (Figure 6 and 7). These results indicate that VIM1 is the major functional member of the VIM family with regards to DNA methylation and epigenetic silencing, while VIM3 and VIM2 (in descending order) play lesser roles in the examined loci. This functional hierarchy might reflect qualitative differences among VIM proteins or relative gene expression levels (VIM1 is the most highly transcribed member of the gene family based on public databases: http://mpss.udel.edu/at and http://bbc.botany.utoronto.ca/efp).
VIM proteins affect CpHpG as well as CpG methlyation at the 180-bp repeats and 45S rRNA genes (Figure 2), while VIM proteins specifically affect CpG methlyation in other loci examined. One possibility is that CpHpG methylation in the centromere and 45S rRNA genes may be reduced as a secondary consequence of a loss of CpG methylation, as reported in met1 mutants [29]. Alternatively, VIM proteins may have locus-specific regulatory mechanisms for maintaining DNA methylation. One indication of locus specificity is the preferential effect of vim1 mutations on centromere methylation and compaction [18], which might result from varying levels of functional redundancy at different genomic locations – for instance, a more diminished role for VIM2 and VIM3 at the centromere. The specificity may be determined by the primary sequence itself or sequence copy number. In addition, the activities of VIM proteins could be influenced by other proteins that have sequence specificity.
The vim1 vim2 vim3 triple mutant showed delayed flowering, and we uncovered FWA hypomethylation and ectopic FWA expression as one possible mechanism for this developmental phenotype. Liu et al. reported that plants overexpressing a VIM1-GFP fusion protein leads to delayed flowering and an elevated level of a key repressor of flowering, FLOWERING LOCUS C (FLC) transcripts [30]. One possibility is that the late flowering phenotypes in vim1 vim2 vim3 triple mutant plants and plants overexpressing VIM1 result from different mechanisms. Another possibility is that the late flowering phenotype in the VIM1 overexpression line might be caused by a dominant negative mechanism. The overexpressed VIM1-GFP fusion proteins could sequester other components for maintaining DNA methylation and epigenetic silencing. Alternatively, the VIM1 transgene induce silencing of endogenous VIM genes, similar to the situation in our vimKD-A and vimKD-B transgenic lines.
The precise role of VIM proteins in epigenetic regulation remains an open question, but two plausible models are supported by previous reports and this study. The mammalian counterparts of VIM proteins exhibit a variety of activities that either directly modify histones [17],[19] or specifically recognize modified histones [31]. Accordingly, we originally proposed that VIM1 affects DNA modification by acting at the chromatin – cytosine methylation interface, possibly by modifying chromatin substrates for DNA methyltransferase activity. The preference for CpG hypomethylation in the vim mutants suggests that the MET1-mediated pathway is primarily affected by any alteration of the chromatin substrates. In the alternative model, which is based on the physical interaction of mammalian VIM homologs with DNMT1 [23]–[25], VIM proteins tether the DNMT1-class CpG methyltransferase MET1 to the replication fork. This model predicts that inactivation of the redundant VIM family proteins would phenocopy met1 mutants – a prediction supported by several observations. First, the vim1 vim3 mutant and the transgenic vim knock-down lines showed a preference for CpG hypomethylation, similar to met1 mutants. Second, these vim mutant lines exhibited CpHpH hypermethylation and a reduction in ROS1 transcript level, two recently described characteristics of met1 null mutants [32],[33]. Third, the developmental phenotypes of the vim1 vim2 vim3 mutant, including late flowering, resembled that of met1 homozygotes [28],[29]. The late flowering phenotype is likely to be caused in part by the ectopic expression of FWA associated with DNA hypomethylation of the upstream repeats observed in both met1 and vim1 vim2 vim3 mutants. Fourth, the reduction in 180-bp centromere and 5S rRNA gene repeat methylation in the vim1 vim2 vim3 triple mutant matched the extreme hypomethylation observed in the met1-3 null mutant. The parallels between the vim1 vim2 vim3 mutant and met1 mutants argue that the VIM proteins are essential components of the MET1-mediated cytosine methylation pathway.
Seeds of vim T-DNA insertion mutants were obtained from the SALK T-DNA collection [34] through the Arabidopsis Biological Resources Center at The Ohio State University. The vim1-2 allele (SALK_050903), the vim2-2 allele (SALK_133677), and the vim3-1 allele (SALK_088570) carry a T-DNA insertion in the fourth exon, the third exon, and the fourth exon of the corresponding gene, respectively. Dr. Jerzy Paszkowski kindly donated the met1-3 mutant seeds. Plants were grown in a controlled environmental chamber at 22°C under long-day conditions (16 h light per day).
Full-length VIM1, VIM2, and VIM3 cDNA clones were PCR-amplified from a wild-type Col first-strand cDNA preparation using primers VIM1-F/VIM1-R, VIM2-F/VIM2-R, and VIM3-F/VIM3-R, respectively. The fragments were cloned into pENTR-D TOPO (Invitrogen, USA) and the resulting VIM inserts were recombined into pEarlyGate104 [35] using Gateway technology (Invitrogen, USA). These constructs were transformed into Agrobacterium tumefaciens (LBA4404) and were introduced into Col WT or Bor-4 vim1-1 plants by in planta transformation [36].
T2 generation transgenic seeds were germinated on 1× MS (Murashige and Skoog) media and grown for 6–10 days in 16 h/8 h (light/dark) growth conditions at 22°C. Seedlings were fixed in a 1× PBS solution containing 4% paraformaldehyde for 1 hour at room temperature and subsequently stained with 10 µg/mL propidium iodide. Images of root nuclei were acquired with a Leica SP2 laser scanning confocal microscope equipped with a 488 nm laser, 561 nm laser, and filter sets suitable for the detection of YFP and propidium iodide. The images were merged and processed using Adobe Photoshop CS3 (Adobe Systems).
Genomic DNA was digested with HpaII, MspI, NlaIII or HaeIII (New England Biolabs, USA) according to the manufacturer's instructions. Radiolabeled probes were generated by random priming, and blots were prepared and hybridized using standard methods. The following probes were generated from purified cloned inserts: 180-bp repeat (CEN) clone, pARR20-1 [37]; 45S rRNA gene clone, pARR17 [37]; and 5S rRNA gene clone, pCT4.1 [38].
Genomic DNA samples were modified by sodium bisulfite using the EpiTech Bisulfite kit (Qiagen, USA) according to the manufacturer's protocols. PCR products were TA-cloned into pGEM-T Easy (Promega, USA) and individual clones were sequenced with the T7 primer. Approximately 24 individual clones were sequenced for each locus from two independent bisulfite sequencing experiments. Detailed bisulfite sequencing data, including the average methylation content for each clone, are provided in Table S1.
1 µg of genomic DNA was digested with HpaII or HhaI (no enzyme for controls). Dilutions of DNA from the digestion reaction were then used for each PCR reaction. PCR conditions were 2 min at 94°C, followed by 27 cycles of 94°C for 30 s, 53°C for 30 s, and 68°C for 1 min for each primer sets.
To check the expression of the VIM genes, total RNA was isolated from 2-week-old leaves and inflorescence tissues from wild-type Col plants. 3-week-old leaves from each genotype were used for checking the expression of other genes. Two or three independent RNA extractions were performed per genotype and a pool of plants was used for each extraction. Aliquots of 1 µg of total RNA were treated with DNaseI (Invitrogen, USA) and 300 ng of DNase-treated total RNA was used as input in RT-PCR reactions using the Superscript III RT (Invitrogen, USA). For the 180-bp centromere repeat, we used ‘GAPC-R & CEN-F’, ‘GAPC-R & CEN-R’, or ‘no primers’ for strand-specific first-strand cDNA synthesis, and all reactions were performed with RT. The ‘no primers’ control was used to detect trace amounts of contaminating DNA, which is a particular problem due to the high-copy number of the repeat templates. For the 5S rRNA genes, we used ‘GAPC-R & 5S-R’ primers for first-strand cDNA synthesis and minus RT negative controls were performed. For other sequences, oligo(dT) primers were used for first-strand cDNA synthesis and minus RT negative controls were performed with primers specific to each sequence. Amplification of GAPC or ACT2 RNA was used as an internal control. All primers used for RT PCR and the other analyses are listed in Table S2.
Small RNA gel blot analysis was performed using the mirVana miRNA isolation kit (Ambion, USA) as described previously [18]. The siR1003 [39] and miR163 riboprobes were generated according to the mirVana probe construction kit (Ambion, USA) and labeled by T7 polymerase transcription in the presence of α-32P UTP.
|
10.1371/journal.pmed.1002249 | Multimorbidity and healthcare utilization among home care clients with dementia in Ontario, Canada: A retrospective analysis of a population-based cohort | For community-dwelling older persons with dementia, the presence of multimorbidity can create complex clinical challenges for both individuals and their physicians, and can contribute to poor outcomes. We quantified the associations between level of multimorbidity (chronic disease burden) and risk of hospitalization and risk of emergency department (ED) visit in a home care cohort with dementia and explored the role of continuity of physician care (COC) in modifying these relationships.
A retrospective cohort study using linked administrative and clinical data from Ontario, Canada, was conducted among 30,112 long-stay home care clients (mean age 83.0 ± 7.7 y) with dementia in 2012. Multivariable Fine–Gray regression models were used to determine associations between level of multimorbidity and 1-y risk of hospitalization and 1-y risk of ED visit, accounting for multiple competing risks (death and long-term care placement). Interaction terms were used to assess potential effect modification by COC.
Multimorbidity was highly prevalent, with 35% (n = 10,568) of the cohort having five or more chronic conditions. In multivariable analyses, risk of hospitalization and risk of ED visit increased monotonically with level of multimorbidity: sub-hazards were 88% greater (sub-hazard ratio [sHR] = 1.88, 95% CI: 1.72–2.05, p < 0.001) and 63% greater (sHR = 1.63; 95% CI: 1.51–1.77, p < 0.001), respectively, among those with five or more conditions, relative to those with dementia alone or with dementia and one other condition. Low (versus high) COC was associated with an increased risk of both hospitalization and ED visit in age- and sex-adjusted analyses only (sHR = 1.11, 95% CI: 1.07–1.16, p < 0.001, for hospitalization; sHR = 1.07, 95% CI: 1.03–1.11, p = 0.001, for ED visit) but did not modify associations between multimorbidity and outcomes (Wald test for interaction, p = 0.566 for hospitalization and p = 0.637 for ED visit). The main limitations of this study include use of fixed (versus time-varying) covariates and focus on all-cause rather than cause-specific hospitalizations and ED visits, which could potentially inform interventions.
Older adults with dementia and multimorbidity pose a particular challenge for health systems. Findings from this study highlight the need to reshape models of care for this complex population, and to further investigate health system and other factors that may modify patients’ risk of health outcomes.
| The co-occurrence of multiple chronic conditions in an individual (multimorbidity) has been linked to poor outcomes including increased hospital use, longer length of stays, and worse cognitive and physical functioning.
Particularly for community-residing older adults with dementia, multimorbidity can result in challenges to both self-care and provided care. Individuals with multimorbidity often receive care from multiple physicians across different care settings each year. This lack of physician continuity may lead to poorer quality of care and outcomes.
Important gaps exist in our understanding of the interplay between multimorbidity, health system use, and continuity of physician care specifically for individuals with dementia.
Our historical cohort study was designed to estimate the risk of acute care hospitalization and emergency department (ED) visit by level of multimorbidity (i.e., chronic disease burden) among persons with dementia in the community. We were especially interested in whether the risk of these health outcomes was lower for those with better continuity of physician care.
We performed a retrospective cohort study of 30,112 home care clients with dementia in Ontario, Canada, using routinely collected health and clinical information linked at the individual level.
We defined the level of multimorbidity (i.e., chronic disease burden in addition to dementia diagnosis) based on a count of the presence of 16 common chronic conditions, and compared time, in days, from a health assessment to initial hospitalization (for any cause) and ED visit (not resulting in an inpatient stay) in persons with different multimorbidity levels. We accounted for other possible outcomes including death or placement in a long-term care facility.
We found that multimorbidity was highly prevalent in this population—89% of the cohort had been diagnosed with two or more conditions in addition to dementia.
In multivariable analyses, we found that the risk of hospitalization and ED visit increased with each higher level of multimorbidity.
These associations were comparable in clients with dementia who had high and low degrees of physician continuity. In other words, continuity of physician care did not modify the association between level of multimorbidity and the outcomes.
Multimorbidity is the norm rather than the exception among older adults with dementia in the home care sector. This increased chronic disease burden is associated with a greater likelihood for costly hospital admissions and emergency visits.
With increases in life expectancy, improvements to disease detection, and a shift to community-based care, use of home care services and the prevalence of multimorbidity among older persons with dementia will likely rise. Data from this study may be useful in identifying at-risk individuals and prioritizing the deployment of limited healthcare resources.
| Dementia (including Alzheimer disease) is a progressively debilitating condition associated with cognitive and functional impairment and behavioral challenges. As a condition affecting primarily older adults [1], most individuals with dementia also have other coexisting chronic conditions, or multimorbidity. This creates complex challenges for clinical care [2]. For example, certain conditions, such as stroke [3] and diabetes [4], have been linked to accelerated cognitive decline. Dementia-related impairments can also hinder a patient’s ability to self-manage concurrent diseases, adhere to therapies, or effectively communicate the signs and symptoms of complications to care providers, which may lead to adverse outcomes [5]. Therefore, common goals for dementia care are to manage coexisting conditions and, where possible, to prevent potentially avoidable care transitions, including hospitalization and institutionalization [6].
Healthcare utilization, including hospital admissions [7–11] and emergency department (ED) visits [10–13], has been shown to be elevated among older adults diagnosed with dementia. Hospital and ED visits are particularly relevant for persons with dementia given their heightened risks for cognitive and functional decline during and after hospitalization [14–16]. Such vulnerability places these patients at further risk of additional care transitions following acute care discharge, including readmission to the ED [12] or hospital or placement in a long-term care (LTC) facility [17], as well as death [18]. Associations between these outcomes and multimorbidity are poorly understood within the dementia population, but may be useful for healthcare providers and policymakers in targeting high-risk individuals for enhanced patient-centered care, and improving patient integration across all sectors of the healthcare system.
A further concern is that older adults with dementia and coexisting conditions may be susceptible to care fragmentation, as they frequently receive care from multiple physicians across the healthcare system each year [19]. This can lead to deficiencies in care delivery, including poor communication between providers and medication errors. A recent US study of older adults with dementia found that lower continuity of physician care (COC) was associated with greater healthcare utilization [20], but important gaps exist in our understanding of utilization patterns among persons with dementia and multimorbidity, particularly in the context of healthcare systems that provide universal access to services.
In the current study, we used linked population-based health administrative and clinical assessment data to quantify associations between multimorbidity and two healthcare utilization outcomes, hospital admissions and ED visits, among community-residing individuals with dementia in Ontario, Canada. We focused specifically on individuals with dementia in the home care setting. In Ontario, as elsewhere, the home care sector is a growing component of healthcare delivery that helps older adults to maintain independence within their home residence while reducing overall health system costs. Individuals with dementia comprise a large proportion of this population [21,22]. As a secondary objective, we tested whether COC had a direct association with healthcare utilization and whether COC modified the associations between multimorbidity and the hospitalization and ED visit outcomes. In a recently published study [23], the association between multimorbidity and hospitalization was less pronounced among individuals with greater physician continuity in the general population. We hypothesized that similar findings of effect modification by COC would be observed among persons with dementia.
This study was approved by the Research Ethics Board of Sunnybrook Health Sciences Centre (Toronto, Canada). As we used health information routinely collected in Ontario, informed consent from study participants was not required. The study is reported per RECORD guidelines (S1 Text). The study protocol is available in S2 Text.
We conducted a retrospective analysis of linked population-based health administrative and clinical assessment data in Ontario, Canada’s largest province and home to over 13 million residents. Almost all Ontarians are covered by a universal health insurance program that pays for all medically necessary inpatient, emergency, and physician services, and includes coverage for medications for individuals 65 y of age and older. Publicly funded home care services and LTC placements are coordinated and delivered through regional Community Care Access Centres. Home care services are provided on a short- or long-stay basis. The latter refers to ongoing supportive care required for more than 60 d in a single episode. In Ontario, it is mandatory for all long-stay home care clients to be assessed with a comprehensive clinical assessment tool, the Resident Assessment Instrument for Home Care (RAI-HC), on a semiannual basis.
All services provided via the public health insurance system are recorded in databases and later linked deterministically using unique encoded identifiers and analyzed at the Institute for Clinical Evaluative Sciences in Toronto. Each dataset used in this study is described in S1 Table.
We identified all Ontario residents aged 50 y and older who received a RAI-HC assessment between January 1, 2012, and June 30, 2012, and had been diagnosed with dementia prior to the assessment date. Dementia was identified based on the presence of (1) a relevant diagnostic code (S2 Table) recorded on a hospital discharge, (2) a relevant diagnostic code recorded on three physician billings separated by at least 30 d occurring within a 2-y period, or (3) dispensing of any cholinesterase inhibitor (whose only indications are for the treatment of Alzheimer disease or dementia associated with Parkinson disease). This case ascertainment algorithm has been validated in Ontario [24]. We restricted our population to individuals with a RAI-HC assessment in order to include only those with dementia living in the community and to obtain additional health and functional status information for the study population that is available only from the RAI-HC. We excluded individuals if they had missing information on age or sex, or were not eligible for healthcare coverage at the time of RAI-HC assessment. One RAI-HC assessment per individual was selected for analysis (index RAI-HC assessment), the nearest to April 1, 2012. Counts according to the study exclusion criteria are shown in S3 Table.
For all individuals included in the study, we determined the presence of 16 comorbid conditions. Each condition was defined at the time of index RAI-HC assessment using historical data. Consistent with previous studies on multimorbidity in Ontario [22,23,25,26], these conditions were selected based on their large economic impact and high prevalence in the general population [27–29], and included the following: acute myocardial infarction, asthma, any cancer, cardiac arrhythmia, chronic coronary syndrome, chronic obstructive pulmonary disorder, congestive heart failure, diabetes, hypertension, non-psychotic mood and anxiety disorders, other mental illnesses (which included schizophrenia, delusions, and other psychoses; personality disorders; and substance abuse), osteoarthritis, osteoporosis, renal failure, rheumatoid arthritis, and stroke (excluding transient ischemic attack). All cases were identified from Ontario Health Insurance Plan database and Discharge Abstract Database (DAD) data using ICD-9 and -10 codes. Validated case ascertainment algorithms (where available) or similar case definition approaches were used (S2 Table). For these conditions, for each individual we defined level of multimorbidity (i.e., chronic disease burden) as the number of prevalent chronic conditions in addition to dementia. This was coded as 0–1, 2, 3, 4, or 5+.
As proxies for disease severity, we derived two independent markers from the RAI-HC data, the Minimum Data Set Health Status Index (MDS-HSI), and the Changes in Health, End-stage disease and Symptoms and Signs (CHESS) scale. The MDS-HSI is a preference-based measure of health-related quality of life that includes six domains (sensation, mobility, emotion, cognition, self-care, and pain) [30]. Values range from 1.00 (perfect health state) to −0.02 (health state worse than death, which is scored at 0). In contrast, CHESS is an ordinal measure used to detect instability in health for older adults and has been shown to be a strong predictor of hospitalization and mortality in older adults [31,32]. We coded values to range from 0 (no health instability) to 4 (high to very high instability).
We measured COC using the Bice–Boxerman continuity of care index [33]. This index measures the extent to which a patient visits the same clinician for ongoing medical care over a defined period. Values range from 0 to 1, where scores approaching 1 reflect a higher concentration of visits to a single physician. We included all ambulatory visits over the 2 y prior to the index RAI-HC assessment and considered all physician specialties because both primary care physicians and specialists play a role in the management of chronic conditions. The Bice–Boxerman index accounts for physician referrals. Each individual was categorized as having either a high COC or low COC, with high COC defined as a Bice-Boxerman index value greater than or equal to the median score in the study population. Alternative COC measures were explored in sensitivity analyses, as noted below.
Age, sex, and date of death (where applicable) were identified from the Ontario Registered Persons Database, and neighborhood-level income quintile and rurality (urban or rural residence) [34] from the 2006 census. Marital status was derived from the index RAI-HC assessment; categories included married, separated/divorced, widowed, and never married. All LTC placements following the index RAI-HC event (where applicable) were derived from Continuing Care Reporting System data. Lastly, the number of acute hospital episodes and the number of unplanned ED visits that each client experienced in the 1 y prior to their index RAI-HC assessment were identified from the DAD and National Ambulatory Care Reporting System datasets, respectively.
We followed each individual prospectively for 1 y following their index RAI-HC assessment to identify the time (in days) to (1) first acute inpatient hospital admission (DAD data) and (2) first unplanned ED visit that did not result in an inpatient stay (National Ambulatory Care Reporting System data). For both measures, all causes were considered.
We described the distribution of baseline sociodemographic characteristics, clinical characteristics, and the frequency of prior health service utilization by level of multimorbidity. We modeled associations between level of multimorbidity and the risk of (1) acute hospitalization and (2) unplanned ED visit with a competing risks regression derived from Fine and Gray’s proportional sub-hazards model [35] using Stata’s stcrreg command. This regression is based on the cumulative incidence function, which quantifies the probability of an event of interest (hospitalization or ED visit) during the study follow-up period, acknowledging the possibility of one or more competing events. Deaths and placements into LTC facilities were considered competing events. Individuals who did not experience any outcome (event, death, or LTC admission) were censored at the end of the 1-y observation period. We derived age- and sex-adjusted associations between level of multimorbidity and each outcome. Multivariable competing risk regressions then assessed the risk of hospitalization or ED visit by level of multimorbidity adjusting for age, sex, income quintile, rurality, marital status, COC, health-related quality of life (MDS-HSI), CHESS, prior hospitalizations, and prior ED visits. Plots of Schoenfeld residuals were used to assess the assumption of proportionality, which was not violated in any of the models. Individuals missing information on any covariate were excluded from each analysis; however, no single covariate had more than 1.8% missing (COC).
To determine whether COC modified the association between level of multimorbidity and the hospitalization and ED visit outcomes, an interaction term was added to each multivariable model. We plotted resulting coefficients for visual representation of effect modification and used Stata’s post-estimation lincom command to assess statistical differences in the risk of each outcome by COC with increasing levels of multimorbidity. Ten sub-hazard ratio (sHR) estimates were derived: one per level of multimorbidity (n = 5) per outcome (n = 2). In addition, for both outcomes, a Wald test was used to assess whether the sHRs associated with low COC were the same for each level of multimorbidity.
Sensitivity analyses assessed the influence of particular subgroups expected to have either a higher or lower baseline risk of admission to hospital or to the ED. Specifically, multivariable analyses were repeated that excluded from our study population individuals whose index RAI-HC assessment was related to either a “change in status” or “review at the return from hospital” (sensitivity analysis 1) and individuals living with modest to very severe cognitive impairment, based on Cognitive Performance Scale value ≥ 4 (sensitivity analysis 2). We also assessed the robustness of our COC measure and effect modification findings by recalculating COC by excluding individuals with <3 physician visits (sensitivity analysis 3) and also by categorizing COC into tertiles (low, medium, and high COC; sensitivity analysis 4).
We identified 30,112 individuals in Ontario with dementia who had a RAI-HC assessment between January 1 and June 30, 2012. They represented 27.5% of all home care clients otherwise eligible for study inclusion (S3 Table). Table 1 presents characteristics of the study population. The mean age of the study population was 83.0 (standard deviation 7.7) y, 63% were women (n = 19,056), and 88% lived in an urban setting (n = 26,461). Eleven percent of the study population (n = 3,309) was diagnosed with dementia alone (n = 755) or with dementia and one other condition (n = 2,554). A total of 89% (n = 26,804) had two or more conditions in addition to dementia, while 35% (n = 10,568) had five or more comorbid conditions in addition to dementia. The most prevalent comorbid conditions were hypertension (82.4%), osteoarthritis (59.7%), and diabetes (34.4%) (S4 Table). Both proxies for disease severity (MDS-HSI and CHESS) showed greater impairment with higher levels of multimorbidity, while prior healthcare utilization (hospitalizations and ED visits in the past 1 y) increased with increasing number of chronic conditions. Median COC in the population was 0.63. Continuity decreased with each additional level of multimorbidity.
Table 2 shows that 29% (n = 8,759) and 34% (n = 10,189) of the study population experienced an acute care hospitalization or ED visit, respectively, as their first event during the 1-y follow-up period (data presented by high and low COC in S5 Table). Both of the proportions increased with higher levels of multimorbidity, from 19% (0–1 chronic conditions) to 37% (5+ chronic conditions) for hospitalization and from 25% (0–1 chronic conditions) to 40% (5+ chronic conditions) for ED visit. Overall, 18% (n = 5,298) of the study population died at any point during the 1-y follow-up period, and 38% (n = 11,349) entered LTC.
Results from age- and sex-adjusted and multivariable competing risk analyses are shown in Table 3. For both hospitalization and ED visit, risk increased monotonically with higher levels of multimorbidity. All comparisons (reference group is 0–1 chronic conditions) were statistically significant (p < 0.05). For hospitalization, risk among those with five or more conditions was more than double (sHR = 2.18, 95% CI: 2.00–2.37) that of individuals with 0–1 conditions in age- and sex-adjusted analyses. In the full model, the sHR was reduced slightly to 1.88 (95% CI: 1.72–2.05). Similarly for ED visit, the age- and sex-adjusted risk was 76% greater among those with five or more conditions (versus 0–1 conditions, sHR = 1.76, 95% CI: 1.63–1.90). In the full model, sub-hazards were 63% greater (sHR = 1.63, 95% CI: 1.51–1.77).
Of note in the multivariable models, men (versus women) had a greater risk of hospitalization (sHR = 1.18, 95% CI: 1.13–1.24) but lower risk of an ED visit that did not result in an inpatient stay (sHR = 0.94, 95% CI: 0.90–0.99). For both outcomes, sub-hazards were not linearly associated with income quintile. Hospitalization risk was no different by rurality, but risk of an ED visit was greater among rural (versus urban) residents (sHR = 1.18, 95% CI: 1.12–1.25). Additionally, clients with previous hospitalizations in the past year were significantly more likely to be admitted to hospital during follow-up but not more likely to experience an ED visit (e.g., for clients with 2+ recent hospitalizations compared with none, sHRs were 1.40, 95% CI: 1.31–1.49, and 1.04, 95% CI: 0.98–1.11, for hospitalization and ED visit risk, respectively, from multivariable analyses). Previous ED visits were associated with a significantly greater risk for both outcomes (e.g., for clients with 2+ recent ED visits compared to none, sHRs were 1.39, 95% CI: 1.32–1.47, and 1.94, 95% CI: 1.85–2.04, for hospitalization and ED visit risk, respectively).
Adjusted for age and sex only, low (versus high) COC was associated with an 11% increased risk of hospitalization (sHR = 1.11, 95% CI: 1.07–1.16) and a 7% increased risk for an ED visit (sHR = 1.07, 95% CI: 1.03–1.11). These associations were not statistically significant, however, in the full multivariable models (Table 3). Fig 1 illustrates the effect modification by COC on level of multimorbidity for both outcomes, Table 4 presents sHRs for low (versus high) COC at each level of multimorbidity (reference = 0–1 chronic conditions), and S1 Fig. shows cumulative incidence estimates. Although point estimates diverged, no comparisons were statistically significant; in other words, COC did not statistically (using a p < 0.05 criterion) modify the association between level of multimorbidity and either outcome. Wald tests for significance of the interaction terms confirmed this finding (p = 0.566 for hospitalizations and p = 0.637 for ED visits).
The results from multiple sensitivity analyses are presented in S6 Table and S2 and S3 Figs. In each model tested, there were no differences in data interpretation to that of our primary analyses.
In our investigation of older community residents with dementia receiving home care services in Ontario, Canada, we observed a large burden of comorbid chronic disease in addition to dementia, and this multimorbidity was associated with an increased risk of subsequent hospital admission and ED visit. In multivariable analyses that accounted for competing risks of death and LTC admission, risks of all-cause hospitalization and ED visit both increased monotonically with each additional diagnosis, and were 88% greater and 63% greater, respectively, among those with five or more comorbid conditions relative to those with dementia alone or with dementia and one other condition. In this population, low physician continuity was associated with elevated risk of hospitalization and ED use in age- and sex-adjusted analyses, but not in multivariate models. Contrary to our hypothesis, greater COC did not modify the association between level of multimorbidity and hospitalization or ED visit risk in this older population with dementia.
In this study, more than one-quarter of otherwise eligible long-stay home care clients aged 50 y and older were diagnosed with dementia. This prevalence is comparable to that found in other studies of the home care sector in Ontario [21,22], and larger than in the general population in the province [36] or elsewhere [37]. The high prevalence of multimorbidity we observed in our cohort is consistent with previous studies on comorbidity among persons with dementia residing in a community setting [2,38,39], where hypertension, osteoarthritis, and diabetes are commonly reported. More than one-third of our study population had five or more chronic conditions in addition to dementia, and only 11% of our sample had dementia alone or with one other condition. Multimorbidity, therefore, is very much the norm rather than the exception in this group of patients.
Detailed examinations of the relationships between multimorbidity with health service utilization within the dementia population are currently limited. Bynum et al. [9] described crude rates of hospitalization for all causes and for ambulatory-care-sensitive conditions stratified by number of chronic conditions. As in the current work, outcomes were more frequent with each additional condition. These findings have important implications for patients with dementia, their families and care providers, and healthcare policymakers. They highlight the need to develop effective interventions targeted to the particular needs of this population. Unique challenges exist in detecting and managing comorbidities for dementia patients [40], including difficulties in communicating medical complaints because of memory and language deficits. Clinical encounters may also be dominated by dealing with dementia and its manifestations, to the exclusion of other conditions. While patient-centered care embracing the principles of chronic disease management should be encouraged [41], the nature of dementia will in itself make this more difficult.
A component of chronic disease management and improved health system integration, measures of COC quantify the dispersion of physician visits among providers. We found that physician continuity decreased with increasing multimorbidity in the dementia population. Interestingly, low (versus high) COC was associated with an 11% increase in hospitalization risk and a 7% increase in ED visit risk in age- and sex-adjusted analyses, but results were nonsignificant in the fully adjusted models. We also found that COC did not modify the associations between level of multimorbidity and hospitalization and ED visit outcomes, a finding verified in multiple sensitivity analyses. In contrast, COC was found to modify observed associations between multimorbidity and hospitalization (adjusting for age and sex) in the general population of Ontario [23], and low levels of COC have been associated with higher rates of hospitalization, ED visits, imaging and laboratory testing, and healthcare spending among dementia patients (relative to propensity-matched Medicare beneficiaries) in the Unites States [20]. Although comparisons between studies are limited due to differences in study populations, together these findings indicate that the relationship between multimorbidity, COC, and healthcare is complex. For dementia care in particular, careful attention will have to be placed on the quality, appropriateness, and quantity of interactions between the lead practitioner, the patient, family caregivers, and other components of the healthcare system such as home care delivery.
In this regard, there are multiple targets for future research. Detailed examination of the longitudinal relationships between the intensity of (formal and informal) long-stay home care services and physician continuity may provide insights into our null findings. Both informational and management continuity [42] may play a role as effect modifiers of the relationship between multimorbidity and health service use, but this has yet to be explored. In addition, differences in health system access may exist in rural versus urban areas, which has implications for home care use, frequency of physician visits, and use of acute and emergency services. We were unable to explore these differences in the current work as most (88%) of our study cohort resided in urban areas. Exploring these and other areas could be useful in identifying which individuals with dementia are most at risk and in need of intervention.
This study has several limitations requiring comment. This study was a retrospective cohort analysis of routinely collected health information and is therefore subject to limitations arising from the nature of the data [43]. We treated multimorbidity level and COC as fixed covariates in all analyses. Inclusion of time-varying variables in competing risk regression, however, can result in biased estimation [44]. We focused only on all-cause hospitalizations, without stratifying by specific reasons for hospitalization and ED visits. Such stratification could potentially inform interventions. We also focused only on clients with dementia. The effects of multimorbidity and physician continuity on a population without dementia may be different. Future studies comparing persons with versus without dementia may help to explain our findings. Strengths of this work include a large and representative population-based sample of community residents with dementia in the home care sector. Results would be expected to be generalizable to other jurisdictions with comparable healthcare systems that provide similar community-based services to persons with dementia. Whether our findings also apply to those living with dementia in the community who have not yet presented to the healthcare system is unknown and should be investigated in future studies. We used validated algorithms to define many chronic diagnoses, and the conditions we investigated are also consistent with those explored by other researchers and governing bodies to understand multimorbidity elsewhere [45,46]. Our models considered multiple competing risks and also adjusted for multiple but independent disease severity indicators that are not readily available in traditional health administrative databases. Linking to the RAI-HC data enabled this inclusion.
With increases in life expectancy and improvements to disease detection, the number of individuals living in the community with dementia and multimorbidity will increase. Our findings regarding the distribution of chronic disease burden and associated elevated risks of acute hospital admission and ED visits may be useful for healthcare providers and policymakers in identifying at-risk individuals with dementia in the community and setting priorities for care strategies. Heightened relational physician continuity did not modify the associations in this study. Additional research is therefore warranted to identify modifiable health system and other factors predictive of health outcomes to facilitate the development of effective interventions aimed at reducing costly health system use for this complex population.
|
10.1371/journal.pgen.1003564 | Disturbed Local Auxin Homeostasis Enhances Cellular Anisotropy and Reveals Alternative Wiring of Auxin-ethylene Crosstalk in Brachypodium distachyon Seminal Roots | Observations gained from model organisms are essential, yet it remains unclear to which degree they are applicable to distant relatives. For example, in the dicotyledon Arabidopsis thaliana (Arabidopsis), auxin biosynthesis via indole-3-pyruvic acid (IPA) is essential for root development and requires redundant TRYPTOPHAN AMINOTRANSFERASE OF ARABIDOPSIS 1 (TAA1) and TAA1-RELATED (TAR) genes. A promoter T-DNA insertion in the monocotyledon Brachypodium distachyon (Brachypodium) TAR2-LIKE gene (BdTAR2L) severely down-regulates expression, suggesting reduced tryptophan aminotransferase activity in this mutant, which thus represents a hypomorphic Bdtar2l allele (Bdtar2lhypo). Counterintuitive however, Bdtar2lhypo mutants display dramatically elongated seminal roots because of enhanced cell elongation. This phenotype is also observed in another, stronger Bdtar2l allele and can be mimicked by treating wild type with L-kynerunine, a specific TAA1/TAR inhibitor. Surprisingly, L-kynerunine-treated as well as Bdtar2l roots display elevated rather than reduced auxin levels. This does not appear to result from compensation by alternative auxin biosynthesis pathways. Rather, expression of YUCCA genes, which are rate-limiting for conversion of IPA to auxin, is increased in Bdtar2l mutants. Consistent with suppression of Bdtar2lhypo root phenotypes upon application of the ethylene precursor 1-aminocyclopropane-1-carboxylic-acid (ACC), BdYUCCA genes are down-regulated upon ACC treatment. Moreover, they are up-regulated in a downstream ethylene-signaling component homolog mutant, Bd ethylene insensitive 2-like 1, which also displays a Bdtar2l root phenotype. In summary, Bdtar2l phenotypes contrast with gradually reduced root growth and auxin levels described for Arabidopsis taa1/tar mutants. This could be explained if in Brachypodium, ethylene inhibits the rate-limiting step of auxin biosynthesis in an IPA-dependent manner to confer auxin levels that are sub-optimal for root cell elongation, as suggested by our observations. Thus, our results reveal a delicate homeostasis of local auxin and ethylene activity to control cell elongation in Brachypodium roots and suggest alternative wiring of auxin-ethylene crosstalk as compared to Arabidopsis.
| The plant hormone auxin is pivotal for root system development. For instance, its local biosynthesis is essential for root formation and growth in the dicotyledon model Arabidopsis. Thus, increasing interference with auxin biosynthesis results in increasingly shorter roots, partly because of reduced cell elongation. In this study, we isolated a hypomorphic mutant in an auxin biosynthesis pathway enzyme in the monocotyledon model Brachypodium. Counterintuitive, this mutant displays a dramatically longer seminal root, because mature cells are thinner, more elongated and therefore more anisotropic than in wild type. Interestingly, this phenotype can be mimicked in wild type by pharmacological interference with production of a key auxin biosynthesis intermediate, but also by interference with the biosynthesis of another plant hormone, ethylene. The latter controls auxin biosynthesis in Arabidopsis roots. Surprisingly however, auxin levels in the Brachypodium mutant are elevated rather than reduced, because of a simultaneous up-regulation of the second, rate-limiting step of the pathway. Ethylene normally represses this second step, suggesting an inverted regulatory relation between the two hormones as compared to Arabidopsis. Our results point to a complex homeostatic crosstalk between auxin and ethylene in Brachypodium roots, which is fundamentally different from Arabidopsis and might be conserved in other monocotyledons.
| The root system plays a fundamental role for plant growth and survival, not only by providing support, water and nutrients for the shoot, but also by participating in secondary functions, such as hormone biosynthesis or storage of photoassimilates [1], [2]. Root system architecture, that is the number and arrangement of different root types and their branching pattern, is highly plastic and determined by developmental and environmental factors that interact to optimize soil exploration. This is particularly important for the capture of growth limiting macronutrients, including nitrogen and phosphorus, whose edaphic distribution strongly influences post-embryonic root development and, therefore, root system architecture [2]–[4]. However, the root system can only respond to variation in such resources within its inherent developmental limits of growth rate and branching capacity, which are genetically determined. Optimization of root system architecture through breeding is therefore of particular interest in crops to increase root system plasticity with respect to biotic and abiotic stresses [5], [6].
Our knowledge about the molecular genetic control of root growth and branching has been largely obtained from analyses of the dicotyledon plant model system Arabidopsis thaliana (Arabidopsis) through mutagenesis approaches [2], [7]. The genes identified through these efforts have greatly benefitted the isolation of corresponding loci in monocotyledons, such as rice or maize [8]–[11]. Many of them encode proteins with regulatory functions, and among them components of plant hormone signaling pathways are particularly preeminent. For example, interference with the auxin-signaling pathway by mutation typically impairs primary root elongation or root branching, and in extreme cases even abolishes root formation [12]–[14]. The same is true for loss-of-function mutations in genes that encode enzymes involved in tryptophan-dependent auxin biosynthesis. In particular, auxin biosynthesis from tryptophan via indole-3-pyruvic acid (IPA) has been shown to be essential for root formation [15], [16]. Two enzyme classes define this pathway: the TRYPTOPHAN AMINOTRANSFERASE OF ARABIDOPSIS 1 (TAA1) and TAA1-RELATED (TAR) proteins, which catalyze the conversion of tryptophan to IPA; and the family of YUCCA cytochrome P450s, which catalyze the conversion of IPA to indole-3-acetic acid (IAA), the major active form of auxin [15]–[18]. Whereas the YUCCA genes were originally identified through a gain-of-function approach that led to auxin over-accumulation [19], TAA1/TAR genes were identified through loss-of-function approaches [16], [20], [21]. For instance, one study isolated the taa1 mutant because of its root growth resistance to the application of 1-aminocyclopropane-1-carboxylic-acid (ACC), a rate-limiting precursor for the biosynthesis of another hormone, ethylene [16]. This phenotype arises as a consequence of reduced auxin biosynthesis, which is normally up-regulated by ethylene through induction of TAA1/TAR gene expression. This finding also illustrates the dosage-dependent action of auxin, because although auxin and its perception are essential for root formation and growth, excess auxin application, biosynthesis or signaling are eventually inhibitory [22]–[24]. Indeed, it has been suggested that depending on the species, auxin levels might be supra-optimal for root growth [25].
Phylogenetic analysis has identified bona fide TAA1/TAR homologous genes in monocotyledons, with varying degrees of redundancy. For instance, whereas maize contains five genes of this family, only two are found in both rice and the monocotyledon model system, Brachypodium distachyon (Brachypodium) [26]. So far, only one TAA1/TAR-related mutant has been identified in monocotyledons, the vanishing tassel 2 (vt2) mutant of maize [26]. Despite the presence of multiple TAA1/TAR homologs in maize, vt2 null mutants display rather severe shoot phenotypes, such as dwarfism, reduced axillary meristem formation and associated impaired inflorescence development. Free auxin levels are reduced to ca. one third of wild type levels in vt2 mutants, suggesting that VT2 encodes the predominant TAA1/TAR activity in maize. Here we report the isolation and characterization of a Brachypodium mutant in the TAR2-LIKE (BdTAR2L) gene. Unlike vt2, this Bdtar2l mutant displays only mild shoot phenotypes. However, we observed dramatic root phenotypes, which surprisingly appear to result from upwardly disturbed auxin homeostasis.
In an effort to identify genetic factors that influence root system architecture in Brachypodium, we monitored seedlings from transgenic lines obtained in our lab through T-DNA transformation in tissue culture. One regenerated line stood out because of the occurrence of longer seminal (primary) roots (Fig. 1A–B), a phenotype that co-segregated recessively with the T-DNA insertion (χ2 test two-tailed p value = 0.7697). Isolation of the flanking genomic DNA by an inverse PCR strategy [27] revealed that this line contains only one T-DNA locus, whose integration site is located in the Bd2g04290 gene. Both the copy number and the insertion site were confirmed by whole genome sequencing of the homozygous mutant line (Fig. S1A). Bd2g04290 is one of the two TAA1/TAR homologs of Brachypodium, the other one being Bd2g34400 [26]. Based on their closest homologs in Arabidopsis, we named them Brachypodium distachyon TAR2-LIKE (BdTAR2L, Bd2g04290) and Brachypodium distachyon TAR1-LIKE (BdTAR1L, Bd2g34400), respectively. Quantitative RT-PCR (qPCR) to monitor expression of both genes in dissected seedling tissues indicated that BdTAR1L expression is dominant in the root meristem, whereas relative BdTAR2L expression increases strongly in the elongating and mature parts of the root, and in the shoot tissues (Fig. S1B). The T-DNA insertion in this Bdtar2l mutant is located 140 bp upstream of the ATG codon, thereby presumably disrupting the 5′ UTR, but not the coding sequence (Fig. 1C). To determine whether and to what extent the T-DNA insertion affects BdTAR2L expression, we quantified BdTAR2L mRNA levels by qPCR in 4-day-old seedlings. Indeed, expression was still detected both in shoot and root tissue, however at severely reduced levels of less than 20% and 5%, respectively, as compared to wild type or an unrelated transformant (the unrelated transformant line was included in all our assays to control for any tissue culture regeneration effects and contains a single copy T-DNA insert in a non-annotated, possibly repetitive region as determined by whole genome sequencing) (Fig. 1D–E). Therefore, it appears that our Bdtar2l mutation represents a hypomorphic allele, which we thus named Bdtar2lhypo. Interestingly, plants that are homozygous for the Bdtar2lhypo mutation display dramatically elongated roots (Fig. 1F–G), compared to which the shoot phenotype is rather mild. We could not detect any difference to wild type in the vegetative growth pattern, but observed a general decrease in overall leaf size to ca. 80% of wild type (Fig. 1H–M). Reproductive development in the mutant progresses normal without any apparent defects in inflorescence development, and plants are fully fertile.
A closer look at the mutant roots revealed that their phenotype is principally due to increased cellular anisotropy, which is most apparent in the post-meristematic, differentiated region. For instance, mature cortical cell length in Bdtar2lhypo roots reaches typically ca. 150% of wild type control (Fig. 2A–B), which would largely account for the overall increase in root length. We did indeed not observe a difference in root meristem size, measured as the number of cells that constitute the division and transition zones of the meristem in the central metaxylem cell file (Fig. 2C–D). Also, metaxylem cell length at equal position in the meristem is similar in Bdtar2lhypo and wild type up to the elongation zone, from where on cells elongate dramatically faster in Bdtar2lhypo than in wild type (Fig. 2C). At the same time, the transverse total as well as stele area of mature roots is reduced in Bdtar2lhypo to ca. 85% of wild type, accompanied by a slight reduction in the number of cells along the circumference of the innermost cortex layer (Fig. 2E–H). A quantitative analysis of transverse sections using an automated segmentation pipeline indicated that the number of cells in the outer six cell layers is indeed slightly reduced in Bdtar2lhypo mutants (Fig. 2I–J). Moreover, except in the epidermal layer, transverse cell area is in tendency smaller in Bdtar2lhypo (Fig. 2K). Therefore, mature root cells are overall thinner and longer than in wild type. Interestingly, this change in cellular anisotropy also manifests in the morphology of the root hairs, which are extensions of the epidermal cells and shorter in Bdtar2lhypo (Fig. 2L–M). In summary, while the decreased diameter of Bdtar2lhypo roots can be explained by a combination of a slight decrease in cell proliferation and in expansion in the radial dimension, their increased length can be attributed to enhanced cell elongation. Thus, the Bdtar2lhypo root phenotype largely results from increased cellular anisotropy.
To independently corroborate the effects of reduced BdTAR2L expression, we obtained another mutant allele from the Brachypodium T-DNA collection in which the gene is disrupted by a T-DNA insertion in the second intron (Fig. S1C) [28]. In semi-quantitative RT-PCR, a cDNA fragment comprising the borders of exons 1 and 2 was nearly undetectable (Fig. S1D), and compared to the Bdtar2lhypo allele, BdTAR2L expression in the root as monitored by qPCR was even more severely reduced, to 1–2% of wild type levels (Fig. 3A). However, since we could not exclude production of some residual full-length transcript, we designated this allele a quasi-null mutant (Bdtar2lqnull). Compared to their wild type background, Bd21-3, Bdtar2lqnull mutants again display an elongated root phenotype, which is however not as drastic as in Bdtar2lhypo mutants (Fig. 3B–C). This could again be largely attributed to increased cell elongation, which reaches about 125% of wild type (Fig. 3D–E). Moreover, Bdtar2lqnull mutants also display shorter root hairs (Fig. 3F). At the same time, transverse root and stele area are reduced to about the levels observed in Bdtar2lhypo mutants, without a change in the number of cortical cell layers (Fig. 3G–I). Therefore, similar to Bdtar2lhypo, Bdtar2lqnull mutants display increased cell elongation and cellular anisotropy in the root. Unlike in Bdtar2lhypo mutants, however, exaggerated root growth is not sustained in Bdtar2lqnull mutants although the enhanced cell elongation is maintained (Fig. 3J). This is because of a gradual consumption of the root meristem as development proceeds (Fig. 3K). Compared to Bdtar2lhypo, Bdtar2lqnull mutants also display more severe shoot phenotypes, notably a clearly reduced shoot length in young seedlings (Fig. 3L) and a dwarf stature as an adult (Fig. 3M), which is accompanied by severely reduced fertility. Collectively, our mutant characterizations therefore suggest that the Bdtar2lhypo and Bdtar2lqnull mutants indeed represent an allelic series that displays the consequences of gradually reduced BdTAR2L dosage.
The low BdTAR2L expression level in the mutants is not compensated by up-regulated BdTAR1L expression (Fig. S1E) and therefore should result in overall decreased tryptophan aminotransferase activity. However, the Bdtar2lhypo long root phenotype is counterintuitive in this respect, because progressive loss-of-function of TAA1/TAR activity in Arabidopsis leads to progressively impaired rather than enhanced root growth [16]. The same is true when Arabidopsis wild type plants are grown on a specific competitive inhibitor of TAA1/TAR enzymes, L-kynerunine [29]. To test whether L-kynerunine also inhibits root growth in Brachypodium, we transferred 2-day-old seedlings onto media with different L-kynerunine concentrations and assayed root growth two days later. Strikingly, root elongation was stimulated rather than inhibited already at concentrations as little as 1 µM (Fig. 4A–B). Higher concentrations, up to 100 µM, strongly promoted root elongation up to 150–200% of the mock controls. Moreover, we observed exaggerated cell elongation upon L-kynerunine treatment (Fig. 4C–D), which therefore mimics the Bdtar2lhypo root phenotype. Interestingly, unlike wild type, the Bdtar2lhypo mutant hardly responded to L-kynerunine treatment (Fig. 4A–D). Finally, similar to the Arabidopsis taa1 mutant [20], both Bdtar2l alleles were hypersensitive to the application of the toxic tryptophan analog, 5-methyl-tryptophan (Fig. 4E–F), which is an artificial substrate for TAA/TAR enzymes. 5-methyl-tryptophan can be detoxified by its conversion to IPA [20], and therefore 5-methyl-tryptophan hypersensitivity is indicative of reduced IPA production. Thus, the data are consistent with the idea that reduced TAA1/TAR activity in the Bdtar2l mutants is indeed responsible for the mutant phenotypes.
While other auxin-dependent processes, such as gravitropism, appeared unaffected in Bdtar2l mutants (Fig. S1F), we also observed a root system branching phenotype. In Bdtar2lhypo mutants, coleoptile node root formation is slightly reduced (Fig. 5A), but unlike the seminal roots, coleoptile node roots elongate normally (Fig. 5B). Contrary to the coleoptile node root phenotype, the number of emerged lateral roots from the seminal root is increased in Bdtar2lhypo (Fig. 5C). This increase is also evident once lateral root number is normalized for total root length (Fig. 5D), even if the total number of lateral roots is small. Because it was difficult to follow this phenotype over a longer period in the tissue culture system (due to the limited growth space on our 20 cm dishes) [30], we employed an alternative assay, i.e. lateral root emergence that has been triggered by removal of the seminal root meristem. In this assay, Bdtar2lhypo mutants showed enhanced lateral root formation capacity (Fig. 5E), again also holding up once normalized for seminal root length (Fig. 5F). Again, this phenotype could be copied by L-kynerunine treatment of wild type (Fig. S1G). Considering that increased Bdtar2lhypo root length can be largely explained by cell elongation, it therefore appears that Bdtar2lhypo mutants have a genuinely higher capacity of seminal root branching.
Collectively, our genetic as well as pharmacological analyses suggest that reduced tryptophan aminotransferase activity in Brachypodium results in increased root cell elongation and anisotropy. This contrasts with gradually reduced root growth in Arabidopsis taa1/tar single and double mutants. As expected, in Arabidopsis this root growth reduction is accompanied by gradually decreased free auxin levels [17]. Thus, the most parsimonious explanation for the Bdtar2l phenotype is that auxin levels might normally be supra-optimal for cell elongation in Brachypodium, similar to what has been proposed for rice [25]. To our surprise then, we found that free auxin levels are elevated rather than reduced in Bdtar2l seminal roots (Fig. 6A; Fig. S1H), in particular in the elongating and mature parts where the expression of BdTAR2L is relatively high as compared to BdTAR1L (Fig. S1A). Consistently, elevated auxin levels where also observed upon L-kynerunine treatment (Fig. S1I). To determine whether this could arise from compensatory up-regulation of proposed alternative auxin biosynthesis pathways [31], we checked the expression of various homologs of corresponding rate-limiting enzyme genes in Bdtar2l roots, i.e. AMIDASE-LIKE 1-LIKE (BdAMI1L), NITRILASE 1-LIKE (BdNIT1L), ALDEHYDE OXIDASE 1-LIKE (BdAO1L) and BdAO2L. However, with the exception of a slight increase in BdAO1L expression, no significant upward changes were detected (Fig. 6B). By contrast, the expression of four YUCCA homologs, selected for the reported root-specific expression of their respective counterparts in rice [32], is significantly up-regulated in Bdtar2l roots (Fig. 6C; Fig. S1J), amounting to more than triple in combined transcript levels in Bdtar2lhypo (Fig. 6C) and one-and-a-half in Bdtar2lqnull, corresponding with the respective auxin levels. The increased BdYUCCA expression could account for the increased auxin levels, because it has been determined that YUCCA gene expression is rate-limiting for auxin biosynthesis via the IPA pathway [17], [33].
Root growth resistance to enhanced ethylene production, conferred by application of ACC, contributed to the isolation of the taa1/tar mutants in Arabidopsis, because ethylene promotes auxin biosynthesis via the IPA pathway through transcriptional regulation of TAA1/TAR and YUCCA genes [16], [34]. By contrast, we found that expression of BdTAR2L and BdTAR1L is only mildly ethylene-responsive (Fig. 6D). Moreover, the expression of the four BdYUCCA genes tested is negatively regulated by ACC application (Fig. 6E). Thus, in Brachypodium, the ethylene pathway might repress rather than promote auxin biosynthesis via the IPA pathway, mainly by down-regulating BdYUCCA expression. A prediction from this observation is that the Bdtar2l root phenotypes might be rescued by enhanced ethylene signaling. To test this notion, we transferred 2-day-old Bdtar2lhypo seedlings onto media containing increasing amounts of ACC and monitored root growth over the two days that followed. Indeed, ACC treatment strongly impaired Bdtar2lhypo root elongation and reduced growth to about the level of wild type mock controls (Fig. 7A–B). Moreover, ACC treatment restored cell elongation to wild type length in Bdtar2lhypo (Fig. 7E–F).
The above results suggested that inhibition of ethylene biosynthesis or signaling in Brachypodium roots should mimic the Bdtar2lhypo root phenotype. We tested this notion by transferring 2-day-old seedlings onto media that contained aminoethoxyvinylglycine (AVG), an inhibitor of a rate-limiting enzyme in ethylene biosynthesis, ACC synthase [35]. Following root growth over the two days that followed revealed that Bdtar2lhypo roots are largely resistant to AVG, while wild type roots display a dramatic increase in elongation that approached the levels observed in Bdtar2lhypo (Fig. 7C–D). Investigation of cortical cells revealed that again this effect could be explained by increased cell elongation (Fig. 7E, G). Higher levels of AVG eventually slowed down elongation rate of Bdtar2lhypo roots, but still promoted root elongation in wild type. A cautionary note on AVG is that it not only inhibits ACC synthase, but also other enzymes that require pyridoxal 5′-phosphate (PLP) as a cofactor [36], [37]. Since the activity of TAA1/TAR enzymes is stimulated by PLP [16], it appears possible that AVG treatment impairs their function to some degree, mimicking L-kynerunine treatment. Thus, for independent confirmation we took advantage of a mutant from the Brachypodium T-DNA collection, in which a homolog of the Arabidopsis gene ETHYLENE INSENSITIVE 2 (EIN2), an essential positive regulator of ethylene signaling [38]–[40], carries a T-DNA insertion in the promoter, 469 bp upstream of the start codon (Fig. 7H). As a consequence, expression of this EIN2-LIKE (BdEIN2L1, Bd4g08380) gene is significantly down-regulated (Fig. 7I). Strikingly, this hypomorphic mutant (Bdein2l1hypo) displays a Bdtar2l root phenotype (Fig. 7J–K), and while this is not accompanied by up-regulation of BdTAR1L or BdTAR2L (Fig. 7L), it is accompanied by increased BdYUCCA expression (Fig. 7M). Finally, similar to Bdtar2l mutants, auxin levels are elevated in the elongating parts of Bdein2l1hypo roots (Fig. S1K), thereby corroborating our above findings.
The observed stimulatory effects of L-kynerunine and AVG treatment on root elongation have not been described for Arabidopsis. However, given the morphological differences between Arabidopsis and Brachypodium roots, in particular the more than three-fold difference in thickness, it is conceivable that the concentration of those substances required for root penetration and biological action might be different as well. The described largely inhibitory effect of those treatments on root elongation in Arabidopsis could therefore have resulted from application of too high concentrations. These considerations prompted us to revisit the response of Arabidopsis to an extended concentration range of both L-kynerunine and AVG. Interestingly, relatively low concentrations as compared to Brachypodium of both treatments indeed slightly promote root elongation (Fig. S1L–M), although by far not as strong as in Brachypodium.
The root systems of dicotyledons and monocotyledons display some fundamental differences in their organization and ontogeny, as exemplified by the respective model systems, Arabidopsis and Brachypodium [30]. Despite these differences, the principal genes involved in root formation, growth vigor and branching are expected to be homologous in the two systems. This is based on experience in other species such as maize, where several causative mutations that affect root system development are in homologs of auxin signaling components [9], [41]. The effect of manipulating the IPA branch of auxin biosynthesis has been investigated in another monocotyledon crop, rice, through gain- and loss-of-function approaches. For instance, both over-expression and down-regulation of the YUCCA homolog OsYUCCA1 by transgenic means results in strongly reduced root growth [42], whereas a knockout in another YUCCA homolog, CONSTITUTIVELY WILTED 1, displays reduced root branching [43]. Compared to those mutants, the enhanced root elongation phenotype of Bdtar2l mutants is unusual. Our initial interpretation was therefore that auxin levels are supra-optimal for cell elongation in the Brachypodium seminal root, as has been suggested for seminal root growth in rice [25]. However, repeated independent measurements of multiple samples clearly indicated that auxin levels are increased rather than decreased in Bdtar2l mutant roots. This is particularly pronounced in the Bdtar2lhypo allele, and in tendency also observed in the Bdtar2lqnull allele, correlating with quantitatively corresponding BdYUCCA up-regulation. The comparatively severe shoot phenotypes of the Bdtar2lqnull allele, its less pronounced root cell elongation, and the observation that the root meristem gradually breaks down as development progresses indicate that compared to the Bdtar2lhypo allele, IPA levels are eventually limiting in Bdtar2lqnull mutants. This idea is supported by the dose-response curve of wild type to L-kynerunine, where increasing amounts promote cell elongation up to a certain threshold, beyond which root growth is inhibited. Further corroborating this idea, a threshold also exists for the Bdtar2lhypo mutant, which moreover is hypersensitive to L-kynerunine treatment as concentrations that still promote root elongation in wild type are inhibitory in Bdtar2lhypo.
A similar dose-response curve is observed for AVG treatment, which inhibits the rate-limiting step in ethylene biosynthesis, but might also impinge on TAA1/TAR activity because of its generic action on enzymes that use PLP as a co-factor [16], [36], [37]. Stimulation of root growth by AVG treatment has also been reported for rice [25], although the reported dosage response is quantitatively different from our assays with Brachypodium. For instance, while in rice 0.05 µM AVG promoted root growth and 1.0 µM was already inhibitory, in Brachypodium 5.0 µM was still stimulating. In part, this could be due to technical issues, for instance the concentration needed in the tissue culture media to reach the same tissue penetration in roots of different thickness or cell permeability. In light of our results, it appears possible that the response of rice to AVG treatment is similar to Brachypodium, i.e. that it could reflect a combined effect of reducing TAA1/TAR as well as ACC synthase activity, thereby boosting auxin levels by removing the inhibitory effect of ethylene on YUCCA expression as long as interference with TAA/TAR1 activity does not lead to limiting IPA levels. The finding that YUCCA expression is rate limiting for auxin biosynthesis in Arabidopsis [17], [33] supports this interpretation, suggesting that this is also likely the case in Brachypodium and/or rice.
Corroborating the effects of AVG application and circumventing its ambiguity, the root phenotype of the Bdein2l1hypo mutant confirms the involvement of the ethylene-signaling pathway in auxin homeostasis. However, based on the observed regulatory logic of this hormone crosstalk, a central finding of our study is that the regulation of the IPA branch of auxin biosynthesis through the ethylene pathway observed in Arabidopsis roots might not be conserved in Brachypodium. This idea is based on several convergent observations, for instance that unlike their Arabidopsis counterparts, expression of BdTAR2L as well as BdTAR1L is hardly ethylene-responsive or that BdYUCCAs are repressed upon ACC treatment and up-regulated in Bdein2l1hypo, consistent with the latter's Bdtar2l phenotype. Moreover, unlike Arabidopsis taa1/tar mutants, Bdtar2lhypo mutants are not ACC-resistant. Rather, ACC treatment essentially restores the Bdtar2lhypo phenotype to wild type. Thus, our data thus support a scenario in which the effects of auxin biosynthesis through the IPA branch on root cell elongation are mediated by the ethylene pathway rather than vice versa. Such an inversion of a regulatory relationship could alternatively reflect a shift in the key nodes of the regulatory network linking auxin and ethylene through feedback loops and is a simple way for evolutionary adaptation. Indeed, feedback of ethylene on auxin biosynthesis by repressing YUCCA expression, rather than promoting TAA1/TAR as well as YUCCA expression as in Arabidopsis, is a central feature of the mutant phenotypes described in our paper (Fig. 8). How this feedback is mediated remains unclear for the moment. The recent discovery of an enzymatic link between auxin and ethylene biosynthesis suggests that this crosstalk might very well respond directly to IPA levels [44]. Hypomorphic mutants, such as those employed in our study, might become a crucial tool in future efforts to elaborate such a scenario.
Molecular biology and genetics procedures, such as genomic DNA isolation, genotyping, sequencing or qPCR were performed according to standard procedures as described [45], [46].
The community standard diploid inbreed Brachypodium distachyon line Bd21 was used for transformation and as a control in all experiments [47], except for the Bdtar2lqnull mutant (stock id JJ9248.0) and the Bdein2l1hypo mutant (stock id JJ110.0), which together with their Bd21-3 wild type background line were obtained from a Brachypodium T-DNA collection (http://brachypodium.pw.usda.gov/TDNA/) [28]. Genotyping, for instance to establish homozygous mutant lines, was performed using oligonucleotides 5′-CGT GAG AGC TAG TGG GAT AG-3′ and 5′-ATG GGT GGC TGA TGG CGT AG-3′ (BdTAR2L wild type allele for Bdtar2lhypo), 5′-CGT GAG AGC TAG TGG GAT AG-3′ and 5′-TTG AAG GAG CCA CTC AGC CGC G-3′ (Bdtar2lhypo T-DNA insertion); 5′-GCG GTT CCC TGT TCA TCT TC-3′ and 5′-CAC AGC GAA ACA ACA CAC AG-3′ (BdTAR2L wild type allele control for Bdtar2lqnull), 5′-GCG GTT CCC TGT TCA TCT TC-3′ and 5′-TAC GAG CCG GAA GCA TA AAG-3′ (Bdtar2lqnull T-DNA insertion); 5′-GTA CCT TTC TCC GTC AAG AG-3′ and 5′-GAA GGA GGC ATC AGG ACA TG-3′ (BdEIN2L1 wild type allele), 5′- GTA CCT TTC TCC GTC AAG AG -3′ and 5′-CTC CGC TCA TGA TCA GAT TG-3′ (Bdein2l1hypo T-DNA insertion); Arabidopsis thaliana experiments were performed with the standard Col-0 accession. For tissue culture growth, the lemma of mature seeds was carefully peeled off with forceps before seed sterilization in 1 ml of 70% ethanol per seed for 1 min. After ethanol removal, seeds were soaked in a solution of 1.3% sodium hypochlorite plus one drop of Tween-20 per 50 ml for 5 min. with gently rocking, then rinsed with sterile deionized water three times. The sterilized seeds were stratified for 2 days at 4°C to ensure synchronous germination on vertically oriented 10 or 24 cm square plates of half-strength Murashige-Skoog (MS) media (2.45 g/l MS salts with vitamins, 1% sucrose, 1% agar, pH 5.7) in a growth chamber under continuous light of 100–120 µE intensity at 22°C. To quantify leaf number, sheath/blade length and blade width, 2-day-old Brachypodium seedlings were transferred into pots with soil, watered every 2–3 days and incubated at 22°C under a 20 h photoperiod. Leaf features were measured 18 days after germination (dag), crown roots were counted 25 dag. Arabidopsis seedlings were grown as described [46].
To determine root length, seedlings growing in vertically oriented plates were either scanned or photographed with a digital camera to measure root length using the ImageJ software, version 1.47b. For lateral root quantification after seminal root meristem removal, 2 mm of the root tip were cut from the seminal root of 4-day-old plants with a scalpel. The number of visible lateral roots was then scored 4 days later. For gravitropism assays, Brachypodium seeds were germinated for 2 days in vertically oriented plates. To induce gravitropic response, plates were then rotated 90° and grown for another 24 hours. Plates were scanned on a flatbed scanner before and after gravitropic stimulation.
Embryonic calli generation of Bd21 was performed according to [48], subsequent transformation with the pVec8GFP plasmid and plant regeneration according to [47], and retrieval and mapping of the region flanking the right border of the T-DNA insert in Bdtar2lhypo mutants according to [27]. A total of 48 transgenic lines were produced, among which the Bdtarlhypo mutant was a chance hit.
Whole genome sequencing of genomic DNA isolated from Brachypodium seedlings was performed on the Illumina HiSeq 2000 platform, generating more than 250 mio. paired-end reads of 100 bp length. The Bowtie 2 software [49] was used for the alignment on the Brachypodium distachyon reference genome (http://mips.helmholtz-muenchen.de/plant/brachypodium/download/index.jsp), revealing coverage of ca. 100 reads per bp. For detection of T-DNA insertions, reads that aligned on the T-DNA reference sequence were selected for alignment on the genome. This procedure confirmed the localization of the Bdtar2lhypo insert on chromosome 2 (position: 3,030,511). The precise position of the control line insert remains undetermined because it could not be mapped to a unique annotated region, however it is clear that it does not disrupt any annotated gene. Finally, coverage of the T-DNA reference sequence was similar to genome coverage, confirming the presence of a single insertion in both sequenced genomes.
The hormone and inhibitor treatments were done on plates, except in the case of qPCR, for which treatments were carried out in liquid media for 3 h. Briefly, Brachypodium seeds were germinated on standard plates as described above. At 2 days after germination, seedlings were then transferred to media containing the respective hormone or inhibitor, or mock. For Arabidopsis treatments, 4-day-old seedlings were transferred.
Auxin measurements were performed on eight independent samples of pooled roots per genotype excised from 4-day-old seedlings as described [50].
Seminal roots of 4-day-old seedlings were fixed in a solution of 1% glutaraldehyde, 4% formaldehyde and 50 mM sodium phosphate buffer (pH 7.2). Fixed roots were thoroughly rinsed four times with water. To determine transverse root and cell area, roots were cut into 0.5–1 cm pieces and embedded in 6% agarose. Sections of 75 µm were obtained approximately 2 cm from the root tip using a Leica-VT 1000S vibratome. Sections were stained with 0.1% toluidine blue solution for 30 s and washed. For quantification of cortical cell length, unstained roots were cleared with 10% potassium hydroxide solution at 95°C for 30 min. Roots were mounted on glass slides with 50% glycerol and photographed either in light field or differential interference contrast using a Leica DM5500B compound microscope. For visualization of meristem structure, seminal roots were stained following the mPS-PI procedure [45] before imaging with a Zeiss LSM 700 confocal microscope. Cortical cell length, root hair length, meristem size and central metaxylem cell length were quantified using the ImageJ software, version 1.47b.
qPCR reactions were performed using a Stratagene MxPro 3005P Real-Time PCR System (Stratagene). Three technical replicates were analyzed for each sample. The specificity of each amplification reaction was verified by DNA melting curve analysis and gel electrophoresis of the amplified products. Not reverse transcribed samples and non-template controls were included in every assay to rule out genomic DNA contamination. The final threshold cycle (Ct), efficiency and initial fluorescence (R0) for every reaction were calculated with the Miner algorithm [51]. Relative expression levels were obtained from the ratio between R0 of the target gene and R0 of the reference gene, UBIQUITIN-CONJUGATING ENZYME 18 (BdUBC18). The following oligonucleotides were used: BdUBC18 (Bd4G00660), 5′-GGA GGC ACC TCA GGT CAT TT-3′ and 5′-ATA GCG GTC ATT GTC TTG CG-3′; BdTAR1L (Bd2G34400), 5′-GAA TCG GGA TGG TGG CCT CG-3′ and 5′-ATT GTC GGA TCG CCG TGA TC-3′; BdTAR2L (Bd2G04290), 5′-GGC TCC ATA CTA CTC TTC GTA TC-3′ and 5′-CAG TAG TAG GCC AGG TCG TG-3′; BdYUCCA1L (Bd1G28967), 5′-GCA ATG GCT CAA GGG AAG TG-3′ and 5′-TGT GGC AGT TTG ATG CTT CC-3′; BdYUCCA7L (Bd1G00587), 5′-GCA GTG GCT CAA GGG AAG C-3′ and 5′-TGT GGT ATG CTG TGG CGA TG-3′; BdYUCCA8L (Bd5G01327), 5′-CCC AGT TCA TCT CCT ACC TC-3′ and 5′-GGT ACT CGA CGG TGG ACT TC-3′; BdYUCCA13L (Bd2G10302), 5′-GTC GTC CGC AGC GAG CTT CA-3′ and 5′-GGG GGT TTG GAG CTT CAT GG-3′; BdAMI1L (Bd5G27490), 5′-CGA CTT CTC CCT CGG AAC TG-3′ and 5′-GTT GCT GAC GCG AGA CAA TG-3′; BdNIT1L (Bd3G49620), 5′-CCC CTG CCA CCA TTG ATA AAG-3′ and 5′-GTC TTC TTT TCC CTT GGC AG-3′; BdAO1L (Bd1G52740), 5′-GGC TGT GGC GAA GGT GGA TG-3′ and 5′-ACC CTC AGT GGT GAT AAC TG-3′; BdAO2L (Bd1G56667), 5′-GTG GAC CCA GTG CAA ATG TG-3′ and 5′-CAT ATA CAG CCT CCC CAG AAG-3′; BdEIN2L1 (Bd4G08380), 5′-AGA ATC TTG CCC AGA TTT GC-3′ and 5′-GCA AAC CAT ATG CCT GTG AG-3′;
|
10.1371/journal.pcbi.1004823 | A New Fiji-Based Algorithm That Systematically Quantifies Nine Synaptic Parameters Provides Insights into Drosophila NMJ Morphometry | The morphology of synapses is of central interest in neuroscience because of the intimate relation with synaptic efficacy. Two decades of gene manipulation studies in different animal models have revealed a repertoire of molecules that contribute to synapse development. However, since such studies often assessed only one, or at best a few, morphological features at a given synapse, it remained unaddressed how different structural aspects relate to one another. Furthermore, such focused and sometimes only qualitative approaches likely left many of the more subtle players unnoticed. Here, we present the image analysis algorithm ‘Drosophila_NMJ_Morphometrics’, available as a Fiji-compatible macro, for quantitative, accurate and objective synapse morphometry of the Drosophila larval neuromuscular junction (NMJ), a well-established glutamatergic model synapse. We developed this methodology for semi-automated multiparametric analyses of NMJ terminals immunolabeled for the commonly used markers Dlg1 and Brp and showed that it also works for Hrp, Csp and Syt. We demonstrate that gender, genetic background and identity of abdominal body segment consistently and significantly contribute to variability in our data, suggesting that controlling for these parameters is important to minimize variability in quantitative analyses. Correlation and principal component analyses (PCA) were performed to investigate which morphometric parameters are inter-dependent and which ones are regulated rather independently. Based on nine acquired parameters, we identified five morphometric groups: NMJ size, geometry, muscle size, number of NMJ islands and number of active zones. Based on our finding that the parameters of the first two principal components hardly correlated with each other, we suggest that different molecular processes underlie these two morphometric groups. Our study sets the stage for systems morphometry approaches at the well-studied Drosophila NMJ.
| Altered synapse function underlies cognitive disorders such as intellectual disability, autism and schizophrenia. The morphology of synapses is crucial for their function but is often described using only a small number of parameters or categories. As a consequence, it is still unknown how different aspects of synapse morphology relate to each other and whether they respond in a coordinated or independent manner. Here, we report a sensitive and multiparametric method for systematic synapse morphometry at the Drosophila Neuromuscular Junction (NMJ), a popular model for mammalian synapse biology. Surveying a large NMJ image repository, we provide insights in the natural variation of NMJ morphology as a result of differences in gender, genetic background and abdominal body segment. We show which synapse parameters correlate and find that parameters fall into five groups. Based on our findings, we propose that two of them, NMJ size and geometry, are controlled by different molecular mechanisms. Our study provides insights into the design principles of a model synapse and tools that can be applied in future studies to identify genes that modulate or co-orchestrate different aspects of synapse morphology.
| Normal brain function relies on functional neuronal networks in which neurons connect and communicate with one another. Communication primarily takes place at chemical synapses, where neurotransmitters are released from the presynaptic compartment of a neuron and activate receptors at the postsynaptic compartment of its target cell. Abnormal synaptic development and function have been found to underlie cognitive disorders such as intellectual disability, autism spectrum disorder and schizophrenia [1–5]. The morphology and function of synapses are highly intertwined [6–9] and morphological aspects have therefore been studied extensively to gain further insight into the regulatory networks underlying synaptic function. Mammalian dendritic spines change shape upon maturation and plasticity from long, thin filopodia-like structures to typical stubby and mushroom-shaped postsynaptic compartments of increased efficacy [10,11]. In Drosophila, synaptic structure and activity is modulated according to circadian timing [12–15] or upon experienced-dependent or stimulated activity [16,17], to name only three examples.
Despite the central interest in synapse morphology in neuroscience -studied at different developmental stages, upon genetic or environmental perturbation and in different organisms- it is still largely unknown how different structural aspects relate to one another and adapt in a coordinated manner when changes are induced. Systematic synapse morphometry could shed light on these poorly understood relationships. In genetically unperturbed conditions, such insights would be crucial to understand the developmental design principles that shape the synapse. This in turn would provide a basis to identify the genetic players that drive the required coordinated structural changes during synaptic development and plasticity with higher sensitivity.
As an initial step into quantitative, correlative synapse morphometry, we have turned to an identifiable, methodologically and genetically accessible synaptic terminal: the Drosophila larval neuromuscular junction (NMJ). The Drosophila NMJ is an extensively studied and well-established in vivo model for glutamatergic synapse biology [18,19]. The synaptic terminal, a branched chain of synaptic boutons, is formed by the motor neuron and gets surrounded by the subsynaptic reticulum (SSR) as it invades its target muscle [20]. Boutons are periodic enlargements [21] that host the presynaptic release sites, ‘active zones’ [20], at which the synaptic vesicles dock to the presynaptic membrane to release their neurotransmitters. Together with the exactly opposed postsynaptic receptor complex, the active zone forms the chemical synapse [20]. Large scale genetic screens at the NMJ have been very successful in identifying genes and molecular mechanisms of synapse development [22–31]. However, so far, these screens have largely relied on visual inspection and semi-quantitative scoring of a limited amount of morphometric features. While this has uncovered main determinants of NMJ morphology, it is likely that the extent of the regulatory networks remained undiscovered.
In this study, we developed a macro in Fiji (an open-source image analysis software [32]) to quantitatively assess nine morphometric features in a large number of glutamatergic NMJs based on high-content fluorescence microscopy images. We found the macro to accurately assess eight of them (NMJ area, perimeter, total length, longest branch length, number of islands, number of branches and branching points and number of active zones), making it suitable for high-throughput analyses of synapse morphology. Here, in preparation for reverse genetic approaches, this method was applied to two isogenic host strains of genomewide RNAi libraries (VDRC [33]; see Methods) to build large wt-like control datasets. Using these data, we followed a systems biology approach by using the differences in gender, abdominal body segment and genetic background as natural sources of biological variation to gain insights into the (in)dependencies and correlations of the measured morphometric NMJ features. Our study is the first to investigate the systems properties of the well-studied Drosophila NMJ, providing new insights into the design principles of a synapse.
We generated a large collection of NMJ images in two different genotypes, the isogenic host strain for the GD and KK RNAi libraries of the Vienna Drosophila Resource Center [33] crossed to a panneuronal elav promotor line (see Methods). The obtained larvae were dissected and stained with antibodies against two key components of the NMJ, the discs large 1 protein (Dlg1 –the ortholog of mammalian PSD-95) and bruchpilot (Brp—sole ortholog of human ELKS/CAST/ERC proteins [34]), to visualize general synapse morphology and active zones [35], respectively. We focused on abdominal segments A2-A5, which are best accessible in larval ‘open book’ preparations. In total we acquired microscopic images of 1576 NMJs in 397 larvae.
It is a laborious undertaking to measure NMJ features (semi) manually, especially when several NMJ features are of interest. To support high-throughput analyses and achieve objective quantification, we set out to develop a macro for computer-assisted morphometry that can accurately quantify high-content, non-confocal images.
The macro ‘Drosophila_NMJ_morphometrics’, was developed using the open source Fiji platform [32] and is made available via figshare, a public repository where users can make their research outputs available: https://figshare.com/s/ec634918c027f62f7f2a [36]. For usage, Fiji needs to be installed and the macro has to be downloaded and saved with the extension (.ijm) into the Fiji Plugins folder. The macro will appear in the Plugins menu under the name Drosophila_NMJ_Morphometrics. Upon running the macro a graphical interface displays the default settings of the macro, which can be adjusted according to the customer’s needs. The ‘help’ option offers additional information. A point-by-point protocol for using the macro is provided in the supplementary material (S1 Protocol).
The macro consists of three sub macro’s that can be used separately or run in a consecutive manner to analyze and process images (via checkbox options of the macro interface). The first sub macro “Convert to stack” identifies all image files available and creates stacks and maximum intensity projections of both channels. The second sub macro “Define ROI” presents the projections to manually delineate the region of interest (ROI). As we were interested in type 1b NMJs on muscle 4, this manual step was required to exclude type 1s synaptic terminals on the same muscle, and occasionally exclude synapses on nearby muscles that are present in the images. The third sub macro “Analyze” applies fully automated analysis through all stacks within the limits of the ROI. For each NMJ, nine morphological parameters are measured (described in more detail below) and processed to an (.txt) output file. Images are processed to a result picture, in which the delineation of the automatically recognized NMJ features is presented.
During image analysis, from each NMJ three structures were derived: 1) NMJ outline, 2) NMJ skeleton and 3) number of Brp-positive active zones. Technical details underlying each derived structure are described in the Methods section Image Analysis. The NMJ outline is used to determine the NMJ area and its perimeter and a subsequent watershed separation provides the number of boutons (Fig 1A, NMJ outline indicated in yellow). From the skeleton (Fig 1A, indicated in blue) five NMJ features are deduced: the total NMJ length, the sum of the length of the longest continuous path connecting any two end points (longest branch length), the number of unconnected Dlg1-stained compartments per NMJ (referred to as ‘islands’), the number of branches and the number of branching points (one branching point connects three or more branches). The number of active zones was determined by counting Brp-positive spots in the Brp-channel (Fig 1A, indicated by white foci). Taken together, the macro determines three derivatives per NMJ, from which it deduces nine morphological NMJ features. Eight of the nine features are based on the Dlg1- and one on the Brp-channel. Fig 1A provides a schematic overview of all nine NMJ features.
The Fiji macro was used to process our dataset of 1576 NMJ images. We checked the quality of our images at low magnification (Fig 1B, input checkpoint 1) and found that for 33 images (2.09%) the NMJ was not fully captured in the acquired stack. Furthermore, a second checkpoint at high magnification (Fig 1B, input checkpoint 2) revealed 35 images (2.22%) in which the NMJ was partially out-of-focus, 25 images (1.59%) with weak staining and 21 images (1.33%) from which we could not guarantee the specificity of type 1b. For the remaining 1468 NMJ images (93.15%) (Fig 1C), the macro-annotated images were used to evaluate the obtained NMJ outline, skeleton and Brp-positive active zones per NMJ image (Fig 1B, output checkpoint 3). Variability in staining intensity led to a relatively high amount of images from which part of the NMJ outline was not recognized (n = 90, 6.13% from 1576), certain areas that lack Brp-positive active zones (n = 121, 8.24%) or a combination of these two events (n = 95, 6.47% from 1576). Images with skeleton misannotations (n = 186, 12.67% from 1576) were manually corrected. In summary, after three rounds of quality checks, we remained with a NMJ dataset of 1295 images for the NMJ-outline features (82.17% from initial; 86.62% from 1468), 1383 images for the NMJ-skeleton features (87.75% from initial; 92.51% from 1468) and 1259 images for the active zones (79.89% from initial; 84.21% from 1468). In total, we obtained 1163 NMJ images from which all features past the three quality rounds (73.79% from initial, 79.22% from 1468).
In absence of truly objective NMJ measures, we compared the results obtained with the macro to the manual counts of two experienced experimenters for 30 NMJ images (S1 Table). We first investigated the sample distributions to determine the deviation between manual and macro assessment over the complete set of images. The 95% confidence intervals largely overlapped with each other for the parameters NMJ area, perimeter, length, longest branch length, islands, branches, branching points, and active zones (S1 Fig). Thus, no significant differences were found between the distributions of macro and manual assessment for these NMJ features (Table 1). However, the macro resulted in a significantly lower amount of bouton counts (macro: 16 boutons per NMJ; manual; 25 per NMJ; p<0.0001) (Table 1). The 95% confidence interval widths were highly comparable between macro and manual counts, indicating that the macro does not add additional noise to the outcome (S1 Fig). Secondly, we investigated the deviation between manual and macro evaluation per given sample, expressed as %deviation or sensitivity and specificity. The %deviation per given sample was often negative for the NMJ perimeter, length and longest branch length (S1 Table), indicating that the macro measures somewhat higher absolute values as compared to the manual counts. On average, the boutons showed a six times higher % deviation between macro and manual counts compared to the NMJ area, perimeter, length and longest branch length (Table 1). Sensitivity (the proportion of positive results that is indeed a true positive) and specificity (the proportion of true positives that is identified as such) was determined per NMJ image for the discrete NMJ features islands, branches, branching points and active zones (S1 Table). On average, all four parameters scored >91% on sensitivity and >92% on specificity (Table 1). Finally, Lin’s concordance correlation coefficient (ccc), which describes the reproducibility between two evaluation methods, was calculated to determine the deviation of the acquired macro data from the perfect concordance (x = y) (Table 1, Fig 2) [37]. On a scale from 0.00 to 1.00, the macro scored ccc’s ≥0.84 for all NMJ features but bouton count. Bouton count resulted in a ccc of 0.22 (C.I.95% 0.10–0.32), which indicates that macro and manual performance are discordant. In summary, the macro assessed nine NMJ features, eight of which were successfully validated with high concordance correlations. We therefore mainly focused on these eight features in all subsequent analyses.
To further validate the macro, we tested the reproducibility of published findings on mutants with altered synaptic parameters for each of the three principal image segmentation procedures performed by our macro (NMJ outline, skeleton and active zones). We and others have shown that Ankyrin 2 (Ank2, CG42734) mutant [38,39] or knockdown [40] flies present with fused boutons and smaller NMJs. Here, we used panneuronal Ank2 knockdown NMJs as a positive control to validate the macro’s NMJ outline (Fig 3A and 3B). The NMJ area was significantly smaller upon Ank2 knockdown by two independent RNAi strains (Ank2-RNAiKK107238 339μm2, padj = 2.18E-08; Ank2-RNAiKK107369 361μm2, padj = 1.20E-05), compared to our genetic background control dataset (452 μm2). The NMJ perimeter was only significantly smaller for the stronger RNAi strain (control 289μm; Ank2-RNAiKK107238 238μm, padj = 1.82E-03). Highwire (hiw, CG32592) is a known regulator of NMJ length and the extent of branching and mutants typically present with long, highly branched NMJs [41]. Our macro reproduced the mutant phenotype in NMJs that have a panneuronal knockdown of Highwire, again by using two independent RNAi strains (Fig 3C and 3D). The NMJ skeleton-derived parameters length (Hiw-RNAiGD28163 197μm, padj = 3.10E-25; Hiw-RNAiGD36085 147μm, padj = 7.31E-07; control 122μm), longest branch length (Hiw-RNAiGD28163 154μm, padj = 2.02E-13; Hiw-RNAiGD36085 122μm, padj = 4.62E-04; control 106μm), number of branches (Hiw-RNAiGD28163 9.33, padj = 2.10E-04; Hiw-RNAiGD36085 7.69, padj = 2.52E-02; control 5.74) and number of branching points (Hiw-RNAiGD28163 3.13, padj = 6.74–04; Hiw-RNAiGD36085 2.73, padj = 3.31E-02; control 1.79) are all significantly higher (120–180%) compared to the genetic background control dataset. Lastly, the GTPase Rab3 is required for proper bruchpilot distribution and the mutant (rup) presents with a reduced number of Brp-positive active zones (81 compared to 298 in control NMJs on muscle 4) [42]. The macro reproduced this phenotype upon panneuronal knockdown (Rab3-RNAiKK100787 138 Brp-positive active zones; control 290 Brp-positive active zones; p = 4.43E-29) (Fig 3E and 3F).
The large collection of objectively quantified NMJ data offered the possibility to look at systematic differences in NMJ morphometry for gender, genetic background and body segment. For this purpose we restricted the dataset to images from which we obtained data for all nine features, including muscle measurements (due to the latter requirement an additional 62 NMJ images were excluded). We divided the dataset into male- (n = 724) and female-specific (n = 377) data and evaluated the differences between both sexes. We found that six features significantly differed from each other (padj < 0.05), with males showing lower average values than females: active zones (♂ 281; ♀ 303; padj = 1.51E-08), NMJ area (♂ 429μm2; ♀ 464μm2; padj = 8.43E-09), perimeter (♂ 289μm; ♀ 306μm; padj = 7.26E-06), NMJ total length (♂ 124μm; ♀ 130μm; padj = 1.65E-04), longest branch length (♂ 107μm; ♀ 114μm; padj = 8.30E-05) and muscle area (♂ 61377μm2; ♀ 66976μm2; padj = 1.98E-15) (Fig 4). In contrast, gender did not significantly impact the number of branches (♂ 5.5; ♀ 5.4; padj = 1.00), branching points (♂ 1.7; ♀ 1.7; padj = 1.00) and Islands (♂ 2.1; ♀ 2.1; padj = 1.00). Taken together, this suggests that the branching geometry is similar for both sexes, whereas size is not.
Next, we aimed to determine the influence that the genetic background might have on our NMJ features, focusing on two genetic backgrounds relevant for large scale reverse genetic screening. We divided our dataset, considering males only, into two genetic backgrounds, deriving from GD (n = 311) versus KK VDRC RNAi libraries (n = 413), and compared these between each other for each NMJ feature (Fig 5). Three of the nine features showed a significant difference between the two tested genetic backgrounds: active zones (GD 267; KK 292; padj = 2.21E-08), NMJ area (GD 396μm2; KK 453μm2; padj = 1.98E-15) and length (GD 121μm; KK 126μm; padj = 3.78E-02). No significant differences were observed for the other six features: longest branch length (GD 105μm; KK 109μm; padj = 1.32E-01, Islands (GD 2.1; KK 2.0; padj = 1.32E-01), branches (GD 5.7; KK 5.3; padj = 3.96E-01), branching points (GD 1.8; KK 1.7; padj = 8.38E-01), muscle area (GD 60765μm2; KK 61838μm2; padj = 3.29E-01) and NMJ perimeter (GD 288μm; KK 289μm; padj = 9.57E-01). This data shows that the genetic background can be a significant source of “variance” at the NMJ.
The literature reports data for abdominal body segments in the range of A2-A5, whereby studies report on evaluated NMJ data at one segment or a combination of different segments [19,43]. We aimed to quantitatively determine whether among these segments NMJs show considerable differences in one or several features. Consequently, we divided the dataset into four groups, each representing one segment. We did find differences among features across the 4 evaluated segments, following different spatial patterns. The number of active zones, branches and branching points showed a relative decrease from anterior to posterior (Fig 6A–6C). However, only segment A2 showed significant differences to (some of) the other segments (Table A in S2 Table). The number of islands followed the same pattern, but the values were not significantly different over the different segments (Fig 6D; Table A in S2 Table). The total length and longest branch length showed the opposite pattern; segment A2 NMJs were significantly shorter than NMJs of segments A3-A5 (Fig 6E and 6F; Table A in S2 Table). The muscle area of segment A2 was also significantly smaller compared to segment A3. The size peaks in segment A3 and significantly decreased in the segments A4 and A5. The muscle size of segment A5 was significantly smaller than that observed for A2 (Fig 6G; Table A in S2 Table). The NMJ area formed the fourth category that showed significant increase from segment A2 to A3 and a significant decrease to the same level from A4 to A5. The NMJ perimeter behaved very similar, although values were not significantly different from one another (Fig 6H and 6I; Table A in S2 Table). An overview of the number of cases, mean, confidence intervals en p-values is provided in S2A Table. Although not always significant, consistent patterns were observed in each group of gender and genetic background (Tables B-E in S2 Table). Taken together, the NMJ features could be subdivided in four groups with different patterns over the abdominal segments A2-A5: i) active zones, branches, branching points and islands, ii) length and longest branch length, iii) muscle area and iv) NMJ area and perimeter.
Finally, we used our morphometric dataset to determine which NMJ features might correlate with each other and which features appear comparatively independent, to reveal coordinated aspects of NMJ morphology. A pair wise correlation analysis was performed, in which the correlations of all possible feature pairs were determined and ordered accordingly (Fig 7A). As one might have expected, the strongest positive correlation was found between branches and branching points (R = 0.92), indicating that these features can almost predict each other. The other group of moderately-to-strongly correlating features included the size-related features NMJ area, perimeter, length and longest branch length (0.45<R<0.82). The number of active zones correlated to a lesser extent with this group (0.36<R<0.47). We only observed a weak correlation between the NMJ area and the muscle area (R = 0.35). Both the features muscle area and number of islands seemed to behave as independent features, lacking any moderate (0.4<R≤0.7) or strong (R>0.7) correlation with any of the other NMJ features.
We applied a principal component analysis (PCA), a statistical method to reduce the dimensionality of a dataset, and summarized our data in five different components. This aggregation was the most acceptable because it explained 91% of the variance of our data and classified each of the measured NMJ features on one of these components (Fig 7B; Table A in S3 Table). The size-related features NMJ area, perimeter, length and longest branch length constitute the first principal component, which explained 38.2% of the total variance. The features branches and branching points contributed most to the second principal component (21.7% of the total variance), thus the second component mainly accounted for NMJ geometry. The first two components explained almost 60% of all variance. The angles between the features contributing to the first versus the second principal components were around 90°, indicating that the variables NMJ area, perimeter, length and longest branch length hardly correlated with the variables branches and branching points. This is in agreement with the above reported correlation coefficients (Fig 7B and 7C). The features islands, muscle and active zones contributed most to the third (13.4%), fourth (10.4%) and fifth (7.7%) principal component, respectively. Based on these results, we defined five morphometric groups with a variety of mutual kinship: 1) NMJ size (NMJ area, perimeter, length and longest branch length), 2) geometry (branches and branching points), 3) islands, 4) muscle area and 5) number of active zones. Important to note is that the active zones also showed a moderate correlation and contribution to the NMJ-size features underlying most of the first component. This suggests that the number of active zones is at least partially coordinated with NMJ size. We obtained comparable results when applying PCA on datasets specific to one combination of gender and genetic background library (Tables A-E in S3 Table) or datasets specific for one abdominal body segment (Tables F-I in S3 Table).
In summary, our data showed that synaptic size varies the most within (natural) populations, followed by the branching geometry. It is remarkable that the size- and geometry-related features hardly correlated, suggesting that these features are differentially regulated during larval NMJ development.
Our macro was designed to cope with the challenges of high-throughput images with limited resolution and quality. However, to ensure wide applicability we also tested our macro on confocal images (S3 Fig). Following a similar strategy as above, manual and macro counts were compared between n = 15 NMJ confocal images, co-labeled for Dlg1 and Brp (Tables A-I in S4 Table). No significant differences and ccc scores ≥0.83 were found for the NMJ features NMJ area, perimeter, length, longest branch length, number of islands, number of branches, number of branching points and number of active zones when manual measurements were compared to the macro assessment (Table 2). We found, however, a significant difference between manual and macro bouton count (p = 0.01; ccc = 0.55), which indeed confirmed that the marker and not the technique is causing this difference. It generally applied that the better the quality of the image, the better the macro performed.
We further tested the applicability of our software to other synaptic markers. Horseradish peroxidase (Hrp) is a neuronal membrane marker, commonly used to stain NMJ presynaptic terminals. Visual inspection of macro-annotated images still revealed errors in bouton counting for boutons that lacked a discernible interbouton space. All other NMJ features were displayed correctly, as is shown in a representative NMJ image (Panel A in S2 Fig). Neither the pre- or postsynaptic marker tested (Dlg1 and Hrp) where suitable for bouton counting with our macro. Since the number of boutons is a frequently assessed parameter in studies of NMJ morphology, we further optimized the macro in order to reliably recognize and count the boutons using the synaptic markers Synaptotagmin (Syt) and Cysteine string protein (Csp), two presynaptic vesicle-associated proteins. Both proved to be very suitable markers to distinguish and count even closely positioned boutons, probably because of the complete lack of staining in interbouton regions (Panels B-C in S2 Fig). For appropriate segmentation of bouton numbers a second macro was created: Drosophila_NMJ_Bouton_Morphometrics. It is available via the same public figshare repository: https://figshare.com/s/ec634918c027f62f7f2a [36].
Drosophila_NMJ_Bouton_Morphometrics allows users to accurately count boutons on Syt or Csp immunostaining, co-labeled with Brp The working procedure is the same as described previously for Drosophila_NMJ_Morphometrics (S1 Protocol). It provides a result file where the NMJ features: number of boutons, NMJ bouton area, NMJ length, NMJ longest branch length, number of islands, number of branches, number of branching points and number of active zones are assessed. To prove the reliability of Drosophila_NMJ_Bouton_Morphometrics bouton counts, the same validation procedure as described previously was used to evaluate manual versus macro bouton counts in n = 26 NMJ confocal images, labeled with Syt (S5 Table and S4 Fig). No significant differences in number of boutons and a ccc score of 0.96 between manual and macro counting were found (Table 3).
Synapse morphology, shaped by synaptic transmission and regulating synaptic efficacy, is of central interest in neuroscience. However, it is still largely unknown how synapses adopt their overall shape. Most studies focused on one or few synaptic features rather than assessing synapse morphology more comprehensively and quantitatively. Here, we used the Drosophila larval NMJ, a widely used model for glutamatergic synapse biology and amenable to powerful genetic approaches, to pave the way for systematic synapse morphometry. We developed and released a Fiji-based macro to automatically and objectively evaluate nine morphological NMJ features. We estimate that the macro would save an experienced researcher up to 15 minutes per NMJ spent on manual image segmentation and analysis. By applying this method on a large number of muscle 4 glutamatergic type 1b NMJs, we quantified significant effects of gender, genetic background and abdominal body segment on multiple aspects of NMJ morphology. Correlation and PCA analyses demonstrated that the nine assessed morphological features can be grouped into five morphometric groups. The two groups that accounted for features involved in synaptic size and geometry contributed most to the first two principal components, which covered 60% of total variance. These two groups hardly correlated with each other. We propose that different molecular mechanisms control the two components, at least at the evaluated muscle 4 NMJ.
Our knowledge on molecules and mechanisms that shape synapses has greatly expanded in the last decade, and genetic screens in Drosophila have made an important contribution. Synapse morphology is a frequently used readout to discover genes required for proper synaptic function. Most studies have used (semi-)quantitative analyses, using e.g. the selection- and line-options of Fiji/ImageJ, but often only upon initial visual detection and performed by hand. Although this has proven to be sufficient to identify genes that if mutated grossly disrupt synapse development, this strategy likely has left more subtle modulators unidentified. Thus, the extent of the synapse regulome, including players that ensure proper orchestration of synapse coordinates, still awaits discovery. Their comprehensive identification needs a sensitive readout and a thorough understanding of how different morphological aspects relate to one another. Sutcliffe et al. created a publically available ImageJ plugin called “DeadEasy Synapse”, which measures the total voxel size of Brp-positive active zones per NMJ [44], and is thus complementary to our macro that instead achieves active zone counts and quantitative assessment of eight further morphometric NMJ features. In addition to Fiji/ImageJ, Cellprofiler is another open-source system for high-throughput image analysis [45]. It has been proven very useful for cell image analysis [46–48], but also for morphological phenotypes measured in C.elegans [49]. Cellprofiler lacks options to trace branch-like skeleton structures, required to measure NMJ length and branching pattern. In this study, we developed such a tool. We show that our Fiji-based macro is semi-automated, sensitive, and objective, whereas manual counting is laborious and can be assumed to be subject to interpersonal differences. The macro generates output files that allow the user to evaluate accuracy of the image segmentation and to correct or exclude (depending on the nature of the limiting feature) annotated images from further analysis. A low quality of the immunohistochemistry resulted in less correctly assessed NMJ images. Whenever the input quality was guaranteed (input checkpoints 1 and 2), we retained 84–92% of NMJ images, depending on the feature of interest. In this study where we took a specific interest in the correlations among all features, we exclusively used the NMJ images in which all features could be assessed with high accuracy. Staining variability is responsible for most of the excluded images. However, it similarly influences manual evaluation and can therefore not be linked to the macro performance.
Based on literature and our experimental observations, we aimed at quantifying nine morphological NMJ features. The macro performance was assessed for both wide field high-content and confocal muscle 4 NMJ images by investigating the deviation between manual and macro evaluations at three different levels: (1) sample distribution, (2) per given sample, and, most importantly, (3) for concordance. We deliberately chose the word ‘deviation’ over ‘error’ to underline that neither method can be considered as objectively true. Whereas successful for eight of these features, we were not able to optimize the bouton count in a satisfying manner for Dlg1, our marker of choice. The successful features were scored objectively and accurately, given the equal confidence interval widths and high concordance correlation coefficients (≥0.84, from which seven features even scored above 0.90). Whereas the macro measurements resulted in somewhat higher absolute values for length-related parameters, the high ccc scores demonstrate that this is a consistent proportional difference compared to the manual evaluation. Consequently, when both mutant and control samples are assessed by the same method, the difference (mutant:control) is equal for both methods. The macro counts somewhat higher absolute values, because it continuously thresholds between fore- and background, whereas manual measurements -in this case- are based on straight lines. The NMJ area is a two dimensional NMJ feature, a small difference in area evaluation therefore has a larger effect, explaining the somewhat lower ccc. This phenomenon is intrinsic to the nature of this parameter, as is also illustrated by a similar deviation between both manual experimentors. We conclude that our methodology shows accuracy and sensitivity comparable to manual evaluation for eight NMJ features.
Manual versus macro assessment of bouton number was not comparable, given the low concordance correlation coefficient of 0.22. We thus excluded the number of boutons from further analysis on this control dataset in this study. The macro uses a watershed transform -an algorithm that separates touching or slightly overlapping particles by identifying their local maxima in the distance function within these objects- on the segmented NMJ outline to distinguish individual boutons on NMJ outline invaginations, characteristic for interbouton regions [50]. We show that Dlg1 is not the optimal marker to determine bouton number, as this (mainly) postsynaptic marker presents with poorly pronounced interbouton constriction. We developed a second macro “Drosophila_NMJ_Bouton_Morphometrics” to assess the number of boutons using anti-Syt and anti-Csp, which successfully segmented boutons. We carefully validated the reliability of this second macro as done previously for our markers of interest (Dlg1 and Brp). Taken together, we developed 2 Fiji-based macros (Drosophila_NMJ_Morphometrics and Drosophila_NMJ_Bouton_Morphometrics) that perform objective and sensitive quantification of nine morphological NMJ features in a high-throughput manner.
Beyond the sensitive read-out that is required to detect subtle differences in synaptic morphology, a thorough understanding of natural variation and contributing factors is required to limit the calling of false positive phenotypes. We therefore quantified the effect of the main variables in our study -gender, genetic background and abdominal body segment- on the eight muscle 4 NMJ features acquired by our macro and the manually acquired muscle area.
Males and females showed considerable differences in size-related NMJ features and in the number of active zones. The female synaptic terminal was almost 5% longer and 8% bigger compared to males and it contained 7% more active zones. In agreement with the bigger size of female flies [51], the area of muscle 4 was 9% bigger in females than in males. Therefore, one might generate false positive results when comparing two datasets with each containing an uncontrolled amount of males-females. Interestingly, mutant screens or gene focused studies do not always report on gender selection or control, although gender can easily be selected for analyses [52]. Furthermore, gender selection is not mentioned in NMJ protocols that are often referred to by these studies [53–58]. Sex-specific differences at specific NMJ terminals were already described: Lnenicka, et al reported that muscle 5 produced larger excitatory postsynaptic potential in females and that type 1s motonerves on muscle 2 and 4 showed a greater charge transfer [59]. Interestingly, except for muscle size, no sex-specific difference in electrophysiology was detected for type 1b neurons on muscle 4, the synaptic terminal we focused on. Synapse morphology of 4-1s and 5-1b was measured, but no differences between males and females were found in this study for NMJ length, number of branches or number of boutons, which led the authors to suggest that the observed differences in transmitter release are due to ultra-structural and/or biochemical differences [59]. Our highly sensitive morphometry on 4-1b uncovered sex-specific NMJ properties that can potentially underlie the reported physiological differences. To our knowledge, no sex-specific regulators of NMJ morphology have been described yet.
Many of the NMJ-size related features were also significantly different between the two isogenic host strains (GD and KK RNAi libraries) we tested. Unquestionably, genetic differences can influence larval growth [60], but since no significant difference was observed in muscle 4 size, we did not find indication of overall differences in animal growth in the two investigated genotypes. A traceable difference between both strains is the yellow+ marker in a yellow mutant background carried by the KK host strain. No NMJ abnormalities have been reported in yellow mutants, but yellow mutant alleles affect male courtship and mating behavior [61], suggesting a potential role in or effect on the nervous system.
Finally, we also demonstrated a significant impact of the abdominal body segments on muscle 4 NMJ morphometry. The observed patterns per morphological feature could be divided into four categories, with the greatest variance in geometry-related features and active zones. In general, muscle 4 type 1b synapses were shorter, more branched and have the highest number of active zones anteriorly.
In summary, all three tested variables in our NMJ analysis—gender, genetic background and abdominal segment- had a significant effect on at least some of the assessed morphological NMJ features on muscle 4. For quantitative evaluations, if aiming at high sensitivity, it is therefore important to take these features into account.
So far, Drosophila NMJ studies were often focused on a particular aspect of NMJ morphology, rather than assessing morphology more comprehensively. Consequently, the interdependencies of morphological features at this synapse, and at others, remained unknown. Carefully evaluating these relationships in unperturbed conditions, we here shed light onto which features are to what extent correlated, and thus provide first insights into the system properties of this important model synapse.
We identified five morphometric groups based on a pair wise correlation and principal component analysis. Interestingly, the features underlying the two groups explaining most of the variance hardly correlated, which led us to speculate that different biological processes underlie NMJ size and geometry. In agreement, both groups behaved differently between sexes and over the abdominal segments. Whereas NMJ geometry showed a decreasing number of branches and branching points from anterior to posterior, size features seemed to increase from abdominal segment 2 to 3.
Surprisingly, the muscle 4 area only correlated to a minor extent with NMJ size (R = 0.349) and very weakly with the number of active zone (R = 0.160). Although a strong correlation was observed between bouton number and muscle 6/7 size during embryonic and larval growth [62], the features seem to be less correlated within third instar larval stage. Whereas we cannot exclude differences between different NMJs/muscles, it seems more likely that the earlier observed correlation over a developmental time period reflect a general trend in growth to which all underlying mechanisms are subject to, rather than a tight causal relationship. Our result is in agreement with an earlier observation in which muscle size only partially (~50%) explained the variation in bouton number when comparing different Drosophila species [60]. Our finding provides an argument not to normalize synaptic size by muscle size, as has been practiced by some studies in third instar Drosophila larvae. These conclusions are applicable to NMJ morphological analyses of muscle 4 with the two markers of choice.
Although we validated our macro and carefully evaluated its performance in each NMJ image, our findings are still subject to technical variation such as variations in specimen dissections, immunolabeling intensities and the genome constitution. However, we have shown that the macro does not produce more variance than a manual counter does and the consistent correlations among gender and genetic background subgroups show the reproducibility of our data and support our conclusions. In our study, we evaluated three natural sources of variation to study the relation between nine synapse parameters: sex, abdominal segment identify and two selected genetic backgrounds. Additional sources causing NMJ structural variation have been demonstrated, such as larval and induced synapse activity [16,17]. We raised larvae under controlled conditions and therefore do not expect a significant variation in our data linked to such mechanisms.
In summary, we developed a sensitive, accurate and semi-automated Fiji-based macro that permits high-throughput systems morphometry on the well-studied Drosophila larval NMJ on muscle 4. This method has the ability to handle several commonly used NMJ markers and microscope techniques. Here, we used a systems biology approach on two comprehensively generated, multiparametric datasets to start to understand how different morphological aspects of the synapse are coordinated. We showed how the nine measured morphometric features relate to one another and defined five morphometric groups in which features showed higher intra- than inter-correlation (Fig 8), suggesting that different molecular mechanisms are at work. The macro that we have developed and the here reported results have shed first light onto the design principles of an important model synapse, paved the way to quantitative synapse morphometry and can be applied to identify genes that couple features within a one morphometric group or that orchestrate the relations between different morphometric groups.
Fly stocks were maintained using standard Drosophila diet (sugar/cornmeal/yeast). Virgins from a w1118; UAS-Dicer-2; elav-Gal4 promotor line were crossed to males from either w1118 (VDRC stock 60000 –genetic background of the GD library) or y,w1118;P{attP,y[+],w[3`] (VDRC stock 60100 –genetic background of the KK library) and maintained at 28°C, 60% humidity. Crosses were performed with consistent amounts of flies and food. Positive control RNAi strains were in a similar manner crossed to virgins of the promotor line. The following RNAi strains were obtained at the Vienna Drosophila research center: Ankyrin2 (FBgn0261788; CG42734) P[KK104937]VIE-260B (VDRC stock KK107238) and P[KK106729]VIE-260B (VDRC stock KK107369); highwire (FBgn0030600; CG32592) w1118; P[GD14101]v28163 (VDRC stock GD28163) and w1118; P[GD14104]v36085 (VDRC stock GD36085) and Rab3 (FBgn0005586; CG7576) P{KK108633}VIE-260B (VDRC stock KK100787). Panneuronally induced knockdown conditions were compared to progenies from the driver crossed to the respective library host strain (GD60000 or KK60100).
Wandering L3 larvae were labeled for their gender and dissected, fixed in 3.7% paraformaldehyde for 30 min and co-labeled for bruchpilot (Brp) and discs large 1 (Dlg1). Brp was revealed using the primary antibody nc82 (1:125) (Developmental Studies Hybridoma Bank) applied overnight at 4°C and the secondary Alexa 488-labeled goat-anti-mouse antibody (1:125) (Invitrogen Molecular Probes). Discs large was visualized using primary antibody anti-Dlg1 (1:25) (Developmental Studies Hybridoma Bank) conjugated with the Zenon Alexa Fluor 568 Mouse IgG1 labeling kit (Invitrogen), applied according to the manufacturer’s protocol. For Hrp, Syt and Csp labeling, larvae were blocked for 1.5h on 5% NGS-PBS-T (0.3% Triton X-100 in PBS). Primary antibody anti-Hrp (rabbit, 1:750) (Jackson ImmunoResearch), anti-Syt (rabbit, 1:100) (kindly provided by H.Bellen) or anti-Csp (mouse, 1:25) was applied overnight at 4°C, followed by the Alexa 568- or 488-labeled secondary antibodies (1:500) (Invitrogen Molecular Probes).
NMJ images were obtained of type 1b NMJs at muscle 4 using an automated Leica DMI6000B high-content microscope. Individual NMJs were imaged at 10x (snapshot; Dlg1 only; 1.096 pixels/μm) and 63x magnification (stack; both channels; 6.932 pixels/μm). A fixed stack size was used, comprising 42 images per channel with a z-step size of 0.3μm and a z-volume of 12.152μm. The 2x42 images were saved as separate tiff files, encoding the NMJ number, z-plane and channel number in the file name. The area of muscle 4 was manually assessed via the segmented line option in Fiji at the lower magnification.
Confocal NMJ images were obtained of type 1b NMJs at muscle 4 using the Olympus FV1000 microscope. Individual NMJs were imaged at 60x (stack; both channels; 4.83 pixels/μm) with a z-step size of 0.91μm and a z-volume adjusted to the depth of the NMJ.
The macro, written in ImageJ macro language, is compatible with the open source Fiji platform [32]. The entire analysis procedure consists of three steps, for which three separate sub-macros were written. This setup was chosen to allow maximum flexibility in the workflow. Sub macros can be executed from the main macro through a graphical user interface (GUI).
The first sub macro “Convert to stack” traverses a directory structure selected by the user in the GUI. Detected unprocessed images belonging to the same NMJ are recognized based on their z-plane and channel number and subsequently converted and saved to a hyperstack (containing all image data in a single tiff file) and maximum-intensity based Z-projection (referred to as ‘flat stack’). In our setup, the macro takes as input two channel z-stacks where the individual z-planes are stored as separate tiff files. The macro can however be adapted to deal with other types of input images.
The second part of the macro (sub macro 2 –“Define ROI”) is a semi-automated step where flat stacks are detected and opened automatically in a consecutive manner. Only images are opened that have not been processed by sub macro 2 before. In every flat stack, the region of interest (ROI) is manually defined by the user using the free hand selection tool. A binary ROI image is created and stored in the image source directory.
The third sub macro (“Analyze”) identifies hyperstacks for which the ROI image is present. NMJ image analysis is performed throughout each stack, within the limits of the image-specific ROI. Macro-annotated images are stored, containing a delineation of the analyzed NMJ. Additionally, a text file containing nine quantified features per NMJ is stored in the main directory.
To enable high-throughput image acquisition and analysis, we used wide field fluorescence imaging to develop the macro. Out-of-focus fluorescence around the NMJ terminal, inherently present in these images, necessitates pre-processing to distinguish foreground signal from background noise. Therefore in the first step of the macro images are filtered applying a rolling ball background subtraction algorithm with a radius of 20 pixels. This algorithm is considered effective and fast for suppression of a non-uniform background with objects of rather constant diameter [63]. The outline of the entire NMJ is defined by an auto-threshold selection based on Renyi’s Entropy algorithm, applied to the Dlg1 or Hrp staining. This algorithm was shown to outperform several other entropy-based threshold selection methods [64], and resulted in consistent and adequate segmentation on a series of test images in the present study. Constriction of the synaptic terminal between boutons provided the basis for the analysis of bouton counting’s. A watershed separation is performed on the binary NMJ outline. Resulting objects exceeding an (empirically determined lower bound) area threshold of 100 pixels are considered to represent boutons. To filter against background noise as for example present in Hrp staining, an optional filter (“Remove small particles”) was implemented to remove particles smaller than 100 pixels.
To measure NMJ length and branching geometry, a binary skeleton for the NMJ is determined. The skeleton is a one-pixel thick axis along the center of the NMJ, calculated using mathematical morphology on the binary image. We found that the auto-threshold described above, used to accurately determine the NMJ outline, and was sometimes too restricted for accurate determination of the skeleton when using Dlg1, Syt or Csp staining. The macro therefore uses auto-threshold selection based on Li’s Minimum Cross Entropy for this purpose. This algorithm generally results in somewhat wider segmentation results, as previously witnessed by the results of Sengur et al. [65]. The Renyi’s Entropy algorithm was used for NMJs stained with Hrp. From the NMJ skeleton five features (length, longest branch length, number of branches, branching points and islands) are calculated.
Subsequently, the number of active zones is counted in the Brp-channel by finding local intensity maxima in the 3D image stack. To reduce the effect of intensity variations over individual active zones, stacks are first filtered applying a 3D grey closing with a small circular structuring element. Identified local maxima are considered to represent one active zone if they do not touch other local maxima (either horizontally/vertically or diagonally) and exceed a minimum intensity level (automatically determined using Huang's fuzzy thresholding method), to prevent background fluctuations to be counted as active zones.
Confocal NMJ images were processed in a similar manner, with maxima noise tolerance ‘100’ and Brp-puncta lower threshold ‘250’.
The nine NMJ features measured by the macro were in parallel manually quantified, blind to the macro results, independently by two experimenters in 30 NMJ images. Images were processed with Fiji. For each channel a projection was created from in-focus planes. The subtract background algorithm was applied to the Dlg1 channel, followed by 3 consecutive applications of the standard FIJI smooth filter (3x3 average filter), and area and perimeter were determined by manually thresholding the NMJ terminals. All length related NMJ features were measured using the freehand lines tool. Active zones were visually assessed in the Brp channel. Macro counts were plotted against averaged manual counts, and Lin’s concordance correlation coefficients was calculated in R, using the epi.ccc function of the epiR package [66]. The % deviation between manual and macro count is calculated as (average manual result—macro result) / average manual result x 100%. Sensitivity is true positives / (true positives + false positives), specificity is true positives / (true positives + false negatives). Active zone results are compared to experimentor #1.
Confocal NMJ images (n = 15) were validated in a similar manner by one experimenter. Picture one was excluded from Brp-analysis because of low staining quality.
Statistics were performed in R (R Development Core Team, 2008) [67]. For comparisons between manual and macro counts, gender and genetic backgrounds, independent 2-group t-tests were applied for normally distributed features (area, active zones, boutons, length, longest branch length, perimeter and muscle) and Mann-Whitney U tested for not normally distributed features (branches, branching points and islands). Whenever required, p-values were adjusted by a Holm-Bonferroni correction and indicated as padj. Anova-Tukey method was used for body segment analysis followed by Tukey’s honest significance test (Tukey’s HSD). Pearson correlations were calculated for the different feature combinations and visualized in an adjusted scatter plot matrix [68]. Principal component analysis [69] was used to study the relationship among different aspects of synapse morphology.
Drosophila_NMJ_Bouton_Morphometrics was specifically developed to assess the number of NMJ boutons of Syt- or CSP- immunostained NMJs. The macro processing and macro analysis follow the same steps as described previously in this section, except that the outline of the entire NMJ is defined by an auto-threshold selection based on the algorithm moments and a dilating step prior to the watershed separation. These increase efficiency of bouton segmentation. The algorithm area in this macro assesses bouton area after the bouton segmentation, and the algorith NMJ perimeter is obsolete and has been removed. Resulting objects exceeding an (empirically determined lower bound) area threshold of 10 pixels are considered to represent boutons. The filter against background noise (“Remove small particles”) should always be activated when running this macro and was implemented to remove particles smaller than 10 pixels.
The validation of bouton counts for the Drosophila_NMJ_Bouton_Morphometrics macro was performed using confocal images of NMJs (n = 26) immunolabeled with anti-Syt antibody. The number of boutons were counted by two experimenters blind to the results of the macro, and where compared with the macro counts using the same procedures as described in macro validation. Two picture where excluded from Syt-analysis because of low staining quality.
|
10.1371/journal.pgen.1007169 | Mutations in THAP1/DYT6 reveal that diverse dystonia genes disrupt similar neuronal pathways and functions | Dystonia is characterized by involuntary muscle contractions. Its many forms are genetically, phenotypically and etiologically diverse and it is unknown whether their pathogenesis converges on shared pathways. Mutations in THAP1 [THAP (Thanatos-associated protein) domain containing, apoptosis associated protein 1], a ubiquitously expressed transcription factor with DNA binding and protein-interaction domains, cause dystonia, DYT6. There is a unique, neuronal 50-kDa Thap1-like immunoreactive species, and Thap1 levels are auto-regulated on the mRNA level. However, THAP1 downstream targets in neurons, and the mechanism via which it causes dystonia are largely unknown. We used RNA-Seq to assay the in vivo effect of a heterozygote Thap1 C54Y or ΔExon2 allele on the gene transcription signatures in neonatal mouse striatum and cerebellum. Enriched pathways and gene ontology terms include eIF2α Signaling, Mitochondrial Dysfunction, Neuron Projection Development, Axonal Guidance Signaling, and Synaptic LongTerm Depression, which are dysregulated in a genotype and tissue-dependent manner. Electrophysiological and neurite outgrowth assays were consistent with those enrichments, and the plasticity defects were partially corrected by salubrinal. Notably, several of these pathways were recently implicated in other forms of inherited dystonia, including DYT1. We conclude that dysfunction of these pathways may represent a point of convergence in the pathophysiology of several forms of inherited dystonia.
| Dystonia is a brain disorder that causes disabling involuntary muscle contractions and abnormal postures. Mutations in THAP1, a zinc-finger transcription factor, cause DYT6, but its neuronal targets and functions are unknown. In this study, we sought to determine the effects of Thap1C54Y and ΔExon2 alleles on the gene transcription signatures at postnatal day 1 (P1) in the mouse striatum and cerebellum in order to correlate function with specific genes or pathways. Our unbiased transcriptomics approach showed that Thap1 mutants revealed multiple signaling pathways involved in neuronal plasticity, axonal guidance, and oxidative stress response, which are also present in other forms of dystonia, particularly DYT1. We conclude that dysfunction of these pathways may represent a point of convergence on the pathogenesis of unrelated forms of inherited dystonia.
| Dystonia is a brain disorder that causes disabling involuntary muscle contractions and abnormal postures. When this is the only feature, it is termed isolated dystonia. The pathogenic molecular mechanisms underlying the neuronal dysfunction that leads to dystonia remain to be elucidated and current treatments are unsatisfactory. The advent of next generation sequencing is rapidly expanding the list of genes causing isolated dystonia, including dominant mutations in THAP1 (DYT6), TOR1A (DYT1), GNAL (DYT25), ANO3 (DYT24), CIZ1 (DYT23) and TUBB4A (DYT4), and recessive mutations in HPCA (DYT2), COL6A3 (DYT27) and PRKRA (DYT16) [1–6], although some of these are still pending confirmation. Apart from rare inherited defects in dopamine synthesis [7], there is no known biological pathway that causally links genetic forms of dystonia. Phenotypic similarities between some inherited dystonias, including the most common DYT1 and DYT6, may suggest a shared underlying pathogenic mechanism. Uncovering such mechanisms would be a significant milestone, and potentially widely applicable for therapeutic development.
DYT6 is caused by mutations in THAP1 [Thanatos-associated (THAP) domain-containing apoptosis-associated protein] [8,9], encoding a ubiquitously expressed transcription factor [10,11]. Similar to DYT1, caused by a mutation in TOR1A encoding TorsinA, the disease phenotype is restricted to the central nervous system despite widespread expression of the mutated protein. Disease-causing mutations in THAP1 are dispersed throughout the coding regions, but most are missense and located in the DNA-binding domain (DBD) [12]. THAP domain DNA-binding activity is zinc-dependent, and the four metal-coordinating residues of the C2CH module are crucial for functional activity [13]. Nonsense mutations, equivalent to a null allele, likely result in the generation of small mRNA species that are subject to nonsense-mediated decay [8]. Little is known about the biological function of THAP1, particularly in neurons, although there is a neuron-specific DNA-binding THAP1-like-immunoreactive species [10]. There is also an alternatively spliced form lacking Exon2 which functionally does not substitute for the full-length protein [14] and is normally present at very low levels in the brain [15].
DYT6 is inherited in an autosomal dominant manner with reduced penetrance. Few brains from DYT6 patients have been examined and, to date, they do not exhibit any characteristic neuropathologic lesions [16]. Structural and functional neuroimaging in DYT6 manifesting and non-manifesting carriers (NMCs) demonstrates abnormalities in cerebello-thalamo-cortical and cortico-striato-pallido-thalamo-cortical pathways [17]. Genetically engineered mice with heterozygote Thap1 mutations, either C54Y or ΔExon2, display structural abnormalities of the deep cerebellar nuclei and deficits on motor tasks without overt dystonia [18]. Both mutations are early embryonic lethal when homozygote [18], and in mouse embryonic stem cells, lead to decreased viability and neuroectodermal differentiation [14]. The cell cycle is the major dysregulated pathway that emerged from microarray assays of HUVECs with up- or down-regulation of THAP1 and of human lymphoblasts harboring a disease-associated intronic variant of THAP1 [19,20], but was not enriched in ES cells [14].
To study the role of Thap1 in brain, we performed unbiased transcriptomic, RNA-Seq profiling of postnatal day 1 striatal and cerebellar tissue in two genetic mouse models of THAP1/DYT6 harboring mutations that alter the DNA binding domain, either (1) Thap1C54Y, a constitutive knock in (KI) of the C54Y causative mutation in the DNA binding domain (DBD) of THAP1 and (2) Thap1-, a constitutive knockout (KO) of exon 2 (ΔExon2) [18]. This was followed by functional studies to validate dysregulated molecular pathways with a focus on those that were either “top hits” and/or overlapped with other dystonias. We, and others, [21] have found that molecular abnormalities are present in the developmental stage, but their relationship to the emergence and persistence of the phenotype remains to be determined.
Thap1C54Y /+ KI mice carry a point mutation in one of three cysteine residues that are part of the zinc binding motif [19], altering binding of THAP1 to DNA [8,22]. Thap1 KO mice, i.e. ΔExon2, deletes part of the DBD, and is referred to herein as Thap1+/- [18]. We performed RNA-Seq analysis of cerebellar and striatal tissue dissected from postnatal day 1 (P1) Thap1+/-(ΔExon2), Thap1C54Y/+ and wild-type (WT) controls (N = 4/genotype, all males). At this age, neurons outnumber glia and the limited studies in Ruiz et al. [18] showed greater changes in mRNA levels than in the adult. There was a higher number of differentially expressed genes (DEGs) in Thap1+/- than in Thap1C54Y/+ in both structures (Fig 1 and S1–S4 Tables). 55 striatal DEGs overlapped between genotypes (Fig 1C, S5 Table) and 35 overlapped in the cerebellum (Fig 1F, S5 Table). The highest ranked DAVID Gene Ontology (GO) terms for the striatal overlapping DEGs were positive regulation of signal transduction, proteasome-mediated ubiquitin-dependent protein catabolic process and lipid storage, while those for overlapping DEGs in the cerebellum included cellular response to amino acid stimulus, and DNA-templated transcription. Among the overlapping DEGs, Cdh4 and Phf13 (up-regulated in striatum), Wibg and Rsph1 (down-regulated in striatum), Ppan (differentially regulated in cerebellum), and Nid2 (down-regulated in cerebellum) contain presumptive THABS motifs (S5 Table), supporting the notion that Thap1 may act as either an activator or repressor [13].
To determine if any of the DEGs are linked to dystonia or related disorders, we cross-matched them with the Emory University genetic dystonia panel (http://geneticslab.emory.edu/tests/MM550) and with the Mount Sinai Genetic Testing Laboratory Movement Disorders and Neuromuscular Disease Panel (S6 Table). We found 59 cross-matched striatal Thap1+/- DEGs and 5 Thap1C54Y/+ DEGs. Cryab and Fus appeared in both sets. In cerebellum, we identified 54 cross-matched DEGs in the Thap1+/- and 7 in Thap1C54Y/+, including 3 overlapping genes in both genotypes, Thap1, Lama1 and Sacs.
As the RNA for this study was derived from whole tissue, we utilized an RNA-sequencing transcriptome and splicing database annotating glia, neurons, and vascular cells from the mouse brain [23], in order to investigate the cell subtype origin of the DEGs in the striatum and cerebellum of the Thap1+/- or Thap1C54Y/+ vs WT mice (S7 Table). The largest cell-type fraction of both the up- and down- regulated genes in the Thap1+/- striatum were neuronal, while the greatest fraction of the up- and down- regulated genes in the cerebellum were expressed in endothelial and oligodendrocyte progenitor cells, respectively. The largest cell-type fraction of both the up- and down- regulated genes in the Thap1C54Y/+ striatum were neuronal and astrocytic in origin, while most of the up- and down- regulated genes in the cerebellum were derived from endothelial and microglial cells (S7 Table). It should be noted that P1 mice will have a greater enrichment for cell growth and gene regulatory pathways. Zhang et al. [23] showed that oligodendrocyte-lineage cell isolation does not occur until P17, the earliest time point when the full collection of oligodendrocyte-lineage cells is present. Therefore, one potential limitation of this study is age at which the transcriptomes were assayed but it is also a strength in terms of looking at altered pathways during this critical period.
Ingenuity Pathway Analysis (IPA) and DAVID GO terms identified biological functions and pathways enriched in DEGs from each dataset. The highest ranked IPA canonical pathways and GO Terms for each set of DEGs are shown in Fig 2 and S1 and S2 Tables. There were marked genotype-dependent differences, but there were overlaps between striatum and cerebellum within each genotype. Based on what were either the most significantly enriched pathways and terms, and/or those which are related to identified abnormalities in DYT6 and DYT1 models, we chose to functionally explore the eIF2α pathway, neuron projection development, synaptic plasticity [long term depression (LTD) and potentiation (LTP)], and mitochondrial Complex I.
To determine if any of the DEGs from the RNA-Seq datasets were bound directly by THAP1, we compared them against two publicly available THAP1 ChIP-Seq datasets, one from human ENCODE K562 cells, and the other from mouse ES cells [14]. The Thap1+/- cerebellum had the greatest number of overlapping genes when compared to both ChIP-Seq datasets, with a total of 32 overlapping genes when compared to the mouse ES cell data and 39 overlapping genes when compared to the K562 dataset (S11 Table). The highest ranking biological functions enriched in the 32-member gene set as identified by DAVID GO are: cellular macromolecules metabolic process, ribosomal small subunit assembly, and negative regulation of protein kinase activity (S11 Table). The highest ranking biological functions in the 39-member gene set as identified by DAVID GO are: cellular process, cellular metabolic process and gene expression (S11 Table).
The eIF2α signaling pathway was one of the top differentially regulated signaling pathways in the striatum and cerebellum of Thap1+/- mice. The eIF2α pathway is a key effector of the cellular response to several stressors, including the accumulation of misfolded proteins in the endoplasmic reticulum (ER), and was linked to TorsinA function soon after TOR1A was identified as the causative gene in DYT1 [24–26]. DYT16 is caused by mutations in PRKRA, a stress-activated modulator of the eIF2α kinase PKR, with evidence of abnormal phosphorylation of PKR and eIF2α in patient fibroblasts under ER stress [27,28]. In addition, a proteomics-based study identified abnormal eIF2α pathway activation in DYT1 mouse and rat brains, which correlated with assays in human brains [24]. Therefore, given the previous evidence suggesting a role of eIF2a signaling dysregulation in dystonia, and our own RNA-seq data, we investigated UPR genes and proteins to assay the baseline status of the UPR in Thap1 recombinant mice.
Initially, we assayed the relative mRNA levels for members of the eIF2α signaling pathway in P1 Thap1+/- cerebellum and striatum by RT-qPCR. We assessed changes in genes known to play a role in UPR or eIF2α signaling pathways using DEGs found to be dysregulated directly from the RNA-seq analysis (eIF3K, eIF2a, eIF4A, eIF4B), as well as upstream and downstream mediators of eIF2α (BiP, Chop, Rsp6, XBP1s, and total XBP1) in Fig 3A. There was dysregulated expression of most components of this signaling pathway in cerebellum (S9 Table), and in the Thap1+/- striatum significant differences were observed in Atf4 expression, and XBP1s/total XBP1 ratios (Fig 3, S9 Table). These data show that there are baseline abnormalities of the eIF2α signaling pathway; however, their contribution to the genotype-dependent phenotypes remains to be determined.
The eIF2α signaling pathway is involved in translational regulation, and notably, the DEGs were enriched for a second translational control pathway mediated by mTOR. These kinase cascades regulate protein function via phosphorylation and protein levels. We assayed components of both pathways at basal level by western blotting of protein lysates derived from P1 Thap1+/- striatum and cerebellum. Consistent with the RNA-Seq and RT-qPCR data, the key effector of the eIF2α pathway, ATF4, was reduced by 25% in the striatum (Fig 3, right panel). Overall, the data lend further support to the presence of dysfunction of the eIF2α pathway in DYT6 brain. Notably, screening designed to identify genes implicated in the response to ER stress in human B cells via genetic interactions identified THAP1 as the top hit and all changes were at the level of protein interactions [29].
To assay the function of the eIF2α pathway in the presence of a Thap1 mutation, we challenged P4 Thap1+/- and WT pups with tunicamycin, which induces ER stress and the unfolded protein response (UPR) in liver, cerebral cortex, and cerebellum at this age [30]. There was a clear engagement of the main ER chaperone, BiP, in striatum of Thap1+/- mice, and cerebellum of both genotypes (Fig 4). We detected genotype-dependent differences in the expression levels of ATF4 at basal level (dextrose-only controls) in the striatum of Thap1+/- mice as compared WT mice, and the most notable genotype-dependent difference following challenge with tunicamycin was a decrease in ATF4 striatal expression in Thap1+/- but not in WT mice (Fig 4). No differences were observed amongst the different groups when we assessed p-eIF2α/eIF2α and p-eIF2α/GAPDH expression levels (S3 Fig). Therefore, we could not find consistent up or down changes in the UPR with Thap1 baseline status and tunicamycin treatments. Nonetheless, our data suggest a dysregulation of the eIF2α signaling pathway. Therefore, we went on to perform physiologic/functional experiments in Figs 5–7.
IPA pathways related to oxidative stress (i.e. Mitochondrial Dysfunction and Oxidative Phosphorylation) were significantly dysregulated in Thap1+/- striatum and cerebellum, and abnormalities in these pathways can contribute to the UPR and to synaptic plasticity [31,32]. Specifically, mitochondrial complex I deficiency (OMIM:252010_3) was enriched (FDR < 0.05) in Thap1+/- cerebellum using Harmonizome [33]. However, we found no genotype-dependent differences in complex I activity in either the striatum (t = 0.89, df = 5, p = 0.42) or cerebellum (t = 0.36, df = 5, p = 0.74).
Synaptic plasticity is a specific neuronal function predicted by IPA and GO to be disrupted in the striatum of Thap1+/- mice, particularly long-term depression (LTD) and the related phenomenon of synaptic depotentiation (S1 Table). This association is again reminiscent of the deficit in striatal LTD in mouse models of DYT1 [34,35], suggesting that this might be a convergent feature among different types of dystonia and might even be related to dysfunction of the eIF2α pathway [26]. Although differential gene expression for Thap1C54Y/+ striatum was less predictive of a synaptic plasticity phenotype (S2 Table), we tested whether persistent synaptically-induced forms of plasticity at glutamatergic synapses–both LTD and long-term potentiation (LTP)–were altered in the striatum of both mouse lines. Extracellular recordings in acute slices from Thap1+/- mice stimulated with high-frequency synaptic stimulation (HFS) revealed greatly reduced LTD, similar to that reported for DYT1 mouse models (Fig 5A1, S9 Table). In contrast, LTP was normal in Thap1+/- striatum (Fig 5A2). The plasticity phenotype was different for the Thap1C54Y/+ mice: LTD showed a non-significant trend towards enhancement (Fig 5B1, S9 Table), but LTP was reduced (Fig 5B2, S9 Table).
The potential for plasticity at inhibitory synapses could distort the interpretation of our striatal field recordings. For example, reduced LTD at excitatory synapses is also consistent with increased GABA-A mediated currents. To address this possibility, we repeated the LTD experiment in the presence of the GABA-A antagonist gabazine (S2 Fig), and confirmed that LTD was reduced in slices from Thap1+/- mice. Moreover, the Thap1C54Y/+ mutation led to significantly enhanced LTD, in agreement with the trend observed in the absence of gabazine (Fig 5B1, S9 Table).
These results indicate that synaptic plasticity is susceptible to the deletion of Thap1 Exon2 or expression of a DYT6-related mutation of Thap1, while the specific nature of the disruption differs between these two manipulations. Note that paired-pulse ratio was not affected in either the Thap1+/- or Thap1C54Y/+ striatum (Fig 5C, S9 Table), indicating that presynaptic function was normal. However, basal synaptic efficiency was enhanced in Thap1C54Y/+ mice [stimulus strength to evoke 1 mV field excitatory postsynaptic potential (EPSP): 0.88 ± 0.11 μA (n = 24) for wildtype, 0.87 ± 0.10 μA (n = 21) for Thap1+/-, and 0.49 ± 0.05 μA (n = 23) for Thap1C54Y/+; S9 Table]. The increased efficiency in Thap1C54Y/+ striatum would be consistent with elevated synapse number. Alternatively, postsynaptic function could be up-regulated in Thap1C54Y/+ mice. Since striatal LTP at excitatory synapses is expressed postsynaptically [36], basal upregulation of postsynaptic function might limit the extent to which these synapses can be further potentiated, consistent with the reduced LTP seen in these mice.
These extracellular recordings reveal plasticity phenotypes that must be expressed by a substantial fraction of MSNs. Future patch-clamp experiments, performed on MSNs expressing either the D1 or D2 receptor, will be useful to determine if the Thap1+/- and Thap1C54Y/+ mutations differentially affect LTD or LTP in MSN subtypes.
Normal regulation of eIF2α by phosphorylation is required for multiple forms of LTD. In the striatum, inhibition of the eIF2α kinase PERK prevents synaptically-induced LTD [26], and eIF2α has been implicated in hippocampal LTD induced by pharmacological stimulation of metabotropic glutamate receptors (mGluR-LTD) [37,38]. To test whether the LTD deficits in Thap1+/- striatum could be due to dysregulation of eIF2α, we evaluated if abnormal LTD were rescued by Sal003, a selective inhibitor of the eIF2α phosphatase (Fig 6). We found differential effects of Sal003 on mGluR-LTD (induced by the group 1 agonist DHPG) and synaptically-induced LTD (induced by high-frequency stimulation; HFS). Sal003 restored mGluR-LTD to wildtype levels (Fig 6A, S9 Table), in agreement with the finding that eIF2α phosphorylation is required for this form of plasticity in the hippocampus [37]. However, Sal003 had no effect on HFS-induced LTD (Fig 6B, S9 Table). This finding contrasts with the ability of Sal003 to restore LTD in DYT1 mutant mice [26], indicating that Thap1 deletion can interfere with LTD by affecting signaling mechanisms in addition to eIF2α dysregulation. Since HFS-induced LTD in striatum depends on the synaptic activation of mGluRs [39,40], it is possible that, independent of its effect on eIF2α signaling,Thap1 deletion interferes with the ability of HFS to activate postsynaptic mGluRs. In fact, we identified numerous dysregulated genes that participate in synaptic transmission and conduction of nerve impulses in Thap1+/- striatum (S1 Table).
Neurite development was another enriched pathway identified across regions in Thap1+/- mice. TorsinA has a proposed role in neuritogenesis, and diffusion tensor imaging abnormalities detected by MRI are seen in different brain regions of patients with DYT6, DYT1 and other dystonias [41–43]. Signaling by the Rho family of GTPases is the top pathway in the Thap1C54Y/+ striatum, and this pathway plays a critical role in neuritogenesis and axonal pathfinding [44]. To investigate the influence of Thap1 on this process, we assayed neurite development in striatal MSNs in vitro. We cultured striatal E16 (embryonic day 16) neurons from individual Thap1+/- and WT embryos and quantified neurite length after 24 hours. Thap1+/- striatal neurons exhibited shorter processes as compared to WT controls. The phenotype was present but less severe in Thap1C54Y/+ striatal neurons (Fig 7, S9 Table). The total number of cells was equal between the genotypes, suggesting normal survival of the heterozygote mutant neurons following plating at equal densities.
We used an unbiased RNA-Seq approach to identify dysregulated genes and pathways in mice harboring either a Thap1 C54Y (disease causing) or a ΔExon2, i.e. null allele. A major goal was to determine if these pathways and resultant phenotypes overlap with abnormalities observed in other dystonias, particularly DYT1 [45], due to their clinical, electrophysiological, structural and functional neuroimaging similarities. RNA-Seq was performed in two brain regions, striatum and cerebellum, consistently implicated in dystonia pathogenesis [46–50], and at a developmental time point when Thap1 levels and transcriptional abnormalities peak [10,15,18], supporting the notion that dystonia is a developmental disorder [51,52]. These initial experiments yielded data consistent with the proposed hypothesis, further supported by ensuing functional validation.
The data herein support our previous conclusion that the C54Y mutation does not equal a DNA-binding loss-of-function mutation. First, there were a far greater number of DEGs in Thap1+/- than Thap1C54Y/+ mice (Fig 1). Second, fold changes were overall lower in Thap1C54Y/+ in comparison to Thap1+/- (S5 Table). Third, only a small number of DEGs were altered in both genotypes (S5 Table). For both mutations, the log2 values are low for what would be predicted for a transcription factor [53], perhaps in the presence of the C54Y mutation due to compensatory auto-up-regulation of Thap1 [18,54]. Reports of dystonia patients with homozygous THAP1 mutations with non-manifesting heterozygous parents [55,56] highlight dosage dependent effects of THAP1 mutations.
THAP1 mutations occur in all domains of the protein, but genotype-phenotype correlations have proven difficult to establish due to the small number of patients with each mutation. The C54Y protein, unlike the WT protein, does not bind to the Tor1a promoter [22], but it may bind DNA at other sites and perhaps aberrantly expands DNA binding beyond THAB motifs [57] and/or alter cofactor binding [58]. Thus, some DYT6 mutations represent a partial or total loss of function, whereas others could lead to a combination of haploinsufficiency and gain of function, accounting for the genotype-dependent DEG disparity. Shared transcriptomic and phenotypic features, despite the many differences between mice carrying the C54Y and null alleles, also support this possibility. Thus, there was overlap of key disordered biological processes and biofunctions in the striatum and cerebellum between genotypes. These results indicate that the choice of animal model must be carefully considered in DYT6 research, as different mutations can yield divergent results, although sometimes leading to dysregulation of the same pathways and processes.
These new data together with published DYT1 studies [24,26], suggest a point of convergence of neuronal dysfunction on the eIF2α pathway in DYT6 and DYT1. To a lesser extent, other translational control pathways (mTOR and eIF4/p70S6K) are also implicated in the dysregulation produced by mutations in Thap1. Notably, as a transcription factor, THAP1 regulates TOR1A transcription in artificial systems, but this has not been verified in vivo [18,22,59].
A proteomics-based study identified abnormal eIF2α pathway activation in DYT1 mouse and rat brain, which correlated with assays in human brain [24]. A second group designed an RNAi-based functional genomic screening in HEK293T cells that also implicated the eIF2α pathway in DYT1 biology. Moreover, pharmacological manipulation of eIF2α signaling restored absent cortico-striatal LTD in DYT1 knock-in mice [26]. Together, these reports support the presence of abnormal eIF2α signaling in DYT1 brain and its possible causal link to DYT1 synaptic defects. EIF2α signaling provides a potential point of merger with another, rarer form of primary dystonia. DYT16 is caused by mutations in PRKRA, an activator of the eIF2α kinase PKR, with evidence of abnormal eIF2α phosphorylation in patient fibroblasts [27,28]. Coding variants in ATF4, a direct target of eIF2α, were found in patients with focal dystonia [26] and lastly, a recent gene-expression analysis in adult cerebellar tissue from a mouse model of DYT11 dystonia also identified genes associated with protein translation among the top down-regulated mRNAs [60].
We report eIF2α-pathway-related molecular and electrophysiological findings in DYT6 mice that have some similarities with those in DYT1, including abnormalities in baseline ATF4 and in LTD. Taken together, these reports suggest that efforts in dystonia research should include the unravelling of the mechanisms underlying these observations, addressing causality and reversibility.
Multiple dysregulated pathways highlighted in the GO and IPA analyses of the RNA-Seq data may contribute to the deficits in synaptic plasticity and neuritogenesis described herein. These include eIF2α signaling, which in addition to being a key component of ER stress responses, regulates important physiological events under homeostatic conditions, including synaptic plasticity [61–65]. ATF4 plays a specific role in neuronal plasticity, postsynaptic development, mushroom spine density, memory, neuronal survival, caspase activation and dopaminergic neuron degeneration [66–69]. Selective knockdown of ATF4 impairs hippocampal LTP in vitro and in vivo [68] and as noted above, pharmacological inhibition of eIF2α dephosphorylation rescues cortico-striatal LTD defects in DYT1 KI mice [26]. Notably, eIF2α signaling was not among the top functional pathways in C54Y mice, and these mice exhibited reduced LTP but normal LTD. This highlights the presence of a synaptic plasticity defect in both dystonia models, but with genotype-driven differences as discussed earlier. Furthermore, these data suggest intriguing similarities between the different dystonia models that deserve further study, as there are other IPA and GO-enriched pathways in both genotypes that impinge on synaptic plasticity. These include LTD and LTP themselves in the ΔExon2 mice, the mTOR translation control pathway [70,71], Dopamine-DARPP-32 feedback in cAMP signaling [72], and G-protein/second messenger/cAMP gene groups.
Finally, we identify deficits in neuritogenesis in vitro as a possible convergence point between Tor1a and Thap1 mutations. Mutant TorsinA inhibits neurite extension in cultured cells [73] and DYT1 mice have thinner and less complex dendrites in Purkinje cells and striatal medium spiny neurons [42,43]. The deficit in neuritogenesis is present in both ΔExon2 and C54Y striatal neurons. Many of the same highlighted GO and IPA pathways that could alter synaptic plasticity may also contribute to defective neuritogenesis, including the overlapping GO neurite projection terms and IPA Axonal Guidance Signaling, eIF2α/ATF4 pathway [74], Rho GTPase signaling, and G-protein and cAMP signaling pathways [75]. The fact that in 9-week-old mice, basal synaptic efficiency was enhanced in corticostriatal inputs of Thap1C54Y mice yet normal in Thap1+/- mice, suggests that any effects on neuritogenesis that persist into adulthood might be offset by additional synaptic changes. Future studies will need to address the temporal and biological relationship between the neurodevelopmental and plasticity phenotypes uncovered in this study.
Other disordered pathways previously implicated in DYT1 and observed in the study reported herein are related to mitochondrial dysfunction [76,77] and lipid metabolism [78,79]. Interestingly, many of the Thap1 DEGs which are included in the Emory University and Mount Sinai Genetic Testing Movement Disorders and Neuromuscular Disease Panels are also linked to those biological processes, suggesting that lipid metabolism and mitochondrial function may deserve further investigation in different forms of dystonia. Furthermore, torsins have recently been implicated as essential regulators of cellular lipid metabolism [78].
In conclusion, using an unbiased transcriptomic analysis in two brain regions from two mouse models of DYT6, we identified eIF2α dysregulation as a potential point of convergence between different forms of dystonia, possibly through its influence on key homeostatic neurodevelopmental events. The identification of similar eIF2α dysregulation and synaptic plasticity defects as previously described in DYT1 mice and rats in the DYT6 animals is a key convergence of biological mechanisms among inherited dystonias, perhaps adding the group of translational dysregulation-associated dystonias (DYT1, DYT3, DYT6, and perhaps DYT16 [24,26,80] to those linked to dopamine dysfunction (DYT5, DYT11, DYT25) [7]. Moreover, the disordered post-synaptic DARPP-32/G-protein/cAMP signaling system(s) potentially overlaps with other dystonias, particularly DYT25 caused by mutations in GNAL [8], suggesting that there are multiple pathways which may contribute to this phenotype. The consolidation of multiple types of dystonia into specific pathogenic mechanisms could facilitate focused research into the etiology of dystonia and the rational design of targeted therapies applicable to this group of movement disorders.
Experimental procedures were carried out in compliance with the United States Public Health Service's Policy on Humane Care and Use of Experimental Animals and were approved by the Institutional Animal Care and Use Committee (IACUC) at Icahn School of Medicine at Mount Sinai (Protocol #07–0483).
The Thap1C54Y/+ and Thap1+/- mice used in this study were congenic C57Bl. The generation of the original mice has been previously provided in detail in Ruiz et al. [18] and as the mutations are early embryonic lethal, breeding strategy was always heterozygote X WT. Animals were maintained on a 12:12 light: dark cycle with ad libitum access to food and water throughout the course of the entire experiment.
Postnatal day 1 (P1) pups were euthanized by decapitation, the brain was dissected, snap-frozen striatal and cerebellar tissues were homogenized in QIAzol Lysis Reagent (Qiagen). Total RNA purification was performed with the miRNeasy mini kit (Qiagen), and was carried out according to the manufacturer’s instructions. Five hundred nanograms of RNA were reversed-transcribed using the High Capacity RNA-to-cDNA Kit (Applied Biosystems, Foster City, CA, USA). The cDNA solution was subjected to real-time qPCR in a Step-One Plus system (Applied Biosystems) using the PerfeCTa SYBR Green FastMix ROX (Quanta BioSciences). Quantitative PCR consisted of 40 cycles, 15 s at 95°C and 30 s at 60°C each, followed by dissociation curve analysis. Primer sequences can be found in S8 Table.
Total RNA from P1 striatal and cerebellar tissues (2–3 μg/ sample) were submitted for further processing to the Genomics Core Facility at the Icahn School of Medicine at Mount Sinai. Sequencing libraries were prepared with the TruSeq RNA Sample Prep Kit v2 protocol (Illumina, San Diego, CA, USA). Briefly, ribosomal RNA was removed using the Ribo-Zero rRNA Removal Kit (Human/Mouse/Rat) (Illumina), the remaining RNA fragmented and cDNA synthesized with random hexamers, end-repaired and ligated with appropriate adaptors for sequencing. After size selection and purification using AMPure XP beads (Beckman Coulter, CA, USA), 6 bp barcode bases were introduced at one end of the adaptors. The size and concentration of the RNA-Seq libraries was measured by Bioanalyzer and Qubit fluorometry (Life Technologies, Grand Island, NY, USA), and the rRNA-depleted libraries sequenced on the Illumina HiSeq 2500 System with 100 nucleotide paired-end reads.
Fastq files were aligned to the Ensembl release 88 (GRCm38.75) version of Human Reference genome (mm10) [81], using STAR read aligner [82]. Accepted mapped reads were summarized separately to gene and exon levels using the featureCounts function of subread [83,84], and used to generate gene and exon count matrices. We examined gene expression data for each sample, and found that the total number of mapped reads [Total Reads (Mean): 39,708,979.5; Uniquely Mapped Reads (Mean): 36,244,152.41] were similar across all samples (S10 Table). There were no obvious outlier samples on visual inspection of principal component analysis or hierarchically clustering of gene expression, and all samples were retained for downstream analyses.
For each of the primary comparisons within the study, the assembled count matrix was filtered to remove transcripts without any counts in any sample. Counts were adjusted by total library size and normalized using DESeq2 [85]. P-values were adjusted using the Benjamini-Hochberg method [86]. Due to the small number of DEGs present in striatum and cerebellum of Thap1C54Y/+ mice, p-value of 0.05 was used as cut off.
Mitochondrial respiratory complex I activity assay (ab109721, Abcam) was performed, utilizing striatal and cerebellar protein lysates from P1 Thap1+/-(ΔExon2) and WT mice (n = 4 per genotype), according to manufacturer's instructions.
Two subcutaneous injections of 3 μg/g tunicamycin (T7765, Sigma-Aldrich) diluted in 150mM dextrose (or dextrose-only control) were applied two hours apart on postnatal day 4 as described [30]. Mice were euthanized 24 hours later, the striatum and cerebellum dissected and snap frozen.
Snap-frozen striatal and cerebellar tissues were homogenized in RIPA buffer with N-ethylmaleimide and protease/phosphatase inhibitors as described previously [24,87]. Protein concentrations were determined using the BCA assay (23225, ThermoFisher Scientific), 30 μg protein lysates were resolved in 10% or 12% Bis/Trisacrylamide gels (BioRad), transferred to nitrocellulose membranes and western blot completed and quantified using the following antibodies: ATF4 / CREB-2 (1:200; sc-200, Santa Cruz), BiP (1:1000; 3177, Cell Signaling), PERK (1:1000; 5683, Cell Signaling), p-PERK (1:1000; 3179, Cell Signaling), eIF2α (1:200; sc-11386, Santa Cruz), p-eIF2α (1:1000; 9721, Cell Signaling), CHOP (1:500; sc-7351, Santa Cruz), mTOR (1:1000; 9862, Cell Signaling), p-eIF4B (Ser406) (1:1000; 5399, Cell Signaling) and GAPDH (1:1000; sc-32233, Santa Cruz). All primary antibodies were incubated in TBS-Tween 5% milk for overnight at 4°C. Membranes were then washed in TBS-Tween. Secondary antibodies anti-rabbit IgG–HRP (PI-1000, Vector Laboratories) or anti-mouse IgG-HRP (PI-2000, Vector Laboratories) were used (1: 5,000) in TBS-Tween 5% milk for 1 hour at room temperature. Immunoreactive proteins were visualized using either Pierce ECL Western Blotting Substrate (32106, ThermoScientific) or Amersham ECL Prime Western Blotting Detection Reagent (RPN2232, GE Health) on a Fujifilm LAS4000 imaging device.
Nine-week-old mice were anesthetized with isoflurane, their brains rapidly removed from the skull and placed in ice-cold modified solution (aCSF) containing (in mM): 215 sucrose, 2.5 KCl, 1.6 NaH2PO4, 4 MgSO4, 1 CaCl2, 4 MgCl2, 20 glucose, 26 NaHCO3 (pH = 7.4, equilibrated with 95% O2 and 5% CO2) and 250 μm thick coronal slices containing the striatum prepared with a Vibratome VT1000S (Leica Microsystems), incubated at 31°C for 30 min and then at room temperature for ≥ 1h in normal aCSF containing (in mM): 120 NaCl, 3.3 KCl, 1.2 Na2HPO4, 26 NaHCO3, 1.3 MgSO4, 1.8 CaCl2, 11 glucose (pH = 7.4 equilibrated with 95% O2 and 5% CO2). Hemi-slices were transferred to a recording chamber constantly oxygenated and perfused with aCSF at ~4mL/min using a peristaltic pump (Masterflex C/L); experiments were performed at 28.0 ± 0.1°C. Recordings were acquired with a GeneClamp 500B amplifier (Axon Instruments) and Digidata 1440A (Molecular Devices). All signals were low-pass filtered at 2 kHz and digitized at 10 kHz. For field EPSP recordings, a patch pipette was fabricated on a micropipette puller (Sutter Instruments), filled with normal aCSF, and placed in the dorsomedial striatum for LTP or dorsolateral striatum for LTD. A concentric bipolar electrode (FHC) was positioned immediately above the corpus callosum. Before and after HFS, the stimulus intensity was set evoke an EPSP that was 50% of the maximal obtainable response. During HFS, stimulus intensity was increased to evoke a maximal response. LTD or LTP was induced by the following high-frequency stimulation (HFS) protocol: four 1-s duration, 100 Hz trains, separated by 10 s. mGluR LTD was induced by a 5-min bath application of 100 μM DHPG [(S)-3,5-Dihydroxyphenylglycine]. Sal003, when present, was applied for 5–10 min before DHPG, and washed out with DHPG. Gabazine, when present, was applied at 10 μM for at least 20 min before the delivery of HFS, and remained in the superfusate for the remainder of the recording. Square-wave current pulses (100 μs pulse width) were delivered through a stimulus isolator (Isoflex, AMPI). Paired-pulse ratio was measured by delivering two stimuli at 20, 50, 100 ms inter-stimulus intervals. Each inter-stimulus interval was repeated three times and the resulting EPSPs were averaged.
Primary striatal cultures were prepared as described [10], fixed with 4% paraformaldehyde 48 hours after plating, and processed for immunostaining following our published protocol [14] using the following primary antibodies: Tau (1:500, MN1000, ThermoFisher Scientific), TUJ1 (1:250, sc-51670, Santa Cruz), and MAP2 (1:1,000; AB5622, Millipore). Cells were visualized under an Olympus IX51 inverted fluorescent microscope.
NeuriteTracer, a neurite tracing plugin for the freely available image-processing program ImageJ was used to analyze fluorescence microscopy images of neurites and nuclei of cultured primary neurons. The plugin was used to count neuronal nuclei, and traces and measure neurite length as described [88]. Ten randomly selected images of each neuronal culture type were processed. DAPI staining was employed as a nuclear stain. The average length was obtained by dividing the total length of the traces by the number of nuclear counts.
GraphPad software (GraphPad Prism 5) was used to perform Student's t-tests for the qPCR and western blot densitometry, two-way ANOVAs followed by Tukey’s post hoc tests were used to analyze the tunicamycin western blot densitometry. Statistical differences in neurite length / outgrowth were assessed by ANOVA with Student’s post hoc t-test. Statistical significance was deemed to be achieved if P < 0.05. Values are presented as mean ± SEM. Electrophysiological data were analyzed by one-way ANOVAs followed, where appropriate, by Newman-Keuls post hoc tests.
All next generation sequencing data are deposited in NCBI-Gene Expression Omnibus database and are accessible through GEO Series accession number GSE98839. The accession number for human ENCODE ChIP-Seq data from K562 cells used in the manuscript for THAP1 is GEO: GSM803408. The accession number for the mouse embryonic stem (ES) cells ChIP-Seq data for Thap1 used in the manuscript is GEO: GSE86911.
|
10.1371/journal.pcbi.1002225 | Replication Timing: A Fingerprint for Cell Identity and Pluripotency | Many types of epigenetic profiling have been used to classify stem cells, stages of cellular differentiation, and cancer subtypes. Existing methods focus on local chromatin features such as DNA methylation and histone modifications that require extensive analysis for genome-wide coverage. Replication timing has emerged as a highly stable cell type-specific epigenetic feature that is regulated at the megabase-level and is easily and comprehensively analyzed genome-wide. Here, we describe a cell classification method using 67 individual replication profiles from 34 mouse and human cell lines and stem cell-derived tissues, including new data for mesendoderm, definitive endoderm, mesoderm and smooth muscle. Using a Monte-Carlo approach for selecting features of replication profiles conserved in each cell type, we identify “replication timing fingerprints” unique to each cell type and apply a k nearest neighbor approach to predict known and unknown cell types. Our method correctly classifies 67/67 independent replication-timing profiles, including those derived from closely related intermediate stages. We also apply this method to derive fingerprints for pluripotency in human and mouse cells. Interestingly, the mouse pluripotency fingerprint overlaps almost completely with previously identified genomic segments that switch from early to late replication as pluripotency is lost. Thereafter, replication timing and transcription within these regions become difficult to reprogram back to pluripotency, suggesting these regions highlight an epigenetic barrier to reprogramming. In addition, the major histone cluster Hist1 consistently becomes later replicating in committed cell types, and several histone H1 genes in this cluster are downregulated during differentiation, suggesting a possible instrument for the chromatin compaction observed during differentiation. Finally, we demonstrate that unknown samples can be classified independently using site-specific PCR against fingerprint regions. In sum, replication fingerprints provide a comprehensive means for cell characterization and are a promising tool for identifying regions with cell type-specific organization.
| While continued advances in stem cell and cancer biology have uncovered a growing list of clinical applications for stem cell technology, errors in indentifying cell lines have undermined a number of recent studies, highlighting a growing need for improvements in cell typing methods for both basic biological and clinical applications of stem cells. Induced pluripotent stem cells (iPSCs)—adult cells reprogrammed to a pluripotent state—show great promise for patient-specific stem cell treatments, but more efficient derivation of iPSCs depends on a more comprehensive understanding of pluripotency. Here, we describe a method to identify sets of regions that replicate at unique times in any given cell type (replication timing fingerprints) using pluripotent stem cells as an example, and show that genes in the pluripotency fingerprint belong to a class previously shown to be resistant to reprogramming in iPSCs, identifying potential new target genes for more efficient iPSC production. We propose that the order in which DNA is replicated (replication timing) provides a novel means for classifying cell types, and can reveal cell type specific features of genome organization.
| In mammals, replication of the genome occurs in large, coordinately firing regions called replication domains [1]–[7]. These domains are typically one to several megabases, roughly align to genomic features such as isochores, and are closely tied to subnuclear position, with transitions to the nuclear interior often coupled to earlier replication, and transitions to the periphery to later replication [4], [5], [8], [9]. Given their connections to subnuclear position and remarkably strong correlation to chromatin interaction maps [3], replication profiles provide a window into large-scale genome organization changes important for establishing cellular identity. The organization of replication domains is cell-type specific, and a larger number of smaller replication domains is a property of embryonic stem cells (ESCs) [3]–[5]. Importantly, in both humans and mice, induced pluripotent stem cells (iPSCs) reprogrammed from fibroblasts display a timing profile almost indistinguishable from ESCs, suggesting that replication profiles may also be used to measure cellular potency [3], [5].
While a wide-range of cell classification methods are actively used, the most common practice for verifying identity is to monitor a handful of molecular markers, some of which are shared with other cell types. Genome-wide classification of features such as DNA methylation [10]–[12], transcription [13], [14] and histone modifications [15], [16] have in principle more potential to accurately distinguish specific cell types. However, these features of chromatin are highly dynamic at any given genomic site [17], and most measurements require high-resolution arrays and costly antibodies. Moreover, recent reports highlight the unstable nature of transcription and related epigenetic marks in multiple embryonic stem cell lines [18], [19]. By contrast, since replication is regulated at the level of large domains, replication profiles are considerably less complex to generate and interpret than other molecular profiles. Timing changes occurring during differentiation are on the order of several hundred kilobases and are highly reproducible between various stem cell lines [3]–[5]. They are also robust to changes in individual chromatin modifications, retaining their normal developmental pattern in G9a(−/−) cells despite strong upregulation of G9a target genes and near-complete loss of H3K9me2 [8].
Here, we describe a method for classifying cell types—replication fingerprinting—based on genome-wide replication timing patterns in mouse and human ESCs and other cell types. We applied the method to 67 (36 mouse and 31 human) whole-genome replication timing datasets to demonstrate the feasibility of classifying cell types using a minimal set of cell type-specific regions. After identification, these regions were used to classify two independent samples using site-specific PCR. We also demonstrate that loss of pluripotency is accompanied by consistent changes in replication timing, implicating the replication program as an important factor in maintaining pluripotency and revealing a novel fingerprint for pluripotent stem cells.
In addition to our previously reported replication profiles, BG02 hESCs were differentiated to mesendoderm and definitive endoderm as previously described [20], as well as ISL+ mesoderm and smooth muscle cultured in defined medium (Methods), and profiled for replication. Replication profiles were generated as described previously [3]–[5], [21]. In brief, nascent DNA fractions were collected in early and late S-phase, differentially labeled, and co-hybridized to a whole-genome CGH microarray. The ratio of early and late fraction abundance for each probe—“replication timing ratio”—represents its relative time of replication. Values from individual probes are then smoothed using LOESS (a locally weighted smoothing function), and plotted on log scale (Figure 1). Replication profiles generated in this way are freely available to view or download at www.ReplicationDomain.org [22], and those analyzed in this report are summarized in Table S1.
Figure 1 illustrates the basic concept of replication fingerprinting. Two exemplary profiles each for D3 embryonic stem cells (ESCs; blue) and D3 ESC-derived neural precursor cells (NPCs; green) are overlaid. Given that most of the genome is conserved in replication timing between any two cell types (e.g. 80% conserved between ESCs and NPCs [4]), the first challenge is to choose genomic regions that are differentially replicated within a set of cell types. We define a “replication fingerprint” of a cell type as a set of genomic regions useful for classification, along with their associated replication timing values. For a simplified example, we show exemplary fingerprint regions for a segment of chromosome 7 (Figure 1A, gray bars). Note that the four regions change dramatically upon differentiation to neural precursors (e.g., ESC2 vs. NPC1; Figure 1A,B), but have replication timing values that are well conserved between replicate experiments (e.g., ESC1 vs. ESC2). We and others have observed similarly widespread changes in replication profiles between any two different cell types profiled to date [1], [3]–[5], [7].
As classification methods require a measure of distance between samples, we defined the distance between replication profiles as the sum of absolute differences in replication timing in fingerprinting regions (Figure 1B). To select an optimal set of fingerprinting regions we maximize a “distance ratio,” representing the ratio of the average distance between unlike cell types to the average distance between equivalent cell types (Figure 1C). This ratio is maximized by selecting regions that are consistently different in replication timing between different cell types, but consistently similar between equivalent types. Importantly, the assignment of unlike vs. equivalent cell types is user-defined and flexible, allowing selection of features that best distinguish any group of cells from any other, such as ESCs from NPCs, normal from disease-related cells, or pluripotent from committed cells.
While Figure 1 shows a simplified example of four regions distinguishing ESCs from NPCs, real-world classification requires the ability to make distinctions genome-wide between many cell types, making manual selection of regions impractical. Therefore, to make the method generally applicable, we developed an automated algorithm based on Monte Carlo sampling [23] to select regions that best distinguish between all available cell types in genome-wide replication datasets. Alternative approaches evaluated for feature selection and classification included Bayesian networks, nearest neighbor methods, decision trees and SVMs, which were comparably successful only for smaller collections of cell types. We chose to explicitly maximize distances between cell types in the method described here in anticipation of translating cell classification to more convenient empirical assays with a limited number of features, because larger timing differences are easier to verify empirically and are more robust to experimental and biological variation.
In practice, replication fingerprinting is a feature selection problem. Although most genome-wide approaches are both simple and comprehensive, we found that genome-wide correlations and distances, while a good first approximation of the relatedness between cell types, are not ideal for classification as the small amount of noise in regions with conserved replication timing is compounded over this relatively large fraction of the genome (Figure S1). We therefore wish to exclude domains that are noisy (having high technical or biological variability), irrelevant (conserved in all cell types), or redundant (containing overlapping information). To achieve this, we first remove regions with conserved replication timing between cell types, resulting in a set of informative regions that can be further optimized by a Monte Carlo selection algorithm.
Figure S2 depicts the Monte Carlo algorithm. To reduce noise from individual probe measurements, replication profiles are first averaged into nonoverlapping windows of approximately 200 kb. This window size represents a balance between sizes of the regions that change replication timing during development (400–800 kb), and the number of probes needed for timing changes to be deemed statistically significant (35–180 probes are contained in each window depending on the probe density of the array platform; see Methods, Table S2). An initial set of regions with the highest replication timing changes in the set of replication profiles are chosen to exclude regions with conserved replication timing, and half of these starting regions are randomly selected to calculate initial distances between cell types. At each iteration of the algorithm, a region can be added to the set of fingerprint regions, removed from the set, or swapped with an unused region. Using a Metropolis-Hastings criterion [23], [24], moves that improve the overall distance ratio are accepted with higher probability than those that do not; after 20,000 or more such moves, a final set of fingerprinting regions is selected.
As depicted in Figure 2, the fingerprinting algorithm selects domains with large and reproducible replication timing differences between cell types, discarding those with minimal or variable changes. Before selecting optimal regions (Figure 2A,C), the average distance between “like” and “unlike” cell types are similar, translating into classification errors for randomly selected domains (Figure 2C) as well as the whole genome (Figure S1, red shading). After selection, the separation in distances between like and unlike types becomes very distinct (Figure 2B, D), even for closely related cell types (Figure 3). These regions similarly highlight distinctions between cell types both in correlations (Figure S3, S4, S5, S6, S7, S8), and distance matrices between cell types (Figure S9, S10, S11, S12).
Since Monte Carlo selection is stochastic, different sets of fingerprinting regions can be selected in different runs. To evaluate the stability of regions included in replication fingerprints, we applied the algorithm 100 times for each type of human and mouse fingerprint constructed (Figure S13). Results demonstrate that fingerprinting regions are well-conserved among multiple rounds of selection, with the top 10–14 regions selected in 100/100 trials in each case. For all subsequent classification, we used regions included in at least 75/100 fingerprinting runs.
As the distances between profiles derive from either the same or different cell types (Figure 2C), their distributions can be used to create a general classifier (Figure 2C,D, Figure 3A), with an error rate proportional to the overlap in distances shared by “like” and “unlike” cell type comparisons (Figure 2C,D, blue shading). This allows us to state a level of confidence for a given prediction, as well as estimate the similarity of a cell type to others. To refine this classification, we applied the k-nearest-neighbor rule [25] (kNN; k = 3) to assign cell types according to the three most similar profiles in the training set. Distances below the threshold – θ = 2.4 in Figure 2D – are hypothesized to derive from similar cell types, and are used with kNN to classify profiles according to the closest profiles in the training set. Distances above the threshold are presumed to derive from different cell types, preventing kNN from classifying highly divergent RT profiles as the cell type of the most similar known profile.
To test the ability of our method to select suitable regions for classification, we applied it to predict the known identity of 9 mouse and 7 human cell types with 36 and 31 total experimental replicates, respectively. Datasets used for prediction are summarized in Table S1, with most described in detail in previous publications [3]–[5]. Rough classification of each experiment into like and unlike cell types by a distance ratio cutoff was accurate in 951/961 (99.0%) human and 1250/1296 (96.5%) mouse comparisons respectively (Figure 3A,B). Refining this classifier by using kNN to assign cell types according to the three most similar profiles in the training set resulted in correct predictions for 36/36 mouse and 31/31 human replication timing profiles (Figure 3C,D). Strikingly, even closely related cell types could be reliably distinguished using this method, such as mouse ESCs and early primitive ectoderm-like stem cells (EPL/EBM3), and two day intermediates of human ESC differentiation into endomesoderm (DE2; day 2) and definitive endoderm (DE4; day 4). Thus, replication profiles appear capable of distinguishing among a wide array of cell types in early mouse and human development.
The use of all experimental data in a selection algorithm often results in overfitting the model to a limited set of observations. For this reason, machine-learning algorithms are commonly trained and tested on different subsets of data (termed cross-validation). To determine whether overfitting is occurring in our selection method and assess the degree to which fingerprinting domains are generally cell type-specific, we performed leave-one-out cross-validation (LOOCV) with each of the available experiments by constructing fingerprints using all but one experimental replicate, and testing classification on the remaining replicate. In all cases (31/31 human, 36/36 mouse), correct predictions in the excluded profile confirmed that fingerprinting regions remained consistent with cell type, and that most cell-line-specific differences were discarded (Figure 3C, LOOCV column). This was also true for a cell line with only one replicate (mouse 46C neural precursor cells), implying that most of the regions of differential replication timing useful for classification are shared between cell lines.
To simulate the classification of a cell type not yet encountered in the training set, we tested predictions after selecting fingerprinting regions with all replicates of a given cell type excluded (Figure 3C, LCTO column). This confirmed that most cell types not yet encountered were correctly classified as “Unseen” (7/7 cell types in human, 7/9 in mouse). However, two cases in which profiles were ambiguous were between neural precursors (NPCs) and mouse epiblast-like stem cells (EpiSCs, EBM6), suggesting that closely related cell types are more accurately distinguished when examples of each type are included in the training set.
One of the most striking features of replication timing is its widespread consolidation into larger replication domains during neural differentiation, concomitant with global compaction of chromatin [3], [4]. This consolidation, along with recovery of ESC replication timing by induced pluripotent stem cells (iPSCs), suggested that replication patterns in specific regions of the genome are associated with the pluripotent state. Further, if certain timing changes are a stable property of cellular commitment, they may provide a unique opportunity to evaluate differentiation capacity using replication-timing patterns. To explore this, we analyzed the differences in replication profiles between collections of pluripotent/reversible (ESCs, iPSCs, EBM3) and committed cell types in 13 human and 21 mouse cell lines (Figure 4A). In each case, we created a stringent consensus fingerprint for classification consisting of regions found in >75/100 runs (18 regions each in mouse and human), and examined genes in the top 200 fingerprint regions (∼2% of the genome) to characterize a more inclusive sample. Genes and regions found to consistently switch to earlier or later replication as pluripotency is lost are provided in Tables S3, S4, S5, S6.
Strikingly, several regions displayed conserved, significant differences in timing between all pluripotent and committed cell types (Figures 4A, S10, S12). As with general fingerprints, classification into pluripotent or committed types could be performed unambiguously (36/36 cases in mouse, 31/31 in human), even with regions selected with the test profile excluded (LOOCV column). Several of the genes consistently switching to later replication in mouse and human pluripotency fingerprints have known roles in maintaining pluripotency (for instance, Dppa2 and Dppa4 in both species, and DKK1 in human; Tables S4 and S6). In addition, two classes of genes stood out from this analysis that showed significant switches to later replication in both species: a large cluster of protocadherins (PCDs), and the majority of the Hist1 cluster of core histone genes (Table S7). The former are developmentally regulated genes with broad involvement in neural development and cell-cell signaling [26], [27], and switch to later replication in all committed mouse and human cell types. The latter Hist1 cluster was later replicating in 8/8 committed cell types in mouse and 5/6 in human (not lymphoblasts), and includes several core histone genes that were downregulated up to 2.5-fold in NPCs. These results are intriguing in light of previous reports of histone downregulation during development [28], as well as a hyperdynamic chromatin phenotype in ESCs that involves higher exchange rates of histone H1 [29] and is required for efficient somatic cell nuclear reprogramming in Xenopus oocytes [30]. Importantly, all of the histone H1 genes are found in this cluster, suggesting that regulation of global H1 abundance may provide a mechanism for the overall chromatin compaction and consolidation of replication timing observed during neural differentiation [3]–[5].
To characterize the genes included in the mouse pluripotency fingerprint, we compared them to a previous class of genes that showed lineage-independent switches to later replication in mouse ESC differentiation, and failed to revert to ESC-like expression in three separately derived samples of partial iPSCs (clusters 15 and 16 in Figure 7 of Hiratani et al., 2010). Remarkably, 200 out of 217 genes in the top 100 mouse pluripotency regions belonged to this class, despite very different methods for deriving them (Figure 5A). All of the fingerprint genes switched to later replication, and at the transition between early and late epiblast stages where cell fates become restricted [5]. Most genes also had reduced expression in late epiblast and neural progenitor stages (average 1.66-fold reduction in transcription from ESC/EBM3 to EBM6/NPCs). Thus, some of these genes may make prime candidates for improving the efficiency of iPSC production, or for reverting human ESCs to a more naive, mouse ESC-like state. However, the overlap between human and mouse pluripotency fingerprint genes, while significant, was much lower (Figure 5A), and this was true even when comparing human ESCs to developmentally analogous mouse EpiSCs [3], [31]. Therefore, many pluripotency-associated genes and loci may be species-specific, consistent with recent studies that underscore considerable differences between mouse and human pluripotency networks [32], [33]. This low alignment is also accounted for by a general drop in overall alignment in regions with the greatest developmental switches in replication timing (Figure 5B), which are those preferentially selected by the fingerprinting algorithm.
Of the genes conserved in the fingerprints of both species (indicated by boldface type in Tables S4 and S6), most belong to the aforementioned large class of protocadherins. However, Dppa2 and Dppa4 are also conserved, as well as genes with no known roles in maintaining pluripotency (Cast, Riok2, Lix1) that reside within the same replication units as pluripotency fingerprint genes in both species. Other core pluripotency genes remain relatively early replicating in both species (Pou5f1[Oct4], Sox2, Nanog), and are likely regulated by other mechanisms. For instance, Sox2 belongs to a class of genes with strong promoters (HCP, or high CPG content promoters) generally unaffected by local replication timing [4], [34].
One potential application of replication fingerprints is in the development of PCR kits for epigenetic classification, particularly for cell types or disease samples with no known aberrations in transcription or sequence. To confirm that fingerprint regions can be translated into a classification scheme using site-specific PCR, we classified two unknown samples representing cell types that were analyzed previously, but that were derived from different cell lines than the original set used for training. The experiment was performed in a blind manner in which the experimenter had no prior knowledge of the regions or cell types being tested. Primers were assembled against sequences within 10–20 kb from the center of each fingerprint region, and the replication times of each region were quantified as the “relative early S phase abundance” (relative abundance of a sequence in nascent strands from early S phase), as previously described [35] (Figure 6A). PCR-based timing values were rescaled for consistency with the original scale of the array datasets used in training, and distances were calculated between the unknown samples and other human profiles in fingerprint regions (Figure 6B). Using the same methods as in prior classifications, these distances correctly identified the two unknown samples as lymphoblasts and hESCs, respectively; the three known datasets with the smallest distances were each of the correct cell type.
Our method for cell typing through replication fingerprinting addresses a well-recognized need for comprehensive methods to assess cellular identity and differentiation potential in stem cell biology. Unlike other molecular markers, replication is regulated at the level of large, multi-megabase domains, making comprehensive, genome-wide profiles relatively simple to generate and interpret [36]. In particular, the robust stability of replication timing profiles in stem cells [8], and wide divergence between cell types make them a promising candidate for classification.
While the functional role for the replication program is not yet understood, its conservation between human and mouse cell culture models of development support its functional significance. We and others have shown a substantial correlation (R2 = 0.42–0.53) in replication patterns between mouse and human cell types, with timing patterns of embryonic stem cells, neural precursor cells, and lymphoblastoid cells most closely aligned to their cognate in the other species [1], [3]. The important role for replication is further corroborated by its remarkably strong link to genome organization [3], and its ability to confirm the mouse epiblast identity of human ESCs genome-wide and with an epigenetic property [3], [31].
By comparison, methods for cell typing using DNA methylation, gene expression, histone modifications or protein markers are well suited to some applications [10]–[16], but may not be informative for certain fractions of the genome, or may rely on genome features that cannot distinguish between similar cell states. We therefore envision replication fingerprinting as a complement to existing cell typing strategies that may be used for samples unsuitable for traditional methods, or for additional confidence in assessing cell identity in cases where this is critical, such as regenerative medicine. One caveat to consider in these applications is that replication profiles, similar to other genome-wide methods, are an ensemble aggregate from many cells, making measurement of homogeneity difficult. In addition, as with other supervised classification approaches, the method is informative only for cell types (classes) available during training. However, our fingerprinting method is in principle applicable to any data type, and may be modified to select discriminating features in other epigenetic profiles.
A major advantage of our fingerprinting method is in selection of a minimal set of regions that allow for classification with a straightforward PCR-based timing assay and a reasonably small set of primers, particularly if only cell-type specific regions are examined. Our results suggest that a standard set of 20 fingerprint loci can be effective for classification, but the number of regions queried can be adjusted based on the confidence level required. The sole requirement for replication profiling is the collection of a sufficient number of proliferating cells for sorting on a flow cytometer. Consistently, just as replication fingerprints can be generated for particular cell types or general categories of cells, features of replication profiles allow for the creation of disease-specific fingerprints, which may be valuable for prognosis.
In addition to cell typing applications, replication profiling is informative for basic biological questions. Here, we have identified regions that may undergo important organizational changes upon differentiation, which include a class of gene that fail to reverse expression in partial iPSCs, and the majority of mouse and human histone H1 genes. Human lymphoblasts retained early replication in H1 genes, which may be explained by their high rate of proliferation. Since highly developmentally plastic regions (including pluripotency fingerprint regions) are poorly conserved (Figure 5B) the evolutionary conservation of cell-type specific timing patterns must be driven by the moderately changing majority of the genome.
The recent derivation of mouse ESC-like human stem cells with various methods raises an intriguing question [37]: will naïve hESCs align better to mESCs than to mEpiSCs for replication timing as they have for transcription? Although pluripotency is currently assessed by marker gene expression or laborious complementation experiments, replication timing assays in regions uniquely early or late replicating in pluripotent cells provide a tractable method to predict the pluripotency of various cell types, as well as insights into conserved genome organizational changes during differentiation.
Mouse replication timing datasets are described in Hiratani et al., 2010. Briefly, mouse embryonic stem cells (ESCs) from D3, TT2, and 46C cell lines were subjected to either 6-day (46C) or 9-day (D3, TT2) neural differentiation protocols to generate neural progenitor cells (NPCs) [4], [5]. For D3, intermediates were also profiled after 3 (EBM3) and 6 (EBM6) days of differentiation. Muscle stem cells (myoblast) and induced pluripotent stem cells (iPSCs) reprogrammed from fibroblasts were collected as described for human and mouse [38]–[40]. For human timing datasets, neural precursors were differentiated from BG01 ESCs as described in Schulz et al., 2004 [3], [41]. Lymphoblast cell lines GM06990 and C0202 were cultured as previously described [2], [42]. Differentiation of BG02 hESCs to mesendoderm (DE2) and definitive endoderm (DE4) was performed by switching from defined media (McLean et al. [20]) to DMEM/F12+100 ng/ml Activin A 20 ng/ml Fgf2 for two and four days, respectively, with 25 ng/ml Wnt3a added on the first day. Mesoderm and smooth muscle cells were derived by adding BMP4 to DE2 cells at 100 ng/ml.
Using custom R/Bioconductor scripts [43], [44], microarray data from Hiratani et al. 2008, Hiratani et al. 2010, and Ryba et al., 2010 were normalized to equivalent scales, and averaged in nonoverlapping windows of approximately 200 kb. Additional profiles for human ESCs, definitive endoderm, mesendoderm, mesoderm, and smooth muscle were derived, normalized and scaled equivalently, as described [45]. Profiles shown in Figure 1 and Figure 4 were smoothed using LOESS with a span of 300 kb.
Selection of fingerprint regions was performed as described using custom R/Bioconductor scripts. Regions of non-conserved RT (2000/10994 mouse, 2000/12625 human) were first selected based on standard deviation, then optimized using a Monte Carlo algorithm (Figure S2). Using the Metropolis-Hastings criterion for Monte Carlo with simulated annealing [23], [24], moves are accepted when exp((dRbest−dR)/T)>i, where dR is the distance ratio of the proposed move, dRbest is the current best distance ratio, T is a temperature parameter that decreases geometrically during the simulation, and i is a random number from 0 to 1.
Cell type classification was performed using absolute distances between experiments measured from replication timing in fingerprint regions, using the k-nearest neighbor rule with k = 3; i.e., each profile was categorized according to the three nearest profiles. Crossvalidation was performed to select an appropriate value for k, with k = 3 chosen as the smallest value that yielded 100% classification accuracy after leave-one-out crossvalidation (LOOCV) to allow classification of cell types with fewer replicates. For LOOCV results, each experiment was sequentially left out during Monte Carlo selection, and the resulting regions were used to predict the identity of the excluded experiment. To test prediction on cell types not yet encountered, all profiles for a given cell type were left out during region selection (LCTO), and cell type was predicted using the resulting regions. All data analysis was performed using custom R scripts and Bioconductor packages [43], [44].
For each fingerprint region depicted in Table 1, 10–20 kb from the center of the region was sent to NCBI Primer-Blast (http://www.ncbi.nlm.nih.gov/tools/primer-blast/) to design several PCR primer sets with product sizes of 150–350 bp, using standard parameters. Forward and reverse primer pairs displaying the greatest specificity were chosen. Primer sets were verified for specificity and product size using the In-Silico PCR tool at the UCSC genome browser (http://genome.ucsc.edu/cgi-bin/hgPcr).
PCR reactions were set up using 1.25 ng genomic DNA and 1 uM each of forward and reverse primers in 12.5 uL scaled according to the instructions of Crimson Taq DNA Polymerase (NEB). Thirty six cycles of PCR (empirically determined to be unsaturated for amplification) were performed according to manufacturer's conditions with annealing temperature of 62°C. One-third of the reaction was analyzed on a 1.5% agarose gel containing ethidium bromide. The gel was scanned by Typhoon Trio (GE Healthcare) and band intensity was quantified by Image Quant TL (GE Healthcare). After the background was subtracted, signal intensity from the early S fraction was divided by the sum of those from early S and late S fractions from each sample, as described [35]. PCR timing values were converted to array RT scale (equivalent root-mean-square) using the scale function in R, and distances were calculated against other cell types as previously performed.
GSE18019, GSE20027
|
10.1371/journal.pcbi.1003888 | Size Does Matter: An Integrative In Vivo-In Silico Approach for the Treatment of Critical Size Bone Defects | Although bone has a unique restorative capacity, i.e., it has the potential to heal scarlessly, the conditions for spontaneous bone healing are not always present, leading to a delayed union or a non-union. In this work, we use an integrative in vivo - in silico approach to investigate the occurrence of non-unions, as well as to design possible treatment strategies thereof. The gap size of the domain geometry of a previously published mathematical model was enlarged in order to study the complex interplay of blood vessel formation, oxygen supply, growth factors and cell proliferation on the final healing outcome in large bone defects. The multiscale oxygen model was not only able to capture the essential aspects of in vivo non-unions, it also assisted in understanding the underlying mechanisms of action, i.e., the delayed vascularization of the central callus region resulted in harsh hypoxic conditions, cell death and finally disrupted bone healing. Inspired by the importance of a timely vascularization, as well as by the limited biological potential of the fracture hematoma, the influence of the host environment on the bone healing process in critical size defects was explored further. Moreover, dependent on the host environment, several treatment strategies were designed and tested for effectiveness. A qualitative correspondence between the predicted outcomes of certain treatment strategies and experimental observations was obtained, clearly illustrating the model's potential. In conclusion, the results of this study demonstrate that due to the complex non-linear dynamics of blood vessel formation, oxygen supply, growth factor production and cell proliferation and the interactions thereof with the host environment, an integrative in silico-in vivo approach is a crucial tool to further unravel the occurrence and treatments of challenging critical sized bone defects.
| In 5–10% of fracture patients, the bone fractures do not heal in the normal healing period (delayed healing) or do not heal at all (non-union). In order to investigate the causes of impaired healing and design potential treatment strategies, we have used a combined experimental and computational approach. More specifically, large bone defects were analyzed in mouse models and simulated by a previously published computational model. After showing that the predictions of the computational model match the observations of the experimental model, we have used the computational model to investigate the underlying mechanisms of action. In particular, the results indicated that the new blood vessels do not reach the central fracture zone in time due to the large defect size, which leads to insufficient oxygen delivery, increased cell death and disrupted bone healing. The healing, however, could be rescued by adequate blood vessel ingrowth from the overlying soft tissues. Moreover, potential treatment strategies were designed based on the influence of these soft tissues. In conclusion, this study demonstrates the potential of a combined experimental and computational approach to contribute to the understanding of pathological processes like the impaired bone regeneration in large bone defects and design future treatments thereof.
| Although bone has a unique restorative capacity, i.e. it has the potential to heal scarlessly, the conditions for spontaneous bone healing are not always present, leading to a delayed union or a non-union. The orthopedic literature does not specify a universally accepted definition of a fracture non-union [1], [2]. The eventual bony union after an atypical long period of healing, in comparison to the normal healing period, is called a delayed union. The absence of healing during at least three to six months defines a fracture non-union in humans. Fracture non-unions (hypertrophic, atrophic or oligotrophic) are classified based on their radiographic and histological appearance [1], [3].
Hypertrophic non-unions are characterized by an abnormal vascularity and abundant callus formation. They are typically caused by excessive motion at the fracture site, which prevents bony bridging although the essential biological factors are present [1]. Atrophic non-unions, however, are the result of inadequate biological conditions and typically appear on radiographs as blunted bony ends. They show little callus formation around the fracture gap, filled with mostly fibrous tissue and little or no evidence of mineral deposition [1]. Oligotrophic non-unions have some radiographic and biological characteristics of both hypertrophic and atrophic non-unions, i.e. they possess the required biological activity but show little or no callus formation [4].
Excess motion, a large interfragmentary gap [5], open fracture [5]–[7], the particular bone [8], location of the trauma within the bone [8], loss of blood supply [9], severe periosteal and soft-tissue trauma [6], [7] are some of the mechanical and biological risk factors for the development of a non-union. Preexisting patient risk factors such as old age [10], cachexia and malnutrition [11], immune compromise [12], genetic disorders (e.g. type 1 neurofibromatosis), osteoporosis [13], anticoagulants [14], anti-inflammatory agents [15], etc. may also affect the fracture healing outcome but are not the primary causes [16]. Besides an extensive amount of experimental research, several computational models have also been developed to further unravel the occurrence of fracture non-unions. For comprehensive reviews on mathematical models of fracture healing, we refer the reader to Geris et al. [17], Isaksson et al. [18] and Pivonka et al. [19]. Despite the large amount of (often phenomenological) information existing in the literature, additional in vivo, in vitro and in silico research is still required to address the key mechanisms that lead to fracture non-unions, determine the factors predictive of fracture complications and establish the optimal therapeutic strategies for each type of fracture non-union.
In this work we propose an integrative in vivo - in silico approach to investigate the occurrence of oligotrophic and atrophic non-unions as well as to design possible treatment strategies thereof. The gap size of the domain geometry of a previously published mathematical model has been enlarged in order to study the complex interplay of blood vessel formation, oxygen supply, growth factors and cell proliferation on the final healing outcome in large bone defects. The simulation results are corroborated by comparison with dedicated experimental data. Next, the mathematical model is used to explain the underlying mechanisms that lead to the experimental observations as well as design different treatment strategies. Finally, the potential of the combined in silico - in vivo approach is demonstrated by applying it to the case of BMP-treated fracture healing.
All animal experiments were conducted according to the regulations and with approval of the Animal Ethics Committee of the KU Leuven.
C57BL/6 mice were purchased from the R. Janvier Breeding Center (France). A segmental defect was created in the right tibia of 14 week-old male mice as described elsewhere [20]. Briefly, animals were anaesthetized with a ketamine-xylazine mixture (100 mg/kg ketamine and 15 mg/kg xylazine) and the right lower leg was shaved. A custom-made external fixator, based on the Ilizarov external fixation device, was fixed to the tibia using 27 G steel needles. Subsequently, the tibia was exposed and a 4.5–5 mm mid-diaphyseal segment was excised with a 6.5 mm diamond saw disk (Codema n.v., Kortrijk, Belgium). A demineralized CopiOs scaffold (2.5×2.5×5 mm3; Zimmer b.v.b.a., Wemmel, Belgium) seeded with 1×106 mouse periosteal cells (passage 4) was implanted, the skin was sutured and animals received postoperative analgesia (buprenorphine, 60 µg/kg body weight). The demineralized CopiOs- scaffold was used to minimize the soft tissue collapse within the critical size defect. After 3, 14 or 56 days animals were sacrificed, the tibia was excised and samples were analyzed by μCT and then processed for histology.
Murine periosteum-derived cells (mPDC) were isolated from the long bones of 8 week-old male mice as previously described [21]. In short, the femurs and tibias isolated from 8 week old male C57BL/6J mice were dissected and digested with collagenase-dispase after protecting the epiphyses with low melting point agarose. After a filtration and washing step, the cells were plated at 1×104 cells per square centimeter and replated when reaching 80–90% confluency. After isolation, cells were pooled per 2–3 mice and cultured in a humidified incubator at 37°C with 5% CO2 in α-minimal essential medium (α-MEM) supplemented with 2 mM glutaMAX-I, 1% penicillin/streptomycin (100 units/ml and 100 µg/ml respectively) and 10% fetal bovine serum (all from Gibco, Life Technologies, Gent, Belgium). When reaching 80–90% confluency, cells were trypsinized and reseeded at 7500 cells/cm2.
To deliver BMP2 at the defect site, mPDCs were transduced 72 hours prior to implantation with an adenoviral vector encoding human BMP2 (a generous gift from Dr. Frank Luyten, KU Leuven, Belgium) at a multiplicity of infection of 50.
Bone formation in large bone defects was followed by radiographic images at different time points after surgery using the Skyscan 1076 high resolution in vivo micro-computed tomography (μCT) scanner (Bruker-μCT, Kontich, Belgium). For bone quantification, samples retrieved at day 56 were scanned using the high resolution SkyScan 1172 μCT system (Bruker-μCT) at a pixel size of 10 µm with 50 kV tube voltage and 0.5 mm aluminum filter. Projection data was reconstructed using the NRecon software and quantification of mineralized tissue was performed using the CTAn software (both from Bruker-μCT).
Isolated bones were fixed in 2% paraformaldehyde overnight and decalcified in EDTA for 14 days at 4°C prior to dehydration, embedded in paraffin and sectioned at 4 µm. Histochemical staining with hematoxylin and eosin (H&E) and immunohistochemical staining for mouse CD31 is routinely performed in our lab and has been described previously [21]. Images were taken on a Zeiss Axioplan 2 light microscope using the Zeiss AxioVision software.
Data are presented as means ± standard error of the means. Data were analyzed by one-way ANOVA using the NCSS statistical software. Differences were considered statistically significant at p<0.05.
The multiscale computational framework for the mathematical modelling of bone fracture healing and its relation to angiogenesis was established earlier and has been described in detail in [22]. The framework consists of (1) a tissue level describing the various key processes of bone fracture healing with 10 continuous variables, (2) a cellular level representing the developing vasculature with discrete endothelial cells and (3) an intracellular level that defines the internal dynamics of the Dll4-Notch signaling pathway in every endothelial cell (Figure 1). The model accounts for the various key processes that occur during the soft and hard callus phase of bone fracture healing (see [22] for a more detailed description). While the model described in [22] already partially accounted for the role of oxygen, we have recently extended the model to capture the various effects of oxygen on cellular processes in a much more complete and refined way [23]. A brief description of the oxygen model is found below and more details are given in Supporting Text S1.
After the initial inflammation phase (which is not included in the current mathematical model), the fracture callus is filled with a cocktail of granulation matrix, stem cells and growth factors. In regions where oxygen is abundantly available (i.e. close to the cortex in the case of normal fracture healing), the mesenchymal stem cells will directly differentiate into osteoblasts and form bone through the intramembranous pathway. In regions where the oxygen tension is lower (i.e. the central fracture callus in the case of normal fracture healing), the mesenchymal stem cells will differentiate to chondrocytes that will form a cartilage template to mechanically stabilize the fracture. This is followed by endochondral ossification during which blood vessels and osteoblasts are attracted to the central fracture callus, resulting in degradation of the cartilage template and bone formation. Finally, the newly formed bone is remodeled (not included in the current mathematical model).
At the tissue level, the fracture healing process is described by calculating the spatiotemporal evolution of the density of mesenchymal stem cells (cm), osteoblasts (cb), chondrocytes (cc), fibroblasts (cf), bone (mb), cartilage (mc), fibrous matrix (mf), osteochondrogenic growth factor (gbc), angiogenic growth factor (gv) and oxygen (n) using 10 non-linear, coupled partial differential equations of the taxis-diffusion-reaction type. At the cellular level, the evolution of the discrete vasculature is determined by sprouting, vascular growth and anastomosis and is modeled by a lattice-free method. At the intracellular level, an agent-based model is used to implement the rules that capture the intracellular dynamics of the Dll4-Notch signaling pathway which determines tip cell selection during sprouting angiogenesis. The oxygen model includes an accurate description of the oxygen dependency of a number of cellular processes, namely osteogenic and chondrogenic differentiation, cell proliferation, cell death, oxygen consumption and the hypoxia-dependent production of an angiogenic growth factor. The cellular consumption of oxygen was described using a Michaelis-Menten kinetic law where the cell-specific maximal oxygen consumption rate has the following relative cellular order: chondrocytes<MSCs<osteoblasts<fibroblasts. The oxygen values at which the considered cell-specific oxygen-dependent processes occur at maximal rate or at which their rate changes are based on a rigorous literature screening of the state-of-the-art experimental knowledge (Figure 2). More specifically, the relative order of the oxygen dependent processes was determined as accurately as possible since it is crucial to the behavior of the oxygen model. The complete description of the set of equations, the boundary and initial conditions, the parameter values, implementation details as well as some underlying assumptions and simplifications can be found in Supporting Text S1 as well as in previous publications [22], [24], [25].
The geometrical domain of the fracture callus, as well as the boundary conditions and initial positions of the endothelial cells (cv) are shown in Figure 3-B. Note that the periosteum near the bone ends is considered to be well vascularized such that a muscular contribution to the vasculature (i.e. the initial position of the endothelial cells) is unnecessary. To simulate the bone regeneration process in a large bone defect, the domain was extended over a distance equal to half the gap size of a murine critical sized defect (5 mm). The effect of the host environment on the fracture healing process is explored with several combinations of boundary conditions, however in the standard compromised condition the influence of the host environment is neglected thereby representing the worst-case scenario (Figure 3-B). It has been shown experimentally that the amount of cells and growth factors is significantly reduced in a large fracture gap [26], [27]. Therefore, in order to simulate this effect, the initial conditions for the MSCs and osteochondrogenic growth factors were decreased tenfold to 2.103 cells/ml and 10 ng/ml respectively in the central callus area (indicated with dots in Figure 3-B). The initial oxygen tension (ninit) in the central callus area is equal to 3.7%. All other model parameters as well as initial and boundary conditions were left unchanged with respect to the normal healing case [23] and can be found in Supporting Text S1 (Figure 3). Note that the computational model does not simulate the presence of the demineralized CopiOs scaffold, which was used to minimize the soft tissue collapse within the critical size defect. Previous results have however shown that the demineralized carrier structure does not contribute nor enhance the bone formation process.
The results of the mathematical model are quantified in terms of tissue fractions, specified for each part of the fracture callus (i.e. endosteal, periosteal and intercortical). The tissue fractions are calculated by the following procedure: first the spatial images are binarized using tissue-specific thresholds (0 means that the tissue is not present, 1 means that the tissue is present in a grid cell). Subsequently, an equal weight is assigned to the different tissues, i.e. if a grid cell contains three tissues, the area of that grid cell is divided by three in the final calculations of the tissue (area) fractions [23].
A qualitatively similar healing progression is predicted by the simulation results as observed experimentally (Figures 4 and 5). At early time points a periosteal reaction, characterized by a thickening of the periosteal layer (Figures 5-B1, B1′ and C1) as well as the presence of a hematoma, a fibrous-like tissue associated with the presence of numerous (red) blood cells (Figures 5-B1, B1″), are observed at the cortical host bone site, both supporting the initial and boundary conditions that were applied in the multiscale model (Figure 3-B). In the center of the large bone defect no signs of tissues or infiltration of blood vessels are detected, only scaffold material together with a low cellularity is observed (Figures 5-B1-center, C1′), corresponding to the predictions of the in silico model (Figure 4-G).
On day 14, a periosteal endochondral ossification reaction is seen, evidenced by the presence of cartilage (large round cells staining grey-blue with H&E) and trabecular-like bone (dense matrix, staining bright pink with H&E, with the clear presence of embedded osteocytes) (Figures 5-B2, B2′; arrow indicates cartilage), while direct bone formation occurs endosteally (Figures 5-B2, B2″). The mathematical model predicts a similar distribution of tissue formation, i.e. direct bone formation near the bony ends and endochondral ossification further away in the fracture callus (Figure 4-B,C). In the center of the defect only a highly dense fibrous tissue is observed in both the experimental, the scaffold remains stained pink-blue with H&E but lack the presence of embedded cells (Figure 5-B2-center), as well as the mathematical model (Figure 4-A). In contrast to the experimental model, the mathematical model does not predict any blood vessels in the central callus area (indicated with dots in Figure 3). These vessels, however, appear to be small and immature whereas the blood vessels that are associated with the sites of bone formation are large and mature (compare Figures 5-C2 and C2′). This discrepancy might be explained by the fact that the mathematical framework only models angiogenesis, i.e. blood vessel growth through the creation of new vessel branches from existing ones, whereas vasculogenesis, i.e. de novo network formation from scattered endothelial cells, is not included here. Indeed, after bone fracture the hematoma will be filled with blood, containing amongst others endothelial precursor cells, which could explain the small, immature blood vessels observed experimentally. We would like to stress, however, that this is a first hypothesis that is currently being explored further.
Notice the closure of the bone marrow canal by new bone on day 56, separating the bone marrow (right) from the scaffold region (left) (Figure 5-B3, B3′, B3″). As such, capping of the bone ends has occurred both in the experimental and the mathematical model (Figure 4-C). The blood vessels in the center are still much smaller compared to those near the edges of the defect (compare Figure 5-C3 and C3′). In the center of the defect no signs of bone formation are detected, only fibrous tissue is seen, at this time point associated with a very low cellular content (Figure 5-B3-center). Also in the mathematical model no additional bone formation is predicted between post fracture day (PFD) 60 and 90, thereby classifying this fracture as a non-union [1], [2].
After this qualitative validation of the model predictions with the experimental observations of bone healing in a large defect, the model was used to understand the mechanisms underlying the occurrence of fracture non-unions. It appears that in the mathematical model, chondrogenic differentiation and cell survival are severely impaired in the central callus area (indicated with dots in Figure 3) due to the harsh hypoxic conditions (optimal oxygen tension for chondrogenic differentiation is 3%, minimal oxygen tension for MSC and chondrocyte survival is 0.5%, see Figure 2) (Figure 4-D,F). Consequently, the angiogenic growth factor (gv), which is the major stimulus for vascular growth and as such endochondral ossification, is not produced in the central callus area (Figure 4-E). As a result, the bone healing stops after capping of the bony ends, resulting in an atrophic non-union (Figure 4-C). Note that the predicted bone front extents further into the callus than observed in the in vivo model. This might be due to some limitations of the computational model. Firstly, in the current model all the progenitor cells can differentiate towards both the chondrogenic and osteogenic lineage, depending on the local growth factor concentrations and oxygen tensions. In reality, however, it has been shown that the progenitors from the endosteal callus can only differentiate towards the osteogenic lineage, resulting in the absence of cartilage in the endosteal callus [28]. Progenitor cells from the periosteum do have the capability to differentiate to both lineages, explaining why endochondral ossification mainly occurs in the periosteal callus [28]. As such, the current simplification of the model leads to an overestimation of the amount and the location of the cartilage matrix, resulting in an overestimation of the predicted bone formation. Secondly, the current model does not account for changes in callus size and shape during the healing process which may also influence the bone formation process.
After establishing the in silico and in vivo non-union model, the in silico model was further used to explore the influence of the gap size on the healing outcome (Figure 6). By increasing the gap size, the bone tissue fraction at PFD 90 is reduced whereas the cartilage fraction remains similar (close to zero) and the fibrous tissue fraction is greatly increased (Figure 6). Although the bone tissue fraction reaches 84% in a 3 mm defect, there is no cortical bridging which indicates the formation of a non-union. The simulation therefore predicts that a murine bone defect becomes critical at 3 mm. In the remaining part of this study we will focus on the bone regeneration process in 5 mm defects, in correspondence with the in vivo set-up described above.
Since for all the different gap sizes explored in Figure 6, the same set of initial and boundary conditions was employed, the occurrence of fracture non-unions might be attributed to an inadequate vascularization of the central callus region. More specifically, the ingrowing vasculature which originates from the bony ends, needs to cover a larger distance in larger defects, resulting in a too late vascularization of the central fracture area (Figure 4-G) and consequently harsh hypoxic conditions (Figure 4-F). As was explained above, these hypoxic conditions lead to cell death thereby arresting the production of angiogenic growth factors and ultimately the bone healing process (Figure 4-C). Clearly, the spatiotemporal patterns of oxygen tension are an important determinant of successful bone repair which prompted us to investigate the complex interplay between oxygen delivery, diffusion and consumption in a critical size defect (5 mm). An extensive sensitivity analysis was performed on the parameter values describing the delivery of oxygen (Gn), the diffusion of oxygen (Dn) and the oxygen consumption by osteoblasts (Qb), chondrocytes (Qc), MSCs (Qm) and fibroblasts (Qf). Moreover, since experimental evidence has shown that the biological potential (e.g. the amount of osteoprogenitor cells and growth factors present) might be greatly reduced in critical size defects [26], [27], we also explored the influence of the initial conditions (cm,init, gbc,init, cf,init, mf,init, ninit) in the central callus area (indicated with dots in Figure 3) on the fracture healing outcome (Table S1 in the supplementary material).
The initial position of the endothelial cells (see Figure S1 in the supplementary material), has a small influence on the final bone tissue fraction (+/−2%). This difference can be attributed to a different spatial filling of the blood vessels in the 2D simulated geometry and is in the same range as the influence of the stochastic component in the description of blood vessel migration on the simulation outcome (+/−3%) [24]. Based on these findings, we consider deviations of more than 2% with respect to 50% of bone tissue fraction to be sufficient to warrant further analysis. In order to gain more understanding in the complex non-linear dynamics of the oxygen model, the mechanisms underlying these significant deviations were investigated and are discussed in more detail below.
The sensitivity analysis revealed a non-linear influence of the initial amount of MSCs (cm,init) on the bone tissue fraction at PFD 90. This can be explained by the fact that on the one hand a low initial concentration of MSCs (cm,init<2.104 cells/ml) reduces the biological potential of the fracture site since less cells can contribute to the bone healing process. On the other hand, a high initial concentration of MSCs (cm,init>2.105 cells/ml) will worsen the detrimental hypoxic conditions in the central callus region due to the increased amount of oxygen consumption. The initial concentration of fibroblasts (cf,init) does not show this non-linear behavior. High initial concentrations of fibroblasts and/or MSCs are detrimental (cf,init>5.105 cells/ml) since the increased oxygen consumption will lower the average oxygen tension in the central callus area. Contrary to the MSCs, low initial concentrations of fibroblasts do not seem to have a major influence on the final amount of bone formation. This is mainly because fibroblasts do not contribute to the biological potential of the hematoma as they cannot differentiate towards the osteogenic or chondrogenic lineage.
The sensitivity analysis also indicates that the amount of osteochondrogenic growth factors present in the fracture hematoma (gbc,init) is a critical determinant of the final amount of bone formation. Indeed, increasing the growth factor concentration results in a significant increase in the amount of bone formation measured after 90 days of healing. This result can be attributed to an increased chondrogenic differentiation which limits the oxygen consumption since chondrocytes consume less oxygen than MSCs. As such, the central hypoxic area will be reduced leading to more bone formation.
After the inflammation phase, the fracture callus is filled with granulation tissue (represented here by mf,init). It appears that a large amount of granulation tissue negatively influences the fracture healing outcome which is due to its inhibitory effect at large matrix densities on the proliferative capacities of MSCs, fibroblasts, chondrocytes and osteoblasts.
Similar to the initial amount of MSCs also the initial oxygen tension (ninit) has a non-linear effect on the final amount of bone formation. Very low oxygen tensions (ninit<0.5%) lead to a larger hypoxic area and less bone formation whereas oxygen tensions above 4% (ninit>4%) hamper the proliferation of chondrocytes, thereby disrupting the cartilage production and consequently the endochondral ossification process. Interestingly, in the intermediate range of oxygen tensions (0.5%<ninit<4%), lower initial oxygen tensions appear to result in more bone formation (Table S1, 0.7% versus 3.7% oxygen tension of the standard compromised condition). Although intuitively we would expect that these low oxygen tensions would lead to worse hypoxic conditions, model analyses show that the average oxygen tension in the fracture callus remains above 0.8% during the entire healing period (note that the low oxygen tensions of the central callus area are averaged with the high oxygen tensions near the bony ends), which is well above the oxygen threshold for chondrocyte and MSC cell death (i.e. 0.5%). As such the oxygen tension is low enough to inhibit extensive proliferation (as the chondrocytes and MSCs preferentially proliferate at 3% and 4% oxygen tension respectively, Figure 2) and therefore avoiding too much oxygen consumption, but high enough to keep a small amount of remaining stem cells alive. Moreover, the oxygen consumption is not only reduced due to the smaller amount of consuming cells. The cellular consumption of oxygen is also oxygen dependent, leading to a lower cellular consumption in low oxygen environments. It is the combination of these effects that limits the drop of the average oxygen tension, allowing the MSCs to survive and contribute to the bone healing process for a longer period of time (40 days for case ninit = 0.7% versus 4 days in the standard condition). A similar reasoning can be made for the case where an initial gradient of oxygen tensions was applied to the central callus region (ninit,gr = 0.8%/mm*x). In this simulation the oxygen tension varied from 0% in the middle of the callus to 4% at the bony ends. The low oxygen tensions in the central area supported the maintenance of a small population of MSCs for a longer period of time (6 days versus 4 days in the standard condition). This resulted in a larger amount of cartilage and finally bone. Note that this specific gradient in oxygen tension is less beneficial for the amount of bone formation at PFD 90 than a uniform distribution of 0.7%, as in the case of the gradient the oxygen tension in the middle of the callus is too low to sustain cell viability.
Besides investigating the influence of the initial conditions, the sensitivity analysis also focused on the complex interplay between oxygen delivery (Gn), diffusion (Dn) and consumption (Qb, Qc, Qm, Qf). Altering the oxygen delivery (Gn) by the vasculature has a large effect on the final amount of bone formation. Very low values of oxygen delivery increase cell death in the central hypoxic area, resulting in the absence of any bone formation. Increasing the value of oxygen delivery slightly improves the fracture healing outcome. Note that, although the bone tissue fraction is 37% in case of Gn = 22.10−12 mol/cell.day and 55% in case of Gn = 3.2.10−12 mol/cell.day, the spatial extent of bone ingrowth at PFD 90 is very similar (results not shown). This is however masked by the increased proliferation and matrix production of fibroblasts who thrive in the well-oxygenated environment created by Gn = 22.10−12 mol/cell.day. As such, the bone tissue fraction for Gn = 22.10−12 mol/cell.day is reduced with respect to Gn = 3.2.10−12 mol/cell.day. The parameter values of the cell-specific oxygen consumption rates (Qb, Qc, Qm, Qf) also influence the outcome of the model significantly. For all cell types, it is beneficial to reduce the oxygen consumption rates since this will limit the decrease in oxygen tension in the central fracture area and consequently the amount of cell death. This benefit is greatest for the MSCs and chondrocytes as these cell types mainly populate the central fracture area and contribute to the hypoxic conditions encountered here. Conversely, the amount of bone formation is greatly reduced when the oxygen consumption rate of the MSCs (Qm) or chondrocytes (Qc) is increased. The model outcome is also negatively affected by a high osteoblastic oxygen consumption rate (Qb) whereas a high fibroblastic consumption rate (Qf) only slightly reduces the final amount of bone. In the first case, the oxygen tension near the bony ends is reduced, resulting in hampered osteogenic differentiation and limited bone formation. In the latter case, the fibroblasts reduce the oxygen tension in the entire callus area (the fibroblasts are initially uniformly distributed in the fracture callus) but this drop is limited due to the small amount of fibroblasts present. Interestingly, a similar reasoning does not hold for the MSCs (although they are also initially uniformly distributed and limited in cell population) since they mainly grow in the central fracture zone whereas the fibroblasts optimally proliferate in a well-oxygenated environment such as the tissues surrounding the bony ends. As such, a high oxygen consumption rate of MSCs severely impairs the bone formation process whereas a high oxygen consumption rate of fibroblasts only slightly reduces the amount of bone formed at PFD 90.
It can be noticed from Table S1 and Figure S2 that the diffusion properties of oxygen have a major impact on the simulation outcome. Reducing the diffusion coefficient of oxygen impairs the bone formation due to the creation of a larger hypoxic zone (Figure S2-A,C). Increasing the diffusion coefficient appears to be beneficial although a closer look at these simulation results reveals that the endochondral process is not captured correctly anymore with bone formation largely preceding the ingrowth of new blood vessels (Figure S2-D,F). Note that also in this case a non-union is formed, since there is no cortical bridging, even though a bone tissue fraction of 89% is reached (Table S1). Increasing the diffusion coefficient even further results in a complete absence of bone formation since the resulting oxygen tensions are too low for any cell type to survive (Figure S2-H) (see Supporting Text S3).
In conclusion, we can state that the initial conditions have an important impact on the final amount of bone formation. They are however not sufficient to result in complete healing of critical size defects due to insufficient vascularization of the central callus area, leading to hypoxic conditions and cell death. As such, an adequate and timely restoration of the vasculature appears to be an important determinant of the healing outcome.
Inspired by the importance of a timely vascularization as well as by the limited biological potential of the fracture hematoma, we explored the influence of the host environment on the bone healing process in critical size defects. It appears that the fracture healing process is intimately linked to the surrounding muscle envelope since clinical evidence has found that open fractures with significant muscle injury complicate fracture healing and are a risk factor for the development of non-unions [5]–[7]. Moreover, tibial shaft fractures, which are only covered by a thin layer of soft tissue, are prone to a number of complications often resulting in additional surgical interventions [5], [9], [29]. There are a number of ways by which the skeletal muscle can contribute to the bone healing process. Firstly, experimental studies have shown that blood vessels originating in the overlying muscle contribute to the vascularization of the fracture callus [30], [31]. Secondly, muscle cells are a source of growth factors (e.g. FGF-2, TGF-β) [32] as well as progenitor cells [33]–[35]. Thirdly, the muscle envelope might provide the adequate biomechanical stimuli required for successful bone healing [36], [37].
In order to further unravel the potential mechanisms of interaction that exist between the bone regeneration process and the overlying skeletal muscle, the role of the skeletal muscle as a source for vascularization, progenitor cells and growth factors or a combination thereof was investigated by applying different boundary conditions to the in silico model (Figure 7–8). More specifically, the contribution of the muscle to the vascularization of the fracture callus was simulated by initializing additional endothelial cells on the border of the central callus area with the muscle, either partially or fully covering the fracture gap. The influence of the muscle as a source of MSCs or growth factors was represented by a Dirichlet boundary condition, applied to the upper border for the entire duration of the simulation (i.e. 90 days) and fully covering the fracture gap (Figure 7). The value of the Dirichlet boundary conditions is equal to the ones applied in the standard case, i.e. 2.104 cells/ml for the MSCs and 2 µg/ml for the osteochondrogenic growth factors [25], [38]–[40]. Since the mathematical model does not take into account any mechanoregulatory stimuli, the influence of mechanoregulatory stimuli generated by the overlying muscle on the bone formation processes cannot be evaluated in this study.
The results, summarized in Figure 8, underline the importance of the host environment for successful fracture healing since all the investigated conditions improve the amount of bone formation with respect to the standard condition or result even in bridging of the critical size defect. Note that the host environment is also more efficient in stimulating the bone regeneration process than the initial conditions tested in Table S1, since the host environment continuously provides the fracture callus with fresh growth factors, cells and blood vessels (or a combination thereof) whereas the initial conditions represent only a single (initial) contribution to the bone regeneration process. In order to limit the length of the paper, we will touch upon the most important findings of Figure 8 and refer the reader to Supporting Text S4 for an in depth discussion of the results. Both the contribution of the muscle as a source of vascularization (case A) as well as of osteoprogenitor cells (case D) results in the formation of a union, whereas a partial supply of blood vessels from the host environment (case B) or muscle derived release of osteochondrogenic growth factors fails to result in a complete bridging of the defect (Figure S3). In most cases, except for cases E and G, the combination of two or more boundary conditions enhances the bone formation process. Indeed, the combined delivery of cells and growth factors results in less bone formation (case E) than the delivery of cells alone (case D). Figure 8 also shows that without vascular ingrowth from the muscular environment, the delivery of cells results in the largest amount of bone formation (case C versus case D). However, if the fracture callus is fully or partially vascularized by the overlying muscle, the delivery of growth factors is more beneficial than the delivery of cells for the final healing outcome (case F versus G, case I versus J).
In conclusion, we can state that the contribution of the host environment, and more specifically its role as a source of vascularization is critical for successful bone healing. Interestingly, the results indicate that the lack of adequate vascularization can be rescued by a continuous delivery of osteoprogenitor cells, potentially in combination with osteochondrogenic growth factors.
Intrigued by the results of the previous section, we wondered if the lack of adequate vascularization could also be rescued by a single contribution of (more optimal) initial conditions. Or, from another perspective, whether the initial conditions that were insufficient to result in successful bone healing in a compromised environment (Table S1), would be able to stimulate the bone regeneration process more in a permissive host environment. In order to answer this question, we use the model in which the fracture callus is partially supplied by blood vessels from the overlying muscle (case B, Figure S3) since this environment is not as compromised as the standard compromised condition (Table S1) but nevertheless results in the formation of a non-union without additional cells or growth factors (Figure 8, case B). We tested three potential treatment strategies: the injection of growth factors, the injection of cells and the injection of a combination product. All the injections take place at day zero, making them initial conditions (Figure 9).
According to the results of Figure 9, all the treatment strategies yield at least the same (within 2% of intrinsic variability) or more bone formation than a non-treated fracture in a permissive environment. Moreover, the permissive environment is clearly beneficial since the amount of bone formation is increased with respect to the compromised environment for all the treatment conditions (Table S1).
The injection of precursor cells does not significantly improve the bone healing outcome since the vascularization of the central callus area is still delayed, resulting in hypoxic conditions and cell death (Figure 11). The injection of osteochondrogenic growth factors is able to heal the critical size defect surrounded by a muscular envelope that partially contributes to its vascularization if the concentration is sufficiently high (Figures 9–10–11). The mechanism of action underlying this result can be explained as follows. The large initial concentration of growth factors will lead to the differentiation of the osteoprogenitor cells into chondrocytes, which consume less oxygen than MSCs. Consequently, the pool of oxygen-consuming MSCs is reduced thereby limiting the oxygen consumption. As such, a large initial concentration of growth factors makes the hypoxic area shrink, finally leading to the successful healing of the critical size bone defect (Figure 11).
The injection of the combination product has improved the amount of bone formation but is not as beneficial as osteochondrogenic growth factor injections alone (Figures 9–10–11). Indeed, the growth factor concentration is not high enough to commit the entire population of MSCs to the chondrogenic lineage. As such, a small amount of MSCs remains undifferentiated and can continue to proliferate and consume oxygen. Since MSCs consume more oxygen than chondrocytes, the remaining MSC pool increases the drop in oxygen tension and consequently cell death. As a result, the amount of bone formation is lower than in the case of growth factor treatment alone.
The predictions of the mathematical model are compared with the results of the in vivo set-up where the influence of BMP-2 overexpressing periosteal cells on bone formation in large defects was explored. Defects, treated with a collagen scaffold containing mPDCs, thereby mimicking the initial conditions of the computational model, show little bone formation (Figure 12-B), which is also predicted by the mathematical model (Figure 9, standard permissive condition). Interestingly, while defects treated with mPDCs show no presence of bone or cartilage in the center of the defect (Figure 12-C1), large amounts of bone and the presence of cartilage (arrow) and bone marrow are noted in the defects treated with BMP-2 overexpressing mPDCs (Figure 12-C2, C3). Clearly, the presence of BMP-2 enhances the bone formation process which results in a clinical union. Similarly, the computational model predicts that the injection of only growth factors is sufficient to heal a large defect in a permissive environment (Figure 9, gbc,init). Note, however, that in the experimental set-up BMP-2 overexpressing cells are implanted whereas computationally an initial bolus injection of growth factors is simulated.
In the experimental model, the sites of bone formation are closely associated with numerous large blood vessels (indicated in dark brown by the CD31 staining, Figure 12-C5, C6), in contrast to the small blood vessels observed in the center of the defects treated with mPDCs only (Figure 12-C4). In the mathematical model, the blood vessel formation is also closely connected to the bone formation process (Figure 10). In the central callus area we hypothesize that the small blood vessels observed in vivo arise through vasculogenesis, which is not accounted for in the mathematical model. However, since these small blood vessels appear to be immature and not associated with bone formation, the mathematical model does predict the correct tissue distribution in the central callus area even in the predicted absence of small blood vessels. As expected, no blood vessels are observed at the site of cartilage formation (Figure 12-C5, arrow).
While the influence of the amount of seeded cells alone or in combination with BMP-2 overexpression was not explored experimentally, the computational simulations predicted an improvement in the amount of bone formation but not a complete healing of a large bone defect (Figure 9, cm,init and cm,init/gbc,init).
Note that in a compromised environment, a large defect will develop into a non-union, irrespective of a growth factor treatment (Table S1, standard compromised condition), whereas in an environment with full vascular ingrowth from the overlying muscle, a union will develop, irrespective of a growth factor treatment (Figure 8, case A). As such, since the computational predictions of the growth factor treatment in a permissive environment reproduce the in vivo observations correctly, one may speculate that the muscle overlying the large defect in the in vivo set-up partially contributed to the vascularization of the fracture callus and consequently the bone healing process. Further characterization of the origin of the vasculature growing towards the defect area would be required to confirm this.
From the results discussed above, we can conclude that a single injection of osteochondrogenic growth factors is able to compensate for the lack of adequate vascularization. Since a single injection of cells fails to promote complete bridging of the critical size defect in a permissive environment, a sequence of cellular injections might be more appropriate strategy.
Finally, we used the in silico model to optimize the treatment strategy of the previous section for critical size defects surrounded by a compromised host environment. As can be concluded from Table S1, the lack of muscular contribution to the vascularization of the fracture callus as well as of osteoprogenitor cells or growth factors, greatly hampers the bone regeneration process and results in the formation of a non-union. Furthermore, the initial conditions can be tuned to improve the amount of bone formation but are insufficient to provide complete healing of the critical size defect (Table S1). This was attributed to the delayed vascularization of the central callus area, leading to hypoxia and cell death. In order to improve the limited biological potential of the fracture callus and host environment, additional progenitor cells or growth factors can be injected in the fracture callus. However, cellular strategies would miss their therapeutic target if injections would take place at day 0, since cell survival would be very limited in these challenging hypoxic conditions. Therefore, we investigated whether a single injection of MSCs, osteochondrogenic growth factors or a combination thereof at a later time point would improve the bone healing outcome, as in this way the blood vessel network will have restored at least partially (Figure 13).
As can be seen in Figure 13, the injection of osteochondrogenic growth factors does not improve the bone healing outcome, except at PFD 0. This can be attributed to the increased chondrogenic differentiation and consequently limited oxygen consumption, as was discussed previously. At the other time points, the delay in vascularization of the central callus area results in hypoxia and cell death. Consequently, the injection of additional growth factors is to no avail since there are no cells present on which they can exert their influence. Interestingly, the time at which the MSCs or the combination product was injected, appears to be a critical determinant for the final amount of bone formation. If the cellular treatment is administered before PFD 35, the amount of bone is reduced (compared to no treatment) since the additional cells increase the oxygen consumption thereby worsening the hypoxic conditions in the central callus area. One can notice a further decrease of the effectiveness of the cellular treatment (cells only as well as the combination with growth factors) for injections at PFD 7 and 14. This can be related to the oxygen tension encountered in the central callus area at the time of injection, and the fact that this oxygen tension evolves with time. More specifically, at PFD 0 the oxygen tension has dropped only slightly and at PFD 28 the vasculature is already growing into the fracture callus so that in both cases the oxygen tension in the central callus area is able to support the injected cells. At intermediate time points, however, the oxygen tension is too low to support the injected cells, explaining why injections at PFD 7 and 14 are the least effective. If cells are administered at PFD 35, the delay of 35 days between the occurrence of the fracture and the start of the cellular therapy allows for a partial restoration of the blood vessel network, which seems to be optimal for the injection of cells only. The effectiveness of the combination product, however, continues to increase when the treatment is further postponed (up to day 56). The non-linearities in the predicted bone tissue fractions as a function of time of administration, as well as the discrepancy in optimal timing between the cellular and combination treatment can again be explained by the evolving oxygen tension of the central callus area which gradually increases as a function of time through a combination of oxygen release from the active vasculature and passive diffusion. More specifically, the average oxygen tension in the central callus area of a large non-treated defect surrounded by a compromised environment increases from 2.2% at 35 days, to 3% at 42 days, 3.9% at 49 days, 5.6% at 56 days and 6.2% at 63 days. At PFD 35 both the oxygen tension as well as the osteochondrogenic growth factor concentration are low in the central callus area so that only limited chondrogenic differentiation occurs upon injection of MSCs. However, the low oxygen tension inhibits extensive cellular proliferation (avoiding too much oxygen consumption), resulting in a small amount of “quiescent” stem cells (similar to ninit = 0.7%, Table S1). When the oxygen tension in the central callus area subsequently increases to 3%, the remaining MSCs differentiate to chondrocytes and contribute to the bone regeneration process. As such, there are two bursts of chondrogenic differentiation which results in an increased amount of bone formation. If the cellular administration occurs after PFD 35 the increased oxygen tension will enhance the chondrogenic differentiation, thereby reducing or even eliminating the pool of “quiescent” stem cells. As such the proliferation and survival of the newly formed chondrocytes will mainly determine the extent of the bone formation. Since the average oxygen tension in the central callus area is higher at PFD 56 and 63 than at PFD 49, more bone will be formed for cellular injections in the former two cases, compared to PFD 49.
By combining osteochondrogenic growth factors with stem cells, all the injected MSCs will directly become chondrocytes thereby depleting the pool of osteoprogenitor cells, irrespective of the starting time of the treatment. Similar to the cellular treatments started after PFD 35 the proliferation and survival of the chondrocytes will mainly determine the extent of the bone formation. It appears that the average oxygen tension at PFD 56 results in an optimal proliferation and survival of the chondrocytes and hence subsequent endochondral bone formation. At PFD 63 the average oxygen tension becomes too high for optimal chondrocytic proliferation thereby reducing the amount of bone formation.
According to the model results, cellular injections are only effective if delayed until a specific time point (i.e. day 35, Figure 13) in order to allow for a partial restoration of the blood vessel network. Note that although the cellular as well as the combination treatment lead to an increased amount of bone compared to no treatment (provided the injection is sufficiently delayed), they nevertheless result in the formation of a non-union (with a maximal amount of bone at 90 days up to 80% and 85% respectively). As such, a single injection is insufficient and future research should focus on an optimal sequence of injections in order to heal critical sized defects in a compromised environments.
This study has used an integrative in vivo - in silico approach to investigate the occurrence of oligotrophic and atrophic non-unions as well as to design possible treatment strategies thereof. An extensive sensitivity analysis was performed in order to study the complex interplay of blood vessel formation, oxygen supply, growth factors and (osteoprogenitor) cells on the final healing outcome in large bone defects. The results of the sensitivity analysis indicated that the initial conditions (osteochondrogenic growth factor, MSCs, oxygen) are necessary for the bone regeneration process but not sufficient for complete bone healing of a critical size defect (5 mm). They do, however, have an important impact on the final amount of bone formation. Interestingly, simulation results of the same oxygen model in a small defect (0.5 mm) were found to be robust to changes in the initial conditions [23].
Although the performed sensitivity analysis yields interesting results, the interpretation thereof should be done carefully due to a number of reasons. For example, the sensitivity analysis that was performed in this study (Table S1) used a one-at-a-time (OAT) design, where the effect of one factor is assessed by varying the value of only that factor and keeping all other factors fixed. The main disadvantage of this simple method is its inability to capture interactions between factors. A simple combination of the ‘optimal’ values of initial conditions (see Table S1: the value that for an OAT design yielded the most bone at PFD 90) indicates for instance that a more adequate design is necessary to unravel these (non-linear) interactions between the different parameters of the oxygen model. Indeed, combining the ‘optimal’ initial conditions of MSCs (cm,init), fibroblasts (cf,init), osteochondrogenic growth factors (gbc,init) and oxygen (ninit) results in 38% of bone after 90 days which is less than for the respective ‘optimal’ initial conditions alone. The conclusions of the sensitivity analysis are also only valid for this specific set of parameter values since, for example, the optimal initial oxygen tension will vary depending on the initial stem cell concentration. Despite its limitations, the OAT-design already indicates some interesting non-linear responses of the model with respect to the initial MSC cell density and the oxygen tension as well as their interactions. Future work should focus on more complex designs, including latin hypercube design and uniform design [41], to calculate quantitative metrics of sensitivity and study these non-linearities and (higher-order) parameter interactions further in order to unravel the underlying mechanisms and define new research hypotheses.
The dynamics of all cellular variables (apart from the endothelial cells) is described by means of continuum equations, meaning amongst others that cell proliferation was captured by means of a logistic growth equation. While this equation accounts for a maximal cell density in the callus area, it does not allow to specify an upper limit to the number of division cycles a cell (such as an MSC) can undergo before senescence. Because of this upper limit, in reality the amount of cells that can be obtained through division is dependent on the original pool size whereas in the mathematical model, a single cell can theoretically divide until the entire callus reaches maximal cell density. The main consequence of this limitation is that our predictions might be too optimistic in that fracture healing might be even more challenging in reality, because a sufficient number of cells (such as MSCs) cannot be reached to heal the fracture. In the future we will try to implement a description that allows to account for a limited number of population doublings, potentially through the extension of the agent-based description of endothelial cells to the skeletal cell types.
Even though the predictions of the current model might be too optimistic, all of the conditions explored in Table S1 nevertheless resulted in the formation of a non-union. Indeed, the simulation predicts that a murine bone defect becomes critical at 3 mm (Figure 6) which corresponds to the experimental observation of Zwingenberger et al. [42]. They report the creation of a persisting femoral bone defect in nude mice when the defect size is 3 mm [42]. The predicted value is also in the same range as other mouse femoral critical defect sizes reported in the literature: 2 mm [43], 3.5 mm [44] and 4 mm [45]. As such, the computational framework is able to model the occurrence of non-unions and can be used to design several treatment strategies depending on the host environment. In our model, a single initial (i.e. at PFD 0) injection of osteochondrogenic growth factors at sufficiently high concentration (gbc,init = 1 µg/ml) directly into a callus surrounded by a permissive environment resulted in complete healing of the critical size defect (Figure 9). The beneficial effect of growth factor delivery was also confirmed by the study of Patel et al. [46]. They report that the BMP-2 release from gelatin microparticles incorporated within the pores of a scaffold that was implanted within a 8 mm rat cranial critical defect resulted in significantly higher bone formation after 12 weeks, i.e. 37.4±18.8% (test) versus 7.8±7.1% (control) bone volume respectively. Similar conclusions were made by Willett et al. who studied the influence of recombinant human BMP-2 (rhBMP-2) delivery on tissue regeneration in a murine composite injury model [47]. The in vivo composite injury model consisted of a critically sized femoral bone defect and an adjacent volumetric muscle injury in the quadriceps (both 8 mm) [47]. They have shown that treated bone defects without volumetric muscle loss were consistently bridged whereas the treatment failed to promote the regeneration process in the challenging composite injury [47]. Although care must be taken when directly comparing these findings to our in silico results (since the exact role of the muscle in the in vivo setting of Willett et al. was not characterized), they do predict the same trends. Indeed, the multiscale model predicts a successful healing in the case of growth factor administration to a critical sized defect that is fully or partially supplied by blood vessels from the overlying muscle (Figure 9). In contrast, in a compromised environment where the role of the muscle as a source of vascularization is lacking, additional injections of growth factors, either at PFD 0 (Table S1) or at later time points (Figure 13) do not induce bony bridging of the large bone defect.
In large bone defects not only the initial concentration of growth factors but also the initial amount of osteoprogenitor cells might be reduced [26], [27]. Consequently, the use of stem cells for the treatment of critical size defects is actively being pursued [48]. The injection of MSCs in the callus area elicited an improved healing response (although without reaching full bridging) in silico if the environment is sufficiently vascularized to sustain the cell viability, which according to the model meant that injections were only effective if delayed until a certain time point (day 35 according to Figure 13). Similar conclusions were drawn by Geris et al. who investigated the occurrence of bone atrophic non-unions by an integrative approach [49]. Based on the recovery of the blood supply to the interfragmentary gap, they predicted with an in silico model that the injection of MSCs at three weeks post-osteotomy would prevent the onset of an atrophic non-union which was also confirmed by experimental results [49]. The necessity of vascularization for successful healing of challenging critical size defects is also substantiated by the results of Table S1 (no contribution of the overlying muscle to the vasculature) and Figure 9 (partial contribution of the overlying muscle to the vasculature), where an initial injection (i.e. at PFD 0) of additional cells in a defect that is insufficiently vascularized does not significantly improve the bone formation outcome. As such, the mathematical model retrieves the beneficial effect of cellular injections in some cases, similar to the experimental observations reported in literature [50], [51], although the effectiveness is strongly dependent on the available vasculature.
Interestingly, the model results indicate that the effectiveness of a therapy (consisting of the injection of cells, growth factors or a combination thereof) is dependent on the timing of the treatment as well as the host environment. The former effect is strongly related to the biological potential of the fracture callus at the time the treatment is applied, while the latter potentially constitutes a source of additional osteoprogenitor cells, growth factors or vascularization. For example, growth factor injections at PFD 0 or at later time points in a compromised host environment lead to only 63% and 52% of bone respectively whereas growth factor injections at PFD 0 in a permissive environment result in the formation of a union. In all three cases the main cause underlying the formation of a non-union in a large defect (without treatment) is the increased cell death in the central (hypoxic) callus area. Since growth factor injections at PFD 0 result in increased chondrogenic differentiation, which in turn limits the oxygen consumption and the decrease of oxygen tension (severity of hypoxia), this treatment increases the amount of bone formation. Note that nevertheless a hypoxic area arises which results in the formation of a non-union. Consequently, a permissive environment that provides additional vascular ingrowth, improves the bone formation outcome even further. Growth factor injections at later time points in a compromised environment are, however, to no avail since there are no cells left in the central callus area.
In summary, we can state that a treatment will be most beneficial if it tackles the underlying mechanism of action causing the hampered bone formation. Although this statement seems a logical and intuitive design rule, the underlying mechanisms of actions are a result of the complex non-linear, oxygen-dependent dynamics of blood vessel formation, oxygen supply, angiogenic growth factor production, cell differentiation, cell proliferation and oxygen consumption. The fact that many cellular processes, like survival, proliferation and differentiation are (non-linearly) dependent on oxygen tension and that they all have a specific range of oxygen tension at which they are ‘optimized’ (maximally affected) (Figures 1–2), makes it virtually impossible to intuitively predict the resulting bone healing outcomes. Instead, it requires a rigorous computational modelling of the governing mechanisms and dependencies (Figure 14).
Taken all the results together, we can conclude that complete cortical bridging of a challenging critical size defect will only occur if growth factors, osteoprogenitor cells and vasculature are present at the same time and place (Figure 14). Indeed, the blood vessels will supply the necessary oxygen to ensure cellular survival whereas the growth factors will promote the correct differentiation cascade finally resulting in the continuation and successful completion of the bone regeneration process. Consequently, the most stringent factor that is lacking in a certain area or at a certain time point will be an ideal candidate for potential treatment strategies. For example, bone tissue engineering treatments where a scaffold seeded with cells and osteochondrogenic growth factors is implanted in a bone defect, should focus on a timely vascularization in order to ensure the survival of the implanted cells. Potential strategies of vascularization include the induction of a Masquelet-membrane [52], [53], the delivery of angiogenic growth factors [46] as well as the in vitro creation of a pre-vascularized construct by co-culture of osteoprogenitor cells with endothelial cells [54]. Encouraging results were for example obtained by Patel et al. who showed that the dual release of vascular endothelial growth factor (VEGF) and bone morphogenetic protein-2 (BMP-2) in a 8 mm rat cranial critical size defect enhanced the bone formation at 4 weeks, suggesting a synergistic effect of these growth factors during early bone regeneration [46]. Note that besides the biological stimuli also mechanoregulatory stimuli influence the bone formation process [36], [37]. The current multiscale model does not take this into account, meaning amongst others one assumes that the fracture is sufficiently stabilized through external or internal fixation such that excessive loading will not play a role in the formation of a non-union (Supporting Text S2).
In conclusion, the multiscale oxygen model was able to capture the essential aspects of in vivo atrophic and oligotrophic non-unions. Interestingly, thorough model analyses assisted in understanding the underlying mechanisms of action, i.e. the delayed vascularization of the central callus region resulted in harsh hypoxic conditions, cell death and finally disrupted bone healing. Since a timely vascularization was found to be critical for the successful healing of large bone defects, the oxygen model was used to design and test potential treatment strategies for both permissive and compromised host environments. A qualitative correspondence between the predicted outcomes of certain treatment strategies and experimental observations was obtained, clearly illustrating the model's potential. Furthermore, the results of this study demonstrate that due to the complex non-linear, oxygen-dependent dynamics of blood vessel formation, oxygen supply, angiogenic growth factor production, cell differentiation, cell proliferation and oxygen consumption, it becomes virtually impossible to determine the effectiveness of a treatment strategy intuitively thereby underlining the importance computational modelling tools. Moreover, the model predictions also showed that the effectiveness of a therapy is strongly influenced by the host environment since it can serve as a source of additional osteoprogenitor cells, growth factors or vascularization to populate the fracture callus and increase the biological potential thereof. Consequently, future research should focus on extensive experimental characterization as well as computational modelling of the host environment and its interaction with potential treatment strategies.
|
10.1371/journal.pcbi.1006453 | Telescope: Characterization of the retrotranscriptome by accurate estimation of transposable element expression | Characterization of Human Endogenous Retrovirus (HERV) expression within the transcriptomic landscape using RNA-seq is complicated by uncertainty in fragment assignment because of sequence similarity. We present Telescope, a computational software tool that provides accurate estimation of transposable element expression (retrotranscriptome) resolved to specific genomic locations. Telescope directly addresses uncertainty in fragment assignment by reassigning ambiguously mapped fragments to the most probable source transcript as determined within a Bayesian statistical model. We demonstrate the utility of our approach through single locus analysis of HERV expression in 13 ENCODE cell types. When examined at this resolution, we find that the magnitude and breadth of the retrotranscriptome can be vastly different among cell types. Furthermore, our approach is robust to differences in sequencing technology, and demonstrates that the retrotranscriptome has potential to be used for cell type identification. We compared our tool with other approaches for quantifying TE expression, and found that Telescope has the greatest resolution, as it estimates expression at specific TE insertions rather than at the TE subfamily level. Telescope performs highly accurate quantification of the retrotranscriptomic landscape in RNA-seq experiments, revealing a differential complexity in the transposable element biology of complex systems not previously observed. Telescope is available at https://github.com/mlbendall/telescope.
| Almost half of the human genome is composed of Transposable elements (TEs), but their contribution to the transcriptome, their cell-type specific expression patterns, and their role in disease remains poorly understood. Recent studies have found many elements to be actively expressed and involved in key cellular processes. For example, human endogenous retroviruses (HERVs) are reported to be involved in human embryonic stem cell differentiation. Discovering which exact HERVs are differentially expressed in RNA-seq data would be a major advance in understanding such processes. However, because HERVs have a high level of sequence similarity it is hard to identify which exact HERV is differentially expressed. To solve this problem, we developed a computer program which addressed uncertainty in fragment assignment by reassigning ambiguously mapped fragments to the most probable source transcript as determined within a Bayesian statistical model. We call this program, “Telescope”. We then used Telescope to identify HERV expression in 13 well-studied cell types from the ENCODE consortium and found that different cell types could be characterized by enrichment for different HERV families, and for locus specific expression. We also showed that Telescope performed better than other methods currently used to determine TE expression. The use of this computational tool to examine new and existing RNA-seq data sets may lead to new understanding of the roles of TEs in health and disease.
| Transposable elements (TEs) represent the largest class of biochemically functional DNA elements in mammalian genomes[1,2] comprising nearly 50% of the human genome. As many of these transcriptionally active elements originated as retroelements, we refer to the set of RNA molecules transcribed from these elements in a population of cells as the retrotranscriptome. The contribution of the retrotranscriptome to the total transcriptome, cell-type specific expression patterns, and the role of retroelement transcripts in disease remain poorly understood[3]. Although most TEs are hypothesized to be transcriptionally silent (due to accumulated mutations), recent studies have found many elements to be actively expressed and involved in key cellular processes. For example, aberrant expression of LINE-1 (L1) elements, the most expansive group of TEs, has been implicated in the pathogenesis of cancer[4–7], while human endogenous retroviruses (HERVs) are reported to be involved in human embryonic stem cell differentiation[8,9] and in the pathogenesis of amyotrophic lateral sclerosis[10]. We, and others, have shown that HIV-1 infection increases HERV transcription[11–15]. These lines of evidence therefore indicate that TEs have important roles in the regulation of human health and disease.
The ability to observe and quantify TE expression, especially the specific genomic locations of active elements, is crucial for understanding the molecular basis underlying a wide range of conditions and diseases[16]. Traditional techniques for interrogating the TE transcriptome include quantitative PCR[17,18] and RNA expression microarrays[19–23]. However, these techniques are unable to discover elements not specifically targeted by the assay, and may fail to detect rare, previously unknown, or weakly expressed transcripts. High-throughput RNA sequencing (RNA-seq) promises to overcome many of these shortcomings, enabling highly sensitive detection of transcripts across a wide dynamic range. Mathematical and computational approaches for transcriptome quantification using RNA-seq are well established[24,25] (reviewed by Garber et al.[26]) and provide researchers with reproducible analytical pipelines[27,28]. Such approaches are highly effective at quantifying transcripts when sequenced fragments can be uniquely aligned to the reference genome, since the original genomic template for each transcript can be unambiguously identified[29,30]. In contrast, sequencing fragments that originate from repetitive sequences often have high scoring alignments to many genomic locations, leading to uncertainty in fragment mapping and the derived transcript counts. Issues arising from these “multimapping” or “ambiguous” fragments are well known and are often addressed by masking repetitive sequences or otherwise discarding ambiguous fragments[31–33]. The disadvantage of ignoring repeats is that interesting biological phenomena, including those involving TEs, are missed[31]. Several approaches have been proposed that account for read mapping uncertainty using statistical models. The most common approach, described by Li et al.[34,35], involves modeling read assignments using a mixture model, with expression levels as mixture weights and fragment assignments as latent variables; model parameters are then estimated using an expectation-maximization algorithm. Several variations on this model have been proposed, such as modeling read counts instead of individual reads (MMSEQ[36]) or using Markov chain Monte Carlo (MCMC) to sample model parameters (BitSeq[37]). A few approaches deviate from the mixture model approach; notably, MMR instead evaluates alignments by minimizing a loss function[38]. To our knowledge, none of these packages have been adapted specifically for quantifying TE expression.
A growing field of study is now interested in using high-throughput sequencing to characterize the retrotranscriptome[8,9,39–41]. Instead of considering repetitive sequences as a source of noise that interferes with gene expression analysis, the TEs themselves are the features of interest. Three general approaches are used to deal with challenges of aligning short sequencing reads to repetitive elements. i) “Family-level” approaches combine read counts across multiple instances of a TE subfamily, since fragments mapping to multiple genomic locations can often be uniquely assigned to a single repeat subfamily. This approach provides valuable information about which TE subfamilies may be differentially regulated, but lacks the resolution needed to identify specific expressed elements. ii) “Heuristic” approaches simplify the problem of multi-mapped fragments by examining alignments and using filtering criteria to resolve ambiguity. Examples of heuristic approaches include discarding ambiguous reads (unique counts), randomly assigning ambiguous reads to one of its best scoring alignments (best counts), or dividing counts among possible alignments (fractional counts). Finally, iii) “statistical” approaches implement a statistical model that estimates the most probable assignment of fragments given underlying assumptions about the generating process and the observed data.
Several existing software packages have implemented these approaches specifically for TE quantification. RepEnrich[42,43] maps reads to “pseudogenomes” composed of multiple loci belonging to the same subfamily, then uses a fractional counts heuristic to resolve any remaining ambiguous fragments. TEtranscripts[44] and SalmonTE[45] are both statistical approaches that use mixture models estimated by expectation-maximization. The main difference between these approaches is that TEtranscripts begins with genome alignment, while SalmonTE adapts the Salmon[46] approach of quasi-alignment to transcriptome sequences. Like MMSEQ, SalmonTE also uses equivalence classes to reduce the effort needed for parameter optimization. By default, all three TE quantification approaches summarize estimates by subfamily.
Here, we introduce Telescope, a tool which provides accurate estimation of TE expression resolved to specific genomic locations. Our approach directly addresses uncertainty in fragment assignment by reassigning ambiguously mapped fragments to the most probable source transcript as determined within a Bayesian statistical model. We implement our approach using a descriptive statistical model of the RNA-seq process and use an iterative algorithm to optimize model parameters. We use Telescope to investigate the expression of HERVs in cell types from the ENCODE consortium.
Resolution of transposable element (including those of human endogenous retroviruses, HERVs) expression from RNA-seq data sets has been complicated by the many similarities of these repetitive elements. Telescope is a computational pipeline program that solves the problem of ambiguously aligned fragments by assigning each sequenced fragment to its most likely transcript of origin. We assume that the number of fragments generated by a transcript is proportional to the amount of transcript present in the sample; thus, the most likely source template for a randomly selected fragment is a function of its alignment uncertainty and the relative transcript abundances. Telescope describes this relationship using a Bayesian mixture model where the estimated parameters include the relative transcript abundances and the latent variables define the possible source templates for each fragment[47].
The first step in this approach is to independently align each fragment to the reference genome; the alignment method should search for multiple valid alignments for each fragment and report all alignments that meet or exceed a minimum score threshold (Fig 1A). Next, alignments are tested for overlap with known TE transcripts; transcript assignments for each fragment are weighted by the score of the corresponding alignment (Fig 1B and 1C). In our test cases, we typically find that less than 50% of the fragments aligning to TEs can be uniquely assigned to a single genomic location and many fragments have more than 20 possible originating transcripts.
Telescope uses a Bayesian mixture model to represent transcript proportions and unobserved source templates and estimates model parameters using an expectation-maximization algorithm. In the expectation step (E-step), the expected value of the source template for each fragment is calculated under current estimates of transcript abundance (Fig 1D). The maximization step (M-step) finds maximum a posteriori estimates of the transcript abundance dependent on the expected values from the E-step (Fig 1E). These steps are repeated until parameter estimates converge (Fig 1D and 1E). Telescope reports the proportion of fragments generated by each transcript and the expected transcript of origin for each fragment (Fig 1F). The final counts estimated by Telescope correspond to actual observations of sequenced fragments and are suitable for normalization and differential analysis by a variety of methods. The software also provides an updated alignment with final fragment assignments that can be examined using common genome visualization tools. Telescope is available at https://github.com/mlbendall/telescope.
The core statistical model implemented in Telescope is based on the read reassignment model described by Francis et al.[47] and is similar to existing models for resolving mapping uncertainty[34,35,44,45]. Three main differences distinguish our model from existing models. First, our model includes a reassignment parameter, theta, that is absent in other models. This parameter effectively penalizes ambiguous alignments and may be important in cases where many highly similar transcripts are present. Second, our model includes an additional mixture component for fragments that map outside of the known transcriptome, accounting for missing transcripts in the annotation. Finally, our model does not use equivalence classes; reassignment occurs at the fragment level.
To demonstrate that our algorithm can truly resolve repetitive element expression to precise genomic locations, we generated sequencing fragments from a single genomic locus in silico and used Telescope to resolve alignment ambiguity and quantify expression. The locus selected was HML2_1q22 (HERV-K102), an HML-2 provirus that is highly similar to several other HML-2 loci[48] and should thus generate many ambiguously mapping fragments. All of the simulated fragments aligned to multiple genomic locations, and most of these (68.4%) had multiple distinct alignments sharing the same “best” alignment score (S1 Fig). Fragments mapped to 71 different HERV proviruses, including 58 HML-2 loci. After using our model to identify the most probable source locus for each fragment, we found that all fragments could be confidently assigned to HML2_1q22 with greater than 99% posterior probability (S1 Fig). This is possible because our model effectively reweights ambiguous alignments by borrowing strength from nearby alignments that are unique or high-scoring. In this case, there were no uniquely aligned fragments within HML2_1q22, but many fragments had best-scoring alignments to this locus. This result demonstrates that our approach can accurately reassign ambiguously mapping fragments and thus enables accurate expression quantification at single-locus resolution.
To investigate HERV expression in a robust way across a diverse platform of cell types we relied on publicly available RNA-seq data. The ENCODE data project is an invaluable source of genomic data from disparate sources and provides the opportunity to mine the transposable element expression in a setting of maximum genomic information. We profiled 13 human cell types, including common lines designated by the ENCODE consortium, as well as primary cell types, and applied our approach to determine HERV expression across the spectrum of human cell types, including normal or transformed, and contrasting cell lines with primary cells (Table 1, S1 Table).
Over 2.7 billion sequenced fragments aligned to human reference hg38 with between 23.6% and 46.1% of the fragments in each sample aligning ambiguously to multiple genomic locations. Telescope intersected the aligned fragments with a set of 14,968 manually curated HERV loci belonging to 60 families (see methods) and identified over 27 million fragments that appear to originate from HERV proviruses. Most (80.1%) of these fragments aligned to multiple genomic locations; we used Telescope to reassign ambiguous fragments to the most likely transcript of origin and estimate expression at specific HERV loci.
We developed genome-wide maps of HERV expression for 8 of the analyzed cell types that had replicates (Table 1, S1 Table), and used CIRCOS[49] to visualize the data (Fig 2). The outer track is a bar chart showing the number of HERV loci in 10 Mbp windows, with the red part of the bar representing the number of loci that are expressed in one or more cell types. The 8 inner rings show the expression levels (log2 counts per million (CPM)) of 1365 HERV loci that were expressed at least one of the cell types examined. Moving from the outer ring to the inner ring are replicates for each of the 8 cell types with replicates: H1-hESC, GM12878, K562, HeLa-S3, HepG2, HUVEC, MCF-7, and NHEK.
We found 1365 HERV loci that were expressed in at least one of the cell types (CPM > 0.5). Not all HERVs were expressed in all cell types, some were widely expressed in all cells, whereas others were only expressed in one or more cell type (Fig 2). There is also a spectrum of differential HERV expression, with some HERVs having significantly higher expression than others. Visual inspection of HERV expression maps suggest that there are certain regions of the genome that have minimal HERV expression, while other regions appear dense in HERV expression (Fig 2). The genomic context of HERV expression can also be inspected more closely in areas of interest, i.e. chromosome 19 (S2 Fig) and chromosome 6 (S3 Fig).
To ascertain global, subfamily and locus level specific HERV expression, we assessed the number of HERVs expressed in each cell type. All cell types expressed HERVs; the number of expressed loci ranged from 216 (in MCF-7), to 533 (H1-hESC) (Fig 3A). The number and proportion of cell type specific locations (expressed in only one cell) differed among cell types. Nearly half (46.3%) of locations expressed in H1-hESC were not expressed in any other cell type, while 89.3% of locations expressed in MCF-7 were also present in other cell types (Fig 3A). This suggests that regulatory networks are shared among some cell types but not others. We next examined the relative contribution of HERV families to overall HERV transcription and found that different cell types could be characterized by enrichment for different HERV families. For example, HERVH accounted for 91.8% of the transcriptomic output in H1-hESC cells, while HERVE was dominant in K562 cells (24.4%) (Fig 4A). Other families, such as HERVL, were evenly distributed across cell types, both in number of expressed locations and in expression levels (Fig 4B). Resolving the most highly expressed locations in each cell type at a locus specific level shows that the distribution of expression varies among cell types. (Fig 3C). For example, HepG2 is characterized by high expression from a single locus, while H1-hESC has many locations that are activated.
Previous work has suggested that estimates of HERV expression are highly sensitive to sequencing technology used, and differences due to sequencing technology can obscure biological differences due to cell type[40]. Since aligning shorter fragments (i.e. single-end reads) tends to produce more ambiguously mapping fragments compared to longer fragments, we hypothesized that Telescope (which resolves ambiguity) would create HERV expression profiles that are robust to differences in sequencing technology. Hierarchical clustering of all 30 polyA RNA-seq HERV profiles shows that replicates from the same cell type cluster most closely with other samples from the same cell type, regardless of the sequencing technology used (Fig 5A). Clusters for all cell types had significant support using multiscale bootstrap resampling (approximately unbiased (AU) > 95%). Principal component analysis (PCA) also indicates that cell type, not sequencing technology, is associated with the strongest differences among expression profiles. The first principal component, accounting for 44% of the total variance in the data, separates H1-hESC samples from all other samples (Fig 5B). The second and third components further separate the samples into the other 12 cell types, and capture 13% and 10% of the total variance, respectively. Interestingly, the second component separates blood-derived cell types (K562, GM12878, CD20+ and CD14+) from the other cell types, suggesting that cells derived from the same tissue may share similarities in HERV expression profiles.
We further explored differences among cell types using differential expression (DE) analysis. Pairwise contrasts between cell types were performed to determine the number of significant DE loci (FDR < 0.1, abs(LFC) > 1.0) (Fig 5C). As found in the unsupervised analysis, HERV expression in H1-hESC was drastically different from other cell types, with between 578 and 1127 significantly DE loci.
Finally, we asked whether other existing approaches for TE quantification would be sufficient to identify cell type specific signal in the data or whether these approaches would be sensitive to other variables. We analyzed the ENCODE datasets using default parameters for five other approaches, including best counts, unique counts, TEtranscripts, RepEnrich, and SalmonTE. Hierarchical clustering of the resulting expression profiles reveal that cell types clusters are only recovered using unique counts and Telescope (S3 Fig), though unique counts tended to have less support for clusters. In contrast, clustering with the other four approaches did not recover all cell type clusters; 7 out of 8 cell types clustered together when using best counts expression profiles, 5 cell types were recovered with TEtranscripts and RepEnrich, and only 1 cell type cluster was recovered with SalmonTE profiles (S4 Fig). Interestingly, clustering of the SalmonTE expression profiles revealed 5 samples that did not cluster with their respective cell types, but instead clustered with other single-end datasets (S4 Fig).
In order to examine the sensitivity and biases of computational approaches for quantifying TE expression, we designed simulation experiments with known expression values. Earlier studies have suggested that the HERV-K(HML-2) subfamily (hereafter referred to as HML-2) is expressed in human tissue and may be relevant to human health[8,10,50,51]. Furthermore, its relatively few subfamily members (~90 distinct genomic loci[48]) and high nucleotide identity make HML-2 a good model for studying TE expression. Here, we report on the performance of each method to detect locus-specific expression of HML-2 by simulating RNA-seq fragments with sequencing error. We simulated 25 independent RNA-seq datasets (see methods) and analyzed each using 7 TE quantification approaches: 1) unique counts, 2) best counts, 3) RepEnrich, 4) TEtranscripts, 5) RSEM, 6) SalmonTE, and 7) Telescope. To ensure equal comparisons, all approaches use the same annotation (S1 File), and modifications to the annotation were made to allow locus-specific quantification (instead of family-level quantification) for RepEnrich, TEtranscripts, and SalmonTE.
For all simulations, we plotted the final counts estimated by each approach compared to the expected read count (Fig 6A–6G). We calculated the precision and recall across all loci and simulations (Fig 6H) and represented the overall accuracy of the approach using the F1 score (Fig 6I). Five out of seven approaches were highly sensitive, with true positive rates above 95% in most simulations. The two exceptions were RepEnrich and unique counts, which both tend to discard many more reads than expected (“Unassigned”, Fig 6A and 6C). The unique counts approach consistently underestimated expression levels with ~40% of all estimates (96 out of 250) missing at least 50% of the true expression (Fig 6A). One striking example of this underestimation was for HML2_8p21e; this locus did not generate any fragment that could be uniquely mapped, thus was never detected by this approach.
Performance of the other five approaches differed primarily in the type and magnitude of misclassification errors. False positives occur when reads are incorrectly assigned to annotated loci that are not expressed, resulting in incorrect detection of unexpressed HERV loci. Best counts had a high false positive rate; on average, 12.1% of fragments were incorrectly assigned to unexpressed loci resulting in false detection of unexpressed loci in all simulations (“Other”, Fig 6B). Similarly, the average proportion of reads assigned to unexpressed HERVs is greater than 5% for TEtranscripts, RSEM, and SalmonTE (“Other”, Fig 6D–6F) but is less than 0.1% for Telescope (“Other”, Fig 6G). On the other hand, false negatives occur when reads originating from non-TE regions are assigned to TEs. Since we expect non-TE reads to be unassigned, the number of false negatives can be measured by the difference between the expected number of non-TE reads and the final number of unassigned reads (“Unassigned”, Fig 6). Best counts and Telescope both tend to correctly discard non-TE reads (“Unassigned”, Fig 6B and 6G), while TEtranscripts, RSEM, and SalmonTE tend to incorrectly assign these reads to annotated TEs (“Unassigned”, Fig 6D–6F). We suspect that the model implemented in TEtranscripts attempts to assign all fragments to annotated transcripts, as there is no category for unannotated regions in their model. For RSEM and SalmonTE, this error may be due to the restricted sequence space used to classify the reads. As these methods are mapping to the transcriptome, the true originating sequence is absent from the index and fragments are forced to map to similar, yet incorrect, sequences. This error could be avoided by developing more complete TE annotations or including additional loci that share sequence similarity with TEs of interest.
Of all methods considered here, Telescope had the highest rate of precision and recall from all other counting methods tested (Fig 6H and 6I). In contrast to the best counts approach (Fig 6I), Telescope assigned only 20 fragments to genomic annotations that were not expressed, while 6061 fragments were assigned incorrectly by best counts. The overall accuracy of Telescope estimates from true expression levels, as measured by F1-score, was the highest of all approaches (Fig 6I). These simulation results demonstrate that Telescope resolves ambiguously aligned fragments and produces unbiased estimates of TE expression that are robust to sequencing error.
Transposable elements represent a major biochemically active group of transcripts that are increasingly recognized as important regulators in complex biological systems and disease. However, difficulties in identifying and quantifying these elements has led to TEs being largely ignored in the literature. Here we present Telescope, a novel software package that can be used to mine new or existing RNA-seq datasets to accurately quantify the expression of TEs. The key advantage of our approach is the capability to localize TE expression to an exact chromosomal location.
Based on our analysis of 13 ENCODE cell types, we have identified 1365 individual HERV loci that are expressed in one or more cell types and generated genomic maps that showing cell type specific HERV expression profiles. The ability to quantify expression at specific loci demonstrates that regulation of HERV expression occurs at the locus level (in addition to subfamily-level regulation), as different expression patterns are observed for loci within the same subfamily. For example, our results confirm previous studies identifying HERVH upregulation in embryonic stem cells [9,39,52] and add to this finding by identifying the precise location of HERVH insertions that produce the most transcripts. This high level of resolution for TE expression enables further investigation into the local genomic context, epigenetic regulation, and coding potential of expressed loci.
An earlier study investigating HERV expression using the same datasets found strong differences in estimated HERV expression profiles depending on the sequencing technology used (paired or single end)[40]. Using Telescope, we did not find this same bias; instead, replicates of the same cell type clustered together, while most variance in the data was among cell types. Four of the other TE quantification approaches tested did not appear biased with respect to sequencing technology, while one (SalmonTE) appeared to separate single-end from paired-end samples. We suspect that this is a result of SalmonTEs pseudoalignment approach, as more ambiguous assignments can occur if pairing information is not considered. Other types of bias, such as fragment bias, have been identified in RNA-seq data[53] and may influence expression estimates in Telescope and other programs. We expect future versions of our software to implement corrections for these biases.
Our simulations show that Telescope is highly sensitive and has low type I and II error rates. Unique counts, a heuristic that is commonly chosen for its unambiguous assignments, was shown to discard much of the data and underestimate true TE expression. Best counts, which is commonly used for convenience, also performed poorly and spuriously identified transcripts that were not expressed. Several software packages, including RepEnrich, TEtranscripts, and SalmonTE, also aim to quantify TE expression, but use a family-level approach that quantifies TE subfamilies instead of individual loci. Our simulations used modified inputs for these approaches that allowed us to compare them to Telescope. Based on our simulation results, we find that our approach achieves high sensitivity while minimizing spurious detections, while all other approaches tend to identify TEs that are not expressed. We conclude that Telescope offers superior accuracy for TE quantification and is the only available software packages that quantifies TE expression at single-locus resolution.
Telescope will have widespread utility in other settings. Studies on TE expression have become prominent in studies of embryonic stem cell development[8][9], neural cell plasticity[54,55], oncogenesis[4–7,56,57], psychiatric and neurological disorders[58–60] and autoimmune diseases[61,62]. As the breadth of knowledge on TEs expands, expression profiling of TEs using Telescope will allow scientists to discover unique and collective TE transcripts involved in the biology of complex systems.
Telescope implements a generative model of RNA-seq relating the probability of observing a sequenced fragment to the proportions of fragments originating from each transcript. Formally, let F = [f1,f2,…,fN] be the set of N observed sequencing fragments. We assume these fragments originate from K annotated transcripts in the transcriptome T = [t0,t1,…,tK]. In practice, annotations fail to identify all possible transcripts that generate fragments, thus we include an additional category, t0, for fragments that cannot be assigned to annotated transcripts. Let G = [G1,G2,…,GN] represent the true generating transcripts for F, where Gi∈T and Gi = tj if fi originates from tj. Since the process of generating F from T cannot be directly observed, the true generating transcripts G are considered to be “missing” data. The objective of our model is to estimate the proportions of T by learning the generating transcripts of F.
The alignment stage identifies one or more possible alignments for each fragment, along with corresponding alignment scores. Telescope uses the alignment score generated by the aligner and reported in the AS tag[63]. This is typically calculated by adding scores and penalties for each position in the alignment; a higher alignment score indicates a better alignment. Let qi = [qi0,qi1,…,qiK] be the set of mapping qualities for fragment fi, where qij = Pr(fi|Gi = tj) represents the conditional probability of observing fi assuming it was generated from tj; we calculate this by scaling the raw alignment score by the maximum alignment score observed for the data. We write the likelihood of observing uniquely aligned fragment fu as a function of the conditional probabilities qu and the relative expression of each transcript for all possible generating transcripts Gu
Pr(fu|π,qu)=∑j=0Kπjquj
where π = [π0,π1,…,πK] represents the fraction of observed fragments originating from each transcript. Note that quj = 0 for all transcripts that are not aligned by fu. For non-unique fragments, we introduce an additional parameter in the above likelihood to reweight each ambiguous alignment among the set of possible alignments. The probability of observing ambiguous fragment fa is given by
Pr(fa|π,θ,qa)=∑j=0Kπjθjqaj
where θ = [θ0,θ1,…,θK] is a reassignment parameter representing the fraction of non-unique reads generated by each transcript.
Using these probabilities of observing ambiguous and unique fragments, we formulate a mixture model describing the likelihood of the data given parameters π and θ. The K mixture weights in the model are given by π, the proportion of all fragments originating from each transcript. To account for uncertainty in the initial fragment assignments, let xi = [xi0,xi1,…,xiK] be a set of partial assignment (or membership) weights for fragment fi, where ∑j=0Kxij=1 and xij = 0 if fi does not align to tj. We assume that xi is distributed according to a multinomial distribution with success probability π. Intuitively, xij represents our confidence that fi was generated by transcript tj. In order to simplify our notation, we introduce an indicator variable y = [y1,y2,…,yN] where yi = 1 if fi is ambiguously aligned and yi = 0 otherwise. The complete data likelihood is
L(π,θ|x,q,y)∝∏i=1N∏j=0K[πjθjyiqij]xij
Telescope iteratively optimizes the likelihood function using an expectation-maximization algorithm[64]. First, the parameters π and θ are initialized by assigning equal weight to all transcripts. In the expectation step, we compute the expected values of xi under current estimates of the model parameters. The expectation is given by the posterior probability of xi:
E[xij]=πjθjyiqij∑k=0Kπkθkyiqik
In the M-step we calculate the maximum a posteriori (MAP) estimates for π and θ
πj^=∑i=1NE[xij]+ajM+∑k=0Kakandθj^=∑i=1NE[xij]yi+bj∑i=1Nyi+∑k=0Kbk
Where M=∑j=0K∑i=1NE[xij] and aj and bj are prior information for transcript tj. Intuitively, these priors are equivalent to adding unique or ambiguous fragments to tj. As currently implemented, the user may provide a prior value for either parameter that is distributed equally among all transcripts. We have found that providing an informative prior for the bj (—theta_prior) is recommended given the repeat content of the human genome, since large values for this parameter prevents convergence to boundary values. Convergence of EM algorithms to local maxima has been shown by Wu[65], and is achieved when the absolute change in parameter estimates is less than a user defined level, typically ϵ<0.001.
A Telescope analysis requires an annotation that defines the transcriptional unit of each TE to be quantified. For HERV proviruses, the prototypical transcriptional unit contains an internal protein-coding region flanked by LTR regulatory regions. Existing annotations, such as those identified by RepeatMasker[33] (using the RepBase database[32]) or Dfam[66] identify sequence regions belonging to TE families but do not seek to annotate transcriptional units. Both databases represent the internal region and corresponding LTRs using separate models, and the regions identified are sometimes discontinuous. Thus, a HERV transcriptional unit is likely to appear as a collection of nearby annotations from the same HERV subfamily.
Transcriptional units for HERV proviruses were defined by combining RepeatMasker annotations belonging to the same HERV subfamily that are located in adjacent or nearby genomic regions. Briefly, repeat families belonging to the same HERV subfamily (internal region plus flanking LTRs) were identified using the RepBase database[32]. RepeatMasker annotations for each repeat subfamily were downloaded using the UCSC table browser[67] and converted to GTF format, merging nearby annotations from the same repeat subfamily. Next, LTRs found flanking internal regions were identified and grouped using BEDtools[68]. HERV transcriptional units containing internal regions were assembled using custom python scripts. Each putative locus was categorized according to provirus organization; loci that did not conform to expected HERV organization or conflicted with other loci were visually inspected using IGV[69] and manually curated. As validation, we compared our annotations to the HERV-K(HML-2) annotations published by Subramanian et al.[48]; the two annotations were concordant. Final annotations were output as GTF (S1 File); all annotations, scripts, and supporting documentation are available at https://github.com/mlbendall/telescope_annotation_db.
We identified 30 ENCODE datasets with available whole-cell bulk RNA-seq data from tier 1 and 2 common cell types (S1 Table). Sequence data was obtained from SRA and extracted using the parallel-fastq-dump package (https://github.com/rvalieris/parallel-fastq-dump). Adapter trimming, quality trimming, and filtering were performed using Flexbar[70] (version 3.0.3). For Telescope analysis, the trimmed and filtered reads from each run were aligned to human reference genome hg38 using bowtie2[71]. Alignment options were specified to perform a sensitive local alignment search (—very-sensitive-local) with up to 100 alignments reported for each fragment pair (-k 100). The minimum alignment score threshold was chosen so that fragments with approximately 95% or greater sequence identity would be reported (—score-min L,0,1.6). Sequence alignment map (SAM/BAM) files from different runs corresponding to the same sample were concatenated to obtain sample-level BAM files. An annotation of HERV locations in hg38 (S1 File) and the BAM file for each sample were provided as inputs for Telescope. Telescope options included up to 200 iterations of the expectation-maximization algorithm (—max_iter 200) and an informative prior on theta (—theta_prior 200000). The “final counts” column in the Telescope report are used as HERV expression data in subsequent analysis.
ENCODE datasets were also analyzed using five other approaches. Unique and best counts approaches use the same alignment and annotation as above and are included as part of the Telescope output. RepEnrich, TEtranscripts, and SalmonTE were all run according to author instructions, with author-provided annotations and default parameters.
Library size for each sample is considered to be the total number of fragments that map to the reference genome. Counts per million (CPM) were calculated by dividing the raw count by the library size and multiplying by 1 million. A CPM cutoff of 0.5 was used to identify expressed loci; since the smallest sample considered has more than 20 million fragments, expressed loci are represented by at least 10 observations. Raw counts output by Telescope were used for differential expression analysis. Size factors for normalization were calculated by dividing the library sizes by their geometric mean. Normalization, dispersion estimation, and generalized linear model fitting was performed using DESeq2[72]; the model was specified with cell type as the only covariate. Contrasts were extracted for each pair of cell types; HERVs with an adjusted p-value < 0.1 and log2FoldChange > 1.0 were considered to be differentially expressed.
Read counts for clustering were transformed using a variance stabilizing transformation in DESeq2[72]. Hierarchical clustering with multiscale bootstrap resampling was performed on transformed counts using correlation distance and UPGMA clustering implemented in pvclust[73]. Uncertainty in hierarchical cluster analysis was assessed by calculating two p-values for each cluster that range from 0 to 1, with 1 indicating strong support for the cluster. The bootstrap probability (BP) is calculated by normal bootstrap resampling and approximately unbiased (AU) probability is computed by multiscale bootstrap resampling[74].
For the simulation study, we simulated 25 independent RNA-seq datasets with 2100 paired-end fragments each. For each dataset, we randomly selected 13 loci to be expressed, including 10 HML-2 proviruses and three “non-TE” loci. HML-2 proviruses were selected from 92 HML-2 loci present in our annotation; non-TE loci were selected from a set of 968 unannotated genomic regions that share sequence similarity with the HML-2 subfamily (S2 File). Non-TE loci are included to examine the type II error rate of the approaches; assigning non-TE fragments to HML-2 loci is considered a false negative. Expression levels for the 10 HML-2 loci in each dataset were randomly chosen, ranging from 30 to 300 fragments per locus. Each of the three non-TE loci were expressed at 150 fragments each. Using this expression pattern, we simulated sequencing fragments with the Bioconductor package for RNA-seq simulation, Polyester[75]. All simulations used the parameters of read length: 75 bp; average fragment size: 250; fragment size standard deviation: 25; and an Illumina error model with an error rate of 5e-3.
Each simulation dataset was analyzed using 7 TE quantification approaches: 1) unique counts, 2) best counts, 3) RepEnrich, 4) TEtranscripts, 5) RSEM, 6) SalmonTE, and 7) Telescope. To ensure a fair comparison among approaches, the same annotation (S1 File) was used as input for all approaches. Note that the HML-2 loci used for simulation are contained in this annotation, but the non-TE loci are absent. For RepEnrich, TEtranscripts, and SalmonTE, the locus identifier was used in place of the family name in order to generate locus-specific estimates. Aside from these changes, each program was run as suggested by the authors. Unique counts was implemented by aligning reads with bowtie2, allowing for multi-mapped reads (-k 100—very-sensitive-local—score-min L,0,1.6) and filtering reads with multiple alignments. The same bowtie2 parameters were used for best counts without specifying -k (—very-sensitive-local—score-min L,0,1.6).
The five software packages include final read counts as part of the output. Read counts for the unique counts and best counts approaches were obtained using htseq-count[76]. After mapping and counting the reads for each annotated HERV, reads can be divided in two categories, depending their origin, HML-2 reads or non-TE reads. Those reads can then be correctly or incorrectly mapped, depending of the outcome of the counting method, leading to 4 different categories: a) reads assigned to HML-2 correctly (True Positive) b) reads assigned to HML-2 incorrectly (False Positive) c) reads not assigned correctly (True Negative) d) reads not assigned incorrectly (False Negative). All classifications were made based on counts and not fragment assignments, as several approaches do not provide final fragment assignments. The classifications were used for recall and precision calculations.
Telescope is implemented in Python, is available as an open-source program under the MIT license, and has been developed and tested on Linux and MacOS. The software package and test data can be found at https://github.com/mlbendall/telescope. We recommend installing Telescope and its dependencies using the bioconda package manager[77].
A complete snakemake[78] pipeline for reproducing the ENCODE analysis is available from https://github.com/mlbendall/TelescopeEncode. Scripts for reproducing the simulations are available from https://github.com/LIniguez/Telescope_simulations. A tutorial for running the single-locus analysis is available from https://github.com/mlbendall/telescope_demo.
|
10.1371/journal.pgen.1003620 | Female Behaviour Drives Expression and Evolution of Gustatory Receptors in Butterflies | Secondary plant compounds are strong deterrents of insect oviposition and feeding, but may also be attractants for specialist herbivores. These insect-plant interactions are mediated by insect gustatory receptors (Grs) and olfactory receptors (Ors). An analysis of the reference genome of the butterfly Heliconius melpomene, which feeds on passion-flower vines (Passiflora spp.), together with whole-genome sequencing within the species and across the Heliconius phylogeny has permitted an unprecedented opportunity to study the patterns of gene duplication and copy-number variation (CNV) among these key sensory genes. We report in silico gene predictions of 73 Gr genes in the H. melpomene reference genome, including putative CO2, sugar, sugar alcohol, fructose, and bitter receptors. The majority of these Grs are the result of gene duplications since Heliconius shared a common ancestor with the monarch butterfly or the silkmoth. Among Grs but not Ors, CNVs are more common within species in those gene lineages that have also duplicated over this evolutionary time-scale, suggesting ongoing rapid gene family evolution. Deep sequencing (∼1 billion reads) of transcriptomes from proboscis and labial palps, antennae, and legs of adult H. melpomene males and females indicates that 67 of the predicted 73 Gr genes and 67 of the 70 predicted Or genes are expressed in these three tissues. Intriguingly, we find that one-third of all Grs show female-biased gene expression (n = 26) and nearly all of these (n = 21) are Heliconius-specific Grs. In fact, a significant excess of Grs that are expressed in female legs but not male legs are the result of recent gene duplication. This difference in Gr gene expression diversity between the sexes is accompanied by a striking sexual dimorphism in the abundance of gustatory sensilla on the forelegs of H. melpomene, suggesting that female oviposition behaviour drives the evolution of new gustatory receptors in butterfly genomes.
| Insects and their chemically-defended hostplants engage in a co-evolutionary arms race but the genetic basis by which suitable host plants are identified by insects is poorly understood. Host plant specializations require specialized sensors by the insects to exploit novel ecological niches. Adult male and female Heliconius butterflies feed on nectar and, unusually for butterflies, on pollen from flowers while their larvae feed on the leaves of passion-flower vines. We have discovered–between sub-species of butterflies-fixed differences in copy-number variation among several putative sugar receptor genes that are located on different chromosomes, raising the possibility of local adaptation around the detection of sugars. We also show that the legs of adult female butterflies, which are used by females when selecting a host plant on which to lay their eggs, express more gustatory (taste) receptor genes than those of male butterflies. These female-biased taste receptors show a significantly higher level of gene duplication than a set of taste receptors expressed in both sexes. Sex-limited behaviour may therefore influence the long-term evolution of physiologically important gene families resulting in a strong genomic signature of ecological adaptation.
| Nearly 50 years ago Ehrlich and Raven proposed that butterflies and their host-plants co-evolve [1]. Based on field observations of egg-laying in adult female butterflies, feeding behavior of caterpillars, and studies of systematics and taxonomy of plants and butterflies themselves, they outlined a scenario in which plant lineages evolved novel defensive compounds which then permitted their radiation into novel ecological space. In turn, insect taxa evolved resistance to those chemical defences, permitting the adaptive radiation of insects to exploit the new plant niche. Ehrlich and Raven's theory of an evolutionary arms-race between insects and plants drew primarily from an examination of butterfly species richness and host-plant specialization. It did not specify the sensory mechanisms or genetic loci mediating these adaptive plant-insect interactions.
Insects possess gustatory hairs or contact chemosensilla derived from mechanosensory bristles, scattered along a variety of appendages [2]–[4]. In adult butterflies and moths, gustatory sensilla are found on the labial palps and proboscis (Figure 1), the legs (Figure 2A) [5], the antennae (Figure 2B) [6], [7], and the ovipositor [8], [9]. In adult Heliconius charithonia legs, the 5 tarsomeres of the male foreleg foretarsus are fused and lack chemosensory sensilla, while female foretarsi bear groups of trichoid sensilla (n = 70–90 sensilla/tarsus) associated with pairs of cuticular spines [10]. Each trichoid sensilla contains five receptor neurons. These sensilla are sensitive to compounds that may be broadly classified as phagostimulants (e.g., sugars and amino acids), which promote feeding behavior, or phagodeterrents (secondary plant compounds), which suppress it [11]; in adult females they may also modulate oviposition [12].
Genes for vision, taste and smell are likely to be crucial genomic loci underlying the spectacular diversity of butterfly-plant interactions. The availability of genomes for two butterfly species, the postman Heliconius melpomene (Nymphalidae) [13] and the monarch (Danaus plexippus) [14], as well as the silkmoth (Bombyx mori) [15], enables us to examine the evolutionary diversification of gustatory (Gr) and olfactory (Or) receptor genes that mediate insect-plant interactions. Each of these species feeds on hosts from different plant families. Silkmoth larvae feed on mulberry (Morus spp., Moraceae) and monarch larvae feed on milkweed (Asclepias spp., Apocynaceae). The larvae of Heliconius feed exclusively on passion flower vines, primarily in the genus Passiflora (Passifloraceae). In addition, adult Heliconius are notable for several derived traits such as augmented UV color vision [16], pollen feeding (Figure 1B) [17], [18], and the ability to sequester substances from their host plants that are toxic to vertebrate predators such as birds [19], [20].
In Drosophila melanogaster, the Gr gene family consists of 60 genes [21]–[24], several of which are alternatively spliced, yielding 68 predicted Gr transcripts [24]. One or more of these Gr proteins including possibly obligatory co-receptors [25]–[27] may be expressed in each gustatory receptor neuron [11]. Originally considered members of the G-protein-coupled receptor (GPCR) family, insect Grs have an inverted orientation in the membrane compared to the GPCR family of vertebrate Grs [28] and are part of the same superfamily as the insect Ors [21]. Signalling pathways for insect Grs may be both G-protein dependent [29], [30], [31] and G-protein independent [32]. For the vast majority of Drosophila Grs the specific compounds to which they are sensitive remain unknown. Nonetheless, several receptors for sugars [33]–[35], CO2 [26], [36], bitter substances [37]–[39] and plant-derived insecticides [25] have been identified in flies.
Knowledge of the Gr gene family for insects outside Drosophila is sparse and has primarily relied on the analyses of individual reference genomes. Expression studies are challenging, due to the very low expression of Grs in gustatory tissues [21], [23]. In addition, Grs and Ors typically have large introns, small exons and undergo fast sequence evolution, making their in silico identification using automated gene prediction algorithms from genomic sequences problematic. Thus, the large repertoire of Grs (and Ors) that have been examined in the reference genomes of the pea aphid [40], the honey bee [41], the red flour beetle Tribolium castaneum [42], the mosquitoes Aedes aegypti [43] and Anopheles gambiae [44], and several Drosophila spp. [45], [46] have required extensive manual curation. In Lepidoptera, a large insect group which includes ∼175,000 species, completely described Gr (and Or) gene models from genomes are rare and limited to B. mori [47], D. plexippus [14] and H. melpomene (Grs, this study; Ors, [13]). In other lepidopteran species, only fragmentary Gr data are available: five sequences in Spodoptera littoralis [48], three in Heliothis virescens [49], two in Manduca sexta [50], [51] and one in Papilio xuthus [52].
Adult females of each Heliconius species only lay eggs on a limited number of host plants [53], and therefore need to recognize different species from among the large and diverse Passifloraceae family, which also show a remarkable diversity of chemical defences [54]. The evolutionary arms race between Heliconius butterflies and their hosts led us to hypothesize that Heliconius Grs (and Ors) might be subject to rapid gene duplication and gene loss as well as copy-number variation (CNV). Recent work taking advantage of published Drosophila genomes has shown a relationship between host specialization and/or endemism and an increased rate of gene loss, as well as a positive relationship between genome size and gene duplication [46], [55]. Moreover, Drosophila Grs appear to be evolving under weaker purifying selection than Ors [55].
We previously used the reference genome sequence for H. melpomene to annotate three chemosensory gene families, encoding the chemosensory proteins (CSPs), the odorant-binding proteins (OBPs), and the olfactory receptors (Ors). This demonstrated a surprising diversity in these gene families. In particular there are more CSPs in the butterfly genomes than in any other insect genome sequenced to date [13]. We build on this work below by characterizing the Gr gene family in the reference H. melpomene melpomene genome and in two other lepidopteran species whose genomes have been sequenced, B. mori (Bombycidae) and D. plexippus (Nymphalidae), by performing in silico gene predictions and phylogenetic analysis. We then analyzed whole-genome sequences of twenty-seven individual butterflies, representing eleven species sampled across all major lineages of the Heliconius phylogeny and including sixteen individuals from two species, H. melpomene and its sister-species H. cydno. We also generated RNA-sequencing expression profiles of the proboscis and labial palps, antennae and legs of individual adult male and female butterflies of the sub-species H. melpomene rosina from Costa Rica (∼1 billion 100 bp reads). We used these data to address four major questions: Are different chemosensory modalities less prone to duplication and loss than others (e.g., taste vs. olfaction)? Is there evidence of lineage-specific differentiation of Gr (and Or) repertoires between genera, species and populations? What is the relationship between CNVs and the retention of paralogous genes over long-term evolutionary timescales? Are the life history differences between males and females reflected in the expression of Grs and Ors as well as in the retention of novel sensory genes in the genome?
We find higher turnover of the Grs than the Ors over longer evolutionary timescales, and evidence for both gene duplication and loss among a clade of intronless Grs between lepidopteran species and within the genus Heliconius. We also find for H. melpomene and its sister species, H. cydno, evidence of copy-number variation (CNVs) within their Gr and Or repertoires. Lastly, our RNA-sequencing suggests both tissue-specific and sex-specific differences in the diversity of expressed Grs and Ors, with female legs expressing a more diverse suite of Grs than male legs. Our data set revealing the expression of 67 of 73 predicted Gr genes and 67 of 70 predicted Or genes in adult H. melpomene butterflies is the most comprehensive profiling of these chemosensory gene families in Lepidoptera to date, and suggests how female host plant-seeking behaviour shapes the evolution of gustatory receptors in butterflies.
In total, we manually annotated 86,870 bp of the H. melpomene melpomene reference genome (Table S1). Our 73 Gr gene models, consisted of 1–11 annotated exons, with the majority having three or four exons; six were intronless. We found genomic evidence (but not RNA-seq evidence) of possible alternative splicing of the last two exons of HmGr18, bringing the total number of predicted Grs to 74. Alternative splicing has not been previously described in the silkmoth B. mori [47], but is known to occur in most other insects examined, including D. melanogaster, Anopheles gambiae, Aedes aegypti and T. castaneum [24], [43], [44]. We also identified eleven new putative Grs in the monarch butterfly genome, DpGr48-56, DpGr66 and DpGr68 (Table S1) [14].
All but five of our gene models contained more than 330 encoded amino acids (AAs) while individual gene models ranged from 258–477 AAs. Several Gr genes contained internal stop codons (Table S1). In at least one case, we found RNA-seq evidence of an expressed pseudogene–HmGr61–with two in-frame stop codons. In other cases, the 5′ end of our assembled transcripts was not long enough to verify the internal stop codons in the genome assembly. The Grs are located on 33 distinct scaffolds, with 58 forming clusters of 2–8 genes on 18 scaffolds, distributed across 14 chromosomes.
To study the patterns of gene duplication and loss more broadly across the Lepidoptera, we next examined the phylogenetic relationships of Grs from the three lepidopteran reference genomes [13]–[15]. Across the gene family phylogeny a large number of duplications among the putative ‘bitter’ gustatory receptors of Heliconius or Danaus have occurred, while the putative CO2 and sugar receptors are evolving more conservatively, with only single copies in the H. melpomene reference genome (see below)(black arcs, Figure 3). A majority (∼64%) of Gr genes found in the H. melpomene genome are the result of gene duplication since Heliconius shared a common ancestor with Danaus or Bombyx. This is in contrast to the more conserved pattern of evolution of the Ors (Figure 4) [13] where a majority (37 of 70 or 53%) of genes show a one-to-one orthologous relationship with either a gene in Danaus, in Bombyx or both.
Within the genus Heliconius there is a great diversity of host plant preferences for different Passiflora species. To look at the relationship between gene duplication and loss over this shorter timescale, we focussed our efforts on a group of six intronless Grs, HmGr22-26 and Gr53, because it is only feasible to identify single-exon genes with high confidence, given that the Illumina whole-genome sequencing approach leads to poorly assembled genomes (Table S2). These genes are also of interest as some members of this group are very highly expressed. Notably HmGr22 is one of the most widely expressed genes in our adult H. melpomene transcriptomes, which was verified by reverse-transcriptase (RT)-PCR and sequencing of the PCR products (Figure 5A). In this regard HmGr22 resembles another intronless Gr, the silkmoth gene BmGr53, which is expressed in adult male and female antennae and larval antennae, maxilla, labrum, mandible, labium, thoracic leg, proleg and gut [32]. The remaining five intronless Grs have much more limited domains of expression in adult H. melpomene (see below). We searched for these genes in de novo assemblies of whole-genome Illumina sequences from eleven species across the Heliconius phylogeny. We investigate whether, as in Drosophila, a high turnover in putative bitter receptors is observed in species with host plant specializations or in species which are endemic and thus smaller in effective population size [46].
Although patterns of host plant use are complex within the genus, some notable host-plant shifts have occurred, leading to the prediction that gene loss may have occurred along more specialized lineages [46]. For example, H. doris unlike many Heliconius, tends to feed on large woody Passiflora that can support their highly gregarious larvae [53]. It also probably has a smaller effective population size than most other Heliconius species. From the 11 species studied, we identified a total of 44 intact or nearly intact intronless Grs, as well as three intronless pseudogenes (Genbank Accession Nos. KC313949-KC313997)(Table S2 and S3). We also identified one intact intronless Gr each in monarch and silkmoth and one intronless Gr pseudogene in monarch. Phylogenetic analysis indicates that six intact intronless Gr genes were present at the base of the genus Heliconius while the intronless Gr pseudogene in monarch was the result of duplication since Heliconius and monarch shared a common ancestor (Figure 5B, Figure 6). Subsequent to the radiation of the genus Heliconius, there have been a number of gene losses. Whereas all members of the melpomene clade (H. melpomene, H. cydno, H. timareta) retained genomic copies of all six genes, members of the erato clade (H. erato, H. clysonymus and H. telesiphe) and sara-sapho clade (H. sara and H. sapho) have lost their copies of Gr22 and Gr25. In addition, members of the so-called primitive clade (H. wallacei, H. hecuba, and H. doris) have lost Gr23, while H. doris and H. wallacei have apparently lost Gr24 independently (Figure 6). The woody plant specialist, H. doris, has retained the fewest intronless Grs, apparently also having lost its copy of Gr53, a pattern mirrored by Drosophila host plant specialists [46]. We have, however, no direct evidence that the intronless Grs are in fact involved in host plant discrimination so the observed patterns of loss may be better explained by other variables such as effective population size.
We next tested whether the greater level of diversification of Grs as compared to Ors over long evolutionary timescales (compare Figure 3 and Figure 4), is similarly reflected in greater population level variation in Gr and Or duplicate genes. To test this hypothesis, we examined the incidence of CNVs among Grs and Ors that exist as single-copy genes in the reference H. melpomene genome with a one-to-one orthologous relationship with a gene in Danaus, Bombyx or both (conserved)(red dots, Figure 3 and 4), or as genes that are Heliconius-specific where no orthologue exists in either Danaus or Bombyx (non-conserved). We used whole genome resequence data (12 genomes) for three subspecies of H. melpomene (H. melpomene amaryllis, n = 4; H. melpomene aglaope, n = 4; and H. melpomene rosina, n = 4)(Figure 7, inset) and one sub-species of H. cydno (H. cydno chioneus, n = 4)(Table S4). We first mapped genomic resequence reads to the H. melpomene melpomene reference genome, and then searched for regions of abnormal coverage using CNVnator [56]. More than half of Gr loci showed presence of CNVs (37 out of 68 loci). However, there were noticeably fewer CNVs in Gr loci that evolve conservatively over the long-term, such as among the putative CO2 receptors, while there was an excess of CNVs in loci that show patterns of Heliconius-specific duplication (11.1% vs. 54.9%, respectively)(Fisher's Exact Test, two-tailed, P = 0.0004) (Table 1)(Figure 7). Intriguingly, many sugar receptor CNVs are sub-species specific; we observed fixed duplications relative to the reference genome in H. melpomene aglaope (HmGr4, Gr5, Gr6, Gr8, Gr45, Gr52) and H. melpomene amaryllis (Gr4, Gr5, Gr6, Gr7, Gr8, Gr45, Gr52), among genes that are found on different chromosomes (Table S5, Figure 7). Although the majority of CNVs are likely to be evolving neutrally, this raises the possibility of local adaptation within the species range around the detection of sugars. As expected given their long-term stability, Ors also show a lower incidence of CNVs (12 out of 67 loci), with no association between gene duplication and CNV incidence at least in H. melpomene (Table 1, Table S6). In H. cydno, a slight excess of Or CNVs was observed in loci that resulted in paralogous genes over longer evolutionary timescales (Fisher's Exact Test, two-tailed, P = 0.0475)(Table 1)(Figure 8).
We have not experimentally verified the incidence of copy number variation in any of these genomes, and some of the regions identified as CNVs are likely to be false positives. To investigate the rate of false positives, we analysed resequence data from the reference genome itself and discovered 3 Gr and 3 Or CNVs, suggesting a false positive rate of around 4%. (We therefore excluded these loci from our statistical tests.) However, the fact that broad patterns of observed CNVs are consistent with the evolutionary patterns at deeper levels supports our conclusion that CNV, in the absence of strong purifying selection, is an important driver of gene family diversification. These results also provide a novel line of evidence that the butterfly Grs have a higher rate of evolutionary turnover as compared to Ors.
The life histories of adult male and female butterflies are similar with respect to the need to find food and potential mates except that adult females are under strong selection to identify suitable host plants for oviposition. To ascertain host-plant identity, female butterflies drum with their legs on the surface of leaves before laying eggs [10]. This behaviour presumably allows the female to taste oviposition stimulants. Consistent with this behaviour, adult nymphalid butterfly legs are known to contain gustatory sensilla [57], and it has been reported that while nymphalid butterfly females have clusters of gustatory sensilla on their foreleg foretarsi, males lack these entirely [10], [58]. Here we confirm this mostly anecdotal evidence for sexual dimorphism using scanning electron microscopy (SEM). The mid- and hindlegs of both male and female H. melpomene have similar numbers of individual gustatory sensilla along their entire lengths, but there is a striking difference in their abundance and distribution on the foretarsi of the female forelegs. Unlike males, females exhibit cuticular spines associated with gustatory (trichoid) sensillae (n∼80 sensilla/foretarsus for females; n = 0/foretarsus for males) (Figure 2A) [10].
We therefore hypothesized that the repertoire of expressed Gr and Or genes in H. melpomene legs might be more diverse in females as compared to males. Furthermore, if female-specific genes are used for assessment of potential host plants, then fast-evolving insect-host interactions might produce rapid duplication of these genes over evolutionary timescales. Accordingly, we examined the expression profiles of Grs and Ors in adult H. melpomene by RNA-sequencing of libraries prepared from mRNAs expressed in adult antennae, labial palps and proboscis, and legs from one deeply-sequenced male and female each of H. melpomene (6 libraries total)(Table S7 and S8). The number of 100 bp reads per individual library ranged from 17.4 to 25.9 million for paired-end sequencing or 74.8–103.9 million for single-end sequencing (Table S8). To confirm these findings, we subsequently made 12 individual libraries from two more males and two more females (Table S7). As coverage was uneven across these libraries, we analysed them by merging biological replicates by sex and tissue type, and then downsampling so that an equal number of reads was analyzed for each treatment. The number of 100 bp reads analyzed for paired-end sequencing ranged from 19.4 to 49.6 million (Table S8). After downsampling, we examined the expression levels of the widely-expressed elongation factor-1 alpha gene in each of the libraries as a control, and found a comparable level of expression between sexes within each tissue type (Table S8). By careful visual examination of the uniquely-mapped reads to our 143 reference Gr and Or sequences, we found evidence of Gr and Or expression in all three adult tissue-types, with both tissue-specific and sex-specific differences as detailed below (Figure 9, Tables S9, S10, S11, S12, S13, S14). In total, we found evidence for expression of 67 of 73 Grs and 67 of 70 Ors identified in the H. melpomene reference genome.
Strikingly, the sexual dimorphism of gustatory sensilla we observed among the foreleg foretarsi is reflected in Gr gene expression patterns. A total of thirty-two Grs are expressed in both male and female H. melpomene leg transcriptomes including three CO2 receptors, HmGr1-3, four putative sugar receptors HmGr4, Gr6, Gr45 and Gr52 and a fructose receptor, HmGr9 (Figure 9A, Table S9, Supplementary Text). Many Grs showed sex-specific expression, however, with many more Grs in female (n = 46) as compared to male leg transcriptomes (n = 33)(Figure 9B, C).
In total 15 of these Grs expressed in female legs, HmGr10, Gr24, Gr26, Gr29, Gr40, Gr41, Gr48, Gr50, Gr51, Gr16, Gr55, Gr57, Gr58, Gr60 and Gr67, are the result of duplications since Heliconius and Danaus shared a common ancestor (Figure 3 small arrows, Figure 9B, Table S9). By contrast, only one of the three male-biased Grs, HmGr19, evolved as a result of recent duplication. There is an excess of Heliconius-specific Grs but not Ors (see below) that are expressed in female legs (Fisher's Exact Test, two-tailed, p = 0.019)(Table 2). Since male H. melpomene do not need to identify host-plants for oviposition, it seems likely that the 17 female-specific Grs in our leg transcriptomes are candidate receptors involved in mediating oviposition (Figure S1).
Besides using their antennae for olfaction, female nymphalid butterflies also taste a host plant by antennal tapping before oviposition. This tapping behaviour presumably allows the host plant chemicals to come into physical contact with gustatory sensilla on the antennae. We therefore examined whether there was any difference in the abundance of gustatory sensilla on the antennae of male and female H. melpomene. Using scanning electron microscopy, we found individual gustatory sensilla scattered along each antennae of both male and female H. melpomene but no obvious sexual dimorphism in their abundance or distribution (Figure 2B). We found 28 Grs expressed in both male and female H. melpomene antennae (Figure 9A, Table S10), including two sugar receptors, HmGr4 and HmGr52, a putative fructose receptor HmGr9 and two CO2 receptors, HmGr1 and Gr3. Besides the sugar and CO2 receptors noted, other conserved genes that are expressed in both male and female antennae include HmGr63, a candidate Gr co-receptor (see Text S1), and HmGr66, a candidate bitter receptor.
We also found 11 Grs expressed in female H. melpomene antennae that did not appear to be expressed in male antennae. Two of these, HmGr47 and Gr68, appeared in the top one-third of the most abundant female antennal Grs in terms of number of reads recovered from the individual butterfly transcriptome. In contrast, just four Grs were expressed in male antennae HmGr11, Gr25, Gr31, and Gr69 but not female antennae (Figure 9B, C, Table S10). Six of the female-biased Grs and two of the male-biased Grs (Gr31, Gr69) expressed in antennae are the result of duplication events since Heliconius and Danaus shared a common ancestor.
By contrast with the leg and antennal tissue, where more Grs are expressed in females compared to males, the labial palps and proboscis (Figure 1) transcriptomes contained the largest number of Grs (n = 35) expressed in both sexes (Figure 9A, C, Table S11). Five of the six candidate sugar receptors in the H. melpomene genome are expressed in both the male and the female transcriptomes along with two of the three conserved CO2 receptors, which may be used to assess floral quality [59] (Figure 3, Table S11). A majority (21 of 35) of Heliconius Grs expressed in both male and female labial palps and proboscis libraries have no existing ortholog in the silkmoth genome, apparently the result of gene loss in B. mori or gene duplication along the lineage leading to Heliconius (Figure 3). This may in part reflect the fact that adult silkmoths have lost the ability to feed. Interestingly, four Grs expressed in both male and female labial palps and proboscis transcriptomes could not be detected in male and female antennae and legs (HmGr12, Gr20, Gr35, and Gr59)(Figure 3, red arrows, Figure 9B). Some of these Grs might play a role in the pollen-feeding behaviour that is specific to Heliconius, and which involves preferences for particular species of flowers in the plant families Rubiaceae, Cucurbitaceae and Verbenaceae (see Discussion).
In addition to the Gr gene expression described above, we examined Or expression in the three adult tissues. The expression of Ors in antennal tissue has been widely studied in a variety of insects including Drosophila and some Lepidoptera [50], [60]. As expected, we observed that Or gene expression was high in the antennae. Unexpectedly, Or expression was about as prevalent as Gr expression in the proboscis and labial palps and leg transcriptomes (Figure 9D, E, F). In total across all three tissues profiled, we found evidence for the expression of nearly all predicted Or genes (67 of 70 genes)(Table S12, S13, S14) in the H. melpomene reference genome [13].
Outside Drosophila, the study of sensory gene family evolution in insects has generally been limited to the comparison of a small number of phylogenetically distant reference genomes. Such studies have commonly involved a comparison of the size of gene families between taxa in order to document lineage-specific expansions (Figure 10), and the comparison of dN/dS ratios to identify branches subject to rapid evolution [61]. Here we have used a similar approach to annotate 73 Grs in the Heliconius melpomene reference genome. However, we have also demonstrated the power of next-generation sequencing to elucidate patterns of evolution and expression of these genes. These data have offered exciting new insights into a set of genes that show both rapid evolution and sex-specific expression patterns, suggesting that female oviposition behaviour drives the evolution of butterfly gustatory receptors.
Previous work in other insects indicates that Grs are an important target for gene duplication and loss between species. Most notably, D. sechellia and D. erecta are host specialists, on Morinda citrifolia and Pandanus candelabrum respectively, while D. simulans is a generalist fly exploiting a broad array of rotting fruit [46]. Host specialization in the former species is associated with an acceleration of gene loss and increased rates of amino acid evolution at receptors that remain intact. Here we have used whole-genome Illumina sequencing of single diploid individuals to similarly document patterns of gene gain and loss across Heliconius. This method yields highly fragmented genome assemblies, but such assemblies have proven very informative, most notably for studying the evolution of the clade of single-exon bitter receptor genes. We identified three gene duplication events along the lineage leading to Heliconius, followed by eight independent instances of clade-specific pseudogenizations or losses of different members of the intronless Grs, Gr22-26 and Gr53, within Heliconius and one instance within Danaus plexippus (Figure 5 and Figure 6). In both Heliconius and Drosophila gene gain and loss appear to primarily affect Grs that are presumed to respond to bitter compounds (Figure 3). To verify whether this pattern holds within the genus Heliconius for the remaining gene family members with more complex intron-exon structure will require better genome assemblies for multiple Heliconius species (Table S2).
These patterns of rapid gene gain and loss are mirrored by within-population variation in copy number. From 16 resequenced genomes for H. melpomene and its sister species H. cydno, we have shown that CNVs occur more commonly among the Grs than the Ors (Figure 7, 8, Table 1). Within the Grs, the bitter receptors of H. melpomene represent a class of genes that are both highly prone to lineage-specific duplication and commonly subject to population-level copy number variation. These putative bitter receptor genes are also more likely to show female-specific expression, especially in the legs, which suggests a role in insect-host chemical interactions (Table 2, Figure 3, Figure S1).
In human genomes, a tendency for CNV-rich areas to display higher dN/dS ratios and yield paralogous genes has been noted [62], along with an enrichment of CNVs in genes involved in immune function and in the senses (specifically in Ors which are unrelated to the insect Ors) [63], [64]. It is also widely known that copy-number variation is an important source of disease-causing mutations in humans [64]. With the exception of insecticide resistance in insects [65], [66], the spectrum of naturally-occurring copy-number variants is only just starting to be explored in Drosophila [67], [68] and non-model systems. Our results demonstrate the great utility of high throughput sequencing to reveal the naturally-occurring spectrum of CNVs that underlie gene family expansions in non-model systems, in traits of ecological relevance.
Heliconius butterflies have complex relationships with their Passifloraceae host plants. Some species are host-specialist, feeding on only one or a few Passiflora species, others specialise on particular sub-genera within Passiflora, while others are generalists, albeit within this one host plant family (Figure 6) [53]. The Passifloraceae is extremely chemically diverse, most notably in their diversity of cyanogenic glycosides that protect the plant from herbivores. It seems likely that coevolution of the butterfly chemosensory and detoxification system on the one hand, with the plant biochemical defense on the other, has played an important role in the evolution of this chemical arsenal. In contrast to the research already carried out on the chemistry of the host plants [54], until recently almost nothing was known about the chemosensory system of Heliconius butterflies. All of these insect host-plant interactions are mediated primarily by adult female butterflies, which must correctly identify suitable host plants for oviposition [69], [70], or risk the survival of their offspring.
Expression data for Grs in the Lepidoptera have been limited until now–especially for adults–due to their low expression level. The largest previous study identified 14 Grs profiled in larval B. mori [32]. We have found evidence for adult expression for most (∼91%) of the 73 predicted Gr genes. This provides a marked contrast to the handful of gustatory receptors that have been identified from traditional expressed sequence tag (EST) projects in other Lepidoptera. Our methods may provide a greatly improved yield of expressed genes because we now have a set of well-annotated target Gr genes against which RNA-seq data can be mapped, together with a greater diversity of transcripts afforded by deep sequencing. Such methods have also permitted us to find widespread expression of their sister gene family, the Ors, in the adult chemosensory tissues examined (68 of 70 or 97% of predicted genes) (Figure 9).
Many of these Gr genes are likely to be involved in the detection of host plant attractants as well as toxic secondary metabolites and thus allow the discrimination of suitable hosts. Most notably, there were a large number of Heliconius-specific Grs with female-biased expression in both legs and antennae (Figure 9). As mentioned previously, these female-biased leg Grs (but not Ors) are also more likely to represent unique duplicates on the Heliconius lineage (Table 2). Female-biased Or expression, as quantified using RNA-seq data, has been reported for Ors expressed in the antennae of the adult mosquito, Anopheles gambiae [71]. Specifically, 22 Ors displayed enhanced expression in mosquito female antennae but not in male antennae. Since adult mosquito females but not males need to find hosts for a blood-meal, and adult butterfly females but not males need to find host plants for egg-laying, this suggests that host-seeking behaviour of female insects may be an important general driver of sensory gene evolution. Indirect evidence for the possible role of some of these Grs in Heliconius host plant detection comes from comparative studies of Grs mediating oviposition behaviour in swallowtail butterflies (Papilionidae). Papilio xuthus PxGr1 a member of the Gr subgroup that contains D. melanogaster Gr43a and HmGr9, has been characterized as a receptor for synephrine, which is an alkaloid found in citrus trees [52]. It is expressed in female P. xuthus tarsi and is necessary for the correct oviposition behavior of swallowtail butterflies [52]. Within the two clades most closely-related to PxGr1, are 9 butterfly-specific Grs: HmGr10, Gr16, Gr55, Gr56 and Gr57, and the newly-described DpGr16, Gr50, Gr52, and Gr54 (Figure 3). Four these Grs, HmGr16, Gr55, Gr56 and Gr57, result from Heliconius-specific gene duplications (i.e., no Danaus or Bombyx homologs). Grs55-57 are also in the top ten most highly expressed Grs in female legs. The identification of these sex-biased leg Grs has provided an important starting point for future ligand specificity studies combining heterologous expression, electrophysiology, RNAi [51], assays of the proboscis-extension reflex, and female oviposition behavior.
Lastly, the patterns of Gr gene expression among different tissues and sexes has permitted us to identify a number of Grs that are strong candidates for mediating the remarkable pollen feeding behaviour that is unique to Heliconius, among the butterflies. The Heliconius proboscis contains at least two types of gustatory sensilla, hair-like sensilla chaetica, and sensilla styloconica (Figure 1). Like other butterflies, Heliconius respond to varying amounts of sugars including sucrose present in floral nectar [72]. Unlike other moths and butterflies, Heliconius actively collect pollen with their proboscides, preferentially from Psychotria (Rubiaceae), Psiguria/Gurania (Cucurbitaceae) and Lantana (Verbenaceae) flowers [17], [18], [73]. Once a pollen load is collected (Figure 1D), the butterflies use a combination of mechanical shearing (coiling and uncoiling of the proboscis) and enzymatic activity (using proteases found in saliva) to release amino acids from the pollen [74]. The RNA-seq data we have collected for H. melpomene proboscis and labial palps should provide a useful resource for future studies examining the molecular basis of this unique digestive trait.
Pollen feeding in adult Heliconius has an important ecological function. Amino acids obtained from pollen are key resources used in male nuptial gifts and egg allocation [18], [75]–[77]. They also permit Heliconius adults to have exceptionally long lifespans. Pollen feeding behavior is not found outside the genus Heliconius, even in the sister genus Eueides, whose larvae share a preference for Passiflora host-plants with Heliconius. In the present study we have identified four Heliconius-specific Grs that are only expressed in the proboscis (HmGr12, Gr20, Gr35 and Gr59) but not in antennae or legs (Figure 9B), suggesting a role for these genes in pollen-feeding behaviour.
Taken together, the whole-genome and whole-transcriptome data suggest that Gr genes in particular are highly evolutionarily labile both on short and long evolutionary timescales, and begin to offer an insight into the likely molecular basis for the rapid coevolution observed between these butterflies and their host plants. Understanding the remarkable diversity underlying this ecological interaction at a molecular level has remained a challenge (but see [32], [52], [78], [79]). Thanks to technological innovations in sequencing, the genetic basis of taste and olfaction involved in host-plant adaptation in Heliconius is beginning to be uncovered.
We have shown that like the opsin visual receptors [80], the chemosensory superfamily composed of constituent Gr and Or families in Lepidoptera show rapid gene family evolution, with higher rates of copy-number variation and gene duplication among the Grs than the Ors, as well as gene losses in the Grs. In particular, there is a group of putative bitter receptors that show female-specific expression in the legs and that are especially prone to gene duplication, providing new material for sensory diversification in the insect-host plant arms race. We have also shown, for the first time, widespread expression of Ors in non-antennal tissues in a lepidopteran. With the most comprehensive data set on Gr and Or expression in butterflies to date we are one step closer to identifying the sensory and molecular genetic basis of the Heliconius-Passiflora co-evolutionary race that inspired Ehrlich and Raven in 1964.
tBLASTn searches were conducted iteratively against the H. melpomene melpomene genome (version v1.1) and haplotype scaffolds [13] using B. mori [28], [47] and D. plexippus Grs [14] as input sequences. For these in silico gene predictions, intron-exon boundaries were identified by first translating the scaffold nucleotides in MEGA version 5 [81], searching for exons identified in the tBLASTn searches, then back translating to identify splice junctions. Intron sequences were then excised to verify that the remaining exonic sequences formed an in-frame coding sequence. Insect Grs are defined by a conserved C-terminal motif TYhhhhhQF, where ‘h’ is any hydrophobic amino acid [21]. We inspected our predicted protein sequences for this motif or variants thereof, specifically ‘S’, ‘M’ or ‘K’ instead of a ‘T’ or ‘L’, ‘T’ or ‘I’ instead of ‘F’. In the handful of cases where we were unable to find the last short exon that contains this motif, final assignment to the Gr gene family was based on using the predicted amino acid sequence as a search string for either tBLASTn or BLASTp against the nr/nt Genbank database. Gene annotations were submitted to the EnsemblMetazoa database http://metazoa.ensembl.org/Heliconius_melpomene/Info/Index as part of the H. melpomene v. 2 genome release (for GeneIDs see Table S1). Chromosomal assignments were based on published mapping of scaffolds in the H. melpomene melpomene reference genome [13].
Following amino acid alignment using ClustalW, preliminary phylogenetic trees were constructed in MEGA using neighbor-joining and pair-wise deletion to identify orthologous relationships with B. mori and D. plexippus Grs. Reciprocal tBLASTn searches against the B. mori and D. plexippus genomes as well as searches using the protein2genome module in EXONERATE [82] were then performed in order to search for ‘missing’ Grs in those genomes. Final phylogenetic analysis was performed using a maximum-likelihood (ML) algorithm and JTT model on an amino acid alignment that was inspected by eye and manually adjusted. These results were compared to a ML tree made from a Clustal-Omega alignment [83] and were found to be nearly identical. Once the initial H. melpomene Gr gene predictions were obtained, EXONERATE, Perl scripts and manual annotations in Apollo [84] were used to produce gff3 files for submission of the annotated H. melpomene genome scaffolds to EMBL-EBI.
Butterfly pupae of H. melpomene rosina were obtained from Suministros Entomológicos Costarricenses, S.A., Costa Rica. Adult males and females were sexed and frozen at −80°C. Total RNAs were extracted separately from antennae, proboscis together with labial palps, and all six legs of three males and three females of H. melpomene using Trizol (Life Technologies, Grand Island, NY). A NucleoSpin RNA II kit (Macherey-Nagel, Bethlehem, PA) was used to purify total RNAs. Each total RNA sample was purified through one NucleoSpin RNA II column. Purified total RNA samples were quantified using a Qubit 2.0 Fluorometer (Life Technologies, Grand Island, NY). The quality of the RNA samples was checked using an Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA). 0.3–4.0 µg of purified total RNAs were used to make cDNA libraries. A TruSeq RNA sample prep kit (Illumina, San Diego, CA) was used to prepare 18 individual cDNA libraries. After being normalized according to their concentrations, the enriched individual libraries were pooled and then run on a 2% agarose gel. cDNA products ranging from 280 to 340 bp with an average of 310 bp were cut out and purified using a Geneclean III kit (MP Biomedicals, Solon, OH) to facilitate post-sequencing assembly. After being re-purified using Agencourt AMPure XP magnetic beads (Beckman Coulter Genomics, Danvers, MA), the cDNA pool was quantified using the Qubit 2.0 Fluorometer, and quality control-checked using the Agilent Bioanalyzer 2100. The cDNA pools were then normalized to 10 nM and run as either two paired-end or three single-end 100 bp runs on a HiSeq 2000 (Illumina, San Diego, CA) by the UCI Genomics High-Throughput Facility.
mRNA sequences were demultiplexed, trimmed and sorted using Python and Perl scripts. A single de novo assembly of the combined libraries was performed using CLC Genomics Workbench 5 to check for missing exons in our gene models. The 73 corrected Gr gene models and 70 Or gene models were then used as an alignment reference to perform unique read mapping of each individual chemosensory transcriptome. To determine if an individual Gr or Or was expressed in a given tissue, each of the 1716 individual Gr and Or mapping alignments was inspected by eye for uniquely mapped reads, and any spuriously-mapped reads (i.e., reads <70 bp in length with indels or sequence mismatches at the ends) were discarded. As a control for potential differences in RNA preparation between samples, we also quantified the number of uniquely mapped fragments to the widely-expressed elongation factor 1-alpha (EF1α) gene transcript and calculated the Fragments Per Kilobase of transcript per Million mapped reads (FPKM) [85]. Illumina reads for each of the libraries were deposited as fastq files in the ArrayExpress archive under the accession number: E-TAB-1500 (Table S7).
One week old adult H. melpomene rosina butterflies were sexed, frozen at −80°C, then dissected and mounted for imaging on an FEI/Philips XL30 FEG scanning electron microscope at UCI's Materials Characterization User Facility. Forelegs, middle legs, hindlegs and antennae were examined for the presence of gustatory sensilla.
We also examined resequenced genomes of twelve H. melpomene and four H. cydno individuals, including H. melpomene aglaope, H. melpomene amaryllis and H. melpomene rosina (Table S4), sequenced by The GenePool, University of Edinburgh, U.K. and the FAS Center, Harvard University, U.S.A., for evidence of copy-number variation (CNV) in the Grs and Ors using CNVnator [56]. These sequences were deposited in the European Nucleotide Archive (ENA) under accession number: ERP002440. The Illumina resequenced genomes were first mapped to the H. melpomene reference genome and the average read depth was calculated along a 100 bp sliding window. The output of CNVnator was parsed for candidate insertion and deletion variants, and those with estimated copy number of >2× were counted as potential duplications and <0.5× as potential deletions.
The GenePool, University of Edinburgh, and the Oxford Genomics Centre, University of Oxford, U.K., produced whole genome 100 bp sequences from H. cydno, H. timareta, H. wallacei, H. doris, H. clysonymus, H. telesiphe, H. erato petiverana, H. sara and H. sapho using the Illumina Pipeline v. 1.5–1.7 with insert sizes ranging from 300 to 400 bp. We deposited sequences for H. sapho and H. sara in the Sequence Read Archive (SRA) under accession number ERP002444. We performed de novo assembly of the short reads using Abyss v. 1.2 [86] implemented in parallel at the School of Life Sciences, University of Cambridge, U.K. Based on previous results [87], recommendations estimated by the software, and comparison of N50 values in preliminary experiments, we chose a k-mer size of 31, a minimum number of pairs required n = 5 and the minimum mean k-mer coverage of a unitig c = 2 (full command: abyss-pe n = 5 k = 31 c = 2 in = ‘for.fastq rev.fastq’). In all assemblies, at least 96% of reads mapped back to the contigs. We created BLAST databases of these whole genome sequence assembly contigs (Table S2) in Geneious Pro v. 5.5.6. The lack of introns in the putative bitter receptor genes Gr22-26 and Gr53 permitted us to easily retrieve them from these BLAST databases. To confirm the identity and improve the quality of the sequences found, we mapped the reads to the assembled exon sequences in CLC Genomics Workbench v. 5.5.1, using the following conservative settings to prevent mis-mapping of paralogous sequences: mismatch, insertion and deletion cost of 3; length fraction and similarity fraction of 0.9. We then inspected all read-mappings by eye. Because the intronless Grs are closely related, we aligned the translated nucleotide sequences in MEGA using the ClustalW algorithm, and also inspected the alignment by eye. For all intronless Gr sequences except for the pseudogenes, sequence length was highly conserved (i.e., there were few indels). To illustrate the high substitution rate of the retrieved pseudogene sequences, we selected the neighbor-joining method for tree reconstruction and performed 500 bootstrap replicates.
To infer the number of intronless Gr gene duplications and losses, we used the program Notung v. 2.6 [88], [89], which reconciles gene trees onto the species tree. The gene tree was made by a maximum likelihood analysis of 1074 nucleotide sites, aligned by Clustal-Omega, and 500 bootstrap replications. The species tree was derived from a phylogeny based on independent nuclear and mitochondrial DNA sequences [90].
We verified the presence of HmGr22 in several adult tissues using reverse-transcriptase PCR and primers for HmGr22 (5′-CCATAATTTTGTCATCCT-3′ and 5′-GATTTCGAAATAAGGTCTGT-3′) and EF1alpha (5′-CGTTTCGAGGAAATCAAGAAGG-3′ and 5′-GACATCTTGTAAGGGAAGACGCAG 3′). RNA was extracted from fresh frozen specimens using Trizol and purified using the Nucleospin RNA II kit, which contains a DNAase-treatment step. RNA concentration was diluted to 12.5 µg/ml. Each 25 µl reaction had 2.5 µl 10× BD Advantage 2 PCR buffer, 2.5 µl dNTPs (2 mM), 0.5 µl (100 µM) forward and 0.5 µl reverse primer, 0.5 µl (1∶20 diluted) Stratagene Affinity Script Reverse Transcriptase, 0.5 µl 50× Advantage 2 Polymerase Mix, 17 µl H2O and 1 µl RNA. The PCR reaction consisted of 38 cycles of 95°C for 30 s, 55°C for 30 s, and 68°C for 55 s. The identity of the RT-PCR products was confirmed by Sanger sequencing.
|
10.1371/journal.ppat.1003215 | Th2 Cell-Intrinsic Hypo-Responsiveness Determines Susceptibility to Helminth Infection | The suppression of protective Type 2 immunity is a principal factor driving the chronicity of helminth infections, and has been attributed to a range of Th2 cell-extrinsic immune-regulators. However, the intrinsic fate of parasite-specific Th2 cells within a chronic immune down-regulatory environment, and the resultant impact such fate changes may have on host resistance is unknown. We used IL-4gfp reporter mice to demonstrate that during chronic helminth infection with the filarial nematode Litomosoides sigmodontis, CD4+ Th2 cells are conditioned towards an intrinsically hypo-responsive phenotype, characterised by a loss of functional ability to proliferate and produce the cytokines IL-4, IL-5 and IL-2. Th2 cell hypo-responsiveness was a key element determining susceptibility to L. sigmodontis infection, and could be reversed in vivo by blockade of PD-1 resulting in long-term recovery of Th2 cell functional quality and enhanced resistance. Contrasting with T cell dysfunction in Type 1 settings, the control of Th2 cell hypo-responsiveness by PD-1 was mediated through PD-L2, and not PD-L1. Thus, intrinsic changes in Th2 cell quality leading to a functionally hypo-responsive phenotype play a key role in determining susceptibility to filarial infection, and the therapeutic manipulation of Th2 cell-intrinsic quality provides a potential avenue for promoting resistance to helminths.
| Helminth parasites mount chronic infections in over 1 billion people worldwide, of which filarial nematode infections account for 120 million. A major barrier to the development of protective Th2 immunity lies in the dominant down-regulatory immune responses invoked during infection. Although this immune suppression is linked with a range of Th2 cell-extrinsic immune regulators, the fate of CD4+ Th2 cells during chronic infection, and the role of Th2 cell-intrinsic regulation in defining protective immunity to infection is largely unknown. In this study, we use a murine model of filarial nematode infection to show that as infection progresses the Th2 effector cells responsible for killing helminths become functionally hypo-responsive, developing a phenotype similar to adaptive tolerance or exhaustion, and their ability to clear infection becomes impaired. We further demonstrate that we can therapeutically manipulate the intrinsic functional quality of hypo-responsive Th2 cells via the PD-1/PD-L2 co-inhibitory pathway to reawaken them and enhance resistance to infection. Thus, our data provide the first demonstration that Th2 cell-intrinsic hypo-responsiveness plays a key role in determining susceptibility to helminth infection.
| Protective immunity to helminth parasites takes decades to acquire, if it develops at all, with over 1 billion people harbouring chronic infections [1]. Protection is mediated by the Th2 arm of immunity [2], which is also responsible for causing allergic diseases such as asthma, atopic dermatitis, and allergic rhinitis, and types of fibrosis. A major reason for the failure in anti-helminth Th2 immunity is that the parasites immunosuppress their host, exemplified by host PBMC losing the ability to proliferate and produce Th2 cytokines, such as IL-4 and IL-5, in response to parasite antigen [3], [4], [5]. Interestingly, this Th2 down-modulation has parallels with the modified Th2 response originally described in association with tolerance to allergens, and characterised by a switch from an inflammatory IgE response to an anti-inflammatory IgG4 and IL-10 response [6], [7]. Thus, the regulatory pathways invoked by helminths can cross-regulate and protect against allergic diseases in humans and animal models [8], [9]. As such, defining the mechanisms of immune down-regulation during helminth infections is of importance for the development of therapeutic strategies or vaccines to induce long-term protective anti-helminth immunity, and novel approaches for the treatment of allergies and fibrosis.
Following the observations that neutralisation of IL-10 or TGF-β can restore the immune-responsiveness of PBMC from helminth-infected individuals [10], [11], studies have focussed on determining the extrinsic regulators that control Th2 cell function. From these, a variety of cell types have been shown to inhibit immunity to helminths and allergens [12], including Foxp3+ regulatory T cells (Tregs) [13], [14], alternatively activated macrophages (AAM) [15], [16], DC [17], [18], and B cells [19], [20]. However, the intrinsic fate of parasite-specific CD4+ Th2 cells within a chronic down-regulatory environment is largely unknown, even though the idea that helminth-elicited T cells become anergised during infection was postulated 20 years ago [21]. It is known that CD8+ T cells develop a functionally hypo-responsive phenotype in chronic Th1 infections, termed exhaustion [22], and human helminth studies provide some evidence for the development of a form of Th2 cell-intrinsic dysfunction. PBMC from filariasis patients display a gene expression profile characteristic of anergic T cells [3], and T cells from individuals with chronic nematode infections show defects in TCR signalling [23]. Recently, a murine study on the down-modulation of pathogenic Th2 responses during Schistosoma mansoni infection provided the first formal demonstration that CD4+ Th2 effector cells can develop an intrinsically hypo-responsive phenotype [24]. Thus, there is a question of whether individuals fail to acquire protective immunity to helminths because their Th2 cells become intrinsically dysfunctional.
We previously used a murine model of filariasis, Litomosoides sigmodontis infection of permissive BALB/c mice, to define the immune regulatory mechanisms that prevent helminth killing. We demonstrated that purified CD4+ T cells lose the ability to proliferate and produce Th2 cytokines to parasite antigen as infection progresses [25]. This loss of function within the CD4+ T cell compartment was independent of Foxp3+ Tregs and IL-10 indicating that, alongside extrinsic regulation by Foxp3+ Tregs [24], [25], [26], susceptibility to filarial infection is associated with an intrinsic functional change within the CD4+ Th2 cells. Thus, in this study we employed IL-4gfp 4get reporter mice [27] to track and determine the fate of Th2 cells during L. sigmodontis infection. We found that, whilst increasing in number, the IL-4gfp+CD4+ Th2 cells became conditioned towards a functionally hypo-responsive phenotype as infection progressed denoted by a progressive loss in their intrinsic ability to produce the cytokines IL-4, IL-5 and IL-2. The onset of hypo-responsiveness was accompanied by increased expression of PD-1 by IL-4gfp+ Th2 cells, and in vivo PD-1 blockade led to increased resistance to infection and a long-term increase in Th2 cell functional quality. In contrast to viral and protozoan infections [28], [29], [30], the control of T cell quality by PD-1 was driven through its interactions with PD-L2, and not PD-L1. Thus, intrinsic changes in Th2 cell functional quality play an important role in defining resistance and susceptibility to filarial nematodes, and it is possible to enhance resistance to infection by therapeutically manipulating Th2 cell quality.
Susceptibility to L. sigmodontis infection is associated with a loss of responsiveness by CD4+ T cells at the infection site, the pleural cavity (PC), such that as infection progresses purified PC CD4+ T cells show reduced L. sigmodontis antigen (LsAg)-specific proliferative and cytokine responses in vitro [25], [26]. This down-modulation within the CD4+ T cell population was independent of extrinsic regulation by CD4+CD25+Foxp3+ Tregs or IL-10, suggesting that it represented either a contraction in the number of Th2 cells, or a qualitative change in the intrinsic function of the responding Th2 cells.
Identifying parasite-specific T cells during helminth infection is challenging due to the polyclonal nature of the response and a lack of knowledge of the specific antigens recognised. Thus, to determine whether changes in Th2 cell quantity or intrinsic functional quality explain the observed CD4+ T cell down-modulation during L. sigmodontis infection we employed BALB/c IL-4gfp 4get reporter mice [27]. IL-4gfp+ T cells elicited during acute infection with the nematode Nippostrongylus brasiliensis are parasite specific [31], and so IL-4gfp expression by CD4+ T cells was used as a tool for tracking L. sigmodontis-specific Th2 cells, with a caveat that IL-4gfp is a surrogate marker and a small proportion of IL-4gfp+ T cells may not be L. sigmodontis specific. Importantly, once committed to the Th2 lineage, T cells store IL-4 mRNA and the production of IL-4 protein is controlled post-transcriptionally [32]. IL-4gfp 4get mice report the presence of IL-4 mRNA independently of IL-4 protein [32], [33], meaning that the number of committed IL-4 mRNA+ Th2 cells can be quantified by GFP expression, whilst further assays can be used to independently assess their functional quality.
BALB/c IL-4gfp mice were infected s.c. with L. sigmodontis larvae. From the skin the larvae migrate via the lymphatics to the PC by d 4 post-infection (pi) where they undergo a series of moults reaching the adult stage around d 25 pi, and become sexually mature with the females releasing transmission stage microfilaria (Mf) approximately 55 d pi. A patent infection is defined as having mature adult parasites within the PC, and Mf circulating within the blood stream [34]. At the infection site there was a gradual increase in the proportion of IL-4gfp+ Th2 cells as infection progressed, culminating in 35% of CD4+ T cells expressing GFP by day 60 (Figure 1A and B). This translated to a significant elevation of total numbers of IL-4gfp+ Th2 cells by d 20 pi, which was maintained until d 60 pi (Fig. 1C). Increases in the proportions of IL-4gfp+ Th2 cells were also seen in the thoracic LN (tLN) and spleen (Figure 1D and F), albeit to a lesser extent, and only resulted in significantly increased total numbers of IL-4gfp+ Th2 cell in the tLN (Figure 1E and G). Thus, the down-modulation of CD4+ T cell responsiveness within the PC was not caused by a loss of Th2 cells.
To determine whether Th2 cell functional quality declined during infection, intra-cellular staining was used to define the proportion of IL-4gfp+ Th2 cells actively producing IL-4 protein (Figure 2A), as well as IL-5 and IL-2 (Figure S1), in response to PMA and ionomycin stimulation. Contrasting with the increase in numbers of IL-4gfp+ Th2 cells within the PC, there was a 69% reduction in the proportion of IL-4gfp+ Th2 cells making IL-4 protein between d 20 and d 40, which was still apparent at d 60 (Figure 2B). The proportion of IL-4gfp+ Th2 cells producing IL-5 protein also declined by 70%, although with delayed kinetics as the reduction did not occur until d 60, indicating a staggered loss of cytokine production (Figure 2C). Similarly, there was a 79% decrease in the proportion making IL-2 between d 20 and d 60 (Figure 2D). Distal to the PC, the proportion of IL-4gfp+ Th2 cells capable of producing IL-4 protein within the tLN remained unaffected (Figure 2E), and despite a transient decrease at d 40 in the spleen the proportion of IL-4+ IL-4gfp+ Th2 cells at d 60 pi was equivalent to d 20 (Figure 2H). In contrast, the production of IL-5 and IL-2 proteins by IL-4gfp+ Th2 cells was impaired at d 60 in both the tLN and spleen (Figure 2 F, G, I and J).
Thus, the decline in CD4+ T cell responsiveness observed during chronic L. sigmodontis infection represents a step-wise intrinsic loss of functional ability of IL-4gfp+ Th2 cells to produce cytokines, rather than a decrease in the total number or proportion of Th2 cells. This hypo-responsive Th2 cell phenotype is most prominent at the infection site, but to a lesser extent radiates out to the draining LN and spleen.
Co-inhibition through the PD-1 pathway leads to the functional exhaustion of CD8+ T cells during chronic immune challenge [22], [28], and is involved in the inhibition of Th2 responses during helminth infections [35], [36], [37], [38]. To investigate whether L. sigmodontis-induced Th2 cell hypo-responsiveness was associated with PD-1 co-inhibition, the expression of PD-1 by IL-4gfp+ Th2 cells was assessed. At d 20 pi, when the IL-4gfp+ Th2 cells were still functionally active, there was no change in the proportion of PC IL-4gfp+ Th2 cells expressing PD-1 (Figure 3A). However, concomitant with the onset of hypo-responsiveness, there was a two-fold increase in the proportion of IL-4gfp+ Th2 cells expressing PD-1 at d 40 pi. PD-1 expression remained elevated until d 60 (Figure 3A), and the mean fluorescence intensity of PD-1 expression on IL-4gfp+ Th2 cells increased with similar kinetics (Figure 3B). When PD-1 expression by IL-4gfp+ Th2 cells was compared to production of IL-4 protein, the majority of IL-4 producing Th2 cells at d 20 were PD-1 negative, and the enhanced PD-1 expression at d 40 and 60 associated with the loss of IL-4 protein (Figure 3C).
T follicular helper (Tfh) cells are IL-4gfp+ and express PD-1 during helminth infection [39], [40], [41], and as expected the increases in tLN IL-4gfp+ T cells observed during L. sigmodontis infection represented an expansion of both IL-4gfp+CXCR5− Th2 cells and IL-4gfphighCXCR5+ Tfh cells (Figure 3D). IL-4gfphighCXCR5+ Tfh cells from naïve mice constitutively expressed high levels of PD-1 and there was no change upon infection (data not shown). In contrast, infection with L. sigmodontis significantly increased the percentage of IL-4gfp+CXCR5− Th2 cells expressing PD-1 from d 40 onwards (Figure 3E).
To test the hypothesis that blocking PD-1 signalling on the hypo-responsive Th2 cells in vitro could re-activate their functional responsiveness, IL-4gfp+ Th2 cells were purified from the PC 60 d post-L. sigmodontis infection and restimulated with LsAg in the presence of an anti-PD-1 blocking mAb [42] using irradiated naïve splenocytes as APC. Due to the low number of IL-4gfp+ T cells present within the PC it was necessary to pool cells from 10–15 mice to obtain sufficient cell numbers. Significantly increased LsAg-specific proliferation and elevated IL-5 production was seen upon addition of the anti-PD-1 mAb (Figure 3F and G), indicating that PD-1 blockade can restore the function of committed Th2 cells. Thus, Th2 cell hypo-responsiveness is associated with increased expression of PD-1 by IL-4gfp+ Th2 cells in both the PC and tLN, and blocking PD-1 in vitro increases the antigen-specific capacity of PC IL-4gfp+ Th2 cells to proliferate and produce IL-5.
To directly test whether co-inhibition through the PD-1 pathway inhibits protective immunity, L. sigmodontis infection was allowed to establish within susceptible BALB/c mice and PD-1 activity blocked from d 28–43 pi using a neutralising anti-PD-1 mAb (Figure 4A). PD-1 blockade impaired the ability of L. sigmodontis to develop a fully patent infection as the incidence and levels of blood Mf were significantly reduced in anti-PD-1 treated mice compared to control mice at d 68 pi (Figure 4B and Table 1). As there was no effect of treatment on the number of adult parasites recovered at d 60 pi (Figure 4C), we scored the uterine egg and Mf contents of female parasites to determine whether PD-1 blockade reduced blood Mf by inhibiting fecundity. Changes in helminth fecundity are often a sensitive and quantitative measure of the efficacy of host immunity, even when it is insufficient to kill adult parasites [43]. There was a significant reduction in the number of healthy uterine eggs within female parasites from PD-1 treated mice at d 60 pi (Figure 4D). In addition, the number of female parasites with Mf within their uteri was reduced three-fold following PD-1 treatment (Table 1), although those with uterine Mf tended to have similar levels as female parasites from the IgG controls (Figure S2A). Thus, in vivo PD-1 blockade promotes host resistance to established L. sigmodontis infection resulting in impaired fitness and fecundity in a proportion of the female parasites, and reduced levels and incidence of circulating transmission stage Mf within the host's blood.
PD-1 blockade could enhance protective immunity to L. sigmodontis by restoring the functional quality of hypo-responsive Th2 cells and/or by increasing the overall quantity of Th2 cells. To address this we treated L. sigmodontis infected BALB/c IL-4gfp reporter mice with an anti-PD-1 blocking mAb from d 28–43 pi and quantified the number and antigen-responsiveness of IL-4gfp+ Th2 cells at d 60 pi (Figure 4A). No differences were found in the proportion or total number of IL-4gfp+ Th2 cells within the PC or tLN at d 60 pi (Figure 5A–D), indicating that PD-1 blockade does not result in a long-term elevation in Th2 cell numbers.
To examine if PD-1 blockade increased the antigen-specific functional quality of the hypo-responsive Th2 cells, IL-4gfp+ Th2 cells were purified from the PC and tLN of control and PD-1 treated mice 60 d pi and equal numbers of Th2 cells were restimulated in vitro with LsAg using naïve irradiated splenocytes as APC. Due to the low number of IL-4gfp+ T cells present within the PC it was necessary to perform the assays on pooled cells from 6–10 mice. PC IL-4gfp+ Th2 cells purified from anti-PD-1 treated mice showed significantly increased antigen-specific proliferation and elevated IL-5 production compared to IL-4gfp+ Th2 cells from control mice (Figure 5E and F). PD-1 blockade also increased the capacity of purified tLN IL-4gfp+ Th2 cells to secrete IL-5 in response to LsAg at d 60 (Figure 5G). LsAg-specific production of IL-4, IL-10 and IFN-γ by tLN and PC IL-4gfp+ Th2 cells was not consistently detectable (data not shown). Thus, PD-1 blockade results in a long-term enhancement of Th2 immunity by augmenting the functional quality of parasite-specific Th2 cells, rather than increasing the overall number of Th2 cells.
During chronic viral infections PD-1 blockade acts directly on exhausted CD8+ T cells to restore their function [28]. Similarly, our data showed that in vitro PD-1 blockade directly enhanced the functional quality of L. sigmodontis-specific hypo-responsive IL-4gfp+ Th2 cells (Figure 3F and G), and in vivo blockade led to increased antigen-specific responsiveness of IL-4gfp+ Th2 cells 20 d after treatment had finished. This predicts that in vivo PD-1 blockade during L. sigmodontis infection would initially enhance the functional quality of existing IL-4gfp+ Th2 effector cells within the PC. However, when PC Th2 cell responses were assayed immediately following treatment (d 40 pi, Figure 4A) there were no increases in the proportion or total numbers of IL-4gfp+ Th2 cells within the PC (Figure 6A and B). There was also no increase in the proportion of IL-4gfp+ Th2 cells capable of producing IL-4 or IL-5 protein following stimulation with PMA and ionomycin (Figure 6C and D).
In contrast, PD-1 blockade caused a 2-fold increase in the proportion and total numbers of IL-4gfp+ T cells within the tLN (Figure 6E and F), although again had no impact on the proportion of IL-4gfp+ Th2 cells producing IL-4 or IL-5 protein (Figure 6G and H). Consistent with the expression pattern of PD-1, the increase in tLN IL-4gfp+ T cells was caused by an expansion of both IL-4gfp+CXCR5− Th2 cells and IL-4gfphighCXCR5+ Tfh cells, with IL-4gfp+CXCR5− Th2 cells accounting for 44% of the expansion (data not shown). These data suggest that, although in the long-term PD-1 blockade increased the antigen-specific functional quality of IL-4gfp+ Th2 cells at the infection site, anti-PD-1 treatment did not recover the responsiveness of IL-4gfp+ Th2 cells to PMA and ionomycin. Instead it initially enhanced Type 2 responses within the tLN resulting in a temporary increase in the number of IL-4gfp+ Th2 and Tfh cells.
PD-1 interacts with two ligands, PD-L1 and PD-L2 [44], and to identify through which ligand PD-1 was acting L. sigmodontis infected BALB/c mice were treated with blocking anti-PD-L1 and/or anti-PD-L2 mAbs [45] from d 28–43 pi (Figure 4A). At d 40 pi, when PD-1 blockade results in an expansion of IL-4gfp+ Th2 cells within the tLN (Figure 6E and F), significantly increased numbers of IL-4 and IL-5 producing CD4+ T cells were found following combined blockade of PD-L1 and PD-L2 (Figure 7A and B). Blockade of PD-L1 or PD-L2 alone had no effect on the numbers of IL-4 and IL-5 producing CD4+ T cells, indicating that PD-L1 and PD-L2 synergise to regulate the expansion of IL-4 and IL-5 secreting T cells within the tLN.
In contrast, blockade of PD-L2 alone was sufficient to enhance resistance and impair the development of patent infections, demonstrated by a significant two-fold reduction in the incidence of mice with Mf within their blood at d 68 pi compared to IgG controls (Table 2). This was associated with a significantly reduced incidence of Mf in the pleural cavity, suggesting a lower release of Mf by female parasites. Interestingly, although mice treated with both anti-PD-L1 and anti-PD-L2 had a lower incidence of blood Mf, PD-L1 neutralisation alone resulted in a trend towards an increased number of mice harbouring blood Mf. This suggests that PD-L1 may promote rather than inhibit protective immunity to L. sigmodontis, but that the inhibitory role of PD-L2 is dominant. Whilst anti-PD-L2 treatment significantly reduced the number of mice developing patent infections, those that presented with circulating Mf had similar levels within the blood and pleural cavity to the IgG control group (Figure S2B and C). The increased resistance resulting from blockade of PD-L2 was associated with significantly elevated production of IL-5 protein (Figure 7C), as well as IL-4 (Figure 7D), by tLN cells restimulated in vitro with LsAg. Thus, while PD-L1 and PD-L2 act synergistically to control Th2 cell expansion in the tLN, PD-L2 plays the dominant role in dampening IL-4 and IL-5 production and resistance towards L. sigmodontis.
Alternatively activated macrophages can inhibit Th2 responses to helminths through PD-L1 and PD-L2 [37] or PD-L2 alone [35]. Expression of PD-L1 on macrophages, independent of alternate activation, also inhibits T cell responses to S. mansoni [36]. AAM are elicited during L. sigmodontis infection resulting in T cell suppression [16], suggesting that they could drive the hypo-responsive Th2 cell phenotype via PD-L1 or PD-L2. To test this, the expression of PD-L1 and PD-L2 on F4/80high pleural cavity AAM was assessed during L. sigmodontis infection. Analysis was performed at d 60 pi when PC derived F4/80high macrophages are known to be alternatively activated [16]. Although F4/80high macrophages from naïve mice did not express PD-L1 or PD-L2 constitutively, levels of both were up-regulated 11-fold following infection (Figure 8A and B). To determine whether AAM inhibit Th2 cells via PD-L1 and/or PD-L2, we purified and restimulated d 60 PC IL-4gfp+CD4+ Th2 cells with LsAg in the presence of d 60 PC AAM or naïve control macrophages. Irradiated naïve splenocytes were used as APC. The ability of the hypo-responsive IL-4gfp+ Th2 cells to proliferate in response to LsAg was then assessed following the addition of neutralising antibodies to PD-1, PD-L1 and PD-L2. In the presence of AAM, the LsAg-specific proliferation of IL-4gfp+CD4+ Th2 cells was significantly reduced compared to culture with naïve control macrophages confirming that the AAM were suppressive (Figure 8C). However, blocking PD-1, PD-L1, PD-L2, or a combination of PD-L1 and PD-L2, failed to restore LsAg-specific proliferation (Fig. 8C), indicating that AAM-mediated suppression of proliferation was not via the PD-1 pathway. Similarly, L. sigmodontis-elicited AAM inhibited the OVA-specific proliferation of naïve DO11.10 T cells independently of PD-1, PD-L1 and PD-L2 (Figure 8D). Consistent with our previous work showing that AAM only inhibit T cell proliferation and not cytokine production [16], the addition of AAM did not reduce Th2 cell production of IL-5 or IL-4 (data not shown). Thus, L. sigmodontis-elicited AAM do not suppress the antigen-specific proliferation of committed Th2 cells or naïve T cells via the PD-1 pathway.
Suppression of protective immunity during helminth infections is known to involve a wide range of Th2 cell extrinsic immune regulators [2], [12]. However, the intrinsic fate of parasite-specific Th2 cells within a chronic immune down-regulatory environment, and the resultant impact such fate changes may have on host resistance is unknown. In this study we used IL-4gfp reporter mice to demonstrate that during chronic filarial nematode infection CD4+ Th2 cells are conditioned towards an intrinsically hypo-responsive phenotype, characterised by a loss of functional ability to proliferate and produce IL-4, IL-5 and IL-2 cytokines. The development of Th2 cell hypo-responsiveness was a key element in determining susceptibility to L. sigmodontis infection, and could be reversed in vivo by blockade of the PD-1/PD-L2 pathway resulting in the long-term recovery of Th2 cell functional quality and enhanced resistance.
The hypo-responsive Th2 cell phenotype during L. sigmodontis infection had some parallels with T cell exhaustion, which leads to impaired Th1 immunity towards viruses [28] and protozoan parasites [29], [30]. Similar to exhaustion, Th2 cell hypo-responsiveness was mediated through PD-1 co-inhibition, and was characterised by a sequential loss of cytokine production with IL-4 being lost prior to IL-5. In the future it will be important to confirm whether Th2 cell hypo-responsiveness also extends to other Th2 cytokines, such as IL-13, as limited cell numbers restricted the cytokines we could measure in this study. Alongside the similarities to exhaustion there were also notable differences. PD-1 predominantly mediates CD8+ T cell exhaustion via interactions with PD-L1 [28], [30], whereas Th2 hypo-responsiveness was driven through interactions with PD-L2. This preferential regulation of Type 2 immunity by PD-L2 is consistent with its expression being specifically induced by IL-4 and STAT-6 signalling, contrasting with PD-L1, which is preferentially regulated by Type 1 stimuli [46], [47]. Also, PD-1 interactions with PD-L1 are actively required for maintaining exhaustion, with PD-L1 blockade immediately boosting the functional quality of exhausted CD8+ T cells [28]. Although in vitro PD-1 blockade enhanced the antigen-specific ability of hypo-responsive Th2 cells to produce Th2 cytokines, stimulating them with PMA and ionomycin, which bypasses the PD-1 pathway, failed to recover their function. Similarly, the Th2 cells remained hypo-responsive to PMA and ionomycin stimulation immediately following in vivo PD-1 blockade, although it led to a recovery in their functional ability to respond to parasite antigens later in infection. Thus, the hypo-responsive Th2 cell phenotype is likely distinct from exhaustion, and appears to be more deep-seated, involving more mechanisms, than PD-1 co-inhibition alone.
The disparity between GFP expression, which marks IL-4 mRNA, and IL-4 protein suggests that the loss of cytokine production by hypo-responsive Th2 cells is due to post-transcriptional regulation, which is an essential step in the production of Th2 cytokines [32]. This has parallels with anergic self-reactive T cells that express mRNA for effector cytokines such as IFN-γ, IL-4, and IL-13, but are unable to produce protein because translation is blocked by AU-rich elements within the cytokine 3′UTRs [48]. The description of an anergic molecular signature within the PBMC of filariasis patients and the findings that addition of IL-2 can restore the in vitro immune responsiveness of human PBMC [3], [49], reinforce the idea that Th2 hypo-responsiveness is a form of anergy. If so, it is more likely to represent a form of adaptive tolerance than classical clonal anergy as it is not rescued by stimulation with PMA and ionomycin, and results in the shutdown of multiple cytokines, not just IL-2 [50]. Similarly, during S. mansoni infection the anergy factor GRAIL is responsible for driving Th2 cells towards an intrinsically hypo-responsive state with characteristics of adaptive tolerance [24]. Interestingly, there are differences in Th2 cell hypo-responsiveness during filariasis and schistomiasis. Firstly, GRAIL is not part of the anergic signature of PBMC from filariasis patients [3]. Secondly, S. mansoni induced Th2 cell hypo-responsiveness does not relate to PD-1 [24]. Thus, while an intrinsic functional shut-down of Th2 cells appears common to different chronic helminth infections, it may involve distinct mechanisms.
Consistent with the hypothesis that multiple factors maintain Th2 cell hypo-responsiveness, in vivo PD-1 blockade failed to initially expand or recover the function of the hypo-responsive Th2 cells at the infection site. Instead it first caused a temporary expansion of CXCR5−IL-4gfp+ Th2 cells and CXCR5+ IL-4gfp+ Tfh cells within the draining LN, followed by the appearance of functionally superior IL-4gfp+ Th2 cells at the infection site 20 days later. In contrast to the PC, CD4+ T cells in the LN did not lose the ability to produce IL-4 protein during L. sigmodontis infection. Tfh cells are the predominant source of IL-4 in the LN and demonstrate distinct control of IL-4 gene expression compared to Th2 cells [39], [40], [41], [51]. Thus, similar to viral infections where Tfh cells do not become exhausted [52], Tfh cells may remain functionally responsive during chronic helminth infection and maintain a source of IL-4. It is interesting to speculate that, rather than directly rescuing the hypo-responsive Th2 cells, PD-1 blockade acted by expanding a reservoir of still responsive IL-4gfp+ T cells within the LN, either Tfh or Th2 cells, that over time replaced the unresponsive Th2 cells at the infection site. Alternatively, PD-1 can inhibit T cell priming [53] and so its blockade may have favoured the generation of new responsive Th2 cells.
The involvement of PD-L2, rather than PD-L1, indicates that professional immune cells regulate CD4+ Th2 cell hypo-responsiveness. Suppression of T cell responses by PD-1 during helminth infections has mainly been attributed to macrophages expressing PD-L1 and/or PD-L2, and the PD-1 pathway has been shown to be an important mechanism of suppression by AAM [35], [36]. Although L. sigmodontis infection induces suppressive AAM [16], the proliferative suppression of Th2 cells and naïve T cells by L. sigmodontis-elicited AAM was independent of PD-1, PD-L1 and PD-L2. Thus, whilst AAM are clearly able to suppress T cells via the PD-1 pathway they do not do so in all Th2 contexts, and it is not their dominant mechanism of suppression during L. sigmodontis infection. Furthermore, as we have previously shown [16], suppression by L. sigmodontis-elicited AAM was restricted to T cell proliferation and did not inhibit the production of Th2 cell cytokines. As reduced Th2 cytokines were a defining characteristic of hypo-responsive Th2 cells it indicates that AAM are not driving hypo-responsiveness, although in vivo AAM studies are required to confirm our in vitro findings. Interestingly, B cell deficient mice are more resistant to primary L. sigmodontis infection indicating a regulatory role for B cells [54]. B cells can express PD-L2 [44] and up-regulate it during L. sigmodontis infection (van der Werf, Taylor, unpublished data) raising the possibility that B cells are involved in conditioning Th2 cells towards hypo-responsiveness. Alternate candidates that may influence the intrinsic functional quality of Th2 cells include DC and Foxp3+ Tregs.
The development of functionally impaired CD4+ Th2 cells provides a potential explanation for why protective memory to helminths takes decades to develop in humans [14]. Hypo-responsive Th2 cells may fail to develop into memory cells as seen with exhausted CD8+ T cells [22], and consistent with this filariasis patients show contractions in their central memory CD4+ T cell pool [55]. Alternatively, a tolerised memory response may develop as anergic T cells can show long-term survival and maintain their unresponsive phenotype even in the absence of antigen [56]. A failed or tolerised memory response may also explain why helminth-infected individuals become rapidly re-infected following drug clearance, even though some aspects of immune suppression are lifted. PD-1 blockade in combination with drug treatments may thus represent a new strategy for restoring protective Th2 memory, particularly as we find PD-1 blockade has a long-term effect on Th2 cell quality and it has been successfully used in clinical trials to treat cancer [57], [58]. Alternate targets include GITR, as providing co-stimulation through GITR increases the functional quality of L. sigmodontis specific Th2 cells [59], and CTLA-4, which promotes the expression of T cell anergy factors and inhibits protective Th2 immunity during filarial infections [3], [26]. The development of Th2 hypo-responsiveness also has implications for vaccine development. Even the best live-attenuated filarial vaccines are only 70% effective [60], meaning that residual infections could condition vaccine-elicited Th2 cells towards hypo-responsiveness resulting in vaccine failure.
Altogether, our data demonstrates that intrinsic changes in Th2 cell quality lead to the development of a functionally hypo-responsive phenotype that plays a key role in determining susceptibility to filarial nematode infection, and that can be therapeutically manipulated to promote resistance. Alongside its relevance to the treatment of helminth infections, a deeper understanding of how Th2 cells are conditioned towards hypo-responsiveness will help define the checkpoints that determine whether a T cell remains inflammatory or becomes tolerised during chronic immune challenge. This may help determine why Th2 cells fail to shutdown naturally in settings of chronic pathology, such as in allergic inflammation or fibrosis, and potentially lead to novel approaches for tolerising pathogenic Th2 cells.
All animal work was approved by the University of Edinburgh Ethics Committee (PL02-10) and by the UK Home Office (PPL60/4104), and conducted in accordance with the Animals (Scientific Procedures) Act 1986.
Female BALB/c and IL-4gfp 4get reporter mice on the BALB/c background (courtesy of Markus Mohrs, The Trudeau Institute) [27] were bred in-house and maintained under specific pathogen-free conditions at the University of Edinburgh. Mice were used at 6–12 weeks of age. The L. sigmodontis life cycle was maintained in gerbils using the mite vector Ornithonyssus bacoti [61]. Mice were infected s.c. on the upper back with 30 L. sigmodontis L3 larvae. Adult parasites were recovered by lavage and fixed in 70% ethanol for morphological analysis. The analysis of fecundity of female L. sigmodontis parasites was performed as previously [43]. The numbers of healthy eggs and Mf within the anterior, median and posterior of the uterus were semi-quantitatively scored on scales of 0–5. Each region's scores were summed giving a total possible score of 15. To quantify blood Mf, 30 µL of tail blood was collected in FACS lysing solution (Becton-Dickinson). L. sigmodontis antigen (LsAg) was prepared by collecting the PBS-soluble fraction of homogenized adult male and female worms.
Mice received i.p. injections of 250 µg of blocking anti-PD-1 mAb (RMP1-14, Bioxcell), 250 µg of blocking anti-PD-L2 mAb (Ty25, Bioxcell) or 200 µg of blocking anti-PD-L1 mAb (MIH5, in house) every three days from d28–43 pi An equivalent dose of rat IgG (Sigma-Aldrich) was used as control.
The parathymic, posterior, mediastinal and paravertebral LN, were taken as a source of tLN draining the PC. PC cells were recovered by lavage. TLN and spleen cells were dissociated and washed in RPMI-1640 (invitrogen) supplemented with 0.5% mouse sera (Caltag-Medsystems), 100 U/ml penicillin, 100 µg/ml streptomycin and 2 mM L-glutamine. To purify GFP+CD4+ T cells from IL-4gfp mice PC or tLN cells were enriched for CD4+ T cells by magnetic negative selection (DynaMag, Dynal) using anti-CD8 (53–6.72), anti-B220 (RAB632), anti-MHC class II (M5/114.15.2), anti-Gr1 (RB6-8C5) and anti-F4/80 (A3-1), followed by sheep anti-rat IgG Dynal Beads (Invitrogen). Cells were stained with allophycocyanine-conjugated anti-CD4 (RM4-5). To purify GFP+CD4+ T cells from anti-PD-1 treated mice cells were stained with phycoerythrin-conjugated anti-CD4 followed by positive magnetic section with anti-phycoerythrin MicroBeads (Milenyi Biotec). GFP+CD4+ T cells were then purified using a FACSAria flow sorter (Becton-Dickinson). On average, sorted cells were 98.3% positive for CD4, of which 97.6% were GFP+. Due to limited cell numbers it was necessary to pool CD4+GFP+ T cells from 10–15 mice to obtain sufficient numbers. Whole tLN cells were cultured at 5×105 cells/well and spleen cells at 1×106 cells/well in 96 well plates (Nunc). Purified GFP+CD4+ T cells were cultured at 5–10×104 cells/well with 1×106 irradiated (30 Gy) naïve splenocytes. For in vitro restimulations, cells were cultured in medium alone or with 10 µg/ml LsAg for 72 hours followed by addition of 1 µCi/well [Methyl-3H]-Thymidine (PerkinElmer) for 16 h to measure proliferation. Blocking antibodies against PD-1, PD-L1 and PD-L2 were used at 20 µg/ml as detailed in the results. For macrophage suppression assays, PC cells were adhered to 96-well flat-bottom plates at 1×105 cells/well for 2 h at 37°C and the non-adherent fraction rinsed off. GFP+CD4+ T cells or DO11.10 CD4+ T cells were added at 5×104 cells/well and after 72 h the cultures were pulsed with thymidine as described. DO11.10 cells were restimulated with 0.5 µg/ml OVA peptide (ISQAVHAAHAEINEAGR) from Advanced Biotechnology Centre (Imperial School of Medicine, London, U.K.). For measurement of intra-cellular cytokines cells were stimulated for 4 hours with 0.5 µg/ml PMA (Sigma-Aldrich) and 1 µg/ml Ionomycin, with 10 µg/ml Brefeldin A added for the final 2 hours (all from Sigma-Aldrich).
The following antibodies were used: Alexafluor700-conjugated anti-CD4 (RM4-5), polyclonal anti-GFP (Ebioscience), Alexafluor488-conjugated goat anti-rabbit IgG (Invitrogen), eFluor450-conjugated anti-IL-2 (JES6-5H4, Ebioscience), phycoerythrin-conjugated anti-IL-4 (11B11, Biolegend), allophycocyanine-conjugated anti-IL-5 (TRFK5, Biolegend), phycoerythrin-conjugated or biotinylated anti-PD-1 (J43, Ebioscience), biotinylated anti-PD-L1 (MIH5, Ebioscience), phycoerythrin-conjugated anti-PD-L2 (Ty25, Ebioscience), biotinylated anti-CXCR5 (RF8B2, BD Biosciences) and allophycocyanine-conjugated streptavidin (Biolegend). Non-specific binding was blocked with 4 µg of rat IgG/1×106 cells. For intracellular cytokine staining dead cells were excluded using Aqua Dead Cell Stainkit (Molecular Probes), and the cells fixed and permeabilized using the BD Cytofix/Cytoperm kit. Staining was compared with the relevant isotype controls to verify specificity. Flowcytometric acquisition was performed on a FACSCANTO II or LSR II (BD Biosciences) and data were analyzed using Flowjo Software (Tree Star). Antibody pairs used for cytokine ELISA were as follow: IL-4 (11B11/BVD6-24G2) and IL-5 (TRFK5/TRFK4). Recombinant murine IL-4 and IL-5 (Sigma-Aldrich) were used as standards. Biotin detection antibodies were used with ExtrAvidin-alkaline phosphatase conjugate (Sigma-Aldrich) and Sigma Fast p-nitrophenyl phosphate substrate (Sigma-Aldrich).
Statistical analysis was performed using JMP version 8 (SAS). Parametric analysis of combined data from multiple repeat experiments, or of experiments containing more than two groups, was performed using ANOVA followed by Tukey's post-hoc tests when required. When using two-way ANOVA to combine data from multiple experiments, experimental effects were controlled for in the analysis and it was verified that there were no significant qualitative interactions between experimental and treatment effects. Mf incidence was analysed using a GLM with a binomial distribution.
|
10.1371/journal.ppat.1000419 | Unique Structure and Stability of HmuY, a Novel Heme-Binding Protein
of Porphyromonas gingivalis | Infection, survival, and proliferation of pathogenic bacteria in humans depend on
their capacity to impair host responses and acquire nutrients in a hostile
environment. Among such nutrients is heme, a co-factor for oxygen storage,
electron transport, photosynthesis, and redox biochemistry, which is
indispensable for life. Porphyromonas gingivalis is the major
human bacterial pathogen responsible for severe periodontitis. It recruits heme
through HmuY, which sequesters heme from host carriers and delivers it to its
cognate outer-membrane transporter, the TonB-dependent receptor HmuR. Here we
report that heme binding does not significantly affect the secondary structure
of HmuY. The crystal structure of heme-bound HmuY reveals a new all-β
fold mimicking a right hand. The thumb and fingers pinch heme iron through two
apical histidine residues, giving rise to highly symmetric octahedral iron
co-ordination. The tetrameric quaternary arrangement of the protein found in the
crystal structure is consistent with experiments in solution. It shows that
thumbs and fingertips, and, by extension, the bound heme groups, are shielded
from competing heme-binding proteins from the host. This may also facilitate
heme transport to HmuR for internalization. HmuY, both in its apo- and in its
heme-bound forms, is resistant to proteolytic digestion by trypsin and the major
secreted proteases of P. gingivalis, gingipains K and R. It is
also stable against thermal and chemical denaturation. In conclusion, these
studies reveal novel molecular properties of HmuY that are consistent with its
role as a putative virulence factor during bacterial infection.
| Pathogenic bacteria cause infection in humans as found in periodontitis, which is
a chronic inflammation of the gums caused by Porphyromonas
gingivalis. As part of the infective process, bacteria must acquire
nutrients to survive and multiply at the infection site, and among such
nutrients is heme. This is an iron-dependent co-factor of several indispensable
enzymes and proteins. P. gingivalis liberates heme from host
heme-binding proteins through the action of proteases and arranges its transport
to the bacterial cell through two proteins, HmuY and HmuR. They grab free heme
and transport it across the bacterial membrane into the cell, respectively. This
function poses stringent conditions on these proteins regarding stability and
resistance toward the host immune system. We report here that HmuY is very
stable and that it displays a novel protein fold, which consists only of
β-strands. It reminds us of a right hand, whose fingers trap heme. Once
heme is bound, HmuY forms tetramers, which have the four heme-binding sites
buried and thus protected from competing host heme-binding proteins. This
feature also facilitates heme transport to HmuR and into the bacterial cell. All
these data may help to develop new antibacterial agents at times in which
resistance toward antibiotics, both at intensive healthcare stations and in the
community, poses serious challenges to human health.
| Periodontitis causes chronic inflammation of the gums and it affects
10–15% of adults worldwide, potentially leading to tissue
destruction and tooth loss, and Porphyromonas gingivalis is its
main etiological agent [1],[2]. In addition, P. gingivalis, an
anaerobic black-pigmented, Gram-negative bacterium, has been implicated in
cardiovascular diseases, respiratory diseases, diabetes, osteoporosis, and pre-term
low birth-weight [3]-[5]. The pathogen cannot synthesize protoporphyrin IX
but acquires exogenous heme (“heme” is here used to refer
indistinctly to either Fe2+- or
Fe3+-protoporphyrin IX), an excess of which is stored in the
characteristic black pigment on the bacterial cell surface [6]–[8]. The
co-factor is obtained from hemoglobin, haptoglobin-hemoglobin, myoglobin, hemopexin,
serum albumin, lactoperoxidase, cytochrome c, and catalase by the action of
hemolysins and proteases [9]–[12]. In addition,
P. gingivalis and other Gram-negative bacteria possess systems
to bind locally liberated heme such as secreted heme-binding proteins and hemophores
[13],[14]. One such hemophore is HasA, employed by
Serratia mercescens to scavenge host heme in order to deliver
it to the receptor, HasR, for internalization [15]. Similarly, hemophores
have been described in Haemophilus influenzae, Yersinia
enterocolitica, Pseudomonas aeruginosa, and
Bacillus anthracis
[13],
[16]–[20]. Further heme is
transported into the cell through outer-membrane receptors [21].
In P. gingivalis, heme is primarily imported by heme-binding
protein, HmuY, and its cognate outer-membrane receptor, HmuR [22]. The latter is involved
in heme transport through the outer membrane and probably depends on the interaction
with protein TonB, which is needed to transduce energy for the passage of heme and
other ligands into the periplasm in most Gram-negative pathogens [11],[14],[22],[23]. The
two Hmu proteins are encoded in tandem by the hmu operon, which
comprises six genes in total, hmuYRSTUV. The locus is regulated by
iron [23]
and by a transcriptional repressor encoded by gene pg1237
[24], and its
disruption leads to a 70% decrease in heme binding and a 45%
decrease in heme uptake [25]. Potential protein pairs with high sequence
similarity to HmuY and HmuR have been identified on contiguous genes in other
bacteroidetes (Microscilla marina, Prevotella
intermedia, and Bacteroides from the species
vulgatus, fragilis, ovatus,
thetaiotaomicron, caccae,
stercori, and coprocola), proteobacteria
(Plesiocystis pacifica, Stigmatella aurantica,
and Myxococcus xanthus), spirochaetes (Leptospira
biflexa), and chlorobi (Chloroherpeton thalassium). This
suggests a widespread mechanism for heme uptake (our unpublished data; [23],[25]). HmuY
is an outer-membrane-associated lipoprotein, which is identical in sequence to a
P. gingivalis envelope protein designated fibroblast activating
factor [26]. This factor induces proliferation and protein
synthesis in normal human gingival fibroblasts, indicating an additional role for
HmuY in the host immune response. The hmuY gene encodes a 23-kDa
protein, with no significant sequence similarity to any other protein, whose 25
first residues are not present in the purified protein. This stretch comprises a
leader sequence, a lipid-binding site, and a potential protease cleavage site [23],[25]. The
protein is functional as a dimer in its heme-depleted form and as a tetramer once
heme is bound [23]. Heme bound to HmuY, with a midpoint potential of
136 mV, displays a low-spin six-fold Fe3+ co-ordination sphere
with the participation of residues His134 and His166, as revealed by point mutation
studies [27].
In order to shed light on the mechanisms of heme binding and transport through HmuY,
we set out to assess the folding and stability properties of HmuY in its
heme-depleted (apo-HmuY) and heme-complexed (holo-HmuY) forms, as well as its
susceptibility to proteolysis by trypsin and gingipains K (Kgp) and R (RgpA and
RgpB). In addition, we solved the X-ray crystal structure of holo-HmuY. Taken
together, these data enabled us to propose a mechanism for Hmu-mediated heme uptake
by P. gingivalis.
Assessment of the biophysical response of apo- and holo-HmuY to thermal and
chemical denaturation may contribute to unravel heme binding and transport at
the infection site. Previously, a far-UV CD spectrum of native apo-HmuY had
shown that the protein has mainly a β structure [23]. Here we found that
heme binding did not affect the spectrum, indicating absence of significant
structural changes of the secondary structure (Figure 1). In unfolding studies (Figure 2), far-UV CD
spectroscopy revealed that thermal denaturation of HmuY (encompassing the
sequence of the purified natural protein; Asp26-Lys216) was irreversible,
leading to protein precipitation (data not shown). In contrast, guanidinium
hydrochloride (GdnHCl)-induced chemical denaturation was reversible and no
significant difference in the equilibrium unfolding profiles of apo- and
holo-HmuY was observed (Figure
2C). Both forms tended to follow one-step unfolding process with a
calculated free energy of denaturation
(ΔGden) of 34.1±12.4 kJ/mol. To
obtain information on local changes of the tertiary structure of HmuY and the
heme cavity, we further studied thermal and chemical denaturation by intrinsic
tryptophan fluorescence spectroscopy. Thermal denaturation is reversible and
gives rise to sigmoidal unfolding curves for both HmuY forms (Figure 2A and 2B). However,
the initial part of the curves, especially that of holo-HmuY, deviates from the
one-step reverse-unfolding mechanism. Exposure of apo-HmuY to increasing
concentrations of GdnHCl reduced fluorescence intensity, giving a sigmoidal
shape to the unfolding curve (Figure 2D). In contrast, the unfolding profile of holo-HmuY (Figure 2E) does not account
for a typical one-step reverse-mechanism transition unless the reference
wavelength is far away from the heme-binding maximum (323 nm). This suggests
that holo-HmuY may exhibit local differences in the tertiary structure when
compared with apo-HmuY similar to HasA and HasAp [18],[28],[29]. Simultaneously, we
examined heme loss by holo-HmuY by absorbance change in the Soret region (Figure 2F and 2I). Holo-HmuY
retained ∼70% of the bound heme up to 3 M GdnHCl, and even in
the presence of 4 or 5 M GdnHCl, it remained partially loaded with the
co-factor. An interesting feature can be observed at intermediate GdnHCl
concentrations (3.6 M) and at ∼50°C. Although the overall
fluorescence spectra are similar, thus indicating that the conformations are
related (Figure 2G and 2H),
the possibility of intermediate species should not be excluded. Changes in the
fluorescence spectra in the pre-transitional region may indicate differences in
local conformation of HmuY upon heme binding and tetramer formation. This
corresponds to the end of the pre-transitional region regardless of the
technique used, as indicated by arrows in Figure 2A, 2B, 2D, 2E, and 2F. It may be
easier to see the intermediates when heme is bound to HmuY (Figure 2E), since the co-factor modulates the
fluorescence characteristics of HmuY tryptophans. We conclude that the first
step of the HmuY-heme complex unfolding is a tetramer-to-dimer transition,
subsequently leading to heme loss. Posterior dimer dissociation and protein
denaturation are probably the limiting steps in this process. Taking all
together, HmuY stability against heat and GdnHCl-induced denaturation is similar
to or higher than that of other stable proteins bound to heme [30].
However, in contrast to the latter, both apo- and holo-HmuY show comparable
resistance to denaturation, although some local changes in the tertiary
structure may occur upon heme binding.
Ligand binding may enhance resistance to proteolysis [31], so trypsin was
assayed as a degrading agent against HmuY. In addition, response to the cysteine
proteases Kgp, RgpA, and RgpB was examined, as these are secreted by P.
gingivalis upon infection and target host hemoproteins [12].
Both apo- and holo-HmuY were fully resistant to digestion by trypsin and
gingipains in their native state under the conditions assayed (Figure 3). In contrast,
protein samples previously subjected to thermal denaturation were completely
degraded. These data strongly suggest that HmuY is very stable and compactly
folded, regardless of the bound co-factor, and insensitive against endogenous
proteases.
The structure of HmuY is asymmetric, with maximal dimensions of
∼55×40×35 Å, and it resembles a right hand
(Figure 4A). It consists
of a roughly globular nucleus reminiscent of a palm, out of which protruding
segments mimicking thumb and fingers emerge (Figure 4A, 4B, and 4C). As anticipated by CD
spectroscopy (see above and [23]), HmuY is an all β-protein
constituted by 15 β-strands. Both the N- and C-terminus are located on
the protein surface corresponding to the palm and they point toward the wrist
(following the analogy with a hand). They precede and succeed, respectively, two
β-strands that participate in a twisted β-sandwich or laterally
open β-barrel made up of two antiparallel β-sheets of,
respectively, five (sheet I; strands
β1+β6+β13−β15;
connectivity +4x, −1, −1, −1) and four
(sheet II; strands β2−β5; connectivity −1,
+2x, +1) strands (Figure 4B and 4C). A large twisted and curled β-ribbon
(β7β8) is inserted into the front of the palm mimicking a thumb
and two further ribbons are found on the back resembling a pinky (strand
β12), a ring finger (β11), a middle finger (β10), and an
index finger (β9) (top to bottom in Figure 4C). Sheet I is curled toward sheet II
and twisted for ∼95°, while sheet II is arched away from sheet I
and twisted for ∼85°. This gives rise to a large hydrophobic
cavity at the interface between sheets, which is the core of the protein and
contributes to most of the palm. It is created by side chains from strands
β1 and β2, the loop connecting
β2 with β3
(Lβ2β3), and strands β3−β6 and
β13−β15. In addition, hydrophobic residues provided by
Lβ6β7, Lβ8β9, and Lβ10β11 close
the β-sandwich on the flank bordered by β5 (sheet II) and
β6 (sheet I) (Figure
4C). Two smaller hydrophobic cores are observed on the convex side of
either sheet by side chains provided by β6,
β13−β15, and Lβ6β7, which folds back
on top of the sheet (sheet I), and Lβ3β4 plus
Lβ5β6 (sheet II). All together, these hydrophobic clusters give
rise to a molecule, which provides a structural explanation for its high
stability and resistance toward denaturation and proteolysis.
Together with the finger-proximal lateral wall of the central hydrophobic core,
thumb and fingers give rise to the heme-binding cavity, which is radically
different in structure and location within the molecule from those found in any
other heme-binding protein described. The cavity occupies a volume of 2,136
Å3 and is made up by 37 residues, including 29
hydrophobic or neutral residues and eight charged residues (Figure 4D). A single
Fe3+-chelating heme b molecule (hemin) is inserted laterally
like a wedge into the cavity, curiously with its charged propionate substituents
pointing toward the palm and the hydrophobic methyl and vinyl substituents
pointing toward the exterior (Figure 4C and 4D). The protein∶heme complex is
characterized by an interaction surface of 532 Å2, which is
61% of the total accessible surface of the co-factor (868
Å2). This value is rather low if compared with other
heme-protein complexes [32] but can be explained in terms of the
quaternary arrangement (see below). Complex interactions are mainly hydrophobic
and entail three hydrogen bonds, one salt bridge, and van-der-Waals interactions
between 13 protein residues (Table 1 and Figure
4D). Further noteworthy are a salt-bridge of Arg79 Nη2 and two
hydrogen bonds of Tyr80 Oη and Tyr13 Oη with the carboxylate
oxygen atoms of the propionate substituent of pyrrol ring a, as well as a
hydrogen bond between Thr124 Oγ1 and the propionate of ring d. However,
the most relevant contacts are the metallo-organic bonds established between the
heme iron and the Nε2 atoms of His134 (2.04 Å away), provided
by thumb-strand β8, and His 166 (2.09 Å), from
ring-finger-strand β11. These two ligands occupy the apical positions of
an octahedral iron co-ordination sphere, whose equatorial ligands are the four
porphyrin nitrogen atoms (at 2.03–2.06 Å). Accordingly, ion
co-ordination is exerted by six nearly equivalent and equidistant
sp2-hybridized nitrogen atoms and is thus highly symmetrical, which
should redound to a very stable complex. In addition, both protein histidine
side chains are in hydrophobic environments: His134 is surrounded by Tyr127,
Met129, Met136, and Pro168 (from a vicinal complex; see next chapter) and His166
by Phe156, Phe164, Pro171, and the methyl and vinyl groups from an adjacent
complex.
The quaternary structure of holo-HmuY is a cross-like tetramer of complexes,
∼95 Å in diameter, created by the combination of a local and a
crystallographic two-fold axis (Figure 4E). This finding fits well with analytical size-exclusion
chromatography studies showing that, upon addition of heme, HmuY eluted as a
tetramer [23]. The tetramer is made up by interactions between
thumb and finger tips and includes β-hairpins β7β8,
β9β10, β11β12, and β14β15 of
each of the four complexes, which contribute through a surface of
1148–1174 Å2 to the oligomer. This is within the
range reported for interaction surfaces usually found in protein complexes [33].
All contacts between vicinal complexes (e.g. the magenta and yellow ones in
Figure 4E) are symmetric
and include eight hydrogen bonds, two salt bridges and four hydrophobic
interactions, two hydrophobic protein-heme and one hydrophobic inter-heme
contacts (Table 1). Fewer
contacts are observed between opposite complexes (e.g. the magenta and cyan ones
in Figure 4E); they just
include three symmetric hydrophobic interactions established by Ala169, Asp132,
and the vinyl substituents of heme pyrrol rings b (Table 1). The overall arrangement entails
that the four heme groups are in direct contact with each other in the center of
the tetramer. Each co-factor molecule has a buried surface of 636–639
Å2, 74% of its total area. The inter-iron
distances are 13 Å (vicinal monomers) and 17 Å (opposite
monomers), within the range of values observed in electron-transport proteins
such as cytochrome c (9–18 Å; [34],[35]),
although no function of HmuY in redox biochemistry or electron transport has
been postulated. As a result of this packing the four heme groups are buried and
thus protected from competing heme-binders.
Structure similarity searches following a variety of algorithms failed to
identify significant matches extending beyond selected parts of the central
β-sandwich, so we conclude that HmuY conforms to a novel all
β-fold. Among structurally characterized heme-binding proteins
participating in iron storage and transport are the archetypes hemoglobin and
myoglobin, which like related globins and serum albumin, are all-α class
proteins [36]–[38]. With respect to
functional analogs of HmuY, only the closely-related hemophores HasA and HasAp
from S. marcescens and P. aeruginosa,
respectively, have been reported for their structure [18],[28].
They consist of a two layer α/β-sandwich with a meander fold
characterized by a twisted antiparallel six-stranded β-sheet with four
helices on its concave side. The overall shape of the molecules is reminiscent
of a fish, which traps a heme in its mouth. Here, the iron is octahedrally
co-ordinated by an apical histidine and a tyrosine. Like HmuY, HasA and HasAp do
not undergo major structural rearrangement upon heme binding.
The only other structurally characterized all-β-structure engaged in heme
transport and storage is serum hemopexin [32],[39].
It consists of two tandem ∼200-residue fourfold β-propeller
domains, which are thick discs consisting of four blades arranged around a
central channel. Each blade is made up of a twisted four-stranded antiparallel
β-sheet [39]. The functionally relevant oligomerization
state is a monomer, and heme binding correlates with major structural
rearrangement [32]. In the heme-bound complex, the two hemopexin
domains are roughly perpendicular to each other and connected by a partially
flexible 20-residue linker (Figure
4F). The heme–binding site resides at the interface between
domains, which is covered by a cluster of conserved aromatic residues. As in
HmuY, heme binding is exerted by two histidine residues, one provided by the
linker, which bind the iron on its two apical positions, and a cluster of basic
residues and tyrosines that bind the heme propionate groups [32].
Further as in HmuY, the heme propionate groups are buried in the molecule and
the hydrophobic substituents of heme rings b and c likewise point toward the
exterior of the molecule. However, beyond these very detailed features and a
generally apolar environment of the heme-binding cavity, there is no further
structural similarity between holo-hemopexin and holo-HmuY.
Accumulating biochemical and genetic evidence [13], [14], [22]–[25],[40],[41], as well as the present data, suggest the
following mechanism for Hmu-mediated heme uptake in P.
gingivalis and, by extension, in other related bacteria (Figure 5). HmuY is synthesized
and exported to the outer membrane, where it would be anchored to a lipid
through an attachment site typical for prokaryotic lipoproteins [42].
Location of HmuY at the outer membrane of intact P. gingivalis
cells and to outer-membrane vesicles has been shown [23],[25],[26]. In addition, the
existence of a membrane-attached HmuY species was substantiated by experiments
in Escherichia coli cells [43]. From the surface,
HmuY would be shed by Kgp to enter the inflamed periodontal tissue at the site
of infection. This step is backed by N-terminal sequencing of HmuY purified from
the culture medium, which revealed that the protein starts with residue Asp26,
and that recombinant HmuY comprising five additional upstream residues of the
gene-encoded sequence, i.e. starting with Met21-Gly-Lys-Lys-Lys-Asp26, was
cleaved at bond Lys25-Asp26 by Kgp in vitro (data not shown).
Moreover, mRNA encoding HmuY was much more highly expressed than any of the
other hmu-operon encoded proteins (7–20 times and
several orders of magnitude more transcription than HmuR and HmuS-V,
respectively [23],[25]). This observation,
as well as our results demonstrating high stability against denaturation and
proteolysis, is consistent with the idea of a protein that is targeted for
secretion as a virulence factor during infection. One of the virulence
mechanisms in P. gingivalis is biofilm formation, which
facilitates the long term survival of the bacterium and induces an inflammatory
reaction in the host. A recent report showed that HmuY is found predominantly in
the biofilm [44]. Furthermore, it was demonstrated that the
P. gingivalis hmuY-knockout mutant cannot grow in the
presence of human serum as a sole heme source, confirming that this protein is
necessary for growth of bacteria under low iron/heme conditions, as found in
deep biofilm layers [23],[45].
Host hemoproteins are present in significant concentrations in the gingival
crevice [46], but HmuY would not be able to compete with
them for heme binding due to its much lower affinity for the co-factor
(Kd∼3 µM, our unpublished data; hemopexin and
hemoglobin, Kd<1 pM, [47],[48];
haptoglobin/hemoglobin complex, Kd<fM, [49]; serum albumin,
Kd∼10 nM, [50]). However,
gingipains, the major proteases produced and secreted by P.
gingivalis, as well as other secreted proteases, can efficiently cleave
hemoglobin, haptoglobin, and hemopexin, thus liberating heme [10],
[51]–[54]. In contrast, we
have shown here that HmuY is resistant to proteolysis by gingipains. In an
alternative or complementary fashion, release of heme from these hemoproteins
could occur spontaneously, as this is a low-energy event [14], or through
conformational changes induced in the host hemoproteins that would make the
co-factor accessible. In any case, HmuY would take up heme, and this would lead
to tetramerization under occlusion of the heme binding sites. Tetrameric HmuY
would protect heme from host scavengers and shepherd it to HmuR. At this point,
heme transfer to the latter encounters two obstacles: the receptor has ten times
lower affinity for heme than HmuY
(Kd = 24 µM; [21],[40],[41]) and
the heme groups are inaccessible within the holo-HmuY tetramer. Accordingly,
this step would probably entail a rupture of the tetramer triggered by the
HmuY-HmuR interaction to expose heme. Similarly, disruption of a tight
heme/carrier complex to enable heme uptake by a receptor have been reported for
the S. marcescens HasA-HasR system [15] and for
hemopexin-hemopexin receptor [32]. On the basis of our mutational analysis of
HmuY heme ligands [27], an initial step in heme transfer could
involve disruption of only one of the two axial histidine ligands as found for
HasA [15],[27]. At this point, a
Fe(III)-to-Fe(II) transition is conceivable [27]. Once bound by
HmuR, heme would be translocated across the outer membrane into the periplasm.
In the absence of the HmuR receptor, heme cannot be efficiently transported into
the cells, which retards growth of the hmuR-knockout mutant
[43]. In addition, this mutant becomes more pigmented,
indicating that HmuY binds and stores the accumulating excess of heme on the
bacterial cell surface as a result of the broken pipeline. Therefore, HmuY,
especially in the form associated with the outer membrane, may also store heme
and protect the bacterial cell from damage induced by free heme. A phenotype of
P. gingivalis hmuY- and hmuY/hmuR-knockout
mutants confirms this hypothesis since both strains are less pigmented than the
wild-type [23]. Heme translocation by HmuR putatively occurs
under assistance of TonB [11],[14]. The pernicious oxidative potential of free
heme would also require the presence of binding proteins to escort it across the
periplasm to the cytoplasm [21]. This step might be performed by the other
hmu operon proteins, which would be required in much less
amount than HmuY: HmuS, which displays sequence similarity with cobN/Mg
chelatase; HmuT and HmuU, which are similar in sequence to permeases; and HmuV,
annotated as an ATP-binding protein engaged in hemin import [25].
Further studies, e.g. of the HmuY/HmuR interaction, are necessary to understand
this novel heme transfer mechanism for bacterial survival.
Pathogenic bacteria have evolved sophisticated mechanisms in response to the
changing environment and the host antimicrobial defense systems. The
multiprotein system possibly encoded by the hmu operon in
proteobacteria, bacteroidetes, spirochaetes, and chlorobi, contributes to heme
uptake and utilization for bacterial survival and infection. As pathogenic
bacteria continue to develop resistance to antibiotics, targeting nutrient
uptake systems may offer novel strategies to combat microorganisms such as
P. gingivalis, a formidable pathogen. In this context,
these data on structure and function of the hmu-encoded
heme-binding protein, HmuY, which may have also a role in the host immune
response and in interaction with host cells, may lead to the development of
novel therapeutic approaches to pathogen incapacitation. The high stability of
HmuY given by its unique structure makes it a suitable candidate for
biotechnological and biomedical applications.
P. gingivalis apo-HmuY lacking the first 25 residues of the
DNA-derived protein sequence (NCBI accession number CAM 31898) and a variant
lacking the first 21 residues but containing a C-terminal HSV and His8-tag, were
expressed using plasmids pHmuY11 or pDB and E. coli ER2566 (New
England Biolabs) cells, and purified as previously reported [23],[43]. A protein variant
incorporating selenomethionine instead of methionine was prepared using plasmid
pHmuY11 and the same cells, which were added to 500 mL of minimal medium lacking
methionine and implemented with 25 mg of selenomethionine (Sigma) 30 min before
induction [55]. Holo-HmuY was reconstituted from heme and
apo-HmuY by incubating 1 equivalent of protein with 1 equivalent of heme (ICN
Biomedicals) at room temperature. Excess heme was removed by gel filtration
through a PD-10 desalting column (Amersham Pharmacia). To purify HmuY from
culture media, P. gingivalis cells were cultured anaerobically
on blood-agar plates and then in basal medium supplemented with hemin or
dipirydyl as described previously [23]. Cultures were
centrifuged at 20,000×g for 20 min at 4°C,
supernatants filtered using membranes with a pore size of 0.22 µm,
dialyzed against 50 mM Tris/HCl buffer, 25 mM NaCl, pH 7.6, and concentrated
using 10-kDa cut-off membranes (Amicon). Concentrated media were further
ultracentrifuged (Beckman) at 100,000×g for 2 h at 4°C and
supernatants were used to purify HmuY.
UV-Vis spectra were recorded with an Agilent 8453E UV-Vis spectrophotometer
(Agilent Technologies). Far-UV CD spectroscopy (205–255 nm) was
carried out using a Jasco J-810 spectropolarimeter and 10-mm-path-length
cuvettes. For thermal denaturation experiments, CD spectra were recorded from
210 to 250 nm. The CD signals at 225 nm were monitored as a function of
temperature from 20 to 80°C. Protein samples were examined in 20 mM
sodium phosphate, 20 mM NaCl, pH 7.4 or 20 mM sodium phosphate, 1 M GdnHCl, pH
6.5.
GdnHCl unfolding experiments were performed according to standard protocols [56]. A
stock of 6 M GdnHCl (MP Biochemicals) was used to prepare solutions in 20 mM
sodium phosphate, 20 mM NaCl, pH 7.4 and variable GdnHCl concentrations (0 to 6
M). Subsequently, concentrations of GdnHCl solutions were determined through
measurement of their refractive index at 25°C using a Zeiss
refractometer. Apo- and holo-HmuY (protein∶co-factor ratio
1∶1) were added to each sample at 2 µM final concentration
and incubated at 25°C for 18 h. GdnHCl-induced chemical denaturation was
monitored at 20°C by CD using a Jasco J-810 spectropolarimeter and by
intrinsic tryptophan fluorescence using a Jasco FP-750 spectrofluorometer.
Samples were excited at 295 nm for fluorescence measurements and the emission
spectra from 300 to 700 nm were recorded (slit width 5 nm). CD and fluorescence
data were transformed to yield the relative fraction of unfolded protein and to
determine the free energy of denaturation.
HmuY was subjected to proteolysis by trypsin, Kgp, RgpA, and RgpB. For tryptic
digestion, two reactions with HmuY in 100 mM Tris/HCl, 20 mM CaCl2,
pH 8.0 at 1∶50 protease∶substrate molar ratio were conducted
in the presence (1∶1 molar ratio) and absence of heme. Fresh portions
of trypsin were added every 1 h during the first 12 h, and then every 12 h. For
assays with gingipains, proteases were pre-incubated in 200 mM HEPES, 10 mM
cysteine, pH 7.6 for 15 min at 37°C prior to addition of HmuY purified
from E. coli cells or P. gingivalis culture
medium (1∶20 protease∶substrate molar ratio) and further
incubation for 1 or 20 h at 37°C. Aliquots were taken from the reaction
mixtures at given time points, and the reaction was inhibited by addition of a
protease-inhibitor cocktail (Roche) and through boiling in SDS-PAGE sample
buffer. Control reactions were performed in the absence of protease. HmuY
samples previously subjected to thermal denaturation (95°C, 10 min) were
also assayed for proteolytic susceptibility. All samples were examined by
15%-Tris/glycine SDS-PAGE and Coomassie Brilliant Blue G-250
staining.
Crystallization assays were performed following the sitting-drop vapor diffusion
method. Reservoir solutions were prepared by a Tecan robot and 200-nL
crystallization drops were dispensed on 96×2-well MRC plates
(Wilden/Innovadyne) by a Cartesian nanodrop robot (Genomic Solutions) at the
joint IBMB-CSIC/IRB/Barcelona Science Park High-Throughput Crystallography
Platform (PAC). Best crystals as thin reddish prisms appeared in a Bruker
steady-temperature crystal farm at 20°C using protein solution (26 mg/mL
in 5 mM Tris/HCl, pH 8.0) and 2.4 M (NH4)2SO4,
0.1 M MES, pH 6.0 as reservoir solution and D(+)-glucose monohydrate as
additive (relative volumetric ratios 1∶1∶0.35). These
conditions were successfully scaled up to the microliter range with Cryschem
crystallization dishes (Hampton Research). Crystals of
selenomethionine-derivatized protein were obtained under similar conditions.
Cryoprotection of protein crystals for diffraction data collection was achieved
through harvesting with 3.0 M (NH4)2SO4, 0.1 M
MES, pH 6.0 and subsequent stepwise replacement of the mother liquor with 3.0 M
(NH4)2SO4, 25% glycerol, 0.1 M
MES, pH 6.0. Complete diffraction datasets were collected at 100 K from a single
flash-cryo-cooled (Oxford Cryosystems) native crystal (at
λ = 1.0000 Å) and from a
derivatized crystal (at λ = 0.9792
Å; absorption peak for selenium determined through a fluorescence scan
performed with a Si-drift chamber detector (Rontec)) on an ADSC Q315R CCD
detector at beam line ID23-1 of the European Synchrotron Radiation Facility
(ESRF, Grenoble, France) within the Block Allocation Group “BAG
Barcelona”. Crystals were tetragonal and harbored one dimer per
asymmetric unit (VM = 2.5
Å3/Da; 58% solvent contents). Diffraction
data were integrated, scaled, merged, and reduced with programs XDS [57] and
SCALA within the CCP4 suite [58] (see Table 2).
The structure was solved by SIRAS by using native and selenomethionine-derivative
diffraction data and programs SHELXD/E under phase extension to 1.8 Å
[59]. The phases obtained were subjected to a density
modification step with program DM within CCP4 and an electron density map was
computed. Subsequently, manual model building on a Silicon-Graphics workstation
using program TURBO-Frodo alternated with crystallographic refinement with
REFMAC5 within the CCP4 suite until completion of the model (see Table 2). This model
contained protein residues Glu35 to Lys216 plus a heme
co-factor for each of the two protein chains within the crystal asymmetric unit
(molecules A and B). The ten N-terminal residues of the protein were flexible
and were not traced. Both molecules were equivalent in practice
(rms deviation for all atoms equaled 0.57 Å), so
discussion of the structure considered only molecule A unless otherwise
stated.
Figures were prepared with programs SETOR [60] and Carrara 4.
Structural similarity searches were performed with programs DALI (http://ekhidna.biocenter.helsinki.fi/ dali_server), SSM
(http://www.ebi.ac.uk/msd-srv/ssm), VAST (http://www.ncbi.nlm.nih.gov/Structure/VAST/vast.shtml), and the
CATHEDRAL server (http://www.cathdb.info/).
Cavity volumes were computed with program PDBsum (http://www.ebi.ac.uk/pdbsum). The final co-ordinates of holo-HmuY
have been deposited with the PDB at www.pdb.org (access code 3H8T).
Inter- and intra-molecular close contacts (<4 Å) and contact
surfaces (with a rolling sphere of 1.4 Å) were determined with program
CNS [61]. The value of interaction surfaces was estimated
taking the half of the difference between the sum of the individual molecular
surfaces and the total complex surface.
|
10.1371/journal.pgen.1000646 | Redundant and Specific Roles of the ARGONAUTE Proteins AGO1 and ZLL in Development and Small RNA-Directed Gene Silencing | The Arabidopsis ARGONAUTE1 (AGO1) and ZWILLE/PINHEAD/AGO10 (ZLL) proteins act in the miRNA and siRNA pathways and are essential for multiple processes in development. Here, we analyze what determines common and specific function of both proteins. Analysis of ago1 mutants with partially compromised AGO1 activity revealed that loss of ZLL function re-establishes both siRNA and miRNA pathways for a subset of AGO1 target genes. Loss of ZLL function in ago1 mutants led to increased AGO1 protein levels, whereas AGO1 mRNA levels were unchanged, implicating ZLL as a negative regulator of AGO1 at the protein level. Since ZLL, unlike AGO1, is not subjected to small RNA-mediated repression itself, this cross regulation has the potential to adjust RNA silencing activity independent of feedback dynamics. Although AGO1 is expressed in a broader pattern than ZLL, expression of AGO1 from the ZLL promoter restored transgene PTGS and most developmental defects of ago1, whereas ZLL rescued only a few AGO1 functions when expressed from the AGO1 promoter, suggesting that the specific functions of AGO1 and ZLL are mainly determined by their protein sequence. Protein domain swapping experiments revealed that the PAZ domain, which in AGO1 is involved in binding small RNAs, is interchangeable between both proteins, suggesting that this common small RNA-binding domain contributes to redundant functions. By contrast, the conserved MID and PIWI domains, which are involved in 5′-end small RNA selectivity and mRNA cleavage, and the non-conserved N-terminal domain, to which no function has been assigned, provide specificity to AGO1 and ZLL protein function.
| In eukaryotes, short RNAs (21–24 nucleotides long) have broad effects on gene expression through the action of ARGONAUTE (AGO) proteins. The model flowering plant Arabidopsis thaliana contains ten AGO proteins, among which AGO1 and ZLL/PNH/AGO10 play a major role in regulating gene expression through small RNA-directed RNA cleavage and translational repression. Here, we address the common and specific effects of zll and ago1 loss of function in Arabidopsis. We show that zll mutations lead to increased AGO1 protein levels and suppress a subset of small RNA-directed gene regulatory defects of weak ago1 mutations. Although AGO1 and ZLL proteins are highly similar in sequence, we show that only the PAZ domain, which in AGO1 is involved in binding small RNAs, can be exchanged between the two proteins. By contrast, the PIWI domain, that is responsible for the RNA cleaving activity of AGO1, the MID domain, which is involved in 5′ nucleotide selection of small RNAs, and the functionally uncharacterized N-terminal domain contribute to their individual functions during small RNA-directed gene regulation and development.
| Small RNA-directed gene regulation is a major process in plant development and viral defense [1],[2]. A central component in these pathways is the activity of ARGONAUTE (AGO) proteins, which bind small RNAs and mediate repression of the complementary RNA targets [3],[4]. In Arabidopsis, 10 AGO genes have been identified [5]. AGO1 [6] associates with numerous microRNAs (miRNAs) and short interfering RNAs (siRNAs) to mediate target repression via mRNA cleavage and inhibition of translation [3],[4],[7]. Binding of AGO1 to miR168, which targets AGO1 mRNA, establishes a homeostatic AGO1 regulatory loop [8],[9]. AGO4 and AGO6 function in small RNA mediated chromatin regulation whereas AGO7 associates specifically with miR390 and directs cleavage of the non-protein coding TAS3 precursor RNA to generate trans-acting short interfering RNAs (tasiRNAs) [5]. Recently, ZLL was implicated in miRNA-directed translational inhibition [7] and repression of miR165/166 levels [10].
AGO1 and ZLL protein sequences are highly similar, including the PAZ and MID domains, which bind small RNAs in AGO1 [11], and the PIWI domain, which is required for target mRNA cleavage in AGO1 [3],[4]. By contrast, their N-terminal domains do not display sequence similarities. Both genes differ in their expression patterns and developmental functions. AGO1 is expressed broadly during plant development, and ago1 loss-of-function mutants display pleiotropic defects in development and in virus defense [6],[12]. Seedlings of the null allele ago1-1 form only a few finger-like leaves and about 10% of seedlings lack a shoot meristem. ago1 mutants are deficient in transgene posttranscriptional gene silencing (PTGS) of L1 35S:GUS, a standard reference for systemic sense transgene PTGS in Arabidopsis [13], the tasiRNA pathway, and cell autonomous miRNA-directed repression [5]. In contrast to ago1-1, the hypomorphic allele ago1-27, which expresses an AGO1 protein with reduced mRNA cleavage activity, displays more subtle developmental defects [12].
Expression of ZLL is limited to the provasculature and, weaker, to the adaxial (upper) sides of leaves, and ceases as tissue differentiation takes place [14],[15]. In the Landsberg erecta (Ler) accession, zll mutant seedlings display differentiated cells or complete organs in place of the shoot meristem stem cells with allele specific penetrance [14]–[16]. Recent studies indicate that ZLL function in the provasculature is necessary and sufficient to maintain shoot meristem stem cells during embryogenesis [17]. Furthermore, ZLL acts in a sequential manner with AGO1 during embryogenesis to potentiate WUSCHEL (WUS) dependent signaling from the stem cell organizer to the stem cells in the developing shoot meristem primordium [17]. ago1 zll double mutants of strong alleles result in early embryo arrest, suggesting that both proteins also have redundant activities during early embryo development [14]. Recent findings demonstrated that both proteins function in miRNA-directed repression of Cu/Zn SUPEROXIDE DISMUTASE 2 (CSD2) and SCARECROW-LIKE 6 (SCL6-IV) mRNAs and proteins [7]. In contrast to ago1 mutants, however, L1 transgene PTGS is not compromised in zll mutants [12].
Here, we address specific and overlapping functions of ZLL and AGO1 in development and RNA silencing pathways. Our results indicate that in ago1 hypomorphic mutants, loss of ZLL function restores leaf development and siRNA and miRNA pathways and leads to increased AGO1 protein levels, implicating ZLL as a negative regulator of AGO1. Analyses of chimeric gene constructs indicate that the PAZ domain, which is thought to mediate small RNA binding, is exchangeable between both proteins, whereas the MID-PIWI and N-terminal domains appear to contribute to their specific functions.
To study genetic interactions between ZLL and AGO1, we analyzed different mutant combinations. Since double mutants of strong zll and ago1 alleles in the Ler ecotype are embryo lethal [14], we analyzed mutant alleles in the Col ecotype, where ZLL loss of function alone does not greatly affect development (Figure 1A and Figure S1), unlike in the Ler accession, where shoot meristem stem cells are defective [14],[15]. Despite the reduced effect of zll mutations in Col compared with Ler, ago1-1 zllago10-1 double mutant embryos also arrested at the late globular stage with defects in cell division, cell elongation, and expression of both WOX5 and WUS genes, which mark root and shoot stem cell niches, respectively (Figure S2). None of these effects were observed in any single mutant, indicating redundant functions of ZLL and AGO1. To avoid embryo lethality obtained in double mutants with the null allele ago1-1 [6] and to enable the analysis of genetic interactions during postembryonic development, we used the hypomorphic ago1-27 mutant in combinations with zllago10-1 and zllago10-3 alleles. ago1-27 mutants are defective in small RNA-directed regulation [9],[12] and, in contrast to the severe growth and developmental defects of ago1-1, display increased leaf margin serration, reduced leaf width, abnormal flower phyllotaxis, and reduced fertility compared to wildtype [12]. By contrast, seedlings of zllago10-1 and zllago10-3 single mutants did not display any noticeable leaf defects (Figure 1 and Figure S1A) [18] and only infrequently a defective shoot meristem (0.2%, n>1000) [14],[15]. Surprisingly, ago1-27 zllago10-1 and ago1-27 zllago10-3 double mutants revealed that both zll mutations partially suppressed the increased leaf margin serration of ago1-27 (Figure 1B and Figure S1A), rather than enhancing it as we expected for two related AGO proteins involved in RNA silencing. By contrast, neither the phyllotaxis nor the fertility defects of ago1-27 were restored by the zll mutations (data not shown).
To study ZLL and AGO1 interactions at the level of RNA silencing, we first analyzed PTGS of the L1 35S:GUS transgene. Our previous studies indicated that PTGS of the L1 35S:GUS transgene was compromised in sgs3, rdr6, hen1, and ago1 mutants but not in zll single mutants [12],[19],[20]. The newly identified ago1-40 EMS mutation causes an A to V amino acid change at position 863 of the protein, resulting in increased mRNA levels and protein activity and decreased siRNA levels for the L1 35S:GUS transgene (Figure 2 and Table S1). Unlike previously identified ago1 mutations that impair L1 PTGS with 100% efficiency, about 50% of ago1-40 adult plants at each generation had triggered PTGS, allowing us to test whether zll mutations affected L1 PTGS in ago1-40. To avoid any potential interference between the 35S promoters embedded in the T-DNA of the available insertional zll mutants in Col and the L1 35S:GUS transgene [21], we backcrossed five times to L1 the EMS-induced zll-3 mutant, which was isolated in the Ler accession [15]. L1/zll-3Col had similar GUS mRNA levels, protein activity and siRNA levels as silenced L1 controls (Figure 2) [12].
GUS mRNA levels in L1/ago1-40 zll-3Col double mutants were reduced in comparison to L1/ago1-40 mutants to nearly the level of silenced L1 controls (Figure 2). This increase in L1 silencing in the double mutant correlated with increased levels of GUS siRNAs. Seven days after germination (DAG), GUS siRNA levels were more than 10-fold higher than in L1/ago1-40 mutants, reaching levels comparable to silenced L1 controls 7 DAG, and by 15 DAG even exceeding L1 control levels (Figure 2). Thus, loss of ZLL function restored L1 gene silencing compromised in ago1-40.
To address whether zll mutations also were able to restore the miRNA pathway in ago1 hypomorphs, we analyzed miRNA levels and miRNA-regulated target genes in the ago1-27 zllago10 double mutants. miR398 levels were reduced and CSD2 mRNA and protein levels were substantially elevated in ago1-27 compared to zllago10-3 and zllago10-1 single mutants and wildtype (Figure 3A and Figure S3). By contrast, in both ago1-27 zllago10-3 and ago1-27 zllago10-1 double mutants, CSD2 mRNA and protein levels were reduced and miR398 levels were elevated, compared to ago1-27 alone (Figure 3A and Figure S3). The zllago10-3 mutation also restored miR164 accumulation and miR164-directed CUC2 silencing to wildtype levels in the ago1-27 background (Figure 3B). To extend our investigation to the whole-genome level, a transcriptome analysis was performed using Col wildtype, ago1-27, zllago10-1 and ago1-27 zllago10-1. Among 46 miRNA targets that were elevated in ago1-27 compared to wildtype but which were not affected in zllago10-1 single mutants, 19 were reduced completely or partially to wildtype levels in the ago1-27 zllago10-1 double mutant (Table S2). Taken together, loss of ZLL function restored L1 PTGS and silencing of approximately half of the miRNA targets deregulated in ago1-27.
The suppression of developmental, L1 silencing and miRNA pathway defects in hypomorphic ago1 mutants by zll mutations raised the question whether ZLL might be a negative regulator of AGO1. To test this hypothesis, we compared AGO1 mRNA and protein levels in ago1, zll and ago1zll double mutants. AGO1 protein levels were increased in both ago1-27 zllago10-3 and ago1-40 zll-3Col double mutants compared to the corresponding ago1 single mutants (Figure 4 and Figure S4). AGO1 mRNA and miR168 levels, however, were not significantly different (Figure 4). This indicates that ZLL is a negative regulator of AGO1 at the protein level, consistent with the role of ZLL in translational inhibition [7].
To determine whether the specific effects of ago1 and zll mutations could be explained by the expression patterns of AGO1 and ZLL, we first compared the expression patterns of pZLL:YFP-ZLL and pAGO1:CFP-AGO1 reporter genes. Both reporter constructs rescued the corresponding mutants, indicating that the fusion proteins are functional (Table S3) [17]. YFP-ZLL and CFP-AGO1 proteins were detected in a largely overlapping punctuate pattern outside the nucleus of expressing cells (Figure 5E–5J). As previously reported, pZLL:YFP-ZLL is initially expressed throughout the embryo, but becomes limited to provascular strands and the adaxial side of the cotyledons at about the globular stage (Figure 5A) [17]. By contrast, pAGO1:CFP-AGO1 is expressed in the whole embryo with the strongest signal in the provascular cells from globular stage to early torpedo stage (Figure 5C). Thus, ZLL and AGO1 expression patterns overlap partially, with the AGO1 expression pattern being broader than the one of ZLL, in agreement with mRNA localization results [14].
To evaluate the significance of the broad AGO1 expression pattern, we expressed AGO1 from the ZLL promoter and found that pZLL:AGO1 by and large restored development of ago1-1 (Table 1 and Figure S5) and ago1-27 (data not shown) mutants and also L1 PTGS in ago1-27 (Figure 6A–6B). However, miR398 accumulation and CSD2 silencing were only partially restored in ago1-27/pZLL:AGO1 (Figure 6C). These results suggest that limiting expression of AGO1 to the ZLL region is sufficient to provide most AGO1 functions in development and RNA silencing. Nevertheless, expression in cells outside the ZLL pattern is required to completely restore AGO1 activity.
Next, we addressed whether differences within ZLL and AGO protein sequences are responsible for differences in their functions by analyzing whether AGO1 could replace ZLL and vice versa. AGO1 expression from the ZLL promoter (pZLL:AGO1) rescued shoot meristem formation in the zll-1 mutant in the majority of cases (Table 2). By contrast, expression of ZLL from the AGO1 promoter (pAGO1:ZLL) in the strong ago1-1 allele resulted only in a slight reduction of leaf radialization compared to untransformed ago1-1 (Figure S6), but did not rescue any other developmental defect. Furthermore, in the ago1-27 hypomorph, pAGO1:ZLL was unable to rescue altered flowering time, reduced rosette size (Figure S7), L1 PTGS (Figure 6A and 6B) or CSD2 regulation (Figure 6C). Thus, whereas AGO1 can largely replace ZLL function in stem cell development, ZLL appears unable to efficiently replace the developmental, miRNA and PTGS functions of AGO1. Intriguingly, although pAGO1:ZLL did not restore CSD2 silencing in ago1-27, it fully restored miR398 accumulation to wildtype levels (Figure 6C). These results suggest that the intrinsic differences of AGO1 and ZLL proteins determine their specific contribution to small RNA and development pathways.
To address whether and if any ZLL and AGO1 protein domains have similar functions, we analyzed the ability of chimeric proteins composed of AGO1 and ZLL domains to rescue the respective mutant defects. As expected from the pZLL:AGO1 result, most chimeric ZLLAGO1 proteins (where one AGO1 protein domain was embedded in a ZLL protein backbone) driven from the ZLL promoter rescued shoot meristem formation of the zll-1 mutant (Table 2). The marked exception was the AGO1 N-terminal domain (pZLL:ZLLAGO1 N′) that could not efficiently replace the corresponding ZLL N-terminal domain (Table 2). This finding was unexpected since the complete AGO1 protein largely replaced ZLL, and might indicate that the function of the N-terminal domain is sensitive to the correct protein context.
On the converse, only the ZLL PAZ domain within the AGO1 backbone (pAGO1:AGO1ZLL PAZ) efficiently rescued developmental defects not only of the ago1-27 hypomorph (Figure 7L and Figure S7) but also of the null ago1-1 allele (Figure 7E, Table 1, and Figure S5). The ZLL PAZ domain also largely restored L1 PTGS and GUS siRNA accumulation, and CSD2 silencing and miR398 accumulation in ago1-27 (Figure 6). PTGS restoration, however, was delayed compared to the developmental rescue (Figures 6A and 7), consistent with previous findings that PTGS is more sensitive than development to compromised AGO1 activity [12]. By contrast, replacing the N-terminal or MID-PIWI domains of AGO1 with the corresponding ZLL regions (pAGO1:AGO1ZLL N′ and pAGO1:AGO1ZLL MID-PIWI) only restored bilateral leaf development but not sterility of ago1-1 mutants (Figure 7C, 7F, and 7G, Table 1, and Figure S5), or any developmental defects of ago1-27 mutants (Figure 7M and 7O and Figure S7). In addition, neither the N-terminal domain nor the MID-PIWI domains of ZLL were able to restore L1 PTGS and GUS siRNA accumulation or CSD2 silencing in ago1-27 (Figure 6). Since previous studies have indicated that PAZ, MID and PIWI domains function together in small RNA binding [11],[22],[23], we constructed a pAGO1:AGO1ZLL PAZ-PIWI chimera where the AGO1 genomic region containing PAZ, MID and PIWI domains was replaced by the corresponding ZLL genomic sequence (Figure S8). pAGO1:AGO1ZLL PAZ-PIWI resulted in similar effects as pAGO1:AGO1ZLL MID-PIWI (Table 1, Figures 6 and 7, and Figure S7). This suggested that the failure of the ZLL MID-PIWI domains to restore the majority of ago1 defects was not due to an incompatibility with the AGO1 PAZ domain or the disruption of the region connecting the PAZ and PIWI domains. Notably, although the pAGO1:AGO1ZLL PAZ-PIWI did not rescue CSD2 silencing, it restored miR398 accumulation in ago1-27 (Figure 6C).
In summary, these results indicate that the ZLL PAZ domain has the capacity to fulfill AGO1 functions in development, the miRNA pathway, and PTGS whereas the ZLL N-terminal and MID-PIWI domains are largely incompatible with AGO1 activity.
As part of the small RNA-directed RNA silencing machinery, the closely related ZLL and AGO1 proteins fulfill important roles during Arabidopsis development. Previous studies of mutant phenotypes indicate the presence of both, redundant, specific, and even opposite functions of ZLL and AGO1. Here, we investigate the diversity of ZLL and AGO1 functions and show that ZLL acts as a negative regulator of AGO1, and that the activities of the two proteins are determined by both functionally equivalent and distinct domains.
We find that double mutant combinations of strong zll and ago1 alleles are embryo lethal with strong patterning defects, revealed by abnormal expression of marker genes for the shoot and root meristem stem cell niche. This indicates that ZLL and AGO1 have a significant set of redundant functions required during early embryo development, in line with previous reports [14]. Although we have been unable to directly determine the small RNAs bound to ZLL due to the instability of the ZLL protein, we present several lines of indirect evidence suggesting that ZLL and AGO1 have partially redundant functions in small RNA-mediated silencing, and that ZLL domains are capable of binding a subset of small RNAs bound by AGO1: (1) Our protein domain swapping experiments indicate that the PAZ domain, which has been shown to bind small RNAs in several AGO proteins [11], is interchangeable between ZLL and AGO1, providing fully active proteins, (2) miR398 accumulation is restored to wildtype levels in an ago1 hypomorph by expression of pAGO1:ZLL, pAGO1:AGO1ZLL PAZ, pAGO1:AGO1ZLL MID-PIWI and pAGO1:AGO1ZLL PAZ-PIWI chimeras, and (3) both AGO1 and ZLL negatively regulate AGO1.
In addition to redundant functions of AGO1 and ZLL, our results using hypomorphic ago1 alleles to circumvent embryo lethality demonstrate opposing effects of ago1 and zll mutations. First, loss of ZLL function re-establishes both PTGS of the L1 transgene and miRNA-directed repression of a subset of target mRNAs deregulated in ago1-27, including miR398- and miR164 directed repression of their CSD2 and CUC2 targets, respectively. Furthermore, we observe partial suppression of hypomorphic ago1 leaf serration defects by zll mutations, which could be due to the partial re-establishment of miR164-directed CUC2 regulation in ago1 zll double mutants (Figures 1 and 3 and Figure S1) [24]. These opposite effects of ago1 and zll mutations are consistent with recent findings showing that mRNAs of leaf polarity-related HD-ZIP transcription factors and the corresponding miR165/166 are affected oppositely in zll and in ago1 single mutants (S. Bosca and T.L. unpublished) [9],[10],[25]. A plausible explanation for the restoration of developmental and RNA silencing defects caused by reduced AGO1 activity is provided by our finding that loss of ZLL activity results in upregulation of AGO1 protein levels in ago1-27. This negative regulation of AGO1 by ZLL suggests that homeostasis of AGO activity involves cross-regulation between different AGO proteins, which in the case of ZLL affects AGO1 protein but not mRNA levels, consistent with the recent implication of ZLL in translational repression [7]. Importantly, since ZLL expression itself is not a target of small RNA-mediated repression whereas AGO1 is [9],[26], ZLL has the potential to provide an input into RNA silencing activity that is independent of negative feedback dynamics and thus might serve to mediate, for example, developmental tuning of RNA silencing.
However, silencing of all miRNA targets deregulated in ago1-27 is not restored by the absence of ZLL function. One possible explanation is that upregulation of AGO1 protein levels in ago1 zll double mutants does not restore AGO1 activity completely to wildtype levels, which might be required for efficient silencing of a subset of target genes. Alternatively, since the miRNA pathway is cell autonomous [27],[28], the re-establishment of silencing of miRNA targets is expected to be limited to tissues where AGO1 and ZLL are co-expressed but will not take place in tissues where only AGO1 is expressed. This explanation is consistent with the pZLL:AGO1 analysis, where limiting AGO1 expression to the ZLL domain in ago1 mutants restored systemic L1 PTGS but did not fully restore miR398 accumulation and CSD2 regulation. Future experiments comparing AGO1, ZLL and miRNA tissue-specific expression will help to discriminate between these two possibilities.
Even though the sequences of ZLL and AGO1 proteins are closely related, the corresponding single mutants display different developmental defects. The pleiotropic ago1 mutants are defective in leaf morphology, general growth, and fertility, whereas zll mutants in the Ler accession display specific developmental defects in shoot apical meristem, flower, and silique development with allele specific penetrance. In contrast to the interchangeable PAZ domain, the non-conserved N-terminal domains, for which a function has yet to be assigned, cannot be exchanged between AGO1 and ZLL without loss of activity. Similarly, exchange of the MID and PIWI domains, which in AGO1 have been shown to provide selectivity for small RNAs possessing a 5′ U [22] and to function as a slicer domain that cleaves mRNA, respectively [3],[4], also cannot provide fully active proteins. This indicates that these domains contribute to functional differences. It is possible that the inability of the ZLL MID-PIWI fragment to replace the AGO1 domains reflects different preferences for 5′ nucleotide selectivity. Since the consensus amino acid residues essential for mRNA cleavage in several AGO1 proteins [29] are present in the ZLL PIWI domain, it is conceivable that both AGO1 and ZLL have the capacity to silence via mRNA cleavage and translational inhibition, but that each protein has a different preference for one of the two mechanisms, in line with recent findings [7].
Future dissection of AGO1 and ZLL properties will help to reveal how the interplay between AGO1 and ZLL proteins influences silencing specificity and efficiency in development.
The following mutants in the Col ecotype have been described previously: ago1-1 [6], ago1-27 [12], and zllago10-1 [18]. ago1-27 zllago10-3 mutants were generated using the zllago10-3 mutant (SALK_519738), which expresses a mis-spliced transcript that lacks part of exons 13 and 14 creating a frame-shift mutant with reduced levels of ZLL mRNA (Figure S1B). ago1-40 displays developmental defects (data not shown) similar in range, although much milder than ago1-27 [12]. zll-1, zll-3, and zll-15 mutants were isolated in the Ler accession as described [15]. Plants on soil were grown as described previously [30]. Plants on agar plates where grown on 1/2×MS supplemented with Gamborg vitamins (Sigma) and 10 g/l saccharose if indicated.
For RNA gel blot analyses, frozen tissue was homogenized in a buffer containing 0.1 M NaCl, 2% SDS, 50 mM Tris-Hcl (pH 9), 10 mM EDTA (pH 8) and 20 mM beta mercaptoethanol and RNAs were extracted two times with phenol. RNA gel blot analyses and quantification of GUS activity were performed as described [31]. Hybridization signals were quantified using a Fuji phosphor imager and normalized to a U6 oligonucleotide probe for miRNA gel blot analyses. GUS mRNA and GUS activity analyses were performed on the aerial parts of 7-day-, 15-day- and 21-day-old seedlings grown on Bouturage media (Duchefa) in 16 hours light, 8 hours dark at 22°C. For the CSD2 and miR398 analyses, seeds were germinated on media [32] without sucrose in both the presence and absence of 0.5 µM CuSO4, and plants were grown in 16 hours light, 8 hours dark at 22°C for 12 days at which time the aerial portion of the seedlings were harvested and homogenized in liquid nitrogen. For the CUC2, miR164, AGO1 and miR168 analyses, plants were grown for 10 days on media [32] in the presence of 0.5 µM CuSO4. For cDNA synthesis, RNAs were extracted with the Plant RNeasy kit (Qiagen), treated with DNAseI (Invitrogen) and l µg of DNA-free RNA was reverse transcribed with oligo-dT (Invitrogen). Quantitative real time (QRT)-PCR, was performed on a MasterCycler ep realplex (Eppendorf) with the RealMAster SYBR ROX mix (5PRIME) according to the manufacturer's protocol. Each reaction was performed on 5 µl of 1∶60 dilution of the cDNA and synthesized in a 20 µl total reaction. Specific oligonucleotide pairs were:
EF1a: 5′- CTGGAGGTTTTGAG GCTGGTAT -3′,
5′- CCAAGGGTGAAAGCAAGAAGA -3′;
CSD2: 5′- CAGAAGATGAGTGCC GTCATGCGG -3′,
5′- CCGAGGTCATCCTTAAGCTCGTG -3′;
CUC2: 5′- GCA CCAACACAACCGTCACAG -3′,
5′- GAATGAGTTAACGTCTAAGCCCAAGG-3′ and
AGO1: 5′- AAGGAGGTCGAGGAGGGTATG -3′,
5′- CAAATTGCTGAGCCAGAACAG -3′. The reactions were incubated at 95°C for 2 minutes to activate the hot-start recombinant Taq DNA polymerase, followed by 45 cycles of 15 seconds (s) at 95°C, 15 s at 60°C and 20 s at 68°C to ensure primer extension and to measure the fluorescence signal. The specificity of the PCR amplification procedures was checked with a heat dissociation protocol (from 60°C to 95°C) after the final cycle of PCR. The efficiencies of the primer sets were evaluated by performing QRT-PCR on several dilutions of a mix of the different strands. The results obtained on the different genotypes were standardized to the expression level of EF1a. For microarray analyses, RNAs were extracted using the RNeasy Plant Mini Kit (Qiagen), labelled according to the manufacturer's instructions using the Quick-Amp One-Color Labelling Kit (Agilent Technologies) and hybridized to Agilent custom microarrays. Three replicates were performed for each genotype.
Protein was extracted in buffer containing 50 mM Tris-HCl pH 7.5, 150 mM NaCl, 10% glycerol, Sigma Protease Inhibitor (CSD2) or 20 mM Tris-HCl pH 7.5, 300 mM NaCl, 5 mM MgCl2, 0.1% IGEPAL CA-630, 5 mM DTT, Sigma Protease Inhibitor (AGO1). Protein concentrations were determined using BioRad DC protein assay. Five µg (CSD2) and 80 ug (AGO1) of protein were resuspended in Laemmli buffer (20 mM Tris-HCl pH 6.8, 2% SDS, 5% glycerol, 40 mM DTT and 0.02% bromophenol blue), heated at 100°C for 5 minutes, and separated on a 15% (CSD2) or 6% (AGO1) SDS-PAGE gel. Proteins were transferred to PVDF membrane (BioRad). For detection, the membrane was blocked in 5% non-fat dry milk in 1×TBS, 0.1% Tween-20 (1×TBST) for 1 hour at room temperature, and incubated with a 1∶1000 dilution of CSD2 primary polyclonal antibody (Agrisera) or 1∶5000 dilution of AGO1 primary antibody ([33], Eurogenetech) in 5% non-fat dry milk and 1×TBST for 1.5 hours at room temperature. The membrane was then rinsed in 1×TBST for 45 minutes before incubation with a secondary peroxidase-conjugated anti-rabbit antibody (Sigma) in 5% non-fat dry milk in 1×TBST at room temperature for one hour. After the membrane was rinsed in 1×TBST for 45 minutes, CSD2 and AGO1 signals were revealed using the Western Lightning kit (PerkinElmer Life Sciences) kit and the Immunstar WesternC kit (Biorad) at the manufacturer's specifications.
For fluorescence studies, embryos where dissected from ovules using fine tip syringes in 10% glycerol, mounted on slides and analyzed using an AxioImager microscope (Zeiss) with YFP or CFP filter sets. Images were taken using Axiovision 4.4 software (Zeiss) and figures were generated using Photoshop 7.0 (Adobe). For confocal pictures, a Leica TCS SP2 AOBS spectral confocal microscope was used. Embryos were stained with DAPI (1 mg/ml) for 5 minutes and mounted in 50% glycerol in 1×PBS.
All AGO1 and ZLL sequences for both the fluorescent protein fusion and chimeric constructs are derived from the Col accession. AGO1 and ZLL chimeric constructs were made by exchanging five genomic domains; the 5′ sequence upstream of the ATG, the N-terminal, PAZ and the MID-PIWI domains and the 3′ region downstream of the stop codon. For cloning, restriction sites were introduced within introns at the appropriate positions (Table S4 and Figure S8). During the course of this work, we re-sequenced the ZLL Ler gene and several new ZLL cDNA clones and found that the original report of six amino acid differences between the ZLL Col and ZLL Ler proteins [15] was in error. The ZLL Ler amino acid sequence is identical to that of ZLL in Col, as previously published [14].
|
10.1371/journal.pgen.1007839 | RSM1, an Arabidopsis MYB protein, interacts with HY5/HYH to modulate seed germination and seedling development in response to abscisic acid and salinity | MYB transcription factors are involved in many biological processes, including metabolism, development and responses to biotic and abiotic stresses. RADIALIS-LIKE SANT/MYB 1 (RSM1) belongs to a MYB-related subfamily, and previous transcriptome analysis suggests that RSM1 may play roles in plant development, stress responses and plant hormone signaling. However, the molecular mechanisms of RSM1 action in response to abiotic stresses remain obscure. We show that down-regulation or up-regulation of RSM1 expression alters the sensitivity of seed germination and cotyledon greening to abscisic acid (ABA), NaCl and mannitol in Arabidopsis. The expression of RSM1 is dynamically regulated by ABA and NaCl. Transcription factors ELONGATED HYPOCOTYL 5 (HY5) and HY5 HOMOLOG (HYH) regulate RSM1 expression via binding to the RSM1 promoter. Genetic analyses reveal that RSM1 mediates multiple functions of HY5 in responses of seed germination, post-germination development to ABA and abiotic stresses, and seedling tolerance to salinity. Pull-down and BiFC assays show that RSM1 interacts with HY5/HYH in vitro and in vivo. RSM1 and HY5/HYH may function as a regulatory module in responses to ABA and abiotic stresses. RSM1 binds to the promoter of ABA INSENSITIVE 5 (ABI5), thereby regulating its expression, while RSM1 interaction also stimulates HY5 binding to the ABI5 promoter. However, no evidence was found in the dual-luciferase transient expression assay to support that RSM enhances the activation of ABI5 expression by HY. In summary, HY5/HYH and RSM1 may converge on the ABI5 promoter and independently or somehow dependently regulate ABI5 expression and ABI5-downstream ABA and abiotic stress-responsive genes, thereby improving the adaption of plants to the environment.
| The phytohormone abscisic acid (ABA) regulates multiple developmental processes in plants, including seed dormancy and germination, growth, and abiotic stress responses. The transcription factor ELONGATED HYPOCOTYL 5 (HY5), a core regulator of light signaling, is involved in ABA and abiotic stress responses by directly regulating the expression of ABA INSENSITIVE 5 (ABI5). In this study, we show that a MYB-related transcription factor, RADIALIS-LIKE SANT/MYB 1 (RSM1), plays important roles in ABA and salinity signaling in Arabidopsis, and we dissect the relationship between RSM1 and HY5. RSM1 interacts with HY5/HYH in vitro and in vivo and they may function as a regulatory module in responses to ABA and abiotic stresses. RSM1 binds to the promoter of ABI5, thereby regulating its expression; moreover, RSM1 interaction stimulates HY5 binding to the ABI5 promoter, but RSM1 was not found to enhance the activation of ABI5 expression by HY5. Genetic analyses reveal that RSM1 mediates the functions of HY5 in responses of seed germination, post-germination developmental responses to ABA and abiotic stresses, and seedling tolerance to salinity. In summary, our work demonstrates that HY5/HYH and RSM1 may bind with the ABI5 promoter to regulate ABI5 expression independently or somehow dependently, thereby controlling the expression of ABI5-downstream target genes in ABA and salinity signaling.
| Plants grow in a continuously changing environment that imposes various stresses. Abiotic stresses such as drought and salinity are amongst these environmental stresses [1, 2]. The plant hormone abscisic acid (ABA) is induced by abiotic stresses, and plays essential roles in plant responses and adaptation to those stresses, in addition to regulating several developmental processes, including seed development and maturation, dormancy and germination, seedling growth, and floral transition [3–6].
A family of novel START domain proteins known as PYR/PYLs/RCARs, are the best characterized ABA receptors, although several others have been documented [7, 8]. In the PYR/PYLs/RCARs-initiated core ABA signaling pathway, PYR/PYLs interact with and inhibit clade-A PP2Cs, including ABI1, ABI2, HAB1 and PP2CA/AHG3 [9, 10]. These PP2Cs negatively regulate ABA responses [7] by de-phosphorylating and inhibiting positive regulators of ABA signaling, e.g. a subfamily of ABA-activated SNF1-related protein kinases 2 (SnRK2s) including SnRK2.2, SnRK2.3 and SnRK2.6 in Arabidopsis [11]. SnRK2 kinases phosphorylate and activate a faimly of basic leucine zipper (bZIP) transcription factors called ABFs/AREBs, which include ABA INSENSITIVE 3 (ABI3), ABI4 and ABI5, as well ascertain ion channels and transporter proteins [12–15]. The ABFs bind to ABA-responsive promoter elements (ABRE) to induce the expression of ABA-inducible genes and thereby control seed germination and seedling development [5].
ELONGATED HYPOCOTYL 5 (HY5), a bZIP transcription factor, is the primary regulator of light signaling pathways in plants [16, 17]. HY5 functions downstream of phytochromes, cryptochromes, and UV-B photoreceptors to mediate photomorphogenesis under red, blue, far-red, and UV-B light [18–24]. Recent studies have revealed that HY5 is also involved in ABA signaling ABA and abiotic stress responses [18, 25–29]. HY5 plays regulatory roles in responses to ABA and NaCl during seed germination and seedling growth [27, 28]. ABA- and salinity-promoted ABI5 expression are both dependent on the presence of HY5 [27, 28]. Upon salinity stress, HY5-interacting protein COP1 is translocated to the cytosol to avoid destroying nucleus-localized HY5, thereby facilitating ABI5 expression [28]. In the mechanism underlying HY5 regulation of ABI5, HY5 may directly bind to the promoter of ABI5 to increase the expression of ABI5 and ABI5 target genes [27]. In addition, ABI5 can bind to its own promoter to promote its expression, while BBX21 negatively regulates ABI5 expression by interfering with HY5 binding to the ABI5 promoter [30]. In addition to its involvement in salinity stress responses, HY5 also regulates plant responses to cold stress and promotes the transcription of chilling responsive anthocyanin synthesis genes [29]. In this regard, HY5 can be an integrator of light signaling, ABA signaling and stress signaling [27]. HY5-HOMOLOG (HYH, AT3G17609), the closest homolog of HY5 in the Arabidopsis genome [31], also encodes a bZIP transcription factor, which plays a role in the phyB signaling pathway. HY5 and HYH may act together to regulate the expression of their target genes and thus mediate many important cellular processes [31].
The MYB family, one of the largest families of transcription factors in Arabidopsis, includes approximately 200 genes [32] with a highly conserved DNA-binding domain (MYB domain). MYB proteins are involved in many processes, including metabolism, cell fate and identity, developmental processes and responses to biotic and abiotic stresses [33]. Arabidopsis RSM1 (RADIALIS-LIKE SANT/MYB 1), encoded by At2g21650, belongs to the MYB-related subfamily of the MYB family [34]. RSM1 has other names, MEE3 (MATERNAL EFFECT EMBRYO ARREST 3) [35] or AtRL2 (ARABIDOPSIS RAD-LIKE 2) [36]. Mutation of the RSM1 gene affects female gametophyte development and embryogenesis in Arabidopsis [37]. RSM1, containing a SANT/MYB DNA-binding domain, is highly homologous to RADIALIS of Antirrhinum majus [38, 39]. Thus, RSM1 and its homologs in Arabidopsis were also designated the RAD-like family, which consists of four members: RADIALIS-LIKE SANT/MYB 1 (RSM1) (At2g21650), RSM2 (At4g39250), RSM3 (At1g75250) and RSM4 (At1g19510) [34]. RSM1-overexpressing seedlings are hookless and defective in gravitropism in the dark, while they display short hypocotyls under red light [34], indicating that RSM1 is involved in seedling photomorphogenesis. A previous work from our laboratory demonstrated that MEE3/RSM1 is a novel repressor of the floral transition by activating transcription of Flowering Locus C (FLC), a key flowering repressor [40]. Additionally, a transcriptome analysis suggested that RSM1 is highly expressed in guard cells and regulated by ABA and cold stress [41–44]. Moreover, transcription of RSM1 was found to be up-regulated 2 hours after treatment with cytokinin BA [45]. Therefore, RSM1 may play roles in plant development, stress responses and plant hormone signaling, but the molecular mechanisms underlying the roles of RSM1 in these processes remain obscure.
In our initial observations, RSM1-overexpressing plants exhibited opposite phenotypes to those of the mutants for HY5/HYH and shared similar phenotypes with those of HY5-overexpressing plants, with regard to seed germination, abiotic stress and ABA responses, seedling photomorphogenesis and the floral transition. These finding inspired us to speculate that the functions of RSM1 may be closely related to HY5/HYH. Questions arose regarding whether and how RSM1 and HY5/HYH are functionally associated in the biological processes listed above, but especially with regard to stress, ABA and light signaling pathways. In this work, we aimed to characterize the roles of RSM1 and homologs in plant responses to ABA and salinity during seed germination and seedling development, as well as to elucidate the relationships of RSM1 and homologs with HY5/HYH and ABA signaling components ABI5, ABI3, and ABI4. Our data suggest that RSM1 interacts with HY5/HYH to regulate responses of seed germination and seedling development to ABA and salinity in Arabidopsis.
To explore the effect of ABA and salt stress, quantitative RT-PCR (qRT-PCR) was conducted with imbibed seeds and seedlings. As shown in Fig 1A, RSM1 expression in imbibed seeds exposed to light was induced at one day of ABA treatment and subsequently repressed. RSM1 has three homologs: RSM2 (At4g39250), RSM3 (At1g75250) and RSM4 (At1g19510) [34]. The amino acid sequences of RSM2, RSM3, and RSM4 are highly similar to that of RSM1, with identities of 73.96%, 70.13% and 68.83%, respectively (S1 Fig). The expression levels of RSM2, RSM3 and RSM4 were also regulated by ABA during seed germination and seedling development (Fig 1B–1D). Similarly, RSM1 expression was also regulated by ABA and high salinity in 7-day-old seedlings of Col-0 wild-type (WT). As shown in Fig 1E, RSM1 expression was inhibited by ABA treatment at multiple time points (Fig 1E). The level of RSM1 expression was moderately induced at the beginning of treatment and then repressed compared to the initial level, when challenged with 200 mM NaCl (Fig 1F).
Taken together, the regulation patterns of RSM1 expression shown above suggest that RSM1 is involved in ABA and salinity stress responses during seed germination and seedling development.
Previous transcriptome analyses [42, 43] and our assays of RSM1 expression (Fig 1) suggest that RSM1 may be involved in ABA and abiotic stress signaling.
Therefore, seed germination and root elongation were measured for RSM1-related genetic materials treated with or without ABA, NaCl or mannitol at different concentrations. T-DNA insertion mutants of RSM1 (CS876657), RSM2 (CS371942) and RSM3 (Salk_069941C) were obtained from ABRC, whereas rsm1 rsm2 double mutants and rsm1 rsm2 rsm3 triple mutants were generated in our laboratory (S1 Fig), and qRT-PCR was used to measure the RSM1 expression level of each mutant using technique. As shown in S1E Fig, RSM1 expression was reduced in the rsm1 mutant, rsm1 rsm2 double mutant and rsm1 rsm2 rsm3 triple mutant in comparison with that of the WT plants. Moreover, transgenic RSM1-overexpressing plants OX-9 and OX-12 [40] were also assessed in this study.
In the absence of treatment, the germination rates of RSM1-related materials (including both mutants and overexpressing plants) showed no clear differences from that of the WT plants (Fig 2A and 2B). In contrast, overexpression of RSM1 gave rise to increased sensitivity to ABA, NaCl and mannitol treatments, whereas the single mutant for RSM1, rsm1, exhibited reduced sensitivity of seed germination to the ABA and NaCl treatments to different extents (Fig 2C–2G). Interestingly, the single mutants rsm2 and rsm3 displayed no clear differences from the WT plants while rsm3 was more sensitive to ABA, NaCl and mannitol earlier in the germination process (Fig 2C and 2E–2G). Additional information regarding the statistical analysis is included in S1 Table. With regard to cotyledon greening, there was no difference among the RSM1-related materials and the WT plants without treatment (S2A Fig). However, RSM1-overexpressing seeds were hypersensitive to ABA, NaCl and mannitol with regard to cotyledon greening, while the rsm1 rsm2 rsm3 triple mutant displayed reduced sensitivity to 1 μM ABA, and the rsm2 and rsm3 single mutants were more sensitive to 100 mM NaCl earlier in the germination process, as compared to control plants (S2B–S2D Fig). Additional information regarding the statistical analysis is included in S2 Table.
To evaluate the responses of early seedling development and growth to stress treatments, we also measured the fresh weight of 7-day-old RSM1-related materials after stratification and treatment with or without 1 μM ABA, 100 mM NaCl or 200 mM mannitol. The fresh weights of OX-9 and OX-12 seedlings were significantly lower than those of other genotypes following treatments with 1 μM ABA or 200 mM mannitol, but they were moderately lower than those of other genotypes following treatment with 100 mM NaCl. However, only the fresh weight of the rsm1 mutant seedlings displayed reduced sensitivity to treatment with 200 mM mannitol in comparison with that of the WT plants (S2E Fig).
In addition, we also determined whether RSM1 is involved in inhibition of root elongation by ABA. Five-day-old seedlings were transferred to plates supplied with or without ABA, NaCl and mannitol, and the root length was measured 5 d after the transfer. In the absence of treatment, the primary root lengths of the RSM1 mutants, RSM1-overexpressing plants, and control plants were similar, whereas the root length of the RSM1-overexpressing plants shorter than that of the WT plants under ABA treatment (S3 Fig). However, there were no clear differences among the the root lengths of any of the tested plants treated with NaCl or mannitol. These results suggest that RSM1 may be weakly involved in the regulation of primary root elongation by ABA.
The potential role of RSM1 in seedling responses to salinity stress was also assessed by examining the survival rates of various genotypes under high-salinity conditions. Seven-day-old seedlings were transferred to plates with or without 200 mM NaCl supplementation before measurements were collected. All seedlings grew well under the control conditions (Fig 3A), and only RSM1-overexpressing plants (OX-9 and OX-12 lines) exhibited a higher rate of survival than that of the WT plants, which suggests that RSM1 overexpression enhances the tolerance of seedlings to high salinity (Fig 3A and 3B). In addition, single mutant rsm3 and the triple mutant displayed reduced survival rates in comparison with that of the WT plants (Fig 3A and 3B). In concordance with the survival rate assay, RSM1-overexpression resulted in reduced relative electrolyte leakage in comparison with that of the WT plants, as revealed in electrolyte leakage assay (Fig 3C). However, the single, double, and triple mutants displayed no obvious differences in ion leakage in comparison with the WT plants (Fig 3C). Intracellular K+/Na+ homeostasis is important for plants responding to salt stress and adaption. K+/Na+ assays showed that salt-stressed RSM1-overexpressing seedlings accumulated more K+ and less Na+, whereas the mutants showed no evident differences in comparison with the WT plants (S4 Fig). These results suggest that RSM1 may enhance the tolerance of plants to high salinity during seedling development.
To analyze the organ specificity of the expression patterns of RSM1, we generated a proRSM1:GUS construct consisting of a 2.4-kb fragment of the RSM1 promoter to drive the GUS reporter gene. The proRSM1:GUS construct was transformed into the Col-0 wild-type background. RSM1 promoter activity was not detected in dry seeds (S5A Fig), but it was detected in all other tested organs, including the cotyledons, hypocotyls, radicles, true leaves and roots of young seedlings (S5B–S5H Fig), as well as rosette leaves (S5I Fig), floral organs (S5J and S5K Fig) and developing siliques (S5L Fig).RSM1 promoter activity was mainly observed in the vascular tissues of several organs (S5E–S5H Fig).
proRSM1:GFP-RSM1 plants were initially used to determine the cellular localization of RSM1, but possibly due to the low expression level of RSM1, no apparent signal was detected. Therefore, 35S:GFP-RSM1 lines were used to determine the cellular localization of RSM1. As shown in S5M Fig, GFP-RSM1 protein was localized in the nucleus and in the vicinity of the plasma membrane of epidermal cells in the cotyledons, hypocotyls, and roots of 5-day-old seedlings. GFP-RSM1 protein also localized in guard cells and the vascular tissues of the true leaves of seedlings (S6A Fig). These findings were in agreement with previous results from our laboratory [40]. In addition, we performed transactivation assay for RSM1 in yeast cells. As illustrated in S6B Fig, RSM1 did not exhibit transactivation activity despite confirmed expression in yeast (S6C Fig). However, the results of the cellular localization assays and bioinformatics analysis suggest that RSM1 is likely to function as a possible transcription factor.
To determine the mechanism by which RSM1 mediates ABA signaling, the expression levels of essential and marker genes in ABA signaling were analyzed in various RSM1-related genotypes during seed germination with or without ABA treatment. ABI5, RD29A, RD29B, AtEM1, AtEM6, RAB18, ABF3 and ABF4 were overall down-regulated in rsm1, rsm1 rsm2 and rsm1 rsm2 rsm3 germinating seeds in the absence or presence of ABA (Fig 4A and S7 Fig). Most of the tested genes were responsive to ABA (S7 Fig) [12, 13, 46–48]. The expression levels of genes upstream of the ABA signaling pathway, including ABI1, ABI2, SnRK2.2 and SnRK2.3 were not found to be apparently regulated by RSM1 (S7 Fig). Based on these observations, we speculate that RSM1 may function by regulating ABA and stress signaling at a particular node such as ABI5, while RSM1 does not seem to regulate major upstream ABA signaling components such as PP2Cs and SnRK2.2s.
To test whether RSM1 regulates ABI5 by binding to its promoter, yeast one-hybrid assays were performed using the 1800 bp genomic sequence before the start codon of the ABI5 gene. The ABI5 promoter was divided into six fragments (A, B, C, D, E and F), which were cloned into yeast one-hybrid reporter constructs to drive expression of the LacZ reporter gene (Fig 4B). As shown in Fig 4C, RSM1 bound only to fragment B, which contained the sequence extending from -703 to -375 bp before the start codon of the ABI5 gene. To map the specific RSM1 binding sites on the ABI5 promoter, fragment B was divided into six short fragments (approximately 60 bp) for electrophoretic mobility shift assays (EMSA). As shown in Fig 4D, the His-RSM1 protein was able to associate with fragment P2 (-648 to -593 bp) of the ABI5 promoter (Fig 4D). Chromatin immunoprecipitation (ChIP) assays were performed with the material from 12-day-old RSM1-overexpressing plants and the WT plants to determine whether RSM1 binds to the ABI5 promoter in vivo. The ChIP assays confirmed that RSM1 associated with fragment B, but not with the control sequence (ACTIN2) or fragment A, in vivo (Fig 4E). Taken together, these findings demonstrate that RSM1 may function as a transcription factor by binding to the ABI5 promoter to regulate ABI5 expression and thus influence the ABA signaling pathway.
The findings described above prompted to address whether RSM1 and ABI5 interact genetically in ABA signaling. To this end, an RSM1 overexpressing line (OX-12) was crossed with the abi5-7 mutant. The germination rates of abi5-7, OX-12 and OX-12 abi5-7 were similar to that of the WT plants in the absence of ABA (Fig 5A and 5B). However, in the presence of ABA, the germination rate of OX-12 was lower than that of the WT plants, whereas that of abi5-7 was higher than that of the WT plants. The differential ABA responses of OX-12 and abi5-7 seeds were more evident at higher concentrations of ABA. OX-12 abi5-7 was similarly less sensitive to ABA (Fig 5C–5E). The cotyledon greening rates of OX-12 abi5-7, OX-12, and abi5-7 were similar to that of the WT plants without ABA (Fig 5F). In the presence of ABA, OX-12 was more sensitive and abi5-7 was less sensitive to ABA regarding cotyledon greening, whereas OX-12 abi5-7 was similarly less sensitive to ABA as compared to abi5-7 (Fig 5G). Additional information regarding the statistical analysis is included in S3 Table. These findings suggest that ABI5 is downstream of RSM1 in the processes of seed germination and seedling development.
In addition to ABI5, ABI3 and ABI4 are also vital positive regulators of ABA signaling during seed germination and seedling development [14, 15, 49]. To analyze the genetic relationship of RSM1 with ABI3 and ABI4, OX-12 was crossed with abi3-8 or abi4-1. The seed germination rates and cotyledon greening rates of various genotypes were examined under varying concentrations of ABA (0, 1 μM and 5 μM). In the absence of ABA, no clear differences in the seed germination rate or cotyledon greening rate were observed among the tested genotypes (S8A, S8B and S8F Fig). However, in the presence of ABA, the germination rate and cotyledon greening rate of OX-12 were lower than those of the WT plants, whereas those of abi3-8 were higher than those of the WT plants. The responses of OX-12 abi3-8 and abi3-8 to the presence of ABA were similar (S8C, S8D, S8G and S8H Fig). Additional information regarding the statistical analysis is included in S4 Table.
The fresh weights of 7-day-old seedlings were measured after stratification and treatment with or without ABA (S8E Fig). The fresh weight of OX-12 seedlings was significantly lower than those of the seedlings of other genotypes under 1 μM ABA treatment, but OX-12 abi3-8 and abi3-8 both showed greater fresh weight under ABA treatment, displaying reduced sensitivity to ABA treatments (S8E Fig). These results suggest that ABI3 is downstream of RSM1 during seed germination and seedling development. Similarly, without ABA, the seed germination rates and cotyledon greening rates of the control, abi4-1, OX-12 and OX-12 abi4-1 plants were similar (S9A, S9B and S9F Fig). However, in the presence of ABA, the germination rate and cotyledon greening rate were lower in OX-12 but higher in abi4-1 as compared to those of the WT plants, while OX-12 abi4-1 and abi4-1 were similarly less sensitive to ABA (S9C–S9E and S9G Fig). Additional information regarding the statistical analysis is included in S5 Table.
The results described above suggest that ABI3, ABI4 and ABI5 are downstream of RSM1 to mediate the function of RSM1 in ABA-regulation of seed germination and post-germination.
Based on initial experiments as discussed earlier, we postulated that RSM1 and HY5/HYH may have a close relationship at the transcriptional regulation and/or protein interaction level.
To test whether HY5/HYH regulate the expression of RSM1, qRT-PCR was employed to measure the transcript levels of RSM1 in HY5/HYH-related genetic materials. As shown in Fig 6A, the transcript level of RSM1 was significantly down-regulated in hy5, hyh and hy5 hyh mutants as compared to that of the WT plants. Although transcription of RSM1 is regulated by HY5/HYH, it was unclear whether HY5 and HYH regulate transcription of RSM1 by binding to its promoter. By analyzing the promoter of RSM1 using an online tool (http://www.softberry.com), we identified a C-box sequence (GACGTC) located between -378 bp and -373 bp in the promoter region of RSM1, which has been predicted and experimentally demonstrated to be a HY5-binding site [50]. To determine how HY5/HYH binds to the RSM1 promoter, an approximately 1100 bp genomic sequence before the start codon of the RSM1 gene was divided it into four fragments, which were utilized in yeast one-hybrid assays (Fig 6B). HY5 and HYH were found to bind specifically to the sequence located between -523 bp and -268 bp in fragment B of the RSM1 promoter, which included the C-box mentioned above (Fig 6C). ChIP and EMSA were performed to obtain further confirmation of this interaction. As shown in Fig 6D, ChIP-qPCR revealed that HY5 could bind to the RSM1 promoter in vivo. In the EMSA assay, three mutated probe sequences, designated as m1, m2 and m3, were designed as unlabeled competitive probes to test the specificity of binding sites on the RSM1 promoter. As shown in Fig 6E–6G, the binding of HY5 and HYH to the C-box-containing sequence was effectively competed by unlabeled probes m1 and m3, which contained a wild-type C-box but not by probe m2, in which the C-box was mutated. Therefore, three different approaches demonstrate that HY5 and HYH specifically bind to the RSM1 promoter, which may facilitate their regulation of the activity of the RSM1 promoter.
Next, the effect of HY5 on RSM1 promoter activity was visualized using GUS reporter gene analyses. RSM1 promoter activity was detected in the cotyledons and hypocotyls of 3-day-old seedlings (S10A Fig) In cotyledons, RSM1 promoter activity was mainly detected in vascular tissues. Visually, proRSM1:GUS activity was notably decreased in the hy5 mutant background (S10B and S10C Fig) as compared to that of the wild-type background (S10A Fig), which suggested that the hy5 mutation reduced RSM1 promoter activity. These findings support the notion that regulation of RSM1 transcription may require the presence of HY5.
As a key positive regulator, HY5 plays an essential role in light signaling [25] in addition to its roles in the regulation of plant development and stress responses [27, 29, 30]. hy5 mutant plants exhibit hyposensitivity to ABA and salt treatments during seed germination and seedling growth [27]. As stated earlier, RSM1-related genetic materials displayed phenotypes relevant to those of hy5 mutants (Figs 2 and 3), which suggested that HY5 and RSM1 might be functionally related in the contexts of seed germination and seedling growth. To address this question, OX-12 was crossed with hy5, hyh or hy5 hyh, and phenotypical analyses were performed on the resulting plants. Wild-type, hy5, hyh, OX-12, OX-12 hy5 and OX-12 hyh exhibited similar germination rates on MS medium (Fig 7A and 7B). With ABA or NaCl supplementation, the germination rates of hy5 and hyh were much higher than those of the WT plants and the other tested genotypes. However, the germination rate of OX-12 hy5 was remarkably repressed relative to that of the WT plants, which mimicked the phenotype of OX-12 (Fig 7C–7G). Additional information regarding the statistical analysis is available in S6 Table. These results suggest that RSM1 is downstream of HY5/HYH and thus mediates the functions of both genes in the regulation of seed germination by ABA and salt stress.
In addition to seed germination, we also tested the tolerance of different genotypes to salt treatment to determine whether RSM1 and HY5/HYH had the same relationship as that revealed for the effects of NaCl on seed germination. To this end, we calculated the survival rates (rates of non-bleached seedlings) of different genotypes after 7-day-old seedlings were transferred to MS medium supplemented with 200 mM NaCl for 3 days. Surprisingly, hy5, hyh and hy5 hyh were sensitive to the NaCl treatment, whereas OX-12 was tolerant to the NaCl treatment in this assay (Fig 8A and 8B), although hy5 was less sensitive whereas OX-12 was sensitive to NaCl treatment in the germination assay. Apparently, OX-12 hy5, OX-12 hyh and OX-12 hy5 hyh mimicked OX-12 with regard to survival rate (Fig 8C and 8D), which indicated that RSM1 is also downstream of HY5/HYH in seedling responses to salinity stress.
These findings suggest that RSM1 is downstream of HY5/HYH in the responses of plants to high salinity in the germination and seedling developmental stages, although it seems that these genes/proteins play opposite roles in each stage.
Considering that RSM1 interacts with HY5 and HYH genetically, we questioned whether they also physically interact. To address this question, we carried out in vitro pull-down assays, in which HY5 or HYH was tagged with glutathione S-transferase (GST) and RSM1 was tagged with His. As shown in Fig 9A and 9B, the in vitro pull-down assays illustrated a direct physical interaction between RSM1 and HY5 or HYH. In vivo bimolecular fluorescence complementation (BiFC) assays were performed to confirm the results of the pull-down assays. YFPN-RSM1 and YFPC-HY5 or YFPC-HYH were transiently co-transformed into and expressed in Arabidopsis mesophyll protoplasts. After overnight incubation in the dark, a YFP signal resulting from complementation between YFPN-RSM1 and YFPC-HY5 or YFPC-HYH was successfully detected in some of the cells by confocal microscopy. These in vivo data confirmed the interaction between RSM1 and HY5 or HYH (Fig 9C). Moreover, BiFC assays with particle bombardment in onion epidermal cells confirmed the interactions described above and clearly showed that the interaction of RSM1 with HY5 or HYH takes place in the nucleus (S11 Fig). In summary, different assays firmly establish that RSM1 can physically interact with HY5 or HYH.
According to previous reports, HY5 directly binds to the promoter of ABI5 to regulate its expression [27, 30]. We have found that RSM1 binds to the ABI5 promoter (-703 bp to -374 bp) (Fig 4), which is different from the fragment (-1754 bp to -1294 bp) bound by HY5. Given that RSM1 interacts with HY5/HYH, we speculated that RSM1 may function as a transcriptional regulator instead of as a transcription factor to regulate HY5/HYH binding to the promoter of its target gene ABI5, so EMSA assays were performed to test this speculation. As shown in Fig 9D, GST-HY5 indeed binds to the fragment (-1754 bp to -1294 bp) of the ABI5 promoter; this fragment was reported previously [27]. His-RSM1 protein was unable to bind to this fragment of the ABI5 promoter (Fig 9D). However, increasing the amounts of His-RSM1 apparently enhanced the binding of GST-HY5 to the ABI5 promoter (Fig 9D). These findings support that RSM1 may function as a partner to enhance HY5 binding to the ABI5 promoter, and it likely does so via direct physical interaction with HY5.
As described above, several independent assays have established that RSM1 interacts with HY5/HYH, while the EMSA assays further indicated that RSM1 enhances HY5 binding to the ABI5 promoter. In this context, the functional implication of the interaction between RSM1 and HY5 remained unclear. To address this issue, qRT-PCR and dual-luciferase transient expression assays were conducted. As shown in Fig 10A, qRT-PCR assay revealed that overexpression of RSM1 (in OX-12 or OX-12 hy5) increased ABI5 transcript level in comparison with that of Col-0 or hy5, in the presence of wild-type HY5 (in Col-0 and OX-12), or in the presence of non-functional mutated hy5 (OX-12 and OX-12 hy5). These results can be explained that the functional HY5 is not required for the activation of ABI5 transcription by RSM1. In another word, RSM1 may have its own transcriptional activation activity for ABI5 transcription. This result is also consistent with the conclusion that HY5 is upstream of RSM1 in ABA and salinity signaling based on the epistasis genetic analyses. When a comparison was made between Col-0 and hy5, the functional HY5 (in Col-0) was still seen to promote ABI5 transcription. These results suggest that both HY5 and RSM1 activate ABI5 transcription, and RSM1 may possibly not enhance the function of HY5. In dual-luciferase transient expression assay, HY5 activated the ABI5 promoter-driven luciferase transcription, whereas RSM1 only had mild stimulation to ABI5 promoter activity. Surprisingly, no apparent additive effect was observed when both HY5 and RSM1 constructs were supplemented (Fig 10B and 10C). This result also supports that RSM1 may not enhance the function of HY5 in activating ABI5 promoter activity.
Transcriptional auto-regulation has been previously reported for ABI5 [30]. Unexpectedly, yeast one-hybrid assays showed that RSM1 may bind directly to its own promoter at the site located between -523 bp to -268 bp in fragment B of the RSM1 promoter (S12A and S12B Fig). To determine the specific binding motif, RSM1 promoter fragment B was divided into four sub-fragments of approximately 50 bp in length for the EMSA assay. As shown in S12C Fig, RSM1 was found to bind specifically to sub-fragments P1 and P2.
GUS reporter gene analysis was performed to evaluate the biological relevance of RSM1 to RSM1 promoter self-binding in the context of RSM1 expression. Five-day-old seedlings of proRSM1:GUS in the Col-0 or rsm1 background were subjected to GUS staining. In comparison with the Col-0 background, plants of the rsm1 background showed dramatically decreased RSM1 promoter activity (S12D–S12F Fig). These results reveal a new regulatory mechanism for RSM1 transcriptional regulation and its role in regulating RSM1 responses to stresses.
As one of the largest plant transcription factor families, MYB transcription factors play important roles in plant growth and abiotic stress responses [33, 35]. Genetic analyses with loss-of-function mutants have shown that RSM1 is possibly required for female gametophyte development [37]. In addition, overexpression analyses suggest that RSM1 is also possibly involved in seedling morphogenesis [34, 42, 43]. In a previous study we demonstrated that RSM1 may act as a novel repressor of the floral transition by activating FLC via direct binding to its promoter [40]. Additionally, several other sources of evidence hinted at possible roles of RSM1 in ABA and abiotic stress signaling. Transcriptome analyses revealed that RSM1 is transcriptionally regulated by ABA and exposure to cold temperatures [34, 35]. RSM1 is down-regulated in XERICO-overexpressing plants which have increased tolerance to drought stress [44]. In addition, RSM1 expression is induced by cytokinins and up-regulated in esk1, a mutant with strong tolerance to freezing [41, 45]. The information described above implicates that RSM1 may be versatile in plant development, plant hormone signaling, and stress responses.
In the present study, the expression patterns of RSM1 revealed by qRT-PCR and GUS reporter gene analyses (Figs 1 and S5) provide important information regarding the functions of RSM1 in relevant developmental processes and responses to environmental stresses. RSM1 is localized in the nuclei of stomatal guard cells and vascular tissues (S5 and S6A Figs), suggesting that it may play a role in guard cell function and vascular transport.
Genetic analyses have systematically revealed the versatility of RSM1 in many aspects. As shown in the present study, RSM1 acts as a positive regulator of ABA signaling, but it has a negative effect on tolerance to salinity and dehydration during seed germination (Fig 2), which may endow seeds with the appropriate level of sensitivity to ABA and stressful environments. In contrast, RSM1 acts as a positive regulator of salt tolerance during seedling development (Fig 3), which facilitates seedling survival under salt stress. The differential roles of RSM1 in salinity tolerance at different developmental stages may reflect different regulatory mechanisms for different biological processes. Our observations also support the previous finding [34] that RSM1 plays a positive role in seedling photomorphogenesis under red light. Furthermore, RSM1 also plays a negative role in the floral transition, as previously reported [40]. These findings show conclusively that RSM1 plays important roles in the regulation of seed germination and seedling development by ABA or abiotic stresses, in addition to several other biological processes.
Note that the loss-of-function mutant rsm1 used in this study only exhibits clear phenotypes for particular biological processes, whereas RSM1-ovexpressing plants display much stronger phenotypes. This phenomenon may be ascribed to the fact that RSM1 is in low abundance in planta, and that the rsm1 mutant is not a null allelic mutant. Unfortunately, the null allelic mutant for RSM1 is arrested at the one-cell zygotic stage, as reported by Pagnussat et al. (2005) [37], thereby rendering it impractical for many functional analyses. From another point of view, the phenomenon described above may also be accounted for by the redundancy of RSM1 and its homologous genes in terms of the biological processes assessed in this study. As described earlier”, RSM1 has three homologous genes: RSM2, RSM3 and RSM4. We have made rsm1 rsm2 double and rsm1 rsm2 rsm3 triple mutants, but these mutants do not show clear phenotypes in some assays. We are currently using the CRISPR/Cas9 approach to construct an rsm1 rsm2 rsm3 rsm4 quadruple mutant.
We have established the involvement of RSM1 in multiple biological processes including ABA and abiotic stress responses during seed germination and post-germination seedling development. Intriguingly, HY5 and HYH are also involved in these processes [17, 25, 27, 29]. In the present study, we provide several independent lines of evidence to support our hypothesis that RSM1 is closely linked to HY5/HYH in these biological processes.
First, RSM1 expression is regulated by HY5/HYH at the transcriptional level. As revealed by qRT-PCR analyses, RSM1 expression is down-regulated in hy5, hyh and hy5 hyh mutants (Fig 6A). We provide another line of evidence from GUS reporter gene analyses to support the notion that RSM1 expression is regulated by HY5 at the transcriptional level. proRSM1:GUS activity is dependent on the presence of HY5 under normal conditions (S10 Fig). Furthermore, our in vitro and in vivo assays, including yeast one-hybrid, EMSA, and ChIP-qPCR assays, revealed the biochemical mechanism underlying regulation of RSM1 by HY5 at the transcriptional level, in which HY5 and HYH bind specifically to the C-box of the RSM1 promoter (Fig 6B–6G). We thus conclude that HY5 and HYH may regulate RSM1 expression at the transcriptional level, by specific binding to the RSM1 promoter. This regulatory mechanism (e.g., promoter binding, and expression regulation) is a common means by which target genes are regulated by transcription factors [51, 52]. Indeed, many genes including RSM1 of the HY5 regulon have been revealed by ChIP-chip assays [53]. Our data described above confirm the observation for HY5 binding to the RSM1 promoter.
Second, RSM1 and HY5/HYH interact directly. Our in vitro pull-down assays (Fig 9A and 9B) showed a direct physical interaction between RSM1 and HY5 or HYH. In vivo bimolecular fluorescence complementation (BiFC) assays in mesophyll protoplasts confirmed the interaction between RSM1 and HY5 or HYH (Fig 9C). In addition, BiFC assay with particle bombardment in onion epidermal cells further confirmed the interactions described above, and showed that the interaction of RSM1 with HY5 or HYH takes place in the nucleus (S11 Fig). As a transcription factor, HY5 regulates ABI5 [27], RSM1 (the present study), and many other target genes [53]. In addition, other factors influence the manner in which HY5 regulates its target genes. As shown previously, transcriptional regulator BBX21 interferes with the binding of HY5 to the ABI5 promoter [30]. In our assay, the binding of HY5 to the ABI5 promoter is stimulated by RSM1 (Fig 9D). Therefore, RSM1 may possibly act as a positive transcriptional regulator in this case. With regard to influencing regulation of ABI5 by HY5, the function of RSM1 may be fulfilled via its direct physical interaction with HY5. However, our dual-luciferase transient expression assay suggests that RSM1 does not enhance the activation of the ABI5 promoter by HY5.
Third, RSM1 genetically interacts with HY5 and HYH. Our genetic analyses uncovered the existence of this genetic relationship. RSM1 resides downstream of HY5 and HYH during seed germination and post-germination seedling development and stress tolerance (Figs 7 and 8), no matter whether RSM1 plays a positive or negative role. The relationships between HY5/HYH and RSM1 at the transcriptional and protein levels may provide the molecular basis for their genetic relationship.
Although RSM1 may be involved in ABA responses, it is unclear whether RSM1 regulates ABA biosynthesis or signaling. No effect of RSM1 on ABA content was observed in this study, raising the possibility that RSM1 likely regulates ABA signaling rather than ABA biosynthesis. Indeed, qRT-PCR assays revealed that RSM1 regulates the transcript levels of many ABA-responsive or stress-responsive genes, such as ABI5, RD29A, RD29B, AtEM1, AtEM6, RAB18, ABF2, ABF3 and ABF4, during seed germination (Fig 4A and S7 Fig). Considering that ABI5 is a crucial positive regulator of ABA signaling [13, 46, 47], our results confirm the role of RSM1 in ABA signaling. Given that RSM1 up-regulates ABI5 expression (Fig 4A), our yeast one-hybrid and ChIP assays show that RSM1 can bind to the promoter of ABI5 to induce transcriptional activation of ABI5 (Fig 4). The question of whether RSM1 acts as a transcription factor to control ABI5 expression is clearly prompted by our findings. Although no transactivation activity was detected for RSM1 in yeast cells (S6 Fig), our data suggest that RSM1 can function as a transcription factor to regulate ABI5 expression, as well as act as a regulator to interact with HY5/HYH.
Our genetic analyses establish the genetic relationship between RSM1 and ABI5 in ABA signaling. We found that ABI5 is downstream of RSM1 in the ABA signaling pathways governing seed germination and seedling development (Fig 5). ABI3 is a B3-domain-containing transcription factor that physically interacts with ABI5 [54], while also functioning as an essential upstream regulator and activator of ABI5 expression in the context of ABA signaling [49]. ABI4, an AP2/ERF transcription factor, is also important for ABA signaling during seed development and germination [14]. Like ABI3, ABI4 acts as a transcription activator to induce ABI5 expression, by binding directly to its promoter [55]. Our genetic analyses demonstrate that similar to ABI5, both ABI3 and ABI4 are downstream of RSM1 to mediate the functions of RSM1 in the regulation of seed germination and post-germination by ABA (S8 and S9 Figs). These results establish that RSM1 plays an important role in ABA signaling during seed germination and early seedling development.
In our assays, RSM1 directly binds to the ABI5 promoter and regulates ABI5 expression (Fig 4). HY5 binds directly to the ABI5 promoter [27]. Interestingly, ABI5 binds to its own promoter [30]. Both HY5 and ABI5 belong to the same bZIP transcription factor family, and preferentially bind to the G-box motif. However, they bind to different G-box motifs; HY5 binds to a typical G-box motif [27] located 500 bp upstream of the ABI5-binding site within the fragment located 1127–1231 bp upstream of the start codon [30]. Several other transcription factors also bind directly to the ABI5 promoter and regulate ABI5expression. FHY3/FAR1 bind to the FHY3/FAR1-binding site (FBS) [56] located 130 bp downstream of the G-box motif to which ABI5 binds [30]. ABI4 binds to a CE1-like element in the 5′-untranslated region of ABI5 and activates its expression [55]. In addition, ABI3, a B3-domain containing transcription factor, functions as an essential regulator upstream of ABI5 [31]. Determining whether ABI3 is also a direct regulator of ABI5 will require further investigation. Therefore, HY5, RSM1, FHY3/FAR1, ABI4 and ABI5 bind directly to the ABI5 promoter, but they seem to occupy different regions. Future studies should assess how these factors are coordinated and whether they could regulate the activity of one another on the ABI5 promoter. Interestingly, BBX21 was recently reported to interfere with HY5 binding to and thereby repressing the ABI5 promoter [30]. BBX21 is the only known negative transcriptional regulator for the ABI5 promoter. Although our dual-luciferase transient expression assay does not show that RSM1 strengthens the activation of ABI5 expression by HY5, our EMSA and protein-protein interaction data suggest that RSM1 may work as a partner to enhance binding of HY5 to the ABI5 promoter, possibly via direct physical interaction with HY5. Apart from regulators upstream of ABI5, many ABA-responsive and stress-responsive genes are present downstream of ABI5 and are directly regulated by ABI5. Thus, the ABI5 promoter may represent a convergence point at which transcriptional regulators of the ABA and abiotic stress signaling pathways integrate environmental stimuli by fine-tuning the expression of ABI5 and ABI5 target genes.
When subjected to abiotic stress or ABA, plants up-regulate expression of HY5/HYH. RSM1 may be up-regulated or down-regulated depending on the duration of exposure to ABA or abiotic stresses. There exists a regulatory mechanism in which HY5/HYH up-regulate RSM1 expression by binding to the RSM1 promoter. The protein RSM1 is also involved in the process of regulation of RSM1 expression via direct binding to its own promoter. Thus, fine-tuning of RSM1 expression may be achieved via the regulatory loop formed by both HY5/HYH and RSM1. As a direct target, ABI5 is up-regulated by HY5 via binding to the ABI5 promoter [27]. RSM1 may also regulate ABI5, and this regulatory mode may be complex. RSM1 may function as a transcription factor by directly binding to the ABI5 promoter to accomplish up-regulation of ABI5 expression. On the other hand, RSM1 may also function as a possible partner, interacting with HY5/HYH, although no clear evidence supports that RSM1 enhances the HY5 activation of ABI5 expression. In summary, through the mechanisms described above, RSM1 and HY5/HYH may converge on the ABI5 promotor, and independently or possibly dependently regulate ABI5 expression and ABI5-targeted ABA-responsive genes, and thereby modulate ABA and abiotic stress responses (Fig 11).
All Arabidopsis plants used in this study were of the Columbia-0 (Col-0) ecotype. The following mutants were used in this work: rsm1 (CS876657) [34], rsm2 (CS371942), rsm2 (Salk_069941C), rsm1 rsm2, rsm1 rsm2 rsm3, hy5-215 (denoted as hy5 in the text and figures) [16], hyh (CS849765) [57], hy5 hyh [58], abi5-7 [59], abi3-8 [59], and abi4-1 [14]. T-DNA insertion mutants rsm1 (CS876657) [34], rsm2 (CS371942), and rsm2 (Salk_069941C) were obtained from the Arabidopsis Biological Resource Center. T-DNA insertions were confirmed by PCR on genomic DNA and sequencing of the left and right borders. rsm1 rsm2 and rsm1 rsm2 rsm3 mutants were generated by genetic crosses and confirmed by genomic PCR. 35S:TAP-HY5/hy5 was obtained from the Xing-Wang Deng laboratory at Peking University. Transgenic RSM1-overexpressing plants OX-9 and OX-12 [40] were also used in this study.
Seeds were surface-sterilized and stratified at 4°C for 3 days, sown onto MS media (pH 5.7–5.9) containing 1.0% sucrose and 0.8% agar, and grown at 22°C under long-day condition (16-h day/ 8-h night) for one week. Ten-day-old seedlings were then transferred to soil and grown at 22°C under long-day condition (16-h day/ 8-h night).
For the GUS reporter gene essays, the ~2.4-kb long promoter of the RSM1 gene was amplified and cloned into the pBI121 vector to generate proRSM1:GUS. The construct was then transformed into Agrobacterium tumefaciens GV3101 and subsequently introduced into Arabidopsis Col-0 by using the floral dip method [60].
Seed germination assays were conducted as described previously [61]. Briefly, the same batches of seeds for all genotypes were surfaced sterilized, stratified at 4°C for 3 days, and plated on MS media (pH 5.7–5.9) containing 1.0% sucrose and 0.8% agar at 22°C under long-day conditions (16 h/8 h light/dark).Seed germination rates and cotyledon greening rates were typically scored and calculated every day for seven days after stratification. See the figure legends for details regarding specific days of counting and treatment with ABA, NaCl or mannitol.
To analyze the root length of seedlings, 5-day-old seedlings grown on MS plates were transferred onto MS plates supplemented with 10 μM ABA, 20 μM ABA (Sigma), 50 mM NaCl, 100 mM NaCl, 100 mM mannitol, or 200 mM mannitol. The plates were vertically placed at 22°C in under 16-h/8-h light/dark long-day conditions for 5 days before the lengths of the primary roots were measured.
Seven-day-old seedlings (n ≥ 25) grown under normal conditions were transferred to MS media supplemented with 0 or 200 mM NaCl and grown for 3–4 days. No less than 25 seedlings were counted for the assessment of the survival rate of each genotype.
For the relative electrolyte leakage assay, salt-treated seedlings were washed with ddH2O and placed into 15 mL BD tubes containing 8 mL ddH2O. The tubes were shaken at 180–220 rpm at 22°C for 1 h, after which measurement S1 (μS/cm) was acquired using a conductivity meter (Mettler Toledo, Columbus, OH, USA). Next, the tubes were boiled for 30 min, cooled and shaken for 1–2 h at 180–220 rpm at 22°C, after which measurement S2 (μS/cm) was acquired. The reading S0 (μS/cm) was acquired from the ddH2O control. The relative electrolyte leakage was calculated as follows: EL (%) = (S1-S0) / (S2-S0).
GUS staining assays were performed as described previously [62] unless stated otherwise. In brief, plant material samples were fixed with iced 90% (v/v) acetone at room temperature for 20 min and washed with iced staining buffer (50 mM sodium phosphate, 0.1% (v/v) Triton X-100, 1 mM Na2EDTA, 1 mM potassium ferricyanide and 1 mM potassium ferrocyanide, pH 7.0) for twice on ice. The washed materials were then incubated in GUS staining solution (staining buffer with 1 mg/mL 5-bromo-4-chloro-3-indolyl-β-glucuronic acid) at 37°C overnight. The tissue samples were cleared of chlorophyll in 7:3 (v/v) ethanol and acetic acid, after which they were twice washed with 70% (v/v) ethanol. Images were taken using a stereomicroscope (Leica, Wetzlar, Germany). The measurement of GUS activity was performed using 4-methylumbelliferyl glucuronide as described by Jefferson [62, 63].
To visualize the subcellular localization of the GFP-RSM1 fusion protein, 35S:GFP-RSM1 seedlings were mounted on slides and examined under a Zeiss LSM 710 confocal microscope. GFP fluorescence was detected at 488 nm (excitation) and 490–550 nm (emission). DAPI was used to mark the nuclei.
Total RNA was extracted from one-day germinated seeds or 7-day-old Arabidopsis seedlings using the EasyPure Plants RNA Kit (TransGen, Beijing, China). After DNA depletion by DNase I (TransGen, Beijing, China), 1 μg total RNA was used to synthesize cDNA using ReverTra Ace qPCR RT Master Kit (Toyobo Co., Ltd., Osaka, Japan). Quantitative real-time PCR analysis was performed using SYBR Premix Ex Taq (Takara, Tokyo, Japan) in an ABI 7500 fast real-time instrument (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s instructions. The relative expression levels were normalized to internal control ACTIN2. The qRT-PCR assays were performed with three biological replicates, and three technical replicates were performed in each biological replicates. Information regarding the primers used in this assay is available in S7 Table.
Yeast one-hybrid assays were performed as described previously [64]. The ABI5 promoter, an approximately 1200 bp sequence located upstream of the ATG start codon, was divided into six fragments, which were designated A-F. The ABI5 promoter fragments were constructed into the pLACZ2U plasmid which has a lacZ reporter gene. The RSM1 promoter (1100 bp) was divided into four fragments, which were designated A-D. The RSM1 promoter fragments were constructed into the pLACZ2U plasmid. RSM1 CDS, HY5 CDS and HYH CDS were each constructed separately into the pB42AD plasmid. Both plasmids were introduced into yeast strain EGY48 grown on SD/gal/raf-trp-ura medium containing 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside (X-GAL) and BU salts. The yeast transformation and liquid assays were performed as described in the Yeast Protocols Handbook (Clontech, Mountain View, CA, USA). Images were taken using a digital camera (Nikon).
Electrophoretic mobility shift assays (EMSAs) were performed according to the result from the yeast one-hybrid assay. Escherichia coli strain BL21 transformed with pET28a-RSM1 was induced to express His-RSM1 with 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). The His-RSM1 fusion protein was purified with Ni-NTA beads (Qiagen, Hilden, Germany). Labeled and unlabeled probes were synthesized by Invitrogen. Protein-DNA binding assays were performed using the Scientific Light-Shift kit (Thermo Fisher Scientific). Briefly, 2 μg of His fusion proteins or GST fusion proteins were incubated together with biotin-labeled probes in a 20 μL reaction mixture containing 10 mM Tris-HCl pH 7.5, 50 mM KCl, 1 mM DTT and 50 ng/μL poly (dI∙dC). The reactions were incubated at 25°C for 20 min and separated on 4–6% native polyacrylamide gels in 0.5×TBE buffer. The gels were electroblotted to Hybond N+ nylon membranes (Millipore, Burlington, MA, USA) in 0.5×TBE for 40 min, after which the labeled probes were detected according to the instructions provided with the EMSA kit.
ChIP assays were performed as previously described [65]. Twelve-day-old Col and OX-12 seedlings grown under long-day conditions (16-h day/8-h night) were harvested and subjected to ChIP-qPCR assays using a rabbit polyclonal antibody against RSM1 to immunoprecipitate genomic DNA segments. The enrichment of DNA was analyzed by qRT-PCR. Information regarding the primers used in this assay is available in S7 Table. The enrichment value (%) was normalized to the amount of input DNA.
The full-length HY5 and RSM1 CDSs were cloned into the pGreen II 62-SK vector to generate the effector vectors, which were driven by the cauliflower mosaic virus 35S promoter. The 2-kb ABI5 promoter sequence was cloned into the pGreen II 0800 vector driving firefly luciferase to generate the proABI5 reporter vector. Renilla luciferase driven by a full-length cauliflower mosaic virus 35S promoter was used as an internal control. Vectors were transformed into Arabidopsis Col-0 WT mesophyll cell protoplasts for transient expression as described previously [66]. The transfected protoplasts were cultured at 22°C in the dark for 12 h, and firefly luciferase and Renilla luciferase activities were measured using the Dual-Luciferase Reporter Assay System according to the instruction manual (Promega, Madison, WI, USA).
Expression constructs for His-RSM1 were generated by cloning the CDS of RSM1 into the BamHI and SacI enzyme sites of vector pET28a (Novagen, Millipore, Burlington, MA, USA). The expression constructs for GST-HY5 and GST-HYH were generated by cloning the corresponding CDSs into the EcoRI and XhoI sites of vector pGEX-4T-1 (Amersham, Little Chalfont, UK). Two micrograms of GST or GST fusion proteins were mixed with 2 μg of His-RSM1 in 500 μL GST binding buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, and 0.1% Nonidet P-40), after which the mixture was rotated at 4°C for 2 h. Before the proteins were mixed, the Glutathione Sepharose 4B beads were washed with GST binding buffer, which was then kept rotating at 4°C for 1 h. After four washes with GST binding buffer, the GST resin was boiled with 1×SDS loading buffer and subjected to SDS-PAGE and western blotting.
The full-length CDSs of HYH and HY5 were amplified and cloned into the SacI and SpeI sites of the pSY735 (C terminus of yellow fluorescent protein [YFPC]) vector to generate plasmids YFPC-HYH and YFPC-HY5. Meanwhile, the full-length CDS of RSM1 was amplified and cloned into the SpeI and BamHI sites of the pSY736 (YFPN) vector [67], resulting in plasmid YFPN-RSM1. The plasmids were extracted and concentrated to 2 mg/mL. The in vivo interactions were assayed by transformation using Arabidopsis protoplasts [66] or particle-mediated transformation using onion epidermal cells [68]. After overnight incubation in the dark, the YFP signal was detected using a Zeiss LSM 710 confocal microscope. DAPI was used to mark the nuclei.
For immunoblotting, seedlings of Arabidopsis (Col-0 and other genotypes) were harvested in protein extraction buffer containing 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 10 mM MgCl2, 1 mM EDTA, 10 mM NaF, 2 mM Na3VO4, 25 mM β-glycerol phosphate, 10% (vol/vol) glycerol, 0.1% (vol/vol) Nonidet P-40, 1 mM PMSF and 1× cOmplete Protease Inhibitor Mixture. In brief, protein samples were separated by SDS-PAGE, after which the separate proteins were transferred to a polyvinylidene fluoride film. The film was blocked with 5% milk, incubated with the selected primary antibody overnight at 4°C, washed three times with 1× PBST (5 min each), and incubated with the selected secondary antibody for 1 h at room temperature. After three washes with 1×PBST (5 min each), the film was illuminated and photographed under a Bio-Rad illumination detection device.
For most experiments, Student’s t test was employed to analyze the significance of differences between each treatment group and the appropriate control group. However, for the analyses of seed germination and cotyledon greening, one-way ANOVA and Fisher’s least significant difference (LSD) test were conducted using IBM SPSS Statistics version 20.0 (IBM Corporation, Armonk, NY, USA).
Sequence data from this article can be found in the Genome Initiative or GenBank/EMBL databases under the following accession numbers: RSM1/MEE3 (At2g21650), RSM2 (At4g39250), RSM3 (At1g75250), RSM4 (At1g19510), HY5 (At5g11260), HYH (At3g17609), ABI5 (At2g36270), ABI3 (At3g24650), ABI4 (At2g40220), RD29A (At5g52310), RD29B (At5g52300), SnRK2.2 (At3g50500), SnRK2.3 (At5g66880), ABI1 (At4g26080), ABI2 (At5g57050), AtEM1 (At3g51810), AtEM6 (At2g40170), RAB18 (At5g66400), ABF2 (At1g45249), ABF3 (At4g34000), ABF4 (At3g19290), ACTIN2 (At3g18780) and PP2A (At1g69960).
|
10.1371/journal.pgen.1005504 | Genome Wide Identification of SARS-CoV Susceptibility Loci Using the Collaborative Cross | New systems genetics approaches are needed to rapidly identify host genes and genetic networks that regulate complex disease outcomes. Using genetically diverse animals from incipient lines of the Collaborative Cross mouse panel, we demonstrate a greatly expanded range of phenotypes relative to classical mouse models of SARS-CoV infection including lung pathology, weight loss and viral titer. Genetic mapping revealed several loci contributing to differential disease responses, including an 8.5Mb locus associated with vascular cuffing on chromosome 3 that contained 23 genes and 13 noncoding RNAs. Integrating phenotypic and genetic data narrowed this region to a single gene, Trim55, an E3 ubiquitin ligase with a role in muscle fiber maintenance. Lung pathology and transcriptomic data from mice genetically deficient in Trim55 were used to validate its role in SARS-CoV-induced vascular cuffing and inflammation. These data establish the Collaborative Cross platform as a powerful genetic resource for uncovering genetic contributions of complex traits in microbial disease severity, inflammation and virus replication in models of outbred populations.
| New emerging pathogens are a significant threat to human health with at least six highly pathogenic viruses, including four respiratory viruses, having spread from animal hosts into the human population within the past 15 years. With the emergence of new pathogens, new and better animal models are needed in order to better understand the disease these pathogens cause; to assist in the rapid development of therapeutics; and importantly to evaluate the role of natural host genetic variation in regulating disease outcome. We used incipient lines of the Collaborative Cross, a newly available recombinant inbred mouse panel, to identify polymorphic host genes that contribute to SARS-CoV pathogenesis. We discovered new animal models that better capture the range of disease found in human SARS patients and also found four novel susceptibility loci governing various aspects of SARS-induced pathogenesis. By integrating statistical, genetic and bioinformatic approaches we were able to narrow candidate genome regions to highly likely candidate genes. We narrowed one locus to a single candidate gene, Trim55, and confirmed its role in the inflammatory response to SARS-CoV infection through the use of knockout mice. This work identifies a novel function for Trim55 and also demonstrates the utility of the CC as a platform for identifying the genetic contributions of complex traits.
| Severe Acute Respiratory Coronavirus (SARS-CoV) emerged in humans in Southeast Asia in 2002 and 2003 after evolving from related coronaviruses circulating in bats [1,2]. SARS-CoV caused an atypical pneumonia that was fatal in 10% of all patients and 50% of elderly patients [3,4]. Patients infected with SARS-CoV experienced fever, difficulty breathing and low blood oxygen saturation levels [5,6]. Severe cases developed diffuse alveolar damage (DAD) and acute respiratory distress syndrome (ARDS) and disease severity was positively associated with increased age [7]. Host genetic background is also thought to influence disease severity but this understanding is complicated by inconsistent sample collection, varying treatment regimens and the limited scope of the SARS epidemic in humans [3,8,9]. Existing animal models of SARS-CoV infection have revealed that this lethal pulmonary infection causes a denuding bronchiolitis and severe pneumonia which oftentimes progresses to acute respiratory failure [10,11,12]. More recently, a second emerging coronavirus designated Middle East Respiratory Coronavirus (MERS-CoV) emerged from bat and camel populations [13,14,15], and has caused ~38% mortality. Given the complex interplay between environmental, viral and host genetic variation in driving viral disease severity, as well as the difficulty of studying those factors in episodic outbreaks of pathogens such as SARS-CoV, MERS-CoV and other highly virulent zoonotic pathogens that cross the species barrier at regular intervals, novel approaches are needed to understand and identify those factors contributing to these diseases.
Host genetics play a critical role in regulating microbial disease severity, evidenced by the identification of highly penetrant host susceptibility alleles within CCR5, FUT2, IL-28B in controlling HIV, norovirus and HCV infection and disease severity, respectively [16,17,18]. However, most microbial infections cause complex disease phenotypes that are regulated by the interactions of oligogenic traits with reduced penetrance, making them extremely difficult to identify and validate in human populations during outbreaks. Mannose binding lectin (MBL) polymorphisms were alternatively associated with successful recovery from SARS-CoV infection and a poor outcome of infection [19,20], reflecting the complexity of performing candidate gene or genome wide association studies with limited human samples. The generation of a mouse adapted strain of SARS-CoV, MA15, allowed for development of a small animal model that replicated both human lung disease and the age-dependency of SARS-CoV pathogenesis [10]. MA15 infection of inbred mice deficient in various immune genes has greatly contributed to our understanding of the host response to SARS-CoV infection [21,22]. However, such studies have focused on extreme abrogation of rationally selected candidate genes and have not evaluated the role of undescribed polymorphisms in genes in a model mimicking the genetic diversity seen in the human population. As a complement to human genome wide association studies, here we apply a new approach designed to dissect the identity and contributions of monogenic and oligogenic variants on multiple traits in complex disease outcomes following acute virus infection in a mouse model of human populations.
The Collaborative Cross (CC), a novel eight-way recombinant inbred (RI) mouse strain panel, has recently become available to the scientific community [23,24,25]. The power of the CC for genetic mapping is enhanced by availability of complete genome sequences of the founder strains and rich bioinformatics resources [26,27,28]. The eight founder strains used to generate the CC (A/J, C57BL/6J, 129S1/SvImJ, NOD/ShiLtJ, NZO/HILtJ, CAST/EiJ, PWK/PhJ and WSB/EiJ) are phenotypically diverse and capture single nucleotide polymorphisms (SNPs) and insertion/deletions (In/Dels) at approximately twice the frequency of common variants in human populations [24,29,30,31,32]. The derivation of CC strains from these multiple founders has proven to be useful for identifying polymorphisms that are responsible for a variety of traits [23]. The CC supports precise genetic mapping and, because the CC strains are genetically reproducible, it also serves as a robust validation platform and reference resource for integrative systems genetics applications.
Here, we studied incipient lines of the CC (the preCC) to identify host genes that contributed to SARS-CoV MA15 infection and pathogenesis. We identified four novel quantitative trail loci (QTLs) contributing to SARS-CoV pathogenesis. Within the HrS1 QTL, a combination of approaches applied to the CC platform predicted a single gene candidate, Trim55, as the principle regulator of vascular cuffing after infection. Vascular cuffing is a commonly reported phenotype observed in response to a variety of insults including chemical injury and infection ([33,34,35]; high levels of vascular cuffing have been observed in models of severely pathogenic SARS-CoV infection [21,22]. Fluid vascular cuffing has been reported to decrease lug compliance suggesting an important physiologic consequence of this response [36]. Using knockout mice, we confirmed the role of Trim55 in immune cell infiltration, demonstrating the utility of the CC platform for identifying single gene candidates that likely regulate novel immune functions in trans-endothelial migration and perivascular cuffing following virus infection.
Mice from the eight founder strains as well as 147 eight to twenty week old female preCC mice were infected with 105 plaque forming units (PFU) of mouse adapted SARS-CoV, designated MA15 [10], and weight loss was observed over the course of a four day infection. At day four post infection mice were euthanized and tissue collected for assessment of viral load in the lung as well as virus-induced inflammation and pathology. A wide range of susceptibilities to SARS-CoV infection was found among the founder strains of the CC and the overall heritability of weight changes following SARS-CoV infection determined to have a coefficient of genetic determination of 0.72. NOD/ShiLtJ mice were resistant to infection and gained weight over the course of the experiment (Figs 1A and S1A). A/J, C57BL/6J,129S1/SvImJ and NZO/HILtJ mice experienced moderate and transient weight loss as previously described [21,22] while CAST/EiJ, PWK/PhJ and WSB/EiJ mice demonstrated extreme susceptibility to SARS-CoV infection including substantial weight loss and death (Figs 1A, S1A and S1B). Subsequent dose response studies using the three highly susceptible wild-derived strains indicated an LD50 of between 100 and 500 PFU for CAST/EiJ mice, between 500 PFU and 1000 PFU for PWK/PhJ and between 103 and 105 PFU for WSB/EiJ mice (S1 Table). PreCC mice infected with SARS-CoV ranged from over 30% weight loss by day four post infection to over 10% weight gain (Fig 1A), exceeding the range of susceptibilities observed in the founder strains. Additionally, 26 preCC mice (18% of the preCC cohort) succumbed to infection prior to the day four harvest point indicating extreme susceptibility to SARS-CoV infection.
Viral load in the lung at day four post infection was determined for each surviving preCC mouse as well as for each of the founder strains. Viral lung titers showed a heritability of 0.60 as measured by the coefficient of genetic determination amongst the 7 surviving founder strains. Amongst the founder strains, PWK/PhJ mice had the lowest viral loads in the lungs, with 1.75x103 PFU per lung at day four post infection (Figs 1B and S1B). PWK/PhJ mice also showed significant weight loss and a low LD50 indicating that viral load was unlikely to be responsible for pathogenesis in these mice. In contrast, C57BL/6J mice had the highest amount of virus at 6.35x106 PFU per lung. Lung titers in the preCC mice ranged from below the limit of detection (100 PFU/lung) to over 108 PFU per lung, greatly exceeding the range of viral loads in the founder strains. Some preCC mice had viral loads in the lung below the 100 PFU limit of detection, despite having substantial weight loss. CAST/EiJ mice are extremely susceptible to SARS-CoV infection and do not survive until the day four post infection timepoint. Fig 1C shows the relationship between weight loss and lung titer at day four post infection. We found no correlation between viral load in the lung at day four post infection and weight loss (r = -0.014, p = 0.8938) when excluding those animals with viral loads below the limit of detection. When those animals were included in the analysis there is a significant, but not very explanatory correlation (r = -0.347, p = 0.00019) between the two phenotypes.
Multiple aspects of lung pathology were assessed in surviving preCC animals including disease and immune infiltrates in the airways, vasculature, alveoli and parenchyma and signs of DAD (S2 Table). A wide variety of lung pathologies were found across the preCC mice including denudation of airway epithelial cells, airway debris, eosinophilia, hyaline membrane formation and vascular cuffing (Fig 2A–2F). Quantification of the overall pathology score along with select data ranges are shown in S2 Fig. Hyaline membrane formation and pulmonary edema with accompanying inflammation in the alveoli was a hallmark of SARS-CoV infection in human cases and is also evident in aged mouse models of disease [11]. In contrast to young founder strain animals, robust hyaline membrane formation was observed in 13% of preCC mice at day four post-infection, demonstrating that improved animal models are one likely outcome of infection studies in the CC. Phenotypic correlations of varying strengths were observed between aspects of lung pathology, inflammation, viral load at day four post infection, as well as weight loss across the course of the study (Fig 3).
We genotyped 140 preCC animals at high density, including several that succumbed to infection prior to the scheduled day four harvest. As previously described [23,27], we conducted QTL mapping using Bagpipe (http://valdarlab.unc.edu/software.html) and the underlying eight founder strain haplotypes present in the CC to identify host genome regions containing polymorphisms significantly associated with SARS-induced disease responses. We identified four QTLs shown in Fig 4, HrS1-4 (Host response to SARS) that contributed to disease associated phenotypes at day four post infection. We identified a significant main effect QTL for vascular cuffing (Chr 3: 18286790–26668414), which explained 26% of the variation in vascular cuffing. We also identified two highly suggestive (genome-wide p-values based on permutations of 0.1>p>0.05) QTL for viral titer (Chr 16: 31583769–36719997) and eosinophil infiltration (Chr 15: 72103120–75803414), explaining 22% and 26% of the variation in these traits respectively. Finally, we also searched for modifier QTL, those QTL additively influencing a trait of interest, but whose presence was initially masked by our three other identified QTL. We identified a significant QTL further influencing vascular cuffing (Chr 13: 52822984–54946286), explaining an additional 21% of the variance in this phenotype. HrS4 was a moderate peak even without considering HrS1 status, suggesting that these interactions are additive. Table 1 details each of the SARS susceptibility QTLs including LOD and p-values. Analysis of other phenotypes did not lead to discovery of QTLs passing the p<0.01 significance threshold.
The genetic architecture of the preCC, with up to eight distinct haplotypes at each locus, provides unique opportunity for narrowing QTL regions to candidate genes or SNPs. To narrow QTL regions we estimated the additive allele effects associated with each haplotype and correlated these to the allelic states at candidate causative polymorphisms. Allele effects [23] describe the estimated effect of each of the eight founder haplotypes on the phenotype (e.g. a large positive allele effect for the PWK/PhJ haplotype suggests that having a PWK/PhJ allele will increase the phenotypic trait value of interest). In our analysis we focused on polymorphisms corresponding to the largest contrast between allele effects at the peak QTL locus. For HrS1 we identified two haplotypes, C57BL/6J and WSB/EiJ increasing vascular cuffing relative to the haplotypes of the other six founder strains. For each of HrS2-4, we identified a single founder haplotype altering the phenotype relative to the seven other founder haplotypes (HrS2: PWK/PhJ haplotype reduced viral titer; HrS3: A/J haplotype increasing eosinophillic infiltration; HrS4: CAST/EiJ haplotype reduced vascular cuffing).
We then used high coverage whole genome sequence from the eight founder strains [37] to identify either private SNPs or small In/Dels in the case of a single causative haplotype, or regions of shared descent (in the case of two causative haplotypes) to narrow down the large QTL regions. HrS1 was initially an 8.38 Mb region which contained 26 genes and 9 non-coding RNAs (ncRNAs). Identification of the sub-regions where C57BL/6J and WSB/EiJ share private, common ancestry reduced this region to 449 kb, which contained only one gene, one pseudogene and one miRNA of unknown function (Trim55, GM7488 and AC107456.1, respectively). Allele effects for all four QTLs can be seen in S3 Fig.
The HrS2 QTL on chromosome 16 was a 5.4 Mb region containing 92 genes and 30 ncRNAs. Across the eight founder strains, there were 95,936 SNPs or small In/Dels, and 33,288 of these were private to PWK/PhJ. Seven ncRNAs and 74 genes had private PWK/PhJ SNPs or In/Dels (S3 Table). We further prioritized these variants based on whether the PWK/PhJ private polymorphisms were likely to cause major functional changes to the gene (missense, nonsense, stop gained/lost, splice alterations or nonsense mediated decay). When we did so, we further reduced this list to 48 candidate genes including several mucins as well as genes involved in T cell activation and apoptosis.
The HrS3 QTL on chromosome 15 was a 3.7 Mb region containing six ncRNAs and 63 genes. There were a total of 71,208 SNPs or small In/Dels in the region, 932 of which were private to A/J. No ncRNAs and only 25 genes contained a private A/J SNP or In/Del, and we further reduced these to one candidate gene with major functional changes (S4 Table), Bai1. Bai1 is a high priority candidate gene given the association between eosinophils and angiogenesis [38]; however we have not chosen to focus on Bai1 at this time because of the limited availability of tools for working on an A/J genetic background.
Finally, HrS4 on chromosome 13 was a 2.12 Mb region containing three ncRNAs and 30 genes. There were a total of 461,46 SNPs or In/Dels in the region, 9,732 being private to CAST/EiJ. 29 of the genes and all three ncRNAs contained private CAST/EiJ polymorphisms (S5 Table). When we further prioritized based on major functional changes, we reduced the region to only one ncRNA and nine genes including Cdhr2, a member of the protocadherin family [39].
We focused our validation efforts on Trim55, the single HrS1 candidate and a member of the TRIM protein superfamily which has not previously been associated with any infectious disease phenotype. Although many TRIM proteins function in innate immunity and inflammation, Trim55 (also known as muscle-specific RING finger 2 or Murf2) has only been studied in the context of muscle development and cardiac function [40,41]. Trim55 is expressed in smooth muscle surrounding blood vessels [42], an appropriate location to influence perivascular cuffing phenotypes. Knockout mice on a C57BL/6J background have previously been reported [43] and were kindly made available to our laboratory. Groups of age matched Trim55-/- and C57BL/6J control mice were infected with 105 PFU of MA15 for four days and monitored daily for weight loss and signs of disease. Trim55-/- and C57BL/6J animals had similar weight loss profiles as well as similar viral loads in the lung at day four post infection (Fig 5A and 5B) and no differences in mortality. Hematoxylin and eosin stained lung sections showed significantly reduced vascular cuffing in the lungs of Trim55-/- (mean score of 0.69) compared to control animals (mean score of 1.15) (p < 0.05 by students t test, Fig 5C), confirming the role of Trim55 in contributing to SARS-CoV-induced vascular cuffing. Additional mice were infected for flow cytometric analysis of inflammatory cell populations in the lung after MA15 infection. While we observed a general trend towards increased numbers of T cells, B cells and macrophages in the lungs of C57BL/6J control mice compared to the Trim55-/- mice, only monocyte numbers were significantly different between the two groups. Total monocytes, as well as the subset of Ly6C positive monocytes, were present in significantly higher numbers in the lungs of infected control mice compared to Trim55-/- mice (Fig 5D).
RNA was isolated from the lungs of mock and infected control and Trim55-/- mice at two and days four post infection. 168 genes were identified as differentially expressed (DE, log2 fold change >2 relative to mock) between the two strains, predominantly at day two post infection (GEO accession GSE64660). We then used Ingenuity Pathway Analysis software to identify functionally enriched gene categories. This analysis identified the granulocyte and agranulocyte diapedesis gene ontology categories as among the most significantly enriched (first and third respectively) from genes with DE between Trim55-/- and B6 controls (Fig 6A). Diapedesis, or extravasation, is the process by which inflammatory monocytes and leukocytes bind to endothelial cells and migrate from the blood stream into surrounding injured tissues. The transcriptional analysis indicates decreased expression of tight junction genes and increased chemokine expression in C57BL/6J mice compared to that observed in Trim55-/- mice. Relative expression of genes involved in granulocyte adhesion and diapedesis at days two and four post infection is shown in Fig 6B and 6C.
Emerging coronaviruses like SARS-CoV and MERS-CoV cause high morbidity and mortality in human populations. Because of limited access to clear human disease responses and samples from acute infections, as well as the limited number of overall infected individuals, it is extremely challenging to define the role of host genetic polymorphism in human disease. Coronavirus pathogenesis is heavily influenced by host genetics, as evidence by the extreme resistance of SJL mice, which encode a defective variant CEACAM1 receptor for mouse hepatitis virus entry and infection [44]. Furthermore, genetic monomorphisms in the cheetah have resulted in extreme hypersensitivity to feline infectious peritonitis coronavirus infection, underscoring the importance of abundant genetic variation in controlling lethal coronavirus infection [45,46]. In this study we examined numerous phenotypes following SARS-CoV infection and identified 4 QTL related to various aspects of SARS-CoV pathogenesis. These data support previous predictions that the CC platform can identify genetic variants contributing moderate effect sizes (e.g. ~20%) to complex immune response traits.
Two of the four identified QTL, on chromosome 3 and 13 respectively, pertained to perivascular cuffing. Perivascular cuffing in the lung is frequently observed during microbial and non-microbial lung disease [34,47,48] and is associated in part with extravasation, the process by which inflammatory cells migrate from the blood to surrounding tissues [49,50]. Previous reports of perivascular cuffing include lymphocyte and granulocyte involvement with limited insights into the genetic underpinnings of this phenotype. In vivo models of SARS-CoV infection have shown that vascular cuffing increases in cases of severe disease [21,22,35] and vascular congestion was observed in human SARS-CoV patients [7]. A recent study of pneumococcal infection [51] identified several QTL governing disease susceptibility including one on chromosome 13. The authors also found an association between perivascular inflammation and susceptibility to infection but did not extend their genetic analysis to that phenotype; there was no overlap between their chromosome 13 QTL and Hrs4. Analysis of pulmonary inflammation following hyperoxia-induced lung injury [52] identified QTL on chromosomes 1, 2, 4, 6 and 7 and informative SNPS helped to identify Chrm2 as the causative gene on chromosome 6. In this study we identified QTL contributing to 26% and 21% of the total vascular cuffing phenotypic variance, respectively. The limited numbers of candidate genes under the larger effect size QTL allowed us to test and validate the role of Trim55 in SARS-CoV-induced perivascular cuffing phenotype.
The CC was conceived to expand upon the genetic variation and mapping precision found within classical recombinant inbred (RI) panels which often suffer from inability to narrow the numbers of candidate genes due to the close genetic relationship of the founding lines. The classical BxD panel—derived from C57BL/6J and DBA2/J founder strains–was used previously to identify Klra8, the resistance gene to mouse cytomegalovirus (Cmv–1) infection [53]. Importantly, the validation experiments were conducted over a decade after the initial identification of the Cmv–1 susceptibility locus [54] as the wide initial QTL interval was not sufficient for identification of specific candidate genes. The Collaborative Cross provides a significant advantage in comparison to two-way crosses and other bi-allelic RI strain panels–as illustrated by our study, allele effects associated with founder haplotypes can provide a substantial reduction in the list of plausible candidate loci. Moreover, the inclusion of a diverse set of founder strains increases the likelihood of variants existing at loci that can influence any given trait. Indeed, in our study five of the eight founder strains contributed minor, causative alleles to the four QTL we identified. As the breeding of the CC lines preserved genetic variation across the genome, the CC lacks genetic blind spots and has multiple variant alleles at each locus. With a wide range of phenotypes [23], the CC recapitulates aspects of the genetic diversity of the human population, making it a powerful system for use in causal genetic analyses.
This study was part of an early pilot project to demonstrate the utility of the CC panel [23]. As such we did not have access to fully inbred animals and were limited to a single animal per genotype. However, the increased control of the experimental conditions in these studies and high frequency of minor alleles within the CC population (each allele is present in roughly 12.5% of CC genomes [23] whereas minor allele frequencies in the human population are typically much lower) allowed us to identify multiple host genome regions contributing to differential SARS-CoV infection. Studies utilizing the full CC panel will be able to use the full potential of a reproducible genetic background to obtain repeated assays and high-precision phenotyping, even our limited proof-of-concept study proved to be adequate to identify multiple host genome regions contributing to differential responses to SARS-CoV infection.
Trim55 is part of the well-known superfamily of TRIM proteins, specifically in the C-II subfamily. This subfamily consists of Trim54, Trim55 and Trim63, and is defined by an N-terminal that contains a Ring Finger domain, B-box 2 domain and a coiled-coil domain [42]. The C-II Trim family genes are solely expressed by muscle cells and to date have only been studied in the frame of muscle cell development and cardiac function. Trim55 and Trim63, also known as Murf1, mediate muscle cell protein turnover through their E3 ubiquitin-ligase activities and function in muscle wasting phenotypes [40,43,55]. Trim55 specifically functions in myosin and myofibril maintenance and knockdown studies correlate Trim55 levels with modified post-translational microtubule modifications and defects in myofibril assembly, critical components in extravasation [55].
Blood vessels are comprised of vascular endothelial cells, connective tissue and smooth muscle cells, all of which must be crossed during inflammatory cell trafficking into the lung. During extravasation, inflammatory cells tumble and bind to adhesion molecules, slowing their motion and expanding surface-surface interactions with endothelial cells [56]. Tissue Necrosis Factor-alpha and thrombin expression levels increase following SARS-CoV infection [12,21] and these proteins have both been shown to increase endothelial permeability [57,58]. Here we observed a complicated picture of altered chemokine and tight junction gene expression in the absence of Trim55 (Fig 6B and 6C). Increased expression of Ccl24, CCR3, IL4 and Pdgfc in C57BL/6J mice compared to that in Trim55-/- mice at day four post infection correlates with increased inflammatory cell recruitment and binding to extracellular matrix proteins. These expression changes are consistent with changes in altered recruitment of inflammatory cells to the lung following SARS-CoV infection. Higher expression of Claudin19 at day two post infection in Trim55 deficient mice likely contributes to decreased tight junction permeability and reduced ability for inflammatory cells in the bloodstream to cross the endothelial barrier. Additionally, one of the high priority candidate genes under the modifier QTL on chromosome 13 is Cdhr2, a cadherin superfamily member that may also play a role in extravasation of inflammatory cells into the infected lung. Different specific VE-cadherin residues are known to regulate leukocyte extravasation and vascular permeability [59], demonstrating the importance of cadherin family members in these processes. More recent work details the role of Cdhr2 in intestinal brush border assembly via adhesion links between adjacent microvilli [60].
Intravascular crawling and signaling through RhoA induces actin, microfilament and microtubule reorganizations and the production of endothelial cell docking structures, which surround the inflammatory cell and span tight junctions [56]. Although controversial, myofibril contractile structures may also contribute in to the assembly of these structures. In any event, inflammatory cell transmigration requires the formation of actin-myosin II contractile structures which are attached to tight junction membranes by VE-cadherins, resulting in increased endothelial tension, and programmed separation and expansion of the tight junctions which allow for leukocyte/monocyte passage into the surrounding tissues [61]. It seems likely that Trim55, with its roles in myosin and myofibril maintenance and microtubule organization, contributes to the programmed formation of endothelial docking structures and regulation of inflammatory cell transmigration; key features associated with the formation of perivascular cuffs around vessels in the lung. Our data (Figs 5 and 6) demonstrate that Trim55 contributes to vascular cuffing following SARS-CoV infection. While the mechanism is not yet fully understood, the data strongly suggest that Trim55 is important for extravasation of inflammatory cells, and thus overall SARS-CoV pathogenesis, by altering intercellular junctions and chemotactic signals. Increased studies of Trim55 and Cdhr2 function within the CC population, either via specific crosses of lines with high and low alleles at the HrS1 and HrS4 loci, or via CRISPR-Cas9 modification of these loci will allow further insight into the role that these two genes play during SARS-CoV pathogenesis and recovery, as well as increasing understanding of the more general process of extravasation.
The Collaborative Cross was conceived of as a resource to drive insight into a variety of biomedically important diseases via the reassortment of genetic variants and expansion of phenotypic ranges [62]. Indeed, previous studies with various preCC subsets have demonstrated expanded phenotypes in preCC mice body weight and hematological parameters [23,63,64], response to Aspergillus [65] and susceptibility to Influenza A infection [27,66]. More recently it has been shown that novel combinations of alleles have also resulted in new models for human disease such a spontaneous colitis [67], and that F1 hybrids of CC mice were used to create an improved mouse model for Ebola virus disease [68] including hemorrhagic signs of disease previously not observed in a small animal model. Within our study of SARS-CoV infection within the preCC, we showed more extreme disease phenotypes than those seen within the eight founder strains of the CC. These disease phenotypes included virus titer, weight loss, pathology and lethality. Further, we saw the emergence of new phenotypes including ARDS and DAD not traditionally seen within young inbred strains [11]. Importantly, our results highlight another exciting aspect of the nature of CC genome: transgressive segregation, or the release of cryptic genetic variation [69,70]. As the three wild-derived CC founders all showed mortality early in the course of SARS-CoV infection, genetic variants within these three strains impacting later-stage SARS-CoV responses would normally not be seen. Only via the reassortment of these alleles into a variety of genetic backgrounds (some resistant to clinical disease, some susceptible) were we able to show that alleles from all three wild-derived founders impacted perivascular cuffing or viral titer levels independent of their effects on clinical disease or SARS-CoV mortality. Collectively, these data support the hypothesis that the CC population represents a robust platform for developing improved animal models that more readily replicate disease phenotypes seen in human populations. All told, our identification of multiple QTL related to SARS-CoV pathogenesis, identification of a novel function for Trim55, and the development of new models of acute lung injury, further solidify the utility of the CC as a valuable community resource for research of infectious diseases and other biological systems driven by complex host response networks.
Mouse studies were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All mouse studies were performed at the University of North Carolina (Animal Welfare Assurance #A3410-01) using protocols approved by the UNC Institutional Animal Care and Use Committee (IACUC).
Recombinant mouse-adapted SARS-CoV (MA15) was propagated and titered on Vero E6 cells. For virus titration half of the right lung was used to assess plaque forming units (PFU) per lung using Vero E6 cells with a detection limit of 100 PFU [71]. All experiments were performed in a Class II biological safety cabinet in a certified biosafety level 3 laboratory containing redundant exhaust fans while wearing personnel protective equipment including Tyvek suits, hoods, and HEPA-filtered powered air-purifying respirators (PAPRs).
8–12 week old female animals from the 8 founder strains (A/J, C57BL/6J, 129S1/SvImJ, NOD/ShiLtJ, NZO/HILtJ, CAST/EiJ, PWK/PhJ, and WSB/EiJ) were obtained from the Jackson labs (jax.org), and bred at UNC Chapel Hill under specific pathogen free conditions. 8–20 week old female pre-CC mice were bred at Oak Ridge National Laboratories under specific pathogen free conditions, and transferred directly into a BSL–3 containment laboratory at UNC Chapel Hill. One preCC mouse per line was infected, for the founder strains at day four n = 2 (A/J), n = 3 (C57BL/6J, 128S1/SvImJ, NOD/ShiLtJ, CAST/EiJ, PWK/PhJ and WSB/EiJ) and n = 5 (NZO/HILtJ). Trim55-/- (Murf2-/-) mice on a C57BL/6 background were a kind gift from Christian Witt at the University of Mannheim. Validation experiments used 8–12 week old female mice. All experiments were approved by the UNC Chapel Hill Institutional Animal Care and Use Committee. Animals were maintained in SealSafe ventilated caging system in a BSL3 laboratory, equipped with redundant fans as previously described by our group.
Animals were lightly anesthetized via inhalation with Isoflurane (Piramal). Following anesthesia, animals were infected intranasally with 105 pfu of mouse adapted SARS-CoV (MA15) in 50 μL of phosphate buffered saline (PBS, Gibco), while mock infected animals received only 50 μL of PBS. Animals were weighed daily and at four days post infection, animals were euthanized via Isoflurane overdose and tissues were taken for various assays. No blinding was used in any animal experiments and animals were not randomized; group sample size was chosen based on availability of age-matched mice. Pearson’s correlation was used to determine any correlation between weight loss and log-transformed viral load in the lung.
The left lung was removed and submerged in 10% buffered formalin (Fisher) without inflation for 1 week. Tissues were embedded in paraffin, and 5 μm sections were prepared by the UNC Lineberger Comprehensive Cancer Center histopathology core facility. To determine the extent of inflammation, sections were stained with hematoxylin and eosin (H & E) and scored in a blinded manner by for a variety of metrics relating to the extent and severity of immune cell infiltration and pathological damage on a 0–3 (none, mild, moderate, severe) scale. Significant differences in lung pathology were determined by a two-sample student’s t test. Images were captured using an Olympus BX41 microscope with an Olympus DP71 camera.
The right lung of each mouse was used for flow cytometric staining of inflammatory cells. Mice were perfused with PBS through the right ventricle before harvest, lung tissue was dissected and digested in RPMI (Gibco) supplemented with DNAse and Collagenase (Roche). Samples were strained using a 70 micron filter (BD) and any residual red blood cells were lysed using ACK lysis buffer. The resulting single cell suspension was stained with two antibody panels using the following stains (1) FITC anti-Ly-6C clone AL21 (BD), PE anti-SigLecF clone E50-2440 (BD), PETR anti-CD11c clone N418 (MP), PerCP anti-B220 clone RA3-6B2 (MP), PE-Cy7 anti-Gr–1 clone RB6-8C5 (eBio), eF450 anti-CD11b clone M1/70 (eBio), APC anti-LCA clone 30-F11 (eBio), APC-eF780 anti-MHC class II clone M5/114 (eBio) or (2) FITC anti-CD94 clone 18d3 (eBio), PE anti-CD3Ɛ clone 145-2C11 (eBio), PETR anti-CD4 clone RM4-5 (MP), PerCP anti-CD8 clone 53–6.7 (BD), PE-Cy7 anti-CD49b clone DX5 (eBio), eF450 anti-LCA clone 30-F11 (eBio), AF647 anti-CD19 clone 6D5 (Biolegend), APC-eF780 anti-B220 clone RA3-6B2 (eBio). While this FACS analysis was solely performed on mice of a C57BL6/J background, these antibodies have all been shown recognize the relevant antigens in each of the CC founder lines. Samples were run on a Beckman Coulter CyAN, and data analyzed within the Summit software. Significant differences in lung inflammatory cell populations were determined by a two-sample student’s t test.
Genotyping and haplotype reconstruction were done as described in [23]. Briefly, each pre-CC animal was genotyped using the Mouse Diversity Array [72] (Affymetrix) at 372,249 well performing SNPs which were polymorphic across the founder strains [31]. Once genotypes were determined, founder strain haplotype probabilities were computed for all genotyped loci using the HAPPY algorithm [73]. Genetic map positions were based on the integrated mouse genetic map using mouse genome build 37 [74].
Linkage mapping was done as described in [23]. Briefly, QTL mapping was conducted using the BAGPIPE package [75] to regress each phenotype on the computed haplotypes in the interval between adjacent genotype markers, producing a LOD score in each interval to evaluate significance. Genome-wide significance was determined by permutation test, with 250 permutations conducted per scan. Phenotype data for mapping either satisfied the assumptions of normality or were log transformed to fit normality (titer data). For the likely regions of identified QTL peaks, SNP data for the eight founder strains from the Sanger Institute Mouse Genomes Project [37] were downloaded and analyzed as described in Ferris et al [27].
Bagpipe is freely available at http://valdarlab.unc.edu/software.html.
At two and four days after infection, mice were euthanized and a lung portion placed in RNAlater (Applied Biosystems/Ambion) and then stored at −80°. The tissues were subsequently homogenized in TriZol (Life Technologies), and RNA extracted as previously described [12]. RNA samples were spectroscopically verified for purity, and the quality of the intact RNA was assessed using an Agilent 2100 Bioanalyzer. cRNA probes were generated from each sample by the use of an Agilent one-color Quick-Amp labeling kit. Each cRNA sample was then hybridized to Agilent mouse whole-genome oligonucleotide microarrays (4 x 44) based on the manufacturer’s instructions. Slides were scanned with an Agilent DNA microarray scanner, and the output images were then analyzed using Agilent Feature Extractor software. Microarray data has been deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus database and is accessible through GEO accession GSE64660.
Raw Agilent Microarray files were feature extracted Agilent feature extractor version 10.7.3.1. Raw Microarray files were background corrected using the “norm-exp” method with an offset of 1 and quantile normalized using Agi4x44PreProcess [76] in the R statistical software environment. Replicate probes were mean summarized, and all probes were required to pass Agilent QC flags for 75% replicates of at least one infected time point (41,267 probes passed). This microarray analysis was performed only on animals with a C56BL6/J background; thus it was not necessary to correct for probes with SNPs caused by the genetic variation of the other founder lines.
Differential expression was determined by comparing MA15 infected samples (C57Bl6/J mice vs. Trim55-/- mice) with mock and each other to fit a linear model for each probe using the R package Limma. Criteria for differential expression was an absolute log2 FC of 1 and a q value of < 0.05 calculated using a moderated t test with subsequent Benjamini-Hochberg correction. Differentially expressed (DE) genes were observed for both C57BL/6J and Trim55-/- infected mice compared to time matched mocks at two and day four post infection. DE analysis was also run on the Trim55-/- infected mice against the C57BL/6J infected mice which provided a direct observation of the transcription signatures in the Trim55-/- against the MA15 infected mouse background. To identify genes with similar patterns of variation at early and late times post infection, day two and day four gene signatures were intersected separately and then combined. There was no intersection of DE genes between the day two and four time points when the Trim55-/- infected mice were run against the C57BL/6J infected mice.
Functional analysis of statistically significant gene expression changes was performed using the Ingenuity Pathways Knowledge Base (IPA; Ingenuity Systems) [77]. Functional enrichment scores were calculated in IPA using all probes that passed our QC filter as the background data set.
|
10.1371/journal.pmed.1002299 | Comparison of artemether-lumefantrine and chloroquine with and without primaquine for the treatment of Plasmodium vivax infection in Ethiopia: A randomized controlled trial | Recent efforts in malaria control have resulted in great gains in reducing the burden of Plasmodium falciparum, but P. vivax has been more refractory. Its ability to form dormant liver stages confounds control and elimination efforts. To compare the efficacy and safety of primaquine regimens for radical cure, we undertook a randomized controlled trial in Ethiopia.
Patients with normal glucose-6-phosphate dehydrogenase status with symptomatic P. vivax mono-infection were enrolled and randomly assigned to receive either chloroquine (CQ) or artemether-lumefantrine (AL), alone or in combination with 14 d of semi-supervised primaquine (PQ) (3.5 mg/kg total). A total of 398 patients (n = 104 in the CQ arm, n = 100 in the AL arm, n = 102 in the CQ+PQ arm, and n = 92 in the AL+PQ arm) were followed for 1 y, and recurrent episodes were treated with the same treatment allocated at enrolment. The primary endpoints were the risk of P. vivax recurrence at day 28 and at day 42.
The risk of recurrent P. vivax infection at day 28 was 4.0% (95% CI 1.5%–10.4%) after CQ treatment and 0% (95% CI 0%–4.0%) after CQ+PQ. The corresponding risks were 12.0% (95% CI 6.8%–20.6%) following AL alone and 2.3% (95% CI 0.6%–9.0%) following AL+PQ. On day 42, the risk was 18.7% (95% CI 12.2%–28.0%) after CQ, 1.2% (95% CI 0.2%–8.0%) after CQ+PQ, 29.9% (95% CI 21.6%–40.5%) after AL, and 5.9% (95% CI 2.4%–13.5%) after AL+PQ (overall p < 0.001). In those not prescribed PQ, the risk of recurrence by day 42 appeared greater following AL treatment than CQ treatment (HR = 1.8 [95% CI 1.0–3.2]; p = 0.059). At the end of follow-up, the incidence rate of P. vivax was 2.2 episodes/person-year for patients treated with CQ compared to 0.4 for patients treated with CQ+PQ (rate ratio: 5.1 [95% CI 2.9–9.1]; p < 0.001) and 2.3 episodes/person-year for AL compared to 0.5 for AL+PQ (rate ratio: 6.4 [95% CI 3.6–11.3]; p < 0.001). There was no difference in the occurrence of adverse events between treatment arms.
The main limitations of the study were the early termination of the trial and the omission of haemoglobin measurement after day 42, resulting in an inability to estimate the cumulative risk of anaemia.
Despite evidence of CQ-resistant P. vivax, the risk of recurrence in this study was greater following treatment with AL unless it was combined with a supervised course of PQ. PQ combined with either CQ or AL was well tolerated and reduced recurrence of vivax malaria by 5-fold at 1 y.
ClinicalTrials.gov NCT01680406
| In areas where Plasmodium vivax is endemic, recurrent parasitaemia arises from chloroquine (CQ) resistance and the reactivation of dormant hypnozoites. Recurrent parasitaemia is associated with significant morbidity and mortality. Providing safe and effective radical cure of blood and liver stages of the parasite is critical for the control and elimination of P. vivax malaria.
Two systematic reviews of P. vivax clinical trials highlighted a declining efficacy for CQ in many P. vivax endemic areas and marked heterogeneity in study design for determining primaquine (PQ) efficacy. The conclusions of these reviews were that clinical trials to determine the efficacy of radical cure should follow up patients for a prolonged period to capture late relapses and should quantify the incidence rate of multiple recurrences, not just the time to the first recurrence.
The rise in CQ-resistant P. vivax provides a good rationale for artemisinin combination therapy for P. vivax malaria; however, the efficacy of artemisinin combination therapies in combination with PQ radical cure is poorly documented.
We randomized patients to four treatment arms: one group received CQ only (the current recommended treatment for P. vivax malaria in Ethiopia), the second group received CQ plus PQ, the third group received artemether-lumefantrine (AL) alone, and the fourth group received AL plus PQ. All patients were followed up for a year and were treated with the same treatment for every P. vivax malaria episode.
We quantified the risk of P. vivax infections at day 28 after treatment and also over 12 months. The risk of recurrence by day 28 and 42 was greater following AL than CQ. The addition of PQ to either CQ or AL reduced the risk of recurrence 3-fold by day 42, and 2- to 3-fold over one year.
Patients treated with PQ had on average only 0.5 P. vivax malaria episodes per year, whereas patients not treated with PQ had on average two episodes per year. The efficacy of PQ treatment for recurrences, which was unsupervised, was 3- to 4-fold lower than that of the initial treatment, which was semi-supervised.
In Ethiopia there is evidence of CQ resistance; nevertheless, in this study CQ monotherapy had greater efficacy than AL therapy at day 42.
The addition of PQ radical cure to either CQ or AL provided major benefits in reducing subsequent recurrent infection.
PQ radical cure should be included in the treatment schedule in Ethiopia and other areas with high relapse risk to reduce relapsing infection and transmission, but further work is needed to improve adherence to the current 14-day regimen to ensure maximum public health impact.
| Almost 3 billion people live at risk of Plasmodium vivax infection [1,2], with over 100 million clinical malaria cases estimated to occur each year [3]. The greatest burden of P. vivax malaria is in the Asia-Pacific region and South America, whereas on the African continent, P. vivax infection is limited mostly to the Horn of Africa. Recent intensification of malaria control efforts has reduced the global burden of P. falciparum malaria, but P. vivax has been more refractory. Vivax malaria is more difficult to cure than falciparum malaria, due to its ability to form dormant liver stages (hypnozoites) that reactivate periodically, causing relapsing infections and onward transmission. Chloroquine (CQ) remains the mainstay of treatment for vivax malaria in most endemic countries, but drug resistance has emerged in South-East Asia and is spreading [4]. Relapsing and increasingly frequent recrudescent infections cause repeated symptomatic illnesses, worsening the risk of anaemia and severe and fatal disease [5,6].
Although P. vivax is rare on the African continent outside the Horn of Africa, in Ethiopia it is responsible for approximately 40% of all clinical malaria [7–9]. National treatment guidelines recommend CQ as first-line treatment for uncomplicated P. vivax malaria. Artemether-lumefantrine (AL) is used widely for mixed-species infections of P. falciparum and P. vivax, and for cases of clinical malaria where diagnostics to determine the Plasmodium species are unavailable. Both CQ and AL have schizonticidal efficacy but lack activity against the liver stages. Primaquine (PQ), an 8-aminoquinoline, is the only currently available hypnozoiticide, but can cause severe haemolysis in glucose-6-phosphate-dehydrogenase (G6PD)–deficient patients; for this reason, programmes are often reluctant to recommend PQ, and healthcare providers are hesitant to prescribe it. Ethiopia has embarked on an ambitious malaria control programme, supporting the country’s Health Sector Development Plan as well as the national child survival strategy. The control and elimination of malaria will require deployment of a safe and effective radical cure of P. vivax. At the time this study was conducted, PQ radical cure was not part of the national antimalarial guidelines for patients living in malaria endemic areas of the country. To compare suitable treatment strategies, we undertook a randomized controlled trial of the efficacy of PQ in combination with CQ or AL at two sites in Oromia Region, Ethiopia.
Ethical approval for the study was granted by the National Research Ethics Review Committee in Ethiopia (3.10/801/05), the Human Research Ethics Committee of the Northern Territory Department of Health and Menzies School of Health Research in Australia (HREC 2013–1938), the US Centers for Disease Control and Prevention Institutional Review Board B (6338.0), and the Columbia University Institutional Review Board (AAAK4706).
The study was designed as an open-label randomized controlled trial with four arms, conducted at the Bishoftu Malaria Control Center and the Batu Health Center in Oromia Region, in central Ethiopia (S1 Text). Bishoftu is approximately 50 km south of Addis Ababa, and Batu a further 115 km south. The main transmission season is between September and November, and healthcare facility data confirmed P. vivax to be the dominant malaria species. Anopheles arabiensis is the primary malaria vector, and the prevalence of malaria in Oromia Region was 0.4% during the 2011 national malaria indicator survey [10]. Patients seeking care with suspected malaria were screened for the following inclusion criteria: slide-confirmed mono-infection with P. vivax, age > 1 y, weight ≥ 5.0 kg, living within 20 km of the enrolling health facility, and axillary temperature ≥ 37.5°C or history of fever during the previous 48 h. Patients were excluded if they were pregnant or breastfeeding, had danger signs or severe manifestations of malaria, signs of severe malnutrition, slide-confirmed infection with any other Plasmodium species besides P. vivax, acute anaemia (haemoglobin [Hb] < 80 g/l) or history of haemolysis, known hypersensitivity to any of the study drugs, other significant comorbidities, or regular medication that could interfere with their antimalarial treatment. All participants or their guardian/caregiver agreed to finger prick sampling and provided written informed consent.
Patients were randomized in a two-stage process: (i) at enrolment, randomization to either CQ or AL schizonticidal treatment and (ii) on day 2, after G6PD testing, randomization into the PQ or no-PQ arms. Patients with severe or intermediate G6PD deficiency were excluded from the second randomization to PQ. The randomization sequence was computer-generated by one of the investigators and kept in sealed opaque envelopes. Randomization was done in blocks of eight for each site separately. Clinicians enrolled the patients, and nurses assigned the patients sequentially according to the sealed envelopes.
The primary endpoint was the cumulative risk of P. vivax recurrence at day 28 and day 42 following treatment of the first episode of malaria, comparing AL with AL+PQ and CQ with CQ+PQ. A secondary endpoint was the cumulative risk at day 28 for AL compared with CQ and for AL+PQ compared with CQ+PQ. Other secondary endpoints included parasite and fever clearance, and cumulative risk and incidence rate of recurrences at the end of the study. Safety endpoints were the proportion of patients with adverse events (AEs) and serious AEs, the fractional change in Hb, the proportion of patients with a >25% drop in Hb between baseline and day 7, and the proportion of patients with anaemia (Hb < 100 g/l) on days 3 and 7. Haematological recovery at day 28 was defined as an Hb value on day 28 above baseline.
The sample size was based on a power calculation assuming an efficacy of 68% at day 42 after CQ monotherapy and that the addition of PQ would increase this to 87%. A total of 97 in each group would achieve 90% power to detect this difference between groups. Adjusting the alpha level to 0.025 for multiple comparisons and estimating 15% loss to follow-up increased the proposed sample size to 120 per treatment arm.
Data were double-entered into a Microsoft Access database, and analyses conducted using STATA 14 (StataCorp, College Station, TX, US), according to an a priori statistical analysis plan (S3 Text). In view of the potential interaction between PQ and CQ [15] and between PQ and AL [16], comparisons between each of the arms were undertaken separately, rather than using a 2 × 2 factorial design approach, which assumes no interaction between the interventions. The cumulative risks were calculated by survival analyses (Kaplan—Meier) at day 28, day 42, and the end of the follow-up, and treatment groups compared using a Cox regression model. The proportional hazards assumption was assessed by visually comparing the log(cumulative hazard) by time of follow-up curves for each co-variable category and subsequently by fitting and comparing models with and without time of follow-up interaction terms. Genotyping results were used to calculate adjusted cumulative risk of recurrence by days 28 and 42 by censoring for heterologous infections, which may represent reinfection or relapse; for this analysis, only PCR-confirmed P. vivax infections at enrolment were included (S2 Text) [17]. The risks of recurrence after primary and secondary treatments were compared at 6 mo to ensure sufficient patient follow-up time. Incidence rates were calculated from microscopy results by dividing the number of P. vivax episodes by the number of person-years of observation. Incidence rate ratios between treatment arms were derived from a negative binomial regression model. The primary analysis was by intention to treat (ITT), and in the secondary analysis a modified ITT approach was used, in which patients receiving incorrect treatment or in whom no treatment information was available were censored on the day of the recurrence. Patients with intermediate or deficient G6PD status were not included in the second stage of randomization; they were included only in descriptive results and were excluded from any relevant comparisons between arms.
The study was conducted between 8 November 2012 and 31 December 2014, during which a total of 1,177 patients were screened, and 398 (33.8%) enrolled in the study (Fig 1). In view of logistical constraints and the lack of further patients with malaria, the study was terminated at the end of the malaria season in 2014, after 82.9% (398/480) of the target sample size had been recruited.
The majority of patients (77.1%, 307/398) were enrolled at the Bishoftu Malaria Control Center. There were no important differences in baseline characteristics between the four treatment arms (Table 1) or in the proportion of patients lost to follow-up by day 42. G6PD deficiency was measured for 393 (98.7%) patients; missing values are mainly due to loss to follow-up before day 2. Nine (2.3%, 9/393) patients had intermediate G6PD results; they were randomized on day 0 for schizonticidal treatment, but did not receive PQ treatment (Table 1). In total, 98.2% (391/398) of patients completed a full course of schizonticidal treatment, of whom six (1.5%) vomited their dose within 1 h of administration (four in the CQ arm and two in the AL arm). One patient vomited doses of AL on day 1 and 2.
There were 17 recurrent P. vivax infections documented within 28 d of follow-up, and a further 34 between day 28 and 42 (27 in the AL arm, 18 in the CQ arm, five in the AL+PQ arm, and one in the CQ+PQ arm) (Table 2). An additional five patients presented with a P. falciparum infection before day 28, and one more between day 28 and 42. Patients in treatment arms that included PQ had significantly fewer recurrent malaria episodes than patients on schizonticidal therapy alone. By day 28, the cumulative risk for P. vivax recurrence was 4.0% (95% CI 1.5%–10.4%) for patients treated with CQ alone compared to 0% (95% CI 0%–4.0%) for those treated with CQ+PQ (p < 0.001). The corresponding risks were 12.0% (95% CI 6.8%–20.6%) following AL alone and 2.3% (95% CI 0.6%–9.0%) following AL+PQ (hazard ratio [HR] = 5.1 [95% CI 1.1–23.5], p = 0.034) (Fig 2; Tables 2 and 3).
By day 42, the risk of recurrence had risen to 18.7% (95% CI 12.2%–28.0%) in the CQ arm and 1.2% (95% CI 0.2%–8.0%) in the CQ+PQ arm (HR = 18.5 [95% CI 2.5–138.5], p = 0.005). The corresponding risk for patients in the AL arm was 29.9% (95% CI 21.6%–40.5%) compared to 5.9% (95% CI 2.4%–13.5%) in the AL+PQ arm (HR = 5.9 [95% CI 2.3–15.3], p < 0.001) (Fig 2; Tables 2 and 3). In those not prescribed PQ, the risk of recurrence by day 42 was greater following AL than CQ, although this difference was of borderline significance (HR = 1.8 [95% CI 1.0–3.2], p = 0.059) (Table 3).
Genotyping was feasible in 90.9% (362/398) of parasite isolates on day 0 and in 90.2% (46/51) of paired isolates from patients with recurrence prior to day 42. In total, 47.8% (22/46) of the paired P. vivax isolates that could be assessed were homologous (S2 Text).
Blood chloroquine concentrations on the day of recurrence could be measured in two of the four patients with recurrent infection before day 28 in the CQ arm, both of whom had CQ blood concentrations greater than 100 nM.
By day 1, 60.2% (124/206) of those treated with CQ were still parasitaemic compared to 31.8% (61/192) of those treated with AL regimens (odds ratio = 3.3 [95% CI 2.2–5.0], p < 0.001). By day 2, the prevalence of parasitaemia was 6.8% in the CQ arms and 2.6% in the AL arms. By day 3, only two patients in the CQ arms remained parasitaemic (Table 4).
Of the 166 patients with documented fever at enrolment, 96.4% (159/165) were afebrile within 24 h, with 98.8% (84/85) in the AL arms compared to 93.8% (75/80) in the CQ arms (p = 0.109) (Table 4).
After 1 y of follow-up, 150 patients had experienced at least one recurrent episode of P. vivax determined by microscopy (57 after CQ, 62 after AL, 14 after CQ+PQ, and 17 after AL+PQ), and a further eight had had P. falciparum infections (three following CQ and five after CQ+PQ) (Table 2). The risk of any recurrence of P. vivax was 61.7% (95% CI 51.9%–71.7%) following CQ alone compared to 72.4% (95% CI 62.5%–81.6%) following AL alone (p = 0.127). Compared to CQ or AL alone, the risk of recurrence was significantly lower when treatment was combined with PQ: 20.5% (95% CI 13.0%–31.5%) following CQ+PQ (HR = 5.4 [95% CI 3.0–9.7] compared to CQ alone, p < 0.001) and 22.0% (95% CI 14.2%–33.1%) following AL+PQ (HR = 5.2 [95% CI 3.0–9.0] compared to AL alone, p < 0.001). There was no difference in the risk of recurrence at the end of the study between patients treated with CQ+PQ and AL+PQ (Fig 2; Tables 2 and 3).
Recurrent P. vivax infections occurred later in the PQ arms than the monotherapy arms. The median time of the first recurrence was 82.5 d (interquartile range [IQR] 61.0–186.5) for CQ+PQ compared to 51 d (IQR 42.0–86.0) for CQ alone (p = 0.006) and 87 d (IQR 35.0–117.0) for AL+PQ compared to 49.5 d (IQR 35.0–98.0) for AL alone (p = 0.040). In the no-PQ arms, 90.8% (108/119) of the recurrent parasitaemias occurred within the first 6 mo.
A total of 322 P. vivax recurrent infections were detected by microscopy during follow-up (134 after AL, 133 after CQ, 29 after AL+PQ, and 26 after CQ+PQ) (Fig 3). The incidence rate was 2.3 (95% CI 1.9–2.7) episodes per person-year of observation (PYO) for AL, 2.2 (95% CI 1.8–2.6) for CQ, 0.5 (95% CI 0.3–0.7) for AL+PQ, and 0.4 (95% CI 0.3–0.6) for CQ+PQ (Table 3).
The initial treatment with PQ was supervised on scheduled clinic days, and adherence assessed by tablet review at the subsequent visit. The total PQ dose administered per unit body weight was 3.4 mg/kg (IQR 2.8–4.0) in the AL+PQ arm and 3.5 mg/kg (IQR 2.6–4.0) in the CQ+PQ arm. The histogram of PQ dose distribution shows distinct populations that correspond to patients who received a total PQ dose below 2.6 mg/kg and the rest (S1 Fig). There were 27 patients in the CQ+PQ arm and 22 patients in the AL+PQ arm who received a dose below 2.6 mg/kg, and these patients appeared to be at greater risk of recurrence by 6 mo compared to those who received a dose >2.6 mg/kg: HR = 3.8 (95% CI 1.1–13.0) for CQ+PQ and HR = 2.8 (95% CI 1.0–8.0) for AL+PQ, p = 0.036 and p = 0.059, respectively.
For all subsequent recurrences, only the first dose of the antimalarial drugs was supervised. The risk of recurrence after 6 mo of follow-up between the primary and secondary treatments did not differ significantly for patients in either the CQ arm (HR = 1.2 [95% CI 0.8–1.84], p = 0.5) or the AL arm (HR = 0.8 [95% CI 0.5–1.3], p = 0.4). However, patients were at significantly greater risk of recurrent P. vivax after the unsupervised PQ retreatment than after the partially supervised initial PQ treatment: 86.8% (95% CI 77.3%–92.5%) versus 63.2% (95% CI 31.7%–93.3%) in patients in the CQ+PQ arm (HR = 3.9 [95% CI 1.3–11.4], p = 0.014) and 84.0% (95% CI 74.0%–90.4%) versus 51.9% (95% CI 24.4%–73.7%) in patients in the AL+PQ arm (HR = 3.5 [95% CI 1.4–8.9], p = 0.008) (S2 Fig).
A total of 358 AEs were reported, but no serious AEs. There was no difference in the occurrence of any of the categories of AEs between treatment arms or by total PQ dosage received. The most frequent AEs were headache (25.5%), fever (18.8%), chills (9.2%), cough (7.0%), vomiting (6.7%), body pain (5.0%), abdominal pain (4.8%), weakness (4.2%) and diarrhoea (4.0%) (Table 5).
The nadir of Hb concentration occurred on day 3 irrespective of treatment. The mean percentage fall from baseline by day 3 was 6.3% (95% CI 5.0%–7.7%) in the CQ arm, 5.7% (95% CI 4.3%–7.1%) in the AL arm, 5.9% (95% CI 4.4%–7.4%) in the CQ+PQ arm, and 7.6% (95% CI 5.1%–10.0%) in the AL+PQ arm (p = 0.577). On day 3, 3.6% (7/195) of patients treated with a PQ regimen had anaemia (Hb < 100 g/l) compared to 4.2% (8/191) of those treated with a no-PQ regimen (p = 0.08). By day 28, haematological recovery occurred in 211 (70.8%) patients. Hb concentration did not fall below 80 g/l in any of the patients during the study period, and no patient required blood transfusion.
This study highlights that the combination of PQ with either CQ or AL reduced the risk of early recurrence (within 42 d) of P. vivax by up to 3-fold, and decreased the risk of recurrence over 1 y by 2- to 3-fold, compared to CQ or AL alone. PQ (3.5 mg/kg over 14 d) was well tolerated, without significant adverse effects. Recent clinical trials have highlighted the declining efficacy of CQ against P. vivax in Ethiopia [18–20], with the risk of recurrence at day 28 ranging from 14% [21] to 22% [20]. Our study provides evidence of low-grade CQ resistance [4,22], with 3.8% of patients having recurrence by day 28 in the presence of adequate CQ blood concentration (>100 nM). However, despite CQ’s compromised efficacy, the risk of recurrence was almost 2-fold greater in patients treated with AL alone compared to those treated with CQ alone, although this difference did not reach statistical significance (p = 0.059). In equatorial regions endemic for P. vivax, including Ethiopia, the risk of P. vivax relapse is generally high, with the first recurrence occurring about 21 d following the initial treatment [23]. Lumefantrine has an elimination half-life of 3–6 d, and, by 16 d, drug concentrations have fallen well below the minimum inhibitory concentration for the parasite and thus are no longer sufficient to prevent relapsing infections. Conversely, the slower elimination of CQ affords prolonged post-treatment prophylaxis, capable of suppressing early relapses [24]. By day 42, both CQ and lumefantrine levels are expected to be below the minimum inhibitory concentration, and, hence, in the absence of PQ, it is not surprising that there was no difference in the incidence of recurrent infections beyond that time.
The prolonged duration of the study and repeated administration of the same treatment regimen for each episode of malaria allowed quantification of the cumulative risk of P. vivax and incidence of recurrence over a 12-mo period. The incidence of infection reflects the likely impact that would result from corresponding policy change. Patients receiving CQ or AL alone had four or five recurrences, with an overall incidence rate of approximately two episodes per PYO. The addition of a supervised 14-d regimen of PQ resulted in a 4-fold reduction in the incidence rate, from 2 to 0.5 episodes per PYO. Radical cure with PQ is a function of the total mg/kg dose administered [25,26], and this was highlighted by a clear dose response in our study. Patients who received a total PQ dose below 2.6 mg/kg had significantly more recurrences than those who received a greater dose: almost 4-fold more in the CQ+PQ arm and 3-fold more in the AL+PQ arm (p < 0.05).
There was also a 3- to 4-fold greater risk of recurrence following treatment of patients’ first recurrent parasitaemia (which was unsupervised) compared to that following treatment of their initial parasitaemia (which was semi-supervised). Although this non-randomized comparison is vulnerable to bias, there was no difference in the risk of recurrence following the first and second treatments with CQ or AL monotherapy, suggesting that the difference was likely due to poor adherence to an unsupervised 14-d regimen of PQ. These findings are consistent with previous studies that have shown PQ adherence falling to less than 30% when unsupervised [27–29]. Significantly lower rates of recurrent parasitaemia were seen in a Thai study when PQ administration was directly observed compared to when it was unsupervised [30], although a difference was not observed in a study in Pakistan [31]. Our study emphasises the need for improved delivery of radical cure, using DOT, shorter treatment courses, and greater emphasis on patient and community education to promote the importance of radical cure and the need to continue treatment after resolution of symptoms [32].
There is reluctance of some healthcare providers to prescribe PQ, mainly due to concerns over its safety and potential for causing severe haemolysis [33]. To ensure safe delivery of PQ, WHO recommends prior testing for G6PD deficiency [34]. However, not many P. vivax endemic countries have implemented this yet. Further work to understand the barriers to routine G6PD testing is currently underway. In our study, PQ was administered to individuals confirmed to be G6PD normal at a total dose of 3.5 mg/kg (a low-dose radical cure regimen) [26]. This regimen was well tolerated, with no significant difference in the number or type of AEs between the PQ and no-PQ study arms. The predominant decline in Hb concentration occurred between enrolment and day 3, although PQ treatment only commenced on day 2. Hb recovery began after day 3; the greatest fall in Hb was thus a consequence of malaria rather than being drug-induced. Furthermore, treatment with PQ resulted in a significant reduction in P. vivax recurrences, each of which were associated with repeated bouts of haemolysis and a cumulative risk of anaemia [5,35].
There were a number of limitations of the study. First, the omission of routine Hb measurements after day 42, except in patients with recurrent malaria, prevented quantification of the long-term effects of recurrent episodes of malaria on the cumulative risk of anaemia. Further studies are needed to define the risk and benefits of alternative PQ dosing strategies in different endemic settings. Second, whilst all of the early recurrent parasitaemias were genotyped and compared with the pre-treatment parasitaemia, this comparison was not undertaken for recurrent infections after day 42. Heterologous recurrences occurring within the first 42 d can be due to either reinfection or relapse events but not true recrudescences. Parasite genotyping allows these events to be censored, enabling a better estimate of early efficacy. However, genotyping of late recurrences cannot distinguish between relapsing infections, which can be either homologous or heterologous, and reinfections. For this reason, we used a conservative approach to quantify and compare the cumulative risk and incidence of all recurrent P. vivax infections irrespective of whether they were homologous or heterologous. Third, the early termination of our trial reduced the final sample size. However, loss to follow-up was lower than assumed in the prior power calculation, and the final sample size enrolled into the study still achieved a power of greater than 80% for the primary outcomes of the study.
In conclusion, although there was evidence of low-grade CQ resistance of P. vivax in our study, the risk of early recurrence appeared greater following AL, likely due to relapsing infections emerging after the post-treatment prophylaxis of AL had waned. When administered as monotherapy, there was little benefit of changing the first-line therapy for vivax malaria from CQ to AL in this region of Ethiopia. However, PQ treatment should be included in future revision of the national treatment guidelines to decrease relapse rates, with likely benefits in reducing both malarial anaemia and transmission potential. Better methods of ensuring adequate adherence will be needed if the public health impact is to be maximised.
|
10.1371/journal.pcbi.1003306 | Modeling Integrated Cellular Machinery Using Hybrid Petri-Boolean Networks | The behavior and phenotypic changes of cells are governed by a cellular circuitry that represents a set of biochemical reactions. Based on biological functions, this circuitry is divided into three types of networks, each encoding for a major biological process: signal transduction, transcription regulation, and metabolism. This division has generally enabled taming computational complexity dealing with the entire system, allowed for using modeling techniques that are specific to each of the components, and achieved separation of the different time scales at which reactions in each of the three networks occur. Nonetheless, with this division comes loss of information and power needed to elucidate certain cellular phenomena. Within the cell, these three types of networks work in tandem, and each produces signals and/or substances that are used by the others to process information and operate normally. Therefore, computational techniques for modeling integrated cellular machinery are needed. In this work, we propose an integrated hybrid model (IHM) that combines Petri nets and Boolean networks to model integrated cellular networks. Coupled with a stochastic simulation mechanism, the model simulates the dynamics of the integrated network, and can be perturbed to generate testable hypotheses. Our model is qualitative and is mostly built upon knowledge from the literature and requires fine-tuning of very few parameters. We validated our model on two systems: the transcriptional regulation of glucose metabolism in human cells, and cellular osmoregulation in S. cerevisiae. The model produced results that are in very good agreement with experimental data, and produces valid hypotheses. The abstract nature of our model and the ease of its construction makes it a very good candidate for modeling integrated networks from qualitative data. The results it produces can guide the practitioner to zoom into components and interconnections and investigate them using such more detailed mathematical models.
| Within the cell of an organism, three networks—signaling, transcriptional, and metabolic—are always at work to determine the response of the cell to signals from its environment, and consequently, its fate. Evidence from experimental studies is painting a picture of complex crosstalk among these networks. Thus, while a wide array of computational techniques exist for analyzing each of these network types, there is clear need for new modeling techniques that allow for simultaneously analyzing integrated networks, which combine elements from all three networks. Here, we provide a step towards achieving this task by combining two population modeling techniques—Petri nets and Boolean networks—to produce an integrated hybrid model. We demonstrate the accuracy and utility of this model on two biological systems: transcriptional regulation of glucose metabolism in human cells, and cellular osmoregulation in yeast.
| While the genome contains all hereditary information, the decisions that a cell makes are governed by a complex cellular machinery that resides above the genome. Modeling this machinery is both important—as it helps understand proper cellular functioning and the implications of aberrations thereof, and a daunting—given the “known unknowns” (e.g., kinetic parameters of given reactions) and the “unknown unknowns” (data incompleteness is the rule, rather than the exception, in biological research).
The cellular machinery can be broken down into three main components—signaling, transcription regulation, and metabolism—each of which consists of a network of molecules and interactions among them. The signaling network is responsible for relaying messages from the external environment of a cell to the nucleus. Inside the nucleus, the transcription regulation network determines, upon receiving signals, which genes are expressed, and to what extent. The metabolic network is the energy and resource management component of the cell, producing energy and products that are required by cellular processes. Various modeling techniques have been used successfully for modeling the dynamics of each of these components individually.
The success of modeling each of the three components individually notwithstanding, these components are interconnected within the cell and their dynamics are intertwined, thus creating a complex network whose modeling and understanding are major endeavors in systems biology. Several biological studies and surveys have highlighted this interconnection inside the cell and the significance of analyzing the components simultaneously rather than individually, including, but not limited to, [1]–[6].
Indeed, several approaches were introduced recently for integrated modeling of biological networks: regulatory FBA (rFBA) [7], steady state regulatory FBA (SR-FBA) [8], integrated FBA (iFBA) [9], integrated dynamic FBA (idFBA) [10], probabilistic regulation of metabolism (PROM) [11], the method of [12], dynamic FBA (dFBA) [13] and a recently published whole cell computational model [14]. One common aspect to all the existing models is the use of flux balance analysis (FBA) for modeling carbon and energy metabolism. FBA is a widely used method that estimates fluxes of metabolic reactions, thereby making it possible to predict the growth rate of an organism or the rate of production of a metabolite of interest. However, FBA is only suitable for determining fluxes at steady state. With exceptions of some modified forms, FBA does not account for regulatory effects such as activation of enzymes by protein kinases or regulation of gene expression [15]. The methods that use the unmodified version of FBA – all but idFBA and dFBA — only capture the steady state of metabolism, therefore not capturing the full dynamic within the cell. These methods mainly acquire the effects of changes that individual components have on each other. On the other hand, the methods that discretize FBA (dFBA and idFBA), are able to reveal not only a more complete profile of the cell, but also the dynamic behavior of the interconnections between the components. For recent surveys of these methods, please see [16], [17].
In this paper, we propose a new Integrated Hybrid Model (IHM) that aims to capture the dynamic behavior within and between the components of the cell, and which belongs to the class of executable models [18]. This model integrates two types of modeling techniques: Petri nets (PNs), which have been used for modeling metabolic networks and signaling networks [19], and Boolean networks, which have been used to model regulatory networks as well as protein signaling networks [20], [21]. One of the first successful Petri net-based models of metabolism was devised by Reddy et al. [22], [23]. Over the recent years, various types of Petri nets have been introduced and extensively used in modeling different metabolic systems [24]–[27]. Signaling pathways, on the other hand, have posed more of a challenge for Petri nets. Their highly interleaved (with possible forward- and backward loops) and parametrized nature makes it a difficult mapping onto a Petri net framework. Despite these limitations, Petri nets have been shown to be applicable in signaling pathways using careful parameterization and execution strategies [28]–[32]. Transcription regulation has been modeled successfully using Boolean networks, starting with the work of [33]. Over the years, with the steady increase in the amount of data on genetic regulation, Boolean networks became a common strategy for modeling this cellular process; e.g., [34]–[36].
Our integrated hybrid model uses Petri nets to model the metabolic and signaling components, and Boolean networks to model the transcriptional component. Further, the model makes connections between the Petri net and Boolean network component using a special modeling part. Our modeling approach assumes knowledge of the connectivity among the various species in the system, and is then minimally parameterized based on qualitative data. The dynamics of the biological system are then obtained by executing the parametrized model. Of the existing approaches, idFBA is comparable to our approach, as it allows for modeling the dynamics by discretizing time and conducting FBA analyses for short time intervals. However, idFBA is applicable where FBA models have been curated (e.g., for single-cell organisms), whereas our modeling approach is applicable more broadly in terms of organism selection, and requires only qualitative data.
We implemented and tested our modeling methodology on two biological systems: (1) the transcriptional regulation of glucose in human physiology, with knowledge based on [1], and (2) osmoregulation in S. cerevisiae, based on the system in [37]. The two systems differ in temporal and spatial scales. For the transcriptional regulation of glucose, the interactions among different components are reflected in the cooperation among multiple cell types, and the mass transportation is through blood vessels in the human body, thus acting at longer time scales than single cell systems. On the other hand, the modeling of osmoregulation in S. cerevisiae encompasses metabolism, signaling and transcriptional regulation, all within a single cell. The exchange of proteins or metabolites is mediated through diffusion and cellular transportation. We choose the two systems to show the diversity of the biological scenarios to which our integrated hybrid model is applicable. The two systems are very well curated and studied, both experimentally and computationally. This makes them ideal for validating our methodology and for comparing with existing modeling frameworks. Our modeling approach produced results that match experimentally derived data (in terms of both validation and prediction). There is an abundance of qualitative data on biological interaction networks, and developing models and methods that utilize such data is desirable. Our proposed method fits within this category which offers a complementary approach, rather than an alternative one, to the FBA-based category of methods as well as other categories such as kinetics-based methods.
Our integrated hybrid model combines two modeling techniques, Petri nets and Boolean networks. We begin by briefly reviewing each of these models, and their use in modeling biological networks, and then describe the new integrated hybrid model.
In our context, a Petri net (PN) is a 4-tuple that defines a weighted, complete, directed, bipartite graph. The disjoint sets and correspond to two types of nodes, places and transitions, respectively. In modeling signal transduction and metabolism, they correspond to chemical species and biochemical reactions that happen among these species. The element is a mapping defined , where is the set of non-negative real numbers. These mappings could be used to encode, for example, stoichiometries of biochemical reactions. Finally, is the initial marking of the Petri net, which assigns a number of tokens to each place. This correspond to the initial concentration of chemical species. The state of a Petri net is given by a vector of length with being the number of tokens in place . In particular, the initial state, , is given by the initial marking . Additionally, a vector of length provides the transition rates for the system, where denotes the rate of transition to simulate the empirical rate constant used in the law of mass action that governs the corresponding reaction. The Petri net can be executed both deterministically and stochastically [38]–[40]. In this work, we utilize a stochastic protocol based on the Gillespie “first reaction” method [41]. The method characterizes the dynamics of each transition by a propensity function . Let be a transition whose inputs is the set and outputs is the set . In state , the propensity of transition is defined byGiven these propensity values, the method determines the putative time at which the next transition fires based on the probability distribution function given byThe transition with the smallest time is then chosen to fire. Firing transition amounts to updating the number of tokens in every place according to the rule and updating the number of tokens in every place according to the rule . Once a transition is executed, the state of the Petri net changes. The execution time is updated by 1, which is, in our case, a slight modification from the original algorithms where time is updated by . Consecutive firings of transitions results in a walk through the state space of the Petri net from the start state . The final dynamics of the system is acquired by averaging several full runs of Gillespie starting from the initial state and executing the same number of steps. A detailed description of Petri nets and its application to systems biology can be found in [19]. See Figure 1 for an illustration.
A Boolean network is a 3-tuple , where is a vector of Boolean variables (that is, variables that take values in the set ) and is a vector of Boolean functions with function , for , associated with variable , and is a vector of length that has a Boolean value for each of the variables and denotes the start state. In modeling transcriptional regulation, each Boolean variable indicates whether a gene is being transcribed at a given time and the Boolean functions stipulate how transcriptional factors regulate the transcription of their targets. The state of a Boolean network is a Boolean vector of size , where is the value of variable . The value of of variable is updated by applying function to the current state of the Boolean network. More formally, let be the state of the Boolean network at time . Then, if function is executed at time , the state of the Boolean network one step later is given by , where for every , and . In particular, . Given a Boolean network representing a set of variables, the dynamics of the system can be simulated by repeatedly executing the Boolean functions and updating the “current” state. In the classical synchronous simulation, the states of all variables are updated simultaneously after all of the functions in have executed. In an asynchronous simulation, only one Boolean function is chosen and executed in a given time step. See Figure 1 for an illustration.
As described above, gene regulatory networks have been successfully modeled using Boolean networks. Signaling and metabolic networks have been successfully modeled using Petri nets. In our integrated hybrid model, the regulatory components of the biological system are modeled using Boolean networks, whereas the other two components are modeled using Petri nets. To facilitate connections between the two components, our model contains, in addition to the Petri net and Boolean network components, a set of Place-to-Boolean and Boolean-to-Place triplets that create a Boolean value based on binarization of the number of tokens and a number of tokens based on a Boolean value, respectively. We now describe our modeling approach formally.
In this section, we demonstrate the application of our new modeling approach on two biological systems: transcriptional regulation of glucose and Osmoregulation in S. cerevisiae. Both biological systems intrinsically involve metabolic, signaling, transcriptional regulatory components and complicated interaction in-between these components. They cannot be comprehensively modeled using traditional frameworks that specifically targets separate cellular components.
In order to assess the ability of IHM to capture the dynamics of complex biological systems we implement an IHM model for the system of transcriptional regulation of glucose metabolism, which was surveyed in [1]. Timely uptake of cellular glucose from the blood, a task regulated by the secretion of insulin and glucagon, is crucial to human metabolism. This system involves the interaction of multiple cellular components in cells of different cell types and cells that span a physical distance. We demonstrate that IHM can readily be adapted to model such a biological scenario, and allows us to investigate issues such as the interplay between AKT and FOXO in this system. Given that the modeled system involves more than a single cell type, it is unclear how to apply FBA-based techniques to it.
Yeast responds to the environmental osmolarity by adjusting the cellular glycerol concentration [37]. Such response is mediated through signaling pathways that sense the extracellular osmotic pressure as well as transcriptional regulation of about 10% of the yeast genes that manipulate the metabolism of glycerol. The effect of the medium osmolarity is first sensed and transmitted by the well-studied HOG/MAPK pathway [64], [65] whose upstream involves two redundant branches—Sho1 branch [66], [67] and Sln1 branch [68]. The HOG signaling pathway is one of the first to sense the osmotic upshift, playing a pivotal role in yeast's adaptation to high osmolarity. Hog1, the end effector of HOG pathway, activates in the nucleus the central transcriptional factors Hot1 [69], Msn2/4 [70] and Ptp2/3 [71]. These transcriptional factors turn on the expression of enzymes that promote glycolysis, which leads to the production of glycerol, an inert osmolyte. The surge in the glycerol concentration increases the cytosolic osmolarity, counteracting the osmotic upshift in the environment and protecting the cell from dehydration. While the effect of Gpd1/Gpp2 (which is a product of Hot1 and MSN2/4) gene controls osmoregulation via glycerol production though metabolic pathway, Ptp2/3 is a much stronger mediator of osmotic stress, as it acts on suppressing the activity of Hog1 transcription factor directly.
We proposed a simple, yet effective, integrated hybrid model (IHM) that allows for simultaneously modeling signaling, metabolic, and regulatory processes within a single framework, while explicitly capturing the dynamics within each component and the interplay among them. As we applied the integrated model to two biological systems, we demonstrated how much our model can capture by mainly relying on the topology of the system (given the simple and general rules for setting most of the model's parameters). In both systems, we were able to successfully validate our results against both experimental data and other models. In the case of transcriptional regulation of glucose, we compared our model against an ODE-based model that only focuses on glucose-insulin interactions, while in our case we consider a larger system. The results compare well against the experimental data. No comparison was done with other integrated models, since it is not clear how to formulate an FBA-based model for this system. In the case of the osmoregulation system, we compared our model against the idFBA approach [10].
The IHM framework has an intuitive graphical representation that makes the construction of the connectivity map of the model a relatively simple task. Further, as experimental evidence becomes available to provide support for new connections or against existing ones, the connectivity map can be readily updated to accommodate this new evidence without having to recreate the model from scratch. Our model is reconstructible and its parameterization is obtainable from qualitative data, which is abundant in the literature and public databases. It is important to note that while the connectivity map is often easy to obtain from the literature and public databases, parameterizing the IHM poses the biggest challenge in terms of obtaining the executable model. In this paper, we parameterized the IHM for both biological systems manually—a task that took very short time to achieve, given that most of the parameters were set using general rules and only a few of them had to be fine-tuned. The results (e.g., the feed/fast cycle in the regulation of glucose metabolism system) are qualitatively robust to most parameter values that we choose, as tested by executing the model with parameters varied around the chosen value. We identify as a direction for future research the task of devising computational techniques for automated parameterization of our IHM using qualitative experimental data. Some techniques for a similar task were recently introduced [72] and we will build on those.
While the aforementioned existing approaches for integrated analysis of biological networks provide promising frameworks, a salient feature of all of them is that they depend on flux-balance analysis (FBA) as a main analytical component. This dependence means that an FBA model must be curated for the system under analysis, which is not clear how to obtain for a system such as the regulation of glucose metabolism, which involves more than a single cell type. Further, this dependence necessarily makes the analysis metabolism-centric and shifts the focus from the other two components. Third, as FBA is aimed at understanding the behavior of the system at steady state, the dynamics of the system cannot be studied, except under the idFBA modeling technique, as it takes a step-wise approach to conducting FBA. Our model, on the other hand, is not based on FBA and, consequently, provides a complementary approach to the FBA-based ones.
Our model builds on the success of Boolean networks and Petri nets for modeling cellular networks. As advances continue to be made for both modeling techniques, our integrated modeling approach would readily benefit from these advances, as different flavors of of Boolean networks (e.g., probabilistic ones) and Petri nets (e.g., colored Petri nets) can be plugged into our model without having to modify the way the connectivity map is constructed or the system is executed. In other words, our model can be viewed as a reconfigurable model, where different components, along with their execution protocols, can be assembled to generate a model of integrated systems.
It is important to note that while we made decisions on the model to fit the two biological systems we studied, other biological systems may require more features in the modeling approach. For example, in the Petri-to-Boolean connections, it might be the case that the state of the Boolean variable is set based on a function of a set of the Petri net places. Our IHM can be easily extended to incorporate such features, with little or no need to modify the execution strategy. That is, the model is easy to extend as long as the syntax of the new features and their effects on the execution strategy are well-defined.
Last but foremost, our IHM approach lends itself in a straightforward manner to hypothesis generation. Perturbation experiments can be simulated in silico by setting the numbers of tokens at Petri net places and Boolean variables to a certain value, and the system can be executed to study the effect. For example, a Boolean variable can be set to 0 to simulate its inhibition, or the number of tokens can be set to a large number in place to represent a constitutive enzyme. Further, new components can be added in or existing ones can be removed easily to study the effect of these components on the overall performance of the system. Finally, while we chose to model transcriptional regulation using Boolean networks here, the entire system (that is, all three types of biological networks) could be represented using a single Petri net. This allows for a more refined simulation of the transcription factors and their targeted genes, but also requires replacing the Boolean functions by Petri net transitions whose parameters must be learned from the data.
|
10.1371/journal.ppat.1007024 | Glycoengineering HIV-1 Env creates ‘supercharged’ and ‘hybrid’ glycans to increase neutralizing antibody potency, breadth and saturation | The extensive glycosylation of HIV-1 envelope (Env) glycoprotein leaves few glycan-free holes large enough to admit broadly neutralizing antibodies (bnAb). Consequently, most bnAbs must inevitably make some glycan contacts and avoid clashes with others. To investigate how Env glycan maturation regulates HIV sensitivity to bnAbs, we modified HIV-1 pseudovirus (PV) using various glycoengineering (GE) tools. Promoting the maturation of α-2,6 sialic acid (SA) glycan termini increased PV sensitivity to two bnAbs that target the V2 apex and one to the interface between Env surface gp120 and transmembrane gp41 subunits, typically by up to 30-fold. These effects were reversible by incubating PV with neuraminidase. The same bnAbs were unusually potent against PBMC-produced HIV-1, suggesting similar α-2,6 hypersialylated glycan termini may occur naturally. Overexpressing β-galactosyltransferase during PV production replaced complex glycans with hybrid glycans, effectively 'thinning' trimer glycan coverage. This increased PV sensitivity to some bnAbs but ablated sensitivity to one bnAb that depends on complex glycans. Other bnAbs preferred small glycans or galactose termini. For some bnAbs, the effects of GE were strain-specific, suggesting that GE had context-dependent effects on glycan clashes. GE was also able to increase the percent maximum neutralization (i.e. saturation) by some bnAbs. Indeed, some bnAb-resistant strains became highly sensitive with GE—thus uncovering previously unknown bnAb breadth. As might be expected, the activities of bnAbs that recognize glycan-deficient or invariant oligomannose epitopes were largely unaffected by GE. Non-neutralizing antibodies were also unaffected by GE, suggesting that trimers remain compact. Unlike mature bnAbs, germline-reverted bnAbs avoided or were indifferent to glycans, suggesting that glycan contacts are acquired as bnAbs mature. Together, our results suggest that glycovariation can greatly impact neutralization and that knowledge of the optimal Env glycoforms recognized by bnAbs may assist rational vaccine design.
| Here we engineered various changes in the sizes and shapes of sugars that decorate HIV surface spike proteins and tested the effects of these changes on virus susceptibility to neutralizing antibodies. In so doing, we were able to define the optimal Env-sugars recognized by prototype bnAbs that recognize various canonical epitope clusters on Env spike proteins. Some bnAbs preferred spike proteins decorated with large, complex glycans. Others preferred smaller glycans that improved their access to underlying protein targets. For similar reasons, germline-reverted versions of bnAbs were also generally more effective when the glycans were small. In some cases, bnAbs acquired an ability to bind to sugars as they matured. A comparison of viruses generated in cell lines and primary cells revealed large differences in bnAb sensitivity, raising questions about clinical relevance of cell line-produced virus for checking vaccine responses and, moreover, the use of these cell lines for manufacturing vaccines. Overall, just as car engines may be modified to be supercharged or hybrid for increased power or efficiency, the sugars of HIV coat proteins may also need to be engineered as 'supercharged' and 'hybrid' or otherwise modified in rational vaccine designs to optimize bnAb recognition.
| Neutralizing antibodies (nAbs) are likely to be an essential part of the immunity conferred by an effective HIV-1 vaccine [1]. NAbs interfere with HIV-1 infection by binding to functional Envelope glycoprotein (Env) spikes consisting of gp120/gp41 trimers arrayed on virion surfaces, thereby blocking receptor engagement and/or membrane fusion. Env trimer surfaces are populated by a dense glycan network that constitutes ~50% of its mass [2–5]. Since anti-glycan Abs are regulated by immunological tolerance, glycans provide a formidable defense against nAbs that, at least initially, must attempt to navigate past to reach underlying protein epitopes [6–8]. Although the space between glycans is sufficient for access by single immunoglobulin domain ligands (e.g., soluble CD4 and llama Abs) and by bovine Abs with very long, protruding heavy chain third complementarity determining loops (CDRH3s) [9–11], glycan-free spaces are typically insufficient for human Abs, that consist of two immunoglobulin chains and less protrusive CDRH3s. Structural and glycan array studies reveal that some bnAbs overcome this problem by contacting composite protein-glycan epitopes [4].
HIV-1 Env's unparalleled sequence diversity presents a daunting challenge for vaccinologists aiming to induce broadly neutralizing antibodies (bnAbs). The heterogeneity of its surface glycans could add an additional layer of difficulty. In mammals, N-linked glycosylation involves >700 genes that can impart a plethora of carbohydrate structures to Asn-X-Ser/Thr sequons (where X can be any amino acid except for proline) [12]. Glycosylation (summarized in Fig 1) begins in the endoplasmic reticulum, where an oligomannose glycan precursor (Glc3Man9GlcNAc2; Glc = glucose, Man = mannose, GlcNAc2 = N-acetylglucosamine) is transferred to nascent proteins prior to folding (Fig 1: top row, fourth glycan from left). Trimming of mannose termini results in Man5GlcNAc2, the simplest oligomannose glycan, which may then be modified to form a variety of hybrid or complex glycans that may be fucosylated, galactosylated, sialylated and/or bisected by a central GlcNAc moiety (the latter is modeled in Fig 1: bottom row, rightmost hybrid glycan).
The 75–105 N-linked glycoforms on the surface of HIV-1 Env trimers range from untrimmed high mannose glycans to complex, multiantennary glycans. One key regulator of their maturation state is the cellular expression of enzymes that catalyze each step (Fig 1), which depends on factors including host genetics, age, infection, and pregnancy [13, 14]. Although early mannose trimming is usually efficient, later steps such as galactosylation and sialylation may be inefficient, so that some complex and hybrid glycans may incompletely mature and lack these termini. Further variation can arise from the covalent bond angle of terminal sialic acid (SA) moieties, which may be predominantly α-2,3 or α-2,6-linked in different species, tissues and cell lines [15–18]. For example, peripheral blood mononuclear cell (PBMC)-derived HIV-1 Env is modified mostly with α-2,6-linked SAs, whereas that produced in human embryonic kidney (HEK) 293T or Chinese hamster ovary (CHO) cells bears mostly or exclusively α-2,3-linked SAs, respectively [17, 19, 20]. Furthermore, 293T and Jurkat cell lines impart a higher proportion of high mannose and hybrid glycans than CHO cells [15]. In humans, α-2,8-linked SA may be linked to α-2,3 or α-2,6-linked SA, usually in neuronal tissue (Fig 1).
The density of surface glycans is so great in some Env domains that α-mannosidase, like bnAbs, has difficulty in gaining access due to steric constraints. This results in an unusually high proportion of immature oligomannose glycoforms (Man5GlcNAc2 –Man9GlcNAc2), including an oligomannose patch common to all forms of Env [17, 21, 22]. In general, as Env sequons "compete" for glycan addition and modification, each may variably become occupied by oligomannose, hybrid or complex glycans or, due to steric competition with neighboring sequons, may occasionally remain unoccupied ("sequon skipping"). Together, the above factors contribute to considerable Env glycodiversity [21, 23].
HIV-1 bnAbs fall into 5 distinct epitope clusters: V2 apex, V3-glycan, CD4 binding site (CD4bs), gp120-gp41 interface and membrane-proximal external region (MPER), whose epitopes collectively cover a large portion of the trimer's exposed surface [2, 4, 24–29]. Since most of these bnAbs make some glycan contacts, it appears possible that glycodiversity could modulate their activities [30].
V2 apex-specific bnAbs include at least five families (PG9, CAP256, CH01, PGT145 and PCT64-35S) that exhibit unusually long (>24 amino acid) anionic CDRH3s that project outward to penetrate Env’s glycan shield and reach underlying protein [31–41]. In contrast, another V2 apex lineage represented by VRC38.01 uses a 16AA non-protruding CDRH3 that binds via side-chain to side-chain contacts [42]. These bnAbs contact the positively charged strand C and conserved glycans, typically at positions N156 and N160 [18, 21, 32, 34, 39, 42]. Previous studies showed that PG9 engages SA termini [39, 43], recognizing α-2,3 SAs in one array [36], but preferentially binding α-2,6 SAs in the context of an intact V1V2 domain [4, 39]. CAP256.09 also binds α-2,6 SAs [4, 40]. In contrast, CH01 recognizes mannosylated V2 peptides [44]. PGT145 does not bind in glycan arrays and appears to be largely insensitive to glycan changes [4]. V2 bnAbs may also ‘clash’ with some glycans. For example, glycans at position 130 of the V1 and C-terminal V2 (V2’) of some strains can limit V2 bnAb sensitivity [42, 45]. Furthermore, productive bnAb-glycan contacts or clashes may be influenced by the glycodiversity mentioned above, resulting in non-sigmoidal neutralization curves and sub-saturating neutralization, even with excess nAb [4, 42, 46, 47].
V3-glycan bnAbs target the intrinsic mannose patch, usually centered around the N332 glycan [37, 48–52]. In the absence of this glycan, proximal glycans, e.g. N137, N156, N295 and N301 can contribute to binding [53]. In some scenarios, however, glycan clashes may regulate neutralization [30]. Some of these bnAbs also recognize complex glycans [4, 37, 48, 50, 54].
Most CD4bs bnAbs are subject to possible clashes with a "glycan fence" that includes the N276 glycan of loop D and V5 loop glycans that surround the underlying receptor binding site [8, 27, 29, 55, 56]. Changes in the composition or maturation state of the glycan fence may regulate virus sensitivity to CD4bs bnAbs [57]. However, some bnAbs (HJ16 and 179NC75) incorporate the N276 glycan in their epitopes [58, 59] and VRC13 contacts several partially mature glycans in arrays [4].
Gp120-gp41 interface bnAbs exhibit diverse glycan dependencies. 35O22 [60] targets a quaternary epitope involving glycans N88, N230, N241 and N625, and binds oligomannose and complex glycans in arrays [4]. PGT151 recognizes tetra-antennary glycans at positions 611 and 637, with the N448 glycan playing a regulatory role [3, 4, 21, 61]. 3BC176 does not bind to glycan arrays but is sterically impacted by the N88 glycan [62, 63]. Conversely, ACS202 and VRC34.01 depend on the N88 glycan, but the latter is sterically impeded by the N611 glycan [64, 65]. 8ANC195 depends on the N234, N276 and N637 glycans and clashes with the N230 glycan [66–68]. Finally, CAP248-2B binds proximal to interface glycans, but appears to be unaffected by glycoform changes [69].
MPER bnAbs 10E8, 2F5, Z13 and 4E10 are not known to recognize glycans [70–72]. However, partial deglycosylation by PNGase F increases 2F5 and 4E10 affinity, perhaps due to the removal of complex glycans at the trimer base [73]. Incomplete neutralization by bnAb 10Ee8 may be due to glycan heterogeneity, particularly at position N625 [74].
Glycoengineering (GE) methods, including several outlined in Fig 1 can alter glycan maturation state and involve the use of: i) glycosylation inhibitors, ii) glycosyltransferase knockout cell lines, iii) in vitro enzyme reactions and iv) glycosyltransferase plasmid co-transfections. To date, only a handful of GE methods have been used to modify HIV-1 [4, 37, 75, 76]. For example, kifunensine, which prevents Man9GlcNAc2 trimming (Fig 1), decreases HIV-1 sensitivity to some V2 apex mAbs [37, 39, 43, 76, 77], but increases sensitivity to PGT125 and 35O22 [60]. Virus production in a knockout cell line lacking functional N-acetylglucosaminyltransferase I (GNT1-) increases virus sensitivity to some mAbs [75, 76], but decreases PGT151 sensitivity [3, 4, 61, 78]. Aside from these and other anecdotal reports, the effects of GE on HIV-1 neutralization have not been comprehensively investigated. To fill in this knowledge gap, we investigated the effects of 16 GE methods on the sensitivities of 293T cell-produced pseudoviruses (PVs) to a large panel of bnAbs. Some bnAbs were dramatically impacted. PG9 and CAP256.09 were up to ~30-fold more potent against PVs produced with co-transfected α-2,6 sialyltransferase. PGT151 and PGT121 were more potent against PVs with terminal SA removed. 35O22 and CH01 were more potent against PV produced in GNT1- cells. The effects of GE on bnAbs VRC38.01, VRC13 and PGT145 were inconsistent between Env strains, suggesting context-specific glycan clashes. Overexpressing β-galactosyltransferase during PV production 'thinned' glycan coverage, by replacing complex glycans with hybrid glycans. This impacted PV sensitivity to some bnAbs. Maximum percent neutralization by excess bnAb was also improved by GE. Remarkably, some otherwise resistant PVs were rendered sensitive by GE. Germline-reverted versions of some bnAbs usually differed from their mature counterparts, showing glycan indifference or avoidance, suggesting that glycan binding is not germline-encoded but rather, it is gained during affinity maturation. Overall, these GE tools provide new ways to improve bnAb-trimer recognition that may be useful for informing the design of vaccine immunogens to try to elicit similar bnAbs.
Various GE methods (Fig 1) were used to modify HIV-1 PV Env glycans and then examine the effects on nAb sensitivity. In some cases, we co-transfected plasmids encoding glycosyltransferases or added decoy substrates during PV production in 293T cells. In other cases, PVs were produced in GNT1- knockout cell line or were incubated with neuraminidase (NA) to remove SA termini.
An E168K+N189A JR-FL mutant was used to fully knock in PG9 bnAb sensitivity (K168 is a critical contact; N189A knocks out a competitive glycan). We used a gp41 tail-truncated clone (gp160ΔCT) to increase expression with only a marginal impact on its neutralization sensitivity [79]. For convenience, we refer to this clone as 'JR-FL' henceforth.
It was previously reported that HIV-1 desialylation increases infectivity [80]. However, with the exception of swainsonine, our GE methods, several of which eliminate sialylation, did not enhance infection (S1A Fig). In fact, kifunensine and GNT1- PV infectivities were both <20% of control levels. This discrepancy could be due to our use of PV instead of replicating virus in the cited study, that we performed NA digests after rather than during virus production, and/or that some glycosylation inhibitors may impact Env expression and, by extrapolation, infectivity.
The neutralizing activities of a representative panel of bnAbs of the 5 major epitope clusters were next tested against GE-modified PVs. Two non-neutralizing mAbs (non-nAbs) (14e and F105) were also included to monitor for any overt changes in trimer folding. An overview of IC50s as a heat map is shown in Fig 2A. Fig 3 shows the effects of modifying early, middle and late stages of glycosylation on JR-FL sensitivity to V2 apex and gp120-gp41 bnAbs and the non-nAb controls. S2 Fig shows the effects of the same modifications on the 3 other bnAb clusters.
Early GE tools included adding inhibitors kifunensine and swainsonine or co-transfecting plasmids expressing N-acetylglucosaminyltransferases 1 and 3 (GNT1 and 3) during PV production in 293T cells. A fifth early GE variant was PV produced in GNT1- cells. The resulting PVs are referred to hereafter as their modification followed by PV. GE-modified PVs were generally resistant to non-nAbs, although the GNT1- PV was mildly sensitive to V3 mAb 14e (Fig 3). Although F105 reduced swainsonine PV infectivity to a plateau at ~80% in the data shown, this was not observed in repeats. Overall, early GE methods did not markedly affected trimer compactness. We next analyzed 4 prototype V2 apex bnAbs: PG9, PGT145, CH01 and VRC38.01. Like PGT145 [4], VRC38.01 was unreactive in oligomannose glycan arrays (S3 Fig). Consistent with previous studies, kifunensine PV was resistant to PG9 and CH01 [37, 42, 77], but had little effect on PGT145 and VRC38.01 [42]. Conversely, GNT1- PV was >100-fold more sensitive to CH01, suggesting that small glycans eliminate binding clashes (Figs 2A and 3). GNT1- PV was also ~10-fold more sensitive PGT145, marginally more sensitive to VRC38.01, and marginally more resistant to PG9. On the other hand, swainsonine, which inhibits D2 and D3 mannose trimming (Fig 1), increased PG9 sensitivity (Figs 2A and 3), suggesting preferential recognition of hybrid glycans [4, 40]. CH01 also preferred swainsonine and GNT3 PVs (Fig 3). Thus, GE here helps to minimize CH01 binding clashes by replacing complex glycans with smaller glycans like Man5GlcNAc2 (GNT1- PV) or hybrid glycans (swainsonine and GNT3 PVs). The natural scarcity of bisected glycans in previous reports [13, 19, 21] suggests that GNT3 is typically poorly expressed in 293T cells. In contrast, GNT1 plasmid co-transfection had little effect on V2 bnAbs—or indeed on other bnAbs (Figs 2A and 3, S2 Fig), suggesting that natural cellular levels of this enzyme are not limiting.
Gp120-gp41 interface bnAbs were heavily impacted by early GE, and in diverse ways (Fig 3). PGT151 potency was reduced against kifunensine, swainsonine and GNT1- PVs, consistent with the importance of tetra-antennary glycan contacts which are eliminated by these GE methods [4, 61, 78]. In stark contrast, kifunensine and GNT1- PVs were highly sensitive to 35O22, facilitating nearly 100% saturating neutralization and suggesting a preference for high mannose glycans. 35O22 was also modestly more potent against swainsonine PV but was less potent against the GNT3 PV. VRC34.01 was also more potent against GNT1- and kifunensine PVs (Figs 2A and 3), although in this case the GNT1- PV was the most sensitive, suggesting that smaller glycans reduce binding clashes. VRC34.01 sensitivity was also improved against the GNT3 PV, perhaps due to the replacement of complex glycans with smaller hybrid glycans (Fig 1). However, swainsonine, which also promotes hybrid glycans, had no effect, suggesting that fine differences in glycan structures are important. Finally, 3BC176 activity was increased against kifunensine, GNT1- and swainsonine PVs but not GNT3 PV. Thus, 35O22, VRC34.01 and 3BC176 activities were all improved by GE tools that reduce glycan size, albeit with unique patterns that reflect their different binding modes.
Perhaps unsurprisingly, V3-glycan bnAbs were generally only modestly affected by early GE, as they target the intrinsic mannose patch (S2 Fig). The same was true for CD4bs and MPER bnAbs, in this case because they generally target protein epitopes (S2 Fig). Nevertheless, almost all of these bnAbs were more potent against the GNT1- PV, suggesting that smaller glycans help to minimize binding clashes. PGT125 and PGT128 were even more potent against kifunensine PV, consistent with a preference for untrimmed high mannose glycan.
The poor neutralizing activities of CH01 and 35O22 under standard conditions were unexpected (Fig 3). We investigated whether this was assay-related by checking the activities of both mAbs in the CF2 (used throughout this study) and TZM-bl assays. Both mAbs incompletely neutralized the JR-FL PV at high mAb concentrations in both assays (S4 Fig), although the residual infectivity in the CF2 assay plateaued at higher levels. In contrast, VRC01 completely neutralized the PV in both assays, albeit with slightly different IC50s. Although our previous studies have shown these assays yield similar IC50s, modest differences may stem from the higher CCR5 surface density of CF2 cells [79].
Modifying intermediate glycosylation steps did not impact PV resistance to non-nAbs 14e and F105. The effects on bnAbs were generally modest (Figs 2A and 3). However, 2-deoxy-2-fluoro-l-fucose (2FF) (which inhibits core fucosylation; Fig 1) dramatically improved CH01 sensitivity, consistent with the increased GNT1- PV sensitivity (Figs 2A and 3), further underlining a preference for small glycans. The absence of core fucose may facilitate greater glycan flexibility, minimizing clashes. Perhaps for similar reasons, 2FF also markedly increased 35O22 sensitivity and, to a lesser extent, PGT128, b12 and VRC01 sensitivities, but decreased sensitivities to PGT145, PGT151 and 3BC176 (Figs 2A and 3, S2 Fig). As with early GE, mAbs directed to the V3-glycan supersite, CD4bs and MPER epitopes were largely unaffected by other intermediate GE methods (S2 Fig). However, fucosyltransferase 8 (FUCT8) co-transfection modestly decreased PGT128 sensitivity, mirroring the opposite effect of 2FF, suggesting that fucose causes a binding clash. The generally modest impact of FUCT8, GNT4 and GNT5 co-transfections suggests ample cellular levels of these enzymes are already expressed in the host cells [13].
Late GE had no effect on non-nAbs and only mild effects on 3 of the 4 V2 bnAbs (Figs 2A and 3). In contrast, β-1,4-galactosyltransferase 1 (B4GALT1) co-transfection increased PG9 potency by ~10-fold, suggesting that it helps promote the development of differentiated hybrid or complex glycan termini (Fig 1) [4, 39]. Conversely, inhibiting galactosylation with 2-deoxy-2-fluoro-d-galactose (2FG) had little impact on V2 bnAbs, except for a moderate decrease in PGT145 sensitivity (Figs 2A and 3).
Sialylation depends on effective galactosylation (Fig 1). Since the latter may be limiting, we promoted Env sialylation by co-transfecting various sialyltransferase plasmids together with B4GALT [13]. Co-transfection of β-galactoside α-2,6-sialyltransferase 1 (ST6GAL1) and B4GALT1 increased PG9 sensitivity ~30-fold (Figs 2A and 3). In stark contrast, co-transfection of β-galactoside α-2,3-sialyltransferase 4 (ST3GAL4) markedly reduced PG9 sensitivity (Fig 2A), suggesting a preference for α-2,6 SA termini [4, 36, 39, 43].
Overexpression of N-acetylneuraminide α-2,8-sialyltransferase 4 (ST8SIA4), which transfers additional SA moieties onto SA termini (Fig 1), also increased PG9 sensitivity (Figs 2A and 3). However, the increase was not as strong as with the B4GALT1-modification alone. Therefore, we suggest that ST8SIA4 (used with B4GALT1) in fact has a mild inhibitory effect. Finally, the removal of both α-2,3- and α-2,6-linked SA termini by NA reduced PG9 sensitivity by >5-fold, further emphasizing the role of SA contacts.
Gp120-gp41 interface bnAbs were dramatically affected by late GE (Fig 2A). Since PGT151 recognizes tri- and tetra-antennary glycans [3, 4, 61] and is adversely affected by kifunensine, swainsonine, GNT1- and 2FF [78], we were surprised that B4GALT1+/-ST6GAL1, ST3GAL4 or ST8SIA4 also decreased PGT151 sensitivity (Figs 2A and 3). 2FG had a mild inhibitory effect, suggesting galactose-dependency. Conversely NA had no impact, suggesting impartiality to SA termini [3, 4, 61].
In stark contrast, the same late GE methods dramatically increased 35O22's potency by >100-fold. This was paradoxical, considering the similar enhancing effects of kifunensine, GNT1- and 2FF (Figs 2A and 3). On the other hand, the slight loss of sensitivity with NA treatment is consistent with previous reports that 35O22 recognizes complex glycans (N88, N241 and N625 in the JR-FL strain) [4, 21, 23]. Later, we address the unexpected effects of late modifications on PGT151 and 35O22 sensitivities.
BnAbs VRC34 and 3BC176 were more modestly affected by late GE, although the spread of effects was somewhat wider for 3BC176, with B4GALT1+ST3GAL4-modified PV being the most sensitive (Figs 2A and 3). B4GALT1+/-ST6GAL1 and NA were found to increase PGT121 sensitivity (S2 Fig). Otherwise late GE generally only moderately affected V3 glycan, CD4bs and MPER bnAbs. However, NA increased 2F5 resistance, perhaps because removing negatively charged SA reduces electrostatic repulsion with the membrane, so that the trimer sits deeper in the membrane.
To contextualize the above findings, we extracted Env from GE-modified virus-like particles (VLPs) and ran them in blue native PAGE (BN-PAGE)-Western blot. Previously, we showed that Env resolves into two bands in BN-PAGE: trimers, largely consisting of functional gp120/gp41, and monomers, largely consisting of high-mannose uncleaved gp160ΔCT [81, 82] (Fig 4). GE caused some marked changes. Kifunensine, GNT1-, GNT3 and NA trimers all migrated relatively slowly (Fig 4, compare lanes 1, 2, 3, 5 and 14). This could be due to bulky untrimmed Man9GlcNAc2 glycans (kifunensine), the added mass of a bisecting glycan (GNT3) and, perhaps most importantly, the lack of negatively charged SAs which assist in trimer migration (kifunensine, GNT1- and NA). Conversely, B4GALT1+ST3GAL4/ST6GAL1/ST8SIA4 trimers all migrated faster than the control (Fig 4, compare lanes 1, 15–17), suggesting that additional, negatively charged SA improves trimer migration. GNT1-, kifunensine, 2FG and B4GALT1+ST6GAL1 trimers were all relatively poorly expressed (Fig 4, compare lanes 1, 2, 3, 12 and 16). The weaker expression GNT1- and kifunensine trimers may account for the low PV infectivities that we observed earlier (S1A Fig).
We also analyzed a concentrated stock of replicating full-length JR-FL virus grown in PBMCs using image enhancement to assist comparison with VLP Env (Fig 4, lanes 18 and 21). The PBMC trimer migrated slightly faster than the VLP trimer. This was unexpected, given that the VLP Env (gp160ΔCT) lacks 148 C-terminal amino acids, resulting in an expected reduction in trimer mass of ~48.9 kDa (~16.3 x 3). We suggest that the relatively fast mobility of PBMC trimer is driven by Env hypersialylation [19].
To further contextualize the neutralization analysis above, we analyzed GE-modified VLP gp120 and gp41 in SDS-PAGE-Western blots. Fine details can be found in S1 Text and Figs A and B contained therein. Briefly, the key findings were as follows: i) B4GALT1 increased Env endo H-sensitivity, consistent with the partial replacement of complex glycans with unfucosylated hybrid glycans. This may explain why B4GALT1 unexpectedly reduced PV sensitivity to multiantennary glycan-preferring bnAb PGT151 and increased sensitivity to the small glycan-preferring bnAb 35O22. ii) GNT3, swainsonine and 2FF also increased Env endo H-sensitivity, albeit less effectively than B4GALT1, suggesting that these methods also replace complex glycans with hybrid glycans, albeit less effectively than B4GALT1. iii) Kifunensine and GNT1- Env consisted of relatively homogeneous, endo H-sensitive high mannose glycans, as expected.
Given that B4GALT1+/-ST6GAL1 impacts PG9, 35O22 and PGT151 sensitivities (Figs 2A and 3) and promotes hybrid glycans (S1 Text), we wondered what effect ST6GAL1 alone might have. Unlike B4GALT1, ST6GAL1 alone did not affect gp120 or gp41 endo H sensitivity, suggesting that it does not promote hybrid glycans (S5A Fig). However, it nevertheless increased PG9 sensitivity to near equivalent levels as B4GALT1+ST6GAL1 PV (S5B Fig). This suggests that cellular galactosylation is sufficient for ST6GAL1 to attach α-2,6 SA termini. However, unlike the B4GALT1 treatments, ST6GAL1 alone did not increase 35O22 sensitivity and only modestly decreased PGT151 sensitivity (S5B Fig). As expected, VRC01 was unaffected. Referring to Fig 1, upon the formation of a hybrid intermediate in the medial Golgi, there is a bifurcation in the pathway, where glycans either mature into complex or hybrid glycoforms. We suggest that B4GALT1 overexpression diverts glycoprotein traffic in the latter direction, so that normally complex glycans (magenta in S5C Fig) are replaced by smaller hybrid glycans (orange in S5C Fig), thus "thinning" Env glycan coverage. Co-transfection of ST6GAL1 and B4GALT1 further improves PG9 sensitivity by promoting sialylation (yellow in S5C Fig). According to Fig 1, swainsonine, 2FF and GNT3 modifications might also divert traffic to the hybrid glycan branch. However, judging from gp41 endo H laddering patterns, B4GALT1 is more effective (S1 Text).
To further assess the effects of GE, we next analyzed the vaccine strain BG505 T332N gp160ΔCT (referred to as 'BG505' hereafter), focusing on the GE tools that markedly impacted JR-FL sensitivity: GNT3, 2FF, swainsonine, B4GALT1, ST3GAL4, ST6GAL1 and NA. Kifunensine and GNT1- PV infection was undetectable (S1B Fig). Several other treatments also reduced infection more significantly than they did for JR-FL (S1A and S1B Fig). The effects of GE on BG505 bnAb sensitivity are shown in Fig 2B and S6 Fig. As for JR-FL, BG505 resistance to 14e or F105 was unaffected by GE (S6 Fig). Also, as for JR-FL, PG9 potency was improved against B4GALT1+ST6GAL1 and swainsonine PVs but was reduced against NA PV. In contrast to JR-FL, however, B4GALT1 PV did not show increased PG9 sensitivity and B4GALT1+ST3GAL4 PV did not reduce PG9 sensitivity. Remarkably, B4GALT1+ST6GAL1 increased CAP256.09 sensitivity by ~30-fold and NA decreased sensitivity by a similar factor, consistent with evidence that CAP256.09 contacts terminal SA [4, 36, 40]. As for PG9, B4GALT1 alone had no effect on CAP256.09, whereas the B4GALT1+ST3GAL4 PV was moderately resistant. PGT145, CH01 and VRC38.01 were largely unaffected by GE, except that 2FF inhibited CH01 (S6 Fig), contrasting sharply with its impact on JR-FL sensitivity (Fig 3), suggesting strain-specific GE effects.
As for the JR-FL strain, BG505 sensitivities to V3-glycan and CD4bs bnAbs were largely unaffected by GE (Fig 2B, S6 Fig). However, VRC13 was a notable exception. Unlike most CD4bs mAbs, VRC13 binds to mono- and biantennary glycans bearing terminal galactose in glycan arrays [4]. Contrasting with the mild effects observed for JR-FL, VRC13 neutralization of the BG505 strain was markedly impacted by GE. NA PV was the most sensitive, followed by the B4GALT1 PV. Conversely, swainsonine and GNT3 PVs were less sensitive.
Unlike JR-FL, B4GALT1 overexpression did not markedly affect BG505 sensitivity to PGT151 or 35O22 (S6 Fig), mirroring the similar lack of impact of B4GALT1 on V2 mAbs (also unlike JR-FL). NA slightly enhanced PGT151 sensitivity (S6 Fig)—also not observed for JR-FL (Fig 2). In contrast, B4GALT1+ST6GAL1 mildly inhibited PGT151, suggesting a negative impact of terminal SA. Swainsonine and, to a lesser extent, GNT3, both reduced PGT151 sensitivity, consistent with the partial replacement of multi-antennary glycans with hybrid glycans [4].
None of the GE modifications greatly impacted VRC34.01 sensitivity (S6 Fig). However, 8ANC195 was hypersensitive to B4GALT1+ST6GAL1 PV and was more resistant to NA PV. This suggests α-2,6-linked SA dependency, as we observed for PG9 and CAP256.09. Finally, several GE modifications increased 2F5 sensitivity (S6 Fig) but had less impact on 4E10 and 10E8. Conversely, NA digestion mildly inhibited 2F5 and 4E10, suggesting that, as for the JR-FL strain, removing negatively charged SA moieties may allow the trimer to sit deeper in the membrane, partially obscuring the MPER.
We next examined whether GE-mediated changes in HIV+ plasma sensitivity matched those of mAbs isolated from the same donors. Donor CAP256 plasma neutralization is overwhelmingly mediated by V2 apex bnAbs of the CAP256 lineage [35]. Donor N152 plasma neutralization is largely mediated by the 10E8 bnAb lineage but is also the source of bnAb 35O22 that only modestly contributes to plasma neutralization [60]. B4GALT1+ST6GAL1 increased the sensitivity of BG505 PV to the CAP256 plasma by 100-fold, mirroring the 30-fold increase in CAP256.09 sensitivity (S7A Fig). However, a BG505 K169E mutant was resistant to both, consistent with the criticial role of the K169 contact (S7A Fig). In contrast, B4GALT1+ST6GAL1 modified JR-FL PV showed modestly improved sensitivity to the N152 plasma, as observed for 10E8, but unlike its dramatic effect on 35O22 sensitivity (S7B and S2 Figs and Fig 3). Thus, the effects of GE on HIV+ plasmas appear to track with predominant bnAb specificities.
To formally check that ST6GAL1-mediated increases in sensitivity to PG9, CAP256.09 and 8ANC195 were due to higher numbers of α-2,6 SA termini, B4GALT1+/-ST6GAL1 PVs were subsequently treated with NA and then re-tested for sensitivity to these bnAbs. NA effectively reversed the effects of B4GALT1+ST6GAL1, confirming the importance of SA for these mAbs (S8 Fig). Nevertheless, B4GALT1+NA and B4GALT1+ST6GAL1+NA PVs were not as sensitive as PV treated with NA alone.
In contrast, the effects of the same treatments on JR-FL sensitivity to 35O22 was permanent, i.e. it was not reversed by NA (S8A Fig), confirming that SA was not involved in the gain of 35O22 sensitivity with B4GALT1. PGT151 activity of B4GALT1+/-ST6GAL1 PV was also not fully recovered by NA, again suggesting a permanent effect, although there was a modest gain of sensitivity for B4GALT1 PV (S8A Fig). As noted above, B4GALT1 appeared to have a relatively modest effect on BG505 sensitivity to PGT151 (compare S8A and S8B Fig)—a point we return to below.
We next examined the effects of GE on two newly reported interface bnAbs ACS202 and CAP248-2B [65, 69]. Details are given in S2 Text and Figs C and D contained therein. In brief, both nAbs were only modestly affected by GE. B4GALT1 reduced ACS202 sensitivity slightly while slightly increasing CAP248-2B sensitivity, suggesting that smaller glycans may resolve clashes in the latter case. Overall, this is consistent with the diverse binding mechanisms of interface bnAbs, where some are markedly affected by GE in different ways (PGT151, 35O22, 8ANC195), whereas others are marginally affected (VRC34.01, ACS202 and CAP248-2B).
We also examined the effects of GE on PG9 and PGT145 developmental relatives. Details are given in S2 Text and Fig E contained therein). In brief, glycan proclivities appeared to be consistent within the mature branches of these lineages.
We next investigated the possibility that negatively charged poly-SA chains might further impact PV sensitivity to V2 apex and interface bnAbs. This may have been overlooked by our use of B4GALT1+ST8SIA4 above (Figs 2A and 3), as this lacked ST6GAL1 that might be needed to create α-2,6 SA termini as substrates for ST8SIA4 (Fig 1). Details are provided in S2 Text and Fig F contained therein. In brief, poly SA chains formed by B4GALT1+ST6GAL1+ST8SIA4 triple transfection did not accentuate the effects already observed with B4GALT1+ST6GAL1 or B4GALT1+ST8SIA4 component double treatments, suggesting that extra charge does not improve bnAb binding.
Given the differing impact of B4GALT1 on JR-FL and BG505 sensitivities to interface bnAbs 35O22 and PGT151 (Figs 2 and 3, S2 and S6 Figs), we wondered if B4GALT1-induced endo H-sensitivity observed for JR-FL might occur with B4GALT1 treatment of other strains (S1 Text, S5 Fig). Fourteen VLPs from various clades were produced with or without B4GALT1, then analyzed by SDS-PAGE-Western blot and probed for gp41. The gp41 bands of untreated VLPs were, in most cases, largely endo H-resistant, whereas their B4GALT1-modified counterparts were endo H-sensitive, suggesting that it inhibits gp41 glycan maturation (S9 Fig).
The sizes of gp41 bands from strains BG505, JR-FL, WITO and 16055 were relatively small, consistent with their truncated gp41 tails (gp160ΔCT) and staining was also relatively strong. Of these four strains, unexpectedly, BG505 and WITO gp41 bands were partially endo H-sensitive even without B4GALT1 modification (S9A Fig). The complex laddering in both cases suggested substantial glycodiversity. It is possible that the particularly high expression of the BG505 and WITO Env clones leads to these unusual band patterns. Overall, these findings raise the possibility that the milder effects of B4GALT1 on BG505 sensitivities to 35O22 and PGT151 as compared to JR-FL may be because hybrid gp41 glycans are already present in the BG505 strain, thus diluting the impact of B4GALT1 co-expression.
We next investigated the effect of GE on the sensitivities of the same 14 strains (Fig 5). Here, the geometric mean IC50 of each bnAb against all PVs was plotted to the right of each graph. Overlapping dots in Fig 5 are resolved in S1 Table, where IC50s are shown in a heat map. Wilcoxon Signed Rank tests were performed using two columns of data, in which the IC50s of a given mAb against control and GE PVs were paired for each strain (S1 Table). Representative mAb titrations are shown in S10 Fig. Env sequences of these and other strains used in this study are shown in S11 Fig. BnAbs were categorized into 4 groups, as follows:
In all but one case (CM244.ec1), B4GALT1+ST6GAL1 PVs were more PG9-sensitive. 10 B4GALT1+ST6GAL1 PVs were also more sensitive to CAP256.09. Remarkably, the otherwise CAP256.09-resistant Q23.17 strain was rendered highly sensitive (S10A Fig, S1 Table). Four other PVs: JR-FL, JR-CSF, WITO and REJO were resistant to CAP256.09, due to their lack of the key 169K contact (S11 Fig). B4GALT1 and ST6GAL1 treatments alone also generally increased PV sensitivities to these mAbs. B4GALT1 was, in many cases, less effective for CAP256.09 than for PG9, perhaps because it promotes hybrid glycans that are better tolerated by PG9 [4]. In contrast, NA generally reduced sensitivity to these bnAbs to varying extents. Overall, the improved sensitivity imparted by α-2,6 SA-modification is conserved across multiple clades, and we classify these bnAbs as α-2,6 SA-dependent.
PGT151 sensitivity was consistently, albeit usually modestly, improved by NA, suggesting a conserved preference for terminal galactose [4, 61]. This was particularly marked for the 16055 strain (Fig 5, S10B Fig). Interestingly, GNT1- and B4GALT1 both consistently reduced PGT151 sensitivity, although the extent varied considerably between strains. GE also affected neutralization saturation. Thus, GNT1- modification of the JR-CSF strain completely eliminated PGT151 sensitivity, whereas B4GALT1+/-ST6GAL1 reduced saturation to ~50% (S10B Fig).
For 8 viruses, PGT151 sensitivities of B4GALT1 PVs were increased by subsequent NA treatment (compare B4GALT1 and B4GALT1+NA in Fig 5). However, in all cases except BG505 and JR-CSF, IC50s did not reach the sensitivity achieved by NA alone. Thus, in most cases, B4GALT1 reduces PGT151 sensitivity in a manner that is not fully recoverable by subsequent NA treatment.
PGT121 was also effective against NA PVs and most B4GALT1+NA PVs. In some cases, resistant (or nearly resistant) strains became sensitive with these (and other) treatments (REJO, CNE58, 16055 and KER2018.11), thus increasing PGT121 breadth (Fig 5, S1 Table and S10C Fig). Three of these 4 strains (CNE58 excepted) lack-the canonical N332 glycan, but two have a glycan at position N334 instead (16055 excepted), raising the possibility that GE can help compensate and restore PGT121 binding (S11 Fig). Overall, these findings suggest that PGT121 prefers galactose termini, consistent with glycan array data [4].
High concentrations of 35O22 did not quite reach an IC50 for all 14 unmodified strains (Fig 5). However, activity was dramatically and consistently increased in 12 GNT1- PVs tested (GNT1- modified BG505 and CNE58 PVs were poorly infectious and were therefore omitted; Fig 5, S1 Table). For 9 strains, B4GALT1 PV also increased 35O22 sensitivity. Notably, B4GALT1+/-ST6GAL1 improved 35O22 saturation of JR-CSF and KER2018.11 PVs, and GNT1- led to a further increase (S10D Fig). This pattern was the exact reverse of that observed for PGT151 (S10B Fig). Since 35O22 binds to complex and high mannose glycans in arrays, we suggest that its greater activity against GNT1- PVs might be due to improved glycan core binding [4]. Overall, GE consistently improved 35O22 neutralization, in part by improving saturation—a point that we return to later below.
Consistent with earlier observations (Fig 3 and S6 Fig), 5 of 5 CH01-sensitive strains became more sensitive with GNT1- modifications. GNT1- modification uncovered CH01 sensitivity in another 5 otherwise resistant strains (Fig 5). In keeping with the lack of CH01 binding to glycan arrays, it appears that smaller Man5GlcNAc2 glycans consistently minimize clashes. However, two strains (REJO and 16055) remained resistant upon GNT1- modification. Since key CH01 contacts (N156 and N160 glycans and K171) [42] are present, this resistance may be due to remaining glycan clashes.
Contrasting with the largely consistent and, in some cases, highly significant (S1 Table) patterns above, the most sensitive GE variant for some bnAbs differed between strains. PGT145 potently neutralized many GNT1- PVs but was less effective on Q23.17. NA-treated PVs were also largely sensitive, except for ZM233.6. However, GNT1- and B4GALT1 versions of ZM233.6 were highly PGT145-sensitive (Fig 5; S1 Table). V2 bnAbs are subject to potential clashes with glycans at position N130 and in the V2' region (residues 183–191, stippled pattern in S11 Fig) [45]. None of the 14 strains used in Fig 5 have a N130 glycan, although BG505, BI369.9A and ZM233.6 each have two sequons in the V2' region, while the other strains have one or none (S11 Fig). The dramatically higher PGT145 sensitivities of GNT1- and B4GALT1-modified BI369.9A and ZM233.6 PVs may be because glycan clashes are eliminated. Unmodified BG505 is PGT145-sensitive without GE modifications, suggesting that the V2' glycans of this strain do not limit PGT145 access. The outlier status of the ZM233.6 strain may be related to its unique lack (among these strains) of the N156 glycan (S11 Fig). Overall, PGT145 appears to be glycan-averse and subject to clashes, consistent with its lack of glycan array binding.
GE also had variable effects on VRC13. Half of the GNT1- PVs tested were sensitive. NA also improved sensitivity in most cases. The glycan fence surrounding the primary receptor site [57], typically consists of 4 to 7 glycans, including variable glycans of the V5 loop (S11 Fig). Although VRC13 may clash with some of these glycans, it may contact others. Therefore, since GE may eliminate clashes or modulate mAb-glycan contacts, it is difficult to unequivocally interpret these patterns.
GE also variably affected VRC38.01. Given the lack of glycan array binding, this may be due to glycan clashes. Thus, GNT1- PVs of the REJO and 16055 strains were far more sensitive than control PVs, while the sensitivities of other strains were less affected, and in one case (T250-4), was lower (Fig 5, S1 Table and S10E Fig). The effects of B4GALT1+ST6GAL1 also varied. As mentioned above, all the strains in this panel lack the N130 glycan that clashes with VRC38.01 binding. These strains also exhibit most if not all known VRC38.01 contacts (S11 Fig) [42]. The resistance of the CH070 and ZM233.6 strains may be related to their lack of a tyrosine at position 173 that may help orient the N156 glycan for binding (both strains) and/or the absence of the N156 glycan (ZM233.6 strain; S11 Fig).
The activities of other, less broadly neutralizing nAbs were also examined against these 14 strains. Although HJ16 neutralized the 16055 PV and was unaffected by GE [21, 23], it remarkably enhanced infection by Q23.17 and KER2018.11 PVs in various formats (S10F Fig). However, this enhancement was reduced (Q23.17) or eliminated (KER2018.11) against GNT1- versions of these PVs (S10F Fig), implying that HJ16 can activate infection by some strains when they carry complex glycans.
WITO sensitivity to 8ANC195 was knocked in by B4GALT1+ST6GAL1, thereby increasing its breadth (S10G Fig). This effect was partially reversed by NA, as above with the BG505 strain (S8 Fig). Thus, although WITO strain bears the N230 glycan thought to clash with this mAb and lacks the N234 glycan thought to be important for binding, B4GALT1+ST6GAL1 modification was sufficient to allow this mAb to neutralize.
In many cases, GE increased the percent maximum neutralization by excess nAb. Perhaps the best examples are 35O22 and CH01 (Fig 2). Using neutralization data from the 14-virus panel (Fig 5), we plotted the % of control and GE-modified PVs neutralized to >65%, >90% and >95% saturation by excess bnAb (10μg/ml). GE dramatically improved saturation by PG9, CAP256.09, 35O22 and CH01 (Fig 6). Notably, PG9 neutralized all B4GALT1+ST6GAL1 PVs to >95% saturation. Although the effects of GE on PGT151 and PGT121 were relatively modest, there was also a positive trend. For PGT145, there was no clear increase in saturation against GNT1- PVs, consistent with the variable effects noted in Fig 5. Overall, despite the small numbers of viruses analyzed here, we infer that, in many cases, GE improves neutralization saturation, in some cases dramatically, either by eliminating glycan clashes and/or by creating optimal glycan structures for optimal binding.
Above, we found that certain bnAbs were more effective against B4GALT1+ST6GAL1 PVs. In primary cells, HIV-1 Env is thought to be naturally modified by terminal α-2,6 SA (contrasting α-2,3 SA, as common for 293T cells) [19]. Furthermore, the relatively fast mobility of PBMC-derived trimers in BN-PAGE (Fig 4) suggest possible hypersialylation. We wondered if these factors might make PBMC virus unusually sensitive to these bnAbs. We therefore investigated the change in IC50 of PVs upon B4GALT1+ST6GAL1-modification and the change in IC50 of 293T cell-produced infectious molecular clones (IMCs) upon PBMC passage.
B4GALT1+ST6GAL1-modification of 45_01DG5 and T278-50 PVs rendered them highly sensitive to the α-2,6 SA-dependent bnAb CAP256.25 (Fig 7). Notably, the 45_01DG5 strain bears a methionine at position 165 of the V2 loop C strand (S11 Fig), suggesting that improved glycan contacts may compensate for suboptimal protein contacts. PBMC passage also increased the IMC sensitivity of both strains to CAP256.25 (Fig 7). B4GALT1+ST6GAL1-modification and PBMC passage also improved YU2 sensitivity to 8ANC195. However, ADA sensitivity was unaffected. The sensitivities of 45_01DG5, T278-50 and JR-CSF to PG9 were also increased by B4GALT1+ST6GAL1 modification and PBMC passage. These gains were similar in magnitude to those observed with 8ANC195 on YU2 but less than those with CAP256.25. Although B4GALT1+ST6GAL1-modification marginally improved YU2 and ADA sensitivities to PG9, there was no clear effect of PBMC passage on IMC sensitivity.
In stark contrast to most of the above findings with α-2,6 SA-reactive bnAbs, PGT145 and VRC01 IMC sensitivities were in no case improved by PBMC passage. In fact, sensitivities to VRC01 were reduced in all cases and T278-50 sensitivity was completely knocked out for both bnAbs. Consistent with these findings, B4GALT1+ST6GAL1-modification also did not enhance sensitivity to these mAbs. Overall, these observations suggest that PBMC passage modifies IMCs with α-2,6 SA in a similar way that B4GALT1+ST6GAL1 modifies PV, increasing sensitivities to bnAbs that depend on these hypersialylated termini. However, the sensitivity of PBMC-grown IMCs was, in most cases, lower than that of B4GALT1+ST6GAL1-modified PV.
Given the key role of glycan contacts for CAP256.09 and PG9, we next investigated their role in sensitivity to germline-reverted versions of these bnAbs. Autologous PV from donor CAP256 sampled at 34 weeks was sensitive to an unmutated common ancestor (UCA) and I1 intermediate CAP256 bnAb and the mature .09 and .25 clones (Fig 8A). However, the effects of GE changed as these Abs matured. The UCA preferentially neutralized the GNT1- and B4GALT1+ST6GAL1 PVs, whereas the I1 intermediate was indifferent to GE and both mature clones showed a strong preference for the B4GALT1+ST6GAL1 PV and weaker neutralization of the GNT1- PV. Together, this suggests that α-2,6 SA binding is not germline-encoded but develops during ontogeny. The stronger GNT1- PV sensitivity to the UCA suggests a benefit of reducing glycan clashes. The higher sensitivity of the B4GALT1+ST6GAL1 PV to the UCA is consistent with B4GALT1-induced glycan "thinning" (replacing complex glycans with hybrid glycans) that may also reduce clashes.
To approximate the PG9 ancestor, we used a germline-reverted heavy chain (gH) co-expressed with the mature light chain (mL) [34, 42]. Since autologous viruses were unavailable, we examined heterologous tier 2 strains. GE did not affect the activity of the revertant against the 16055 PV (Fig 8B). However, Q23.17 GNT1- PV and, to a lesser extent, B4GALT1+ST6GAL1-modified Q23.17 PV were relatively sensitive (Fig 8B). In contrast, the B4GALT1+ST6GAL1 PV of both strains was most sensitive to mature PG9. Overall, this suggests that the SA binding is not a feature of CAP256 or PG9 mAb ancestors, where smaller glycans may be more important to minimize clashes, at least in some settings.
To further investigate the impact of GE on the activities of bnAb ancestors, we used a panel of 9 V2 nAb-sensitive strains [34, 36, 40, 42, 45] that all lack the N130 glycan and have short, sparsely glycosylated V2' regions (S11 Fig). Additional data for the JR-FL strain is shown in S12 Fig. As we observed above (Fig 5), CAP256.09 preferentially neutralized B4GALT1+ST6GAL1 PVs, but the UCA and I1 variants preferentially neutralized the GNT1- and B4GALT1+ST6GAL1-modified autologous PVs (Fig 8A and 8C, S2 Table). The I1 intermediate also neutralized GNT1- modified 16055 strain and (marginally) the T250-4 strain, but no others. Similarly, the PG9 revertant did not exhibit the B4GALT1+ST6GAL1 preference of its mature counterpart (Fig 8B and 8C). For some strains, GNT1- PVs were slightly more sensitive, although this difference was not statistically significant when all 9 strains were considered (Fig 8C, S2 Table).
The high sensitivities of many GNT1- PVs to mature CH04 (a clonal variant of CH01) were mirrored by high sensitivities of GNT1- PVs of some strains to its UCA (WITO, KER2018 and Q23.17; Fig 8C). The effects of GNT1- on PGT145 mHgL revertant sensitivity were mixed, but generally matched those of mature PGT145. For example, the revertant neutralized the KER2018.11 and JR-FL GNT1- PVs more effectively, as did its mature counterpart, although revertant neutralization did not reach an IC50 titer against KER2018.11 (S12A Fig).
KER2018.11, C1080, Q23.17 and JR-FL GNT1- PVs were more sensitive to the VRC38.01 revertant (Fig 8C, S12B Fig). Similarly, a mixed chain VRC13 mHgL revertant neutralized the JR-FL GNT1- PV more effectively, as did its mature counterpart (S12C Fig). Overall, the sensitivities of GNT1- PVs to PGT145, VRC38.01 and VRC13 varied between strains, generally in concert with their mature counterparts, suggesting that strain-specific clashes are important throughout their development and may, in some cases, be alleviated by PV production in GNT1- cells.
Finally, we examined the effects of GE on JR-FL sensitivity to a somatic ancestor and a revertant of PGT121. Neutralization by the 3H3L ancestor derived from deep sequencing [83] was not appreciably enhanced by NA, unlike mature PGT121 (S12D Fig). Moreover, it was most effective against the GNT1- PV, contrasting sharply with mature PGT121. Reverted forms of PGT121, including CDR3mat [84, 85] all failed to neutralize (S12D Fig).
The increased sensitivity of GNT1- trimers to germline-reverted bnAbs raises the possibility that they could be useful as vaccine priming immunogens. To be effective in a vaccine context, it may be important to ascertain that GNT1- trimers are compact and V3-resistant like their unmodified counterparts. This would assuage any concerns of 'off target' responses that could drain focus from desired targets. We investigated this question using HIV-1+ donor plasmas. The N90 plasma (source of the VRC38 nAb lineage [42]) neutralized GNT1- JR-FL PV ~30-fold more effectively than the control (Fig 9A). The weakly neutralizing 1648 plasma exhibited a similar increase [86] (Fig 9B). In contrast, the CAP256 plasma neutralized the T250-4 control and GNT1- PVs equivalently [35] (Fig 9C). These differences could reflect the differing glycan proclivities of the bnAbs they contain. However, the high potencies of the N90 and 1648 plasmas against GNT1- PV could also be due to the increased sensitivity of the modified PVs to V3-directed non-nAbs. To investigate, we used peptides to adsorb V3 non-nAbs and a JR-FL A328G mutant which is known to have an overtly V3-sensitive tier 1 phenotype as a reference [79]. V3 mAb 14e neutralized the GNT1- PV somewhat more effectively than the control, whereas the A328G tier 1 mutant was highly sensitive (Fig 9D). Non-nAb F105 (CD4bs) also potently neutralized the A328G mutant, but not the GNT1- PV (Fig 9E). The A328G mutant was also sensitive to the N90 plasma, but slightly less so than the GNT1- PV (Fig 9F). Conversely, the A328G PV was more sensitive to the 1648 plasma than the GNT1- PV, suggesting that V3 neutralization dominates when tier 2 nAb titers are weak (Fig 9G).
Added V3 peptides fully adsorbed 14e but not F105 activity, as expected (Fig 9D and 9E). They also adsorbed A328G neutralization by the N90 and 1648 plasmas. In stark contrast, however, the sensitivities of GNT1- and control PVs were both unaffected (Fig 9F and 9G). This is important, as it implies that the increased sensitivity of GNT1- modified JR-FL PV to the N90 plasma is largely due to increased sensitivity to tier 2 bnAbs, rather than to increased sensitivity to V3 non-nAbs. Quantitatively, if the IC50 of 14e against GNT1- PV of ~10μg/ml (Fig 9D) and infected plasmas contain an estimated ~100–1000μg/ml total of anti-gp120 Abs, the maximum plasma ID50 of 14e-like Abs against the GNT1- PV could be 1:100. This suggests that the increased N90 plasma ID50 (~1:10,000) against the GNT1- PV (Fig 8A) is not due to V3 non-nAbs, as also confirmed by the lack of effect of interfering V3 peptide. In contrast, the ~5-fold reduced A328G mutant sensitivity to the N90 plasma with added V3 peptides suggests that V3 Abs contribute a sizeable fraction of the activity against this tier 1 virus (Fig 9F). Overall, these findings suggest that GNT1- trimers retain a largely V3-resistant, compact tier 2 conformation but can be more sensitive to tier 2 bnAbs and their revertants, making them attractive for use in vaccine priming.
Here we sought to better understand the glycan structures recognized by bnAbs and how this glycoreactivity evolves and might be applied to vaccine design. Our findings suggest that bnAb precursors initially avoid glycans. UCA binding to native trimers may be facilitated when glycan sequons are unusually absent or skipped over (i.e. "glycan holes") [79, 87, 88] or when glycan maturation is stunted, minimizing clashes. Efforts to understand the early events in bnAb development are complicated by several factors. First, in many cases, "UCA-triggering" Env strains are unknown, as are the glycans they carry and, indeed, their form (e.g. gp120/gp41 trimer, gp160 or gp120 monomer). Second, many bnAb ancestors are mere approximations [34], raising questions about how well they represent the behavior of genuine ancestors. Third, UCAs typically exhibit little, if any, neutralizing activity against the triggering viruses, suggesting that more sensitive assays may be needed. In vitro trimer binding assays may be more sensitive and informative. In a recent study, the PCT64 UCA did not neutralize autologous virus, but weakly bound to an autologous GNT1- Env trimer, supporting the preference of bnAb ancestors for minimally glycosylated Env [41]. Similarly, CH04 UCA binding to SOSIP trimers was detected when neutralization was not [34]. Accordingly, we are now investigating the binding of bnAb ancestors to GE-modified VLPs by ELISA. Even these assays may be too stringent: evidence of inferred bnAb ancestor triggering even when they fail to detectably bind to the antigen in vitro attests to the exquisite sensitivity of the earliest stages of bnAb maturation [84, 85, 88].
During maturation, some nAbs acquire an ability to bind glycans. Indeed, the paratope electropositivity of VRC01 class and some V2 bnAbs increases with maturation [34, 35, 42, 89], ostensibly to facilitate interactions with heavily glycosylated HIV-1 Env spikes that often bear negatively charged SA termini. For VRC01-class bnAbs, this may help accommodate the N276 glycan [21]. For CAP256 bnAbs, α-2,6 hypersialylated trimers may promote the development of SA contacts. In contrast, for some nAbs such as CH01 and 35O22, the preference for small glycans does not change and clashes apparently remain unresolved. PGT151 and PGT121 both preferred NA-treated PV. Electrostatic incompatibility with SA does not explain this behavior, as these mAbs recognize complex glycans with galactose termini [4, 37, 48, 50, 54] and do not preferentially neutralize GNT1- PV.
Overall, there was a remarkable agreement between bnAb GE preferences and their reactivity in glycan arrays [4], with some exceptions: despite the SA-dependency of 8ANC195 observed here, this nAb did not bind in glycan arrays [4]. Similarly, although 35O22 bound complex glycans with SA antennae in arrays, it nevertheless neutralized GNT1- PV highly effectively, perhaps suggesting contact with glycan cores.
Our observation that PG9, CAP256.09 and 8ANC195 "punched above their weight" against PBMC-passaged virus is consistent with an earlier report in which PG9 outperformed b12, 4E10, 2F5 and VRC01 [90]. This, coupled with fast mobility of PBMC trimers in BN-PAGE and similar increased potency of these bnAbs against B4GALT1+ST6GAL1-modified PV all lead to the same conclusion that these bnAbs are more effective against α-2,6 hypersialylated trimers. This has two consequences for vaccine development. First, that these bnAbs may be particularly effective in a clinical setting. Second, that α-2,6 hypersialylated trimers may be ideal for boosting as they better match PBMC virus and may promote the development of α-2,6 SA-dependent bnAbs or else provide a "closer to real life" glycosylation profile to enable other bnAbs to navigate past them.
Another explanation for the outperformance of these mAbs against PBMC-passaged virus could be increased V2 tyrosine sulfation at positions Y173 and Y177, which may increase sensitivity to trimer-preferring bnAbs [91]. However, the lack of increased PGT145 potency with PBMC-passaging suggests that α-2,6 SA modification has a more decisive role in regulating bnAb sensitivity. Sulfation may, however, contribute to the generally higher nAb resistance of PBMC viruses [90, 92, 93]. Indeed, the fast migration of PBMC trimers in BN-PAGE could be due to increased sialylation and/or sulfation.
Our findings raise the question of what factors are important for bnAb development in natural infection. Clearly, the answer may be multi-factorial and may include virus sequence diversity and host antibody repertoire differences. Host-encoded glycovariation may also be a factor. Thus, it may be that the donors who developed CAP256 and PG9 lineages impart α-2,6 SA termini with relatively high efficiency. Conversely, PGT121 and PGT151 may have developed in donors where sialylation was inefficient. 35O22 and CH01 may have developed against viral strains with glycan holes that eliminate clashes or else the donor glycosylation machinery might naturally express Env trimers bearing smaller glycans. Indeed, PBMC-based HIV-1 neutralization assays are notoriously subject to significant inter-donor variability that could in part stem from host glycosylation differences that could toggle bnAb epitope exposure on the Env trimers they express.
Our key messages are summarized in Fig 10. GE was able to increase bnAb potency, saturation and breadth (Fig 10A), revealing the most sensitive glycoforms for prototype bnAbs (Fig 10B) and suggesting new prime-boost vaccine strategies (Fig 10C). The diffuse gp120 bands observed in Western blots suggest a possible swarm of glycovariants [21, 23]. BnAbs may not be able to neutralize all these variants, providing an avenue for virus 'escape' without mutation. However, GE could resolve this and improve bnAb saturation, either by optimizing glycan contacts and/or by eliminating clashes. Remarkably, several otherwise bnAb-resistant strains were rendered sensitive by GE, uncovering previously unappreciated breadth. Quantifying the extent of this increased breadth will require an analysis of larger numbers of GE-modified PVs.
Our findings provide new opportunities for affinity-based prime-boost native trimer vaccine strategies [84, 85] (Fig 10C). Thus, priming immunogens may be derived from hypersensitive strains that are i) modified to introduce glycan holes at the desired target, ii) mutated to maximize nAb affinity, iii) GE-modified to eliminate glycan clashes and iv) modified to plug glycan holes at unwanted sites. For example, VRC01 priming immunogens might ideally lack the N276 and N463 glycans [7, 28, 94], and be GNT1- modified to further reduce clashes, permitting different angles of approach and improving electrostatic compatibility. In support of GNT1- trimers as priming immunogens, we found that they are recognized poorly by non-nAbs–implying that off-target non-neutralizing responses should be limited. GNT1- modified Q23.17 trimers were hypersensitive to PG9, CH04 and VRC38.01 germline-revertants, raising particular interest in this strain for vaccine development. Although concerns have been raised over its tier 1B classification [95], its resistance to V3 non-nAbs and V2 bnAb hypersensitivity suggest "closed" trimers that may benefit from a naturally short and unglycosylated V2’ region [45, 88].
Intermediate boosting immunogens (Fig 10C) could be the most sensitive GE-modified glycoform for the target bnAb, to encourage the development of glycan contacts. Thus, for example, NA-treated trimers could promote the maturation of PGT121 and PGT151-like bnAbs. Previous work showed that desialylation improves gp120 recognition by some ligands and also improves immunogenicity by reducing electrostatic surface potential [96]. Desialylation may also modify antigen capture and presentation in ways that might promote B cell responses [96]. As mentioned above, the unexpected "glycan thinning" effect of B4GALT1 overexpression (Fig 10B) may be useful for intermediate boosting, to both minimize glycan clashes and also promote SA-reactivity (Fig 10C).
Mass spectrometry analysis of Env glycans may in future allow us to better understand effects of B4GALT1 and other GE-modifications, providing a clearer basis for their effects and their utility in vaccines. In the current study, we were limited to Western blot analyses both by the relatively poor Env yield, and by the co-presence of immature and unprocessed forms of Env that may contaminate trimer glycoanalysis. However, recent improvements in trimer expression and purification methods should facilitate these analyses in the near future [23, 97].
As final boosts (Fig 10C), α-2,6 hypersialylated trimers may be ideal for reasons already mentioned above. The use of CHO or 293T cells for HIV-1 vaccine production may be limited by their tendency to attach α-2,3 SA termini. To address this problem, trimers might be either produced in these cells along with co-expressed ST6GAL1 or could be enzymatically modified after expression to exchange α-2,3 SA with α-2,6 SA termini.
The common development of V2 bnAbs in natural infection, usually with relatively little somatic hypermutation [35, 36, 98] is increasing interest in this site as a vaccine target. Indeed, V2-hypersensitive strain trimers [34, 36, 42] were recently shown to elicit V2 nAbs [45, 88]. Although the long CDRH3 loops of PG9 and CAP256.09 set a high bar for their use as vaccine blueprints, their relatively high potency in PBMC assays raises their significance, especially if other SA-dependent bnAbs with more common-in-repertoire features can be found. By comparison, V2 bnAbs like CH01, VRC38.01, BG1 and PCT64-35S have shorter CDRH3 loops. VRC38.01 may be of particular interest because it is only moderately impacted by glycan variation, thus blocking one avenue of viral escape [39, 43].
In future, GE-modified baits may help to recover additional bnAbs from HIV-1-infected donors to inform vaccine design. GE also provides a way to investigate the effects of glycovariation on the tier 2 nAbs now increasingly being elicited by leading candidate vaccines [57]. This may enable us to develop prime-boost vaccination strategies to elicit broadly protective nAbs.
All human plasmas were archived (i.e. they were not drawn for this project). Normal (uninfected) human PBMCs with no donor identifiers, sourced from Duke University and NIH blood bank, were used to propagate replicating JR-FL and IMC viruses. Institutional Review Board (IRB) approval for this project was obtained through the San Diego Biomedical Research Institute IRB Committee (approval number: IRB-14-04-JB; Federal Wide Assurance number: 00021327).
MAbs were obtained from their producers and the NIH AIDS Reagent Repository. MAbs included the following (originators given in parentheses): 19b, 39F, F2A3, C011 and 14e (J. Robinson), directed to the gp120 V3 loop [99]; b12 (D. Burton), VRC01 and VRC13 (J. Mascola), 8ANC131 (M. Nussenzweig), HJ16 (A. Lanzavecchia), F105 (M. Posner), directed to epitopes that overlap the CD4bs [67, 99–101]; PGT121, PGT125 and PGT128 (D. Burton) directed to epitopes involving the base of the V3 loop of gp120 and the N332 glycan [37]; VRC38.01, CAP256.09 and CAP256.25 (J. Mascola), PG9, PG16, PGT145 and PGDM1400 (D. Burton), CH01 and CH04 (B. Haynes), directed to V2 apex epitopes [31, 33, 35, 37, 38, 42, 47, 77]; PGT151 (D. Burton), 35O22 (M. Connors), VRC34.01 (J. Mascola), 8ANC195 and 3BC176 (M. Nussenzweig), ACS202 (R. Sanders) and CAP248-2B (P. Moore) directed to the gp120-gp41 interface [60–62, 64, 65, 68, 69]; 4E10 and 2F5 (H. Katinger) and 10E8 (M. Connors), directed to the gp41 MPER [72]. Information on these mAbs can be found at the web link: (www.hiv.lanl.gov). Germline revertants and ancestors were also obtained for mAb lineages CAP256 [35], PG9 [34], PGT145 [34] VRC38 [42], CH04 [33], VRC13 [102] and PGT121 [83–85]. In some cases, variable and J segments were reverted to inferred germline residues, leaving the CDR3 intact. In other cases, UCAs were inferred from nAb ancestors recovered from donors; other ancestors were recovered by deep sequencing.
Plasmid pCAGGS was used to express JR-FL gp160ΔCT on VLP surfaces [99]. Gp160ΔCT is truncated at amino acid 709, leaving a 3 amino acid gp41 cytoplasmic tail. This increases native trimer expression and can be used to produce PVs with similar neutralization sensitivity profiles compared to their full-length gp160 counterparts [99]. Mutants were generated by QuikChange (Agilent Technologies) and were numbered according to the HXB2 reference strain [86]. "SOS" mutations (A501C and T605C) introduce an intermolecular disulfide bond between gp120 and gp41 [99]. The E168K mutation knocks in PG9 epitope and increases trimer expression [99], and the N189A mutation removes a sequon that is competitive with N188, and improves sigmoidal neutralization of V2-targeting nAbs such as PG9.
Plasmids expressing other Env gp160s were obtained from the NIH AIDS repository. Env-deficient sub-genomic plasmid pNL4-3.Luc.R-E- [99], pMuLV Gag (expresses endogenous murine leukemia virus Gag, driven by a CMV promoter) and pMV-Rev 0932 (expresses codon-optimized HIV-1 Rev, driven by a CMV promoter).
We also investigated a series of other Envs, some of which were full-length and others had truncated cytoplasmic tails (gp160ΔCT), as follows. In the following list, clade assignments are given in parentheses. BG505 T332N gp160ΔCT (A), KER2018.11 gp160 (A), BI369.9A gp160 (A), Q23.17 gp160 (A), CM244.ec1 gp160 (AE), T250-4 gp160 (AG), JR-CSF gp160 (B), WITO 4160.33 gp160ΔCT (B), REJO4541 gp160 (B), CH070.1 gp160 (BC), ZM233.6 gp160 (C), CNE58 gp160 (C), 16055–2 clone 3 gp160ΔCT (C), T278-50 gp160 (AG), and 45_01DG5 gp160 (B).
Glycosyltransferase plasmids pEE6.4_B4GALT1 (expresses β-1,4 galactosyltransferase 1), pEE14.4_ST6GAL1 (expresses β-galactoside α-2,6-sialyltransferase 1) and pEE6.4_GNT3 (expresses N-acetylglucosaminyltransferase 3) were reported previously [13]. Others were obtained from the DNASU repository: β-1,3-N-acetylglucosaminyltransferases 1–5 (GNT1, 2, 4 and 5), α-1,6-fucosyltransferase (FUCT8), β-galactoside α-2,3-sialyltransferase 4 (ST3GAL4) or α-N-acetyl-neuraminide α-2,8-sialyltransferase 4 (ST8SIA4).
Glycosyltransferase plasmids were co-transfected at a ratio of 1% total transfected DNA. The exception to this was B4GALT1+sialyltransferase (ST3GAL4, ST6GAL1 or ST8SIA4) co-expression, where B4GALT1 was transfected at 1% and the sialyltransferase at 2.5% total transfected DNA. For increasing galactosylation and increasing sialylation, 5 mM D-galactose (Sigma) was added to the medium prior, during and post-transfection in conjunction with co-transfection of B4GALT1. Decoy substrates for blocking fucosylation, 2-deoxy-2-fluoro-l-fucose (2FF) (Carbosynth) and galactosylation, 2-deoxy-2-fluoro-d-galactose (2FG) (Carbosynth), were added 4h post transfection at 0.4 mM and 1mM, respectively. To block the action of mannosidase 1, kifunensine was added at 25 μM during and post transfection. Swainsonine was added at 20 μM during and post transfection to block mannosidase 2 activity. Terminal SAs were cleaved by incubating 25 μl of 1,000x concentrated PV with 2 μl of NA from C. perfringens (Sigma, Cat# N5631, 5u/ml) and 1 μl of NA from A. ureafaciens (Roche, Cat # 10269611001, 10U/ml)) for 1 h at 37°C. Both NAs cleave α-2,3, α-2,6 and α-2,8 SA, but C. perfringens NA preferentially cleaves α-2,3 SA, whereas A. ureafaciens NA preferentially cleaves α-2,6 SA. PV was then washed with PBS, pelleted and resuspended at 1000 x.
VLPs were produced by co-transfecting human embryonic kidney 293T cells (ATCC) with an Env-expressing plasmid (typically pCAGGS JR-FL gp160ΔCT SOS E168K and mutants thereof), pMuLV Gag and pMV-Rev 0932 using polyethyleneimine (PEI Max, Polysciences, Inc.), as described previously [99]. Two days later, supernatants were collected, precleared by low speed centrifugation, filtered, and pelleted at 50,000 x g in a Sorvall SS34 rotor. To remove residual medium, VLP pellets were diluted with 1ml of PBS, then re-centrifuged at 15,000 rpm and resuspended in PBS at 1,000 x the original concentration.
Replicating JR-FL virus was propagated from a stock provided by the NIH AIDS Reagent Repository (CAT#395, donated by Dr Irvin Chen), by cell-free infection of uninfected human peripheral blood mononucleocytes (PBMCs) activated in RPMI medium containing 20% FBS, 50μg/ml gentamycin, 5μg/ml PHA-P (Sigma Cat# L1668) and 5% IL-2 for 12–24 hours followed by washing and resuspension in RPMI supplemented with only 20% FBS and 5% IL-2 for infection. Virus supernatant was inactivated using 1mM aldrithiol and then was concentrated in the same manner as VLPs for use in gel analyses.
HIV-1-infected donor plasmas N90, N152, CAP256, BB34, 1688, 1702, and N160, and uninfected control plasma 210 have all been described previously [35, 60, 72, 86]. These were obtained from the Laboratory of Immunoregulation, NIAID (N90 and N152), The National Institute for Communicable Diseases, Johannesburg, South Africa (CAP256, BB34) and Zeptometrix, Inc., Buffalo, New York, USA (1648, 1688, 1702, N160, 210).
Blue native polyacrylamide gel electrophoresis (BN-PAGE) was performed as described previously [99]. Briefly, VLPs were solubilized in 0.12% Triton X-100 in 1 mM EDTA. An equal volume of 2x sample buffer (100 mM morpholinepropanesulfonic acid (MOPS), 100 mM Tris HCl, pH 7.7, 40% glycerol, and 0.1% Coomassie blue) was added. Samples were then loaded onto a 4–12% Bis-Tris NuPAGE gel (Invitrogen) and separated at 4°C for 3 hours at 100V. Proteins were then transferred to polyvinylidene difluoride (PVDF) membrane, destained, immersed in blocking buffer (4% nonfat milk in PBST) and probed with an anti-gp120 cocktail (39F, F2A3, C011 and 14e at 1μg/ml) and/or an anti-gp41 cocktail (2F5 and 4E10 at 1μg/ml). Blots were then probed by an anti-human Fc alkaline phosphatase conjugate (Accurate Chemicals) and developed using SigmaFast BCIP/NBT substrate (Sigma).
Env proteins were resolved by reducing SDS-PAGE. Briefly, samples were reduced and denatured by heating at 90°C for 10 min in LDS buffer (Invitrogen) prior to loading onto 4–12% Bis-Tris NuPAGE gels (Invitrogen). SDS-PAGE-Western blots were performed as described above for the BN-PAGE Western blotting method. For the cleavage of oligomannose and hybrid glycans, endonuclease H (endo H) (New England Biolabs) was added to samples after reduction and denaturation, and incubated for 37°C for 1 h prior to SDS-PAGE-Western blotting.
Wilcoxon Signed Rank tests were performed on data for each mAb-virus pair, organized into two columns to compare IC50s under control and GE-modified conditions.
Oligomannose arrays were printed using Man5GlcNAc2, Man6GlcNAc2, Man7GlcNAc2D1, Man7GlcNAc2D3, Man8GlcNAc2D1D3 (D denotes the arms bearing terminal mannose groups; see Fig 1), and Man9GlcNAc2 at 33μM (Z Biotech). GlcNAc2 was printed at the same concentration. Print buffer without glycans was included as a background control. Each array was hydrated for 2 min in ultrapure water and then blocked for 1 h with hydrazide glycan blocking buffer (Zbiotech), rotating at 40 rpm in the dark. Arrays were inserted into a SlideArray holder (SlideArray) to partition the array into 24 subarrays. MAbs were diluted to 50 μg/mL in hydrazide glycan assay buffer. PGT128 and biotinylated Concanavalin A were used as positive controls and V3 mAb 19b was used as a negative control. Each mAb was incubated on an individual subarray for 1h and then washed 5 times with PBS/0.05% tween 20 (PBST). Subarrays that received biotinylated Concanavalin A were incubated with streptavidin-Cy3 (Sigma). All other wells were incubated with anti-IgG-Cy3 (Sigma) for 1h while rotating at 40 rpm covered from light. The arrays were washed 5 times with 70μL of PBST and then washed once with 0.01X PBS and then dried. The arrays were scanned with a GenePix 4000B (Molecular Devices) scanner at wavelength 532nm using GenePix Pro7 software. The fluorescence within each feature was background subtracted using the local method in GenePix Pro7 software (Molecular Devices). Glycan specific binding = (glycan binding background-subtracted fluorescence) − (print buffer alone background-subtracted fluorescence).
A JR-FL native Env trimer structure (PDB: 5FUU) [3] was used to model native spike glycans. First, atomic clashes in the 5FUU structure were relieved and missing side-chains were rebuilt by running a constrained ROSETTA-relax simulation. Each sequon was decorated with a Man9GlcNAc2 glycan. The glycan at position N637 in gp41 is absent, per evidence that one or other glycans at N625 and N637 remain unoccupied [18]. For any overlapping sequons, e.g. those at N188 and 189, only the first sequon is occupied. GlycanRelax [103] was used to approximate glycan conformational behavior. For each model, 10 separate GlycanRelax trajectories of 10,000 cycles of MonteCarlo trials were carried out. Each gp120 glycan could move independently throughout the GlycanRelax minimization. A single low energy model was generated using PyMOL Software (Version 1.5.0.4 Schrödinger, LLC).
|
10.1371/journal.pntd.0002663 | Odorant and Gustatory Receptors in the Tsetse Fly Glossina morsitans morsitans | Tsetse flies use olfactory and gustatory responses, through odorant and gustatory receptors (ORs and GRs), to interact with their environment. Glossina morsitans morsitans genome ORs and GRs were annotated using homologs of these genes in Drosophila melanogaster and an ab initio approach based on OR and GR specific motifs in G. m. morsitans gene models coupled to gene ontology (GO). Phylogenetic relationships among the ORs or GRs and the homologs were determined using Maximum Likelihood estimates. Relative expression levels among the G. m. morsitans ORs or GRs were established using RNA-seq data derived from adult female fly. Overall, 46 and 14 putative G. m. morsitans ORs and GRs respectively were recovered. These were reduced by 12 and 59 ORs and GRs respectively compared to D. melanogaster. Six of the ORs were homologous to a single D. melanogaster OR (DmOr67d) associated with mating deterrence in females. Sweet taste GRs, present in all the other Diptera, were not recovered in G. m. morsitans. The GRs associated with detection of CO2 were conserved in G. m. morsitans relative to D. melanogaster. RNA-sequence data analysis revealed expression of GmmOR15 locus represented over 90% of expression profiles for the ORs. The G. m. morsitans ORs or GRs were phylogenetically closer to those in D. melanogaster than to other insects assessed. We found the chemoreceptor repertoire in G. m. morsitans smaller than other Diptera, and we postulate that this may be related to the restricted diet of blood-meal for both sexes of tsetse flies. However, the clade of some specific receptors has been expanded, indicative of their potential importance in chemoreception in the tsetse.
| Tsetse flies navigate their environments using chemosensory receptors, which permit them to locate hosts, mating partners, and resting and larviposition sites. The genome of G. m. morsitans was interrogated for coding genes of odorant receptors (ORs) and gustatory receptors (GRs) that express in antennae and maxillary palp, and detect the volatile and soluble chemical signals. The signals are then transmitted to the central nervous system and translated to phenotypes. Majority of these genes in G. m. morsitans were spread across different scaffolds, but a few were found to occur in clusters, which suggested possible co-regulation of their expression. The number of ORs and GRs were much reduced in the G. m. morsitans genome, including the apparent loss of receptors for sugar when compared to selected Diptera. There was also an apparent numerical expansion of some receptors, presumably to maximize on their restricted blood-meal diet. The annotation of the chemoreceptor package of G. m. morsitans provides a resource for investigating key activities of tsetse flies that could be exploited for their control.
| Trypanosomiasis management has been a longstanding development preoccupation in sub-Saharan Africa, with tsetse fly control constituting the cornerstone in this effort [1]. Since all tsetse species are able to transmit trypanosomes, the critical determinant of transmission is their obligate blood feeding. Tsetse flies select their hosts through visual and olfactory signals, a process that is mediated by olfactory and gustatory receptors. Tsetse flies navigate their environment by detecting and responding to volatiles and non-volatile cues (odors and tastants). Artificial bait technologies, based on tsetse olfactory responses to natural cues and blends of synthetic versions that mimic those of their natural hosts in the field, have successfully been applied in tsetse control because of their relatively high specificity, low cost, community acceptability, and ability to slow down tsetse re-invasion from adjacent areas [2], [3]. These technologies are environment friendly [4], and applicable for riverine and savanna species of tsetse flies [5], [6]. The attractants include various phenolic derivatives [7]–[9], carbon dioxide, acetone, 1-octen-3-ol, and vertebrate host breath, skin and urine extracts [10]–[12]. Interestingly, 1-octen-3-ol is a constituent of the chemical profile from Lantana camara, an invasive plant to which tsetse flies are attracted [13]. The response to olfactory cues has also been exploited in design of tsetse repellents [14], [15]. The repellents include guaiacol (methylphenols), δ-octalactone and methylketones [16]–[18] and 2-methoxy-4-methylphenol [14]. Natural differential responses among tsetse species and even between sexes and allopatric populations of the same species have been observed [18]–[22], which have stimulated research and design to enhance the efficiencies of the existing attractant-based bait technologies, to develop new ones based on repellent blends (‘push’ tactics) from refractory animals, and to integrate these into ‘push-pull’ strategies. Different Glossina species exhibit different olfactory uniqueness' and this may partly account for the observed graduation of preferences for particular hosts. For instance, riverine tsetse species (such as G. fuscipes fuscipes, G. palpalis and G. tachinoides) prefer feeding on reptilian hosts compared to their savanna relatives (G. morsitans morsitans, G. pallidipes) that feed largely on ungulates and other large mammals [6]. Larvipostion pheromones (n-pentadecane and n-dodecane) from exudates of mature larvae are also known to attract and induce gravid G. m. morsitans and G. m. centralis females to aggregate and deposit larvae [23]. Research on response to tastants in tsetse flies are limited, but point to their potential application in tsetse control [10], [24]. In all, responses to odors and tastants in tsetse have established utility in tsetse control that can be augmented with better understanding of the molecular factors that underpin these responses.
Molecular factors mediating the olfactory and gustatory responses in the tsetse flies are poorly understood. However, research on other insects indicates that the odors and tastants in the environment are generally detected in peripheral sensory neurons by distinct odorant and gustatory receptors (ORs and GRs) [25]–[28]. These receptors are divergent members of a superfamily characterized by seven transmembrane domains, and share low sequence conservation among them except at the C-terminus region that coincides with the seventh trans-membrane domain [29]. The ORs and GRs are thought to have evolved as parallel chemoreceptors across diverse organisms [26]. Each OR is expressed in olfactory receptor neurons (ORNs) within maxillary palpi and antennae [25], [30]–[32]. The ORs generally have multiple introns and are very divergent with poor structural conservation within and across insect orders and species [33], [34], which potentially reflect diverse olfaction related preferences within the orders and species. However, a canonical co-receptor commonly referred to as Orco remains highly conserved across insect orders [35]–[38], a phenomenon that may be associated with its role in proper tuning of odor specificity and activation necessary for appropriate signal transduction in the neurons [39]. The GRs are generally expressed in gustatory receptor neurons (GRNs) within gustatory organs [40] in response to soluble taste and contact pheromones [41], [42]. However, some GRs are expressed in antennal dendrites and respond to carbon dioxide, potentially implicating them in olfaction [40], [43]. The GRs are more conserved in sequence and structure than the ORs [44], [45] probably due to comparatively smaller search space among cues associated with GRs than ORs. The diversity among the ORs and GRs in tsetse can potentially shed light on the natural differential responses observed among them [12], [17], [18], [20]–[29], with potential application in tsetse control. To improve or develop new approaches of vector management, an understanding of the molecular attributes of GRs and ORs and their potential roles in tsetse ecology is essential.
This study was initiated to (1) comparatively annotate and catalogue ORs and GRs in G. m. morsitans (GMOY1.1), (2) establish evolutionary distance between G. m. morsitans ORs or GRs and those in especially D. melanogaster, and (3) examine relative expression of the ORs and GRs in the G. m. morsitans. The assembly has been estimated to be over 99% complete based on the software Core Eukaryotic Genes Mapping Approach (CEGMA) [46] and manually sequenced BACs data. The assembly is currently undergoing genome-wide manual curation and annotation by the International Glossina Genome Initiative (IGGI) consortium.
Coding sequences (CDS) of ORs and GRs in Drosophila melanogaster were obtained from FlyBase5.13 [47] and used to isolate their respective homologs in the G. m. morsitans genome (GMOY1.1) at VectorBase [48] using tBLASTx algorithm [49]. Scaffolds encoding the homologs were searched for and retrieved at a cut-off e-value <1.0e-05. Whole transcriptome illumina 84 million RNA sequence reads generated from female G. m. morsitans [50] were mapped onto the scaffolds using default settings in CLC Genomics workbench suite Version 4.8 (CLC Bio, Aarhus, Denmark). Gene loci of putative Glossina homologs were curated in the scaffold sequences flanking the tBLASTx hits, and intron/exons modeled using the RNA-seq mappings. The predicted gene models were viewed and edited using Artemis v13.2.12 [51] where, intron/exon boundaries were edited using motifs GT for 5′ donor site, and AG for 3′ acceptor site. The start codon (ATG) for each gene model was fixed at the 5′ end and the reading frame terminated at the first of any of the stop codons (TAA, TGA, or TAG). Sequences shorter than average size of known insect ORs (370 aa) were marked as incomplete if they lacked start or stop codons. Sequences with poorly conserved functional domains were considered as pseudogenes.
The homologs were validated through sequence-based searches for presence of ORs or GRs specific 7tm-6-olf-recpt or 7tm-7-olf-recpt [29], [52] domains respectively. The homologs were probed for the domains using DELTA BLAST algorithm [53] against the conserved domains databases (CDD) [54], and presence of alpha helix trans-membrane domains validated using TMHMM server v2.0 [55]. Additionally, all the putative ORs or GRs were validated, using BLAST2GO analyses [56] against the non-redundant Swiss-Prot database [57]. The curated gene models were assigned annotation identifiers by comparing them with automated transcript feature models obtained from the Glossina community annotation portal at VectorBase [48] and edited using Artemis genome viewer tool [51]. The models without automated prediction matches and identifiers were manually built using the Artemis gene build tool window [51] and given unique temporary annotation identifiers. In this respect, features for gene, exons, mRNA, and CDS were created for such gene models. The Glossina gene models were assigned putative gene names where GmmOR* and GmmGR* were adopted for G. m. morsitans odorant receptors and gustatory receptors respectively (the asterisk (*) being an identifier number). The annotated gene model features were submitted to the VectorBase community annotation portal for G. m. morsitans [48] for integration into genome database; nevertheless, a list of annotated amino acid coding sequences is presented in supplementary Dataset S1, and a list of associated gene identities in Table S2. The G. m. morsitans receptor repertoires were evaluated against those documented for D. melanogaster, Anopheles gambiae, Aedes aegypti, Apis mellifera, Nasonia vitripennis, Camponotus floridanus, Harpegnathos saltator and Tribolium casteneum (references in Table 1).
MUltiple Sequence Comparison by Log-Expectation (MUSCLE) tool [58] was used to align GmmORs and GmmGRs with homologs in D. melanogaster, and the alignments edited using Jalview web-server [59]. The secondary structures in the alignments were predicted using JPred program [60]. Phylogenetic cluster inference was done using Maximum Likelihood approach with best fitting Wheelan and Goldman+Freq (WAG+F) model [61], which was chosen as the best ranked from a panel of all amino acid model tests run in MEGA5 [62]. The initial tree was automatically generated and bootstrapped with 500 iterations. The evolutionary rate difference among sites was modeled using a discrete Gamma distribution (5 categories (+G, parameter = 4.2651)). The rate variation model allowed for some sites to be evolutionarily invariable ([+I], 0.8705% sites). All positions with less than 95% site coverage were eliminated and branch nodes determination set at very strong. Evolutionary analyses were conducted using the MEGA5 suite [62].
The expression profiles of G. m. morsitans ORs and GRs gene loci were determined using whole transcriptome 84 million illumina RNA-sequence reads [50]. The RNA-seq reads were mapped onto the G. m. morsitans ORs or GRs nucleotide coding sequences (CDS) in CLC Genomics Workbench (CLC Bio, Aarhus, Denmark) via RNA-seq analysis pipeline with default settings. The expression profiles were presented as reads per kilobase of exon model per million mapped reads (RPKM) for each receptor sequence [63].
Most of the gene loci of G. m. morsitans ORs and GRs were scattered amongst the scaffolds. Fifty percent of G. m. morsitans OR genes were encoded as single-copies on their respective scaffolds. The remainder were encoded in pairs or triplets per scaffold. Five G. m. morsitans OR loci (GmmOR6/7/8, GmmOR18/19, GmmOR22/25, GmmOR27/28 and GmmOR41/42) were located in tandem on their respective scaffolds. Similarly, five G. m. morsitans GR genes clustered on a single scaffold. The rest were encoded as single-copies on their respective scaffolds. All G. m. morsitans GR loci were annotated as complete genes.
Numbers of OR and GR gene loci recovered in G. m. morsitans, relative to those published in other insects are summarized in Table 1. Similar to most insects, the G. m. morsitans has more ORs loci than GRs loci, with the exception of D. melanogaster where the numbers are equal. However, the G. m. morsitans ORs are fewer than those documented in all the insects evaluated, including D. melanogaster. A similar trend was exhibited in G. m. morsitans GRs, except in relation to A. mellifera. Annotation of G. m. morsitans ORs and GRs are summarized in Table 2. The lengths of G. m. morsitans ORs varied between 260 and 541 amino acids, while those of G. m. morsitans GRs ranged from 309 to 514 amino acids. The number of exons ranged between two and eight or 12 in GRs and ORs respectively. The predicted genome structures are given in Figure S1. The frequency of detectable trans-membrane domains was also variable, with proteins having six trans–membrane domains representing about one half of all genes. The G. m. morsitans ORs (57%, 26 out of 46) were homologous to nine D. melanogaster ORs. Similarly, most of the G. m morsitans GRs (57%, 8 out of 14) were homologous to three D. melanogaster GRs genes. The remainder of the G. m. morsitans GRs had one-to-one homology with a single D. melanogaster specific homolog. Reciprocal blasts onto non-redundant protein databases for both G. m. morsitans ORs and GRs are summarized in Supplementary material – Table S1). GmmGR3 and GmmGR4 were also homologous to An. gambiae orthologs, while GmmGR5, GmmGR8 and GmmGR13 had homologs to genes in other Drosophila species. The G. m. morsitans ORs pseudogenes were scanty, representing 7% of the ORs genes recovered. Only GmmOR5 had alternative splice variants. The 7tm-6-olfct-rcpt domain was detected in all G. m. morsitans ORs, and the 7tm-7-chem-rcpt domain was detected in five ORs (GmmOR17, GmmOR21, GmmOR24, GmmOR38 and GmmOR39). The 7tm-7-chem-rcpt domain was also detected in all the G. m. morsitans GRs.
Phylogenetic relationships between G. m. morsitans ORs and GRs and their counterparts in D. melanogaster are summarized in Figure 1. Most of the G. m. morsitans ORs and GRs clustered with their respective ORs and GRs orthologs with a bootstrap support of over 80%. The G. m. morsitans OR14, OR15 and OR16 were homologous to a drosophila larvae receptor, Or45a. The G. m. morsitans co-receptor (Orco) (GmmOR1) had 100% bootstrap support homology to D. melanogaster homolog, Or63b, and was a single copy in the genome, similar to other insects investigated (data not shown). There was an expanded cluster of ORs in G. m. morsitans (GmmOR41-46), relative to a single D. melanogaster homolog, Or67d (Figure 1A), which also had multiple copies in An. gambiae, Cu. quinquefasciatus, Ae. aegypti, Tribolium casteneum (Data not shown). The G. m. morsitans and D. melanogaster GRs clustered into four groups (Figure 1B). Four G. m. morsitans GRs (GmmGR1-4) clustered with homologs of CO2 receptors, Gr21a and Gr63a in D. melanogaster; GmmGR6-7 and GmmGR14, though distantly, clustered with an unusual splice variant DmelGr28a/28b; GmmGR5, 8–12 were homologous to bitter taste-related sensors in D. melanogaster; and GmmGR13 clustered distantly to DmelGr58a/58b homologs, whose functions are unknown.
Relative expression profiles of the G. m. morsitans ORs and GRs gene loci are summarized in Figure 2. Among the G. m. morsitans ORs, expression of GmmOR15 was surprisingly most predominant, accounting for more than 90% of the total RNA-sequence data supporting expression of the ORs. GmmOR15 is homologous to Or45a gene in D. melanogaster. About 5% of RNA-sequence data provided supporting evidence for expression of GmmOR2, GmmOR1 (Orco homolog), GmmOR43 and GmmOR9. Expressions of GmmOR8, GmmOR11, GmmOR25, GmmOR31, and GmmOR39 were not detected in the available RNA-sequence dataset (Figure 2A). Amongst the GRs, GmmGR1-4 had the best RNA-sequence data expression support (Figure 2B).
Specific groups of the G. m. morsitans ORs and GRs were clustered within selected scaffolds. Similar clusters of genes performing common and related functions have been observed among chemosensory genes in D. melanogaster [41], [42], [44], and more recently among twelve G. m. morsitans major milk proteins associated with lactation [50]. Since genes within clusters are generally co-regulated and can lead to joint gene expression [29], [34], [64], the individual clusters of ORs and GRs might be under common regulatory mechanisms and in response to common or related stimuli. The ORs and GRs in G. m. morsitans were fewer than those documented in most insects evaluated (Table 1) [65], [66]. Additionally, specific ORs and GRs in D. melanogaster (nine and three ORs and GRs respectively) appear to have been expanded in G. m. morsitans, representing more than half of the chemoreceptors.
The factors underlying the apparent reductions and expansions of these receptors in the tsetse are unknown. However, it can be postulated that the obligate blood feeding of the tsetse fly (restricted to vertebrate hosts) relative to D. melanogaster (with expansive fruit species hosts) might have necessitated evolutionary selection for specific chemoreceptor loci relevant to discriminate among limited host choices. We know also that environmental factors can determine host choice, as tsetse have been shown to have an acquired preference to specific hosts encountered early in life [67]. Notably, other blood-feeders, such as mosquitoes also seek a variety of plant sources for sugar as energy source, while tsetse flies derive their energy from the amino acids proline and alanine [68]. The G. m. morsitans OR15 (GmmOR15) accounted for more than 90% of the OR expression data. This OR is homologous to DmelOr45a, whose product has been, associated with an escape response in D. melanogaster larvae [69]. The function of this OR in tsetse was not determined; nonetheless it is notable that the source of RNA sequence data was a reproductively active adult female. Hence, it is possible that the GmmOR15 is in some way associated with larval activity.
Similarly, the GmmGR1-4 cluster was most prominent among the GRs homologous to CO2 receptors in D. melanogaster. These GRs may be associated with host seeking and may have a duplicate role in olfaction. These receptors may putatively be associated with attractive responses elicited by the savanna tsetse species, including G. m. morsitans [10]. From the foregoing, it is evident that tsetse seems to prioritize and invest on a select few chemoreceptor genes towards their adaptive behaviors. Indeed, a heavy investment in specific genes is not uncommon in insects [70]–[73]. The G. m. morsitans OR1 (homologous to Orco) was the most conserved amongst the G. m. morsitans ORs, not surprising since such conservation has been observed in other insects [74] probably due to its critical role in modulating responses of the other receptors.
In conclusion, when examined against other blood feeders, which also take sugar sources from plants (e.g. An. gambiae and Ae. aegypti), the G. m. morsitans has a reduced repertoire of ORs and GRs genes. There is a complete loss of receptors for sugar, and a heavy investment in some chemoreceptors, such as those associated with detection of CO2. These observations offer opportunities to develop control tools exploiting these unique adaptations.
|
10.1371/journal.pgen.1000545 | Enhanced Disease Susceptibility 1 and Salicylic Acid Act Redundantly to Regulate Resistance Gene-Mediated Signaling | Resistance (R) protein–associated pathways are well known to participate in defense against a variety of microbial pathogens. Salicylic acid (SA) and its associated proteinaceous signaling components, including enhanced disease susceptibility 1 (EDS1), non–race-specific disease resistance 1 (NDR1), phytoalexin deficient 4 (PAD4), senescence associated gene 101 (SAG101), and EDS5, have been identified as components of resistance derived from many R proteins. Here, we show that EDS1 and SA fulfill redundant functions in defense signaling mediated by R proteins, which were thought to function independent of EDS1 and/or SA. Simultaneous mutations in EDS1 and the SA–synthesizing enzyme SID2 compromised hypersensitive response and/or resistance mediated by R proteins that contain coiled coil domains at their N-terminal ends. Furthermore, the expression of R genes and the associated defense signaling induced in response to a reduction in the level of oleic acid were also suppressed by compromising SA biosynthesis in the eds1 mutant background. The functional redundancy with SA was specific to EDS1. Results presented here redefine our understanding of the roles of EDS1 and SA in plant defense.
| Salicylic acid and enhanced disease susceptibility 1 are important components of resistance gene-mediated defense signaling against diverse pathogens in a variety of plants. Present understanding of plant defense signaling pathways places salicylic acid and enhanced disease susceptibility 1 downstream of resistant protein activation. In addition, enhanced disease susceptibility 1 is primarily thought to function in the signaling initiated via Toll-interleukin 1-receptor type of resistance proteins. Here, we show that salicylic acid and enhanced disease susceptibility 1 serve redundant functions in defense signaling mediated by coiled-coil-domain containing resistance proteins that were thought to function independent of enhanced disease susceptibility 1. Furthermore, resistance signaling induced under low oleic acid conditions also requires enhanced disease susceptibility 1 and salicylic acid in a redundant manner, but these components are required upstream of resistance gene expression. Together, these results show that the functional redundancy between salicylic acid and enhanced disease susceptibility 1 has precluded their detection as required components of many resistance protein–signaling pathways.
| Plants have evolved highly specific mechanisms to resist pathogens. One of the common ways to counter pathogen growth involves the deployment of resistant (R) proteins, which confer protection against specific races of pathogens carrying corresponding avirulence (Avr) genes [1]. Following recognition of the pathogen, one or more signal transduction pathways are induced in the host plant and these lead to the prevention of colonization by the pathogen. Induction of defense responses is often accompanied by localized cell death at the site of pathogen entry. This phenomenon, termed the hypersensitive response (HR), is one of the earliest visible manifestations of induced defense reactions and resembles programmed cell death in animals [1]–[6]. Concurrent with HR development, defense reactions are triggered in both local and distant parts of the plant and accompanied by a local and systemic increase in endogenous salicylic acid (SA) levels and the upregulation of a large set of defense genes, including those encoding pathogenesis-related (PR) proteins [7]–[9].
The SA signal transduction pathway plays a key role in plant defense signaling (see reviews in [10]–[12]). Arabidopsis mutants that are impaired in SA responsiveness, such as npr1 (Nonexpressor of PR; [13]–[15]), or are defective in pathogen-induced SA accumulation, such as eds1 (Enhanced Disease Susceptibility 1; [16]), eds5 (Enhanced Disease Susceptibility 5; [17]), sid2 (isochorishmate synthase; [18]) and pad4 (Phytoalexin Deficient 4; [19]), exhibit enhanced susceptibility to pathogen infection and show impaired PR gene expression. The EDS1, EDS5, PAD4, NPR1 and SID2 proteins participate in both basal disease resistance to virulent pathogens as well as R protein-mediated resistance to avirulent pathogens [20]. Defense signaling mediated via a majority of R proteins, which contain Toll-interleukin1-like (TIR) domains at their N-terminal ends, is dependent on EDS1 [21]. Conversely, the NDR1 (Non-race-specific Disease Resistance) protein is required for many R proteins that contain coiled-coil (CC) domains at their N-terminal ends. However, several CC-nucleotide binding site (NBS)-leucine rich repeat (LRR) type of R proteins, including RPP8, RPP13-Nd, HRT, and RPP7, signal resistance via a pathway(s) that is independent of NDR1 [21], [22]–[24]. Strikingly, the CC-NBS-LRR gene HRT, which confers resistance to Turnip Crinkle Virus (TCV), is dependent on EDS1 [23]. Besides HRT, the only other CC domain-containing R protein that utilizes an EDS1-dependent pathway is RPW8, which confers broad-spectrum resistance to powdery mildew [25]. However, RPW8 is not a typical NBS-LRR type of R protein; it contains an N-terminal transmembrane domain in addition to the CC domain. Although several components contributing to resistance against pathogens have been identified, the molecular signaling underlying R gene-mediated resistance still remains obscure. Furthermore, potential relationship(s) among different downstream components and how they relay information leading to resistance remains unknown.
The EDS1 and PAD4 proteins are structurally related to lipase/esterase-like proteins although their lipase-like biochemical functions have not been demonstrated [16],[19]. EDS1 interacts with PAD4 and SAG (senescence associated gene) 101 and the combined activities of these proteins are required for HR formation and to restrict the growth of virulent bacterial strains [26]. PAD4 and SAG101 also restrict the post-invasive growth of non-pathogenic fungi in Arabidopsis [27].
In addition to the major phytohormone-mediated defense pathways, fatty acid (FA)-derived signals have emerged as important mediators of defense signaling [28]–[35]. The Arabidopsis SSI2/FAB2-encoded stearoyl-acyl carrier protein-desaturase (SACPD) converts stearic acid (18∶0) to oleic acid (18∶1). A mutation in SSI2 results in the accumulation of 18∶0 and a reduction in 18∶1 levels. The mutant plants show stunting, spontaneous lesion formation, constitutive PR gene expression, and enhanced resistance to bacterial and oomycete pathogens [29],[36]. Characterization of ssi2 suppressor mutants has shown that the altered defense-related phenotypes are the result of the reduction in the levels of the unsaturated FA, 18∶1 [30], [31], [35], [37]–[40]. The altered defense-related phenotypes in ssi2 plants can be rescued by restoring the 18∶1 levels via second site mutations in genes encoding a glycerol-3-phosphate (G3P) acyltransferase [ACT1, 30], a G3P dehydrogenase [GLY1, 31], and an acyl carrier protein [ACP4, 35]. A mutation in act1 disrupts the acylation of G3P with 18∶1 resulting in the increased accumulation of 18∶1, thereby restoring wild-type (wt) phenotypes in ssi2 plants. ACT1 preferentially utilizes 18∶1 conjugated to the ACP4 isoform in Arabidopsis [35]. Thus, a mutation in acp4 produces similar phenotypes as the act1 mutant and suppresses ssi2-mediated signaling by increasing 18∶1 levels [35]. A mutation in GLY1 also restores 18∶1 levels in ssi2 gly1 plants because it disrupts the formation of G3P from dihydroxyacetone phosphate [31]. Reduced availability of G3P in turn impairs the ACT1-catalyzed reaction resulting in accumulation of 18∶1 in ssi2 gly1 plants. Concurrently, increasing the endogenous G3P levels via exogenous application of glycerol reduces 18∶1 levels and induces ssi2-like phenotypes in wt plants [31],[40]. This effect of glycerol is highly specific because ssi2-associated phenotypes are not induced upon glycerol treatment of act1 (defective in the acylation of G3P with 18∶1) or gli1 (defective in the phosphorylation of glycerol to G3P) mutants [40].
Recently, we showed that a reduction in 18∶1 levels upregulates the expression of several R genes in an SA-independent manner [37]. Furthermore, we showed that pathogen resistance induced via this mode bypasses the requirement for components that are normally required for signaling downstream of R protein activation. For example, resistance to TCV mediated by the R gene HRT (HR to TCV), requires the recessive locus rrt (regulates resistance to TCV), SA, EDS1 and PAD4 [23]. Exogenous application of SA induces the expression of HRT and overcomes the requirement for rrt. However, exogenous SA is unable to induce HRT or confer resistance in pad4 background [23]. Interestingly, even though a reduction in 18∶1 levels also upregulates HRT expression to confer resistance to TCV, this mode of resistance is independent of PAD4, SA, EDS1 and EDS5, which are required for HRT-mediated resistance to TCV [37]. Remarkably, induction of R genes in response to reduced 18∶1 is conserved in plants as diverse as Arabidopsis and soybean [41]. Furthermore, this low 18∶1-mediated induction of defense responses was also demonstrated in rice recently [42]. Together, these studies strengthen the conserved role of 18∶1 in plant defense signaling.
Here, we show that R gene expression induced in response to a reduction in 18∶1 levels and the associated defense signaling can be suppressed by simultaneous mutations in EDS1 and the genes governing synthesis of SA. We also show that EDS1 and SA function redundantly in R gene-mediated resistance against bacterial, viral and oomycete pathogens and that EDS1 also regulates signaling mediated by CC domain containing R proteins.
Signaling mediated by many R genes is known to require EDS1 and/or NDR1. Previously, we have shown that ssi2 eds1 plants continue to express R genes at high levels, including those that are dependent on EDS1 for their signaling [37]. To determine if NDR1 played a role in ssi2-triggered phenotypes, we generated ssi2 ndr1 plants. The double-recessive plants segregated in a Mendelian fashion and all ssi2 ndr1 plants showed ssi2-like morphology in the F2, F3 and F4 generations (Figure 1A; Table S1). Although the ssi2 ndr1 plants accumulated significantly less SA/SAG (Figure 1C), compared to ssi2 plants, they showed ssi2-like PR-1 and R gene expression (Figure 1D and 1E, Figure S1A). Exogenous glycerol application, which reduces 18∶1 levels, also induced R gene expression in eds1 and ndr1 plants (data not shown). Together, these results suggest that R gene expression induced by low 18∶1 levels does not require EDS1 or NDR1.
The SA/SAG levels in ssi2 eds1 and ssi2 ndr1 plants were significantly higher compared to those in wt plants (Figure 1C). To determine whether high SA in these genotypes was responsible for increased R gene expression, we generated ssi2 eds1 sid2 and ssi2 ndr1 sid2 plants. Interestingly, only the ssi2 eds1 sid2 plants showed wt-like morphology and did not develop visible or microscopic cell death (Figure 1A and 1B). In contrast, ssi2 sid2, ssi2 ndr1, ssi2 ndr1 sid2 or ssi2 eds1 plants exhibited ssi2-like phenotypes. PR-1 gene expression was restored to wt-like levels in the ssi2 eds1 sid2 and ssi2 ndr1 sid2 plants, due to the sid2-derived reduction in SA levels (Figure 1D). In contrast, expression of the SA-independent PR-2 gene was restored to basal levels only in ssi2 eds1 sid2 [43], but not in ssi2 sid2 or ssi2 ndr1 sid2 plants (Figure 1D, Table S2). Most importantly, ssi2 eds1 sid2 showed basal expression of R genes, unlike ssi2 ndr1 sid2 plants (Figure 1E and 1F; Figure S1A, S1B; Table S1). R gene induction was further confirmed by comparing the transcript profiles of 162 NBS-LRR genes in ssi2 sid2 with that of wt plants using Affymetrix ATH1 GeneChips arrays. Twenty-one NB-LRR genes were specifically expressed at 2-fold or higher levels in ssi2 sid2 plants as compared to wt (Col-0) or eds1 plants (P<0.05) (Table S2). All 21 NB-LRR genes were expressed at low levels in ssi2 eds1 sid2 plants, further confirming the results from the RT-PCR analysis. Transcriptional profiling performed using Affymetrix arrays showed that the induction of several R genes (RPM1, RPS2, RPP5, RPS4) was lower than 2-fold in ssi2 or ssi2 sid2 compared to wt plants (Table S2, data not shown for ssi2). To determine if this low-level induction translated to a significant increase in R protein levels, we analyzed the levels of RPM1 in ssi2 plants. Indeed, ssi2 plants accumulated significantly higher levels of the RPM1-Myc protein (Figure 1G).
To rule out the effects of the varied ecotypes of the ssi2 sid2 eds1 (Nössen, Col-0, Ler) plants we introduced eds1-1 (Ws-0 ecotype) and eds1-2 (Ler ecotype) alleles in ssi2 sid2 and ssi2 nahG (Nössen ecotype) backgrounds (Table S1). All combinations of ssi2 with eds1-1/eds1-2 and sid2/nahG produced similar phenotypes (data not shown). FA profiling showed that the ssi2 eds1 sid2 plants contained low 18∶1 levels, similar to ssi2 plants (Table S3). We thus concluded that EDS1 and SA function downstream of 18∶1 levels, but upstream of R gene expression. Furthermore, ssi2 eds1 sid2 plants were wt-like, even though neither ssi2 eds1 nor ssi2 sid2 were restored for defense signaling. Therefore, EDS1 and SA likely fulfill redundant functions in defense signaling induced in response to a reduction in 18∶1 levels.
To further test the redundancy for EDS1 and SA, ssi2 eds1 sid2 plants were treated with SA or its active analog benzo(1,2,3)thiadiazole-7-carbothioic acid (BTH). Application of SA or BTH induced lesion formation on ssi2 eds1 sid2 plants but not on wt, eds1, sid2, eds1 sid2 or EDS1 SID2 F2 plants (Figure 2A and 2B, data not shown for eds1 sid2 and EDS1 SID2). Also, application of SA or BTH induced R gene expression in ssi2 eds1 sid2 plants (Figure 2C). Thus, application of SA restored ssi2-like phenotypes in ssi2 eds1 sid2 plants. Since glycerol application mimics the effects of the ssi2 mutation, we generated eds1 sid2 plants and evaluated them for their ability to induce R genes in response to glycerol. Exogenous application of glycerol lowered 18∶1 levels in all genotypes, but induced the expression of R genes only in wt, eds1, sid2 and EDS1 SID2 F2 plants (Figure 2D, Figure S1C). Only a marginal or no increase in R gene expression was observed in the eds1 sid2 plants (Figure 2D). These results confirmed that EDS1 and SA function redundantly downstream of signaling induced by low 18∶1 levels, but upstream of R gene expression.
We next evaluated the effect of simultaneous mutations in EDS1- and SA-signaling pathways on resistance to TCV in the ssi2 background. We reported previously that resistance to TCV is dependent on the R gene, HRT, and a recessive locus rrt [23]. However, the ssi2 mutation overcomes the requirement for rrt in HRT-containing plants [23],[37]. Furthermore, the ssi2 mutation only confers resistance to TCV when HRT is present (Figure 3A). The ssi2 mutation also overrides a requirement for EDS1 and SA and consequently ssi2 HRT eds1 as well as ssi2 HRT sid2 plants exhibit resistance to TCV [37] (Figure 3A). Unlike HRT ssi2, HRT ssi2 eds1 or HRT ssi2 sid2 plants, the HRT ssi2 eds1 sid2 plants showed susceptibility to TCV; ∼85% HRT ssi2 eds1 sid2 plants were susceptible to TCV as against ∼2–4% of HRT ssi2 sid2 or HRT ssi2 eds1 plants (Figure 3A). TCV-induced expression of PR-1 is also independent of EDS1 and SA. However, TCV inoculation failed to induce PR-1 expression in HRT ssi2 eds1 sid2 plants, unlike in HRT ssi2 sid2 plants (Figure 3B). These results showed that both EDS1 and SA have redundant functions in ssi2-mediated resistance to TCV in HRT plants.
To determine the redundancy of EDS1 and SA in signaling mediated by CC-NBS-LRR R proteins, we tested the effects of mutations in EDS1- and/or SID2 on HR to TCV. Earlier, we showed that HRT-mediated HR to TCV and PR-1 gene expression is not affected by mutations in the EDS1 or SID2 genes [23]. Consistent with previous results, Di-17 (HRT-containing resistant ecotype), HRT sid2 and HRT eds1 plants revealed discrete and similar-sized HR lesions on TCV-inoculated leaves (Figure 3C and 3D). In comparison, HR in HRT eds1 sid2 plants was diffused and formed larger lesions (Figure 3C and 3D). Increased lesion size in HRT eds1 sid2 plants correlated with increased accumulation of the TCV coat protein (CP) and TCV CP transcript (Figure 3E and 3F). Analysis of PR-1 and PR-2 gene expression indicated that TCV-inoculated HRT eds1 sid2 plants accumulated lower levels of PR-1 and PR-2 transcripts, unlike Di-17, HRT eds1 or HRT sid2 plants (Figure 3G and 3H). In contrast to PR, HRT expression remained unaltered in HRT eds1 sid2 plants (Figure 3H). Together, these results suggested that EDS1 and SA function redundantly in HRT-mediated signaling leading to HR formation and expression of PR-1. The functional redundancy with SA was specific to EDS1 and did not extend to PAD4; HRT pad4 sid2 plants showed normal replication of the virus and wt-like HR and PR-1 gene expression (Figure 3C–3G).
A majority of CC-domain containing R proteins, including RPS2, have been reported as not requiring EDS1 for resistance signaling [21]. To determine the effect of simultaneous mutations in EDS1 and SID2 on RPS2-mediated resistance, we compared defense phenotypes produced in single or double mutant plants with that of plants lacking a functional RPS2 gene. Since different alleles of RPS2 confer varying levels of resistance to Pseudomonas syringae (containing AvrRPT2) [44], we screened and isolated an EDS1 knockout (KO) mutant (designated eds1-22) in the Col-0 background and crossed it into the sid2 background (Col-0 ecotype). Inoculation with P. syringae expressing AvrRPT2 induced severe chlorosis on eds1-22 sid2 leaves (Figure 4A). Similar results were obtained when P. syringae expressing AvrRPT2 was inoculated into eds1-1 sid2 double mutant plants (Figure S2A). Interestingly, these phenotypes were very similar to those produced on plants lacking a functional RPS2 (rsp2-101c), while eds1 and sid2 showed no or very mild symptoms, respectively (Figure 4A, Figure S2A). The appearance of symptoms correlated with bacterial growth; eds1-22 sid2 plants and the rps2 mutant supported maximum growth of the pathogen, followed by sid2 plants (Figure 4B). Similarly, the eds1-1 sid2 double mutant plants supported more pathogen growth compared to eds1-1 or sid2 plants (data not shown). Together, these data suggest that the simultaneous loss of EDS1- and SA-dependent signals is required to mimic a phenotype produced by the loss of the cognate R gene, RPS2.
To determine if the loss of both EDS1- and SA-dependent signaling impaired resistance by affecting the RPS2 protein, we analyzed R protein levels in eds1-1 and sid2 single and eds1-1 sid2 double mutant plants. Analysis of RPS2 tagged with HA epitope at various times did not detect any significant changes in RPS2 levels in response to inoculation with P. syringae expressing AvrRPT2 (Figure 4C). Therefore, RPS2 levels in mutant plants were analyzed at only 12 and 24 h post-pathogen inoculation. The RPS2-HA levels in eds1-1, sid2 or eds1-1 sid2 plants were similar to that in wt plants (Figure 4D). These results suggested that abrogation of resistance in eds1 sid2 double mutants was not due to a defect in the accumulation of the R protein.
We next evaluated the effects of mutations in EDS1 and SID2 on RPP8-mediated resistance to Hyalopernospora arabidopsidis biotype Emco5 encoding Atr8. RPP8 (encodes a CC-NBS-LRR type R protein)-mediated resistance signaling was previously reported to be independent of both EDS1 and SA [21],[24]. As expected, RPP8 plants (ecotype Ler) inoculated with the Emco5 isolate showed localized HR and did not support growth of the pathogen (Figure 5A). Consistent with earlier reports [21],[24], RPP8 eds1-2 plants also did not support the growth of Emco5, although they did develop trailing necrosis (Figure 5A and 5B). The presence of the nahG transgene did not alter HR formation or pathogen response in the RPP8 nahG plants (Ler ecotype). In contrast, eds1-2 nahG plants were affected in both HR as well as resistance; eds1-2 nahG plants not only showed extensive trailing necrosis but also supported growth and sporulation of the pathogen (Figure 5A–5C). Although RPP8 EDS1 nahG and RPP8 eds1-2 nahG plants showed contrasting phenotypes (Figure 5A–5C), we still wanted to rule out the possibility that susceptibility of eds1 nahG plants was not due to the accumulation of catechol, which is formed upon degradation of SA by NAHG. Estimation of SA levels in Emco5 inoculated RPP8 (Ler) plants showed marginal increase in SA and no significant increase in SAG levels compared to mock-inoculated plants (data not shown). This suggests that Emco5 inoculated nahG plants are unlikely to show a significant increase in catechol levels. In addition to this, we tested two independent lines of RPP8 eds1-2 sid2 (in the ssi2 background) plants and both showed increased susceptibility to Emco5 (Figure 5D). In comparison, RPP8 eds1-2 or RPP8 sid2 genotypes did not support any growth or sporulation of the pathogen (Figure 5D). Taken together, these results show that EDS1 and SA have redundant functions in RPP8-mediated resistance to H. arabidopsidis Emco5.
To determine the relation between EDS1- and SA-derived signaling, we compared PR-1 gene expression and resistance in plants that were either overexpressing EDS1 or were pretreated with SA. EDS1 overexpression was achieved by expressing EDS1 (At3g48090 from the Col-0 ecotype) under control of the CaMV 35S promoter in Col-0 plants (Figure 6A). The 35S-EDS1 plants analyzed in the T2 and T3 generations showed wt-like morphology (data not shown), wt-like expression of the PR-1 gene (Figure 6A) and accumulated wt-like levels of SA/SAG (data not shown). In comparison, exogenous application of SA induced PR-1 and EDS1 gene expression [data not shown; 16].
Analysis of RPS4 (encodes a TIR-NBS-LRR type R protein)-mediated resistance showed that exogenous application of SA enhanced resistance to P. syringae (expressing AvrRPS4) in wt as well as eds1-22 plants, although wt plants were more resistant to AvrRPS4 bacteria than the eds1-22 plants (Figure 6B). Overexpression of EDS1, on the other hand, did not alter the response to AvrRPS4 bacteria. Strikingly, exogenous application of SA on 35S-EDS1 plants enhanced resistance even more than in the SA-treated wt or eds1-22 plants. Together, these results suggest that EDS1- and SA-derived signaling contribute additively towards pathogen resistance.
We next evaluated the effect of the eds1 sid2 mutations on basal resistance to virulent P. syringae, since both EDS1 and SID2 are known to contribute to basal defense as well. The eds1-1, eds1-22, sid2 and eds1 sid2 plants all showed enhanced susceptibility to virulent bacteria as compared to the respective wt ecotypes (Figure 7A). Interestingly, unlike in the case of the avirulent bacteria, growth of virulent bacteria was similar in eds1 sid2 double mutant plants as compared to that in eds1 or sid2 single mutant plants. These results suggested that loss-of-function mutations in EDS1 and SID2 do not additively reduce basal resistance to virulent P. syringae. Similar to the results obtained with the bacterial pathogen, the loss of both EDS1- and SA-dependent signals did not additively lower basal resistance to TCV either (Figure 7B). This further suggested that the redundant functions of EDS1 and SA might be relevant only for R gene-mediated signaling.
Besides SID2, mutations in FAD7 FAD8, which catalyze desaturation of 18∶2 to 18∶3 on membrane glycerolipids, also lower the SA levels in ssi2 plants [40]. To test if fad7 or fad7 fad8 mutations produced a similar effect as sid2, these mutations were mobilized into the ssi2 eds1 background. The ssi2 eds1 fad7 and ssi2 eds1 fad7 fad8 plants were bigger in size compared to ssi2 fad7 or ssi2 fad7 fad8 plants (Figure S3A). The ssi2 eds1 fad7 fad8 were wt-like in morphology and showed no or greatly reduced cell death lesions (Figure S3A, S3B). PR-1 expression was greatly reduced or abolished in ssi2 eds1 fad7 and ssi2 eds1 fad7 fad8 plants, respectively (Figure S3C) and correlated with their endogenous SA/SAG levels; the ssi2 eds1 fad7 and ssi2 eds1 fad7 fad8 plants showed greatly reduced or basal levels of SA and SAG, respectively (Figure S3D, S3E). Expression of some R genes (SSI4, RPS2, RPP5) was nominally or moderately reduced in ssi2 eds1 fad7 plants (Figure S3D, S3E). By comparison, all R genes tested were expressed at basal levels in ssi2 eds1 fad7 fad8 plants (Figure S3F). These results showed that presence of fad7 fad8 mutations restored the altered defense phenotypes of ssi2 eds1 plants. FA profiling did not detect any significant increase in 18∶1 levels in ssi2 eds1 fad7 and ssi2 eds1 fad7 fad8 plants, compared to ssi2 fad7 and ssi2 fad7 fad8, respectively (Table S4). This suggested that restoration of defense phenotypes in ssi2 eds1 fad7 fad8 was not the result of restored 18∶1 levels, but rather the reduction of SA levels in the eds1 background.
Mutations in EDS5 and PAD4 also lower SA/SAG levels in ssi2 plants [40]. To determine if mutations in these can substitute for sid2 triple mutants containing ssi2 eds1 pad4 and ssi2 eds1 eds5 were generated. The ssi2 eds1 pad4 plants were morphologically similar to ssi2 eds1 or ssi2 pad4 plants and showed spontaneous cell death and increased expression of PR-1 gene (Figure 8A–8C). In comparison, ssi2 eds1 eds5 showed wt-like morphology, greatly reduced cell death and basal expression of PR-1 gene (Figure 8A–8C). Quantification of endogenous SA levels showed that both ssi2 eds1 eds5 and ssi2 eds1 pad4 accumulated lower SA/SAG levels compared to ssi2 eds5 and ssi2 pad4, respectively (Figure 8D and 8E). However, while ssi2 eds1 eds5 plants accumulated basal levels of SA/SAG, the ssi2 eds1 pad4 accumulated significantly higher levels of SA/SAG compared to wt, ssi2 sid2 and ssi2 eds1 eds5 plants (Figure 8D and 8E). Analysis of R gene expression showed greatly reduced levels in ssi2 eds1 eds5 plants but the ssi2 eds1 pad4 expressed ssi2-like levels of R genes (Figure 8F, Figure S1D). Taken together, these results suggest that the suppression of SA levels was required for the normalization of defense phenotypes in the ssi2 eds1 background.
Besides EDS1, the SA signaling pathway is also regulated by PAD4 and EDS5 and via the physical association of EDS1 with SAG101 and PAD4 [17],[19],[45]. To determine if PAD4, SAG101 or EDS5 also function redundantly with SA, we introduced the pad4, sag101 and eds5 mutations in the ssi2 and ssi2 sid2 backgrounds.
The ssi2 sag101, ssi2 pad4 and ssi2 eds5 plants showed ssi2-like morphology, visible and microscopic cell death and constitutive PR-1 gene expression (Figure S4A, S4B, S4C and Figure S5A, S5B, S5C). Consistent with these phenotypes, the ssi2 sag101, ssi2 pad4, ssi2 eds5 plants showed increased expression of R genes (Figure S4D and Figure S5D) and accumulated elevated levels of SA and SAG (Figure S4E, S4F and Figure S5E, S5F). Notably, the SA levels in ssi2 sag101 plants were ∼6-fold lower than in ssi2 plants, suggesting that SAG101 contributed to the accumulation of SA in ssi2 plants. To determine if the reduced SA in the sag101 background could restore wt-like phenotypes in ssi2 eds1 plants, triple mutant ssi2 eds1 sag101 plants were generated. Although the ssi2 eds1 sag101 plants accumulated significantly lower levels of SA/SAG (Figure S4E, S4F), these plants were only slightly bigger than ssi2 eds1 or ssi2 sid2 plants (Figure S4A), showed spontaneous cell death (Figure S4B) and expressed PR-1 (Figure S4C) and R genes constitutively (Figure S4D). We next analyzed the triple mutant ssi2 sag101 sid2, ssi2 pad4 sid2 and ssi2 eds5 sid2 plants. All the triple mutants contained wt-like levels of SA and SAG (Figure S4E, S4F and Figure S5E, S5F). The ssi2 sag101 sid2 plants were morphologically similar to ssi2 plants, showed spontaneous cell death and expressed R genes constitutively (Figure S4A, S4B, S4C, S4D). In comparison, the ssi2 pad4 sid2 and ssi2 eds5 sid2 plants were bigger in morphology. However, plants of both genotypes showed cell death (Figure S5A, S5B) and expressed R genes constitutively (Figure S5D). Together, these data suggest that the functional redundancy with SA was specific only to EDS1 and did not extend to PAD4, SAG101 or EDS5.
SA is long known as an essential modulator of R gene-derived signaling in pathogen defense. Several molecular components, including EDS1, have been identified as essential effectors of SA-derived signaling [23],[26],[45]. Since SA upregulates expression of EDS1, both SA and EDS1 are thought to function in a positive feedback loop and EDS1 is widely considered an upstream effector of SA [16],[19],[23],[45]. Recent data has shown that EDS1 signals resistance via both SA-dependent as well as SA-independent pathways [46]. Strikingly, EDS1-dependent but SA-independent branch of EDS1 pathway still requires SA pathway for full expression of resistance [46]. In this study, we have characterized the relationship between EDS1 and SA. We show that the two components act in a redundant, and not necessarily sequential manner to regulate R gene expression induced in response to a reduction in the levels of the FA 18∶1. Furthermore, EDS1 and SA also function redundantly in R gene-mediated defense against viral, bacterial and oomycete pathogens. It appears that the redundant functions of EDS1 and SA may have prevented their identification as required components for signaling mediated by CC-NBS-LRR R proteins. Indeed, RPS2-mediated signaling is fully compromised only in eds1 sid2 and not in the single mutant plants. Similarly, HRT-mediated signaling leading to HR formation and PR-1 gene expression is only affected in eds1 sid2 plants, while eds1 or sid2 plants behave similar to wt plants. Furthermore, RPP8-mediated resistance, which was previously reported not to require EDS1 or SA [21],[24], is compromised in plants lacking both EDS1 and SA. In contrast to their effect on R gene-mediated resistance, loss of both EDS1- and SA-dependent signals did not additively lower basal resistance to P. syringae or TCV. Together, these data suggests that the redundant functions of EDS1 and SA might be relevant only for R gene-mediated signaling.
In contrast to SA application, overexpression of EDS1 was unable to confer increased resistance to the avirulent pathogen P. syringae. Furthermore, unlike SA, overexpression of EDS1 was not associated with the induction of PR-1 gene expression. These findings, together with the observation that SA was able to induce EDS1 expression and that SA application on wt plants resulted in higher resistance than that in eds1, suggests that SA feedback regulates EDS1-derived signaling in a unidirectional manner (Figure 9B). Thus, SA application induces both SA- and EDS1-derived signaling, the additive effects of which enhance resistance in wt plants much more than in eds1-22 plants. Furthermore, the combined effects of SA pretreatment and EDS1 overexpression induced much better resistance than the individual effects of each. This is consistent with a previous report that 35S-EDS1 plants induce rapid and stronger expression of PR-1 in response to pathogen inoculation [47]. The additive effects of EDS1 and SA was also supported by the observation that eds1 sid2 plants showed pronounced chlorosis upon inoculation with AvrRPS4 expressing pathogen, which is recognized by a TIR-NBS-LRR protein RPS4 (Figure S2B). Since mutations in SA-independent branch of EDS1 pathway and sid2 have additive effects on R gene-mediated resistance [46], it is possible that overexpression of EDS1 triggers signaling via both SA-dependent and/or -independent branches of EDS1 pathway.
Although the Col-0 ecotype is thought to contain two functional alleles of EDS1 [26], a KO mutation in At3g48090 was sufficient to compromise both basal and R gene (RPS4)-mediated resistance. However, the Col-0 eds1-22 mutant consistently supported less growth of virulent or avirulent pathogens compared to eds1-1 or eds1-2 plants. This suggests that the second EDS1 allele in the Col-0 ecotype might also contribute towards the resistance response. This is consistent with another study where constitutive defense phenotypes due to the overexpression of the SNC1 gene, encoding a TIR-NBS-LRR R protein, are not completely suppressed by a mutation in eds1 in the Col-0 background but restored by the eds1 mutation in the Ws background [48].
The inability to accumulate SA together with a mutation in EDS1 was also required to suppress constitutive defense signaling resulting from the overexpression of R genes induced in response to reduced 18∶1 levels. Although eds1 or sid2 plants were entirely competent in inducing R gene expression in response to a reduction in 18∶1, eds1 sid2 plants were not. Thus, ssi2 eds1 sid2 as well as glycerol-treated eds1 sid2 plants showed wt-like expression of R genes while ssi2 eds1, ssi2 sid2 and glycerol-treated eds1 or sid2 plants showed increased expression of R genes. Moreover, treatment of ssi2 eds1 sid2 plants with exogenous SA restored R transcript induction and cell death in these plants. The fact that glycerol treatment is unable to induce R gene expression in eds1 sid2 plants supports the possibility that EDS1 and SA function upstream of, and not merely serve as a feedback loop in, R gene induction. Signaling induced by low 18∶1 levels continues to function in the absence of SA, suggesting a novel SA-independent role for EDS1 in defense signaling.
Since ssi2 eds1 sid2 plants contain a mixed ecotypic background (Nö, Ws/Ler, Col-0, ecotypes), it is possible that ecotypic variations in various genetic backgrounds resulted in the restoration of ssi2-triggered defense phenotypes. Indeed, phenotypic variations amongst different Arabidopsis ecotypes have been associated with many physiological processes [48]–[51]. Moreover, certain alleles can express themselves only in specific ecotypic backgrounds [48],[51]. However, since ssi2 EDS1 SID2, ssi2 EDS1 sid2 or ssi2 eds1 SID2 plants (F2 population) always exhibited ssi2-like phenotypes, it is highly unlikely that ecotypic variations resulted in the restoration of phenotypes in ssi2 eds1 sid2 plants. The effect of ecotypic variations on the observed phenotypes can be further ruled out for the following reasons. First, the effects of different mutations were assessed in multiple backgrounds. For example, we used both eds1-1 (Ws-0 ecotype) and eds1-2 (Ler ecotype) alleles in ssi2 sid2 (Nö, Col-0 ecotypes) and ssi2 nahG (Nö ecotype) backgrounds and all combinations of ssi2 with eds1-1/eds1-2 and sid2/nahG produced similar phenotypes (Table S1). Second, all defense phenotypes were assessed over three generations using multiple progeny. Third, similar results were obtained when different ecotypic backgrounds were evaluated for their response to different pathogens. For example, eds1 nahG or eds1 sid2 backgrounds conferred increased susceptibility to H. arabidopsidis, P. syringae and TCV, even though only the genotypes used for TCV were of mixed ecotypic backgrounds. Fourth, F2 plants containing wild-type alleles behaved like wild-type parents. Finally, the effects of various mutant backgrounds on ssi2 phenotypes were also confirmed by glycerol application on individual mutants.
Although glycerol treatment failed to induce R gene expression in eds1 sid2 plants, it did induce cell death. This is in contrast to the absence of a cell death phenotype in ssi2 eds1 sid2 leaves. One possibility is that the glycerol-triggered cell death is not due to a reduction in 18∶1 levels. However, significant overlap between ssi2- and exogenous glycerol-triggered signaling pathways lessens such a possibility [40]. An alternate possibility is that, while EDS1 affects a majority of the responses induced by low 18∶1 levels, the cell death phenotype is also governed by some additional molecular factor(s). This is supported by the fact that ssi2 pad4 sid2 plants exhibit improved morphology and reduced cell death even though they are not restored for other defense-related phenotypes.
Since the overexpression of R genes can initiate defense signaling in the absence of a pathogen [48],[52], it is possible that the induced defense responses in ssi2 plants are the result of increased R gene expression. This idea is supported by the fact that ssi2-related phenotypes can be normalized by restoring R gene expression to wt-like levels, irrespective of their 18∶1 levels. Thus, wt-like defense phenotypes are restored in suppressors containing high 18∶1 levels, such as ssi2 act1, ssi2 gly1 or ssi2 acp4 [30],[31],[35], as well as in suppressor containing low 18∶1 levels, such as ssi2 eds1 sid2 (this work) and restored in defective crosstalk (rdc) 2 (unpublished data) (Figure 9A). We have also characterized additional ssi2 suppressors that show wt-like phenotypes even though they contain low 18∶1 levels and express R genes constitutively (rdc3, rdc4). Together, these results suggest that the ssi2-associated phenotypes can be restored by normalizing R gene expression to wt-like levels either by increasing 18∶1 levels, impairing factors downstream of signaling induced by low 18∶1 levels, or impairing events downstream of R gene expression induced by low 18∶1 levels.
In addition to 18∶1 levels or R gene expression, ssi2-related defense signaling could also be normalized by altering some factor(s) that function downstream of R gene induction. Indeed, our preliminary characterizations have identified additional ssi2 suppressors that yield wt-like phenotypes with regards to defense signaling but continue to express R genes at high levels. Reduced 18∶1 levels may induce defense signaling by directly regulating the transcription of activators or suppressors of defense gene expression. This is supported by the fact that 18∶1-mediated activation of a transcription factor induces the expression of genes required for neuronal differentiation [53]. Similarly, in Sacharromyces cerevisiae as well as mammalian cells, binding of 18∶1 to specific transcription factors induces the transcription of genes carrying 18∶1 responsive elements in their promoters [54],[55]. On the other hand, expression of the oncogene HER2 is inhibited via the 18∶1-upregulated expression of its transcriptional repressor [56]. Reduced 18∶1 might also directly activate/inhibit/alter protein activities. For example, 18∶1 is known to activate the Arabidopsis phospholipase D [57] and inhibit glucose-6-phosphate transporter activity in Brassica embryos [58]. Indeed, we have also identified several Arabidopsis proteins for which enzymatic activities are inhibited upon binding to 18∶1 (unpublished data).
In conclusion, results presented here redefine the currently accepted pathway for SA-mediated signaling by showing that EDS1 and SA play a redundant role in plant defense mediated by R proteins and in signaling induced by low 18∶1 fatty acid levels. Further biochemical characterization should help determine if 18∶1 binds to EDS1 and if cellular levels of 18∶1 modulate the as yet undetected lipase activity of EDS1.
Plants were grown in MTPS 144 Conviron (Winnipeg, MB, Canada) walk-in-chambers at 22°C, 65% relative humidity and 14 hour photoperiod. The photon flux density of the day period was 106.9 µmoles m−2 s−1 and was measured using a digital light meter (Phytotronic Inc, Earth city, MO). All crosses were performed by emasculating the flowers of the recipient genotype and pollinatng with the pollen from the donor. All the genotypes and crosses analyzed in this work, their genetic background and number of single, double, or triple mutant plants studied are listed in Table S1. In most cases, single, double, or triple mutant plants were obtained from more than one combination of crosses and showed similar morphological, molecular and biochemical phenotypes. F2 plants showing the wt genotype at the mutant locus were used as controls in all experiments. The wt and mutant alleles were identified by PCR, CAPS, or dCAPS analysis and/or based on the FA profile [30],[31],[38],[40]. The EDS1 KO mutant in At3g48090 was, isolated by screening SALK_071051 insertion line, obtained from ABRC. The EDS1 KO was designated eds1-22, based on the previous designation assigned to SALK_071051 T-DNA KO line [48]. The At3g48090 gene showed 98.8% identity at amino acid level to EDS1 allele from Ler ecotype. The homozygous insertion lines were verified by sequencing PCR products obtained with primers specific for the T-DNA left border in combination with an EDS1-specific primer. The eds1-22 lines did not show any detectable expression of EDS1.
Small-scale extraction of RNA from one or two leaves was performed with the TRIzol reagent (Invitrogen, CA), following the manufacturer's instructions. Northern blot analysis and synthesis of random-primed probes for PR-1 and PR-2 were carried out as described previously [29].
RNA quality and concentration were determined by gel electrophoresis and determination of A260. Reverse transcription (RT) and first strand cDNA synthesis were carried out using Superscript II (Invitrogen, CA). Two-to-three independent RNA preparations were used for RT-PCR and each of these were analyzed at least twice by RT–PCR. The RT–PCR was carried out for 35 cycles in order to determine absolute levels of transcripts. The number of amplification cycles was reduced to 21–25 in order to evaluate and quantify differences among transcript levels before they reached saturation. The amplified products were quantified using ImageQuant TL image analysis software (GE, USA). Gene-specific primers used for RT–PCR analysis are described in Table S5.
The leaves were vacuum-infiltrated with trypan-blue stain prepared in 10 mL acidic phenol, 10 mL glycerol, and 20 mL sterile water with 10 mg of trypan blue. The samples were placed in a heated water bath (90°C) for 2 min and incubated at room temperature for 2–12 h. The samples were destained using chloral hydrate (25 g/10 mL sterile water; Sigma), mounted on slides and observed for cell death with a compound microscope. The samples were photographed using an AxioCam camera (Zeiss, Germany) and images were analyzed using Openlab 3.5.2 (Improvision) software.
The asexual conidiospores of H. arabidopsidis Emco5 expressing Atr8 were maintained on the susceptible host Nössen (Nö) or Nö NahG. The spores were removed by agitating the infected leaves in water and suspended to a final concentration of 105 spores/mL. Two-week-old seedlings were sprayed with spore suspension and transferred to a MTR30 reach-in chamber (Conviron, Canada) maintained at 17°C, 98% relative humidity and 8 h photoperiod. Plants were scored at ∼10–14 dpi and the conidiophores were counted under a dissecting microscope.
The bacterial strain DC3000 derivatives containing pVSP61 (empty vector), AvrRpt2 or AvrRps4 were grown overnight in King's B medium containing rifampicin (Sigma, MO). The bacterial cells were harvested, washed and suspended in10 mM MgCl2. The cells were diluted to a final density of 105 to 107 CFU/mL (A600) and used for infiltration. The bacterial suspension was injected into the abaxial surface of the leaf using a needle-less syringae. Three leaf discs from the inoculated leaves were collected at 0 and 3 dpi. The leaf discs were homogenized in 10 mM MgCl2, diluted 103 or 104 fold and plated on King's B medium.
Transcripts synthesized in vitro from a cloned cDNA of TCV using T7 RNA polymerase were used for viral infections [59],[60]. For inoculations, the viral transcript was suspended at a concentration of 0.05 µg/µL in inoculation buffer, and the inoculation was performed as described earlier [56]. After viral inoculations, the plants were transferred to a Conviron MTR30 reach-in chamber maintained at 22°C, 65% relative humidity and 14 hour photoperiod. HR was determined visually three-to-four days post-inoculation (dpi). Resistance and susceptibility was scored at 14 to 21 dpi and confirmed by northern gel blot analysis. Susceptible plants showed stunted growth, crinkling of leaves and drooping of the bolt.
Total RNA isolated from four-week-old plants using TRIZOL as outlined above. The experiment was carried out in triplicate and a separate group of plants was used for each set. RNA was processed and hybridized to the Affimetric Arabidopsis ATH1 genome array GeneChip following the manufacturers instructions (http://www.affymetrix.com/Auth/support/downloads/manuals/expression_analysis_technical_manual.pdf). All probe sets on the Genechips were assigned hybridization signal above background using Affymetrix Expression Console Software v1.0 (http://www.affymetrix.com/Auth/support/downloads/manuals/expression_console_userguide.pdf). Data was analyzed by one-way Anova followed by post hoc two sample t-tests. The P values were calculated individually and in pair-wise combination for each probe set. The identities of 162 NBS-LRR genes were obtained from the Arabidopsis information resource (TAIR; www.arabidopsis.org) and disease resistance gene homolog databases (http://niblrrs.ucdavis.edu/).
FA analysis was carried out as described previously [61]. For FA profiling, one or few leaves of four-week-old plants were placed in 2 ml of 3% H2SO4 in methanol containing 0.001% butylated hydroxytoluene (BHT). After 30 minutes incubation at 80°C, 1 mL of hexane with 0.001% BHT was added. The hexane phase was then transferred to vials for gas chromatography (GC). One-microliter samples were analyzed by GC on a Varian FAME 0.25 mm×50 m column and quantified with flame ionization detection. The identities of the peaks were determined by comparing the retention times with known FA standards. Mole values were calculated by dividing peak area by molecular weight of the FA.
SA and SAG quantifications were carried out from ∼300 mg of leaf tissue as described before [23].
SA treatments were carried out by spraying or subirrigating 3-week-old plants with 500 µM SA or 100 µM BTH. For glycerol treatment, plants were sprayed with 50 mM solution prepared in sterile water.
Total protein was extracted in buffer containing 50 mM Tris pH 8.0, 1 mM EDTA, 12 mM β-mercaptoethanol and 10 µg ml−1 phenylmethylsulfonyl fluoride. Proteins were fractionated on a 10–12% SDS-PAGE to confirm the quality. An antigen-coated enzyme-linked immunosorbent assay was used to determine levels of TCV CP in the infected plants as described before [62].
For protein gel blot analysis, leaf tissue from 4-week-old plants was extracted with a buffer containing 50 mM Tris-HCl, pH 7.5, 10% glycerol, 150 mM NaCl, 10 mM MgCl2, 5 mM EDTA, 5 mM DTT, and 1× proteinase inhibitor (Sigma). Protein concentrations were determined by the Bradford assay (Bio-Rad, CA). For immunodetection, 10–50-µg protein samples were electrophoresed on 10–15% polyacrylamide gels and run in the presence of 0.38 M Tris and 0.1% SDS. Proteins were transferred from the gels to polyvinylidene difluoride membranes by electroblotting, incubated with primary anti-HA antibody (Sigma) and alkaline phosphatase-conjugated secondary antibody (Sigma). Immunoblots were developed using color detection.
|
10.1371/journal.ppat.1002247 | Vaccinia Virus Protein C6 Is a Virulence Factor that Binds TBK-1 Adaptor Proteins and Inhibits Activation of IRF3 and IRF7 | Recognition of viruses by pattern recognition receptors (PRRs) causes interferon-β (IFN-β) induction, a key event in the anti-viral innate immune response, and also a target of viral immune evasion. Here the vaccinia virus (VACV) protein C6 is identified as an inhibitor of PRR-induced IFN-β expression by a functional screen of select VACV open reading frames expressed individually in mammalian cells. C6 is a member of a family of Bcl-2-like poxvirus proteins, many of which have been shown to inhibit innate immune signalling pathways. PRRs activate both NF-κB and IFN regulatory factors (IRFs) to activate the IFN-β promoter induction. Data presented here show that C6 inhibits IRF3 activation and translocation into the nucleus, but does not inhibit NF-κB activation. C6 inhibits IRF3 and IRF7 activation downstream of the kinases TANK binding kinase 1 (TBK1) and IκB kinase-ε (IKKε), which phosphorylate and activate these IRFs. However, C6 does not inhibit TBK1- and IKKε-independent IRF7 activation or the induction of promoters by constitutively active forms of IRF3 or IRF7, indicating that C6 acts at the level of the TBK1/IKKε complex. Consistent with this notion, C6 immunoprecipitated with the TBK1 complex scaffold proteins TANK, SINTBAD and NAP1. C6 is expressed early during infection and is present in both nucleus and cytoplasm. Mutant viruses in which the C6L gene is deleted, or mutated so that the C6 protein is not expressed, replicated normally in cell culture but were attenuated in two in vivo models of infection compared to wild type and revertant controls. Thus C6 contributes to VACV virulence and might do so via the inhibition of PRR-induced activation of IRF3 and IRF7.
| A key event in the innate immune response to virus infection is the detection of pathogen-associated molecular patterns (PAMPs) such as viral DNA and RNA by cellular pattern recognition receptors (PRRs). This leads to expression of interferon-β (IFN-β) by an infected cell. Many viruses have evolved mechanisms to evade the induction of IFN-β. Here a screen of poorly characterized vaccinia virus (VACV) proteins identified protein C6 as an inhibitor of IFN-β induction by PRRs. Data presented show that C6 prevents the activation of the transcription factors IRF3 and IRF7 by the kinases TBK1 and IKKε, which are key components at the point of convergence of several PRR signalling pathways. C6 interacts with the scaffold proteins NAP1, TANK and SINTBAD, which are components of the protein complexes containing TBK1 and IKKε, and this interaction might modulate the activity of these kinases. C6 is expressed early during infection and contributes to virulence because viruses that do not express C6 are attenuated in two in vivo models compared to wild type and revertant control viruses.
| Mammalian cells respond to viral infection by producing pro-inflammatory cytokines and chemokines and also interferons (IFNs), of which type I IFNs, consisting of IFN-β and several IFNα proteins, are particularly important. IFN-α and IFN-β then act in an autocrine and paracrine manner to switch on hundreds of target genes which contribute to anti-viral innate immunity by blocking virus replication and alerting neighbouring cells to the danger of infection (reviewed in [1]). In addition to their role in innate immunity, type I IFNs also promote adaptive immune responses by priming T helper cells and cytotoxic T cells [2]. The initial production of type I IFNs is due to the activation of IFN regulatory factors (IRFs), and in particular IRF3, downstream of pattern recognition receptors (PRRs), which recognize viral DNA, RNA and proteins. PRRs that detect the presence of foreign RNA include the RIG-I-like receptors (RLRs) melanoma differentiation-associated gene 5 (MDA5) and retinoic acid induced gene I (RIG-I), which sense intracellular double-stranded (ds) RNA and single-stranded (ss) RNA containing a 5′ triphosphate, respectively [3]–[6]. Other PRRs that aid the detection of viruses include the endosomal toll-like receptors (TLRs), namely TLR3 which senses dsRNA, TLR7 and TLR8 which sense ssRNA and TLR9 which recognizes unmethylated DNA (reviewed in [7]). Intracellular DNA sensors such as AIM2, RNA polymerase III, DAI and IFI16 are also involved in sensing DNA viruses by recognizing the presence of dsDNA in the cytosol [8]–[15]. RNA polymerase III, DAI and IFI16 signal to cause type I IFN production, while AIM2 activates the inflammasome leading to processing of pro-interleukin (IL)-1β and release of IL-1β [9], [11], [12], [15]. RNA polymerase III is unusual in that it does not signal directly in response to DNA, but instead transcribes AT-rich DNA into RNA species, which are then recognized by RIG-I [8], [10].
The signalling pathways activated by the RLRs, the IFN-inducing intracellular DNA receptors and by TLR3 converge at the level of the kinases TNF receptor associated factor (TRAF) family member NF-κB activator (TANK)-binding kinase 1 (TBK1) and IκB kinase-ε (IKKε). These kinases exist in complexes with the scaffold proteins TANK, NAP1 (NAK-associated protein 1) or SINTBAD (similar to NAP1 TBK1 adaptor) [16]–[18]. To activate these kinases, RLRs and consequently RNA polymerase III signal via the adaptor protein MAVS (mitochondrial antiviral signalling) [19]–[21], other intracellular DNA sensors employ STING (stimulator of IFN genes) [22], [23], while TLR3 uses TRIF (TIR-domain containing adaptor molecule inducing IFN-β) [24], [25]. PRRs also require TRAF3 for the activation of TBK1 and IKKε (reviewed in [26]). Once activated by PRR signalling, TBK1 and IKKε phosphorylate IRF3, causing its translocation to the nucleus and the transcriptional activation of promoters containing appropriate binding sites, such as the IFN-β and CCL5 promoters, and a subset of promoters containing IFN-stimulated response elements (ISREs) [27], [28]. A related transcription factor of the IRF family, IRF7, also plays an important role in anti-viral responses and can be activated in a similar manner to IRF3 during viral infection [29]. However, while IRF3 is expressed constitutively, IRF7 is present at low levels in most cells, but is induced by type I IFNs in a positive feedback loop. Thus, IRF7 is particularly important for the continued expression of IFN-β during viral infection, and also contributes to induction of IFN-β by co-operation with IRF3 [29], [30]. In addition, IRF7 is essential for the induction of IFN-α genes that are not induced by IRF3 [29]. In plasmacytoid dendritic cells, an alternative TBK1- and IKKε-independent signalling pathway resulting in the activation of IRF7 is employed by TLR7, TLR8 and TLR9. These endosomal TLRs signal through the adaptor protein MyD88 (myeloid differentiation factor 88), leading to activation of the kinase IKKα which then phosphorylates IRF7 [29], [31], [32]. This is unusual, because in other PRR signalling pathways IKKα and IKKβ are involved in the phosphorylation of the inhibitor of NF-κB (IκB), causing its degradation and the subsequent activation of NF-κB. NF-κB is another transcription factor activated by PRRs, and is critical for innate immunity. NF-κB and IRF3 (or IRF7) co-operate with the activating protein 1 (AP-1) transcription factor family to induce the transcription of the IFN-β promoter.
A functional type I IFN response provides a potent means of controlling virus infections [33] and consequently viruses have evolved numerous counter-measures to stop the production or action of IFNs or IFN-induced anti-viral proteins (for review see [34]). These strategies include blockage of antiviral PRR signalling pathways (reviewed in [34], [35]). Viruses with a large DNA genome, such as poxviruses, encode an extensive array of immunomodulatory proteins. Vaccinia virus (VACV), an orthopoxvirus used as a vaccine to eradicate smallpox, has many immune evasion mechanisms, and these include intracellular proteins that block PRR signalling, secreted factors that sequester IFNs and proinflammatory cytokines, and proteins that inhibit the effector actions of an IFN response (reviewed in [36]). However, the exact function of many of the approximately 200 virus proteins remains unclear. Here, a functional screen was used to identify VACV proteins that inhibit the induction of the type I IFN response after PRR activation. It is shown that protein C6, the product of the C6L gene, is an inhibitor of IFN-β promoter activation. The C6 protein is a member of a family of VACV proteins that includes B14, A52 and K7 [37], [38]. The crystal structures of B14, A52 and K7 were solved [39], [40] and showed that they, and also VACV proteins N1 [41], [42] and F1 [43], adopt a Bcl-2-like fold. Functional characterisation showed that only F1 and N1 inhibit apoptosis [42], and consistent with this these proteins have a surface groove for binding BH3 peptides from pro-apoptotic Bcl-2 proteins [42], [43]. In contrast, proteins B14, A52 and K7 lack this groove and inhibit innate immune signalling pathways instead [40], [44]. Interestingly, the protein N1 is both anti-apoptotic and inhibits NF-κB activation induced by IL-1 [40], [42], [45].
In this paper we demonstrate that C6 inhibits the activation of IRF3 and IRF7 downstream of the kinases TBK1 and IKKε, while C6 does not inhibit signalling pathways using IKKα for IRF7 activation. Inhibition of IRF3 and IRF7 by C6 may be mediated by its interaction with the scaffold proteins TANK, NAP1 and SINTBAD. Consistent with the ability of C6 to inhibit IFN-β expression, recombinant viruses that do not express C6 are attenuated in vivo compared to the wild type and revertant viruses. C6 represents the first viral protein shown to target the TBK1 scaffold proteins.
To uncover novel VACV proteins that inhibit innate immune signalling pathways, a functional screen of proteins from VACV strain Western Reserve (WR) was used to identify those that inhibit type I IFN induction. For this, poorly characterized proteins encoded in the terminal regions of the VACV WR genome were selected, because these regions are rich in immunomodulatory proteins [46]. Proteins encoded within the highly conserved central region of the VACV genome were excluded from the screen, as were proteins with well-characterized functions, secreted proteins and those smaller than 8 kDa. This selection process identified 49 ORFs, and these were amplified from VACV WR strain genomic DNA, and cloned into mammalian expression vectors. Plasmids encoding these ORFs were transfected individually into HEK293 cells, and the effect on the IFN-β promoter following PRR stimulation was measured by reporter gene assays. ORF VACVWR022 (gene C6L in VACV Copenhagen strain) encoding the protein C6 emerged from this screen as an inhibitor of IFN-β promoter activation. Expression of C6 inhibited the activation of the IFN-β promoter by transfected poly(dA-dT) (which acts via intracellular DNA sensors) or poly(I∶C) (which acts via RLRs) in HEK293 cells (Figure 1A, B), and by infection of cells with Sendai virus which activates RIG-I signalling [4] (Figure 1C). C6 also inhibited poly(I∶C)-induced IFN-β promoter activation in mouse NIH3T3 cells (Figure 1D). Furthermore, the presence of C6 inhibited the expression of endogenous IFN-β mRNA in Sendai virus-infected cells (Figure 1E), as well as the secretion of the chemokine CCL5 from infected cells (Figure 1F). Thus, the C6 protein significantly reduced the expression of IFN-β and CCL5 after stimulation of PRRs by ligands or viral infection.
During viral infection, IRF3 and NF-κB co-operate to activate the IFN-β and the CCL5 promoters. IRF3 is phosphorylated by the kinases TBK1 and IKKε, leading to its dimerization and translocation to the nucleus. In a similar way, the kinases IKKα and IKKβ phosphorylate IκB, which then releases activated NF-κB and allows its nuclear accumulation. To investigate whether the signalling pathways leading to NF-κB or IRF3 activation are inhibited by C6, the translocation of the NF-κB subunit p65 from the cytoplasm to the nucleus was measured by confocal microscopy. For this, HEK293T cells were transfected with a plasmid expressing GFP-tagged C6, or a control plasmid expressing GFP for 16 h. The cells were then infected with Sendai virus for 6 h, or stimulated with IL-1 for 15 min, fixed and stained for endogenous p65. During Sendai virus infection, approximately 20% of control cells expressing GFP displayed p65 accumulation in the nucleus (Figure 2A, B), and the presence of GFP-tagged C6 did not affect the extent of p65 nuclear translocation. Furthermore, the expression of C6 did not affect the nuclear accumulation of p65 in cells stimulated with IL-1, an activator of NF-κB, which, regardless of whether the cells expressed GFP or GFP-tagged C6, caused p65 nuclear translocation in more than 80% of cells (Figure 2A, B).
The effect of C6 on the expression of a luciferase reporter gene under the control of an NF-κB-dependent promoter was examined next. Over-expression of C6 did not prevent activation of the NF-κB-dependent promoter stimulated by IL-1 or tumour necrosis factor (TNF)-α in HEK293 cells (Figure 2C), or by poly(I∶C) in HEK293 cells expressing TLR3 (Figure 2D). In contrast, the VACV protein B14, another member of the poxviral Bcl-2-like protein family, inhibited NF-κB promoter activation under these conditions (Figure 2C, D) as shown previously [47].
The activation and nuclear translocation of IRF3 was investigated next. HEK293T cells transfected with V5-tagged C6 or V5-tagged GFP were infected with Sendai virus for 6 h, fixed and stained for IRF3. In cells infected with Sendai virus, IRF3 translocated to the nucleus in approximately 30% of cells expressing V5-tagged GFP (Figure 3A, B). However, in cells expressing V5-tagged C6, the translocation of IRF3 was impaired, and only 5% of C6-expressing cells displayed nuclear accumulation of IRF3 (Figure 3A, B). Similar results were obtained with GFP-tagged C6 (data not shown). This indicates that the inhibition of promoter induction by C6 is due to the prevention of the activation and/or nuclear translocation of IRF3, and not due to an effect on p65 activation.
To measure phosphorylation-dependent transactivation activity of IRF3, a luciferase-based IRF3 transactivation assay was utilized. This assay uses a fusion protein consisting of the DNA-binding domain of Gal4 and the transactivation domain of IRF3. When the IRF3 transactivation domain is phosphorylated by upstream signalling events, it induces expression of a luciferase reporter gene under the control of a Gal4-dependent promoter. Using this assay, C6 inhibited poly(dA-dT)-stimulated IRF3 transactivation (Figure 3C), providing further evidence that C6 inhibits the activation of IRF3 by PRRs. To determine which step of the signalling cascade that leads to IRF3 activation is targeted by C6, the ectopic expression of signalling proteins that act upstream of IRF3 activation was used to drive the IRF3 transactivation assay. The RLR adaptor MAVS and the kinases TBK1 and IKKε all promoted IRF3 activation in this assay when overexpressed (Figure 3D, E). Co-expression of C6 inhibited the activation of IRF3 in a dose-dependent manner in each case (Figure 3D, E), indicating that C6 acts at the level of these signalling components or further ‘downstream’ to prevent IRF3 activation.
To gain further mechanistic insight, the ability of C6 to inhibit the function of IRF3 once it is activated by phosphorylation was measured. To do this, a constitutively active form of IRF3 (IRF3-5D) was used in which serine to aspartate mutations mimic the phosphorylation of five key residues in the IRF3 sequence [48]. Over-expression of IRF3-5D induced the expression of a luciferase reporter driven by an ISRE element derived from the ISG15 promoter (Figure 3F), which was shown previously to be transcriptionally activated by IRFs [27]. C6 was unable to prevent the activation of the ISRE by over-expression of constitutively active IRF3 (Figure 3F), suggesting that C6 acts to prevent the activation of IRF3, but is unable to interfere with IRF3 function once it is activated by phosphorylation.
IRF7 is a transcription factor that is structurally and functionally related to IRF3 and also participates in the induction of the IFN-β promoter in response to PRR signalling (reviewed in [49]). IRF7 can be phosphorylated and activated by two distinct pathways. TLR3 and cytosolic PRRs, such as sensors of poly(dA-dT), act via TBK1 and IKKε, while the endosomal TLRs TLR8 and TLR9 activate IRF7 using a signalling pathway independent of TBK1 and IKKε, but involving MyD88 and IKKα [29], [32], [50].
The ability of C6 to inhibit the TBK1/IKKε-dependent and -independent signalling pathways to IRF activation was compared by employing a luciferase-based IRF7 transactivation assay. Activation of the TBK1/IKKε-dependent pathway following infection of HEK293 cells with Sendai virus (outlined in Figure 4A) was inhibited by C6 (Figure 4B), as was the activation of IRF7 by the downstream signalling components MAVS, TBK1 and IKKε (Figure 4C, D). To determine whether C6 could inhibit the TBK1/IKKε-independent pathway to IRF7 activation, HEK293 cells stably expressing TLR8 were used. Stimulation of TLR8 by the agonists CL075 or R848 activates IRF7 via MyD88 and IKKα (outlined in Figure 4E), however, this was not inhibited by C6 (Fig. 4F). Similarly, C6 was unable to inhibit IRF7 activation following over-expression of MyD88 or IKKα (Figure 4G). As observed for IRF3, C6 was unable to inhibit the activation of the ISRE element in response to constitutively active IRF7 (IRF7-4D, Figure 4H).
Taken together, these data indicate that C6 inhibits the activation of IRF3 and IRF7 by TBK1- and IKKε-dependent signalling pathways, implying that C6 acts on these kinase complexes, rather than acting on the transcription factors directly.
Published sequence data show that C6 has orthologues in other orthopoxviruses (OPVs), the capripoxvirus and deerpoxvirus (www.poxvirus.org), and within the OPV genus the conservation is high (89–97% amino acid identity). To investigate if C6 function is also conserved, the ability of the C6 orthologue from monkeypox virus (MPXV) strain ZAI 1979-005 (92% amino acid identity) to inhibit poly(dA-dT)-induced IRF3 transactivation was investigated (Figure S1A). The MPXV ORF encoding the C6 orthologue was amplified from DNA extracted from MPXV-infected HeLa cells. When a MPXV C6 expression vector was transfected into cells, the MPXV C6 protein, like the VACV C6 protein, inhibited the pathway at the level of TBK1 and IKKε, because MPXV C6 inhibited the activation of IRF3 caused by the over-expression of either of the two kinases, or of the adaptor protein MAVS (Figure S1B). Thus, the MPXV C6 orthologue behaved like VACV C6 in the assays tested.
To investigate how C6 antagonises activation of the pathway at the level of the TBK1- and IKKε-containing complexes, interactions between C6 and components of the kinase-containing complexes were sought by immunoprecipitation. HEK293 cells were transfected with plasmids encoding FLAG-tagged proteins and then infected with a VACV expressing HA-tagged C6. Immunoprecipitation with anti-FLAG antibody co-precipitated C6 with the scaffold proteins NAP1, SINTBAD and TANK but not with a FLAG-tagged control protein, FLAG-GFP (Figure 5A). In contrast, no interaction with TBK1 or IKKε was detected (data not shown). A reciprocal immunoprecipitation using lysates from cells over-expressing Streptavidin-tagged C6 and FLAG-tagged scaffold proteins also showed an interaction between C6 and the adaptors since immunoprecipitation of C6 co-precipitated NAP1, SINTBAD and TANK, but not GFP (Figure 5B).
It has been proposed that TBK1 and IKKε form distinct complexes, either as homodimers or heterodimers, which would contain a specific scaffold protein (namely TANK, NAP1 or SINTBAD [17], [51], [52]). To investigate whether the interaction between C6 and NAP1, TANK or SINTBAD affected the formation of signalling complexes, the scaffold-kinase interactions were investigated in the presence or absence of C6. A LUMIER interaction assay was used, in which a FLAG-tagged scaffold protein and luciferase-tagged TBK1 were co-expressed in the presence or absence of C6, and the amount of luciferase co-immunoprecipitated with the FLAG-tagged allele was quantified [17], [53]. C6 did not prevent the association between TBK1 and NAP1, SINTBAD or TANK (Figure 5C, E, G). In contrast, expression of isolated TBK1-binding domains (TBDs) inhibited the formation of the scaffold-kinase complexes (Figure 5D, F, H) as described previously [17]. Similar results were obtained for the interactions between the scaffold proteins and IKKε, which were not disrupted by C6 (Figure S2A–C). Thus, C6 appears to associate with the scaffold proteins TANK, NAP1 and SINTBAD, without disrupting the formation of the signalling complexes containing the kinases TBK1 or IKKε.
The expression of C6 protein during infection was investigated by infecting BSC-1 cells with VACV WR in the presence or absence of cytosine arabinoside (AraC), an inhibitor of viral DNA replication and hence intermediate and late protein expression. Using a polyclonal antiserum raised against C6 protein expressed in Escherichia coli, a 17-kDa C6 protein was detected starting from 2 h post infection, with continued expression at all time points thereafter (Figure 6A). Also, C6 was detected in the presence of AraC confirming its expression prior to DNA replication and hence as an early protein during infection, consistent with previous data for C6L mRNA expression [54]. In contrast, the expression of the late protein D8 [55] was blocked by the presence of AraC.
To characterize the contribution of the C6 protein to VACV replication, spread and virulence, recombinant viruses that did or did not express the C6 protein were generated. These viruses included a C6L deletion virus (vΔC6) lacking the C6 ORF, a plaque purified wild type virus (vC6WR) that was isolated from the same intermediate virus as the deletion mutant during transient dominant selection (see methods), a revertant virus where the C6L ORF was re-inserted into the deletion virus at its natural locus (vC6Rev), and an additional recombinant virus (vC6FS) where the C6 translational initiation codon was disrupted by the insertion of an adenine nucleotide. As expected, this virus, as well as vΔC6, did not express C6 protein whereas vC6WR and vC6Rev both did (Figure 6B). For localisation and interaction studies a virus expressing HA-tagged C6 was constructed (vC6HA). Reduced levels of C6 were detected from this recombinant virus using the anti-C6 serum, perhaps due to the lower expression of this protein compared to wild-type C6 or poorer detection by the antiserum. Nevertheless this protein was detected using an antibody against the HA epitope (Figure 6B), and shown to be functional in that it was capable of inhibiting IFN-β promoter induction when expressed from a plasmid (data not shown).
The intracellular localisation of C6 was investigated by biochemical fractionation of cells into cytoplasmic and nuclear fractions, followed by immunoblotting using the anti-C6 serum (Figure 6C). The integrity of nuclear and cytoplasmic fractions was confirmed by blotting for lamin A and C, and for tubulin, respectively. The expression of C6 was detected in both nuclear and cytoplasmic fractions of cells infected with the wild-type virus expressing C6 (vC6WR), the revertant virus (vC6Rev) or the virus expressing HA-tagged C6 (vC6HA) (Figure 6C). The anti-C6 serum was not suitable for the detection of wild-type C6 by immunofluorescence. However, both nuclear and cytoplasmic localisation of C6 was also observed by confocal microscopy when cells infected with vC6HA were stained with an antibody against the HA tag (Figure 6D).
The isolation of the deletion mutant and vC6FS virus indicated that C6 is not essential for VACV replication. To determine whether loss of C6 had an effect on virus replication kinetics or spread, the plaque size and virus growth in cell culture were analysed. The size of the plaques formed 72 h post infection with the various recombinant viruses was measured in three different cell lines: African green monkey BSC-1 cells, rabbit RK-13 cells and human TK-143 cells. The absence of C6 had no effect on the mean plaque size in any of the cell types studied (Figure S3A). To assess viral replication, BSC-1 cells were infected at a high (10) or low (0.01) multiplicity of infection (m. o. i.) with the set of recombinant viruses, and virus in the intracellular and extracellular fractions at various time points post infection was titrated by plaque assay (Figure S3B–E). No significant difference was observed between the titres of either the extracellular or intracellular forms of the recombinant viruses.
The contribution of C6 to VACV virulence was tested in two murine models of infection. In the intranasal (i.n.) model, groups of 10 BALB/c mice were infected with the recombinant viruses at 5×103 plaque forming units (p.f.u.) per animal and weight loss and signs of illness were recorded and compared. A significant difference in weight loss was observed between the viruses that did not express C6 (vΔC6 and vC6FS) and those that did express C6 (vC6WR and vC6Rev) between days 6 and 12 post infection (Figure 7A), with the mice infected with the C6 deletion viruses losing less weight overall and gaining weight more quickly during recovery. The viruses lacking C6 also caused significantly fewer signs of illness in infected mice between day 5 and day 12 after infection (Figure 7B).
In addition, the recombinant viruses were used to infect groups of 10 BL/6 mice intradermally with 104 p.f.u. virus per ear, in both ears, and the sizes of the resulting lesions were measured and compared. The lesions induced by vΔC6 and vC6FS were significantly smaller than those induced by vC6WR and vC6Rev between 6 and 26 days post infection (Figure 7C). In addition the lesions induced by the viruses lacking C6 began to heal sooner (11 days post infection) than those induced by the viruses that did express C6 (14 days post infection) (Figure 7C). Thus, these data show that a virus lacking C6 is attenuated in vivo and indicate C6 is a virulence factor in two different models of infection.
Here the VACV protein C6 is described as a novel modulator of the innate immune system. Data presented show that C6 inhibits IFN-β expression by preventing the activation of the transcription factors IRF3 and IRF7, while not affecting NF-κB activation. C6 acts at the level of the kinases TBK1 and IKKε, and is able to associate with the kinase-associated scaffold proteins NAP1, TANK and SINTBAD. The immunomodulatory function of C6 is likely to be important during infection, as a deletion virus lacking C6 is attenuated in mouse models in vivo.
C6 was identified as an inhibitor of the initiation of the IFN-β response in a screen of poorly characterised VACV proteins. That C6 might be an immunomodulator had been suggested by the previous observations that it belongs to a family of poxvirus proteins [37] whose members (A46, A52, B14, N1 and K7) were shown subsequently to belong to the Bcl-2 protein family and to have immunomodulatory activity (reviewed in [38]). While the family members share structural similarity, their degree of amino acid similarity is low indicating they diverged long ago and although they share an ability to manipulate innate immune signalling pathways, they differ in their targets and mechanisms of action. While A46 acts at the interface between TLRs and their adaptors [56], A52 targets the more downstream signalling factors TRAF6 and IL-1 receptor associated kinase 2 (IRAK2) [57]. B14, K7 and now C6 all appear to target the kinase complexes located at the point of convergence between several different PRR signalling pathways. However, each of the viral proteins targets distinct components of these complexes. B14 binds IKKβ and thereby inhibits IκB phosphorylation and NF-κB activation [47]. In contrast, K7 inhibits IRF3 phosphorylation by binding to the helicase DDX3, which is part of the complexes containing TBK1 and IKKε [58].
In this paper C6 is shown to inhibit the activation of IRF3 and IRF7 in a different way to K7, namely by interacting with TANK, NAP1 and SINTBAD. These three proteins act as scaffold proteins that associate constitutively with TBK1 and IKKε [17], [59], [60]. They are essential for the innate immune response to several different viruses and PAMPs, and in particular for the activation of IRFs, but not NF-κB, upon stimulation with Sendai virus or poly(I∶C) [16]–[18]. The observation that the scaffold proteins are targeted by a viral immunomodulator provides additional evidence for the importance of these proteins in the antiviral response. The precise function of TANK, NAP1 and SINTBAD in the process of TBK1 and IKKε activation has not been defined fully, but there is evidence that the adaptor proteins link the kinases to the upstream signalling pathways, possibly by interaction with TRAF3, which is a component of TLR and RLR signalling pathways [16]. The recruitment of the kinase complexes to TRAF3 would then require the scaffold proteins, and lead to the phosphorylation and activation of TBK1 and IKKε, ultimately leading to the phosphorylation of IRF3 and IRF7. However, how exactly complex formation is linked to the activation of the kinases, and which functions of the adaptors may be redundant or unique, has yet to be elucidated.
NAP1, TANK and SINTBAD are related in domain structure, possessing N-terminal coiled-coil regions that are important for homodimerization, and a central TBD, which mediates the interaction with TBK1 and IKKε [17]. While C6 interacted with all three scaffold proteins, it did not affect their interaction with either TBK1 or IKKε. In contrast, truncated proteins containing only the TBD of either NAP1, TANK or SINTBAD inhibited all the scaffold-kinase interactions in this assay, as described previously [17]. Thus the exact mechanism whereby C6 disrupts IRF activation remains to be determined. It is possible that C6 prevents the association of the scaffold-kinase complexes with TRAF3, or else prevents the activation of TBK1 and IKKε once the complexes are formed. Also, it is possible that the interaction between C6 and the scaffold proteins is indirect and is mediated by additional proteins that may be part of the kinase signalling complexes. Further analysis of the effect of C6 on these protein complexes may shed some light on the mechanism by which TBK1 and IKKε are activated - and inhibited - during viral infection.
VACV is not the only virus that inhibits the TRAF3-scaffold-kinase axis. Recently it was shown that the M protein of severe acute respiratory syndrome (SARS) coronavirus targets a complex containing TRAF3, and prevents the association of TRAF3 with TBK1, IKKε or TANK [61]. Like C6, the M protein inhibits the induction of the IFN-β promoter by inhibiting IRF3 activation [61]. However to our knowledge C6 is the first viral protein shown to associate with all three scaffold proteins.
VACV expresses several proteins that inhibit IRF activation, including the related Bcl-2-like protein K7, which also targets a TBK1-containing protein complex [58]. However, despite this, the effects of K7 and C6 are evidently not duplicative, because when K7 is still expressed, loss of C6 caused a marked virus attenuation in two models of infection. Similarly, there are several VACV proteins that inhibit NF-κB activation, for instance A52, A46, N1, B14. Yet deletion of any single member of this group causes virus attenuation suggesting non-redundant functions. Possible explanations for this non-redundancy might be cross talk between different pathways, so the outcome of blocking a pathway is influenced by the point at which a virus inhibitor functions to block the pathway. Alternatively, the virus proteins might have multiple functions as has been demonstrated for VACV protein N1. The need for the virus to express so many different non-redundant viral inhibitors of host signalling cascades may be due to the host innate immune system being able to partially compensate for the inhibition of an individual signalling component by using parallel pathways all ultimately leading to the induction of type I IFNs and pro-inflammatory cytokines. Furthermore, the importance of one particular signalling pathway or component may vary depending on the cell type infected or stage of infection, thus requiring the inhibition of several, seemingly redundant signalling proteins. Finally, it is plausible that the inhibition exerted by a single viral protein in not complete, particularly at early stages of infection, thus requiring the expression of several different factors targeting components of the same pathway to have an additive effect.
The characterization of poxvirus proteins that inhibit the innate immune system is of interest, since elucidating the mechanism of viral inhibition often reveals new insights into how innate immunity operates. Furthermore, VACV strains are in development as vaccine delivery systems against smallpox and other pathogens (reviewed in [62]), and as vectors for the targeting of tumours and for gene therapy (reviewed in [63]). It is shown here that deletion of a single ORF can attenuate the virus, thus possibly making it safer for clinical use. Also, the deletion of ORFs that encode inhibitors of the innate immune system would be predicted to make the virus more immunogenic, and thus make a more effective vaccine. In this context, it is interesting to note that C6 is conserved in most OPVs and that the MPXV orthologue of C6 is functionally equivalent to the C6 protein from VACV strains WR.
For the screen of VACV candidate immunomodulators, 49 ORFs were selected from the VACV WR strain genome and amplified by PCR from VACV WR genomic DNA isolated by phenol-chloroform extraction from purified viral cores. Candidate ORFs including C6L and B14R were cloned into the expression vector pCMV-HA (Clontech). C6 was also subcloned into pLENTI-Dest-V5 (Invitrogen) for immunoflourescence experiments. The ORF encoding MPXV C6 was amplified by PCR from DNA extracted from MPXV-infected HeLa cells (a kind gift from K. Rubins, Whitehead Institute) and cloned into pCMV-HA (Clontech). IFN-β-promoter luciferase reporter was a gift from T. Taniguchi (University of Tokyo, Japan) and NF-κB luciferase was from R. Hofmeister (University of Regensburg, Germany). ISRE-Luciferase and pFR-Luciferase were purchased from Promega. GL3-Renilla vector was made by replacing the firefly luciferase ORF from pGL3-control (Promega) with the renilla luciferase ORF from pRL-TK (Promega). FLAG- and luciferase fusions with signalling proteins for the LUMIER assay were described in [17]. IKKα was from Tularik Inc. Vectors expressing IRF3-Gal4, IRF7-Gal4, TBK1, IKKε and TRIF were a kind gift from K.A. Fitzgerald (University of Massachusetts Medical School, USA), MAVS was from T.J. Chen (University of Texas Southwestern Medical Centre, USA), MyD88 was from M Muzio (Milan, Italy), TLR3 was from D. T. Golenbock (University of Massachusetts Medical School, USA), and IRF3-5D and IRF7-4D were from J. Hiscott (McGill University, Montreal, Canada).
For construction of the C6 deletion virus 250-bp flanking regions of the C6L gene were amplified by PCR from VACV WR genomic DNA, ligated together and inserted into a plasmid containing the Escherichia coli guanylphosphoribosyl transferase (Ecogpt) gene fused in-frame with the enhanced green fluorescent protein (EGFP) gene (Z11ΔC6). For construction of Z11C6rev, Z11C6FS and Z11C6HA, C6L, C6L with an additional adenine nucleotide in the start codon or C6L with a C-terminal HA tag respectively, plus C6L flanking regions were amplified from VACV WR genomic DNA and inserted into the Z11 plasmid.
C6 polyclonal antiserum was raised against C6 protein purified from Escherichia coli and injected into rabbits (Eurogentec). Other antibodies were from the following sources: IRF3 (IBL), V5 (Cell Signaling), p65 (Santa Cruz), IgG from rabbit serum (Sigma), TBK1 (Cell Signaling), IKKε (Abcam), TANK (Abcam), FLAG (Sigma), Lamins A+C (Abcam), tubulin (Upstate Biotech). The mouse monoclonal antibody AB1.1 against D8 has been described previously [64]. Poly(I∶C) and poly(dA-dT) were from Sigma, TNF, IL-1, CLO75 and R848 were from Invivogen.
HEK293 cells were grown in Dulbecco's Modified Eagle's Medium (DMEM, GIBCO) supplemented with 10% fetal bovine serum (FBS, Biosera) and 10 µg/ml ciprofloxacin (Sigma). BSC-1 cells were maintained in DMEM supplemented with 10% FBS and penicillin/streptomycin (P/S) (50 µg/ml). RK-13 and TK-143 cells were maintained in minimum essential medium (MEM) supplemented with 10% FBS and P/S (50 µg/ml). NIH 3T3 cells were maintained in DMEM supplemented with 10% newborn bovine serum (NBS) and P/S (50 µg/ml). HeLa cells were maintained in MEM supplemented with 10% FBS, 1∶100 non-essential amino acids (NEAA) (Sigma) and P/S (50 µg/ml).
C6 recombinant viruses were constructed using the transient dominant selection method [65]. For construction of vΔC6, RK-13 cells were infected with VACV strain WR at 0.01 p.f.u. per cell and then transfected with the Z11ΔC6 plasmid using polyethylenimine (PEI) (1 mg/ml) according to the manufacturer's instructions. Progeny virus was harvested after 48 h and used to infect RK-13 cells in the presence of mycophenolic acid (MPA, 25 µg/ml), hypoxanthine (HX, 15 µg/ml) and xanthine (X, 250 µg/ml). EGFP-positive plaques were selected and purified by three rounds of infection using RK-13 cells in the presence of MPA, HX and X as above. Intermediate virus was resolved in BSC-1 cells by three rounds of infection in the absence of MPA, HX and X. The genotype of resolved viruses was analysed by PCR following proteinase K-treatment of infected RK-13 cells. Revertant viruses were constructed in a similar manner by transfection of plasmid Z11C6rev (for vC6 rev), Z11C6FS (for vC6FS) or Z11C6HA (for vC6HA) into vΔC6 infected cells.
Luciferase reporter gene assays were performed in HEK293 cells seeded in 96-well plates and transfected with 0.8 µl Genejuice (Merck) per well. Firefly reporter plasmid (60 ng), 20 ng GL3-Renilla control plasmid and 150 ng expression vector or empty vector control were used per well. For the IRF3 and IRF7 reporter gene assays, 60 ng pFR-Luciferase and 20 ng pGL3-Renilla were transfected together with 4 ng of IRF3-Gal4 or IRF7-Gal4, and 150 ng expression vector or control plasmid. For luciferase assays in NIH3T3 cells, cells were seeded in 96-well plates and transfected with 60 ng IFN-β-luciferase, 10 ng pRL-TK, and 250 ng expression vector or empty vector control. Cells were lysed in Passive Lysis Buffer (Promega), and firefly luciferase activity was normalized to renilla luciferase activity. Experiments were performed in triplicate and repeated at least 3 times.
RNA from HEK293 cells grown in 12-well plates was extracted using the RNeasy kit (QIAGEN), and converted to cDNA using the Quantitect RT kit (QIAGEN). IFN-β mRNA was quantified by real-time PCR with the TaqMan gene expression assay Hs00277188_s1 and a β-actin endogenous control VIC-MGB probe (6-carboxyrhodamine–minor groove binder; Applied Biosystems). Experiments were performed in triplicate.
Cell culture supernatants from HEK293 cells grown in 96-well plates were assayed for CCL5 protein using Duoset reagents (R&D Biosystems).
HEK293 cells were grown on glass coverslips and fixed with 4% paraformaldehyde in PBS. Cells were permeabilized in 0.5% Triton in PBS, pre-incubated for 1 h in blocking buffer (5% BSA, 0.05% Tween-20 in PBS), stained for 3 h with primary antibody (1∶300 in blocking buffer) and for 1 h with Alexa488 or Alexa647-labelled secondary antibodies (1∶500, Invitrogen). Coverslips were mounted in MOWIOL 4-88 (Calbiochem) containing DAPI (4,6-diamidino-2-phenylindole; 1 µg/ml). Images were taken on an Olympus FV1000 scanning confocal microscope.
For co-immunoprecipitation, HEK293T cells were grown in 10-cm dishes and co-transfected with vectors expressing FLAG-tagged proteins and a vector expressing TAP-tagged (consisting of FLAG and Streptavidin epitopes) C6, using Fugene-6 (Roche), or transfected with the vectors expressing FLAG-tagged proteins alone and then infected 48 h later with vC6HA (2 p.f.u. per cell) for 16 h. Cells were lysed in lysis buffer (0.1 (v/v) % Triton X-100, 150 mM NaCl, 10% glycerol, 10 mM CaCl2, 20 mM Tris-HCl pH 7.4 and protease inhibitors), pre-cleared by centrifugation and incubated with 30 µl of anti-FLAG M2 agarose beads (Sigma), or strepavidin agarose beads (Thermofisher) for 3 h. Immunoprecipitates were washed 3 times in lysis buffer, and eluted from the beads by boiling in sample buffer containing SDS. Proteins were resolved by SDS-polyacrylamide gel electrophoresis (PAGE) and detected by immunoblotting.
For the LUMIER assay, HEK293 cells were grown in 6-well plates and transfected with 0.5 µg FLAG-tagged plasmid, 0.5 µg luciferase-tagged plasmid and 3 µg C6 expression vector, TBD expression vector or empty vector control using GeneJuice (Merck). Cells were harvested 24 h later in Passive Lysis Buffer (Promega), and subjected to immunoprecipitation using 0.3 µl FLAG antibody pre-coupled to Protein A sepharose beads. Immunoprecipitates were washed 5 times in lysis buffer and eluted using 100 µM FLAG peptide (Sigma) in PBS, and renilla luciferase activity was measured.
HeLa cells were infected with recombinant VACVs at 5 p.f.u. per cell for 16 h. The cells were washed twice in ice-cold LS buffer (20 mM Hepes pH 7.8, 0.5 mM DTT, 0.5 mM MgCl2 in water) and allowed to swell on ice for 20 min. The cells were gently scraped and disrupted by Dounce homogenisation on ice. The lysates were centrifuged at 600× g for 2 min at 4°C to pellet the nuclei. The supernatant (cytoplasmic fraction) was removed. The nuclei were washed five times in PBS, placed in nuclei resuspension buffer (50 mM Tris-HCl pH 8, 0.5 mM MgCl2, 20 mM iodoactetamide supplemented with protease inhibitor (Roche)) and sonicated. Proteins were resolved by SDS-PAGE and detected by immunoblotting.
For the single-step growth curves BSC-1 cells were infected with 10 p.f.u. per cell. At 0 h, 12 h and 24 h post infection the medium was removed and the cells were collected by centrifugation at 500× g for 10 min. The supernatant was removed and extracellular virus titres were determined by plaque assay on BSC-1 cells. For intracellular virus, cells were scraped, collected by centrifugation and subjected to three rounds of freeze-thawing before determining viral titre by plaque assay. For the multi-step growth curve BSC-1 cells were infected with 0.01 p.f.u. per cell and intracellular and extracellular virus was harvested at 0, 12, 48 and 72 h post infection as described above.
RK-13, BSC-1 and TK-143 cell monolayers were infected in duplicate with virus at 50 p.f.u. per well for 72 h to allow formation of well separated plaques. The cells were washed once with PBS and stained for 30 min with crystal violet (5% (v/v) crystal violet solution (Sigma), 25% (v/v) ethanol). Wells were then washed with water and the sizes of six plaques per well were measured using Axiovision 4.6 software and a Zeiss Axiovert 200 M microscope.
Female BALB/c mice (n = 10, 6–8 weeks old) were infected intranasally (i.n.) with 5×103 p.f.u. and monitored as described previously [66], [67]. Female C57BL/6 mice (n = 10, 6–8 weeks old) were inoculated intradermally (i.d.) in the ear pinnae with 104 p.f.u. as described previously [68], [69].
Data were analysed using Unpaired Student's T tests. Statistical significance is expressed as follows: * P<0.05, ** P<0.01, *** P<0.001.
VACV WR C6, P17362.1; TBK1, AF191838_1; IKKε, NP_054721; NAP1, AAO05967; TANK, NP_001186064; SINTBAD, NP_055541; IKKα, NP_001269; MyD88, AAC50954; MAVS, Q7Z434.2; IRF3, AAH09395; IRF7, NP_001563; NF-κB p65, CAA80524; IFN-β, NC_000009.11; CCL5, NC_000017.10.
|
10.1371/journal.ppat.1000090 | Real-Time High Resolution 3D Imaging of the Lyme Disease Spirochete Adhering to and Escaping from the Vasculature of a Living Host | Pathogenic spirochetes are bacteria that cause a number of emerging and re-emerging diseases worldwide, including syphilis, leptospirosis, relapsing fever, and Lyme borreliosis. They navigate efficiently through dense extracellular matrix and cross the blood–brain barrier by unknown mechanisms. Due to their slender morphology, spirochetes are difficult to visualize by standard light microscopy, impeding studies of their behavior in situ. We engineered a fluorescent infectious strain of Borrelia burgdorferi, the Lyme disease pathogen, which expressed green fluorescent protein (GFP). Real-time 3D and 4D quantitative analysis of fluorescent spirochete dissemination from the microvasculature of living mice at high resolution revealed that dissemination was a multi-stage process that included transient tethering-type associations, short-term dragging interactions, and stationary adhesion. Stationary adhesions and extravasating spirochetes were most commonly observed at endothelial junctions, and translational motility of spirochetes appeared to play an integral role in transendothelial migration. To our knowledge, this is the first report of high resolution 3D and 4D visualization of dissemination of a bacterial pathogen in a living mammalian host, and provides the first direct insight into spirochete dissemination in vivo.
| Pathogenic spirochetes are bacteria that cause a number of emerging and re-emerging diseases worldwide, including syphilis, leptospirosis, relapsing fever, and Lyme disease. They exhibit an unusual form of motility and can infect many different tissues; however, the mechanism by which they disseminate from the blood to target sites is unknown. Direct visualization of bacterial pathogens at the single cell level in living hosts is an important goal of microbiology, since this approach is likely to yield critical insight into disease processes. We engineered a fluorescent strain of Borrelia burgdorferi, a Lyme disease pathogen, and used conventional and spinning disk confocal intravital microscopy to directly visualize these bacteria in real time and 3D in living mice. We found that spirochete interaction with and dissemination out of the vasculature was a multi-stage process of unexpected complexity and that spirochete movement appeared to play an integral role in dissemination. To our knowledge, this is the first report of high resolution 3D visualization of dissemination of a bacterial pathogen in a living mammalian host, and provides the first direct insight into spirochete dissemination in vivo.
| Pathogenic spirochetes are bacteria that cause a number of emerging and re-emerging diseases worldwide, including syphilis, leptospirosis, relapsing fever and Lyme borreliosis [1]–[6]. Many clinically-important spirochetes cross the blood-brain barrier and exhibit an unusual form of motility that is predicted to permit efficient movement through dense extracellular matrix in host tissues [6]–[9]. Spirochetes of the Borrelia burgdorferi sensu lato species complex are the causative agents of Lyme borreliosis [1],[10]. B. burgdorferi are transmitted to the skin of mammalian hosts through the bite of an infected tick. Subsequently they enter the vascular circulation and disseminate hematogenously to multiple tissues by unknown mechanisms. Untreated Lyme borreliosis can result in arthritis, carditis and neurological complications.
B. burgdorferi and other spirochetes interact with endothelial cells under static conditions in vitro [11]–[13]. However, spirochete-vascular interactions have never been examined in the host itself, or under the fluid shear forces that are present at dissemination sites. Indeed, host-pathogen interactions under shear stress conditions are still poorly understood for most bacterial pathogens that invade or disseminate in the mucosa or blood vessels, despite the importance of shear forces present in these environments.
Early studies with cultured endothelial cells found that treatments or mutations that rendered B. burgdorferi non-motile impaired invasion but not interaction [14]–[17], suggesting that the spirochete's ability to bore through dense tissues using translational motility might be important for vascular invasion. However, all previous investigations of B. burgdorferi dissemination were performed ex vivo in the absence of shear stress, using endothelial cell monolayers incubated with B. burgdorferi for periods as long as 24 hours, and employed non-dynamic visualization techniques such as electron microscopy which precluded observation of spirochete movement [14], [15], [17]–[19]. Conflicting reports found that extravasating B. burgdorferi were localized exclusively in either endothelial junctions or cells [14],[15],[18]. The role of host cells in transmigration was also controversial, because electron microscopy studies revealed no unambiguous evidence of endocytosis, and since the host microfilament toxin cytochalasin D did not inhibit spirochete internalization [14],[15]. Thus, the mechanism of B. burgdorferi dissemination in mammalian hosts remains a mystery.
It has been challenging to study host-spirochete interactions in a living host because their slender (<1 µm) morphology makes them difficult to visualize by standard light microscopy. Direct observation of Lyme borreliosis spirochete interactions with mammalian cells has been limited to cell culture models or host tissues removed from their native context. Several green fluorescent protein (GFP) alleles have been expressed in B. burgdorferi, usually in the context of reporter constructs used to monitor gene transcription and plasmid maintenance in spirochetes grown in culture [20]–[23]. Recently, a GFP reporter construct was used to monitor gene expression during B. burgdorferi infection, in tissues that were excised from the host before visualization [24]; however, detection of the GFP allele used in this study required relatively long, two second exposure times [23].
Intravital microscopy (IVM) is a powerful tool for studying the cellular dynamics of the immune and cardiovascular systems and tumor metastasis in the context of a living organism [25],[26]. It has also been used to visualize tissue localization dynamics of bacterial pathogens in vertebrate hosts [27],[28]. However, the small size of most pathogens and the spatial resolution limits of conventional epifluorescence IVM have impeded analysis of live host-pathogen interactions at the single cell level. The application of spinning disk confocal microscopy in an intravital setting finally enabled real-time visualization of transmission of malaria parasites to a living host [29], but unambiguous analysis of dissemination by smaller pathogens such as bacteria requires the ability to perform three-dimensional (3D) microscopy in vivo. In the current study we report the first use of spinning disk confocal microscopy to visualize dynamic host-pathogen interactions in three and four dimensions, revealing many aspects of the B. burgdorferi dissemination process that have not been previously observed, in vitro or in vivo. The ability to examine bacterial pathogenesis over time and in the three-dimensional space of living hosts will greatly enhance our understanding of many infectious diseases.
B. burgdorferi are difficult to modify genetically, and transformation with recombinant constructs often results in the loss of plasmids that are required for infectivity in the mouse [30]. We were able to engineer infectious and non-infectious B. burgdorferi expressing a highly fluorescent GFP allele optimized for bacterial expression [20],[31] (Fig. 1A); this allele is distinct from the egfp [21],[23] and gfpmut1 [21],[23] alleles that have also been expressed in B. burgdorferi. The resulting infectious (GCB726) and non-infectious (GCB705) strains displayed similar levels of GFP fluorescence, which could be detected with very short exposure times of less than 100 ms with a conventional epifluorescence microscope. The infectious strain contained the full complement of B. burgdorferi plasmids required for infectivity (see Materials and Methods).
To confirm that fluorescent strain GCB726 was infectious, and to determine if GFP expression could be stably maintained without antibiotic selection in infected murine hosts, two C3H/HeN and two C57 BL/6 mice were inoculated with 5.5×104 fluorescent spirochetes. Two ear punches were collected for each mouse 13 days (C3H mice) or 28 days post-infection (C57 mice), and cultured in B. burgdorferi growth medium with or without gentamycin selection. Spirochetes were recovered from all ear punches in both the presence and absence of antibiotic, confirming that strain GCB726 was infectious in both C3H and C57 mice. Furthermore, 98.7 −/+ 1.3% of the B. burgdorferi cultivated in the absence of gentamycin retained robust levels of GFP expression (Fig. 1B), indicating that the GFP-expressing plasmid was stably maintained in the context of the murine host.
We next investigated whether fluorescent B. burgdorferi could be exploited for real-time studies of host-spirochete interactions using intravital microscopy (IVM). Fluorescent infectious spirochetes were observed in situ in the ears of living C3H mice 20 and 27 days post-infection and in the ears of C57 mice 28 days after infection, using both conventional epifluorescence and spinning disk confocal IVM. Consistent with a previous report that B. burgdorferi localize to the perivascular connective tissue [32], fluorescent B. burgdorferi were observed outside, but usually close to blood vessels, and frequently translated back and forth repetitively over a relatively lengthy distance (Fig. 1C and Video S1). In situ, B. burgdorferi exhibited all of the translational (running) and non-translational (flexing) modes of movement that are characteristically observed in culture medium [7],[33].
Although spirochetes traveled more slowly and reversed directions more frequently when passing around visible obstacles such as blood vessels, they could achieve speeds of up to 4.0 µm/s in the ear, a speed which is very similar to the previously reported 4.25 µm/s in vitro rate [34]. Interestingly, although B. burgdorferi in situ often reversed their direction of movement every few seconds, which is typical of translational motility in vitro [35], they also exhibited sustained unidirectional movement for periods as long as 40 seconds.
In order to investigate the behavior of spirochetes in the host microvasculature, fluorescent B. burgdorferi were injected directly into the bloodstream of C57 mice via the jugular or femoral veins, and were visualized in real-time using both conventional epifluorescence and spinning disk confocal IVM. Prior to inoculation, fluorescent spirochetes were cultured for 48 hours in the presence of 1% mouse blood to promote adaptation to the host environment, since growth in blood is known to regulate the expression of many B. burgdorferi genes [36].
Vascular interactions were analyzed in flank skin, where the best optical clarity was obtained (see Fig. 2 and Video S2). The relevance of this site as a target for B. burgdorferi dissemination was confirmed by recovery of spirochetes from cultures of skin taken from mice 28 days post-infection. Fluorescent B. burgdorferi maintained a stable density in the bloodstream for longer than four hours after injection, but interactions with the microvasculature were analyzed between 5 and 45 minutes after injection during which time endothelial activation was not observed and spirochete titers and rates of interaction were stable (see Text S1 for data and discussion related to the lack of endothelial activation).
Examination of interactions between fluorescent B. burgdorferi and the microvasculature in more than 40 mice yielded several general observations. First, at similar blood densities, non-infectious fluorescent spirochetes did not associate with blood vessels, though they remained in the circulation, indicating that vascular interactions were not an artifact caused by high blood titers or by mechanical impediments to cell flow.
Second, infectious spirochetes associated with capillaries, postcapillary venules and larger veins, but not with arterioles (Fig. 2, Videos S2, S3 and S4). Vessel identity was determined by measurement of vessel diameters and by observation of blood flow patterns in the immediate vascular network (convergence indicates venules while divergence identifies arterioles). The observation that spirochetes did not interact with the lumenal surface of arterioles differs from the conclusions derived from a previous report of mice infected intradermally with B. burgdorferi, which found that after several weeks of infection spirochetes were preferentially localized to the walls of arterial vessels [32]. One possible explanation for this discrepancy is that in the previous study spirochetes might have migrated into the walls of arterial vessels from extravascular tissues, an event that is unlikely to have occurred during the short time frame of our experiments. Additionally, colonization of the connective tissue-rich walls of arteries could be promoted by bacterial adaptation to the host environment during longterm infections. It was unlikely that the inability of B. burgdorferi to interact with arterioles in the time frame of our experiments was due to differences in expression of host cell ligands in arterial and venous vessels, since spirochetes readily associated with the arterial endothelium under conditions of reduced blood flow. Therefore, reduced spirochete interactions in arterioles may have resulted from the elevated shear forces present in these vessels.
Third, interacting spirochetes in capillaries sometimes moved back and forth with and against the direction of blood flow (Fig. 2, Video S2). In contrast B. burgdorferi usually moved with the direction of flow in venules and veins where blood flow was more rapid, but moved freely in multiple directions under conditions of reduced blood flow. From these qualitative observations we infer that spirochete interactions with the host microvasculature are strongly affected by blood flow.
Finally, B. burgdorferi grown in the presence or absence of blood did not exhibit significant differences in the total number or type of interactions. This suggested that blood-stimulated gene expression in B. burgdorferi [36] was not a requirement for microvascular interaction.
Quantitative analysis of B. burgdorferi interactions in postcapillary venules (where interaction rates could be most accurately quantified) revealed two major types of associations: short-term interactions and stationary adhesions (Fig. 3A). Associations were characterized and quantified in venules because interactions in larger veins were too numerous for accurate quantification and blood flow in capillaries could be blocked by trapped spirochetes. Interactions were quantified using conventional video-based epifluorescence IVM, which captures rapid adhesion events more effectively than spinning disk confocal IVM. All spirochetes that paused and associated, even briefly, with the vessel wall were counted, and the length of time required to travel 100 µm along the vessel wall was measured. B. burgdorferi that did not associate with vessels moved very rapidly, and were visible only as blurs; therefore, non-interacting spirochetes could not be quantified. However, the total number of B. burgdorferi in the bloodstream could be quantified by counting the number of spirochetes in blood samples using cell-counting chambers and was similar in all experiments.
Short-term interactions included two sub-groups: transient and dragging interactions. Transient interactions, which constituted the majority of interactions, were defined as those where B. burgdorferi slowed, associated briefly with the endothelium then detached in a tethering-type interaction cycle (for examples, see Videos S3 and S4). Transiently associating spirochetes took less than 1 second to travel 100 µm in the vessel, but moved at least 8 times more slowly than the speed of blood flow, and frequently interacted only partially with the endothelium. Transient interactions could occur at the tip of the bacterium or elsewhere on the bacterial cell body, implying that the ends of B. burgdorferi are not the exclusive sites of tethering.
Transiently associating spirochetes that interacted with the endothelium along much of their length often slowed further and began dragging or crawling along the vessel wall. Short- and long-drag interactions were those in which spirochetes took 1–3s and 3–20s, respectively, to travel 100 µm along the vessel wall (for an example, see Video S4). B. burgdorferi dragging along the endothelium frequently slowed and stopped before dragging further along the wall in the direction of blood flow. The crawling movement observed at this stage of interaction may be similar to the crawling motion described for Leptospira, which results from translational motility in the context of simultaneous tethering of the spirochete at multiple distinct interaction sites [37].
Spirochetes that remained stationary at a single position on the vessel wall and did not translate along the vessel for at least 20 seconds were defined as stationary adhesions (for an example, see Video S3). Once stationary, these bacteria usually remained in the same place for at least two minutes (average: 10 minutes), and were aligned lengthwise along the vessel wall in the direction of blood flow. One end of stationary adhesions was usually less adherent and more mobile than the other, and sometimes exhibited a probing-type behavior that was more consistent with active gyration of the free spirochete end than with a passive rearrangement due to blood flow. The most stably adhered end was always pointing in the opposite direction to blood flow.
To more closely examine B. burgdorferi interactions with the host microvasculature we performed 3D reconstruction on z-series micrographs of spirochetes and PECAM-1-stained vessels obtained used spinning disk confocal IVM. PECAM-1 is expressed on endothelial cells and concentrates at endothelial cell junctions [38], making it useful for visualizing both the lumenal endothelial surface of vessels, as well as the more PECAM-1-intensive intercellular junctions. PECAM-1 was visualized using an Alexa Fluor 555-conjugated antibody to PECAM-1 that has been used previously to study junctional extravasation of leukocytes in vivo [39].
Three-dimensional visualization of short-term interactions and stationary adhesions in venules revealed that these two classes of associating spirochetes differed with respect to their position relative to the PECAM-1-stained endothelium (Fig. 3B and Video S5). Greater than 93% of short-term interactions were localized on the lumenal surface of the vessel wall and were not observed to project into the PECAM-1-stained endothelium in three dimensional reconstructions (see positions 1 and 2 of the dragging spirochete in Fig. 3B lower and side panels). In contrast, a large majority of stationary adhesions (79%, Fig. 3C) were embedded in the PECAM-1-stained endothelium, either partially (57%) or along their entire length (21%) (Fig. 3C; see positions 3 and 4 of the stationary spirochete in Fig. 3B lower and side panels). Video S5 presents a reconstructed three-dimensional view of a typical short-term interaction and stationary adhesion in a venule, with the short-term interaction visible only on the lumenal surface of the endothelium, and the adhering spirochete visible in both the lumen and projecting through the PECAM-1. The right hand panel of Fig. 3B shows lumenal and exterior views of these spirochetes from a 3D reconstruction. Because the PECAM-1 antibody stains the lumenal surfaces of endothelial cells, these data indicate that stationary adhesions are embedded more deeply in the endothelium than transiently interacting spirochetes, but do not imply that stationary adhesions project beyond the external boundary of vessels.
Also of interest, in the majority of stationary adhesions (71%), one end of the bacterium projected further into the PECAM-1 than the other end. The most deeply embedded portion of the adhesion was usually the most stable, since more superficially attached regions of the spirochete exhibited a greater range of movement (see Video S3 for an example). This observation suggested that stationary adhesions might be slowly extravasating through the endothelium via the more deeply embedded tip. However, we did not detect any consistent outward migration of stationary adhesions during the experimental time period studied (up to 45 minutes), although it remains possible that such emigration might take much longer to occur.
Finally, the localization of endothelium-interacting spirochetes was more precisely determined by examining the position of these interactions with respect to endothelial junctions (which are stained more intensely by anti-PECAM-1 antibody than the non-junctional surface of endothelial cells) (Fig. 4). To determine if PECAM-1 redistribution occurred in response to B. burgdorferi, we visualized junctions using PECAM-1 antibody before and after injection of infectious spirochetes, and examined junctional staining patterns from 5–45 minutes after spirochete injection. During this time frame, we observed no PECAM-1 redistribution in 95% of venules examined (n = 20 venules in 5 mice). Since junctional staining was sometimes incomplete, localization of interacting spirochetes was assigned only when the junctional boundaries of endothelial cells in the area of interaction were clearly demarcated. For the majority of stationary adhesions (∼70%), the most stable, deeply embedded region of adhesion occurred at junctions. However, about 25% were adhered primarily to endothelial cells (Fig. 4B). In contrast, the vast majority of transient and dragging short term interactions (93%) occurred on the surface of cells, with only 7% being found at junctions. Since cell surfaces make the largest contribution to total endothelial surface area, short-term interaction with cells may be a stochastic event, whereas stationary adhesion to junctions is likely the result of preferential localization.
B. burgdorferi escaped the microvasculature in an end-first fashion, and therefore projected out of the planes of view where most interactions were observed; thus, 2D visualization alone was insufficient for unambiguous analysis of this final stage of dissemination. Using 4D spinning disk confocal IVM (3D time courses), we measured the percentage of each spirochete's length that projected beyond the PECAM-1-labeled endothelium in successive z-series, and calculated the time span and speed of escape (Fig. 5A–C). As for stationary adhesions, most escaping spirochetes (83%) extravasated through endothelial junctions. Transmigrating spirochetes preceded by stationary adhesion were not observed in this study, although the time necessary to acquire successive z-series in 4D IVM might have precluded detection of short-lived adhesions that began extravasating. Escape took an average of 10.8 minutes, at an average net displacement velocity of 3.4 µm/min (Fig. 5c). The initial and final stages of the escape process were too rapid to capture visually in 4D, since they were faster than the 1–2 minutes necessary to acquire individual z-series (for sample 2D footage of the final stage of escape, see Video S6). Little net displacement occurred during the longer middle phase of escape, even though many bacteria in this phase exhibited obvious reciprocal translational motility (Fig. 5D and Video S7). Reciprocally translating spirochetes could move in either direction as quickly as 624 µm/min, a speed which greatly exceeded their net displacement velocity. The great speed of these bacteria in situ might thus have accounted for our difficulty in capturing the initial and final stages of extravasation. The abridged timelapse shown in Fig. 5A illustrates the typical triphasic escape dynamic. In this case, 41% and 43% of the spirochete length passed out of the PECAM-1-stained endothelium in the first and last 2 minutes of extravasation, respectively, whereas only 16% of the spirochete traversed the PECAM-1 layer in the intervening 16 minutes. It was unlikely that the speeds of the initial and final stages of extravasation were the result of passive drifting of the bacteria through the endothelium, leading us to conclude that transmigration was largely driven by spirochete motility. Furthermore, the speed of the final escape phase, in which spirochetes appeared to burst away from the vessel (Video S6), suggested that the reciprocal translational motility observed in the middle phase was the result of partial adhesion of either the middle or the lagging portion of the spirochete to the endothelium [37]. Together, these observations suggested a prominent role for spirochete motility in the final stage of dissemination.
In this work technological advances in confocal microscopy have been coupled with intravital imaging methodologies to allow for the first time, high resolution, three and four dimensional, real-time visualization of the interaction of a bacterial pathogen with its living host. We have used this technology to study the interaction of the Lyme borreliosis spirochete B. burgdorferi with the microvasculature of one of its natural hosts.
One of the central events in the development of spirochetal diseases is hematogenous dissemination [40]. Previous investigations of dissemination by pathogenic B. burgdorferi were performed in a static environment using endothelial cell monolayers incubated with B. burgdorferi for several hours or longer, and methodologies that precluded direct observation of spirochete behavior [14], [15], [17]–[19]. In contrast, dynamic, 3D and 4D analyses of interactions in a living host under shear stress conditions indicate that B. burgdorferi escape from the microvasculature is a multi-stage process (as summarized in Fig. 6).
Spirochetes first transiently tether to the endothelium, usually at cell surfaces and not intercellular junctions (Fig. 6A), then drag and crawl along the vessel wall while interacting with the endothelium along much of their length (Fig. 6B). In contrast, stationary adhesions are usually established, at intercellular junctions (Fig. 6C), which also appear to be the major site for B. burgdorferi extravasation (Fig. 6D). Both stationary adhesion and extravasation may, therefore, be mediated by host and spirochete molecules distinct from those involved in short-term interactions. It remains unclear if stationary adhesions represent an obligate step in the progression toward vascular escape, or if they act as facilitators of this event by modifying the endothelium (see below). Similarly, multiple types of interactions are observed during leukocyte trafficking under the shear stress conditions of blood flow, which depends on a progressive association between different classes of endothelial and leukocyte molecules as the interacting cell slows down and locates an extravasation site [41]. It is probable that B. burgdorferi interactions with, and escape from the endothelium entail a similar progression.
Previous reports indicate that B. burgdorferi invade cultured endothelial cells by both intracellular and intercellular routes [14],[15],[18]. Treponema pallidum generally migrate through endothelial monolayers via intercellular junctions, whereas Leptospira primarily invade endothelial cells themselves; however, these spirochetes have also been observed in endothelial cells and junctions, respectively [12],[13]. Our results indicate that although the major extravasation route of B. burgdorferi in vivo is the intercellular junctions, a small percentage can also emigrate through endothelial cells. This conclusion raises the interesting possibility that other spirochetes could also exhibit the same versatile invasive capacity in vivo.
B. burgdorferi are known to interact with multiple host molecules that could mediate interaction with and invasion of the host endothelium in vivo; these include fibronectin, plasminogen, glycosaminoglycans, and integrins such as the vitronectin and fibronectin receptors [18], [42]–[47]. Indeed, we are currently using the technology described here to further study B. burgdorferi adhesion and have thus far identified several of the host and bacterial molecules involved at specific steps in the adhesion process (manuscript in preparation). These observations support a progressive model of spirochete adhesion under shear stress conditions in which different classes of host and B. burgdorferi proteins mediate distinct phases of interaction.
One of the most interesting, unprecedented and difficult findings to interpret in our study was that stationary adhesions projected deep into and sometimes through the PECAM-1-stained region of vessels, a phenomenon we refer to as “embedding.” Embedding could occur along the entire length of the spirochete, or at one end only. Interestingly, we found that B. burgdorferi embedded in the PECAM-1 region along their entire length adhered for much longer periods than partially embedded bacteria, and were frequently observed protruding through both sides of the PECAM-1 signal (e.g. see the stationary spirochete in Fig. 3B, lower panel), suggesting that they had migrated more deeply into junctions or endothelial cells than partially embedded adhesions. This observation may be consistent with the results of early electron microscopy studies demonstrating that B. burgdorferi can invade or be taken up by endothelial cells in monolayer cultures [15],[19], and is intriguing in light of previous proposals that spirochete evasion of the host immune system is mediated by “seeding” bacteria that escape immune surveillance in physically protected sites (reviewed in [48]).
Endothelial cells can be as thin as 0.1 µm [49] and the PECAM-1 antibody used in this study stained a 3 µm-thick region of the vessel wall. The observation that stationary adhesions often project beyond the PECAM-1-stained region suggests the possibility that these spirochetes are invading junctions or endothelial cells. However, the measured thickness of the PECAM-1 signal may overestimate the dimensions of the endothelium due to motion artifacts caused by respiration of the immobilized mouse. Therefore, we can only conclude that the apparently embedded state of stationary adhesions results from more intimate adhesion to the endothelium than that observed for short-term interactions. Additional higher resolution studies of stationary adhesions, performed under shear stress conditions, will be required to shed light on the true position of the spirochetes relative to the endothelium and the intriguing possibility that stationary adhesion of spirochetes might provide a protective mechanism for evasion of the immune response.
Spirochete escape from the microvasculature was a rare event, even after intravenous inoculation with large doses of B. burgdorferi. Three dimensional timelapse data captured for 30 emigrating spirochetes revealed that B. burgdorferi escaping the microvasculature traversed the vessel wall end-first (Fig. 6D). Interestingly, multiple emigrating spirochetes were sometimes observed in the same vessel (data not shown). It appears unlikely that cases of multiple escape were the result of endothelial activation in response to B. burgdorferi, since these could be observed immediately after intravenous injection of spirochetes. Another possibility is that the presence of nearby stationary adhesions facilitated transmigration, since these adhesions were more abundant in vessels with escaping spirochetes, and as the site of transmigration was frequently in close proximity to a stationary adhesion (data not shown). Stationary adhesions adjacent to escape sites might modify their immediate vascular environment to promote emigration of other spirochetes.
Spirochetes emigrating end-first frequently exhibited a reciprocal translational form of movement that might drive much of the escape process. The average displacement velocity of emigrating spirochetes was 4-fold less than the average displacement velocity of B. burgdorferi translating in the extravascular tissues of the ear, suggesting that the endothelium presents significant physical barriers to transmigration. This conclusion is supported by the observation that the middle phase of escape was very slow relative to early and late stages. Previous work with Leptospira in vitro indicates that even cells adhered to immobile surfaces at a single point can move in a reciprocating fashion referred to as “staple movement”, likely as a result of a rapid (11 µm/s) lateral displacement of the adhesion site within the spirochete outer membrane [37],[50]. Such a model predicts that disruption of the adhesion site would cause a rapid change in the motile behavior of spirochetes, which is consistent with our observation that escaping B. burgdorferi “burst” out of the endothelium after a protracted period of reciprocal motility.
Early studies of B. burgdorferi, T. pallidum and Leptospira transmigration performed with endothelial monolayers found that non-motile spirochetes could not invade endothelium [12]–[17]. This conclusion is supported by the real-time imaging data reported in this study. It will, therefore, be important to directly examine the role of spirochete motility in emigration by coupling the use of the well-characterized B. burgdorferi motility and chemotaxis mutants [35], [51]–[53] with the technology reported here.
We found that live fluorescent spirochetes can easily be observed in situ in living mice one month after subcutaneous and peritoneal inoculation. Furthermore, intravital microscopy can be performed in many of the tissues targeted by B. burgdorferi and other spirochetes, including the brain, liver, lung and joint cartilage [25]. It is, therefore, clear that the methodology described here could be a powerful tool for addressing a broad range of questions about host-pathogen interactions.
In addition to the types of experiments reported here, this methodology could be exploited for the study of a variety of bacterial pathogens in terms of their invasion of the vascular system, interactions with cellular components of the innate and acquired immune responses, for monitoring gene expression and migration patterns in different tissues over the course of infection and for analysis of chemotactic behavior in the host. The methodology may also be useful for monitoring events that occur immediately after needle or tick bite inoculation, routes of spirochete entry that more closely recapitulate the natural infection process than the intravenous injection of high numbers of blood adapted spirochetes.
In summary, dynamic and high resolution three-dimensional analyses of B. burgdorferi behavior in a living host have revealed numerous previously unobserved aspects of spirochete interaction with, and escape from, the host vasculature. The application of this powerful approach to the study of other micro-organisms is certain to enhance our understanding of the broad and always unpredictable repertoire of pathogenic agents and their interactions with their living hosts.
The terminator sequences (T1 x 4), rbs, B. burgdorferi flaB promoter and GFP coding sequences from pCE320(gfp)-PflaB [20] were PCR-amplified with flanking SacI and KpnI sites, using primers B696 (5′-ccggagctcatgataagctgtcaaacatgag-3′) and B697 (5′-ccggtacctcagatctatttgtatagttcatc-3′), and cloned into pCR Blunt II-TOPO (Invitrogen Canada, Burlington, ON) with the insert SacI site proximal to the vector PstI site, to make plasmid pTM41. This insert could not be cloned into the gentamycin-resistant version of the pBSV2 shuttle vector (pBSV2G) [54], presumably because replication origins and copy number sometimes affect the expression and toxicity of fluorescent proteins in E. coli. Therefore, a modified shuttle vector, pTM49, was constructed, in which the colEI ori of pBSV2G was removed by restriction digestion with enzymes MluI and SnaBI, and replaced with an MluI/SnaBI fragment from pCR Blunt II-TOPO containing the pUC ori. The (T1 x 4)-PflaB-gfp cassette from pTM41 was cloned into the SacI/KpnI sites of pTM49 to generate pTM61.
All strains were grown in BSK-II medium prepared in-house [55]. Electrocompetent infectious B. burgdorferi strain B31 5A4 NP1 [56] and non-infectious strain B31-A [57] (both B31-derived) were prepared as described [58]. Liquid plating transformations were performed with 50 µg pTM61 in the presence of 100 µg/ml gentamycin as described [59],[60]. Gentamycin-resistant B. burgdorferi clones were screened for: 1) the presence of aacC1 sequences by colony screening PCR performed with primers B348 and B349 as described [61]; and 2) GFP expression by conventional epifluorescence microscopy. The presence of the pTM61 plasmid in non-integrated form in fluorescent strains was confirmed by agarose gel electrophoresis of total genomic DNA prepared on a small scale as described [62]. PCR screening for native plasmid content was performed as described [61],[63] and indicated that one fluorescent infectious B. burgdorferi clone (GCB726) contained all endogenous plasmids except cp9, which was displaced by the cp9-based pTM61 construct. Non-infectious strain GCB705 was used for experiments with non-infectious B. burgdorferi. PCR screening for native plasmid content indicated that GCB705 contained the same plasmids as the B31-A parent [61] (lp17, lp28-2, lp28-3, lp38, lp54, lp56, cp26, cp32-1, cp32-2/7, cp32-3 and cp32-9, but not lp21, lp25, lp28-1, lp28-4, lp36, cp9, cp32-6 or cp32-8). Plasmids lp25, lp28-1 and lp36 are known to be essential for infectivity [63],[64].
All animal studies were carried out in accordance with approved protocols from the University of Calgary Animal Research Centre. C3H/HeN (Harlan, Indianapolis, IN) and C57 BL/6 (Jackson Laboratory, Bar Harbor, ME) mice were infected by both intraperitoneal (5×104 cells/ml) and subcutaneous (5×103 cells/ml) needle inoculation. Ear punches and flank skin samples were cultured in Barbour-Stoenner-Kelly II (BSK-II) medium supplemented with 6% rabbit serum (Cedarlane Laboratories Ltd., Burlington, ON) with or without 100 µg/ml gentamycin. The percentage of ex vivo spirochetes that continued to express robust levels of GFP was calculated by counting the number of fluorescent spirochetes at 100 ms exposures compared to the number detected by phase contrast visualization.
For each experiment, infectious or non-infectious strains expressing GFP were freshly inoculated from glycerol stocks into 15 ml BSK-II medium containing 6% rabbit serum and 100 µg/ml gentamycin. B. burgdorferi were grown to 5×107/ml, then diluted to 1–2×106/ml in BSK-II medium containing 6% rabbit serum, 100 µg/ml gentamycin, 1× Borrelia antibiotic mixture (20 µg/ml phosphomycin, 50 µg/ml rifampicin and 2.5 µg/ml amphotericin B, prepared from individual antibiotics obtained from Sigma) and 1% C57 BL/6 mouse blood. Spirochetes were grown in the mouse blood for 48 hours at 35°C to a final density of ∼5×107/ml. B. burgdorferi were pelleted (6,000×g for 15 min at 4°C), washed twice in PBS (Invitrogen Canada, Burlington, ON), and resuspended to 2×109 B. burgdorferi/ml in PBS. All experiments were performed at a final density of ∼1×107 spirochetes/ml of blood to facilitate quantitative analysis of interactions.
Animals were anaesthetized by intraperitoneal injection of a mixture of 10 mg/kg xylazine hydrochloride (MTC Pharmaceuticals, Cambridge, ON) and 200 mg/kg ketamine hydrochloride (Rogar/STB, London, ON). As previously described [65], a depilatory solution (Nair; Armkel LLC) was applied to the dorsal and ventral surfaces of the ear. After 10 min, the solution was gently removed using 0.9% normal saline and cotton swabs. The ear was mounted against the adjustable plexiglass microscope pedestal and held in place under a coverslip. Mouse rectal temperature was monitored via rectal thermometer and maintained at 37°C using a self-regulating heating mat.
The microcirculation of the ventral abdominal skin was prepared for microscopy as previously described [66]. Mice were anaesthetized and body temperature was monitored as described above. Briefly, after shaving a midline abdominal incision was made extending from the pelvic region up to the level of the clavicle. The skin was separated from the underlying tissue, remaining attached laterally to ensure the blood supply remained intact. The area of skin was then extended over a viewing pedestal and secured along the edges using 4.0 sutures. The loose connective tissue lying on top of the dermal microvasculature was carefully removed by dissection under an operating microscope. The exposed dermal microvasculature was immersed in isotonic saline and covered with a coverslip held in place with vacuum grease. The right jugular vein was cannulated to administer additional anaesthetic and fluorescent dyes. To visualize B. burgdorferi-endothelial interactions, 4×108 spirochetes in 200 µl of PBS were injected directly into the jugular or femoral veins of anaesthetized mice. Three to six dermal venules (15–45 µm in diameter) were selected in each experiment.
Conventional epifluorescence microscopy was performed with a Leica DM IRE2 inverted microscope (Leica Microsystems, Frankfurt, Germany) equipped with an Orca ER cooled CCD camera (Hamamatsu, McHenry IL), using a 63× oil immersion objective, a narrow band GFP filter (480 −/+ 10 nm excitation wavelength; 510 −/+ nm emission wavelength: Chroma Technology Corp, Rockingham, VT) and exposure times of 100 ms. Sixteen-bit images were acquired using OpenLab 5.0.2 (Improvision Inc., Lexington, MA), and exported images in .tiff format were converted to 8-bit, colorized using indexed color and cropped in Adobe Photoshop CS prior to export and conversion to CYMK mode in Adobe Illustrator CS (Adobe Systems Inc., San Jose, CA). Identical image capture and adjustment settings were used for all images.
Conventional epifluorescence intravital microscopy was performed using a Zeiss Axioskop microscope equipped with a 40× Wetzlar water immersion lens (Carl Zeiss Canada Ltd., Toronto, ON). Manual focusing was used to ensure that spirochetes remained in the focal plane throughout recording. A video camera (HS model 5100; Panasonic, Osaka, Japan) was used to project the images onto a monitor, and the images were recorded at 29.97 fps for off-line video playback analysis using a videocassette recorder. VHS analogue videos of conventional IVM experiments were converted to digital format using Windows Movie Maker (Microsoft Corporation, Redmond WA), and converted to .swf format using Macromedia Flash Professional 8 (Macromedia Inc., San Francisco, CA) without altering frame rate or editing frame sequence. Leukocyte recruitment was monitored by rhodamine staining of leukocytes, as previously described [67].
Spinning disk confocal intravital microscopy [68] was performed using an Olympus BX51 (Olympus, Center Valley, PA) upright microscope equipped with a 20×/0.95 XLUM Plan Fl water immersion objective. The microscope was equipped with a confocal light path (WaveFx, Quorum, Guelph, ON) based on a modified Yokogawa CSU-10 head (Yokogawa Electric Corporation, Tokyo, Japan). Endothelial cells and junctions were labeled with a monoclonal anti-PECAM-1 antibody (Fitzgerald Industries International, Inc., Concord, MA), conjugated to Alexa Fluor 555 (Molecular Probes, Invitrogen Canada, Burlington, ON). One hundred µl of Alexa-conjugated anti-PECAM-1 were injected per mouse (50 µg/mouse). In some experiments, 50 µl of 5 mg/ml FITC-albumin in normal saline (Sigma-Aldrich Canada Ltd., Oakville, ON) was injected to visualize blood vessels (250 µg/mouse). Laser excitation at 488 and 561 nm (Cobalt, Stockholm, Sweden), was used in rapid succession and fluorescence in red and green channels was visualized with the appropriate long pass filters (Semrock, Rochester, NY). Emission wavelengths for red and green channels were 593 nm and 520 nm, respectively, and no overlapping signal was detected in either channel. Exposure time for both wavelengths was 168 ms. A 512×512 pixels back-thinned EMCCD camera (C9100-13, Hamamatsu, Bridgewater, NJ) was used for fluorescence detection. Volocity Acquisition software (Improvision Inc., Lexington, MA) was used to drive the confocal microscope. Sensitivity settings were 255 and 251 for red and green, respectively, and autocontrast was used. Images were captured at 16 bits/channel in RGB. For timelapse series, manual focusing was used to ensure that spirochetes remained in the focal plane throughout recording. Red and green channels were overlaid using brightest point settings before export in .tiff or .mov format. Overlaid GFP and Alexa Fluor 555 .tiff images exported from Volocity were cropped in Adobe Photoshop CS without manipulation of signal levels or contrast prior to export and conversion to CYMK mode in Adobe Illustrator CS. Exported .mov files were imported without editing directly into Macromedia Flash Professional 8 for labeling and export as .swf files.
Z-series were collected using spinning disk confocal IVM, with images captured in both red and green channels for each slice. All image acquisition settings were as described above, except where noted in the Specific Image Acquisition Settings section, below. The localization of B. burgdorferi relative to the lumen, endothelium, endothelial junctions and extravascular tissue was scored for each z-slice in the series, using xy, xz and yz images constructed from the z-section series using Volocity 4.0.2. Scoring was performed independently by two individuals. Three-dimensional volume rendering (voltex) reconstruction of spirochetes in venules was performed in Amira 4.1.1 (Mercury Computer Systems, Chelmsford, MA) using series of GFP and Alexa Fluor 555 .tiff images exported separately from Volocity. Alpha and gamma settings were 1, and GFP and Alexa Fluor sensitivities were, respectively, 30–170 and 30–225. Animated rotation views of 3D volume rendering were exported as .mpeg files prior to import and labeling in Macromedia Flash Professional 8.
All images were acquired and processed as described in the preceding sections. All timelapse series were captured at 0.94 fps, and exported at 5 fps, except the timelapse presented in Video S1, which was captured at 5.9 fps and exported at 50 fps. Other parameters: Fig. 1C: 0.485 µm/pixel (x and y); Fig. 2: 0.485 µm/pixel (x and y); Fig. 3B: 37 z-slices (45.3 sec/series), 5 µm step size, 0.485 µm/pixel (x and y), 1 µm/pixel (z); Fig. 4A: 19 z-slices (23 sec/series), 1 µm step size, 0.485 µm/pixel (x and y), 1 µm/pixel (z); Fig. 5A: 81 z-slices/time point (71 sec/stack), 0.5 µm step size, 0.485 µm/pixel (x and y), 0.5 µm/pixel (z); Video S1: settings same as in Fig. 1C; Video S2: settings same as in Fig.2; Video S3: 0.485 µm/pixel (x and y); Video S5: settings same as in Fig. 3b; Video S6: 0.485 µm/pixel (x and y).
For quantitative analysis, average and standard error values for different variables were calculated and plotted graphically for all vessels from all mice using GraphPad Prism 4.03 (GraphPad Software, Inc., San Diego, CA). Statistical significance was calculated in GraphPad Prism using a two-tailed non-parametric t-test with a 95% confidence interval. |
10.1371/journal.pgen.1004408 | Coordination of Wing and Whole-Body Development at Developmental Milestones Ensures Robustness against Environmental and Physiological Perturbations | Development produces correctly patterned tissues under a wide range of conditions that alter the rate of development in the whole body. We propose two hypotheses through which tissue patterning could be coordinated with whole-body development to generate this robustness. Our first hypothesis states that tissue patterning is tightly coordinated with whole-body development over time. The second hypothesis is that tissue patterning aligns at developmental milestones. To distinguish between our two hypotheses, we developed a staging scheme for the wing imaginal discs of Drosophila larvae using the expression of canonical patterning genes, linking our scheme to three whole-body developmental events: moulting, larval wandering and pupariation. We used our scheme to explore how the progression of pattern changes when developmental time is altered either by changing temperature or by altering the timing of hormone synthesis that drives developmental progression. We found the expression pattern in the wing disc always aligned at moulting and pupariation, indicating that these key developmental events represent milestones. Between these milestones, the progression of pattern showed greater variability in response to changes in temperature and alterations in physiology. Furthermore, our data showed that discs from wandering larvae showed greater variability in patterning stage. Thus for wing disc patterning, wandering does not appear to be a developmental milestone. Our findings reveal that tissue patterning remains robust against environmental and physiological perturbations by aligning at developmental milestones. Furthermore, our work provides an important glimpse into how the development of individual tissues is coordinated with the body as a whole.
| Between distantly related species, development converges at common morphological and genetic stages, called developmental milestones, to ensure the establishment of a basic body plan. Beyond these milestones greater variability in developmental processes builds species-specific form. We reasoned that developmental milestones might also act within a species to achieve robustness against environmental or physiological perturbation. To address this, we first developed a staging scheme for the progression of pattern in the wing disc across developmental time. We then explored how perturbing environmental or physiological stimuli known to alter the rate of development affected the progression of pattern in the wing disc. We found two developmental milestones, the moult to the third instar and pupariation, where wing disc patterning aligned with the development of the whole body. This suggests that robustness against environmental and physiological conditions is achieved by coordinating tissue with whole-body development at developmental milestones.
| Organisms require robust developmental processes to guarantee that developing tissues pattern correctly in the face of a wide range of environmental and physiological perturbations [1], [2]. A developmental process can be considered robust if variation in this process is uncorrelated with variation in genetic, environmental or physiological conditions [3]. To achieve robustness, the developmental processes that generate individual organs must, at some level, be integrated across the whole body to ensure that a correctly patterned and proportioned adult is produced at the end of development. It is therefore thought that the progression of gene expression that occurs in tissues as they pattern needs to be somehow integrated with the systemic hormone levels that trigger transitions between developmental stages (hereafter termed developmental events) across the whole body [4], [5]. The timing of these developmental events changes with environmental and physiological conditions but how this affects tissue development is not fully understood.
There are several hypotheses to explain how tissue patterning is integrated with whole-body development under different environmental and physiological conditions. One hypothesis is that tissue patterning and whole-body development progress synchronously, so that the rate of the former matches the rate of the latter. If this were the case, a change in the duration of development would extend or contract the progression of patterning in a linear manner (Figure 1a). Consequently, normalizing the progression of pattern to a developmental endpoint, that is using relative rather than absolute developmental time, would produce the same progression of patterning independent of the duration of development (Figure 1b).
Alternatively, tissue patterning may only be coordinated with whole-body development at key developmental events (Figure 1c), for example moulting in holometabolous insects, or the onset of puberty in humans. Although not all developmental events act to coordinate, those that do are often referred to as developmental milestones [6]. Thus if the duration of development varies, the progression of patterning would nonetheless converge at these milestones while showing greater variability between them. Consequently, normalizing the progression of pattern to relative developmental time would produce patterns that overlapped only at developmental milestones (Figure 1d). This would essentially mean that if patterning were to drift in rate with respect to whole-body development, developmental milestones would ensure that the rate of patterning would decelerate or accelerate to achieve the correct stage by the onset of the milestone.
Problematically, it has been difficult to test these alternative hypotheses because, while the process of patterning has been described in exquisite detail in a variety of tissues, the dynamics of patterning is rarely tied to organismal age or whole-body physiology. Several authors have explored how genetic background contributes to the robustness of development (see examples [7], [8]). Their approaches have focussed on the endpoints of development and on changes in the sequences of specific patterning cascades. Furthermore, studies in organisms ranging from insects to nematodes to vertebrates have explored the progression of gene expression in relation to embryonic stage to identify developmental milestones, called phylotypic stages, where gene expression converges upon an embryonic stage common across species [6], [9]–[12]. Such developmental milestones are thought to constrain development like an hourglass, as development across species varies more both before and after the milestones [6], [9], [10], [12]. However, these studies do not address how environmental/physiological conditions affect the progression and sequence of pattern, and how this is coordinated with whole-body development within a species. We therefore took advantage of the extensive knowledge of tissue patterning and whole body physiology of the fruit fly, Drosophila melanogaster, to elucidate the extent to which tissue patterning is coupled with whole-body development.
In Drosophila, the juvenile period comprises three larval moults. This is followed by a wandering stage where larvae leave the food and search for a pupariation site. Larval development ends with pupariation, whereupon the fly metamorphoses into its adult form. These events provide useful markers of whole-body development. Each of these developmental events (moulting, wandering and pupariation) is regulated by pulses in the titre of the steroid hormone ecdysone [13], synthesized by the prothoracic gland.
Most of the adult tissues of Drosophila arise from pouches of cells that grow and pattern within the body of the developing larvae, the imaginal discs [14]–[17]. Pulses of ecdysone have also been shown to regulate some stages of imaginal disc development. Early in the third larval instar a pulse of ecdysone controls the expression of three patterning gene products, Cut (Ct), Senseless (Sens) and Wingless (Wg), in response to nutrition [17]. After pupariation, ecdysone regulates Sens expression to control the differentiation of sensory organs in the wing [15], [16]. Thus, these pulses of ecdysone have been interpreted to be checkpoints that coordinate the patterning and development of tissues with whole-body developmental events [5], [18]. Nevertheless, it remains to be determined if this coordination between tissues and the whole body is necessary and happens at all developmental events, or only at specific developmental milestones.
The rate of developmental progression and the timing of these developmental events can be altered both environmentally and by genetically manipulating the timing of ecdysone synthesis. For example, Drosophila larvae raised at lower temperatures take longer to eclose as adults [19]–[23] while larvae reared at higher temperatures eclose more quickly [20], [22]. Similarly, altering the timing of ecdysone synthesis, by suppressing or activating insulin signalling in the prothoracic gland, also changes developmental timing and retards or accelerates eclosion [17], [24], [25].
To test the extent to which whole-body development and the progression of pattern in individual tissues are coordinated, we first generated a staging scheme to describe how patterning progresses over time in the wing imaginal discs of third instar larvae. This staging scheme was based on the changes in expression pattern of key patterning genes. We then altered developmental rate either environmentally, by using temperature manipulations, or physiologically, by altering the timing of ecdysone synthesis. We compared the progression of patterning, as determined by our staging scheme, in larvae that differ in their developmental rates. Our results indicate that the progression of patterning is coordinated with some, but not all, developmental events and varies between events.
To compose our developmental staging scheme for wing discs, we used immunocytochemistry to identify changes in the expression of eleven patterning gene products at five hour intervals from 0–40 h after third instar ecdysis (AL3E), at wandering, and at pupariation for a total of eleven time points (Figure 2 and Supplementary Figs S1, S2, S3). We used wing discs from larvae of an isogenic wild-type strain Samarkand (SAM) reared at 25°C (wild type at 25°C). Three of these time points coincided with three developmental events – the moult to the third instar, wandering and pupariation. We have a strong understanding of the physiology underlying these developmental events, and so assaying patterning at these time points allowed us to test for coordination between tissue patterning and whole-body development. Collectively, we used the progression of patterning in wild type at 25°C as a baseline for all comparisons in this work.
We identified elements of pattern that we could reliably distinguish across discs of a given time point (Figs 2, 3). New elements of pattern included the addition of a new region of expression, for instance the appearance of expression in a cell or in cells that previously had not expressed a particular gene product; the refinement of an expression field from diffuse expression in a group of cells to more focussed expression in a reduced subset of cells; or the disappearance of expression in a region that had previously expressed that gene product. For each patterning gene product, we discerned the time each patterning element arose, thereby characterizing the transitions in pattern for each gene. From this, we defined stages for each gene product (referred to as gene-specific stages) (Figure 3).
Not all gene products displayed clear gene-specific stages. Engrailed and Patched did not undergo patterning transitions in the third larval instar, consistent with previous studies [17]. Scabrous localization within single cells appeared to be restricted to vesicles, making changes in pattern hard to identify. Hindsight expression in the wing disc was difficult to distinguish from expression in associated tracheal cells. Finally, the patterning transitions for Delta and Notch (N) occurred at the same time. For these reasons, we chose to exclude Engrailed, Patched, Scabrous, Hindsight, and Delta from our characterizations of overall disc stage.
We tabulated the gene-specific stages for each time point from the remaining six gene products, Achaete (Ac), Ct, N, Sens, Dachshund (Dac) and Wg. These combinations of gene-specific stages allowed us to define eleven disc stages (A-K), corresponding to each of the eleven time points sampled from wild-type larvae at 25°C (Figure 3 and see Materials and Methods).
Two of the gene products, Ac and Sens were staged simultaneously in individual discs (Ac is a mouse monoclonal antibody and Sens is a guinea pig antibody). Using these two gene products alone, we can assign discs to nine of the eleven disc stages (Figure 4a). The bubbles in Figure 4a represent the proportion of discs at each time point that fall into a particular disc stage based on their Ac and Sens pattern combined. These data show that using Ac and Sens alone, for five time points all discs are categorized into a single stage. For the remaining six time points sampled, most discs (67–89%) can be attributed to one disc stage, with a smaller proportion of discs (<24%) falling into one or two additional stages. Thus, staging with Ac and Sens alone provides a reliable measure of disc stage across developmental time.
We expected that adding more markers to our staging scheme would increase its resolution. Problematically, due to the nature of antibody staining, it was not possible to stain a single disc for more than two gene products. Consequently, we cannot assign an individual disc to a particular developmental stage with the complete set of markers. To circumvent this problem, we simulated what a disc would look like if we could stain the same disc for all six gene products. We first tabulated the observed stages for each gene product at each time point. The number of discs scored for each gene product ranged from five to sixteen (Supplementary Table S1), depending on the time point and the gene product. We then randomly sampled from this table to simulate all the possible combinations of gene-specific stages for a single disc dissected at this time point. We repeated these permutations 1000 times to generate 1000 simulated discs for each time point. We then applied a Naïve Bayes Classifier (NBC) to the simulated data set to assign each simulated disc to a developmental stage, based on our staging scheme. The NBC analysis does not return a p-value, but instead provides the probability that a disc of a given time point would be assigned to a particular disc stage.
The results of this analysis are represented using a bubble plot (Figure 4b-d). In this plot, the area of each bubble is the proportion of the 1000 simulated discs that were assigned to each disc stage, using the NBC. As a proof of principle, we applied our analysis to the staging scheme devised from the Ac and Sens data. The plot generated from the simulated discs looks very similar to the staging scheme derived from the sampled disc data (Figure 4a, b), although the NBC appears to slightly overestimate the amount of variation in the data (dashed boxes in Figure 4b). Overall, however, our stimulated data set represents well the patterns seen from the sampled discs.
Next, we simulated discs with all six patterning gene products and applied the NBC (Figure 4c). Using all six gene products, we could resolve eleven disc stages in the simulated discs. For six time points, there is a single bubble, indicating that all the simulated discs at that time point share a stage-specific combination of gene-product patterns. This suggests that the criteria for classification are unambiguous at that time point. In the remaining time points, the NBC assigned discs to two or three stages. This indicates that the discs dissected at these time points did not all share the characteristics used to define a single stage. That is, there is variability in patterning among discs dissected at the same time point. Nevertheless, even at these time points the NBC classified the majority of simulated discs (65–94%) to a single stage. Further, the amount of variation for these time points was reduced if the complete data set was used in the simulation instead of using Ac and Sens alone.
We repeated the NBC analysis using only the expression patterns of Ac, Sens and Dac to classify the discs. The results were nearly identical from the complete gene set simulations (Figure 4d), except the NBC classified all discs at 30 h AL3E (stage G) as stage F. This is because stages G and F share the same Ac, Sens and Dac expression pattern, and so the NBC classified the discs into the earliest stage by default. This combination of three gene products provides greater resolution than Ac and Sens alone and was one of the combinations that identified most of the disc stages from the moult to the third instar until pupariation. Hereafter, to minimize the number of gene products necessary to stage wing discs, we established the staging scheme composed from Ac, Sens and Dac as the baseline for all subsequent comparisons. Additionally, we choose to use Wg for the first time point because Ac, Sens and Dac were not expressed at the moult to the third instar.
Once we had a method of defining the developmental stage of a disc, we then asked whether the progression of pattern through these developmental stages was tightly coordinated with whole-body development when developmental rate was altered by changes in rearing temperature. Rearing wild-type larvae at 18°C lengthened the time to adult eclosion from larval hatching, while rearing larvae at 29°C shortened the time, compared to wild-type larvae raised at 25°C (Figure 5). Surprisingly, however, the duration of the third larval instar was slightly longer at 29°C than at 25°C (Figure 5), as was the time to larval wandering from the beginning of the third instar. Thus, for the purposes of our study, larvae reared at 29°C were slow developers.
To assay whether the progression of disc patterning relative to whole-body development was affected by rearing temperature, we used a bubble plot to chart wing disc stage, as assigned by the NBC classifier applied to a permuted data set, expressed in relative developmental time (normalized to pupariation), at 18°C, 25°C and 29°C. At all three temperatures, patterning in the discs was the same at the moult to the third instar and at pupariation (Figure 6a, b and Supplementary Figure S4a, b). At 18°C the progression of disc patterning when normalized to pupariation time was largely the same as at 25°C, indicated by the overlapping bubble plots at the two temperatures (Figure 6a and Supplementary Figure S5). In contrast, at 29°C patterning was initially delayed, evident from discs dissected at the same relative developmental time showing earlier patterning stages at 29°C than at 25°C (Figure 6b and Supplementary Figure S4b). The rate of patterning progression accelerated later in the third instar, however, to achieve the final disc stage at pupariation (Figure 6b and Supplementary Figure S4b). Further, there was more variation in developmental stage among discs dissected at larval wandering at 29°C, compared to 25°C (Figure 6b). Earlier in development, the variation and delay observed in disc stage at 29°C was due to Ac and Sens expression, both of which belong to the Notch signalling pathway (Supplementary Figure S6). In contrast, at wandering much of the delay was caused by variation observed in Sens and Dac expression patterns (Supplementary Figs S6, S7).
The timing of ecdysone synthesis is thought to be key to coordinating whole-body developmental events (moulting, larval wandering and pupariation) with imaginal disc development. To test this hypothesis, we first altered the timing of ecdysone synthesis by downregulating or upregulating insulin signalling in the prothoracic gland, lengthening or shortening the duration of the third larval instar respectively (Figure 5) [17]. To downregulate insulin signalling in the prothoracic gland, we used the P0206 GAL4 driver to overexpress PTEN (P0206>PTEN); to upregulate insulin signalling in this tissue, we expressed InR using the phm GAL4 driver (phm>InR). Together with changes in the duration of development, the rate of patterning in the wing discs was also affected. Early in the third larval instar, patterning appeared to be retarded in both phm>InR and P0206>PTEN larvae, while patterning progressed at an accelerated rate later in development (Figure 6c, d and Supplementary Figure S4c, d).
To explore how wing disc patterning progressed relative to whole-body development, we again used a bubble plot to chart wing disc stage in phm>InR and P0206>PTEN, as assigned by the NBC classifier applied to a permuted data set, against relative developmental time. We used wild-type SAM larvae reared at 25°C for comparison. Under all experimental conditions, wing discs displayed the same pattern at the beginning (moulting) and end (pupariation) of the third larval instar. However, a bubble plot of relative developmental time (normalized to pupariation) against disc stage indicated that in both P0206>PTEN and phm>InR larvae, disc patterning is initially delayed and showed increased variability compared to 25°C wild-type larvae at the same relative developmental time (Figure 6c, d and Supplementary Figs. S4, S8, S9). This delay is more evident in phm>InR discs (Figure 6d), where it is due to changes in the relative progression of Ac, Sens and Dac expression (Supplementary Figure S9), than in P0206>PTEN discs (Figure 6c), where it is primarily due to changes in the progression of Ac and Sens expression (Supplementary Figure S8). Furthermore, in phm>InR discs from wandering larvae, patterning was substantially delayed when compared to wild type at 25°C (Figure 6d).
Some of the observed changes in wing disc patterning progression early in the third instar in P0206>PTEN and phm>InR larvae may result from genetic background effects. Both parental lines, yw; UAS PTEN (referred to as >PTEN) and yw flp; UAS InR29.4 (referred to as >InR), showed small but significant differences in pupariation time compared to the wild type at 25°C (Figure 5). Additionally, in both >PTEN and >InR larvae, we observed early delays in wing patterning relative to wild type at 25°C, due to retardation in the progression of all three gene products – Ac, Sens and Dac (Supplementary Figs S4e, f, S10, S11). However, after 50% developmental time wing disc patterning was the same in all three lines (wild type at 25°C, >PTEN, >InR). Further, wing disc patterning was the same in all three lines at moulting and pupariation, and largely overlapped at wandering (Figure 6e, f). A comparison of wing disc patterning in P0206>PTEN and phm>InR larvae to their genetic controls suggests that the delays observed before 50% relative developmental time are due to genetic background effects while the delays after this period are due to changes in physiology (Supplementary Figure S12).
Our data demonstrate that altering developmental timing of the whole body changes the progression of patterning in Ac, Sens and Dac. Next, we explored whether gene-specific stages of Sens correlated with gene-specific stages of Ac across treatments and genotypes independently of developmental time (Figure 7). We found that overall, Ac and Sens stages were tightly correlated and showed little significant variation with temperature, physiology or genotype. There were some exceptions; for Ac stages 4 and 5 we found that Sens stages were significantly delayed in P0206>PTEN larvae when compared to wild-type larvae at 25°C (Figure 7c). The >InR larvae showed similar delays in Sens with respect to Ac at stage 5 (Figure 7f). In contrast at Ac stage 6, Sens was accelerated in the wild-type larvae at 18°C and in the P0206>PTEN larvae (Figure 7a, c). Thus, Sens stages show some degree of plasticity with respect to Ac stages, but only at Ac stages 4–6.
In this study, we set out to examine the extent to which tissue development is coordinated with the development of the whole body. We tested two alternative hypotheses: 1) the progression of pattern is tightly coordinated with whole-body development at all times, and 2) patterning is coordinated only at developmental milestones.
Previous studies demonstrated that the development of tissues could regulate the timing of whole animal development. Specifically, larvae with slow growing discs greatly delay the development of the whole body [26]–[29]. Discs induce these delays by regulating the timing of a specific developmental event that occurs early in the third instar, termed critical weight [26], [28]. Slowing disc growth after critical weight has no effect on developmental timing [28]. Delaying patterning in the imaginal discs has also been shown to retard the development of the whole body. If the spread of Wg protein is restricted in the imaginal discs by replacing wild-type Wg with a membrane-tethered Wg allele, larvae delay the onset of pupariation [30]. We do not yet know whether Wg signalling in the discs affects developmental timing by affecting disc growth rate nor do we know which developmental events are affected by altered Wg signalling.
Further, there is ample evidence from many insects that ecdysone controls the timing of development in the various tissues of the body [13]. In third instar larvae, ecdysone signalling stimulates neurogenesis in the optic lobe via the Notch/Delta pathway [31]. The pulses of ecdysone that stimulate the onset of pupal development are also known to initiate patterning of the sensory tissues of the wing [16]. Thus, it seemed likely that ecdysone pulses at other stages could act as milestones to coordinate both tissue and whole-body development.
We found that patterning, as determined by disc stage, aligned at the moult to third instar and at pupariation in all conditions studied. It is important to note, however, that considerable patterning occurs in wing discs before the third instar [32]. Furthermore, pupariation is not an endpoint for disc pattern, as the patterning of sensory structures and the specification of the wing veins continue on during pre-pupal and pupal development [15], [16], [33]. Thus pupariation appears to be characterized by an alignment but not termination of patterning progression.
In contrast, disc patterning among wandering larvae showed variability, both within the wild type at 25°C and across experimental treatments. Variation in disc stage at wandering within the reference genotype at 25°C is likely to be due to the fact that the wandering stage lasts approximately 8 hours and therefore occupies a slightly longer time interval than the other intervals of the staging scheme. This, however, does not explain the difference in disc stage at wandering across experimental treatments; discs from phm>InR larvae were mostly at disc stage H at wandering, whereas the wild-type discs at 25°C were mostly at disc stage J. Thus, we conclude that wing patterning is not coordinated with whole-body development at wandering. This was surprising, as wandering is commonly used to stage larvae to ostensibly the same developmental point (for examples see [34]–[36]). Overall, our data supports hypothesis two: patterning aligns with whole-body development at specific developmental milestones, the moult and pupariation, and shows greater variation between these milestones.
Variability in pattern between the moult and pupariation showed common characteristics across treatments and genotypes. Generally, patterning showed delays relative to whole-body development early in the third instar. Disc patterning accelerated relative to whole-body development towards the end of the third instar to reach the final stage at pupariation (Figure 6 and Supplementary Figure S4). Our data highlight the possibility that because perturbations in pattern occur through delays early in the third instar, there is an intrinsic checkpoint late in the third instar that regulates pattern in the discs so that they reach a common patterning stage at pupariation.
The progression of pattern also varied with genetic background. This variation between control genotypes was most apparent early in development. In contrast to the environmental/physiological treatments, patterning was, however, aligned at wandering. This observation suggests that our staging scheme would vary somewhat with the genotype chosen as the reference background. Genetic variation in the mechanisms controlling developmental robustness has been previously described in the context of evolutionary studies. For instance, in Caenorhabditis elegans the types of deviations observed during the highly robust process of vulval development depend on genetic background [37]. We expect that genetic variation in the progression of patterning systems is common, but that it is often undetected due to alignment at developmental milestones.
Many of the delays in the progression of pattern that we observed across developmental time were due to delays in two genes from the same pathway, Ac and Sens [38], [39]. This likely reflects the observation that the progression of patterning in these two genes was correlated, independent of developmental time. Consequently, when one gene was delayed, so was the other. In contrast, delays in Dac expression tended to occur at later stages of development. Taken together, this raises the question of how environmental perturbations might affect gene expression within or between signalling pathways as an interesting avenue for future study.
Collectively, our data reveal that tissue patterning is coordinated with some but not all whole-body developmental events. This raises two questions: first, across all of development which whole-body developmental events are developmental milestones for tissues? Second, do all tissues align their development to the same milestones?
Because many developmental events are regulated by ecdysone, whether or not a tissue aligns its pattern to a particular developmental event may be due to its sensitivity to ecdysone at that time. The response of a tissue to a given ecdysone pulse is likely to be tied to its function. If we had examined the development of tissues that have functions in the larvae, we might have found tighter coordination with wandering. For example, the pulse of ecdysone that initiates larval wandering also coordinates the onset of autophagy in the fat body [40]. Autophagy in this tissue is thought to sustain the growth and development of other tissues during non-feeding stages [41]. In the salivary glands, a pulse of ecdysone in the mid-third instar stimulates glue production, while the pulse at larval wandering induces movement of the glue from the cells into the lumen of the gland [42]. This glue is then expelled in response to the ecdysone pulse at pupariation to cement the animal to the substrate. Consequently, development of the fat body and salivary glands may be tightly coordinated with larval wandering. In contrast, tissues like the imaginal discs, whose differentiation into their adult form only starts after pupariation, may not need to respond to these earlier ecdysone pulses.
Despite the striking effects that environmental and physiological changes induce in developmental timing, the resulting adults bear correctly patterned structures. We originally presumed that this was because developmental time and patterning of the tissues was tightly coordinated. Using our staging scheme, however, we have shown that patterning and whole-body development are coordinated only at moulting and pupariation, suggesting these events mark milestones during development. A third event, wandering, does not appear to act as a developmental milestone, at least as far as wing disc patterning is concerned. We also found that the progression of pattern in the wing disc is far more plastic than originally supposed. Further, we found that both the duration of developmental intervals and rates of patterning can be slowed down or sped up. Thus underlying the robustness of the adult phenotype, we have revealed that developmental milestones coordinate wing disc and whole-body development to cope with environmental and physiological variation.
We used an isogenic wild-type strain, Samarkand (SAM), reared at 25°C to develop the staging scheme, representing the baseline for all comparisons (referred to as wild type at 25°C). To manipulate developmental time environmentally, we reared wild-type SAM flies at 18°C and 29°C (wild type at 18°C and wild type at 29°C). To alter the timing of ecdysone synthesis and manipulate developmental time physiologically, we used the progeny from phm-GAL4 crossed with yw flp; UAS InR29.4 (phm>InR) and from P0206-GAL4 crossed with yw; UAS PTEN (P0206>PTEN) to up- or down-regulate insulin signalling in the prothoracic gland, respectively. Even though P0206-GAL4 is a weaker GAL4 driver for the prothoracic gland and also drives expression in the corpora allata, we chose to use it to drive UAS PTEN because phm>PTEN larvae die as first instar larvae [25]. We used the parental lines yw; UAS PTEN (>PTEN) or yw flp; UAS InR29.4 (>InR) as additional controls for genetic background effects.
Flies were raised from timed egg collections (2–6 hours) on standard cornmeal/molasses medium at low density (200 eggs per 60×15 mm Petri dish) in a 12 h light-dark cycle with 70% humidity, and maintained at 25°C unless stated otherwise. Larvae that were reared at 18°C or 29°C were maintained in incubators without lights due to equipment constraints.
Larvae were staged into 1-hour cohorts at ecdysis to the third larval instar and wing-imaginal discs were dissected at the following times (in h AL3E): wild type at 25°C: 0, 5, 10, 15, 20, 25, 30, 35, 40, 46 (wandering) and 49 (pupariation) h AL3E; wild type at 18°C: 0, 10, 20, 30, 50, 70, 96 (wandering) and 101 (pupariation) h AL3E; wild type at 29°C: 0, 10, 20, 30, 35, 40 h, 48 (wandering) and 52 (pupariation) h AL3E; P0206>PTEN: 0, 10, 20, 30, 40, 60, 73 (wandering) and 80 (pupariation) h AL3E; phm>InR: at 0, 10, 20 and 30 h AL3E, 32 (wandering) and 36 (pupariation) h AL3E; >PTEN control: 0, 10, 20, 30, 48 (wandering) and 53 (pupariation) h AL3E; >InR control: 0, 10, 20, 30, 48 (wandering) and 51 (pupariation) h AL3E.
We measured the average time to wandering and pupariation by counting the number of larvae wandering/pupariating within a cohort every two hours. To measure the average eclosion time, we allowed flies to oviposit for 2–6 hours in food bottles. Larvae were maintained at low densities, and we checked for adult eclosion every 12 h.
To develop our staging scheme, we examined the expression of eleven patterning gene products in the wing discs of wild-type larvae at 25°C by immunocytochemistry: Achaete (Ac), Cut (Ct), Delta (Dl), Hindsight (Hnt), Notch (N), Scabrous (Sca), Senseless (Sens), Dachshund (Dac), Engrailed (En), Patched (Ptc) and Wingless (Wg). These patterning gene products represent the main cascades involved in wing disc patterning: the Notch signalling pathway (represented by Ac, Ct, Dl, Hnt, N, Sca and Sens), the Hedgehog signalling pathway (represented by Dac, En and Ptc) and the Wnt/Wg signalling pathway (represented by Wg). In the wing discs of larvae with altered developmental time (wild type at 18°C, wild type at 29°C, P0206>PTEN and phm>InR) as well as the genetic controls (>PTEN and >InR), we examined the expression of four gene products: Wg for the 0 h AL3E time point, and Ac, Sens and Dac for all time points. Although it was impossible to simultaneously stain for all gene products at all time points for all genotypes under all conditions, we minimized the effects of variation between experimental blocks by conducting experiments between at least two genotypes/conditions in parallel. Further, for any given time point for each of the genotypes/conditions, we stained for different patterning gene products on different days.
For each time point, wing imaginal discs from 10 larvae were dissected in cold phosphate buffered saline (PBS) and fixed for 30 min in 4% paraformaldehyde in PBS. Number of dissected discs varies from 5–16 depending on the treatment/genotype (Supplementary Tables S1 and S2). The tissue was washed in PBT (PBS +1% Triton X-100) at room temperature, blocked in PBT-NDS (2% Normal Donkey Serum in PBT) for 30 min and then incubated in a primary antibody solution (Supplementary Table S3) overnight at 4°C. After washing with PBT, tissue was incubated with fluorescently-conjugated secondary antibody overnight at 4°C. Tissue was rinsed with PBT and wing discs were mounted on a poly-L-lysine-coated coverslip using Fluoromount-G (SouthernBiotech). Samples were imaged using a Zeiss LSM 510 confocal microscope and images were processed using ImageJ.
The expression patterns of each of the gene products examined had previously been characterised in the literature: individual cells, patches of cells, or stripes [34], [43]–[45] (Supplementary Figure S1). To compose the staging scheme, we initially conducted a qualitative analysis of the patterns observed for each gene product at each time point (Figure 2 and Supplementary Figure S2) and described their progression. We then quantified these expression patterns in two ways (Supplementary Figs S2, S3). First, we divided the area of gene product expression by the total area of the disc, to generate a measure of pattern area. Second, we quantified the number of specific elements (cells, patches of cells or stripes) that each expression pattern exhibited. By both quantifying gene product expression and characterising the addition of new pattern elements through time, we were able to identify the gene products that varied the most during the third instar as well as those patterning elements that changed through a stepwise progression. We then used the change in patterns of these gene products to generate a staging scheme.
We used a Naïve Bayes Classifier (NBC) to test the power of our staging scheme to classify dissected discs from each time point into their correct stage. We first tabulated the observed gene-specific stages for all the patterning-gene products in the dissected discs from each time point. We then permuted the data from each time point 1000 times to simulate a population of 1000 discs with the range and frequency of gene-specific stages that was characteristic of wing discs from that time point. We then trained an NBC using our staging scheme and applied it to the permuted data set to determine what proportion of the 1000 simulated discs from each time point would be classified into the ‘correct’ stage. We repeated this analysis to assign stages to discs dissected from larvae reared under all experimental conditions.
All data analyses and statistics were conducted using R. The R scripts used to analyse the data, as well as the complete data, are available for download from Dryad (doi:10.5061/dryad.fq134).
|
10.1371/journal.pntd.0005010 | Validation of a Rapid Rabies Diagnostic Tool for Field Surveillance in Developing Countries | One root cause of the neglect of rabies is the lack of adequate diagnostic tests in the context of low income countries. A rapid, performance friendly and low cost method to detect rabies virus (RABV) in brain samples will contribute positively to surveillance and consequently to accurate data reporting, which is presently missing in the majority of rabies endemic countries.
We evaluated a rapid immunodiagnostic test (RIDT) in comparison with the standard fluorescent antibody test (FAT) and confirmed the detection of the viral RNA by real time reverse transcription polymerase chain reaction (RT-qPCR). Our analysis is a multicentre approach to validate the performance of the RIDT in both a field laboratory (N’Djamena, Chad) and an international reference laboratory (Institut Pasteur, Paris, France). In the field laboratory, 48 samples from dogs were tested and in the reference laboratory setting, a total of 73 samples was tested, representing a wide diversity of RABV in terms of animal species tested (13 different species), geographical origin of isolates with special emphasis on Africa, and different phylogenetic clades. Under reference laboratory conditions, specificity was 93.3% and sensitivity was 95.3% compared to the gold standard FAT test. Under field laboratory conditions, the RIDT yielded a higher reliability than the FAT test particularly on fresh and decomposed samples. Viral RNA was later extracted directly from the test filter paper and further used successfully for sequencing and genotyping.
The RIDT shows excellent performance qualities both in regard to user friendliness and reliability of the result. In addition, the test cassettes can be used as a vehicle to ship viral RNA to reference laboratories for further laboratory confirmation of the diagnosis and for epidemiological investigations using nucleotide sequencing. The potential for satisfactory use in remote locations is therefore very high to improve the global knowledge of rabies epidemiology. However, we suggest some changes to the protocol, as well as careful further validation, before promotion and wider use.
| The high fatality and burden of rabies stands in contrast to the very low performance of laboratory-based surveillance in resource-challenged countries. The absence of reliable human and animal rabies incidence data ultimately result in neglect of disease prevention and control and the perpetuation of RABV transmission despite the existence of powerful management tools. Rapid, easy to perform rabies diagnostic tests that do not require expensive equipment or special storage conditions, which can be reliably performed by trained ordinary veterinary professionals, are needed urgently for use in low income countries. Such novel methods will help to accurately assess the global rabies burden and are necessary to monitor rabies control and elimination. The present study evaluates the performance and reliability of a rapid, easy to use rabies diagnostic tool. Overall, the validated test was in high accordance with the standard reference method for the detection of RABV by immunofluorescence microscopy and showed even higher reliability when applied in resource poor laboratory conditions. The obtained results support the high potential for the use of this test in the field but suggest a change of the original technical protocol and a need for wider validation.
| Rabies is a viral zoonotic encephalomyelitis transmitted to humans after exposure to infected mammals, mainly dogs, through bites, scratches or licks on damaged skin or mucous membranes. This disease still continues to represent a public health concern worldwide, with an estimate of 60,000 human deaths per year, mainly in low income countries. Because of limited control measures in many countries and a lack of governmental concern, rabies remains a neglected tropical disease. This neglect is especially deplorable given the entirely preventable nature of the disease through vaccination of dogs and timely adherence to post-exposure prophylaxis (PEP) of exposed victims [1, 2]. In this way, human deaths due to this zoonotic disease could be reduced by over 95% [3, 4].
Lack of surveillance represents one major element of negligence, leading to missing data on disease incidence, imprecise estimates of the economic impact and a general underestimation of the true worldwide burden of rabies [4, 5, 6, 7]. This means that advocacy for rabies control cannot be supported with solid evidence and the necessity for action is not perceived at the decision maker level, for instance governmental authorities [8]. Poor surveillance is primarily a result of lack of political commitment and resource attribution for the control of rabies, and thus a cycle of neglect perpetuates. The vicious cycle is reinforced by the disregard of disease control in domestic dogs, which constitute negligible economic value and the fact that rabies affects largely marginalized communities with difficult access to healthcare. Deficiencies in basic healthcare do not only contribute to hinder access to PEP but also lead to the misdiagnosis of rabies in the face of other causes of encephalitis, such as cerebral malaria, as reported in Malawi [9]. Surveillance is fundamental to accurate burden of disease measures, for advocacy of disease control and also a prerequisite for disease elimination [10, 11]. Currently, rabies is believed to be underreported at the extent of 1:60 in humans and this rate could even be much higher for animal rabies incidence [6].
One possible point of leverage to break the cycle of underreporting and neglect is the reinforcement and simplification of diagnostic capacities and tools. Infrastructures required for the current standard diagnostic tests are expensive, and their methodologies and interpretation need thoroughly experienced personnel. Antigen detection of RABV using the direct fluorescent antibody test (FAT) is the World Health Organization (WHO) and World Organisation for Animal Health (OIE) reference test [6, 12, 13] and is routinely performed in many developed countries. However, it is difficult to establish in developing countries because fluorescence microscopes are expensive and the required maintenance is demanding. Also, the immunofluorescence conjugate necessary for the test is costly and has to be transported and stored refrigerated. Finally, accurate reading of the test needs stringent quality control of the test performance and very experienced personnel. Similar constraints are encountered with the direct rapid immunohistochemical test (DRIT) [14, 15]. Although the DRIT can be read using a light microscope, the test methodology requires a meticulous protocol which currently lacks commercialized biotinylated anti-rabies antibodies and has to be carried out by trained personnel. For rabies diagnosis however, a simpler field test is desirable for various reasons. In the current situation the benefits of rabies diagnosis are not well perceived by the public and rabies suspicious animals are often killed immediately and rapidly disposed [16]. Also, a short time lag between suspicion and confirmation of a rabies case is important for early adherence to PEP or the cost savings in case of a negative diagnostic result. Finally, transport of samples over long distances in climatically warm settings increases the risk of poor sample quality, which adversely affects FAT test results [17]. Proximity to the public through decentralized laboratory facilities is therefore vital for good sample quality, as well as rapid detection and response.
A rapid immunodiagnostic test (RIDT) based on the lateral flow principle was first described and evaluated in 2007 on a limited panel of RABV samples [18]. The same study reported on the detection limit and potential risk of cross reactivity. Further laboratory evaluation was conducted more recently on the use of this RIDT for the detection of RABV circulating in Europe, and extended to the detection of other species of lyssaviruses [19, 20]. Both studies showed positive results regarding sensitivity and specificity of such tests compared to the FAT. To date, only two studies have been conducted under field conditions, both suggesting positive results for the use of RIDT [21, 22].
In our study, we evaluated the practicability and the performance of this RDIT identified as Anigen Rapid Rabies Test (Anigen test) (Bionote Inc.) in different settings: under field conditions with its application to the surveillance of rabid animals in N’Djamena, Chad and in laboratory settings with a panel of selected RABV isolates. Lastly, we evaluated this tool for a novel application in rabies surveillance, with its use as a vehicle for viral RNA storage and conservation, and demonstrated that recovery and detection of RNA present on the strip of positive samples was possible. The Anigen test appears as a promising tool for the post-mortem diagnosis of animal rabies, and the molecular detection and genotyping of positive test strips.
During June 2012, the RIDT Anigen test, a chromatographic immunoassay-based on lateral flow technology manufactured by BioNote, Inc (Gyeongi-do, Republic of Korea) [23], was added into the routine diagnostic procedure of the rabies laboratory of the Institut de Recherche en Elevage pour le Développement (IRED) in N’Djamena, Chad. It was utilized in parallel with the FAT test, which had been used since 2001. Rabies-suspect animals were presented to the IRED by their owners or by the bite victim. No active surveillance was initiated throughout the study. However, awareness was intensified prior to the study period, during May 2012, by a poster campaign sensitizing the public in N’Djamena to seek medical treatment after a dog bite and to send the biting animal to the IRED in case of rabies suspicion.
For the validation of the Anigen test in the field, diagnostic results from June 2012 to February 2015 were included. Only samples originating from dogs were considered for inclusion according to the manufacturer’s recommendation [23]. During the 33 months of the study period, a total of 49 rabies -suspect dog heads were submitted to IRED for diagnostic testing. The origin of the samples is detailed in Table 1. Most were in fresh condition on arrival and upon testing (85%, n = 42). Five of the samples were decomposed, while in one case, the sample quality was not noted. Only one sample was so decomposed that it was impossible to analyse and was excluded from the study. The final sample size of field isolates at IRED was 48 (Table 1).
The Anigen test was further validated at the National Reference Centre (NRC-R) and WHO Collaborating Center for Rabies at the Institut Pasteur in Paris, France, on 73 samples selected from the collections housed in both of these centers, from 12 different species originating from various countries and belonging to different phylogenetic clades (S1 Table). All these samples were previously analysed by FAT. Thirty of them were negative and the remaining 43 were positive. The positive samples represented a large diversity of RABV. All these 73 samples were stored at -80°C for archive before analysis.
In addition, the limit of detection of the RIDT was evaluated at NRC-R using a panel of 8 different isolates of RABV adapted and amplified on baby hamster kidney cells (BSR cells). Viral suspensions were titrated on the same cells using 5-fold serial dilutions in cell culture medium and expressed as fluorescent focus units per mL (FFU/mL). For the RIDT evaluation, titrated RABV suspensions were first tested at several concentrations using the buffer available from the RIDT kit as a diluent.
The FAT, the gold standard technique for post-mortem diagnosis of rabies [12] was performed at the NRC-R under quality assurance (accreditation ISO/IEC 17025), as previously described [13]. In the rabies laboratory of N’Djamena, the FAT was performed with some deviations regarding the standard procedure: lack of positive and negative control samples inclusion, absence of routine quality assessments, and storage of the immunofluorescent conjugate past the expiration date. In this setting, two microscopic slides were prepared, with two brain impressions per slide. If no viral characteristic fluorescent inclusions were observed on all four impressions, the sample was considered negative. Doubtful results were declared positive due to the potential fatal consequences of a false negative result for the bite victims. However, due to some deviations regarding the standard procedure, it was not possible to consider FAT performed ad IRED as the gold standard for the specificity and sensitivity analysis.
The Anigen test is a simple and rapid diagnostic tool, presenting as an all-in-one included kit. Once the brain is extracted, it is used without additional material and equipment except for one dilution step requiring an additional vial of phosphate-buffered saline (PBS) prepared according to the manufacturer’s recommendation. However, for our study, we omitted the first dilution step, only using the vial with buffer provided by the kit to simplify the test procedure in view of future application under realistic field conditions. The same procedure was used for the Anigen test at NRC-R and at IRED. If it was possible to anatomically identify the regions of the brain in a sample, the test was performed with a small section of the brainstem (approximately 0.1 g), otherwise the same amount of material was taken from different parts of decomposed brain samples. The brain sample was mixed directly in the tube containing the buffer with the swab, all included in the kit, for about one minute, until most of the brain material was well dissolved and then put on to the test plate using the transfer pipette provided in the kit. Four drops were deposited on the strip (corresponding to nearly 100 μL). The test could be interpreted when the coloured liquid reached the top of the test and the purple indicator colour had vanished from the filter paper background. As described by the manufacturer, a positive test result was indicated by two purple lines, one in the test zone and the other in the control zone. If a line only appeared in the control zone, the test was considered negative. In cases where only the test line was coloured rather than the control band, the test was declared invalid and was performed once again. The test took approximately 5 to 10 min after deposit of sample and the interpretation was not be performed after 10 min, according to manufacturer recommendations. Following these recommendations, the test was suitable for dog, raccoon dog and cattle samples (animals which were used originally in the validation of the method [18], and should be tested immediately after collection. In this study, two different batches were used for the validation of the RIDT assay, with batches n°1801076 and n°1801111 for the field and the laboratory validation, respectively.
Brain impressions were performed directly on FTA Whatman cards, a support dedicated to the storage and preservation of RNA [24]. Prior to use, the cards were stored at room temperature in a sealed plastic bag in a dry and clean area. The samples were prepared by diluting a small section (approximately 0.1 g) of the brain in 1 ml PBS (10%). After thorough mixing, the brain homogenate was loaded onto the card with a pipette until the sample indicator circle on the filter was covered. The cards were then dried 24 hours at room temperature before being put separately in transparent plastic bags for transportation.
To prepare the samples on FTA Whatman filter paper, 1 cm2 was cut-off from the area containing the brain impression and incubated during 1 hour in Tri-Reagent LS (Molecular Research Center, Cincinnati, Ohio, USA) or overnight in cell culture medium (DMEM) (Life Technologies, Saint Aubin, France), then placed in Tri-Reagent. To obtain viral RNA directly from the Anigen test strip, the cassettes were opened, the filter paper was removed and the area where the sample was deposited was collected and placed into 1 mL of Tri-Reagent LS. For both FTA and Anigen test supports, total RNA extraction was performed as previously described, following manufacturer recommendations [25].
Viral RNA detection was performed using a one-step dual combine pan-lyssavirus RT-qPCR assay recently described [26], targeting a conserved region among the polymerase. Briefly, this assay includes a pan-RABV RT-qPCR probe-based technique, able to detect all representatives of the broad genetic diversity of RABV, using two degenerated TaqMan probes. In parallel, a SYBR Green RT-qPCR assay is able to detect all the other lyssaviruses tested, in addition to RABV isolates. Both of these assays, which were optimized to a final reaction volume of 20 μL, were performed using 5 μL of RNA template (previously diluted 1:10 in nuclease-free water). For each assay, appropriate controls were used. Details of the combined pan-lyssavirus RT-qPCR assay are in S2 Table and in reference [26].
A selected panel of RNA extracts from Anigen test, which were found positive with the dual combine RT-qPCR assay, were evaluated with RT-PCR to generate amplicons suitable for genotyping by sequencing (at least 500 nt in length). Briefly, a volume of 6 μl of total RNA extraction was used for reverse transcription as previously described [25]. RNA was incubated at 65°C for 10 min with 2 μL of pd(N)6 random primers (200 μg/mL; Roche Diagnostics) and 2 μL of sterilized distilled water and then were stored on ice. Each tube was incubated with 200 U of Superscript II RT (Invitrogen), 80 U of RNasin (Promega), and 10 nmol of each nucleotide triphosphate (Eurobio), in a final volume of 30 μL for 90 min at 42°C, for reverse transcription. Two microliters of complementary DNA (cDNA) were then amplified by PCR targeting the nucleoprotein gene of RABV, as described in [27].
The RIDT assay was evaluated by the NRC-R in an inter-laboratory trial organized during 2015 by the European Union reference laboratory for rabies, which is located in Nancy, France [28]. The FAT technique was also evaluated in parallel in this trial. The test panel consisted of nine anonymous samples of freeze-dried homogenized brains, either uninfected or infected with various lyssavirus species. Details of this trial have been provided elsewhere [28].
Results obtained with FAT and Anigen techniques were compared using the McNemar and Kappa statistic tests in Stata, and were analyzed to determine the intrinsic parameters of the RIDT assay. However and conversely to the FAT technique done at the NRC-R, the immunofluorescence assay performed at the IRED could not be considered as the reference technique due to several deviations compared to the standard procedure. In case of discrepancy between RIDT and FAT, samples were tested for RNA detection with RT-qPCR assay performed on FTA Whatman cards impregnated with the brain of the corresponding sample. For the determination of the sensitivity and the specificity of RIDT, true positivity and true negativity was defined according to the result that was shared by at least two tests among FAT, RIDT and viral RNA detection.
For the majority of the total sample size (n = 121) tested at IRED and at NRC-R, the RIDT was successfully performed, with the presence of a line clearly visible in the control zone after 5 to 15 min of migration once the sample was deposited (Fig 1A and 1B). For only a few samples (n<10), the test was repeated, due to abnormal or incomplete migration (absence of the line in the control zone). When they scored positive with RIDT, most samples exhibited a line with strong intensity in the test zone (Fig 1B). In a few cases the test bands showed even higher intensity than the control band. Also, for some samples tested at NRC-R, the line in the test area was only faintly visible, despite a strong intensity of the line in the control zone (Fig 1A).
A total of eight titrated suspensions from different RABV adapted to cell culture was selected to determine the limit of detection of the RIDT (Table 2). A volume of 100 μL of each of them, diluted or not, were tested. The lowest number of fluorescent focus-forming units (FFU) detected with this assay was 105 FFU, and was obtained for RABV 9704ARG and 04030PHI. Isolates 9147FRA and 9508CZK exhibited a positive signal with 106 FFU. Lastly, no positive signal was obtained with the initial viral suspension for virus 8743THA and 9001FRA, indicating that the limit of detection was > 8.1 x 106 FFU and > 2.4 x 105 FFU, respectively.
Seventy-three samples from NRC-R, including forty-three positive samples representing a large diversity in term of host species, geographical location and genetic diversity (S1 and S3 Tables) were tested. Compared to the gold standard FAT, the RIDT demonstrated an accordance of 95%. The specificity was 93.3% with only two false positive results among the 30 FTA-negative specimens, noticed for samples 150057 and 150125 which were originated from a dog and a cat, respectively (Table 3, S1 and S3 Tables). The sensitivity of the RIDT was 95.3%, with only two false negative results observed for isolates 9217ALL and 9312MAU, a red fox from Germany and a dog from Mauritania, respectively (Table 3, S1 and S3 Tables).
Among the 48 samples included for evaluation at IRED only 3 were not concordant between FAT and RIDT, yielding an accordance of 94% (Table 3). Two of the discordant samples (samples 343 and 389) were decomposed and the quality remained unknown for the last one (sample 362) (Table 1). The FAT was impossible to perform on one (sample 389) of the 3 and for the two others (samples 343 and 362), the result was positive (S4 Table). For all these 3 specimens, RIDT tested negative (S4 Table). In these three cases, where RIDT and FAT did not yield the same result, viral detection performed by RT-qPCR on the FTA Whatman card could not detect viral RNA after multiple attempts confirming the negative RIDT result (S4 Table). For sensitivity and specificity of RIDT and FAT under field conditions, true positivity and true negativity was defined according to the result that was shared by at least two tests among FAT, RIDT and viral RNA detection on the FTA Whatman card. The RIDT showed a higher specificity (100%) than the FAT (78.5%) at IRED. Accordance with the overall true test results of the 48 samples from IRED was 100% for RIDT and 94% for FAT.
The McNemar test showed no significant difference between the FAT and RIDT (Exact McNemar significance probability = 0.5) on samples test at IRED and the Kappa value was 0.86, indicating excellent agreement between the tests. The exact McNemar significance probability for the comparison of FAT and RIDT performed at NRC-R was 1 and the Kappa value was 0.89. Overall, of the 121 samples analysed at NRC-R and IRED, the McNemar significance probability was found to be 0.45 and the Kappa value was 0.87.
A total of 51 samples were tested at NRC-R for viral RNA detection using RT-qPCR on the Anigen test strip, which were previously found positive for the post-mortem diagnosis of rabies (Table 4, S1 and S4 Tables). The FAT scored also positive for all of these specimens. Among them, 32 originating from IRED were used during the field evaluation of the RIDT, whereas 19 were obtained in NRC-R during the laboratory evaluation. Positive detection was obtained for 26 (81.2%), 18 (94.7%) and 44 (86.3%) samples from IRED, NRC-R and the two combined, respectively (Table 4). In parallel, detection of viral RNA was also performed at NRC-R on FTA Whatman cards for 31 samples, which were positive after analysis with both FAT and RIDT at IRED (S4 Table). In this case, a perfect concordance (100%) was noticed. In addition, viral RNA from 2 and 6 samples found previously negative with RIDT were not detected after RT-qPCR analysis performed on the Anigen test strips and FTA Whatman cards, respectively (Table 4). When compared to the FTA Whatman card, RT-qPCR performed on the Anigen test strip exhibited a sensitivity of 80.6% (Table 5).
A limited panel of viral RNA samples extracted from Anigen test strips, which were previously confirmed positive by RT-qPCR, were secondarily tested for genotyping. A total of 14 samples (4 originating from IRED and 10 from NRC-R) were analyzed, among them 13 (93%) provided PCR amplicons (at least 500 nucleotides in length) targeting regions of the nucleoprotein gene commonly used for genotyping after sequencing (S3 and S4 Tables). The only sample found negative (isolate 9702IND) was weakly positive after FAT and RIDT tests, which could then probably explain the absence of PCR amplification.
Finally, we evaluated the RIDT Anigen test in an inter-laboratory trial, in parallel of the FAT on nine anonymous samples. The results obtained were concordant with those expected (Table 6) [28]. In particular, we were able to detect three different RABV isolates (strains CVS27 14–10, GS7 and DR627) constituting the panel, as well as 4 other lyssavirus species, including Duvenhage virus (DUVV), European bat lyssavirus 1 (EBLV-1) and 2 (EBLV-2), and Bokeloh bat lyssavirus (BBLV).
The aim of our study was to evaluate both in the laboratory and under field conditions the RIDT Anigen test in comparison with the FAT, to investigate the intrinsic parameters of this rapid technique, as well as the relevance of its application in sub-Saharan Africa, a region with the highest estimated per capita death rate due to rabies [4] but with poor data reporting situation [29].
We first investigated the limit of detection of this RIDT using serial dilutions of different titrated suspensions of RABV. This value varied among the isolates tested but remaining relatively high, ranked from 105 to 107 FFU.
We then evaluated the use of RIDT for post-mortem diagnosis of animal rabies in the laboratory and under field conditions, and compared it to the FAT assay. Our results demonstrate that the lateral flow test performs similar to the FAT. The accordance between RIDT and FAT was high under both conditions (≥94%), with a specificity of the RIDT from 93.3% to 100%. The sensitivity of this technique was also high in laboratory settings, with 95.3%, and approached 100% under field conditions. Our results indicate the high potential of this test for use in the resource challenged African context. Importantly, we show that the intrinsic performance of RIDT under limited laboratory conditions could be higher compared to the FAT test. Conversely to RIDT, several factors could affect negatively FAT results, including storage and quality of the fluorescent conjugate, maintenance of the fluorescence microscope and experience of the reader. However, given the limited sample size, the explanatory power is not overly strong and further evaluation is highly encouraged.
Lastly, we tested the RIDT Anigen test in parallel of the FAT technique in an international inter-laboratory trial organized by the European Union reference laboratory for rabies, Nancy, France and all results were found concordant.
As underlined by the results obtained under field conditions, the advantages of the immunochromatographic test method are manifold. Samples can be analysed one by one proportionate to the diagnostic demand. This is also true for FAT, however, the conjugate used for the FAT can only be stored for a limited time to ensure the quality of the test. Similarly, storage for the reagents is a cause for quality concern for the DRIT. For both DRIT and FAT, negative and positive controls have to be included in the test procedure for standardization, which is not needed for the RIDT [19]. Storage of the RIDT can be done at room temperature and does not require refrigeration, as is necessary for the conjugates used for the DRIT and FAT test. The tests used at IRED were stored at 20°C in an air conditioned room. There are no data on the reliability of the tests stored at temperatures at above 30°C, as would be encountered in many tropical countries.
The FAT depends heavily on the quality of the immunofluorescent conjugate, the maintenance of the fluorescent microscope and also on an experienced technician reading the microscope slides. In contrast, RIDT is a very easy-to-use kit, which does not require a high level of expertise. This technique is simple to perform and to interpret.
For the dog samples tested in Chad, the test had a specificity and sensitivity of 100%. However, we showed the utility of the Anigen test for many different wild and domestic mammalian species. Our results confirm data obtained from previous studies, and suggest that the spectrum of species which could be tested with this RIDT is larger than recommended by the manufacturer (e.g., dogs, cattle and raccoon dogs) [19, 20, 21].
In our study, we mainly focused on RABV species. In particular, we were able to detect RABV isolates belonging to all of the major phylogenetic clades defined previously [30], with the exception of the Africa 3 clade, which was not been tested in our panel. However, positive detection of isolates belonging to this clade has been already demonstrated [20]. In addition, we were also able to detect 4 different other lyssavirus species, in addition to RABV, when evaluated this technique in an inter-laboratory trial. These results are concordant with those obtained in two other previous studies using different lyssavirus species including African and bat-related lyssaviruses (including Lagos bat virus, Mokola virus and Australian bat lyssavirus), demonstrating that it can also be applicable for the detection of non-RABV lyssaviruses [19, 20, 31].
Several RIDT kits for RABV detection were evaluated using brain samples [18, 20, 21, 32, 33]. A very recent study evaluated 6 commercially available RIDT kits in parallel [34]. Sensitivity of the Anigen test in that study was observed to be unsatisfactory [34].
We modified the test procedure by omitting a dilution step and placing the brain sample directly into the buffer vial provided by the test. The advantage of this approach is that the test can be used with no additional material other than that provided in the kit. This change in methodology might explain the better sensitivity of the test and the higher intensity of the test band compared to other studies [19, 22, 35], because the RABV antigen level is higher without a second dilution. Sharma et al. [35] found that the intensity of the test band decreases with dilution.
Autolysis of samples is less a concern for sensitivity of the RIDT compared to the FAT and PCR [17, 19, 36], which is illustrated by our results. Sample storage in glycerol was suggested to interfere with the optimal test performance, affecting the intensity of the test line [19].
The successful detection of RABV RNA from the Anigen test strip in over 86.3% of samples tested is a promising result and highlights the potential use of the kit as a vehicle for sample submission for further confirmatory diagnostic or genotyping analysis. This potential has also been reported by others [34]. However, the sensitivity was lower when compared to the use of the FTA Whatman card, a dedicated support for storage and preservation of nucleic acids.
The price of less than 10 euros is less expensive compared to the cost of performing FAT [19, 22], but still poses an affordability problem in developing countries.
Further validation has to be conducted with RIDT, especially if the results of this test will guide decision making for PEP. We demonstrated that the sensitivity of RIDT, even high, was not complete compared to FAT. To avoid getting false negative results with this technique, we suggest to confirm all negative results using WHO and OIE reference techniques, such as FAT, before excluding RABV infection in diagnostic samples.
An efficient diagnosis method is just part of the entire process of surveillance and control needed to eliminate rabies, as comparable to translation of efficiency of a vaccine, to the ultimate immunity of the target population [37]. Therefore, all components of the surveillance system, in which the test would be promoted and used, have to be strengthened in parallel [38].
Specificity and sensitivity of the evaluated Anigen test are only slightly reduced compared to the known reference tests for rabies virus detection in brain samples. The results are promising for field use, where the test could help to establish rapid preliminary diagnostic results, which would be further confirmed using WHO and OIE recommended tests at central laboratories. However, we suggest important changes to the test protocol: skip the dilution step of brain biopsy in PBS and perform the brain homogenate with the swab directly into the specimen tube containing 1 ml of assay diluent, both provided in the kit. We also recommend to provide a more precise sketch depicting the brain sampling method. Rapid rabies tests cannot substitute for the current reference tests, but are crucial for the success of rabies surveillance systems in developing countries. Further, we demonstrated here that the test cassettes can be used as a vehicle to ship viral RNA to reference laboratories for further laboratory confirmation of the diagnosis and for epidemiological investigations.
|
10.1371/journal.pntd.0003301 | Urbanization Increases Aedes albopictus Larval Habitats and Accelerates Mosquito Development and Survivorship | Aedes albopictus is a very invasive and aggressive insect vector that causes outbreaks of dengue fever, chikungunya disease, and yellow fever in many countries. Vector ecology and disease epidemiology are strongly affected by environmental changes. Urbanization is a worldwide trend and is one of the most ecologically modifying phenomena. The purpose of this study is to determine how environmental changes due to urbanization affect the ecology of Aedes albopictus.
Aquatic habitats and Aedes albopictus larval population surveys were conducted from May to November 2013 in three areas representing rural, suburban, and urban settings in Guangzhou, China. Ae. albopictus adults were collected monthly using BG-Sentinel traps. Ae. albopictus larva and adult life-table experiments were conducted with 20 replicates in each of the three study areas.
The urban area had the highest and the rural area had the lowest number of aquatic habitats that tested positive for Ae. albopictus larvae. Densities in the larval stages varied among the areas, but the urban area had almost two-fold higher densities in pupae and three-fold higher in adult populations compared with the suburban and rural areas. Larvae developed faster and the adult emergence rate was higher in the urban area than in suburban and rural areas. The survival time of adult mosquitoes was also longer in the urban area than it was in suburban and rural areas. Study regions, surface area, water depth, water clearance, surface type, and canopy coverage were important factors associated with the presence of Ae. albopictus larvae.
Urbanization substantially increased the density, larval development rate, and adult survival time of Ae. albopictus, which in turn potentially increased the vector capacity, and therefore, disease transmissibility. Mosquito ecology and its correlation with dengue virus transmission should be compared in different environmental settings.
| Aedes albopictus has expanded its ecological habitat range throughout the world. Although Ae. albopictus was previously considered a rural vector, this species has adapted well to suburban and urban environments, and it has become the most important and sometimes the sole vector of dengue virus transmission in urban areas. Dengue is a vector-borne disease that has become a severe global public health problem during the last decade. We explored the effect of ecology in different ecological settings (urban, suburban, and rural) on Ae. albopictus larval habitat and mosquito development in Guangzhou, where recently dengue has caused serious public health concerns. The environmental changes caused by urbanization had a significant impact on the ecology of Ae. albopictus. Compared with rural and suburban areas, urban areas had more Ae. albopictus larval habitats, shorter larval development time, higher adult emergence rate, and longer lifespan. These results imply that urbanization significantly increases the potential for dengue outbreaks. Because urbanization is a global trend resulting from economic development, the elucidation of Ae. albopictus adaptation to different environments in China also reveals the potential for this important vector to colonize other parts of the world.
| Aedes albopictus (Skuse) (Diptera: Culicidae), the Asian tiger mosquito, is an aggressive, strongly anthropophilic, exophagic, and exophilic mosquito. As an important vector of dengue fever, chikungunya disease, and yellow fever, Ae. albopictus has emerged as a global public health threat [1]–[3]. Ae. albopictus is indigenous to both tropical and temperate regions of Southeast Asia and islands of the western Pacific and Indian Oceans, but it has recently expanded its range to every continent except Antarctica [4], [5]. Unlike wetland mosquito species that oviposit and develop in habitats that are large, predictable, and easy to identify, Ae. albopictus is difficult to locate and control because this species utilizes small, different types of habitats including small containers and spare tires [6]–[8].
Ae. albopictus originated at the edges of forests and bred in natural habitats (e.g., tree holes, bamboo stumps, and bromeliads) and was previously considered a rural vector [9]. However, this species has adapted well to urban environments with larvae now breeding in artificial containers (e.g., tires, cemetery urns, and water storage containers) and has become the most important and sometimes sole vector in urban areas [8], [10], [11]. Ae. albopictus is found almost everywhere, especially in urban areas in southern and southwestern China [12]–[15]. The frequent outbreaks of dengue fever in the cities in southern (mainly Guangdong province) and southeastern coastal (mainly Fujian and Zhejiang provinces) China in the past few decades have caused serious public health concerns [16]–[18]. Although Aedes albopictus is described as a minor vector of dengue and possibly chikungunya in the world, it is emerging as a major dengue vector in China and was responsible for most outbreaks of dengue in China [19], [20] and chikungunya in 2010 in Guangdong, China [21].
Similar to other mosquito vectors, Ae. albopictus needs aquatic habitats to breed and develop, and therefore, it is sensitive to environmental changes [8], [22], [23]. Destruction of breeding habitats is an important strategy to reduce the Aedes mosquito population; eliminating suitable breeding habitats reduces larval development and thus the adult mosquito population. Equally importantly, environmental changes, such as changes in temperature, affect habitat productivity, larval and adult development times, and survival, which in turn directly and indirectly affect disease transmissibility [24]–[32].
Urbanization refers to the increasing population of urban areas. Urbanization predominantly results in the physical growth of urban areas, leading to environmental changes. Urbanization is a global trend that results from economic development. Asian countries including China and India, countries in Southeast Asia, and African countries such as Nigeria are the fastest growing areas in the world, and the unprecedented movement of people into these areas is predicted to intensify in the future [33]. Many problems have emerged as a result of urbanization, including environmental pollution, crowding, and the destruction of natural ecology. The socioeconomic effects of urbanization have been extensively studied by socio-ecologists [34]–[36]; however, the ecological effects and their impact on vector biology and vector-borne infectious disease transmission remain unclear. Most dengue fever outbreaks occur in the urban areas of China, and these outbreaks have become more frequent over the past decade [18], [20], [37]. There is an accelerating trend of urbanization in China; will this process of urbanization accelerate dengue fever outbreaks? Changes in environmental conditions as a result of urbanization may directly and/or indirectly affect the ecology of mosquitoes, e.g., larval habitat availability and suitability, development, and survivorship. Because Ae. albopictus has invaded Europe (e.g., Italy and France) and the Americas (e.g., USA), which increases the global vulnerability to dengue fever outbreak, therefore, it is crucial to evaluate its adaptations to urban environments.
We hypothesized that urbanization increases Ae. albopictus larval habitats and survivorship and accelerates the development of larvae and adults. This study explored the ecology of Ae. albopictus in different settings (urban, suburban, and rural) in the Great Guangzhou area, China. Field surveys of larval habitat availability, larval development and adult mosquito life-table experiments were conducted in semi-natural conditions to test the hypothesis.
The field surveys of larval habitat availability and semi-natural condition larval development and adult mosquito life-table experiments were carried out in Guangzhou, the capital city of Guangdong province, China. Guangzhou is the largest city in southern China, and it is located in the Pearl River Delta, where numerous cities form a Canton-Macao-Hong Kong economic development zone. The annual average temperature in Guangzhou is 21.6°C, and its annual rainfall is approximately 1,980 mm. This climate is ideal for the development and reproduction of Ae. albopictus. The city has experienced rapid expansion during the recent regional economic development. Several major dengue fever outbreaks have occurred in this area since 1980, and Ae. albopictus is the sole dengue vector [13], [38], [39]. Therefore, Great Guangzhou is an ideal place to study the impacts of urbanization on Ae. albopictus.
The study was conducted in three areas that represented urban, suburban, and rural settings in Guangzhou (Figure 1 and Figure S1). Each study area was approximately 1.8 km2. The distance between each area was approximately 24 km. Tonghe (113°19′E, 23°11′N, 31 m above sea level (a.s.l.)) is an urban area with a population density of >3,000 people/km2. The land use types are primarily residential and commercial buildings and public services such as schools and hospitals, filled with trees and grasses. Liangtian (113°23′E, 23°21′N, 25 m a.s.l.) is a suburban area with a population density of approximately 1,000 people/km2, and land use includes a mixture of residential, manufacturing, and farmland. Dengcun (113°33′E, 23°30′N, 42 m a.s.l.) is a rural area and has a population density of <100 people/km2, where land use is primarily agricultural (rice and vegetable planting) and forest.
Aquatic habitat surveys were conducted in the three areas from May to November 2013. We surveyed all aquatic containers in the study areas monthly with three teams of four trained personnel per team. All properties within the site (i.e., residential, abandoned, commercial, and public services) and alleyways were surveyed, except for parcels whose owners refused access or places that were inaccessible due to physical barriers (e.g., fallen structures). We provided a detailed explanation of the purpose of the study to the residents, and after obtaining consent, we inspected the indoor, outdoor, and surrounding areas for aquatic habitats. The location and the physical characteristics of the habitats were recorded. Their chemical and biological characteristics were sampled, and mosquito larval availability and counts were recorded.
The geographic location of each habitat was located using a hand-held GPS unit (GARMIN Corporation, Taibei, Taiwan). For each habitat identified, the depth and surface area of the water was measured; for small containers (<0.25 L), water was emptied into a separate container to measure the actual volume. Habitats were subjectively characterized by coverage of canopy (direct sunlight, full shade that can be exposed to sunshine, full shade that cannot be exposed to sunshine), habitat type, substrate type (soil, sand, leaf, moss, no substrate), and turbidity (clear = colorless, tinted = in between, polluted = opaque and odoriferous).
Immature Ae. albopictus samples were classified as young larvae (1–2 instar), old larvae (3–4 instar), or pupae. Immature mosquito abundance was determined using the standard 350 ml dippers. Once dipped, larvae and pupae were collected using a pipette, and individual numbers were counted. Samples were transported to the laboratory, where they were reared until emergence for species identification. All mosquitoes that emerged were pooled by site (Urban, Suburban, and Rural) and species.
The BG-Sentinel trap with lure (bought from Solbrite Resources Pte Ltd, Singapore; produced by BioGents, Regensburg, Germany) was used for adult surveillance in this study because it is a very efficient tool for capturing adult Ae. albopictus [40]–[42]. During the surveys, 12 BG-Sentinel traps were placed in each study area. In each study site, we chose three typical environmental settings for the traps: in the urban area, a residential area, public park, and commercial district; in the suburban area, a residential area, factory, and garden; and in the rural area, a residential area, farmland, and forest. The distance between two traps was at least 50 meters. Traps were placed in the same location for three consecutive days during the first week of each month; they were shifted to different locations for another three days during the third week of each month. The adult population was monitored continuously from July to November 2013.
Trapped mosquitoes were collected every 24 hours, transported to the laboratory, and frozen for species identification. Frozen mosquitoes were placed on a piece of white filter paper in a petri dish on a chill table, and the species was identified morphologically using taxonomic keys [43]. Blood-fed females were identified visually by their dilated red abdomens, and they were stored at −80°C for further analysis.
Mosquitoes used for adult life-table experiments were all F0 individuals who originated from different habitats in the study areas. Newly emerged (<24-h-old) adults were transferred to a 30×30×30 cm microcosm covered with nylon netting; 20 females and males each were placed in each microcosm. We used the thumbtack to fix the twine on the ceiling, and cages were lifted 1.0 m above the ground. The twine was smeared with grease to prevent the reach of ants and other insects which may cause interference with the experiment. Cotton wool soaked in 10% sucrose solution was supplied to the mosquitoes daily. Dead mosquitoes were recorded and removed from the cage daily. There was no other mosquito coils or spray near the cages during the experimental period. The experiments were performed in July–August 2013 and repeated in October–November 2013. There were 15 replicates in each site during each season.
Air temperature, humidity, and water temperature were measured using the HOBO data loggers. Data were offloaded using a Hobo Shuttle Data Transporter (Shuttle, Onset Computer Corporation, Bourne, MA) and then downloaded to the computer using BoxCar Pro 4.0 software (Onset Computer Corporation). The monthly rainfall amount in each area was obtained from local meteorological stations (Figure S2).
The daily average, minimum, and maximum temperatures and relative humidity were calculated from the hourly records. ANOVA post hoc Tukey's honestly significant difference (HSD) tests were used to determine the statistical significance of differences in mean temperature and relative humidity in different areas for each season. Water temperature in larval habitats was analyzed in the same manner. Mosquito larval density was standardized as the number of larvae per liter of water. Differences in immature mosquito density among different areas were tested using the Tukey's HSD test after logarithmic transformation of larval densities. Differences in adult mosquito density among different areas were tested using one-way analysis of variance (ANOVA) with repeated measures after square root transformation of raw data. Survival rates of Ae. albopictus larvae were calculated as the proportion of first-instar larvae that survived to emergence of adult. Mean larval development time was defined as the average duration from first-instar larvae to emergence of adult, and was computed separately for each sex. Kaplan-Meier survival analysis was used to determine the effect of different environmental conditions on adult mosquito daily survivorship. Stepwise logistic regression was used to identify the factors significantly influencing the occurrence of immature Ae. albopictus in aquatic habitats. We used the χ2-test to determine the significance of differences in stage-specific survival rates of immature mosquito from different places, and Tukey's HSD tests of ANOVA post hoc were used to determine the statistical significance of differences in stage-specific development times. Statistical analysis was performed using JMP statistical software (JMP 9.0, SAS Institute Inc., USA).
All entomological surveys and collections conducted on private lands or in private residential areas were done with the owners'/residents' permission, consent and presence. These studies did not involve endangered or protected species.
During our survey period, we found 2639, 2523, and 1760 aquatic habitats in urban, suburban, and rural areas, respectively. χ2-test indicated that habitat Ae. albopictus positive rate varied significantly (P<0.0001) among the three areas, urban area had the highest positive rate (44.0%), then suburban area (37.7%), and rural area had the lowest rate (31.5%). A wide variety of container types were present in the three study sites (Table 1). The most abundant container types in urban areas were plastic buckets (412), and the least abundant were tarps (6). Flower pots, disposable food tins and gutters were also abundant in urban areas (Table 1). The most abundant container types in suburban areas were disposable food tins (665), and the least abundant were pools (2). Abandoned tires, plastic buckets, and clay pottery were also abundant in the suburban area (Table 1). The most abundant container types in rural areas were clay pots (445) and plastic buckets (324); disposable food tins (276) were also found frequently in this area. Overall, the variety of containers and habitat types was less abundant in rural areas (Table 1).
Immature Ae. albopictus were most often found in abandoned tires (positive rate 67.3%) and flower pots (65.1%) in urban areas (Table 1, Figure S3). In suburban areas, Ae. albopictus larvae were common in abandoned tires (54.2%) and clay pots (53.4%) (Table 1). Whereas in rural areas, Ae. albopictus larvae were frequently found in plastic buckets (29.3%) and plastic basins (27.7%) (Table 1).
The number of Ae. albopictus-positive habitats varied over time and between different study areas (Figure 2). The urban area had the highest aquatic habitat positive rate in every month except October. Over the seven-month survey period, the aquatic habitat positive rate in urban areas (monthly-mean ± SD 43.8±4.4%) was significantly higher than in rural areas (28.4±7.6%) (Tukey's HSD test, P<0.05) but not significantly different from suburban areas (36.9±7.3%).
Densities of immature mosquitoes also varied significantly among study areas (Table 2). The urban area had significantly higher 1–2 instar larvae density than that in the suburban and rural areas, but the difference in 1–2 instar larvae density between suburban and rural areas was statistically insignificant (Tukey HSD test, Table 2). The density of 3–4 instar larvae in urban and suburban areas was significantly higher than that in rural areas, but the difference in 3–4 instar larvae density between urban and suburban areas was statistically insignificant (Tukey HSD test, Table 2). Urban areas also had a significantly higher pupae density than that in suburban and rural areas, but the difference in pupae density between suburban and rural areas was statistically insignificant (Tukey HSD test, Table 2).
The monthly average density of Ae. albopictus adults was significantly higher in urban areas than that in suburban and rural areas, and it was significantly higher in suburban area than that in rural areas (ANOVA with repeated measures, P<0.05) (Figure 3). The monthly density of adult Ae. albopictus was significantly higher in urban areas than that in the other two sites in all months (Tukey HSD test, P<0.05); and it was significantly higher in the suburban area than that in the rural area every month (Tukey HSD test, P<0.05) except November (P>0.05).
Ae. albopictus adult emergence rates were significantly different among urban, suburban, and rural areas regardless of in natural habitats or with food supplement groups (χ2-test, all P<0.001) (Figure 4). In the natural habitat, Ae. albopictus adult emergence rates was the highest in the urban area (51.5%), then suburban area (19.3%), with the lowest rate in the rural area (13.9%) (χ2-test, all P<0.001) (Figure 4A). In the food supplement group, urban areas had the highest adult emergence rate (χ2-test, all P<0.001), but the difference in adult emergence rates between suburban and rural areas were statistically insignificant (P>0.05) (Figure 4B).
For the natural habitat group, larval development time in urban areas was significantly shorter than that in both the suburban and rural areas (male F = 19.0, d.f. = 2, 92, P<0.001; female F = 20.5, d.f. = 2, 98, P<0.001) (Figure 4, Table S1). The mean developing time from 1st instar larval to adult were 21.4 and 24.2 days for males and females, respectively, in urban areas, but those values were 28.3 days (male) and 32.8 days (female) in suburban areas and 31.7 days (male) and 34.0 days (female) in rural areas. Habitat water temperature in urban areas (25.8±2.7°C) was significantly higher than that in suburban (20.9±3.4°C) and rural areas (20.5±4.1°C) (F = 48.5, d.f. = 2, 174, P<0.001). For the food supplemental group, the larval development time from 1st instar larvae to adults in urban areas was significantly shorter than that in suburban and rural areas (male F = 20.3, d.f. = 2, 23, P<0.001; female F = 9.8, d.f. = 2, 23, P<0.001) (Figure 4, Table S1). Overall, the larvae to adult development time was >50% shorter in control groups than it was in natural habitat groups in all study sites (Figure 4, Table S1).
The larval stage-specific development time were shown in Figure 5. Young larval (1st and 2nd instar) and pupa developed significantly faster in urban areas than that in suburban and rural areas (Tukey HSD test, all P<0.001); old larval (3rd and 4th instar) showed similar development time in the three areas (Figure 5). The stage-specific survival rates also varied among the three areas (Figure 5). Overall, young larvae survived significantly better in urban areas than in suburban and rural areas (χ2-test, all P<0.001). Whereas, survival rates in old larvae and pupae did not show such a difference (all P>0.05) (Figure 5).
From August to September, the life span of female adult mosquito was significantly longer in urban areas than that in suburban and rural areas, but the difference in median survival time between suburban and rural areas was insignificant (Figure 6, Table S2). Adult male mosquito survival time was significantly different among study sites (χ2 = 17.4, d.f. = 2, P<0.001). Male survival time was longest in the suburban area and shortest in the rural area (Figure 6, Table S2). The average outdoor temperatures in urban (29.4±1.7°C) and suburban areas (29.2±0.9°C) were significantly higher than that in rural areas (28.1±1.8°C) (F = 5.4, d.f. = 2, 106, P = 0.0016) (Table S2). Relative humidity were significantly different among rural (87.5±9.4%), urban (82.1±11.1%), and suburban areas (75.9±7.6%) (F = 10.5, d.f. = 2, 106, P<0.001) (Table S2). The mean daily survival rates were similar in all study sites and similar between males and females (Tukey HSD test, all P>0.05) (Table S2). Survival curves were similar in females between urban and suburban areas but different from those in rural areas (Figure 6C and 6D).
From October to November, the median survival of adult female mosquitoes in urban and suburban areas was significantly longer than in rural areas but the difference between urban and suburban areas was insignificant (Figure 6, Table S2). The median survival of males was significantly different among the three sites (χ2 = 181.1, d.f. = 2, P<0.001), with the longest and shortest survival times in urban and suburban areas, respectively (Figure 6, Table S2). The average outdoor temperature in urban areas (24.8±2.6°C) was significantly higher than that in suburban (22.8±3.6°C) and rural areas (21.9±2.4°C), and there was no difference in temperature between suburban and rural areas (Table S2). There was no significant difference in the relative humidity among the three areas. (F = 1.9, d.f. = 2, 134, P = 0.15). Mean daily survival rates were similar in all study sites but differed between males and females (Figure 6, Table S2). Survival curves were similar in females between rural and suburban areas but very different in those from urban areas, which showed prolonged survivorship (Figure 6D).
Stepwise logistic regression revealed that six factors were significantly associated with the presence of immature mosquitoes in the study sites (Table 3). Habitat Ae. albopictus larval presence rate was significantly greater in the urban area than in suburban and rural areas (OR = 1.71, P<0.001), and the suburban area was significantly higher than the rural area (OR = 1.67, P<0.001). The presence of Ae. albopictus larvae was significantly in negative correlation with habitat water depth (OR = 0.03, P<0.001); whereas, it showed positive correlation with habitat water surface area (OR = 3.95, P<0.001). The presence of Aedes larvae was significantly greater in clean water than that in tinted or polluted water (OR = 1.889, P<0.001), and greater in tinted water than polluted water (OR = 1.78, P = 0.034). Shading (regard less of fully shaded or half-shaded), compare to open area, was positively affecting the presence of Aedes larvae (OR = 2.29, P<0.001). The presence of Ae. albopictus larvae was also positively correlated with habitats that have leaves on water surfaces (OR = 2.25, P<0.001), and with habitats that have soil and moss substrates (OR = 1.71, P<0.001).
Outbreaks of dengue fever in China were reported in Hainan province and southern Guangdong province in the 1980s and have been reported in Zhejiang province in 2004, illustrating a 2,000 km expansion from subtropical to temperate areas over 30 years [16]. Among these outbreaks, Ae. albopictus was the only vector reported [13], [20], [44]. Although the causes of dengue fever outbreaks are multi-factorial, environmental changes such as urbanization may be one of the leading factors. We found that in urban areas, there are more Ae. albopictus habitats. In addition, urban areas promoted faster larval and pupal development, and higher larval-to-adult survival rate compared to rural areas. Ae. albopictus mosquito is strongly anthropophilic and has a higher blood-feeding rate in urban areas, where human population density is great, than that in rural areas [29], making it a more susceptible vector in urban areas. Because there is no effective drug therapy or vaccine for dengue fever, vector population control is by far the only effective method for reducing dengue virus transmission. In this context, understanding the vector ecology and biology is essential for developing dengue control strategies. Unfortunately, it is unclear how urbanization impacts the ecology of Ae. albopictus, and the lack of this key knowledge hinders disease control efforts.
We found that, in the similar sampling area, the total number of potential habitats and the number of Aedes-positive habitats were significantly higher in urban areas than in suburban and rural areas. Urban areas have 10-fold higher human population density and more frequent human activities than do suburban and rural areas, leading to a larger number of artificial containers such as abandoned tires, disposable food tins, and flowerpots, which are all favorable breeding habitats for Ae. albopictus [7], [23], [45]. Larger size and higher density of human populations also mean more opportunities for Ae. albopictus blood feeding. Previous study found that Ae. albopictus has a higher blood-feeding rate in urban areas than in rural areas, most likely due to host availability [29]. Additionally, the existence of stable and abundant artificial containers produced by human activities serve as larval development sites, facilitating large mosquito densities in urban areas [11].
Urbanization shifts mosquito breeding sites from natural habitats to artificial habitats. These artificial habitats are usually small containers such as used tires and disposable containers, and they are often directly exposed to sunlight. Therefore, the water temperature in these habitats is higher than in rural areas. In our study sites, the average water temperature of urban habitats was 5°C higher than in suburban and rural areas. Similarly, vegetation changes and land use changes in urban areas may affect the radiation budget and energy balance of the land surface and thus may modify the microenvironments, e.g., food sources that enhance larval survival. These changes facilitate the development of immature Ae. albopictus, i.e., shorten the larval-to-adult development time and enhance the larva survival rate. Our findings are consistent with other studies conducted in different countries and for different mosquito species [24], [30], [46]–[48]. Compared with the food supplemental group, we found that added food sources significantly affect the developmental time and survival rate of immature mosquitoes, which implies that the habitat types in different areas may affect larval development differently due to the difference in availability of nutrients. However, the effects were more pronounced in urban areas than in suburban and rural areas, implying that other factor such as water temperature may play a more important role than food supply in urban areas. These results demonstrated that larvae develop and survive better in urbanized areas, in other words, Aedes larvae is better adapted to urban environment.
Similar to a study conducted in the United States [49], the urban area had a higher pupal and larval density than other two areas; thus, the urban habitats had a higher capacity to support larval development. The reason might be that urban areas had less predators, more nutrition from a “dirtier” environment, or even less drift from agricultural insecticides. Pupal productivity is a good indicator of the abundance of adult mosquitoes [50]–[52]. The surveillance of adult mosquitoes in this study supports this conclusion, i.e., urbanization leads to a higher population density of adult Ae. albopictus.
Higher mosquito density does not necessarily lead to increased disease transmission if adult mosquitoes have a very short life span. We found that both male and female mosquitoes in urban areas had the longest life spans. This result may be due to environmental factors such as air temperature and humidity. The average temperature in urban areas is higher than in suburban and rural areas. Longer adult life spans may enhance disease transmission, although the exact correlation between vector capacity and adult life span needs to be further explored. In this study we fed the adult mosquito with 10% sugar solution without blood, which might have led to exerted stress on the females during multiple gonotrophic cycles and affected the longevity of the female mosquitoes. We observed that the mortality of adult mosquitoes in rural area changed dramatically around day 15, because the air temperature in rural area showed a 5.7°C increase from day 11 to day 15 compared to the first 11 days. This drastic increase in temperature might have influenced the mortality rate of the adult mosquito.
In our survey, we found that the distribution of immature Ae. albopictus was not random. Habitat surface area, canopy coverage, water turbidity, water depth, and substrate type were all important factors influencing habitat selection. These findings confirmed other studies reporting preferences for urban areas [11], [49], shaded containers [49], clean water [53], water with foliage [49], and larger surface area [49], [54]. These results illustrated the complex ecology of Ae. albopictus, which makes controlling this mosquito species difficult in light of its recent global expansion.
In conclusion, the results of this study indicated that urbanization has a significant impact on the ecology of Aedes albopictus. In the urbanizing and urbanized area, the changed environment became more suitable for the growth and development of Ae. albopictus, the condensed population produced more kinds of containers for larval habitats and more blood sources for adult replication. This might be the reason for quick adaptation of Ae. albopictus in urban areas. The epidemic of dengue is largely dependent on vector population. Developing countries such as China and other Southeastern Asian countries experiencing rapid urbanization are under sustained risk of dengue outbreaks.
|
10.1371/journal.ppat.1003329 | Immunodomination during Peripheral Vaccinia Virus Infection | Immunodominance is a fundamental property of CD8+ T cell responses to viruses and vaccines. It had been observed that route of administration alters immunodominance after vaccinia virus (VACV) infection, but only a few epitopes were examined and no mechanism was provided. We re-visited this issue, examining a panel of 15 VACV epitopes and four routes, namely intradermal (i.d.), subcutaneous (s.c.), intraperitoneal (i.p.) and intravenous (i.v.) injection. We found that immunodominance is sharpened following peripheral routes of infection (i.d. and s.c.) compared with those that allow systemic virus dissemination (i.p. and i.v.). This increased immunodominance was demonstrated with native epitopes of VACV and with herpes simplex virus glycoprotein B when expressed from VACV. Responses to some subdominant epitopes were altered by as much as fourfold. Tracking of virus, examination of priming sites, and experiments restricting virus spread showed that priming of CD8+ T cells in the spleen was necessary, but not sufficient to broaden responses. Further, we directly demonstrated that immunodomination occurs more readily when priming is mainly in lymph nodes. Finally, we were able to reduce immunodominance after i.d., but not i.p. infection, using a VACV expressing the costimulators CD80 (B7-1) and CD86 (B7-2), which is notable because VACV-based vaccines incorporating these molecules are in clinical trials. Taken together, our data indicate that resources for CD8+ T cell priming are limiting in local draining lymph nodes, leading to greater immunodomination. Further, we provide evidence that costimulation can be a limiting factor that contributes to immunodomination. These results shed light on a possible mechanism of immunodomination and highlight the need to consider multiple epitopes across the spectrum of immunogenicities in studies aimed at understanding CD8+ T cell immunity to viruses.
| During an infection, the adaptive immune system responds to many epitopes of the pathogen but the strength of these responses varies widely. This unequal distribution of responses across a range of epitopes is known as immunodominance and understanding why it occurs is a fundamental problem in immunology. It is also relevant to vectored vaccines where the intention is to raise immunity against an antigen of choice, but responses to vector epitopes may dominate. We show that the route of infection changes the extent to which the strongest epitope can dominate CD8+ T cell responses to vaccinia virus (VACV). The cause of this phenomenon is linked to virus spread and therefore the different lymphoid organs that prime T cell responses for each route. We also show that local draining lymph nodes are sites of more robust competition between T cells compared with the spleen, explaining why immunodominance differs according to route. Finally the normally heightened immunodominance after peripheral VACV infection can be reduced by expression the costimulators CD80 (B7-1) and CD86 (B7-2) from the virus. In summary, we have carefully dissected immunodominance using VACV as a model and in doing so exposed general features of CD8+ T cell immunity to pathogens.
| Immunodominance is a term used to describe the preferential recognition of some epitopes over others in a complex antigen and is a fundamental property of all immune responses. CD8+ T cell responses to viruses are no exception and immunodominance has been noted for many viruses in mice and humans [1], [2]. Immunodominance arises due to factors that affect either 1) the amount of peptide-MHC (pMHC) complexes, including abundance of parent antigen, ease of processing and affinity of peptides for MHC [3]–[17] or 2) the quantity or quality of T cells in the naive repertoire that recognize these pMHC complexes [5], [8], [10], [11], [18]–[29]. An additional determinant that emerges from the intersection of the factors above is immunodomination, which is the ability of T cells with dominant specificities to inhibit responses to less-dominant epitopes. This is observed most clearly in secondary infections, where some memory T cells are clearly less able to compete [30]–[34]. However, it must also operate in primary infection, because deletion of immunodominant epitopes allows responses to subdominant epitopes to increase [10], [30], [35]. Further in some, but not all cases pre-priming of individual epitopes can lead to radically altered dominance hierarchies, presumably because the already primed T cells have an advantage over other specificities [5], [36], [37]. Finally, competition amongst the various clones recognising the same specificity can be directly observed during infection by monitoring the expansion adoptively transferred TCR transgenic T cells compared with the endogenous polyclonal response [38]. While the mechanism of immunodomination remains obscure, it can be relieved if the epitopes are presented on separate antigen presenting cells (APCs). Therefore it is most likely due to competition for resources either on APCs or released by APCs in the immediate environment, but these remain undefined [36], [38]–[40].
Vaccinia virus (VACV) was used as the live vaccine to eradicate smallpox and attenuated strains are now being used as vectors for recombinant vaccines. In understanding both the historical and contemporary usage of VACV, CD8+ T cell responses are of interest. Further, VACV provides an excellent model for studies of immunodominance because it has a large genome with many mapped epitopes and infections are entirely acute. These attributes set it apart from the well-studied RNA viruses such as influenza or lymphocytic choriomeningitis viruses, with genomes less than a tenth the size and the herpesviruses, all of which cause latent/persistent infections. Of the roughly 50 CD8+ T cell epitopes for VACV identified in the C57BL/6 mouse, B8R20 (TSYKFESV) is by far the most dominant [41]–[45]. Depending on the estimate of the total anti-VACV CD8+ T cell population, 10–25% of all anti-viral CD8+ T cells are specific for this epitope during acute infection and we refer to this as the immunodominant epitope (IDE) [42]. The rest of the mapped epitopes can be considered to be sub-dominant epitopes (SDE) and more than 20 of these induce easily measurable responses that range from 2–0.2% of CD8+ T cell at acute times [41]. In the work characterizing the first five CD8+ T cell epitopes for VACV in C57BL/6, it was noted that different routes of infection appeared to alter immunodominance with intradermal (i.d.) infection favoring B820, the IDE, over the SDE compared to intraperitoneal (i.p.) infection [42]. Responses to an IDE were also favored by the i.d. route in VACV infections of BALB/c mice [46]. That this phenomenon was observed in two strains of mice (with different sets of epitopes) suggests that it was genuinely linked to immunodominance.
Here we confirm, explore and explain the route-related effects on CD8+ T cell immunodominance during VACV infection. We find that immunodominance is linked to the sites of antigen presentation and that individual draining lymph nodes (LN) are environments where there is more competition between T cells of differing specificities than the spleen. Finally we show that the effects of this competition that occurs after i.d. (but not i.p.) infection are reduced by a VACV that expresses the costimulators CD80 and CD86 (B7-1 and 2).
To extend published results suggesting an effect of vaccination route on CD8+ T cell immunodominance, groups of C57BL/6 mice were infected with 1×106 plaque forming units (pfu) vaccinia virus Western Reserve (WR) strain by i.p., i.d., intravenous (i.v.) and subcutaneous (s.c.) injection. After 7 days, CD8+ T cell responses to 15 VACV epitopes (Table 1) were measured using brief ex vivo stimulation of splenocytes followed by intracellular staining for IFN-γ (ICS) (Figure 1A). This method allows very accurate enumeration of epitope-specific CD8+ T cells at acute times after infection [47]. At a broad scale, the overall hierarchy of the 15 epitopes was similar for the four routes, with B820 being the IDE and the other 14 epitopes being SDEs. However, on closer inspection the previously noted increased dominance of the IDE, B820 over the SDE in i.d. compared with i.p. infected mice was repeated here. There were statistically significant differences for several epitopes (including B8) between i.v., i.p. and i.d. routes, but none differed between i.d. and s.c. infections (statistics not shown). To see the ratio of IDE to SDE responses more clearly, the data were plotted to show B820–specific responses as a fraction of the sum of responses to all epitopes (Figure 1B). When viewed this way, responses in mice infected by the two peripheral routes (i.d. and s.c.) had identical IDE∶SDE ratios of around 50%, but use of the systemic i.p. and to a slightly greater extent i.v. route reduced the dominance of B820 to around 30%. Differences between the peripheral and systemic routes were statistically significant (p<0.01). When the data were analyzed to show the total number of CD8+ T cells responding to each epitope, it could be seen that summed responses to the subdominant epitopes are reduced when peripheral routes were used (Figure 1C). The basic phenomenon of skewing towards B820 after i.d. compared with i.p. infection has also been observed in memory CD8+ T cell responses, as measured at 28 days after infection and so is not an artifact of the acute infection (not shown).
Next we wanted to explore whether this phenomenon was related to the particular selection of epitopes used or would apply to other IDE and SDE when expressed as foreign antigens from VACV. One of the most immunodominant CD8+ T cell epitopes mapped to date is the gB498–505 epitope (SSIEFARL) of herpes simplex virus glycoprotein B (HSVgB) and we had two recombinant VACVs expressing this epitope. These express full length HSVgB (VACVfullgB) or the gB498 epitope as an endoplasmic reticulum-targeted ‘minigene’ (VACV-ESminigB) [48]. Infection of mice with these viruses showed that gB498 was a SDE when it required processing from full HSVgB, but became co-dominant with B820 when expressed as an ER-targeted minigene (Figure 1D). These viruses then provided the opportunity to test the effect of route of infection on immunodominance using a single epitope that was either an IDE or a SDE depending on the context. CD8+ T cell responses to the gB498 epitope were then compared to the sum of responses to all 15 VACV epitopes for the two viruses after i.d. and i.v. infection. In the context of full HSVgB, where gB498 ranks as an SDE, responses to this epitope were enhanced roughly two-fold by i.v. compared with i.d. infection, but the opposite result (albeit with a narrower difference) was obtained for the IDE minigene version (Figure 1E). Taken together, the conclusion from these experiments is that VACV infection by peripheral, compared with systemic routes sharpens immunodominance and this can predict which route will maximise responses to foreign antigens expressed from VACV vectors.
VACV has more opportunity to spread after i.p. and i.v. injection, compared with infection by a peripheral route like i.d. [49]–[51]. We reasoned that virus dissemination might be linked to differences in immunodominance if it affected the range of secondary lymph organs that have access to virus antigen. To test the extent to which the inoculated virus spread to lymph organs, mice were infected by i.d., i.p. and i.v. routes and six hours later, virus titers were measured in cervical, mediastinal and mesenteric LNs and spleen (Figure 2A). As expected, after i.v. injection virus titers were highest in the spleen but virus was also found in each LN. For i.d. and i.p. routes, the highest titers of virus were in the cervical and mediastinal LN respectively, which is consistent with the lymphatic drainage of these injection sites. For i.d. infection, spread of inoculum beyond the cervical LN was seen for two of five mice with low titers found in the spleen. After i.p. infection virus was found in the spleen in four of five and mesenteric LN in one of five mice. Next, spread of virus during the first day of infection was explored by quantifying virus 24 hours after infection by the same three routes (Figure 2B). At this later time point virus spread remained wide after i.v. injection (though titers dropped substantially) and this broad distribution was also seen for mice infected by the i.p. route. In fact the titers were generally higher after i.p. compared with i.v. infection. In contrast, virus was entirely restricted to the cervical lymph node after i.d. infection. These results demonstrate that there was wider spread of virus and potential for antigen presentation after i.v. and i.p. than i.d. infection.
Having shown wider spread of virus after i.v. and i.p. routes we wanted to know if this was reflected in the sites where CD8+ T cells were primed. The first method used was the in vivo cytotoxicity (CTL) assay, which has been used previously to track sites of T cell priming [52]. In the first set of experiments, specific killing of B820-pulsed targets in various lymphoid organs was determined at different times after i.d. infection (Figure S1 in Supporting Information). This showed that at two days after i.d. infection, B820-specific killing activity was already found in the cervical LN, but it took several more days until it was detected at similar levels in other LN and spleen, similar to a report with dermal HSV infection [52]. The killing seen at the other LN and spleen on later days presumably reflects the recirculation of primed CD8+ T cells once they leave the site of initial priming. We then examined killing two days after infection by the i.v. and i.p. routes (Figure 3A). Two days after infection by the i.v. route, B820-specific killing was high in all LN and spleen, suggesting priming of CD8+ T cells in all these sites. Infection by the i.p. route produced the highest B820-specific killing in the mediastinal LN, but killing was also strong (around 20%) in spleen and other LN. These data suggested a difference in the amount of priming at sites beyond the local draining LN after i.p., compared with i.d. infection and very widespread priming after i.v. injection.
It was unclear in these experiments if the roughly 10% killing in the spleen and non-draining LN after i.d. infection was due to the early migration of some effectors or was background as an artifact of the assay. To look at the sites of priming even earlier, a recombinant VACV (NP-S-GFP) expressing the ovalbumin257 (SIINFEKL; OVA257) epitope [53], [54] was used in combination with transferred naive OT-I T cells. The dominance of B820 after infection of mice with VACV NP-S-GFP by i.v., i.p. and i.d. routes was similar to that seen for non-recombinant WR (Figure S2 in Supporting Information). To detect the earliest events of priming we looked for CD69 up-regulation on the OT-I cells 24 hours after infection of mice by the three routes (Figure 3B). After i.d. infection, priming of OT-I T cells, as indicated by CD69 up-regulation, was very low in all sites other than the cervical LN. In contrast, i.p. infection primed over 60% of OT-I cells both in the mediastinal LN and the spleen. As expected, i.v. infection was able to prime OT-I in all sites. So by this method, priming of CD8+ T cells at sites beyond the local draining LN is efficient after the systemic i.v. and i.p. routes, but not the peripheral i.d. route of infection. While we saw differences in the spread of priming sites, it remained possible that the length of antigen presentation also differed and that this might drive changes in immunodominance. To test this we used the OT-I model again, but this time did transfers at various times after infection (day 1, 3 and 5) by i.d. or i.v. route and looked for CD69 up regulation 24 hours later (Figure 3C). Overall, there was no difference in kinetics of presentation between the routes and priming of OT-I was observed on all days, though it began to wane at the latest time (day 5 to 6). However, we were surprised to see that priming in the spleen was relatively poor at all times after i.v. infection in this experiment, which was in contrast to findings earlier after infection (Figure 3B). To ensure that this was sound, an experiment was done to include a day 1 readout of OT-I activation side-by-side with later times after i.v. infection and this confirmed that priming in the spleen declines more rapidly than in LN (not shown). Interestingly, by day 3–4, there is little difference in priming in the spleen between the i.v. and i.d. infected mice. All together these experiments show that a difference between the systemic and peripheral routes is the extent to which antigen spread allows CD8+ T cell priming to take place at early times in lymphoid organs beyond the LN and perhaps especially in the spleen.
Next we wanted to explore whether the number of priming sites and/or levels of presentation lead to reduced immunodominance. First, peripheral routes were explored and priming in more LN was achieved by simultaneous injections (two i.d. and two s.c.) at four different sites. Seven days later, responses to the panel of 15 VACV peptides were determined. In mice that received these multiple concurrent injections, responses to the VACV epitopes were similar to those in mice infected at a single site, with B820–specific CD8+ T cells accounting for 50% of the total epitope-specific response (Figure 4A and compare with Figure 1B and C). This suggested that if priming was restricted to LN, increasing the number of lymphoid organs where priming occurs does not reduce immunodominance. In the second experiment, the dose of virus given was reduced to 1×103 PFU, which we reasoned would stop the spread of virus to the spleen after i.p. injection and also greatly limit the number of APCs irrespective of route. First, VACV NP-S-GFP and OT-I transfer was used to directly examine sites of priming after this low-dose infection (Figure 4B). At 24 hours after i.p. infection, only modest priming (5% of OT-I activated) was observed with this dose and it was exclusively in the mediastinal LN. When a similar low dose was used by the i.v. route, again priming was poor (10% of OT-I), but in this case it was found only in the spleen. Therefore the reduced dose clearly limited the spread of antigen and the levels of presentation. Next we examined the effect of this reduced dose on immunodominance and found that mice infected with a low-dose by the i.p. or i.v. routes had CD8+ T cell responses more heavily dominated by the IDE, which now accounted for 50% and 60% of the epitope-specific response respectively (Figure 4C). This is similar, if not more extreme domination by B820 as seen for i.d. infection at the standard high dose. By contrast, reducing the dose to 1×103 PFU in i.d. infection did not further sharpen immunodominance for this route. The lower dose given by all routes resulted in much lower total numbers of epitope-specific CD8+ T cells and again in the case of i.p. and i.v. routes it was SDE-specific responses that were most reduced (Figure 4D). For example reducing dose in the i.v. route roughly halved the number of B820-specific CD8+ T cells, but reduced the number of SDE-specific T cells by around three quarters. In case the dose changed the kinetics of response for the various epitopes, the immunodominance hierarchy was examined at days 6, 7, 8, 9 and 11 after i.p. infection and the dominance of B820 was remarkably stable across these five times (Figure S3 in Supporting Information). From these data it was concluded that simply increasing the number of lymph organs where priming occurs does not reduce immunodominance, neither does priming in the spleen guarantee this outcome. Rather, only where priming in the spleen is substantial can immunodominance be reduced. This suggests a requirement for a higher number of APCs being involved in priming in one organ (as offered by the spleen), rather than there being a special property of splenic versus LN APCs.
The results thus far suggested that the sharpened immunodominance after i.d. infection could be due to a limiting resource needed for priming (e.g. number of APCs) in LN and therefore greater potential for immunodomination. To examine immunodomination directly, cross competition between transferred OT-I cells responding to OVA257 expressed from VACV NP-S-GFP and endogenous CD8+ T cells responding to the VACV epitopes was examined. This approach was originally used to examine competition between T cell clones for OVA257, but OT-Is can also cross-compete with native virus epitopes co-expressed with OVA257 [38], [55]. Two doses (1×103 and 1×105) of congenically marked OT-I T cells (CD45.2+) were transferred into B6.SJL mice (CD45.1+), which were infected with VACV NP-S-GFP a day later. After seven days, splenic responses to the set of 15 VACV peptides (CD45.1+) and the number of responding OT-I T cells (CD45.2+) were measured by ICS. This experiment was done with mice infected by the i.d. route and the i.v. route to represent the greatest and least domination by B820 in the experiments shown thus far. When data were analyzed either as a percent of CD8+ T cells (Figure 5A) or as total number of CD8+ T cells (Figure 5B), responses to VACV epitopes were significantly reduced by the presence of competing OT-I T cells after i.d., but not i.v. infection. This is despite the relatively larger expansion of the transferred OT-I cells in i.v. infected mice. No significant suppression of VACV-specific responses by OT-I transfer was seen when mice were infected with a control virus that did not express the OVA257 peptide (Figure S4 in Supporting Information). It was possible that the transfer of so many OT-I cells might influence the amount of virus growth and therefore antigen presentation during i.d. infection. This was tested by removing the ears from i.d. infected mice that had received OT-I or no transferred cells and determining virus titers by plaque assay. There was no difference in the amount of infectious virus found in ears of mice that received OT-I cells, compared with controls (Figure 5C). Together these data suggest that CD8+ T cell responses are more prone to immunodomination after i.d., compared with i.v. infection.
Immunodomination is most likely the result of T cells competing for limited resources on or very close to APCs, but none have been identified. One essential resource required for CD8+ T cell priming that has been proposed, but not shown to be involved in immunodominance is costimulation [56]–[58]. These studies used mice deficient in CD28 or treated mice with soluble reagents that block the interaction between CD28 and the costimulators CD80 (B7-1) and CD86 (B7-2). However, this approach only tests the effect of eliminating costimulation and so cannot reveal whether costimulation under normal conditions can be limiting and might be a resource for which CD8+ T cells compete. To test this possibility, we used a recombinant VACV expressing CD80 and CD86 (VACV-CD80&86) and infected mice by the i.d. and i.p. routes. Priming of CD8+ T cells by VACV is thought to be largely via direct presentation, so this virus should increase the amount of costimulation available on each APC [53], [59], [60]. Consistent with a role for limiting CD80 and CD86 playing a role in immunodomination, mice infected i.d. with VACV-CD80&86 had higher responses to SDEs and the ratio of IDE to SDE was significantly lower than in mice infected with the control VACV (Figure 6). By contrast, in the i.p. infected mice there was no enhanced response to any epitope and immunodominance was unaltered. Further when analyzed by total number of VACV-specific CD8+ T cells, in mice infected by the id. but not i.p. route expression of CD80 and CD86 improved the sum of responses to SDEs, but not to B820. From these results we conclude that expression of costimulators by a recombinant VACV reduces immunodomination.
Infection route has been suggested to alter several aspects of CD8+ T cell responses to a variety of viruses including priming mechanism, magnitude and quality [61]–[65]. Here we demonstrate clearly that for VACV, immunodominance also needs to be considered. This leads to the first important conclusion of this work, which is that where magnitude of response is a primary read-out, examining responses to a single epitope can be misleading. For example, if B820 was used as a sole epitope in the experiment shown in Figure 1A, one would conclude that i.d. or s.c. injections of VACV were most immunogenic. In contrast if the majority of the SDE were chosen, the opposite conclusion would be drawn. The size of differences in response for individual SDE varied, but were up to four-fold for C4125 across the routes. Analyzing the total number of epitope-specific CD8+ T cells removes the apparent improvement of B820-specific response seen by the peripheral routes, but makes the suppressive effect of these routes on the SDE more pronounced, with the difference for C4125 being more than seven-fold. So while as an overall picture, the change in dominance profile across the doses seems quite modest compared with the very big differences in virus spread and number of priming sites, effects were substantial for some individual epitopes. Much VACV immunology in the past has used recombinant viruses and responses to a single foreign epitope were monitored. In the light of our results some conclusions from these earlier experiments may need to be reconsidered. Indeed the experiments shown here with HSVgB498 demonstrate that different forms of antigen can change the dominance ranking of an epitope dramatically in the context of a recombinant VACV. This in turn alters its competitiveness as an immunogen differentially according to route. There are also lessons here for studies of immunodominance using other viruses where few epitopes are known or used.
The original experiments that indicated a role for route in VACV immunodominance used only dermal scarification and i.p. routes [42]. By including i.v. and s.c. routes here, the association between virus spread and immunodominance was noted and then confirmed virologically. The lack of spread after i.d. infection was not surprising [49]. However, the much larger amount of virus found in all organs 24 hours after i.p., compared with i.v. injection was less expected. The reduction in virus in all lymph organs from 6 hrs to 24 hrs after i.v. injection suggests that these sites do not sustain VACV replication. Therefore the higher amounts of virus found a day after i.p. injection are most likely the result of continued draining of virus generated at other sites, rather than infection of the lymph organs. This is an advantageous feature of the VACV model in that findings made in lymph organs are not complicated by these also being major sites of virus infection. In terms of CD8+ T cell immunity, presence of any virus in lymphoid organs was a useful a guide for defining priming sites, even if the amounts measured were poorly predictive of antigen presentation levels. For example, despite very low levels of infectious virus being found in the various LN after i.v. infection, priming was robust at all these sites. Perhaps in this situation many APCs are reached by the ample (1×106 PFU) inoculum and despite undergoing abortive infections, these cells persist long enough to prime effectively. Alternatively there may be a reservoir of antigen that is cross presented, but the evidence thus far suggests that direct priming is more important for VACV-specific CD8+ T cells [60]. The kinetics of antigen presentation as determined for the i.d. and i.v. routes were remarkably similar with sustained activation of OT-I cells seen until day 5–6, though it was decreasing by this time. This result rules out the premature loss of antigen presentation in LN, perhaps as a result of killing of APCs by IDE-specific T cells, as a mechanism for increased immunodominance associated with the i.d. route. Further the one organ where presentation seemed to decay the fastest was the spleen after i.v. infection and antigen presentation levels there became very similar for i.d. and i.v. routes by day 3–4. This points to early antigen presentation events being more important in setting the immunodominance hierarchy, possibly reflecting the requirement for only a brief encounter with antigen to drive CD8+ T cell responses [66], [67].
We considered the possibility that there was something qualitatively different between the APCs in the spleen and LN that leads to reduced immunodominance. If this were correct, reduced immunodominance should be the hallmark of any responses primed in the spleen. However, reducing virus dose by i.v. and i.p. routes and thereby restricted priming to the spleen and a LN respectively, sharpened immunodominance in both cases. On the other hand, the lower doses also greatly limited the number of APCs available at any priming site, as demonstrated by poor priming of OT-I cells. Strikingly, restricting priming to the mediastinal LN with low dose i.p. infection lead to exactly the same ratio of IDE∶SDE as found for any dose of virus injected i.d. where priming also occurs in a single LN. Further, increasing the number of LNs involved in priming and thereby total APC numbers by infecting multiple peripheral sites did not reduce immunodominance. To reduce immunodominance, a large number of APCs in a single lymph organ were required and this was only provided by the spleen when there was an abundance of antigen. Finally, reducing the virus dose given i.d. did not further sharpen immunodominance, which together with other results here suggests there is a limit to immunodomination. We speculate that the architecture and size of LNs limit APCs and/or some associated essential resources required for priming over a wide range of antigen doses and increase competition between T cells. The possibility of a more competitive environment in LN was confirmed by subjecting VACV-specific CD8+ T cells to rivalry from transferred OT-I T cells. The finding that VACV responses were more easily suppressed by OT-I after i.d. but not i.v. infection is consistent with greater competition across specificities when priming is restricted to LN. This leads to the conclusion that immunodomination in LNs, which are the main priming sites after i.d. infection, suppresses responses to SDE.
Having established i.d. infection with VACV as a setting where immunodomination can occur in primary responses, we decided to take advantage of this model to examine the role of costimulation in immunodominance. Our data suggest that increasing levels of costimulators to APCs that directly prime CD8+ T cells in a LN reduces the level of immunodomination by the IDE. Conversely, there was no advantage for SDE (or the IDE) when these constimulators were expressed by a virus given by the i.p. route. This effect for the i.d. route might be achieved either by reducing competition for costimulation on individual APCs or possibly extending the number of APCs that have adequate levels of costimulators to prime CD8+ T cells. Several groups have demonstrated that CD8+ T cell responses can occur in the absence of costimulation via CD80/CD86 and CD28, but that these responses are substantially compromised [58], [68]–[71]. The only report to examine multiple epitopes came to the conclusion that costimulation affected IDE and SDE equally [58]. This is not necessarily inconsistent with our findings or conclusions. We are not suggesting that IDE and SDE have a differential requirement for costimulation, but rather that in the context of priming in LNs, costimulation is limiting and IDE have an advantage in competing for this resource. The VACV strains that express CD80 and CD86 used here were originally developed to be improved vaccine vectors [72]. Other reports with similar viruses suggest that they enhance magnitude and avidity of CD8+ T cell responses to model antigens that are co-expressed from the same virus [73], [74]. While we have not tested avidity, our data on magnitude are consistent with these published reports: the epitopes examined previously would rank as SDE and the immunizations published were done using a peripheral route (s.c.). However, the assumption made until now was that all CD8+ T cell responses could be boosted by expression of the costimulators. Here we show that the benefit of expressing costimulators, at least in terms of magnitude of response may be for SDE only. This reinforces again the importance of examining responses to multiple epitopes before drawing conclusions about the benefit of different immunization strategies based on pre-clinical models in mice.
In conclusion, we show here that route-related changes in immunodominance after primary infection with VACV are the result of differential spread of virus antigen, which determines the sites of CD8+ T cell priming. CD8+ T cell priming is more competitive when it is mainly limited to LNs and consequently subdominant specificities are subject to greater immunodomination at these sites. Further, we identify costimulatory molecules as one of the resources that might be limited in LN and therefore drive immunodomination. These data have implications for the interpretation of preclinical vaccinology of vectored vaccines. Beyond these insights this work has ramifications for viral immunology in general, demonstrating clearly the importance of putting responses to any single epitope into the broader context of responses to the whole virus.
The majority of viruses used here were kind gifts: WR (Bernard Moss, NIH); VACV NP-S-GFP [54], [75], WR B7-1&B7-2 (called VACV-CD80&86 here) [72] and VSC-8 (as a TK− control virus for VACV-CD80&86) (Jon Yewdell and Jack Bennink, NIH); VACV-ESminigB [48] (S. Tevethia, Penn State Medical College). VACV-fullgB was made by standard homologous recombination methods using plasmid pSC11 [76] to insert the full coding sequence of HSVgB under the control of the p7.5 promoter into the thymidine kinase gene of VACV WR. All recombinant antigens in viruses used here were expressed by the p7.5 promoter from a disrupted thymidine kinase gene of strain WR. Immortalized cell lines, BHK-21 and BS-C-1, were maintained in Dulbecco's Modified Eagle medium (DMEM, Invitrogen) with glutamine and 10% or 2% fetal bovine serum (FBS). VACVs were grown in BHK-21 and purified by centrifugation through a 36% sucrose cushion, then infectivity was titrated on BS-C-1 cells using standard methods.
All experiments were done according to Australian NHMRC guidelines contained within the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes and under approvals F-BMB-38.8 and A2011-01 from the Australian National University Animal Ethics and Experimentation Committee.
Specific pathogen-free C57BL/6, B6.SJL and OT-I transgenic mice were obtained from the Animal Resource Centre (Perth, Australia) and APF (Canberra, Australia). Unless stated, mice were infected with 1×106 PFU of VACV by i.p., i.v. or s.c. in 200 µl of PBS or by i.d. [77] in 5 or 10 µl of PBS.
Mice were euthanized 7 days after infection and spleens were taken for analysis of CD8+ T cell responses by intracellular cytokine staining (ICS) as described previously [47]. Briefly, splenocytes were plated at 1×106 cells/well into round-bottom 96-well plates. Synthetic peptides (Table 1) were added to a final concentration of 10−7 M and plates were incubated at 37°C with 5% CO2. After 1 hour, 50 µg/ml brefeldin A (Sigma-Aldrich, St. Louis, MO) was added. Plates were incubated for another 3 hours, then spun at 4°C to remove the medium. Cells were resuspended in 40 µl of 1/150-diluted anti-CD8-PE antibody (clone 53-6.7; BD Biosciences, San Jose, CA). After 30 min incubation on ice, cells were washed, resuspended in 50 µl of 1% paraformaldehyde, and incubated at room temperature for 20 min. After two washes, cells were stained with 40 µl of 1/200-diluted anti-IFN-γ allophycocyanin antibody (clone XMG1.2; BD Biosciences) with 0.25% saponin (Sigma-Aldrich) overnight at 4°C. Cells were washed three times before acquisition using a FACS LSRII (BD Biosciences). Analysis was done in Flowjo software (Tree Star, Ashland, OR). Events were gated for live lymphocytes on forward scatter (FSC) × side scatter (SSC), followed by CD8+ cells × IFN-γ. Backgrounds as determined for samples without peptide were usually in the order of 0.1–0.2% and were subtracted from the values presented for test samples.
Organs harvested from infected mice were homogenized in 1 ml glass homogenizers (Wheaton) then rapidly frozen and thawed 3 times in liquid nitrogen and a 37°C water bath. The homogenized organs were then 10-fold serial diluted in DMEM with 2% FBS before adding to the 6-well plates with BS-C-1 cell monolayer. After 90 min of incubation at 37°C with 5% CO2, the virus inoculates were removed and replaced by 2 ml/well of 0.4% Sodium carboxymethyl cellulose (CMC, Sigma-Aldrich) in DMEM with 2% FBS. Plates were incubated at 37°C with 5% CO2 for 3 days, then crystal violet (Sigma-Aldrich) used to stain. Plaques were counted and virus titers were determined according to the dilution factor.
CD8+ T cells were prepared from the spleens and lymph nodes of naive OT-I mice using magnetic bead-based negative selection (Miltenyi), and resuspended in PBS with 2% FBS for i.v. transfer. Each B6.SJL mouse received about 5×106 OT-I CD8+ T cells and rested for >18 hours. 24 hours after infection with 106 PFU of VACV NP-S-GFP by various routes, these infected and naive mice were sacrificed. Single cell suspensions were made from the spleens and lymph nodes, and stained with anti-CD8-PE antibody (clone 53-6.7; BD Biosciences, San Jose, CA), anti-CD45.2-APC antibody (clone 104; BioLegend, San Diego, CA) and anti-CD69-PE-Cy7 (clone H1.2F3; BD Biosciences, San Jose, CA) before acquiring by FACS LSRII (BD Biosciences). The activation of OVA-specific T cells is determined by CD69 expression on CD45.2+CD8+ cells.
Target cells for cytotoxicity detection were prepared from naive C57BL/6 mice and split into two populations for peptide-pulsing with B820 and SIINFEKL respectively at 37°C for 1 hour. These two populations were then labeled with different concentration of CFSE (Sigma-Aldrich), 9 µM for B820 (CFSEHi) and 0.9 µM for SIINFEKL (CFSELo), before mixing equal number of cells together for i.v. transfer. A total of 2×107 cells were injected into each mouse which was infected with 1×106 PFU of VACV WR for days as described. After 4 hours, the mice were sacrificed for their spleens and lymph nodes. The cell numbers in two CFSE-positive populations were acquired by FACS LSRII (BD Biosciences) for analysis. The specific lysis was calculated by the following formula: [1−(CFSELo/CFSEHi)naive/(CFSELo/CFSEHi)infected]×100.
Unless stated otherwise, statistical comparisons were done using an unpaired t test with Welch's correction because populations had unequal variance. All tests were analyzed with the aid of GraphPad Prism software (GraphPad, La Jolla, CA).
|
10.1371/journal.pgen.1001337 | Quantifying the Underestimation of Relative Risks from Genome-Wide Association Studies | Genome-wide association studies (GWAS) have identified hundreds of associated loci across many common diseases. Most risk variants identified by GWAS will merely be tags for as-yet-unknown causal variants. It is therefore possible that identification of the causal variant, by fine mapping, will identify alleles with larger effects on genetic risk than those currently estimated from GWAS replication studies. We show that under plausible assumptions, whilst the majority of the per-allele relative risks (RR) estimated from GWAS data will be close to the true risk at the causal variant, some could be considerable underestimates. For example, for an estimated RR in the range 1.2–1.3, there is approximately a 38% chance that it exceeds 1.4 and a 10% chance that it is over 2. We show how these probabilities can vary depending on the true effects associated with low-frequency variants and on the minor allele frequency (MAF) of the most associated SNP. We investigate the consequences of the underestimation of effect sizes for predictions of an individual's disease risk and interpret our results for the design of fine mapping experiments. Although these effects mean that the amount of heritability explained by known GWAS loci is expected to be larger than current projections, this increase is likely to explain a relatively small amount of the so-called “missing” heritability.
| Genome-wide association studies (GWAS) exploit the correlation in genetic diversity along chromosomes in order to detect effects on disease risk without having to type causal loci directly. The inevitable downside of this approach is that, when the correlation between the marker and the causal variant is imperfect, the risk associated with carrying the predisposing allele is diluted and its effect is underestimated. Using simulations, where we know the true risk at the causal locus, we quantify the extent of this underestimation. We show that, for loci which have a modest effect on disease risk and are common in the population, the risk estimated from the most associated SNP is very close to the truth approximately two thirds of the time. Although the extent of the underestimation depends on assumptions about the frequency and strength of the risk allele, we predict that fine mapping of GWAS loci will, in rare cases, identify causal variants with considerably higher risk. Using three common diseases as examples, we investigate the expected cumulative effects of underestimation at multiple loci on our ability to stratify individuals by disease risk and to explain disease heritability.
| Genome-wide association studies (GWAS) have been extremely successful across many diseases in identifying loci harbouring genetic variants that affect disease susceptibility. Virtually all associated variants identified from GWAS to date have relatively small effects: each additional copy of the risk allele typically increases disease risk by 10%–30% (see for example [1]). It has become clear that the variants discovered thus far account for only a small proportion of the genetic basis of each of the diseases, and there has been considerable speculation about where the “missing” heritability might lie [1].
One of several important factors in the success of the GWAS design has been the pattern of linkage disequilibrium in human populations. The strong correlations between nearby SNPs mean that commercially available genotyping chips, which assay 300,000–1,000,000 SNPs, can capture much of the common variation in the human genome, particularly in Caucasian populations [2]. Because genotypes at the causative loci will often be correlated with those at SNPs that are typed on the genotyping chip, it is typically not necessary to assay the true causative variant directly in order to detect a genetic association with disease.
While linkage disequilibrium is extremely helpful for GWAS discovery, the downside is that in most reported regions of association, the true causal variant or variants remain unknown. Therefore it is possible that many of the associated SNPs are only surrogates for the true causal variant(s). When it comes to quantifying the genetic effect, the genotype at the reported SNP acts as a noisy measurement of the genotype at the causal variant. This noise can dilute the apparent strength of the effect, and obscure the true relationship between genotype and phenotype. As we progress towards the identification of the causal variants, estimates of effect sizes for associated loci will thus tend to increase. In turn, the proportion of disease susceptibility explained by GWAS loci will also increase. Thus in addition to other plausible sources, such as secondary signals in GWAS loci, rare variants (<1% frequency), copy number polymorphisms, and epigenetic effects, some of the missing heritability is actually contained in loci already identified by GWAS, and is driven by common variation (>1% frequency).
In this paper we use an extensive simulation study to investigate, and quantify, this phenomenon. We show that estimates of the size of the genetic effect based on the best SNP from the GWAS genotyping chip can often closely approximate the effect size at the true causal SNP. In some cases the causal SNP has a large effect and is poorly tagged, leading to substantial underestimation of the true effect size. We investigate how much of the “missing” heritability could thus be hidden in reported GWAS loci, under several sets of assumptions about the nature of the effects at true causal SNPs. Our results also inform the design and value of fine mapping experiments in GWAS loci.
Patterns of linkage disequilibrium (LD) in human populations are complicated, and preclude analytical results, so we adopted a simulation approach (see Methods for details). We describe the approach informally before describing our results. First, we chose each allele at each SNP in the HapMap ENCODE regions in turn, assuming it to be causative with a given effect size. We then used a previously reported simulation scheme (HAPGEN, [3]) to simulate a large population of chromosomes with European ancestry, whose patterns of LD match those in the CEU HapMap analysis panel. From this population a case-control sample is taken, with the controls sampled randomly from the population and the cases chosen by oversampling chromosomes carrying the causal allele in the appropriate way given its frequency and assumed effect size. To simulate a GWAS, we considered samples of 2000 cases and 2000 controls typed on the Affymetrix GeneChip Human Mapping 500K Array Set (see www.affymetrix.com). No single sample size can model all reported GWAS, but this size is typical of many. (Later, when considering associated loci from specific diseases that have been studied extensively, we simulated GWAS of larger size.) To simulate a GWAS on a particular commercial chip, we examined data at only those SNPs on the chip in question and checked to see whether any of these SNPs showed a p-value for association <10−6. If this occurred we then modelled a replication study, using an additional 2000 cases and 2000 controls for definiteness. We took the best SNP from the simulated GWAS and examined it in the simulated replication sample to check whether it had p<0.01 in this replication sample. In what follows we only considered those simulations where the best SNP on the genotyping chip met both these criteria, as these model the ascertainment implicit in reported GWAS associations. For these simulations, we estimated the effect size at this associated SNP, which we call the hit SNP, in the replication datasets and compared it with the true effect size at the causal variant used for the simulation. The fact that we estimate the effect size from the replication data set is important, because it minimises the effect of “winner's curse”, which would otherwise lead to the effect sizes being over estimated [4]. Simulated GWAS and replication samples were generated for a range of assumed true effect sizes.
Reported genome-wide association studies differ in many particular details, including the choice of genotyping chip used and the sizes of the discovery and replication samples. Specific assumptions are necessary for any simulation study, and ours aim to capture the general features of many reported studies. Investigation of different simulation scenarios, including different genotyping chips and sample sizes, did not change the broad conclusions that follow (data not shown).
To begin, we compare the estimated effect size at the replicated hit SNP with the true effect size at the causal SNP in the simulation. Figure 1 illustrates this comparison for three different values of the true effect size. For each we see a peak of estimates around the true effect size assumed at the causal SNP. But note also that there is often underestimation of the true effect size (mean estimated effect size 1.24, 1.86 and 3.32 for true relative risk of 1.25, 2 and 4 respectively), and that this underestimation can be substantial when the true effect is large. For example, when the true relative risk is 4, the estimated effect size was less than two in 12% of simulations of successful GWAS discovery of the effect.
In Figure 2 we plot the relative under- (or over-) estimation of the effect size (estimated effect size divided by true effect size) as a function of the correlation (as measured by the r2 which is the square of Pearson's correlation coefficient) between the hit SNP and the true causal variant. The underestimation is seen to be due to imperfect tagging: when the true causal variant is not well tagged by SNPs on the genotyping chip (the correlation is weak), the estimated effect at the hit SNP is often much lower than the true effect. Conversely, when the causal SNP is well tagged by a SNP on the chip, the estimated effects cluster around the true effect size. Note that while underestimation decreases as the correlation between the hit SNP and the causal SNP increases, there remains systematic underestimation even when the hit SNP has r2≈0.8 with the causative SNP. For example in one third of simulations when the true effect is two, the estimated effect will be under 1.8. Note also that when the true effect size is large, significant and replicable associations can be detected when the best tag SNP only has r2≈0.2 with the causal variant (Figure 2, relative risk = 4).
Imperfect tagging and an ascertainment effect also explain the feature of the plots whereby the underestimation is much less for smaller true effect sizes. If the true effect is small and the true causal variant is not well-tagged on the genotyping chip, there will not be enough power for the GWAS and subsequent replication to reach significance [5], with the result that the corresponding simulation will not contribute to the plot. But if the true effect is large there may still be power to see a significant result when the true variant is not well tagged, so the simulation contributes to the plot and shows the underestimation. Put another way, if the true effect is small, it will only be detected in an association study if the causal SNP is well tagged, and in this case the effect size will be estimated reasonably well. This second ascertainment effect explains the lack of underestimation at hit SNPs not strongly correlated to the causal SNP in the left panel of the Figure 2. Lastly, as low frequency SNPs are less well tagged by other SNPs [6], the extent of the underestimation also depends on the frequency of the risk allele (see Figure S1). Interestingly, the effect sizes at rare alleles are underestimated to a great extent, but only when the true effect size is large enough for the tag SNP of a rare allele to be detected and replicated in the simulated GWAS.
The results above describe the distribution of estimated effect sizes as a function of known true effect sizes and the frequency of the risk allele. In practice we are actually interested in the reverse question, namely what true effect sizes are plausible in the light of the effect size actually estimated from a GWAS and follow-up study? We will see that this requires assumptions about the true distribution of effect sizes. Indeed, writing RR for relative risk, and RAF (risk allele frequency) for the allele frequency at the risk allele, application of Bayes' theorem gives(1)where “true” refers to the value at the causal SNP and “observed” refers to the value at the hit SNP. Our simulation study allows us to estimate the first factor on the right hand side of (1), and we do so by discretising both the observed and true RR and RAF and creating a matrix of counts based on our simulations over the ENCODE regions. The second factor on the right hand side of (1) is the assumed joint distribution of true risk allele frequencies and effect sizes, which is of course unknown.
We proceed by making two different sets of assumptions about these unknowns. In each case we assume that the distribution of risk allele frequencies is given by the empirical distribution of allele frequencies in the ENCODE regions. In effect this assumes that any SNP variant is, a priori, equally likely to affect disease status. What differs between the sets of assumptions is the assumed effect size of a particular variant. Our first set of assumptions posits that the distribution of effect sizes is the same for all putative causal variants, regardless of their allele frequency, and that effect sizes are close to those observed in GWAS studies. The second set of assumptions explicitly assumes that there might be substantially larger effects at variants with smaller minor allele frequency. These priors are described in detail in the Methods section.
Different sets of assumptions about true effect sizes and risk allele frequencies necessarily lead to different conclusions, and it is impossible to study all possibilities. A number of theoretical analyses [7], [8], [9], [10] have argued for a relationship between effect size, disease model, and minor allele frequency (MAF). As there is no consensus on the exact form and extent of the relationship we do not rely on them explicitly here, and instead our approach aims to capture two different perspectives on unknown effect sizes, with the subsequent analyses indicating a range of possibilities. The first perspective is that the range of true effect sizes will be close to those estimated from current GWAS. The second captures the possibility that low-frequency variants may have considerably larger effect sizes.
Under either set of assumptions, we can use our simulation study, and Bayes' Theorem (1) to estimate the conditional distribution of true effect sizes and risk allele frequency (RAF) in the light of the observed data at the GWAS hit SNP. Figure 3 illustrates this, showing estimates of the posterior distribution of the true effect size conditional on observing a risk estimate between 1.2 and 1.3, for different observed risk allele frequencies, and under the two different prior assumptions on effect size distributions.
A common feature of the histograms in Figure 3 is that the mode of the posterior distribution on the true effect size is on, or very closes, to the observed estimate. That is, current estimates from GWAS studies of effect sizes from a common SNP, in the range 1.2–1.3 are most likely to be very close to truth. As expected, estimated effects within this range are more likely to be 1.3 than 1.2, because larger effects are more likely to generate a signal of association strong enough to pass the p-value thresholds commonly implemented in GWAS. This explains the left hand tail of the distributions represented in Figure 3.
Figure 3 also shows that there is some probability that the effect size at the causal variant is greater than estimated from the most associated SNP. Interestingly, the observed risk allele frequency impacts our posterior belief about the true effect size, under either set of prior assumptions, with underestimation be more marked when the risk allele at the hit SNP is rarer. Under the conservative prior, when the risk allele at the hit SNP has less than 20% frequency in the control population, the probability that the relative risk is above 1.325 is 55%, compared to 35% when the risk allele frequency is between 20–50%. The corresponding numbers for the MAF-dependent prior are 77% and 49%. There are several different phenomena at work here. If the hit SNP is the causal SNP then, assuming that the association is strong enough to be detected and replicated in the GWAS, there is no systematic under estimation (and very little over estimation as we assume the effect size is estimated from the replication sample). However, conditional on the hit SNP not being causal, the distribution of LD with true causal SNP, and therefore the propensity for under estimation, depends on its allele frequency. The posterior distribution on the true effect size given the observed frequency and effect of the hit SNP can be viewed as a mixture of these two scenarios, weighted by their conditional probability. Rarer SNPs are less likely to be tagged well by single markers, and as noted above, poor tagging leads to underestimation of effect sizes. In contrast, for a common SNP, the associated allele is more likely to be well correlated with the causal allele, so there is relatively less under estimation. Under the MAF-dependent prior, when the associated allele is low-frequency the causative allele will tend to be low-frequency as well, and so potentially of larger effect. In the scenario where we believe in larger effects at rare causal alleles and have observed a SNP with low RAF with estimated relative risk between 1.2 and 1.3 there is a 24% chance that the source of the signal is a variant which actually doubles or more than doubles risk with each copy of the risk allele.
Our observations are similar when the observed risk allele is the most common allele in the population (RAF>50%) and therefore the minor allele is protective (Figure S2). Qualitatively, the same conclusions also apply when the estimated effect size at the hit SNP is weaker, for example in the range 1.05 to 1.2 (Figure S3).
One consequence of the potential underestimation of effect sizes from GWAS findings is that as we move to better identification of the actual causal variants, through fine mapping and/or functional studies of associated regions, our estimates of their effect sizes might well increase. Assuming a multiplicative model of risk across loci, these small expected changes could combine to increase the relative risk of disease in those individuals with highest genetic risk of disease.
To investigate this, we simulated genotypes at known associated loci in a population of individuals (assuming Hardy Weinberg equilibrium and no linkage disequilibrium across loci) for each of breast cancer, type 2 diabetes and Crohn's disease, based on reported risk allele frequencies [11], [12], [13] (see Tables S3, S4, S5 for a list of loci). First treating the causal loci and relative risks for each disease as given by current GWAS estimates, we measured the average risk of individuals in the top x%, by risk, of the population (for differing values of x) and compared this to the mean risk in the population. We then repeated this simulation, allowing for the uncertainty in the estimation of true effect sizes by averaging over the uncertainty in both the RAF and effect size of the causal variant on the basis of the posterior distributions of these, given the GWAS findings, under the two priors described above. We assumed that risks combined multiplicatively across loci. For NOD2 and IL23R in Crohn's disease where the causal variant is thought to be known, here and below, we used the effect sizes for the known variant, and did not average over uncertainty in these. Because all three diseases have been extensively studied, we approximated the GWAS discovery process as corresponding to a GWAS discovery sample of 5000 cases and 5000 controls, and a replication sample of 10,000 cases and controls. The actual discovery process for each of the diseases is complicated, often involving meta-analysis and/or multistage discovery, and not straightforward to model accurately, but the approach we use should capture the fact that GWAS-discovery were ascertained through study of large numbers of samples.
The results of the three simulations are given in Table 1.The unadjusted simulations give estimates of how much more at risk individuals with the greatest genetic propensity to disease are, based only on GWAS loci, relative to the average person in the population. As expected, the fold change in risk of individuals carrying a large fraction of risk variants is dependent on the number and magnitude of known loci. For example, individuals in the top 0.1% of risk for Crohn's disease are 20 times more likely than the average person to develop the condition, whereas for breast cancer, where the number of common loci and associated relative risks is typically smaller, the equivalent number is just over two-fold.
The second and third simulations attempt to average over the possible outcomes of our future efforts to map causal mutations, to reveal the likely gains in our ability to stratify individuals on the basis of risk. These use the methodology above, under both prior distributions, to average over the posterior distribution of the allele frequency and effect size at the causal SNPs underlying reported GWAS loci for the three diseases. These adjusted estimates are also shown in Table 1. Across diseases we see that there is a significant increase in the risk associated with carrying multiple risk variants. In particular we see that the biggest differences in risk are for those individuals in the extreme tail. It is these individuals who carry the stronger, likely rarer, risk alleles which are currently insufficiently characterised by the most significant signal of association in some regions identified to be important in disease. For example, the risk of an individual in the top 0.1% of the population for genetic risk typed at the causal loci underlying currently known GWAS loci will likely be increased by a factor of 3–6.5, 5–12, or 25–50, compared to an average individual, for breast cancer, type 2 diabetes and Crohn's disease. These are notably greater increases in risk than current prediction based in the hit SNPs from GWAS loci which would be 2.4, 3.5 and 20 respectively.
We have shown above that as we move to identification of the true causal variants underlying GWAS associations, through fine mapping and functional studies, their effect sizes will tend to increase, in a minority of cases substantially, compared to current estimates from GWAS. This will, in turn, increase the amount of heritability explained by these diseases. We can use the approach developed here to try to quantify this effect.
We investigated this question in the context of the three diseases just described, namely breast cancer, type 2 diabetes, and Crohn's disease. For each disease we took the set of hit SNPs from published associated loci [11], [12], [13] (see Tables S3, S4, S5), and for our two prior distributions on effect sizes we estimated the posterior distribution of both the effect size and the allele frequency for the causal SNP at each locus, as described in the previous section. One commonly used measure of heritability is sibling recurrence risk ratio, often denoted by λS: the relative increase in risk to an individual if their sibling has the disease compared to the baseline risk in the population as a whole [14]. Assuming, as is usual for heritability calculations [15], that there is no interaction between loci, λS can be calculated as a function of the risk allele frequency and effect size for each causal variant. In order to allow for the uncertainty in the allele frequency and likely underestimation of the effect size at the causal variants underlying GWAS associations, we averaged this expression over the posterior distribution of these quantities, given the GWAS findings (see Methods for details).
The results are shown in Figure 4. For each disease they show that the heritability due to already identified GWAS loci will be higher than current estimates, under either set of assumptions about true effect sizes, but particularly under the MAF-dependent prior. Whereas at the time of writing the current estimates of the contribution to λS from GWAS loci are 1.03, 1.08, and 1.49 for breast cancer, type 2 diabetes, and Crohn's disease, these may well be 1.06, 1.14, and 1.61 (mean under the conservative prior) and they could plausibly be as high as 1.21, 1.39 and 2.46 (mean under the MAF-dependent prior). Whilst some of the “missing” heritability is thus disguised rather than missing, we note that this effect is unlikely to account for the extent of the gap between estimates of sibling relative risk (2, 1.8, and 10, respectively, from family studies [16], [17], [18]) and those explained by currently known loci. We return below to a discussion of the discrepancy.
The correlation between alleles along the human genome has allowed GWAS to look for regions associated with disease without having to either genotype all known genetic variation or guess a priori which regions of the genome may be important. Although this approach has been a significant success, there is a predictable downside of using a subset of variation to tag, or predict, untyped diversity: for the vast majority of the SNPs identified as mediating disease risk, we are left uncertain as to whether they are causally involved in the pathway from genotype to phenotype, or, much more plausibly, are just a surrogate for the causal variation.
GWAS associations will thus typically relate to a noisy measurement of the causal variant. One consequence of this is that the size of the genetic effect associated with GWAS loci may be underestimated. We quantified this through an extensive simulation study designed to mimic patterns of linkage disequilibrium in European Caucasian populations. We draw two broad conclusions from these analyses. Firstly, a significant proportion of estimated relative risks will be biased downwards because the hit SNP is a powerful, but imperfect, tag for the true causal variation. In most cases this effect will be relatively minor, but in some instances, the best associated SNP may actually be a poor predictor of a, putatively rarer, SNP with a much larger effect, in which case the effect size estimated from the GWAS finding will substantially underestimate the true effect size.
The exact proportion of reported associations which fall into these two categories depends on properties of the design of the study from which the SNP was identified, and on one's belief about how likely low frequency (>1%) variants of large effect are to cause common diseases. The statistical power afforded by any particular association strategy sets a lower limit on the size of effect that can be under-estimated because an imperfect tag of an allele with a small effect size will simply fail to achieve genome-wide significance. Other properties of GWAS strategy, such as sample ancestry and the number of markers typed, also change our interpretation of observed effect sizes because they influence the distribution of linkage disequilibrium between putative hit SNPs and causal variants.
Our findings show that at any particular locus, especially if the associated SNP has a low MAF, the true effect could be quite large. But we would not expect this to be widespread. Were many true effects this large it would be extremely surprising for so few of them to have been observed: although any one such causal SNP may not be well tagged on the genotyping chips used for GWAS, some of them will happen to be at least moderately well tagged, and their detection would lead to much larger estimates than have been seen from current studies. In the context of this study these early observations suggest that, of the two prior distributions we investigated, it is the conservative prior that may better reflect the true distribution of effect sizes attributed to low and common frequency variants.
One way of viewing the posterior distribution on the true effects shown in Figure 3 is as a probability distribution on the outcome of efforts to fine map current regions of association. In this light, our results inform questions of the design and value of fine mapping experiments. First, simulations similar to those described above (assuming causal variation to be distributed like SNPs in ENCODE regions) suggest that less than 8% of the time will the hit SNP actually be the causal SNP. We note that there may be more reward in terms of gains in predictive ability and increases in effect size from fine mapping SNPs with lower minor allele frequency because they are, on average, more likely to be in poor LD with an unobserved causal variant. On the other hand, our simulations show that although they are unlikely to be causal, most common hit SNPs are likely to be very good surrogates markers for their causal variant. Indeed, in 25% of cases, the hit SNP will be a near-perfect surrogate (ie r2>0.99) for the causal variant. Should this be the case, further genotyping will not reveal other SNPs with stronger associations, unless sample sizes are extremely large.
Here we have quantified the increased spread of genetic risk with genotypes just at known loci, and only considering a multiplicative disease model. But even in this restricted setting, there will be substantial differences in risk between high- and low-risk groups based on these genetic factors. For example the propensity of individuals in the top 0.1% of the population distribution of genetic risk of type 2 diabetes will be increased by a factor of 5–10, compared to the average. For breast cancer, in the analogous top-risk group this risk will be increased by a factor of 3–5 (on the basis of common variation). Importantly, with the growth of GWAS findings, both in terms of numbers of diseases and numbers of loci for particular diseases, more and more of the population will be in this most at risk category for at least one disease: assuming 100 independent diseases, nearly 10% of the population will be in the top 0.1% of risk of at least one disease. Knowing which individuals these are and what diseases they are most at risk of is therefore potentially useful information, both to the individual and at the population level. The issues involved in utilising such information in screening programmes (discussed for example in [13]) are complicated, but our results strengthen the arguments for consideration of this possibility.
We have shown that some of the “missing” heritability for common disease actually resides in known GWAS loci and have estimated this deficit for three particular diseases. While rather more heritability is likely to be explained by known GWAS loci than has been reported, this effect alone falls well short of explaining all the missing heritability. Note, however, that there are other reasons why existing loci may explain more heritability than currently thought. Current calculations (by others, and above) focus on a single causal variant in each associated region: more variants within regions will explain more heritability. They also ignore possible non-multiplicative disease effects, and also ignore interactions between variants at different loci. Power to detect either is low [19], so it is misleading to put much weight on the failure of existing designs to find such effects. As others have noted [20], parts of the missing heritability could be due to multiple rare variants of large effect, associations with other forms of genetic variation such as copy number polymorphisms, and epigenetic effects. Indeed it would be surprising if each did not play some role. Another possibility is that estimates of the “genetic” component of disease susceptibility, from epidemiological studies, confound shared environment with shared DNA, and so inflate heritability estimates [21], [22].
In order to model the signal of association generated by disease-causing mutations, we chose to simulate data exploiting empirical surveys of human diversity. For this purpose we used data from the 10 ENCODE regions [23] within the CEU analysis panel of HapMap II [5], which have undergone SNP ascertainment by resequencing 48 individuals of diverse ancestry. These regions therefore show a fuller spectrum of SNPs than are represented in the HapMap data at large, and haplotypes are expected to be accurate due to the trio design of the CEU HapMap panel [24]. The regions over which we simulate data are centred on each of the 10 ENCODE regions (listed in Table S1) and include 500kb of flanking HapMap variation at the boundaries of each region.
As the typical sample size of most GWAS is much larger than the number of CEU HapMap individuals, we simulated 100,000 chromosomes using the HAPGEN software package. These 100,000 haplotypes we call the reference panel. GWAS case and control samples were then subsampled from the reference panel, as described below. HAPGEN uses a population genetic model that incorporates the processes of mutation and fine-scale recombination to generate individuals from an existing set of known haplotypes. We ran HAPGEN with an effective population size of 11418 (as recommended for the CEU population), a population scaled mutation rate of 1 per SNP, a population scaled recombination rate from estimates described in [25], with the known set of haplotypes taken from the CEU analysis panel of HapMap II as described above (see http://www.stats.ox.ac.uk/~marchini/software/gwas/hapgen.html).
For SNPs greater than 1% in frequency in the ENCODE regions we performed two hypothetical GWAS by letting each of the two alleles be causal in turn. We denote the causal allele by A and the protective allele by a. To generate the control sample we sampled the required number of haplotypes, without replacement, from the reference panel and combined these in pairs to form diploid individuals. This mimics the common use of population controls, rather than controls explicitly chosen for not having the disease under study. For the case sample, we sampled pairs of haplotypes from the reference panel according to the genotype frequencies at the causal SNP dictated by the assumed disease model: If δ is the risk of the AA genotype, and α is the risk of the Aa genotype, both relative to the aa genotype, then we sample case individuals (without replacement) on the basis of their genotypes at the SNP assumed to be causal with success probabilities proportional to:(2)where f is the frequency of the risk allele A in the reference panel. Throughout, for definiteness, we adopted a multiplicative model for disease risk (additive on the log scale) defined by δ = α2. We refer to α as the relative risk (RR) or effect size associated with the causal variant. To approximate a GWAS, we thinned the generated data set to include only those SNPs present on the Affymetrix 500K array that had a minor allele frequency in sampled controls of greater than 1%. This set may or may not include the assumed causal SNP.
For analyses involving only simulated data, we sampled 2,000 cases and 2,000 controls from the reference panel to emulate a typical large GWAS. For the subsequent analyses of heritability and individual risk profiling for type 2 diabetes, breast cancer and Crohn's disease that studied particular reported associations, we simulated 5,000 cases and 5,000 controls to obtain results more comparable to the size of study from which the associations were ascertained. We simulated under a range of relative risks at 24 grid points from 1.05 to 6. In attempting to simulate the signal of disease at rare alleles (1% to 5%) in a GWAS of 5000 cases and controls there were a small number of simulations in which there were insufficient haplotypes in our reference panel to generate the required number of genotypes at the causal SNP for large effect sizes. These simulations were discarded, but as the numbers were small (3% when the RR = 4 and 11% when RR = 6) we do not believe this greatly affects the results presented below.
Following common practice, for each simulated case control sample, we tested for association between genotype and case control status using the Cochran Armitage trend test [26] at each SNP with frequency greater than 1% in the simulated panel of chromosomes. We calculated the p-value of this test statistic which is distributed with 1 degree of freedom under the null hypothesis of no association. If any test across the region obtained a p-value<10−6 the location of the most significant SNP (termed the hit SNP) was recorded and we simulated this SNP in an independent replication sample.
We simulated the replication experiment in three stages. First we simulated the frequency of the causal allele in cases and controls in the replication population. We then simulated the frequency of the hit SNP conditional on the frequency of the causal allele. Finally, we simulated the genotype counts for a sample of cases and controls in this replication population.
We motivated sampling of the frequency of the causal allele in controls in the replication population by thinking of the replication sample as an additional sample from the same population as the original GWAS sample. (Other assumptions are possible here, but seem unlikely to affect the main conclusions.) Specifically, we placed a uniform prior distribution on the unobserved population frequency and sampled a value, f ′, from the posterior distribution of this frequency given the data in the reference panel. (Given the large size of the reference panel, the frequency in the replication sample will be very close to that in the reference panel.) Conditional on f ′, the population replication frequency in cases was calculated from equation (2). To obtain the replication population frequencies at the hit SNP we estimated the conditional distribution in the reference sample of alleles at the hit SNP in each of cases and controls, given those at the causal SNP, and used these for the replication sample. This corresponds to assuming that the LD between the causal and hit SNP in the replication sample will be the same as that in the reference sample. Finally, conditional on the population replication frequencies in cases and controls, we take multinomial samples of the required size to mimic the replication case and control samples. A test of association using the trend test was performed at the hit SNP on the simulated replication samples and deemed a significant replication if the p-value was less than 10−2.
We estimate the effect size, or relative risk, α, at the hit SNP by maximum likelihood under the model described above by equation (2). For studies with population controls this can be achieved in practice by fitting a logistic regression model for case status [27].
We implement two different sets of prior assumption on the effect size and its relationship with minor allele frequency. Our first set of assumptions is that if α is the effect size at a causal variant, then log(α) is normally distributed with mean 0 and standard deviation 0.2, independent of RAF. We refer to this as the conservative prior, since it places little weight on relative risks greater than 1.5. To get a sense for this distribution, it assumes that 81% of true effect sizes are less than 1.3 with 96% less than 1.5, and 99.9% less than 2. A further discussion of the choice of prior on effect sizes can be found in [19] and [28].
Our second set of assumptions, which we call the MAF-dependent prior, again assumes a normal distribution for log(α) with mean 0, but here the standard deviation, σ, is allowed to depend on the RAF. The dependence of the distribution of the effect size on allele frequency has no theoretical justification, but is chosen on pragmatic grounds to give a gradual increase in the average effect size as the alleles at causal SNP become rarer in the population. It is implemented by increasing σ by a weight defined by an exponential density with parameters chosen such that, when the RAF is near 0.5 (a common SNP), this prior is approximately the same as the conservative prior, with σ = 0.2. As the RAF approaches 0 or 1 (corresponding to rarer SNPs), then considerably more weight is put on larger RRs. See Figure S4 and Table S2 for details. For example, when the MAF is less than 5% the second prior gives an approximately 45% chance that the risk associated with each copy of the causal allele is larger than 2.5. Note that we used an empirical prior on the frequency of the risk allele (Figure S5) by choosing each allele, at each SNP, with in the ENCODE region to be causal in turn.
A commonly used measure of heritability is based on considering the risk of disease to an individual conditional on them having an affected () sibling relative to the unconditional probability (which is just the prevalence of the disease):We can calculate the above, using assuming that , and by summing over the genotypes of the siblings and of the mother and father (see [15]):If we divide through by the square of the risk associated with most protective genotype (which we can define to be ) then we can write the above in terms of the per allele relative risk , and assume the genotype probabilities follow Hardy-Weinberg equilibrium with risk allele frequency as above:By making the further assumption that loci are independent an estimate of the heritability explained by a set of hit SNPs can be obtained by multiplying together the λS values calculated at each individual locus. We calculated sibling relative risk in this manner using estimates of RR and RAF of replicated loci from studies of Type 2 diabetes, Crohn's disease and breast cancer (see Tables S3, S4, S5).
We then simulated 100,000 times from the posterior of true RR and RAF of each locus conditional upon the reported RR and RAF, using the simulation approach and the two different priors as described in the paper. For each set of simulations, for each disease, we recalculated λS at each locus and multiplied over loci, giving a sample from the posterior distribution of sibling risk that could be explained by the current set of report loci if the causal loci where typed directly.
|
10.1371/journal.pbio.1001640 | Molecular Composition and Ultrastructure of the Caveolar Coat Complex | Caveolae are an abundant feature of the plasma membrane of many mammalian cell types, and have key roles in mechano-transduction, metabolic regulation, and vascular permeability. Caveolin and cavin proteins, as well as EHD2 and pacsin 2, are all present in caveolae. How these proteins assemble to form a protein interaction network for caveolar morphogenesis is not known. Using in vivo crosslinking, velocity gradient centrifugation, immuno-isolation, and tandem mass spectrometry, we determine that cavins and caveolins assemble into a homogenous 80S complex, which we term the caveolar coat complex. There are no further abundant components within this complex, and the complex excludes EHD2 and pacsin 2. Cavin 1 forms trimers and interacts with caveolin 1 with a molar ratio of about 1∶4. Cavins 2 and 3 compete for binding sites within the overall coat complex, and form distinct subcomplexes with cavin 1. The core interactions between caveolin 1 and cavin 1 are independent of cavin 2, cavin 3, and EHD2 expression, and the cavins themselves can still interact in the absence of caveolin 1. Using immuno-electron microscopy as well as a recently developed protein tag for electron microscopy (MiniSOG), we demonstrate that caveolar coat complexes form a distinct coat all around the caveolar bulb. In contrast, and consistent with our biochemical data, EHD2 defines a different domain at the caveolar neck. 3D electron tomograms of the caveolar coat, labeled using cavin-MiniSOG, show that the caveolar coat is composed of repeating units of a unitary caveolar coat complex.
| Caveolae are flask-shaped invaginations in the plasma membrane of many mammalian cell types, and are particularly abundant in fat cells, muscle cells, and the cells that line blood vessels. Although caveolae are likely to be important for cellular responses to mechanical stress, intracellular trafficking, and signaling events, we still lack an understanding of the precise molecular mechanisms for how they form and carry out these functions. Here we address the question of how caveolae are made. Recent years have seen a considerable expansion of the catalogue of known protein components present in caveolae. Our study shows that the main protein components, cavins and caveolins, assemble into one specific complex. We reveal how different amounts of two caveolar proteins, cavin 2 and cavin 3, may be incorporated into this single type of complex, thereby potentially conferring different functional properties on caveolae. Using electron microscopy, we demonstrate that the protein complex is distributed all around the membrane bulb of caveolae, and so can be truly described as the caveolar coat. The caveolar coat excludes the protein EHD2, which regulates the dynamics of caveolae—this protein has a distinct distribution at the caveolar neck. These findings provide the basis for a more complete understanding of the network of protein interactions that produces caveolae.
| Caveolae, plasma membrane invaginations with a diameter of about 50–80 nm and a characteristic flask-like shape, were first identified 60 y ago by electron microscopy [1],[2]. They are found in many different cell types, and are particularly abundant in endothelial cells, adipocytes, and muscle cells [3]–[5]. Caveolin 1 is the major integral membrane protein in caveolae, and is essential for their formation in nonmuscle cells [6]–[8]. There are two further caveolin proteins. Caveolin 2 has the same distribution as caveolin 1, and hetero-oligomerises with this protein, but is not essential for forming caveolae [9],[10]. Caveolin 3 is only expressed in striated muscle and is required for making caveolae in this tissue [11]. Mice lacking caveolin 1 have multiple phenotypes including hyperglycaemia, lipidosis, and changes in endothelial permeability [8],[12]–, and humans with a loss of function mutation in caveolin 1 are severely lipodystrophic [16]. The molecular and cellular causes of these phenotypes are not completely understood, but caveolae have been proposed to act in a variety of ways, including as endocytic vesicles, as mechanoprotective or mechanosensing membrane reservoirs, as regulators of lipid transport, and as scaffolds for signaling events [3]–[5],[17],[18].
Caveolins were long believed to be the sole protein component of caveolae, and they clearly have a central role in biogenesis of these structures. Direct evidence for this is provided by experiments showing that expression of caveolin in bacteria is sufficient to generate caveolae-like membrane vesicles [19]. Recently, however, the list of caveolar components has been considerably expanded, with the identification of cavin proteins [5],[20]–[26], EHD2 [27]–[29], and pacsin 2 [29],[30]. This implies that caveolar biogenesis and function involves a complex set of proteins, but how these proteins assemble physically and spatially to generate caveolae has yet to be fully elucidated.
The cavins (cavin 1, 2, 3, and 4) localise to caveolae, and are important for their formation and dynamics [5],[20]–[26]. It should be noted that the cavin nomenclature is not the same as the standard gene names for this family (cavin 1, PTRF; cavin 2, SDPR; cavin 3, PRKCDBP; cavin 4, MURC; Cavin 3 is also frequently referred to as SRBC [5]). Cavin 1 is expressed in all cell types that express caveolins, and is essential for making caveolae in vivo [24],[31]. Phenotypes of cavin 1 knockout mice resemble those of caveolin 1 caveolin 3 double knockout mice, implying a central role for cavin 1 in caveolar biogenesis [14],[31],[32]. In contrast to cavin 1, expression of cavins 2, 3, and 4 is more cell and tissue-specific, with cavin 4 only being expressed in striated muscle [22],[33]. Intriguingly, cavin 2 is required for morphogenesis of caveolae in the endothelia of some tissues but not others, and cavin 3 appears to be dispensable for forming caveolae [34]. Cavins are present in large complexes that can be detected on sucrose gradients, and the apparent size of these complexes differs between tissues [34],[35]. Cavins can be co-immunoprecipitated with each other, and cavins 1 and 2 interact directly [21]. Overexpression of cavin 2 distorts and elongates caveolae, while cavin 3 depletion reduces intracellular transport of caveolar vesicles [20],[21]. These data suggest that in vivo caveolae may contain different complements of cavin proteins, and that cavins 2 and 3 may regulate caveolar function and dynamics in a cell-type-specific manner. How different complements of cavins can be incorporated into morphologically uniform caveolae remains unclear.
Caveolae contain an estimated 140–180 caveolin molecules [36] and oligomerisation of caveolins is likely to be critical for caveolae formation. Oligomerisation is cholesterol-dependent, and occurs initially in the trans-Golgi network, resulting in 8S complexes with an estimated 14–16 caveolin molecules [10],[35],[37]. Upon vesicular transport to the plasma membrane, such caveolin oligomers somehow assemble into higher order oligomeric complexes, which are likely to constitute a key structural unit of the caveolar coat [38]. Cavin proteins first co-localise with caveolins after delivery to the plasma membrane, but the nature of this association is not clear, as cavins and caveolins do not co-fractionate on sucrose gradients of detergent-solubilised cell lysates [35]. Moreover, whether the additional caveolar proteins EHD2 and pacsin 2, both of which are likely to regulate caveolar function or dynamics in some way [27]–[30], associate with caveolins and/or cavins directly or with other determinants within caveolar membranes is unknown.
Electron microscopy (EM) techniques have been used to try and ascertain whether there is a protein coat that surrounds caveolae, and to determine its organisation. Platinum and chromium coating of plasma membrane fragments suggests the presence of spiral ridges or striations on the bulb that can be detected by scanning EM [6],[39]. A similar distribution of densities is seen after high-pressure freezing and freeze-substitution [40], and in conventionally stained ultrathin sections periodic local maxima in electron density are observed around the caveolar membrane [41]. However, the organisation of such protein densities as well as the actual shape of caveolae is contingent on the fixation method used [42], and so the relationship between the striations observed by scanning EM and the protein densities seen by transmission EM is not clear. It has been suggested that the caveolar neck forms a separate domain distinct from the caveolar bulb, but neither the identity of the protein components around the bulb nor those around the neck are fully defined [41],[43]. Although some studies report that caveolins are found all around the caveolar bulb [44],[45], others report a more restricted distribution to the sides or neck of caveolae [46]. Inherent limitations of immuno-labeling, including the possibility that epitopes may not be equally accessible all over the caveolar surface and the reduced spatial resolution provided by combining primary and secondary antibodies, mean that it has been hard to address this issue unequivocally [43],[47]. Finally, the subcaveolar distribution of more recently identified components of caveolae, such as the cavins and EHD2, has yet to be addressed.
In the work reported here we have addressed fundamental questions central to an understanding of the protein machinery responsible for generating caveolae. We determine the identity and biochemical properties of the complexes into which caveolins and cavins assemble. We find that cavins and caveolins, but not EHD2 and pacsin 2, are found in a specific 80S complex, which we term the caveolar coat complex. Both immuno-EM and EM labeling with MiniSOG fusion proteins [48] show that this unitary complex does indeed localize all around the caveolar bulb. EHD2 defines a spatially and biochemically distinct domain at the neck of caveolae, and is not required for formation of the coat complex. These data provide conceptual advances in our understanding of how caveolae are generated.
Caveolin 1 has been reported to exist as a labile high molecular weight complex of about 70S, as determined by velocity gradient centrifugation [35],[43]. Such oligomeric complexes of caveolin 1 are readily lost upon extraction of cells with detergents known to fully solubilize membranes rich in cholesterol and sphingolipids, and the apparent size of caveolin 1 complexes is highly sensitive to the nature of the detergent used for solubilisation (Figure S1) [35]. We looked for ways to stabilise complexes containing caveolin prior to cell lysis. These experiments were carried out in a clonal HeLa cell line stably expressing caveolin-1-GFP at about 20% of the level of endogenous caveolin 1 (Figure S2A).
We found that cross-linking of live HeLa cells with the membrane permeable and reversible cross-linker DSP (dithiobis(succinimidylpropionate)) efficiently and reproducibly stabilised a high molecular weight complex containing caveolin 1 (Figure 1A). Upon cross-linking, caveolin 1 was found almost exclusively in a single sharp peak in fractions 8–10 of 10–40% sucrose velocity gradients (Figure 1A and Figure S1). This was the case even when cells were lysed in 2% w/v (70 mM) octyl-glucoside (OG), a condition where without cross-linker most caveolin 1 is found in the top four fractions of the gradient (Figure S1) [35].
In the presence of cross-linker, cavin 1 co-fractionated precisely with caveolin 1 (Figure 1A), whereas without cross-linking cavin 1 was found in a broad peak in the centre of the gradient (centred on fractions 6/7, about 60S; [35]) and did not co-fractionate with caveolin 1 (Figure 1A). Using the profile of cellular 80S ribosomes and purified 60S ribosomal subunits as a reference, we estimated that the cross-linked high molecular weight caveolin and cavin complexes have a sedimentation rate of about 80S (Figure S2B). 80S caveolin complexes were also detected in control cell lines expressing flotillin-2-GFP or GFP alone, and so were not dependent on the presence of caveolin-1-GFP (Figure 1A). Caveolin-1-GFP had the same distribution as endogenous caveolin 1 in the gradient (Figure 1A). Cross-linking did not change the distribution of caveolin 1 or cavin 1 when studied by immunofluorescence, and did not alter the appearance or brightness of either protein as they co-localised in puncta that are likely to correspond to individual caveolae (Figure S3A and S3B). Therefore, cross-linking with DSP does not itself induce redistribution of cavins or caveolin 1 within cells before lysis. These results suggest that cross-linking of live cells with DSP stabilises caveolin/cavin interactions that are otherwise lost during solubilisation with detergents, and thereby allows identification of a large 80S complex containing cavins and caveolins.
To determine the protein composition of the 80S complex, and to confirm that co-fractionation of caveolin 1 and cavin 1 after cross-linking indeed reflects co-assembly of both proteins into the same complex, caveolin-1-GFP was immuno-isolated from pooled fractions 8–10 (HMW, high molecular weight fractions) using magnetic anti-GFP beads (Figure 1B). Immuno-isolation from the same fractions of gradients of cell lysates from flotillin-2-GFP or GFP expressing cells served as controls. The complexes were washed extensively and eluted with a pH shift. Eluates were reduced with DTT to disassemble DSP cross-links, and analysed by SDS-PAGE and silver staining. This revealed the presence of four major bands specific to the caveolin-1-GFP immunoprecipitate (Figure 1B). Western blotting (Figure S1C), and excision of the relevant bands for analysis by mass spectrometry, both confirmed that these correspond to caveolin-1-GFP, cavin 1, cavin 3, and caveolins 1 and 2. Tandem mass spectrometry of the immuno-precipitated 80S complex identified all of the above proteins, and these were the only abundant proteins detected. Cavin 2 was also detected in the complex, though at significantly lower levels (Figure 1C, File S1). Western blotting of the isolated complex confirmed that all caveolar proteins co-purified specifically with caveolin-1-GFP, and not with affinity-purified flotillin-2-GFP complexes or mock purifications from GFP control cells (Figure S2C).
Cavin 1 yielded the most tryptic peptides identified by mass spectrometry analysis of isolated 80S complexes, and was the strongest band on silver-stained gels (Figure 1B and 1C, File S1). In addition, immuno-isolation of caveolin-1-GFP from the HMW fractions caused cavin 1 to be efficiently depleted from these fractions, showing that the large majority of cellular cavin 1 is present in complexes with caveolin 1 (Figure 1D). Together these data show that caveolins 1 and 2 and cavins 1, 2, and 3 assemble into an 80S complex, and that cavin 1 is a major component of this complex. Moreover, we can state that there are no further abundant protein components in the isolated complex. We hereafter refer to this complex as the caveolar coat complex.
To study the role of cavins 1, 2, and 3 in the assembly of the caveolar coat complex, we generated separate HeLa cell lines stably expressing each protein with a C-terminal TEV-GFP-10×His tag (from now on referred to as cavin-1, -2, or -3-GFP). All cavin fusion proteins localised to caveolae by light microscopy (Figure S4A, Table S1), and cavins 1, 2, and 3 co-localised extensively with each other (Figure S4B). Fusion proteins were expressed at low levels, with cavin-1-GFP being expressed at about 20% of endogenous cavin 1 (Figure S4C). Endogenous cavin 2 is difficult to detect in HeLa cells with available antibodies (although it is present, albeit at low levels, as it is detected by mass spectrometry, Figure 1E and File S1), so cavin-2-GFP is likely to be present at significantly higher levels than endogenous cavin 2 in the cavin-2-GFP cell line. The expression of endogenous cavin 3 was specifically down-regulated in several independent cavin-3-GFP-expressing clonal cell lines, resulting in cell lines expressing cavin-3-GFP instead of cavin 3. In the cell line used for the experiments presented here, expression of cavin-3-GFP was similar to that of cavin 3 in control cells (Figure S4C).
We analysed the distribution of the three cavin-GFP constructs in gradients from cells that had been cross-linked with DSP prior to cell lysis. Upon cross-linking, all cavin-GFP fusion proteins, endogenous cavin 1 and cavin 3, as well as caveolin 1 co-fractionated in the 80S fractions 8–10 (Figure 2A and 2B). Cavin-2-GFP exhibited an additional minor peak in fraction 3, and its expression led to the dissociation of small amounts of cavin 1 from the caveolar coat complex into fractions 5–7. In contrast, cavin-1-GFP and cavin-3-GFP were exclusively found in the caveolar coat complex. These data show that the cavin-GFP fusion proteins are incorporated into the caveolar coat complex just like endogenous cavins. To check that the coat complex does not reflect association or cross-linking of the cavin proteins after lysis, HeLa cell lines stably transfected with either cavin-3-GFP or cavin-3-mCherry were grown in the same dish, cross-linked, and lysed. Subsequent immunoisolation with anti-GFP antibodies yielded complexes devoid of cavin 3-mCherry (Figure S3C), arguing that cavin complexes do not form after cell lysis.
We asked whether the composition and stoichiometry of the complex is the same whichever cavin is used to isolate it. Western blotting of the complex, immuno-isolated from pooled fractions 8–10 (HMW fractions) of the gradients separately, from each cavin-GFP cell line showed that the isolated complexes are indeed indistinguishable with respect to the relative amounts of cavin 1 and caveolin 1 present in each complex (Figure 2C). To further demonstrate that the additional caveolar proteins pacsin 2 and EHD2 [27]–[30] are excluded from the coat complex, we carried out Western blotting of the isolated complex from each cavin-GFP cell line. As predicted by the mass spectrometry data (File S1), pacsin 2 and EHD2 could not be detected in isolated caveolar coat complexes (Figure 2C).
Tandem mass spectrometry was used to further characterise the caveolar coat complex isolated from each cavin-GFP and the caveolin-1-GFP cell line. All of the core caveolin and cavin proteins were identified, and in line with our Western blotting data, peptides corresponding to cavin 1, cavin 3, and caveolin 1 were found in approximately equal numbers in all four immuno-isolates (Figure 2D). Cavin 1 peptides were slightly more abundant in the cavin-1-GFP cell line, and cavin 2 peptides were notably more abundant in the cavin-2-GFP cell line—consistent with overexpression of this latter fusion protein. These mass spectrometry data, coupled with the Western blot analysis in Figure 2A and 2C, show that the composition of the caveolar coat complex is constant whichever component is used for immuno-isolation, which in turn implies that the caveolar coat complex represents one specific species of macromolecular assembly.
We sought to determine the stoichiometry of the components of the caveolar coat complex. To this end, the complex was directly isolated from lysates of cross-linked cells, again using immuno-precipitation of either cavin-1-GFP, cavin-1-GFP, or cavin-3-GFP from the relevant cell lines. Isolated complexes were separated by SDS-PAGE, and stained with the quantitative protein dye Sypro Ruby. Cavin-1-GFP, cavin-2-GFP, cavin-3-GFP, cavin 1, cavin 3, and caveolin 1 (and caveolin 2, which is not well resolved from caveolin 1) were clearly visible in such gels, with little background from contaminating proteins (Figure 3A). We used densitometric gel scans of each immuno-precipitate from at least six separate gels and experiments to measure the relative amount of each component present. Molecular weights calculated from amino acid sequence were used to derive an estimate of the relative molar ratios between exogenously expressed cavin-GFPs, endogenous cavin 1 and cavin 3, and endogenous caveolin for all immuno-isolates (Figure 3B–E). Firstly, we found that both cavin-2-GFP and cavin-3-GFP are present in the complex at a molar ratio of slightly less than 1∶3 with cavin 1 (Figure 3B), so cavin 1 is clearly the most abundant of the three cavins. Secondly, we calculated a molar ratio of total cavin 1 to total caveolin of 1∶4 (Figure 3C)—that is, one cavin 1 may bind to four caveolin molecules (the analysis does not discriminate between caveolin 1 and caveolin 2). This ratio was constant whichever cavin was used for immuno-isolation of the complex. Thirdly, the ratio of the total amount of cavin (i.e., cavin 1+cavin 2+cavin 3) to caveolin was also constant whichever cavin was used for immuno-isolation (Figure 3D). This suggests that there are a fixed number of binding sites in the complex.
If the ratio between total cavin and caveolin in the caveolar coat complex is fixed, then one would predict that overexpression of one cavin may reduce the abundance of another cavin within the complex. Indeed, the amount of cavin 3 present when the complex was immuno-isolated from cells overexpressing cavin-2-GFP was significantly reduced compared to that observed when cavin-1-GFP or cavin-3-GFP was used for immuno-isolation (Figure 3E). This implies that cavin 2 and cavin 3 may compete for binding sites within the coat complex. In order to test this, we examined the effects of overexpressing cavin-2-mCherry and cavin-3-mCherry at high levels, using immunofluorescence to assay whether overexpression perturbs the distribution of other cavins. Overexpressing cavin-2-mCherry caused a loss of cavin 3 from caveolar puncta without perturbing the distribution of cavin 1 (Figure 3F and 3G), and overexpressing cavin-3-mCherry caused loss of cavin 2, again without altering the distribution of cavin 1 (Figure 3H and 3I). Therefore, cavin 2 can displace cavin 3 from the caveolar coat complex, and vice versa. Variation in the relative amounts of cavin 2 and cavin 3 occurs between tissues in vivo [22],[34], so this competition is likely to have physiological relevance.
The combined data imply that the single species of caveolar coat complex is composed of a defined number of cavins and caveolins. The core interaction between cavin 1 and caveolin occurs with a stoichiometry of around 1 cavin 1∶4 caveolin molecules. Changes in relative abundance imply that cavin 2 and cavin 3 compete for a defined number of binding sites within this single type of large 80S complex.
Given the above, we reasoned that the 80S caveolar coat complex might be constructed from specific cavin and caveolin subcomplexes. Partially disassembled coat complexes could yield additional information on the nature of such subcomplexes. To pursue this possibility, we quantified the distribution of each cavin fusion protein in sucrose velocity gradients from the appropriate cell lines after lysis with 1% Triton X-100 without prior cross-linking. We observed a bimodal distribution for caveolin 1 in all cell lines, with a minor peak in fraction 3 and a major peak in fraction 7 (Figure 4A and 4B). We suggest that this latter caveolin 1 peak is the 70S species identified previously [35]. Cavin-1-GFP co-fractionated with endogenous cavin 1, as expected, and formed complexes of about 60S [35]. Cavin-1-GFP expression had no effect on the distribution of endogenous cavin 1 and caveolin 1, as compared to control cells expressing GFP alone. Interestingly, however, cavin-2-GFP peaked in fraction 3, whilst cavin-3-GFP peaked in fractions 6 and 7. Moreover, expression of cavin-2-GFP resulted in a shift of endogenous cavin 1 towards low molecular weight fractions, while cavin-3-GFP expression caused cavin 1 to shift towards high molecular weight fractions. This implies that cavin 2 and cavin 3 form distinct subcomplexes with cavin 1, with the former being smaller or less stable in detergent than the latter. Affinity purification of cavin-GFP complexes from gradient fractions 3–5 or 6–8 confirmed this idea (Figure 4C). Cavin-2-GFP was much more abundant in the low molecular weight pool 3–5, while the amounts of cavin-1-GFP and cavin-3-GFP isolated from the two pools were approximately equal. In addition, while all cavin-GFP molecules co-immunoprecipitated endogenous cavin 1, interactions with caveolin 1 were only observed in fractions 6–8, and much more caveolin 1 associated with cavin-1-GFP and cavin-3-GFP than with cavin-2-GFP. We conclude that cavin 2 and cavin 3 form separate subcomplexes with cavin 1 that are distinct in terms of size and/or stability, as well as their affinity for caveolin 1.
If cavin 2 and 3 do indeed form separate subcomplexes with cavin 1, then one would predict that cavin 1 may form complexes with cavin 2 or 3 even in the absence of caveolin 1, and that cavin 2 and 3 should not co-precipitate unless cavin 1 is present. We carried out experiments to test these predictions. There is a marked reduction in the expression of cavin 2 and cavin 3 in cavin 1 or caveolin 1 gene knockout mice and cell lines [23],[24],[31],[34], so we transiently transfected plasmids for overexpressing cavins 1, 2, or 3 as mCherry or GFP fusion proteins into mouse embryonic fibroblasts (MEFs) from cavin 1 and caveolin 1 knockout mice [7],[31],[34]. Western blotting of cell lysates from the transfected cells showed that the fusion proteins of all three cavins could be detected, although expression of cavin-3-mCherry was very low unless cavin-1-GFP was also present (Figure S5A). In caveolin 1 knockout MEFs, immuno-isolation of cavin-1-GFP co-precipitated cavin-2-mCherry and cavin-3-mCherry (Figure S5A). This is consistent with previous studies showing that cavins form high molecular weight complexes in the absence of caveolin 1, and that cavins 1 and 2 bind to each other directly in vitro [21],[22],[34]. In order to ascertain whether cavins 2 and 3 can interact without cavin 1, we compared co-precipiation of cavin-2-GFP and cavin-3-mCherry in control and cavin 1 knockout MEFs (Figure S5B). In control MEFs co-precipitation was detected, but this was lost when cavin 1 was absent, arguing that cavin 2 and cavin 3 do not interact directly, and so do indeed form separate subcomplexes with cavin 1.
To further identify protein–protein interactions within the caveolar coat complex, we used Western blotting to look for interactions between the isolated components after immunoisolation of cavin-1-GFP, cavin-2-GFP, or cavin-3-GFP followed by titration of DTT to partially dissociate cross-links. Complexes purified by isolation of cavin-2-GFP are shown in Figure 4D, and immunoisolation of all three cavins is shown in Figure S6. We found that the disassembly of cross-linked complexes with titration of DTT is precisely the same whichever cavin is used for immuno-isolation, providing additional confirmation that there is one single type of caveolar coat complex. Under nonreducing conditions (without DTT), around 50% of the total caveolin 1 in the caveolar coat complex was found in oligomers of about 350 kDa and more. The remainder of the caveolin 1 was present mostly as either monomers or dimers (Figure 4D). Upon titration of DTT, high molecular weight oligomeric forms of caveolin 1 were reduced to a distinct caveolin oligomer of about 350–400 kDa, which was stable in up to 10 mM DTT and even clearly identifiable under fully reducing conditions (not shown). More minor cross-linked species consistent with the presence of caveolin 1 oligomers increasing in size from 2 to 8 caveolin 1 monomers were not stable in DTT. This suggests that the 350–400 kDa oligomeric form of caveolin 1 is a major component of the caveolar coat complex [49],[50]. We then analyzed cavin 1 (Figure 4D and Figure S6). In nonreduced samples, the large majority of cavin 1 was cross-linked into oligomers of about 180 kDa and more. Monomeric cavin 1 (55 kDa) and a minor cavin 1 species of about 85 kDa were also observed. Interestingly, progressive addition of DTT revealed a relatively stable oligomeric form of cavin 1 of about 180 kDa, a size indicative of a cavin 1 trimer. This form of cavin 1 was found in all immunoisolates and was stable in up to 10 mM DTT (Figure 4D and Figure S6).
Altogether, combining the data on disassembly of the non-cross-linked caveolar coat complex in detergent, co-precipitation in the absence of cavin 1, and partial reduction of cross-links yields specific conclusions: Subcomplexes containing cavin 1 and cavin 2 can be separated from subcomplexes containing cavin 1 and cavin 3, and cavins 2 and 3 do not enter the same complex unless cavin 1 is also present. Partial reduction of cross-links shows that cavin 1 forms a relatively stable trimer, and this trimer is likely to be a core element of the caveolar coat complex.
So as to characterise the role of cavins 2 and 3 in the caveolar coat complex in more detail, we used siRNAs to deplete these proteins from the cavin-1-GFP cell line. In parallel, we also used siRNAs targetting EHD2, to ask whether this protein controls the assembly of the complex, and siRNAs targetting flotillin proteins as a negative control [51]. Depletion of all targeted proteins was highly efficient, as judged by Western blotting (Figure 5A and Figure S7A). Velocity gradient centrifugation showed that 80S complexes were clearly still present in all siRNA-treated cells, and the large majority of cavin 1, cavin 3, and caveolin 1 were still found in the HMW, 80S, fractions 8–10 (Figure 5B and Figure S7B). Lack of cavin 3 caused a marginal destabilisation of the 80S complex, as in this case some cavin 1 and caveolin 1 were detected in lower density fractions in longer exposures of the relevant Western blots (LMW pool, Figure 5B and Figure S7B). In these longer exposures the cavin 1 trimer described above is detected, even though these samples were run under fully reducing conditions. Most importantly, however, quantification of the ratio between the amounts of total cavin 1 and caveolin 1 present in the high molecular weight fractions (i.e., in the caveolar coat complex) confirmed that this was unchanged by any of the siRNA treatments (Figure 5C), which shows that the stoichiometry of a core interaction between cavin 1 and caveolin 1 is independent of the presence of cavins 2 or 3.
Likewise, depletion of EHD2 had no detectable effect on the formation of the caveolar coat complex, having no effect on the behaviour of the complex in velocity gradients or on the relative amounts of cavin and caveolin proteins present (Figure 5B and 5C, Figure S7). EHD2, therefore, not only is not present in the complex, but also does not regulate its formation. Given that EHD2 controls the dynamics and plasma membrane association of caveolae [27],[28], this implies that the caveolar coat complex is the same whether caveolae are continuous with the plasma membrane or form intracellular membrane vesicles.
If the caveolar coat complex does indeed coat caveolae, then it should be found all around the caveolar bulb. We aimed to determine the localisation of the complex within caveolae at high spatial resolution. Firstly, the cell lines expressing caveolin-1-GFP, cavin-1-GFP, cavin-1-GFP, and cavin-3-GFP described above were studied by immuno-electron microscopy. Pre-embedding labeling using anti-GFP antibodies and nanogold-conjugated secondary antibodies, followed by silver enhancement, allowed highly specific labeling of caveolae (Figure 6A). We acquired images of more than 50 caveolae stained with gold particles per cell line, and superimposed both the membrane profiles and the position of gold particles for each case (Figure 6B and Figure S8). The aggregated images revealed that all of the cavin-GFP fusions, as well as caveolin-1-GFP, localised around the caveolar bulb, with no discernable bias towards the membrane proximal or distal region. In order to check that endogenous proteins have the same distribution as GFP fusions, we carried out immuno-labeling of untransfected cells with caveolin 1 and cavin 1 antibodies. The distribution of endogenous caveolin 1 and cavin 1 determined using this approach was indistinguishable from that of caveolin-1-GFP and cavin-1-GFP (Figure 6C). These data argue that caveolar coat complexes are distributed all around the membrane bulb of caveolae.
Our biochemical data show that EHD2 is not present in the caveolar coat complex. To determine the distribution of EHD2 within caveolae, we used cells expressing GFP-EHD2 and immuno-labeling as above (Figure 6D). In clear contrast to the caveolar coat complexes, GFP-EHD2 was enriched around the neck of caveolae (Figure 6C). Therefore, caveolae are likely to have separate sets of proteins coating the bulb and neck regions, and these different distributions can be resolved by immuno-electron microscopy.
Immuno-labeling has inherent limitations, including the possibility that epitopes may not be uniformly accessible or may only tolerate weak fixation, and the fact that complexes of primary and secondary antibodies reduce spatial resolution compared to the fine structural details otherwise delivered by electron microscopy. We aimed to directly visualise caveolar coat complexes by electron microscopy in situ. MiniSOG (for Mini Singlet Oxygen Generator) is a relatively small (106 amino acids) fluorescent flavoprotein that efficiently generates reactive oxygen species upon illumination with blue light. Local production of reactive oxygen species can be used to convert diaminobenzidine (DAB) into an osmiophilic electron-dense polymer. This allows proteins to be localised by EM at high spatial resolution [48],[52]. We generated separate cell lines expressing cavins 1, 2, and 3 as MiniSOG-mCherry fusion proteins (henceforth referred to as cavin-MiniSOG). In order to facilitate analysis of multiple caveolae, we used retinal pigment epithelial (RPE) cells, where caveolae polarise to the rear of the cell, as seen in other polarized cells such as fibroblasts (Figure S9B) [41],[53]. This yields defined regions of the cell that are very rich in caveolae. Cavin fusion proteins were expressed at low levels relative to the endogenous proteins, and localised to caveolae in a manner indistinguishable from the endogenous proteins by light microscopy (Figure S9A, Table S1).
Photooxidation of glutaraldehyde-fixed cavin-MiniSOG expressing RPE cells in the presence of DAB resulted in the deposition of a brown reaction product (Figure 7A). Correlative electron micrographs of plastic embedded and osmium-stained sections revealed a high density of caveolae in such regions (Figure 7B). Caveolae were strongly labeled with an electron-dense stain, regardless of which cavin-MiniSOG was used for photooxidisation (Figure 7B and 7C). The staining was highly specific, as caveolar membranes from adjacent cells not expressing cavin-MiniSOG were not stained, the stain was restricted to regions of the cell enriched in caveolae, and the only cellular membranes stained were caveolae (Figure 7 and Figure S10). MiniSOG-generated stain from all three cavin fusion proteins was distributed right around the caveolar bulb, and was present at the same density at the lateral sides and at the apex of the bulb (Figure 7D). This is consistent with the immuno-electron microscopy presented in Figure 6, and again implies that caveolar coat complexes are present all around the caveolar bulb.
We acquired high-resolution micrographs to study the ultrastructural properties of the caveolar coat complex labeled using cavin-MiniSOG. In thin sections, the label was clearly not continuously distributed along the caveolar membrane, but rather formed a punctate, sometimes spike-like coat. This was observed for all cavin-MiniSOG proteins (Figure 7E). In some caveolae, individual puncta formed periodic densities with approximately regular spacing (Figure 7E and 7F). The spacing of periodic densities around the bulb was not measurably different whether cavin-1-MiniSOG, cavin-2-MiniSOG, or cavin-3-MiniSOG were used (Figure 7F). Quantification of the spacing of these local increases in density revealed a periodicity of 10–16 nm. The shortest distance we were able to measure in electron micrographs of thin sections was around 8 nm (Figure 7G). This shows that caveolar coat complexes form local densities on caveolar membranes with an apparent spacing of 10–16 nm and that MiniSOG labeling allows proteins to be localized with low nanometer precision. The fact that we could observe regular spacing between densities is suggestive of regularity in the coat.
In order to reveal the organisation of the coat in three dimensions, dual-tilt tomograms were recorded from representative regions. Tomography confirmed that cavin-MiniSOG labeling extends all around the bulb (Figure 8A and Movie S1), and that caveolar coat complexes form periodic density maxima (Figure 8B and 8C). In order to estimate the degree of resolution of the MiniSOG label in our tomograms, line scans through individual densities were performed. Line scans perpendicular to the membrane showed that densities peaked sharply and exhibited a half maximum width of about 8–10 nm (Figure 8B). Line scans along caveolar membranes resolved densities separated by about 10 nm, confirming our previous data on thin sections (Figure 8C: compare to 7F). We carried out three-dimensional reconstructions of regions of the caveolar surface where such periodic density maxima were well resolved (Figure 8D). These reconstructions revealed both local maxima and apparent linear striations within the coat (Figure 8D and Movie S2). In some regions, the density maxima had a regular, lattice-like distribution, suggesting that the distribution of MiniSOG reflects an underlying higher-order lattice organisation of the caveolar coat. The three-dimensional reconstructions also reinforced the firm conclusion that the MiniSOG label, and hence the caveolar coat complex, are found all around the caveolar bulb without specific enrichment in the sides, apex, or neck of caveolae.
We show that caveolins and cavins can be purified as a single species of protein complex that excludes EHD2 and pacsin 2. We term this complex the caveolar coat complex. Purification of the caveolar coat complex, and a comprehensive quantitative analysis of its composition, allows us to put forward a model of the basic stoichiometry. Cavin 1 is a core component of the complex, and several independent experiments provide evidence for cavin 1 forming a trimer: Firstly, we identified a 180 kDa cavin 1 species, a size compatible with a trimer, which was relatively stable. Secondly, cavin 2 and cavin 3 interact with cavin 1 at a molar ratio of about 1∶3, suggesting that one cavin 2 or cavin 3 molecule associates with the cavin 1 trimer. Thirdly, cavin 1 is predicted to form a three-stranded coiled coil via its N-terminal domain (http://groups.csail.mit.edu/cb/multicoil/cgi-bin/multicoil.cgi) [54]. We therefore suggest that the core of the coat complex is composed of a cavin 1 trimer. For each cavin 1 molecule in the complex, there are likely to be 4 caveolins. Both in our experiments and previously [49],[50],[55], an SDS-resistant oligomeric state of caveolin 1 has been detected, with an apparent molecular mass of 350–400 KDa. Although this seems slightly too large to be the 12 caveolins predicted by our stoichiometric measurements and the presence of cavin 1 trimers, it is possible that SDS-resistant complexes run anomalously on SDS-PAGE due to pronounced secondary or tertiary structure. Our data imply that the basic unit of 1× cavin 2 or 3∶3× cavin 1∶12× caveolin must assemble into larger multimers to generate the 80S caveolar coat complex. Whether the 80S complex represents all of the coat present on an individual caveolar bulb or an intermediate level of structural organisation is not yet clear.
Previous studies have shown that cavins and caveolin 1 in detergent-solubilised lysates fractionate differently on gradients and do not efficiently co-precipitate, which could be interpreted as arguing for lack of interaction in intact cells [22],[35]. Although our cross-linking data strongly suggest that cavin 1 and caveolin 1 do in fact interact, it will be important to demonstrate this directly. Higher resolution structural information on cavins and caveolins individually and in complexes is required. This may allow elucidation of the precise nature of the molecular contacts made between cavin 1 and caveolin 1. Nevertheless, it is clear that a core complex containing cavin 1 and caveolin 1 is not dependent on the presence of cavins 2 or 3, as the ratio between cavin 1 and caveolin 1 in this complex does not change when cavin 2 or cavin 3 are not present.
Biochemical and imaging experiments argue that cavin 2 and cavin 3 compete for binding to the cavin 1 trimer, as cavin 2 can displace cavin 3 from the complex and vice versa. This implies that, when cavin 2 and cavin 3 are both present in the same cell, the 80S caveolar coat complex will contain both cavin 1 trimers bound to cavin 2, and cavin 1 trimers bound to cavin 3. The observations that cavin 1 will co-precipitate cavin 2 or cavin 3 even in the absence of caveolin 1, but cavins 2 and 3 do not co-precipitate in the absence of cavin 1, are consistent with this. We hypothesise that it is the balance between cavin 2 and cavin 3 that confers additional structural and functional properties on the coat complex. Either overexpression of cavin 2 or siRNA-mediated knockdown of cavin 3 caused a slight but measurable dissociation of cavin 1 and caveolin 1 from the coat complex. Moreover, cavin 2 and cavin 3 formed separate complexes with cavin 1 under noncrosslinking conditions, with the former being smaller or less stable than the latter. These combined data suggest that shifting the ratio between cavin 2 and cavin 3 within the 80S complex towards having relatively more cavin 2 could make the complex less stable, and vice versa. This might be a means to modulate caveolar functions in different tissues, as the ratio between cavin 2 and cavin 3 varies in vivo [22],[34], and the presence of cavin 2 is associated with apparently smaller cavin complexes in vivo [34]. Therefore, the findings presented here correlate with, and imply molecular explanations for, the in vivo data. Further in vivo experiments will be needed to address the functional and physiological consequences of the variation of the cavin 2 and cavin 3 complement within the caveolar coat complex.
Both MiniSOG-tagged cavins and immuno-EM show that the caveolar coat complex is found all around the caveolar bulb, while the caveolar neck is likely to contain additional proteins including EHD2. Notably, our data do not provide any evidence for EHD2 making direct contact with the caveolar coat complex [27],[28], leading to the concept that the neck region constitutes a separate subdomain within caveolae that is distinct from the rest of the bulb [41]. The observation that siRNAs that efficiently target expression of EHD2 have no effect on the size or composition of the caveolar coat complex provides additional evidence that EHD2 has a separate role within caveolae. The previous finding that EHD2 controls the plasma membrane association and dynamics of caveolae [27],[28] leads to the additional conclusion that the coat complex is likely to be the same whether caveolae exist as characteristic flask-shaped invaginations of the plasma membrane, or as intracellular membrane vesicles.
The production of reactive oxygen species by MiniSOG, and consequent deposition of osmiophilic DAB polymer, provides a highly specific label for electron microscopy [48], and our data highlight its utility as a probe for cell biological structures. Local maxima in density produced by MiniSOG are likely to reflect increased local concentration of the tag and hence the tagged protein. Our images suggest that diffusion of reactive oxygen and DAB product away from the MiniSOG is limited, as line scans across the consequent electron density reveal that periodic changes in density over a distance scale of 10 nm can be clearly resolved. Nevertheless, diffusion of singlet oxygen or DAB reaction product is likely to provide a limit to the resolution of MiniSOG generated label. The ultimate resolution achievable by the MiniSOG molecular contrasting system remains to be determined, but structural details measuring a few nanometers have been observed [48].
We show that the caveolar coat complex generates a lattice on the surface of caveolae that can be detected using MiniSOG fusions, as local density maxima in thin sections and in tomographic reconstructions of the caveolar surface. We present evidence from both TEM on thin sections and 3D tomography that the coat is regular, with a periodic spacing of 10–15 nm. Ridges or striations can be observed, as shown in Figure 8D, and some regions of the surface show regular arrays of maxima. Our data, however, do not completely resolve the high-resolution internal geometry of this lattice. It is possible that regions containing less well defined densities and occasional gaps in the lattice are due to local cellular factors restricting or enhancing diffusion of the MiniSOG-produced electron dense stain. It is also possible that the caveolar coat is not as well ordered or arrayed as, for example, clathrin-coated pits. Nevertheless, the key point is that the coat represents a single type of complex coating the caveolar bulb, rather than previously plausible alternative possibilities such as cavins and caveolins being present in different complexes with different distributions.
The observation of local maxima in MiniSOG density with a spacing of around 10–15 nm agrees very well with local density maxima around the caveolar bulb in ultrathin plastic sections prepared from samples stained conventionally [41], and with the spacing of striations on the surface of caveolae revealed by platinum coating of membrane fragments [6],[43]. It is therefore likely that the caveolar coat complex described here is responsible for previously reported, but molecularly undefined, ultrastructural features on the surface of caveolae. Our data, revealing the identity, basic stoichiometry, and distribution of this unitary complex around the caveolar bulb, open the way for further structural characterisation of the protein machinery for generating caveolae.
The following antibodies were used: Mouse anti-GFP (Roche, 11814460001), rabbit anti-PTRF (cavin 1) (Abcam, ab48824), rabbit anti-SRBC (PRKCDBP; cavin 3) (Abcam, ab83913), goat anti-SDPR (cavin 2) (R&D Systems, AF5759), goat anti-EHD2 (Abcam, ab23935), rabbit anti-Caveolin 1 (BD, 610060), mouse anti-flotillin-1 (BD, 610821), mouse anti-flotillin-2 (BD, 610384), mouse anti-clathrin heavy chain (×11), rabbit anti-GFP antibody (Abcam, ab6556), and rabbit anti-RFP (MBL, PM005).
Constructs for human PTRF-mCherry (cavin 1), human SDPR-mCherry (cavin 2), and human SRBC-mCherry (cavin 3) have been described [21]. To generate TEV-GFP-10×His fusion constructs, a TEV-GFP-10×His cassette was inserted into the BamHI/NotI sites of pClontech N1. The respective cDNAs were inserted in frame. To generate MiniSOG-mCherry cavin constructs, MiniSOG cDNA was inserted into pClontech N1 via BamHI/AgeI. Constructs were transfected into HeLa and RPE cells using FugeneHD (Promega) according to the manufacturer's recommendations. Clonal HeLa cell lines expressing cavin-TEV-GFP-10×His proteins were produced from single cell clones and cultured in DMEM, 10% FCS, penicillin/streptomycin supplemented with 0.4 mg/ml G418 (Sigma). RPE cells expressing cavin-MiniSOG-mCherry were selected by FACS and cultured in 50% DMEM/50% F12 medium, 10% FCS, 5 mM glutamine, penicillin/streptomycin, supplemented with 0.4 mg/ml G418.
For crosslinking studies, semiconfluent cultures of HeLa cells were washed twice with ice-cold PBS and incubated on ice with 1.2 mM DSP (Pierce) in ice-cold PBS for 1 h. A 100× DSP stock solution was prepared in DMSO fresh prior to use. After 1 h, DSP was quenched by addition of 1 M Tris pH 7.4 to a final concentration of 100 mM for 15 min. Cells were briefly rinsed in 100 mM Tris pH 8 and immediately scraped into lysis buffer (LB): 50 mM Tris pH 8, 300 mM NaCl, 5 mM EDTA, protease inhibitor cocktail (Roche). Dependent on the experiment, either 1% (v/v) Triton X-100, 2% (w/v) octyl-glucoside (OG), or a combination of 1% Triton X-100/1% OG were added to LB. Cell lysates were incubated on ice for 30 min and spun at 14,000 rpm in a table top centrifuge for 30 min at 4°C, followed by a second centrifugation for 10 min. Lysates were added atop a linear 10–40% (w/v) sucrose gradient prepared in LB plus 0.2% Triton X-100. Gradients were spun in a SW40Ti rotor at 37,000 rpm for 6 h at 4°C. Twelve 1 ml fractions were collected from the bottom of the gradient by tube puncture. For Western blotting, equal volumes (usually 250 µl) of each fraction were precipitated with MeOH/Chloroform. The pellet was dissolved in 1×LDS loading buffer (Invitrogen) and boiled for 2 min. Proteins were separated on NuPAGE 4–20% Tris/Glycine or 4–12% Bis/Tris gels (Invitrogen) and blotted onto PVDF membranes (Millipore).
For immunoisolation of GFP-tagged proteins, magnetic anti-GFP microbeads, and μcolumns (Miltenyi Biotech) were used. For immunoisolation of the HMW complex from sucrose gradients, fractions 8–10 were pooled (total 3 ml) and incubated with 20 µl anti-GFP beads for 2–4 h at 4°C rotating. Alternatively, 1 ml of total cell lysate was incubated with 10 µl anti-GFP beads. Lysates were applied to μcolumns and washed eight times with 2 ml of LB/1% Triton X-100 at room temperature. A final wash was performed with LB without detergent added. Protein complexes were eluted from the column with 140 µl 0.1 M TEA, pH 11.8, immediately neutralized by addition of 70 µl 1 M Tris, pH 7.4, and subjected to tandem mass spectrometry. Alternatively, protein complexes were eluted with elution buffer (Miltenyi Biotech) and separated by SDS-PAGE.
Immunoisolates from total cell lysates were separated on 4–12% NuPAGE Bis/Tris gels. Gels were washed twice in distilled H20 for 5 min each and stained with the fluorescent protein dye SYPRO RUBY (Lonza) for 1 h at room temperature. Gels were washed several times in distilled H20 and the fluorescence scanned on a Chemidoc XRS+ Molecular Imager. The intensities of protein bands corresponding to cavin-GFP fusion proteins, cavin 1, cavin 3, and caveolins were determined using Image Lad software. Bands corresponding to alpha and beta caveolin 1 as well as caveolin 2 could not be resolved clearly and were thus quantified as one band. All values were corrected for by subtracting background fluorescence from the same molecular weight regions of GFP control samples. To calculate relative molar ratios between the protein components of the complex, the following molecular masses were used: Cavin1-GFP, 80 kDa; cavin-2-GFP, 90 kDa; cavin-3-GFP, 65 kDa; cavin 1, 55 kDa, cavin 3, 36 kDa, caveolin, 20 kDa. For documentation, data were exported to Prism Graphpad.
On-target Plus SMART pool siRNAs against human cavin 2, human cavin 3, and human EHD2 were from Thermo Scientific (L-015910, L-016416, and L-016660, respectively). siRNAs against human flotillin 1 and flotillin 2 (J-010636-05, J-010636-06, J-003666-09, J-003666-10) were pooled and used as a control throughout. Cavin-1-TEV-GFP-10×HIS HeLa cell lines were transfected at 30% confluency using Oligofectamine (Invitrogen) and a total of 100 nM siRNA per transfection. Four to five days posttransfection, cells were cross-linked with 1.2 mM DSP as described above and then lysed in LB/1% OG/1% Triton X-100. Cell lysates were cleared by centrifugation at 14,000 rpm for 30 min and loaded atop 10–40% sucrose gradients. Gradients were spun as described above. To quantify the relative amounts of cavin 1 and caveolin 1 in the low molecular weight (LMW; fractions 3–5) and high molecular weight fractions (HMW; fractions 8–10), equal volumes of each fraction were pooled, MeOH/chloroform precipitated, and analysed by Western blotting. The ratio of caveolin 1 to cavin 1 in the HMW pool 8–10 was calculated from three independent experiments, using densitometry in ImageJ.
Wild-type MEFs, caveolin 1 −/− MEFs, or cavin 1 −/− MEFs [34] were co-transfected with equal amounts of pDNA using electroporation. Different combinations of the following constructs were used: Cavin-1-TEV-GFP-10×HIS, cavin-2- TEV-GFP-10×HIS, cavin3-miniSOG-mCherry, cavin-2-miniSOG-mCherry. Per co-transfection, 1×106 cells were transfected with 2.5 µg of each pDNA. Twenty-four h posttransfection, cells were cross-linked with 3 mM DSP as described above and lysed in LB/1% OG/1% Triton X-100. Lysates were cleared by centrifugation at 14,000 rpm and incubated with 10 µl anti-GFP microbeads (Miltenyi Biotec) for 2 h at 4°C. Immunprecipitates were washed five times with LB/1% Triton X-100 and eluted with 60 µl elution buffer (Miltenyi Biotec). Equal volumes were analysed by Western blotting.
HeLa or RPE cells were fixed in 4% paraformaldehyde in PBS pH 7.4 for 10 min and stained with primary antibodies o/n in PBS, 3% FCS, 0.2% saponin. Cells were washed with PBS and incubated in secondary antibodies for 1 h. TIRF microscopy was carried out using an Olympus IX71. Confocal micrographs were captured on a ZEISS 510 LSM using standard filter sets. Co-localisation was quantifed using the Pearson correlation coefficient, as implemented in the “Colocalization Finder “plugin for Image J (http://rsbweb.nih.gov/ij/plugins/colocalization-finder.html).
HeLa cells were grown on glass bottom petri dishes (MatTek) and fixed in 4% Paraformaldehyde in 0.1 M phosphate buffer pH 7.4 overnight at 4°C. After several buffer washes, followed by inactivation of reactive aldehyde groups using 0.1% sodium borohydride in phosphate buffer for 15 min, cells were permeabilised using 0.03% saponin in 20 mM phosphate buffer, 150 mM sodium chloride for 30 min. Cells were incubated in normal goat serum (Aurion) for 40 min prior to incubation in rabbit anti-GFP antibody (Abcam) used at 1∶800 for 4.5 h at room temperature. After thorough washing, cells were incubated with 1∶200 dilution of goat anti-rabbit ultrasmall gold (Aurion) overnight at 4°C. After washing cells were fixed with 2% glutaraldehyde in 0.1 M phosphate buffer for 30 min and washed with distilled water followed by silver enhancement of gold using R-Gent SE-EM (Aurion) reagents. Cells were then postfixed with 0.5% osmium tetraoxide in 0.1 M phosphate buffer on ice for 15 min. Cells were then dehydrated in an ascending ethanol series and embedded in CY212 resin. Ultrathin sections were stained with saturated aqueous uranyl acetate and Reynolds lead citrate and examined using a Philips 208 EM operated at 80 kV.
For photooxidation, cells cultured in MatTec glass bottom dishes were fixed at room temperature with 2% glutaraldehyde (EM grade, EMS Corp.), 2.5 mM CaCl2 in 0.1 M cacodylate buffer pH 7.4 (CB), and immediately transferred onto ice for 1 h. Cells were rinsed five times with ice-cold CB and blocked with 50 mM glycine, 10 mM potassium cyanide, and 10 mM aminotriazole in CB for 15 min on ice. Cells were washed five times with CB and transferred onto a cooled stage on a Leica SPE II confocal microscope. A freshly prepared solution of 0.5 mg/ml diaminobenzidine (DAB, Sigma) in CB was added to the cells. Areas of interest were photooxidized by illumination with blue light, using a 150W xenon lamp, a standard FITC filter set, and a 63× objective NA 1.3. After about 3–4 min, a brownish precipitate formed in place of the fluorescence. Cells were removed from the stage, washed five times with CB, and poststained with 1% osmium tetraoxide in CB for 30 min at room temperature. Cells were washed five times with water, followed by dehydration in 20, 50, 70, 90, and 100% EtOH. Cells were infiltrated in Durcupan ACM resin (EMS Corp.). Photooxidized areas were sawed out of the dish and sectioned. For 2D transmission electron microscopy (TEM), 80 nm sections were sectioned. Electron micrographs were recorded at 80 or 120 kV on a FEI T12 TEM. Images were recorded with Serial EM software and a 2k×2k Gatan CCD camera.
Three-dimensional electron tomography was carried out on 250–300 nm sections at 150 or 300 kV using a FEI Titan TEM. Sections were carbon-coated, glow-discharged, and dipped into a solution of 0.1% BSA and 5 nm colloidal gold particles. Dual tilt series were recorded at +/−60° with 1° intervals and a pixel size of 0.5 nm (at 18k). Images were captured using a 4k×4k Gatan Ultrascan 4000 camera. Reconstruction was accomplished using a combination of IMOD [56] and TxBR [57] reconstruction packages. Rough alignment of the two tilt series was done with IMOD software package, and fine alignment and reconstruction was done using the TxBR package.
|
10.1371/journal.ppat.1003073 | Lymphocytic Choriomeningitis Virus Infection in FVB Mouse Produces Hemorrhagic Disease | The viral family Arenaviridae includes a number of viruses that can cause hemorrhagic fever in humans. Arenavirus infection often involves multiple organs and can lead to capillary instability, impaired hemostasis, and death. Preclinical testing for development of antiviral or therapeutics is in part hampered due to a lack of an immunologically well-defined rodent model that exhibits similar acute hemorrhagic illness or sequelae compared to the human disease. We have identified the FVB mouse strain, which succumbs to a hemorrhagic fever-like illness when infected with lymphocytic choriomeningitis virus (LCMV). FVB mice infected with LCMV demonstrate high mortality associated with thrombocytopenia, hepatocellular and splenic necrosis, and cutaneous hemorrhage. Investigation of inflammatory mediators revealed increased IFN-γ, IL-6 and IL-17, along with increased chemokine production, at early times after LCMV infection, which suggests that a viral-induced host immune response is the cause of the pathology. Depletion of T cells at time of infection prevented mortality in all treated animals. Antisense-targeted reduction of IL-17 cytokine responsiveness provided significant protection from hemorrhagic pathology. F1 mice derived from FVB×C57BL/6 mating exhibit disease signs and mortality concomitant with the FVB challenged mice, extending this model to more widely available immunological tools. This report offers a novel animal model for arenavirus research and pre-clinical therapeutic testing.
| Arenaviruses are carried by rodents, and in South America and West Africa can cause a fatal hemorrhagic fever syndrome in humans. Food, water or household items contaminated with rodent urine can be a source for transmission. General supportive care, anti-fever medication and the antiviral drug Ribaviran are used, however no treatment has proven effective. Due to the lack of small animal models capable of reproducing the human disease, development of an effective therapeutic has been slow. Here we report that a common laboratory arenavirus isolate, lymphocytic choriomeningitis virus, known to cause only a mild infection in humans and a chronic, wasting disease in most laboratory strains of mice produces a hemorrhagic-like disease in the FVB mouse strain. These mice exhibit signs of bleeding, multi–organ involvement, changes in blood diagnostics and mortality indicative of hemorrhagic fever syndrome following infection. We also show that a drug approach to reduce inflammation as a result of immune responses to the virus reduced disease signs and improved survival. Our study provides a small animal model for testing new treatment approaches and points to drug targets that lessen disease severity and improve survival from arenavirus induced hemorrhagic fever.
| Viral hemorrhagic fevers (VHFs) are induced by viruses that belong to one of four families, Arenaviridae, Bunyaviridae, Filoviridae, Flaviviridae. The clinical symptoms of hemorrhagic fever vary depending on the severity and etiological agent but generally fever and bleeding are prominent manifestations of the disease. Hemorrhagic fever viruses, including arenaviruses, pose a significant public health threat both as emerging infectious diseases and as potential bioterrorism agents [1]. The majority of viruses in the Arenaviridae family require maximum biosafety containment (BSL-4) for handling which limits access for most researchers. In addition, the available animal models that induce hemorrhagic fever like symptoms require marmosets, hamsters, guinea pigs, primates, or immunocompromised mice [2], [3]. The lack of a non-immunocompromised mouse model for viral hemorrhagic fever makes it difficult to conduct pre-clinical drug screening. Mice are ideal for use in pre-clinical drug development because of their low cost and the extensive knowledge and reagents available for the species. There is a dire need for VHF therapeutic as there is no FDA approved drug available for hemorrhagic fever disease.
Ideally, the clinical course and signs produced in the animal model will parallel those observed in the human disease. The key characteristics of human viral hemorrhagic fever of arenaviral origin are multiorgan infections with hepatocellular necrosis and thrombocytopenia [2], [4], [5]. In this article, we report that the FVB strain of mice exhibits extreme susceptibility to hemorrhagic fever-like signs after LCMV-Clone 13 (LCMV-13) infection. FVB mice demonstrate thrombocytopenia, hepatocellular necrosis, petechiae, and death. This is in contrast to the C57BL/6 mouse strain's response to LCMV-13, which progresses to a chronic wasting disease [6]. FVB mice showed greatly increased (IL-6, IL-17 and IFN-γ) cytokine and (CXCL1 and MCP-1 and 3) chemokine production profiles early after infection compared to C57BL/6 mice and systemic TNF-α during the hemorrhagic phase of the disease. To investigate the underlying mechanism of the FVB pathology, separate groups of mice were either depleted of CD4+ or CD8+cells at time of infection. We found that mice deficient in either CD4+ or CD8+ cells maintained normal liver function and survived LCMV-13 infection. Furthermore, drawing from data produced in previous mouse HFV challenge experiments (e.g. Ebola or Marburg) we chose to examine the role of IL-17 responses in this arenavirus model. FVB mice treated to block IL-17 responses at the time of infection exhibited increased survival from LCMV-13 challenge compared to untreated mice. These data, to the best of our knowledge, describe the first mouse model of arenavirus induced hemorrhagic fever and support the possibility that T cell-mediated immunopathology plays a role in the underlying cause of HFV disease.
We initially sought to examine the anti-arenavirus activity of a modified PMO chemistry (PMOplus) that is complimentary to a sequence conserved within numerous arenavirus isolates, termed AVI-7012. The PMOplus chemistry has been shown to be efficacious in NHP filovirus lethal challenge models [7]. Conservation of the targeted sequence aligned to the L and S genomic segment and anti-genomic RNA is shown in Table 1. A similar targeting strategy demonstrated inhibition of arenavirus replication in vitro and in vivo utilizing a PMO conjugated to an arginine-rich peptide to enhance cellular uptake [8]. AVI-7012 (4 mg/kg) administered to C57BL/6 mice either prior to or following infection with LCMV-13 exhibited significant antiviral activity compared to PBS or scramble control treated mice when viral RNA was measured in kidney, spleen, brain and liver tissue (Figure 1). In pursuit of examining the uptake of PMOplus chemistry in vivo by cells that are known to support LCMV replication we employed a transgenic mouse model, which expresses enhanced green fluorescent protein (EGFP) as a positive readout of antisense activity via corrective splicing of the EGFP open reading frame [9]. As is common for many transgenic models this was produced on the FVB background strain of mouse.
Four days post-infection, initial observations showed the infected EGFP mice react in an atypical manner compared to C57BL/6 mice at this point in a LCMV-13 infection. EGFP mice exhibited ruffled fur and showed signs of lethargy. Moreover, overt signs of severe disease were apparent 2–4 days later with some but not all of the infected EGFP animals presenting the following signs: mucosal, cutaneous and organ hemorrhaging (Figure 2) and decreased blood pressure, discordination, unresponsiveness, hypothermia, and seizures. The anomalous results observed following this LCMV-13 infection prompted us to carry out a second study without antisense treatment and with standard strains of C57BL/6 and FVB mice using the same virus stock in order to determine if the disease manifestation was due to the EGFP transgene. FVB mice displayed disease signs similar to those of the EGFP mice and on day 6 post-infection, death ensued and by day 8 post-infection only 1 out of 8 FVB mice infected with LCMV-13 had survived (Figure 3a). The FVB survivor did not clear virus, but instead harbored a long-term chronic infection with high viral load detected in multiple organs as late as day 36 post-infection. In agreement with previous reports, C57BL/6 mice showed 100% survival after LCMV-13 infection (Figure 3a). However, infection with high dose of the less pathogenic Armstrong strain of LCMV did not lead to any signs of hemorrhagic disease (Figure 3b). Weight loss in LCMV-13 infected FVB mice commenced after day 3 post-infection and FVB mice plateaued in weight loss at day 6 post-infection at 10.2% net weight loss (Figure 3c). The body weight pattern of C57BL/6 mice was strikingly different with a sharp drop in weight of 9.8% by day 2 post-infection, then a rebound in weight followed by a dramatic weight loss of 19.8% by day 9 post-infection (Figure 3c). FVB mice displayed a pantropic infection with virus being detected in multiple organs. However, even though FVB mice developed a moribund state while C57BL/6 mice did not, viral load was comparable or less in spleens, lungs, kidneys, and livers of FVB compared to C57BL/6 mice (Figure 3d–e). Yet C57BL/6 mice show no clinical symptoms of hemorrhagic-like disease, which indicates the pathology associated with LCMV infection in FVB mice is not solely virus mediated. We further examined the relationship of disease to viral load by targeting virus replication with antisense to Arenavirus 5′ termini, which has been shown to inhibit LCMV replication both in vivo and in vitro (Figure 1; [8]). Figure 1 shows that AVI-7012 can inhibit viral load in C57BL/6 mice, however, when FVB infected mice were treated to reduce viral load they did not exhibit a concomitant increase in survival (Figure 3f)
Considering the clear signs of hemorrhage in the FVB mice following infection we next sought to assess the hematologic parameters in FVB and C57BL/6 mice infected with LCMV-13. Similarly to the clinical manifestation to most arenaviral hemorrhagic diseases, platelet count differences were the most striking. FVB mice showed reduced platelet counts with a range of 146–324 K/ml compared to platelet counts of 1318 K/ml and 1216 K/ml for LCMV-13 infected C57BL/6 and naïve FVB, respectively (Figure 4a). Lymphocyte and granulocyte counts were dramatically increased in blood of LCMV-13 infected mice compared to naïve FVB and infected C57BL/6 mice (Figure 4a). However, spleen size was significantly smaller in LCMV-13 infected FVB mice, independent of gender, compared to infected C57BL/6 mice as well as total T cell counts in the spleen (Figure 4b and c). No mortality was observed in either gender at the challenge inoculum of 104 p.f.u.. In accordance with the smaller spleens was the histopathological finding of severe splenic necrosis in LCMV-13 infected FVB mice while infected non-diseased FVB mice displayed no detectable necrosis (Figure 5a upper panels). Liver pathology in infected diseased mice revealed many single cells undergoing necrosis/apoptosis and randomly scattered zones of parenchymal necrosis with associated degenerate neutrophils (Figure 5a middle panels). Histological signs of modest alveolar edema and/or atelectasis were observed in the lungs of the infected FVB diseased mice but were absent in the infected non-diseased mice. Tissues from infected C57BL/6 mice showed no such disease indication except the livers of some C57BL/6 mice displayed rare tiny foci of hepatocellular degeneration and necrosis with scattered neutrophils (data not shown).
The following clinical biochemistry parameters were analyzed: alkaline phosphatase (ALK), alanine aminotransferase (ALT), aspartate aminotransferase (AST), calcium, cholesterol, triglycerides, albumin, creatinine, glucose, phosphorous, total bilirubin (TBIL), blood urea nitrogen (BUN), and total protein. LCMV-13 infected FVB mice had increased levels of ALT, AST, TBIL, and BUN compared to control mice indicating severe kidney dysfunction and hepatocyte destruction (Figure 5b). Calcium, cholesterol, triglycerides, albumin, creatinine, glucose, total protein, alkaline phosphatase, and phosphorous were normal in LCMV-13 infected FVB mice compared to controls (data not shown). These data combined demonstrate a pantropic infection leading to thrombocytopenia, cutaneous hemorrhaging, hepatic dysfunction, and ultimate death.
LCMV induces a well-characterized immunoregulatory state in most immunocompetent inbred mouse strains, including C57BL/6 mice [10] with subclinical disease signs. That FVB mice progress to hemorrhagic state and succumb implies a role of the immune response in the manifestation of the FVB hemorrhagic disease. To gain some insight into the inflammatory response prior to onset of hemorrhagic fever symptoms, we assayed for systemic cytokines in LCMV-13 infected FVB mice bled at day 3 post-infection and found increased levels of multiple pro-inflammatory cytokines and chemokines (Figure 6a). LCMV-13 infected FVB mice showed increased levels of IL-6 compared to C57BL/6 mice (2152 pg/ml versus 187 pg/ml), IFN-γ (2184 pg/ml versus 6399 pg/ml), CXCL1 (2530 pg/ml versus 154 pg/ml), MCP-1 (12342 pg/ml versus 6413 pg/ml), and MCP-3 (6284 pg/ml versus 1589 pg/ml). Strikingly, on day 1 post-infection FVB mice exhibited systemic IL-17A (78 pg/ml) while levels remained undetectable in C57BL/6 mice. At times during severe disease in FVB mice (day 6–8 post-infection), increased levels of systemic TNF-α were found compared to C57BL/6 mice (Figure 6b). TNF-α levels in LCMV-13 infected FVB mice were between 61–92 pg/ml whereas C57BL/6 mice had undetectable levels of TNF-α.
In order to investigate the immune component of the FVB-related hemorrhagic disease further, we depleted either CD4+ or CD8+ cells in FVB mice with anti-CD8 or anti–CD4. Mice treated with anti-CD4 or anti-CD8 antibody at time of infection and 1 day after demonstrated 100% survival up to day 16 post-infection compared to 0% survival for PBS treated FVB mice (Figure 7a). While peak weight loss for anti-CD8 treated mice was similar to PBS treated mice (17.5+/−0.9% for anti-CD8 treated versus 14.6+/−0.2% for PBS treated), anti-CD8 treated mice regained weight after day 10 post-infection and plateaued at ∼10% lost body weight (Figure 7b). While peak weight loss was not as severe in anti-CD4 treated mice (13.5+/−5.8%), a similar trend was seen between anti-CD4 and anti-CD8 treated mice in that they began to regain weight around the same time PBS treated mice succumbed to disease. Anti-CD8 treated mice had lower AST (2245 u/ml vs 390 u/ml for PBS and anti-CD8 treated mice, respectively) and ALT (1584 u/ml vs 364 u/ml for PBS and anti-CD8 treated mice, respectively) readings than PBS treated mice indicating increased liver function (Figure 7c). Viral load in liver was similar between anti-CD8 treated and untreated mice (Figure 7d), which again suggests viral replication alone is not causing hemorrhagic disease.
Taken together, hemorrhagic disease in this model appears to be caused by a skewed immune response. IL-17 has been shown to play a role in tissue destruction and we identified increased levels of systemic IL-17 early after infection (Figure 6). Previous data from our lab has shown antisense ablation of IL-17 receptor C (IL17RC) can prevent mortality in a mouse model of Ebola hemorrhagic disease. We, therefore, probed the role of IL-17 in arenavirus hemorrhagic disease. A delivery peptide conjugated antisense phosphorodiamidate morpholino oligomer (PPMO) was designed to target the splice-donor site of exon 12 of IL17RC (IL17RC SD12), thereby disrupting the translational reading frame of IL17RC, leading to reduced levels of surface IL17RC. Mice were treated with 7.5 mg/kg of IL17RC SD12 at day 0, day 1, and day 2 post-infection and monitored for disease symptoms. While only 12.5% of animals survived when treated with PBS, 66.7% of animals treated with IL17RC SD12 were still living at day 8 post-infection (Figure 8a). Viral load was significantly reduced in liver, lung, kidney, and brain with IL17RC SD12 treatment compared to untreated FVB (Figure 8b). Furthermore, IL17RC treatment protected liver and kidney function as seen by the reduced levels of ALT, AST, total bilirubin, and alkaline phosphatase found in serum (Figure 8c). Combined, these data suggest a role for IL-17 in mediating arenaviral hemorrhagic induced disease in FVB mice.
The FVB strain is useful for production of transgenic mice on inbred genetic backgrounds due to robust fecundity, and fertilized eggs contain large and prominant pronuclei facilitating microinjection of [11]. However, few immunological tools are available to probe the factors influencing hemorrhagic disease on the FVB H-2q background. We, therefore, bred C57BL/6 (H-2b) and FVB mice to create F1 hybrids. F1 hybrids were infected with a dose range of LCMV-13 and monitored for hemorrhagic disease symptoms. Mice infected with 2×106 p.f.u. showed 100% mortality by day 9 post-infection, 66% mortality with 6×105 p.f.u., and no mortality at 2×105 p.f.u. (Figure 9a). We further confirmed hemorrhagic disease in F1 hybrids by infecting mice with either LCMV-13 or LCMV-Armstrong. Day 7 post-infection, C57BL/6 mice infected with LCMV-13 and F1 hybrids infected with LCMV-Armstrong demonstrated no clinical symptoms except for weight loss while F1 hybrids infected with LCMV-13 demonstrated a significant decrease in body temperature. F1 hybrids infected with LCMV-13 had a body temperature of 29.5+/−0.6 degrees Celsius while F1 hybrids infected with LCMV-Armstrong maintained normal body temperature of 36.8+/−0.4 degrees Celsius (Figure 9b). We then assessed LCMV-specific T cell responses in F1 hybrids. As can be seen in Figure 9d, F1 hybrids infected with LCMV-13 showed a significant reduction in CD44hi CD8 T cell numbers (2.4+/−1.2×106 cells versus 5.8+/−2.5×106 cells for LCMV-13 versus Armstrong infected F1 hybrids, respectively). LCMV-13 infected F1 hybrids also demonstrated reduced LCMV-specific T cells as assessed by both MHC pentamer staining IFN-γ production (Figure 9e). F1 hybrids infected with LCMV-13 had significantly reduced numbers of Db/NP396–404 specific and Db/GP33–41 specific CD8 T cells as assessed by MHC pentamer staining and IFN-γ production, respectively (Figures 9e). These results confirm the findings from Figure 4 showing that mice undergoing hemorrhagic disease have lower T cell numbers late in infection.
One of the requirements for FDA approval of antiviral therapeutics is drug testing in accepted animal models that reproduce human disease as closely as possible. It has been thought that one of the fundamental features of LCMV biology is its ability to establish chronic infections in mice. This is in contrast to the disease course of LCMV in rhesus macaques, which succumb to hemorrhagic fever [12]. It has been reported that under certain circumstances LCMV can lead to mortality in mice. The New Zealand Black strain succumbs to a pulmonary disease much like hantavirus pulmonary syndrome [13]. Likewise, C57BL/6 mice infected with a medium dose of LCMV-13 displayed a similar lung pulmonary edema and interstitial mononuclear infiltration as NZB mice and 23% of those mice died [14]. Two other LCMV infection models have been reported to produce mortality with similar disease signs to the FVB model and a possible link to IL-17 production. The earliest was reported by Sarawar et. al, 1994. Here a subclinical infection of LCMV followed by low dose i.p. exposure to Staphylococcus aureus Enterotoxin B (SEB) resulted in a disease characteristic of hemorrhagic toxic shock leading to significant mortality in Vβ8.1 transgenic mice [15]. Although, Sarawar et. al would have not been able to measure IL-17 at the time, it has been shown in later studies that SEB will potently induce IL-17 expression in mice [16]. Additionally, large amounts of IFN-γ and IL-6, along with a transient increase in TNF-α were detected. Recently it has been shown the inflammatory effect of IL-17 on endothelial activation and neutrophil recruitment acts synergistically with TNF-α [17]. Both of these cytokines were produced in significant amounts in the FVB-LCMV hemorrhagic model and could account for the exaggerated inflammatory response in both models. Moreover, prior depletion of T cells gave similar results whereby the lethal effects of the LCMV infection with SEB were also greatly diminished. Although the source and precise role of IL-17 in the FVB-LCMV hemorrhagic model remains to be determined, anomalous production of IL-17 has been reported for mice deficient in T-bet and eomesodermin when infected with LCMV. These mice fail to differentiate LCMV-specific CD8+ killers T cells, required for defense against the virus, but instead produce a CD8+ IL-17-secreting lineage [18]. Upon viral infection, these mice develop a CD8+ T cell-dependent, progressive inflammatory and wasting syndrome characterized by multi-organ infiltration of neutrophils.
There is currently no mouse model that demonstrates multiple symptoms of the human disease of arenaviral hemorrhagic fever [2]. The disease of FVB mice infected with LCMV-13 described in this report mimics LCMV disease in macaques and many of the clinical signs of Argentine hemorrhagic fever (Table 2). However, the disease progression in FVB departs from the sequelae observed for Lassa Fever (Table 2). Viral load was not a good predictor of disease in our model while disease outcome is often predicted by viremia level in Lassa fever patients [19], [20], [21]. The smallest animal model currently used for arenaviral hemorrhagic fever is the guinea pig [2]. While this is a useful model in some respects, the major limitations of using guinea pigs for therapeutic testing and experimentation is the lack of information and reagents available for guinea pig analysis. Using the FVB inbred mouse strain opens up access to the plethora of tools available for mouse research. In addition, the cost and lower biosafety level (BSL-2+) using mice and LCMV-13, respectively, allows for high throughput screening.
That FVB mice display an acute lethal disease after LCMV-13 infection while C57BL/6 mice develop a chronic infection reinforces the role of the host genetic factors in skewing the arenaviral disease course. Early in infection, we found increased levels of proinflammatory cytokines and chemokines in LCMV-13 infected FVB mice compared to C57BL/6 mice, which points to an immune component in the onset of hemorrhagic fever disease. It has been shown that LCMV-13 infected mice that were depleted of platelets develop lethal hemorrhagic anemia that is dependent on virus induced type I interferons [22] indicating a role for proinflammatory response in hemorrhagic disease. In addition, high levels of IL-6 and IFN-γ were found in rhesus macaques that succumb to lethal disease after LCMV infection [12]. One report has suggested that suppression of pro-inflammatory responses is partly responsible for the terminal shock associated with arenavirus infection in guinea pigs [23]. However, results from this study showed increased IFN-γ and MCP-1 in high pathogenic pichinde virus infected guinea pigs day 2 post-infection, indicating an early, robust proinflammatory response. Similarly, many of the proinflammatory cytokines and chemokines that were upregulated in LCMV-13 infected FVB mice early in infection were below C57BL/6 levels late in infection (data not shown).
That either CD4 or CD8 antibody treatment prevents death in all LCMV-13 infected FVB mice also supports an immune mediated component in the development of hemorrhagic disease. We believe that the disease is T cell mediated as waiting 3 days after infection before anti-CD8 treatment continued to protect mice from lethality (data not shown). Presumably LCMV has gone systemic by 3 days post-infection and the protective capability of anti-CD8 is T cell depletion rather than a CD8+ LCMV reservoir. The finding that CD4+ and CD8+ T cells are involved in the pathogenisis of LCMV-13 in FVB mice is in striking contrast to the C57BL/6 model. LCMV-13 induces an exhausted T cell phenotype where numerous inhibitory receptors are upregulated on CD8+ T cells, which lead to a chronic wasting disease [24]. However, LCMV-Armstrong infection in FVB mice mimics the disease progression of C57BL/6 with peak weight loss 8–9 days post-infection and then clearance of virus (Figures 3 and 9). This suggests that high viremia can skew the immune response. South American arenaviruses and Lassa induce splenic and lymphoid necrosis with varying degrees of lymphoid depletion [25]. While our T cell depletion studies point to a T cell-mediated pathogenesis in the FVB model, many signs in the FVB mice are consistent with human arenavirus immunosuppression such as splenic necrosis (Figure 5a), splenic involution (Figure 4b), and reduced T cell numbers (Figures 4c and 9). Our model suggests that skewing of the T cell response, possibly Th17, promotes unchecked inflammation, which then leads to splenic necrosis, lymphoid apoptosis, and lymphopenia. Our results are similar to Flatz et. al where mice that have humanized MHC class I develop severe Lassa fever whereas T cell depletion prevents disease [24]. We propose a three-tiered model similar to Flatz et. al where three outcomes are possible 1) potent T cell response controls virus (LCMV-Armstrong), 2) intermediate T cell response fails to control virus and triggers severe disease (FVB mice), 3) depletion of T cells allows persistence of Lassa virus with mild disease (T cell-depleted FVB mice).
The FVB mouse, named for its susceptibility to the B strain of Friend leukemia virus, exhibits a predisposition to several viral induced pathologies compared to other mouse laboratory strains [26]. Some examples are the neurological and immunological sequela observed subsequent to infection with either MOMuLV, a retrovirus, Theiler's a picornovirus and Minute virus, a parvovirus [27], [28], [29]. In contrast, the C57BL/6 (H-2Db) mouse strain is resistant to disease or persistence following infection with these viruses.
In the case of Theiler's virus disease, resistance has been linked to the MHC loci [30]. Specifically it has been shown that the FVB×C57BL/6 F1 and FVB H2-Db transgenic mice are resistance to persistent Theiler virus infection and development of inflammatory lesions. This indicates that the H-2Db allele confers dominance over H-2q. Our observation that the FVB×C57BL/6 F1 does not recapitulate resistance to LCMV infection or onset of hemorrhagic disease suggests that H-2Db is not dominant in this condition of viral induced immunopathology. Although it is yet to be determined what immune related gene(s) influence the FVB and F1 mice susceptibility to hemorrhagic disease, a probative advantage prevails with the F1 hybrid. While maintaining the FVB-like hemorrhagic disease the F1 possesses an H2-Db immune system, which will allow further dissection of the mechanism due to the availability of immunological reagents.
In summary, we have discovered a unique model for arenaviral hemorrhagic fever that could have broad applicability for arenaviral therapeutic development and arenavirus research. While our model does not mimic all hemorrhagic fever symptoms from viruses in the family Arenaviridae, the development of a hemorrhagic fever mouse model with an intact immune system represents a major advancement for arenavirus research and preclinical testing. In addition, our data suggests an immune mediated component to the onset of arenavirus hemorrhagic fever. If there is an immune component to the susceptibility to hemorrhagic fever, as our data suggests, the FVB and F1 hybrid LCMV-13 infection models will provide the tools necessary to decipher what the key factors are in initiating arenaviral hemorrhagic fever.
Animal experiments were conducted to comply with the Public Health Service (PHS) Policy on the Humane Care and Use of Laboratory Animals, the US Department of Agriculture's (USDA) Animal Welfare Act & Regulations (9CFR Chapter 1, 2.31), the Animal Care and Use Review Office (ACURO), a component of the US Army Medical Department and Medical Research and Material Command USAMRMC and the United States Government Principles for the Utilization and Care of Vertebrate Animals Used in Research, Teaching and Testing with prior approvals for established protocols from the Oregon State University Institutional Animal Care and Use Committee and ACURO when appropriate.
C57BL/6 and FVB mice were purchased from Jackson Laboratories. C57BL/6 and FVB mice were bred to create F1 hybrids at Oregon State University Department of Animal Resources. All mice were housed at Oregon State University Department of Laboratory Animal Resources facility and experiments were conducted according to approved Institutional Animal Care and Use Committee protocols. Mice were used at 5–8 wks of age. Mice were infected with 1–2×106 p.f.u. LCMV-clone 13 or LCMV-Armstrong. Standardized recording of death and disease symptoms was performed on a daily basis. Symptoms of severe disease were hunched posture for more than 24 hours without movement, discordination, and shaking upon movement. For CD8 depletion, mice were treated with 0.5 mg clone 53–6.7 anti-CD8 antibody at time of infection and 24 hours post-infection. For anti-CD4 depletion, mice were treated with 0.3 mg CD4 antibody clone GK1.5. For IL17RC SD12 antisense treatments, mice were dosed with 7.5 mg/kg intraperitoneal route on days 0, 1, and 2. IL17RC SD12 sequence is CTG GAC ACA GAG GTT GG. The PMOplus and PPMO compounds used were manufactured as previously described, respectively [7], [8].
Tissues were weighed and 300 ml of DMEM was added per tube. A stainless steal bead was added and tissues were homogenized in a Tissue Lyser (Qiagen) for 3 min at 20 Hz. Kidneys were digested in DMEM+1 mg/ml collagenase for 30 min prior to homogenization. After homogenization, tissues were spun down at 14 k for 5 min and 50 µl sup was taken for RNA isolation with Magmax Blood RNA Isolation Kit (Ambion) according to manufacturer's instructions. Viral load of the spleen, kidney, liver,and lung was determined by qRT-PCR amplification of viral GPC [forward primer (5′-GCAAAGACCGGCGAAACTAG-3′), reverse primer (5′-CGGCTTCCTGTTCGATTTGGT-3′) and a taqman probe (5′-CCCAAGTGCTGGCTTGTCACCAAT-3′)]. To translate the qRT-PCR results from a cycle threshold (CT) value into copy number, a standard curve was generated using PCR product of the GP amplicon. The GP amplicon PCR product was run on a gel, excised and purified, then spectrophotometry was used to quantify a copy number for that CT value. Plaque assays were conducted as previously described [31].
Blood was collected through retroorbital puncture using capillary tubes on anesthetized mice directly before sacrifice. Serum was collected after spinning at 8 k for 5 min. Blood and serum were sent to Charles River Laboratories for CBC with differential and Clinical Chemistry profiles.
Serum was assayed for cytokine and chemokines using Flow Cytomix bead system (Bender Medsystem) according to manufacturer's protocol.
Db/GP33, Db/GP276, and Db/NP396 pentamers were purchased from Proimmune (Sarasota, FL). Single cell suspensions were stained with pentamer according to manufacturers protocol and then subsequently stained for CD8, CD44, and CD19 antibodies (BD Biosciences, San Jose, CA). Pentamer positive cells were detected on FC500 (Beckman Coulter, Indianapolis, ID) and analyzed on FlowJo software (Ashland, OR). For ICS analysis, splenocytes were stimulated overnight with 1 µg/ml of the indicated peptides. BFA (Ebioscience) was added 4 hours prior to harvest and cells surface stained for CD44 and CD8, fixed in cytofix/perm and washed in perm/wash buffer (BD Biosciences). Cells were then incubated with IFNγ antibody (BD Biosciences).
|
10.1371/journal.pntd.0005074 | One Health Interactions of Chagas Disease Vectors, Canid Hosts, and Human Residents along the Texas-Mexico Border | Chagas disease (Trypanosoma cruzi infection) is the leading cause of non-ischemic dilated cardiomyopathy in Latin America. Texas, particularly the southern region, has compounding factors that could contribute to T. cruzi transmission; however, epidemiologic studies are lacking. The aim of this study was to ascertain the prevalence of T. cruzi in three different mammalian species (coyotes, stray domestic dogs, and humans) and vectors (Triatoma species) to understand the burden of Chagas disease among sylvatic, peridomestic, and domestic cycles.
To determine prevalence of infection, we tested sera from coyotes, stray domestic dogs housed in public shelters, and residents participating in related research studies and found 8%, 3.8%, and 0.36% positive for T. cruzi, respectively. PCR was used to determine the prevalence of T. cruzi DNA in vectors collected in peridomestic locations in the region, with 56.5% testing positive for the parasite, further confirming risk of transmission in the region.
Our findings contribute to the growing body of evidence for autochthonous Chagas disease transmission in south Texas. Considering this region has a population of 1.3 million, and up to 30% of T. cruzi infected individuals developing severe cardiac disease, it is imperative that we identify high risk groups for surveillance and treatment purposes.
| In this study, we contribute to the growing body of evidence for autochthonous Chagas disease transmission in south Texas along the US-Mexico border. We found that coyotes, shelter dogs, and vectors in this region demonstrated high infection rates of T. cruzi. Random sampling of residents also revealed a higher than expected disease burden that had previously been undiagnosed. With up to 30% of infected individuals developing potentially fatal cardiac disease, it is imperative that we identify and treat patients before irreversible clinical manifestations have occurred. Future prospective studies are necessary to elucidate and validate the disease burden in this area.
| Chagas disease (Trypanosoma cruzi infection) can cause fatal cardiomyopathy in up to 30% of infected people [1]. Transmission to mammals occurs via vector, oral, congenital, and/or transfusion/transplantation routes [2]. The triatomine vector, or “kissing bug,” serves as the predominate mode of transmission, particularly in established sylvatic and/or domestic transmission cycles [3]. Over 100 different wildlife mammalian species are competent reservoirs of disease and have been implicated in propagation of sylvatic transmission cycles in nature [4]. Canines, in particular, are important components of peridomestic transmission, resulting in a bridge between sylvatic and domestic transmission cycles [5–7]. Finally, human infections can occur when vectors establish nests inside or near the home, and vectors feed on both humans and domesticated animals [7, 8].
Disease prevalence is highest in impoverished regions of endemic countries due to a plethora of societal factors, including substandard living conditions that result in increased exposure to vectors [9]. While the southern United States is not traditionally considered an endemic area, recent evidence has implicated the establishment of vector transmission cycles, particularly in Texas [10, 11]. Historical evidence of T. cruzi infected vectors and mammalian reservoirs date back to the early 1900s [12]. While the first documented locally acquired human case was published in Corpus Christi, Texas in 1955, the south Texas region, including the Rio Grande Valley, has been the subject of investigation by public health authorities dating back to the 1940s [12].
South Texas has compounding factors that could contribute to this area being a high-risk region for transmission. Within the state, sylvatic transmission cycles have been reported with seven different vector species and 27 sylvatic mammalian reservoirs [10]. The potential for sylvatic spillover to humans in this region has been implicated from increased outdoor exposure and interactions in rural environments [13]. In addition, colonias (primarily Hispanic communities) in this region of Texas have unprecedented poverty rates and living conditions that allow for easy access for vectors to enter and colonize homes, which might place residents at an increased risk of domestic transmission [5, 14]. Despite this compounding evidence of increased potential for Chagas disease in the region, epidemiologic assessments are lacking. The aim of our current assessment was to ascertain the prevalence of T. cruzi in three different mammalian species (coyotes, stray domestic dogs, and humans) and vectors (Triatoma species) to understand the disease burden attributable to Chagas disease among sylvatic, peridomestic, and domestic cycles.
Texas Department of State Health Services in the lower Rio Grande Valley originally collected terminal samples of coyote sera as part of their rabies control programs in 2005–2006, and secondary aliquots from these specimens were shared for T. cruzi testing for the purposes of this study. Canine sera collection and Chagas disease testing were approved by the University of Texas Health Science Center Animal Welfare Committee (AWC-07-147 and AWC-03-029). For the human seroprevalence aspects of our study, the original Cameron County Hispanic Cohort study was reviewed and approved by the University of Texas Health Science Center at Houston Committee for the Protection of Human Subjects (HSC-SPH-03-007B), and Chagas disease testing on coded samples was approved under Baylor College of Medicine Institutional Review Board (H-32192).
We conducted a retrospective analysis of previously collected sera from coyotes, stray domestic dogs housed in public shelters, and residents participating in related research studies. With regards to the coyote specimens, secondary aliquots from specimens noted above were shared by the Texas Department of State Health Services for T. cruzi testing. For domestic dog specimens, sera were collected in 2007 and 2009 from juvenile (less than 6 months of age and over 8 weeks of age based on tooth development) stray dogs housed in public shelters at one of two locations (Brownsville in Cameron County and Edinburg in Hidalgo County). The rationale for collecting samples from dogs under 6 months of age was to identify new, acute cases of infection so that incidence, as opposed to prevalence, could be determined. We purposefully excluded puppies under 8 weeks of age to eliminate issues related to the possible transfer of Chagas-positive maternal antibodies.
Investigators from the University of Texas Health Science Center at Houston, School of Public Health, Brownsville Regional Campus, collected sera from an established cohort living in Cameron County, TX. The participants were recruited from randomly selected households between 2005 and 2008 as a means of assessing the general health of residents along the US-Mexico border. Potential participants were not excluded based on race/ethnicity, with all race/ethnicities eligible for study inclusion. Data from the original health questionnaire and echocardiograms performed by the Cameron County Cohort (CCC) study were available for descriptive analysis [15].
From 2012 to 2013, we received 115 Triatomine insects that were collected in peridomestic areas by citizens across 6 counties in south Texas. Insect specimens were shipped, typically live, to The University of Texas Rio Grande Valley for further processing. PCR testing was performed in collaboration with Baylor College of Medicine Laboratory for Vector-Borne and Zoonotic Diseases.
Serum samples were thawed and analyzed using Chagas Stat-Pak and DPP assays (Chembio Diagnostic Systems, Inc, Medford, NY). These rapid immunochromatographic assays test for antibodies against T. cruzi. These highly sensitive and specific assays were designed for feasibility in field-testing of both human and canine blood [6, 16–18]. Tests were examined visually and scored as negative or positive, following manufacturer’s directions. A positive sample was defined as being positive on both assays. Negative samples included those that were positive on only one diagnostic but negative on the second diagnostic. Any equivocal samples were retested for further clarification. Due to the samples being retrospectively tested without potential for prospective clinical intervention and the exploratory nature of the project, additional confirmation testing with alternate diagnostics was not performed.
For T. cruzi testing and taxonomic species identification of Triatoma insects, the posterior third of the insects’ abdomen was homogenized with a 5 mm stainless steel bead in AL buffer (Qiagen, Valencia, CA) in TissueLyser II (Retsch, Haan, Germany) for 3 min at 25 Hz. Following manufacturer’s instructions, DNA was then extracted using DNeasy Blood & Tissue kit (Qiagen, Valencia, CA). T. cruzi DNA detection and insect-specific mitochondrial 16S DNA for speciation were performed using PCR and sequencing as previously described [8, 19].
Descriptive statistics were used to identify prevalence infection rates with 95% confidence interval (CI) and stratified by pertinent variables. For domestic dogs, positive infection was translated to incidence since all dogs would have acquired infection in the first 6 months of life. Statistical analysis was performed using STATA v12 (College Station, TX). Spatial analysis was performed using MapInfo Professional v11.5 (Stamford, CT).
Coyote samples collected in the Rio Grande Valley had an overall seroprevalence rate of 8% (16 out of 199; 95% CI = 4.2% to 11.8%) (Table 1). Sampled coyotes were evenly distributed by gender (45% female) and all but one were adults. There was no difference in seropositivity by year of sampling. Interestingly, seroprevalence varied with regards to county of collection, with the highest seroprevalence identified in Zapata County (16%; 10/64), followed by Jim Hogg County (14%; 3/22), Dimmit County (10%; 2/20), and Webb County (1%; 1/83) (Fig 1). No positive coyotes were identified in Cameron, Hidalgo, Starr, or Wallacy counties, although sample sizes from each of these counties were low (range 1 to 4, total tested = 10).
Samples collected from juvenile domestic dogs from neighboring Hidalgo and Cameron counties had an overall serologic incidence of 3.8% (8 out of 209 samples; 95% CI = 1.2% to 6.4%). We found a pronounced increase (4.4 fold) in Chagas incidence when comparing sampling in 2007 to 2009 (Fisher’s exact test, p-value = 0.04, 95% CI = 1.1 to 18.0), with 2% (3/152) of dogs positive in 2007 versus 9% (5/57) found positive in 2009.
Of 841 human sera samples tested from participants in the CCC, 3 individuals (0.4%; 95% CI = 0% to 0.8%) tested positive on both Stat-Pak and DPP assays. Limited residential history, medical histories and socioeconomic variables were reported as listed below. The precise origin and duration of their infection is unknown.
CCC Participant 1 was a 76-year-old female born in Canary, Texas (now known as Livingston, Texas) with a 52-year residential history in Brownsville, Texas. Case-patient 1 reported no current employment with an annual disability-benefit income of $3,336. Her medical history included diabetes, stroke, and hypertension. Case-patient 1’s mother was born in Texas while her father was born in central Mexico (Guanajuato). No data regarding any abnormal cardiac findings were available for this case-patient. On follow-up, participant’s husband reported that the participant had died recently with an apparent cause of death reported as leukemia.
CCC Participant 2 was a 45-year-old male born in San Luis Potosi, San Luis Potosi, Mexico with a 6-year residential history in Brownsville, Texas. In addition, he reported a prior 6-year residential history (while attending school) in the Brownsville, Texas border town of Matamoros, Tamaulipas, Mexico. Case-patient 2 was employed at the time of enrollment, reporting an annual income of $12,000. His past medical and social histories included diabetes and smoking. Both parents were born in north-central Mexico (San Luis Potosi). An echocardiogram performed on this participant showed normal left ventricular and right ventricular systolic function, mild concentric left ventricular hypertrophy, grade 1 left ventricular diastolic dysfunction, and no significant valvular abnormalities. The participant reported no symptoms related to any type of infection, and no additional cardiac evaluations were performed.
CCC Participant 3 was a 63-year-old male born in Matamoros, Tamaulipas, Mexico with a 22-year history of living in Brownsville, Texas. Case-patient 3 was retired with a prior occupational history in agriculture (occupational duration unknown) and a current annual income of $10,248. His medical history was negative for pre-existing conditions or co-morbidities. Case-patient 3’s parents were born in northern Mexico (Nuevo León). An echocardiogram performed at the same time as the original blood collection demonstrated normal biventricular systolic function, mild concentric left ventricular hypertrophy, grade 1 left ventricular diastolic dysfunction, and no significant valvular abnormalities. Similarly, the participant reported no symptoms, and no additional cardiac evaluations were performed.
Finally, to determine the likelihood of infection in vectors in the region, PCR was performed on 115 insects (Triatoma species) collected around homes across 6 counties of south Texas. We found 65 (56.5%) positive for T. cruzi DNA, with prevalence ranked by county as follows: Brooks County (84%; 21/25), Hidalgo County (60%; 6/10), Jim Wells County (50%; 12/24), Kleberg County (47%; 22/47), Dimmit County (33%; 2/6), and Cameron County (0%; 0/1); 2 positive insects did not have a georeference provided. The most common insect collected was Triatoma gerstaeckeri (96.5% of insects; 62/111 T. cruzi positive), followed by T. lecticularia (2.6% of insects; 2/3 T. cruzi positive) and T. sanguisuga (0.9% of insects; 1/1 T. cruzi positive).
Chagas disease transmission has been identified along the Texas-Mexico border dating back to the 1970s [20, 21]. Our current study is the first to assess the infection status of vectors and seroprevalence among mammalian and human populations all living in the same geographic region of south Texas. Seroprevalence was highest among the sylvatic adult coyote reservoir (8%), moderate among peridomestic juvenile dogs in community shelters (3.8%), and lowest among local residents (0.36%), with one of the three positive CCC participants having a life-long history of living in Texas. In addition to finding evidence of infection in canines and humans, we found a high percentage (56.5%) of vectors carrying the parasite, further solidifying the risk of Chagas disease transmission in the region. Prior case reports have suggested the potential for domestic transmission along the eastern side of the Texas-Mexico border [5, 20], and now our larger regional assessment confirms this risk. Compounding evidence of poverty, substandard housing, rural residential exposure to sylvatic animals, and high infection prevalence of multiple species all can contribute to an increased risk of Chagas disease transmission to local residents [10, 14, 22].
Coyotes (Canis latrans) are den dwelling animals native to North America. Habitat preferences include caves and natural holes, or abandoned domestic structures such as drainage pipes, vacant homesteads and railroad tracks [23]. Similarly, triatomine vectors prefer natural or domestic habitats, living in large numbers within dens that provide constant access to a host meal source [3]. Our finding of 8% seroprevalence among coyote populations in the Rio Grande Valley is slightly lower than a prior study in 1978 which found a 12.8% (20 out of 156) prevalence of infection [20]. A second study published in 1984 found a 14% seroprevalence rate in coyotes from across Texas; however, none of the eastern Rio Grande Valley counties were included in this sampling [24]. Tennessee, Georgia, and Virginia are other southern states with known T. cruzi positive coyote populations [25–27]. Comparable to our study, these more recent studies found seroprevalence rates between 7–10%, suggesting that infection rates might be decreasing with time or current diagnostic tests have better sensitivity-specificity.
Dog (Canis lupus familiaris) populations in the United States can be feral or domesticated; however, both groups can serve as bridge hosts for transferring Chagas disease between sylvatic environments and humans. Dogs serve as important sentinel for disease surveillance purposes as their infection rates can be early predictors of transmission risk to humans, especially considering dogs develop clinical cardiac disease quicker than humans [5, 21, 28–30]. Using public health veterinary shelters as a sampling venue is a convenient methodology to capture feral, community-owned, and domesticated dog populations. The shelter dogs in our study of the Rio Grande Valley had a seroprevalence of 3.8%, which is considerably lower than other published infection prevalence estimates among shelter dog populations from across the state. Over 48 different dog breeds in Texas have demonstrated natural infection with T. cruzi, with prevalence estimates ranging from 8.8–20.3% [31, 32]. In the greater Brownsville, Texas area, infection prevalence of shelter dogs has ranged from 7.5% in 2003 to 6.7% in 2014 [5, 32]. While our prevalence is slightly lower than other studies, the reason is most likely related to our decision to sample dogs that were under 6 months of age, allowing us to estimate incidence related to recent vector-borne or congenitally-acquired infection. By estimating incidence, we can better understand the annual contribution of disease transmission in this geographic area.
The epidemiology and seroprevalence of human infection in the southern United States is largely unknown. Even in endemic areas, human seroprevalence is typically lower than sylvatic and domestic animals due to multiple factors, including increased mammalian-vector habitat exposure, mammalian predilection for oral ingestion of the triatomine vector, and varying defecation behaviors of different triatomine species [3, 30, 33]. While sylvatic transmission cycles between wildlife and vectors have been established in the southern United States, we are still in our infancy of understanding disease burden and transmission source in infected populations. A prior study conducted in 1977 found a seroprevalence of 2.4% (12 out of 500) among eastern Rio Grande Valley residents [20], which is a sharp contrast to our finding of 0.4% (3 out of 841). Our study sampling included random selection of participants, while their study biased their results by recruiting patients at Texas Chest Hospital in Harlingen. It is likely our sampling methodologies influenced the varying rates, especially as other historical random-selection population studies reported 0.01–0.9% seroprevalence [12]. Despite our selection methodology differences, both Burkholder et al.’s study and ours included long-time residents of the Rio Grande Valley, with one positive participant in our study very likely acquiring the infection in Texas. Based on our findings of a seroprevalence estimate of 0.4%, and considering a population of 1.3 million for the Rio Grande Valley, we can estimate that ~4,600 people in this region are currently infected with Chagas, with ~1,300 at risk for developing Chagas-related cardiomyopathy. If this estimate is accurate, then the burden of Chagas disease in the Rio Grande Valley is 23 times higher than what we had previously estimated based on our findings of 1 out of 6,500 (0.02%) blood donors in Texas testing positive for the disease [34]. Future studies should aim to further clarify the true disease burden and rate of autochthonous transmission in the Rio Grande Valley, an area with documented sylvatic and domestic T. cruzi transmission [5].
Our study had a few important limitations notable for discussion. The current World Health Organization guidelines require a minimum of two positive results on different antibody-based assays for diagnostic confirmation [35]. While we used two different assays, neither are currently FDA approved in the United States; however, Stat-Pak rapid immunochromatographic assay has demonstrated efficacy in all three populations of mammals in multiple studies [6, 16–18, 27]. For the purposes of this retrospective study we felt confident in the test results, especially as they were relatively consistent with other published literature. In addition to our finding of a high rate of infection (56.5%) among local vector species, other studies have also confirmed high rates of infection (51–82%) in Triatomine vectors throughout Texas [7, 8, 10]. Provided the retrospective nature of our study, the obvious lack of travel history in these coyote and dog populations, and the establishment of known T. cruzi positive vector populations in our study, we would argue that these are true infections acquired via local vector-sylvatic mammal transmission cycles. Another possible limitation, due to our retrospective sampling of frozen sera collected 8–10 years prior, is the potential for antibody decay resulting in a lower prevalence rate. Handling of the specimens included freezing aliquots to -80°C immediately following collection, constant monitoring of freezer temperature, and adhering to discipline standards during the serum thawing process in an effort to maintain sample preservation. Finally, we cannot rule-out the potential for cross-reaction with leishmaniasis. Rare reports of cutaneous leishmaniasis have been reported in the state [36]; however, none of our three Chagas-positive study participants presented with skin ulcers, lowering the potential for cross-reaction.
In conclusion, we contribute to the growing body of evidence for autochthonous Chagas disease transmission among mammals in south Texas. Coyotes, shelter dogs, and vectors in this region continue to demonstrate high infection rates of T. cruzi. Random sampling of residents also revealed a higher than expected disease burden that had previously been undiagnosed, with one human patient suspected of having locally acquired the disease. With up to 30% of infected individuals developing a potentially fatal cardiac disease, it is imperative that we identify and treat patients before irreversible clinical manifestations have occurred. Future prospective studies are necessary to elucidate and validate the disease burden in the Rio Grande Valley.
|
10.1371/journal.ppat.1000972 | Identification of GBV-D, a Novel GB-like Flavivirus from Old World Frugivorous Bats (Pteropus giganteus) in Bangladesh | Bats are reservoirs for a wide range of zoonotic agents including lyssa-, henipah-, SARS-like corona-, Marburg-, Ebola-, and astroviruses. In an effort to survey for the presence of other infectious agents, known and unknown, we screened sera from 16 Pteropus giganteus bats from Faridpur, Bangladesh, using high-throughput pyrosequencing. Sequence analyses indicated the presence of a previously undescribed virus that has approximately 50% identity at the amino acid level to GB virus A and C (GBV-A and -C). Viral nucleic acid was present in 5 of 98 sera (5%) from a single colony of free-ranging bats. Infection was not associated with evidence of hepatitis or hepatic dysfunction. Phylogenetic analysis indicates that this first GBV-like flavivirus reported in bats constitutes a distinct species within the Flaviviridae family and is ancestral to the GBV-A and -C virus clades.
| Bats are important reservoirs for emerging zoonotic viruses with significant impact on human health including lyssaviruses, filoviruses, henipaviruses and coronaviruses. Opportunities for transmission to humans are particularly prominent in countries like Bangladesh, where people live in close association with bats. Whereas previous studies of bats have employed assays that test for known pathogens, we present the first application of an unbiased molecular approach to pathogen discovery in this reservoir for emerging zoonotic disease. Unbiased pyrosequencing of serum from Pteropus giganteus bats enabled identification of a novel flavivirus related to Hepatitis C and GB viruses. Viral nucleic acid was present in 5 of 98 (5%) sera, and in the saliva of one animal. Sequence identification of two strains of the virus, tentatively named GBV-D, suggests P. giganteus as a natural reservoir. Detection of viral nucleic acid in saliva provides a plausible route for zoonotic transmission. Phylogenetic analysis indicates that GBV-D is ancestral to GBV-A and -C, and separate from the recently classified genus Hepacivirus. Our findings provide new insight into the range of known hosts for GB-like viruses and demonstrate the power of unbiased sequencing to characterize the diversity of potentially zoonotic pathogens carried by bats and other reservoirs.
| Bats (order Chiroptera), after rodents, comprise the most diverse group of mammals with more than 1,100 species. They are present on six continents, often have substantial habitat overlap with humans [1] and harbor several zoonotic viruses causing significant human morbidity and mortality, including Ebola- and Marburgvirus, Nipah virus (NiV), and SARS-like coronaviruses [2]–[5]. Proximity of bats to human populations may facilitate the zoonotic transmission of viruses either through direct contact, via amplifying domestic animal hosts, or through food-borne routes [6]–[8].
The current study was set up as part of a viral discovery effort to target key wildlife reservoirs in emerging disease hotspots. Bangladesh is a ‘hotspot’ for emerging zoonotic diseases [9], with a relatively high diversity of wildlife that likely harbors new zoonotic pathogens, one of the densest human populations on the planet, and a high level of connectivity between people, domestic animals and wildlife. In Bangladesh and India, frugivorous Pteropus giganteus bats have been identified as a reservoir for NiV [10], [11], which has been recognized as the cause of several outbreaks of encephalitis [12]–[14]. Pteropus giganteus bats are common throughout the Indian subcontinent, living in close association with humans and feeding on cultivated fruit [14]. NiV transmission from bats to humans has been linked with the harvest and consumption of raw date palm sap, which becomes contaminated with bat feces, urine or saliva overnight when bats such as P. giganteus come to feed from the collecting pots [14], [15]. Date palm sap or other foods eaten by both bats and people, may also serve as a vehicle for transmission of other bat-borne agents.
Several zoonotic flaviviruses, including Japanese encephalitis virus, West Nile virus, and Kyasanur forest virus have been identified in bats; however, to date, GB viruses have not [1]. GB viruses A and C (GBV-A and -C) represent two recently identified species that are currently unassigned members of the family Flaviviridae [16]. GBV-A viruses have been described in New World primates and are not known to infect humans [17]–[19], while GBV-C (also known as Hepatitis G virus (HGV)) have frequently been isolated from humans in many regions of the World, including India and Bangladesh [19]–[23], and from wild chimpanzees (Pan troglodytes) in Africa [24], [25]. Here we describe discovery of a virus in the serum of healthy bats in Bangladesh, tentatively named GB virus D (GBV-D), that is distantly related to GBV-A and -C and represents a new member of the family Flaviviridae.
Every effort was made to minimize bat stress and avoid injury during capture, restraint, and sampling procedures. This study was conducted following Wildlife Trust institutional guidelines under IACUC approval G2907 issued by Tufts New England Medical Center, Boston, Massachusetts.
As part of a longitudinal surveillance study of Nipah virus in bats, 98 free-ranging P. giganteus bats were caught from a colony of approximately 1800 individuals in the Faridpur district of Bangladesh in December 2007 (Figure 1). Each bat was anesthetized using isoflurane gas; morphometric measurements (weight, forearm length, head length, and body condition) were taken and bats were aged [10]. Each bat was marked for future identification using an RFID microchip (AVID corp, www.avidid.com) implanted subcutaneously between the scapulae. Three mL of blood were collected and placed into serum separator tubes (vacutainer; Becton Dickinson, Franklin Lakes, NJ, USA). Serum was allowed to separate overnight at 4°C then drawn off without centrifugation and immediately frozen using a liquid nitrogen dry shipper. To inactivate potentially infectious agents, serum samples were heat-treated at 56°C for 30 min and then stored at −70°C. For RNA extraction, 250 µL of serum was added to 750 µL Tri-Reagent LS (Molecular Research Center, Cincinnati, OH, USA). Saliva was collected from the bat's throat using a sterile cotton swab. Urine was collected either by catching urine in a 1.0 mL sterile cryovial while the bat was urinating, or by urethral swab. Urine and saliva swabs were immediately placed into 1 mL Tri-Reagent LS and frozen in liquid nitrogen.
Total RNA from serum was extracted for UHTS analysis to screen for the presence of microorganisms. Five microliters of total RNA from each bat were combined into 4 pools: 4 pregnant bats; 4 non-pregnant female bats, and 2 pools of 4 adult male bats, respectively. Reverse transcription (RT) was performed on DNase I-treated (DNA-free, Ambion Inc., Austin, TX, USA) RNA pools to generate cDNA using Superscript II RT (Invitrogen, Carlsbad, CA, USA) and random octamers linked to a defined arbitrary, 17-mer primer sequence tail (MWG, Huntsville, AL, USA) [26]. After RNase H treatment cDNA was amplified by the polymerase chain reaction (PCR), applying a 9∶1 mixture of the defined 17-mer primer sequence and the random octamer-linked 17-mer primer sequence, respectively [27]. Products of >70 base pairs (bp) were selected by column purification (MinElute, Qiagen, Hilden, Germany) and ligated to specific linkers for sequencing on the 454 Genome Sequencer FLX (454 Life Sciences, Branford, CT, USA) without DNA fragmentation [28], [29]. Sequences were analyzed using software applications implemented at the GreenePortal website (http://tako.cpmc.columbia.edu/Tools/).
Multiple forward and reverse primers for RT-PCR (available upon request) were designed using the sequences obtained by UHTS in order to fill gaps between fragments. Amplifications were performed with Bio-X-act (Bioline, London, UK) according to manufacturer's protocols. Products were size fractionated by electrophoresis and directly sequenced in both directions with ABI PRISM Big Dye Terminator 1.1 Cycle Sequencing kits (Perkin-Elmer Applied Biosystems, Foster City, CA, USA) at a commercial facility (Genewiz, South Plainfield, NJ, USA). Additional methods applied to obtain the genome sequence included touch-down PCR [30], 2-step walking PCR [31], and 3′- and 5′- RACE (Invitrogen).
A real time Taqman PCR assay was developed to screen bat samples for GBV-D. Reactions were performed in a 25 µL volume by using commercial Taqman Universal Master Mix (Applied Biosystems, Foster City, CA, USA). Primers and probe were designed to target a 60 nt region in the NS4A gene region: Fadi-forward, 5′- gCAgCTgCgTgTgCCA; Fadi-reverse, 5′- ACACCCATgATgTTACCACgAC; Fadi-probe, 5′- FAM- AggACCCggTCgCTCCAgCA-T-BQX (TIB Molbiol, Adelphia, NJ, USA). Cycling conditions were: 50°C for 2 min, and 95°C for 10 min, followed by 45 cycles at 95°C for 15 sec and 60°C for 1 min. Thermal cycling was performed in an ABI 7300 real-time PCR system (Applied Biosystems).
A liver function panel was conducted at the International Center for Diarrheal Disease Research (Dhaka, Bangaldesh) using non heat-treated bat sera (Automated Chemistry Analyzer AU 640, Olympus Corporation, Tokyo, Japan). The following parameters were analyzed: total protein, albumin, globulin, albumin∶globulin ratio, total cholesterol, total bilirubin , alkaline phosphatase, alanine transferase, aspartate aminotransferase, gamma glutamyltransferase , and lactate dehydrogenase.
Sequence alignments were generated with ClustalW software [32] and phylogenetic relationships deduced using Geneious software [33]. Statistical significance was assessed by bootstrap re-sampling of 1000 pseudoreplicate data sets. Sequence relations were determined from p-distance matrices calculated with pairwise deletion for missing data and homogeneous patterns among lineages based on ClustalW alignments as implemented in MEGA software [34]. Sliding window similarity analysis was performed using SimPlot [35]. Potential signalase cleavage sites, glycosylation sites, and phosphorylation sites were analyzed using the respective prediction servers available at the Center for Biological Sequence Analysis (http://www.cbs.dtu.dk/services/).
Total RNA from the serum of healthy bats captured at a roost in the Faridpur district of Bangladesh was extracted for UHTS analysis. Extracts of 16 individual bats were combined into 4 pools consisting of 4 pregnant adult bats, 4 non-pregnant adult female bats, or 2×4 adult male bats. Each pool yielded between 1,400 and 2,000 assembled contigs or singlton reads (representing 50,000–75,000 reads ranging in size from 31–328 nt). Two reads of 238 and 215 nucleotides (nt) derived from the pregnant bat pool had distant homology to GBV-A sequences at the deduced amino acid (aa) level in the E2 and NS4A gene regions respectively (BLASTX); no homology was detected by searches at the nt level (BLASTN; local copy of the executables with standard settings except that the reward for a nucleotide match was set to 2 instead of 1). No viral sequences were detected in other pools at the nt or aa levels. Screening of the individual RNA preparations from the pregnant bat pool using primers derived from the UHTS reads confirmed the presence of the GBV-like sequence in the serum of bat 93. A quantitative real time PCR assay indicated a load of approximately 30 000 RNA copies in bat-93 serum extract, and identified an additional 4 positive bat sera from the original 98 samples (5/98; 5%), indicating serum loads ranging from 350 to 70,000 RNA copies per assay. These positive samples came from male bats that were not included in the initial UHTS pools. Extracts of saliva from the five positive bats indicated a load of approximately 200 RNA copies in bat 93; no signal was obtained with urine extracts from the five positive bats.
Near full-length genome sequence was generated from bat-93 and a second positive serum (bat 68), applying primers crossing gaps between UHTS reads as well as touch-down PCR [30], 2-step walking PCR [31], and 3′- and 5′-RACE (Invitrogen) protocols. The two genome sequences were 96% identical at the nt level (GenBank Accession nos. GU566734 and GU566735), indicating two strains of the same virus. Comparison of deduced polyprotein sequence to other GBV and hepaciviruses indicated highest nt and aa sequence identities to GBV-A and -C (Table 1, Figure 2). The genomic sequence of the GBV-like virus identified in P. giganteus bats, tentatively named GBV-D, comprises 9,633 nt with 52 nt of potentially 5′-untranslated region (UTR), one continuous open reading frame (ORF) of 9318 nt (3106 aa) and 265 nt of 3′-UTR (Figure 3).
Mature structural proteins in GB viruses, as well as other flaviviruses, are the product of cleavage by host signal peptidase [36]. In GBV-D the first potential signal sequence cleavage site is present after a stretch of 57, largely basic aa (6 kDa, pI = 12), followed by sequence homologous to E1 (pfam 01539, http://pfam.sanger.ac.uk/) (Figure 3). The single glycosylation site N177IT present in that sequence is located in a position comparable to GBV-C, -A, -B and HCV glycosylation sites. Identification of the downstream E2 termini is less apparent as the next 580 aa contain multiple potential signal sequences and 10 potential glycosylation sites that indicate no homology to hepaciviral E2/NS1 (pfam 01560), until the sequence aligns with N-terminal NS2 motifs (pfam 01538) (Figure 2, Figure 3). However, despite similarity to pfam 01538 no signal sequence compatible with cleavage at A759/A was found; cleavage may occur at G826/R, which combined with potential signalase cleavage at A584/F may indicate the existence of a heavily glycosylated potential 26 kDa product instead of the p7 trans-membrane protein identified in HCV [37]–[39] or the 13 kDa variant described in GBV-B [40], [41]. Conserved C-terminal motifs of the autocatalytic NS2/NS3 endoprotease domain are compatible with NS2/NS3 cleavage at S1067/A and comparable to other GBV and HCV [42]. Figure 3 indicates potential cleavage sites for NS3 (peptidase S29, pfam 02907; DEAD box helicase, pfam 07652; helicase C, pfam 00271), NS4A (pfam 01006), NS4B (pfam 01001), NS5A (domain-1a zinc finger, pfam 08300; domain-1b, pfam 08301), and NS5B (pfam 00998).
Conserved aa motifs were recognized in NS proteins. RNA-dependent RNA polymerase (RdRp) motifs in RdRp block III that are conserved with respect to other GBV and hepaciviruses were identified in NS5B (Figure 3) [43]–[46]. Potential phosphorylation sites are present at multiple serine (9), threonine (14) and tyrosine (4) residues in NS5A, compatible with its possible function as a phosphorylation-regulated mediator of viral replication [47]. However, significant conservation of primary sequence is not obvious for phosphorylation sites, proline-rich, or interferon-sensitivity determining region motifs [48]–[50]. The C-terminal portion of NS3 has homology to conserved NTPase/helicase motifs [51]; the N-terminal portion includes conserved active triad residues H1123, D1147, S1204 of serine protease [52], the viral protease responsible for cleavage of mature non-structural proteins [53]. Likewise, the active triad H991, E1011, C1032 of the cis-acting protease activity in the C-terminal portion of NS2 is conserved with respect to other GBV and HCV [42]. The only other discernable motif identified was a well-conserved N75 C/D C motif at the N-terminus of E1 (Figure 3) [54].
Phylogenetic analysis of GBV-D was performed in comparison to selected representatives of GBV-A, GBV-B, GBV-C and HCV. Analysis of NS5B aa sequence (Figure 4A) confirmed a closer relationship of GBV-D to GBV-A and -C than to GBV-B or HCV as also indicated by pairwise sequence comparisons (Table 1). The same relationships were also apparent when NS3, or the complete polyprotein sequence were analyzed (Figure 4B and C, respectively). All three trees show GBV-D consistently at the root of the GBV-A/-C viruses, indicating an independent phylogenetic clade compatible with a separate species distinct from the recently created genus Hepacivirus [16].
A liver serum chemistry panel was conducted on sera from 15 bats, the five GBV-D infected and 10 non-infected animals. Standard assays to detect hepatitis and/or impaired liver function were performed [55]. Levels of total protein, alanine transferase, aspartate aminotransferase and total cholesterol were within published ranges reported for P. giganteus, except for bat 33 (infected) and bat 73 (uninfected), which had modest elevation in aspartate aminotransferase. Reference values for albumin, globulin, albumin∶globulin ratio, total bilirubin, alkaline phosphatase, gamma glutamyltransferase and lactate dehydrogenase are not available for P. giganteus, however, values were comparable to those reported for other Pteropus species [56]. Mean values did not significantly differ between infected and uninfected bats (Table 2).
Molecular analyses of sera from Pteropus giganteus bats from Faridpur, Bangladesh led to the identification of a 9,633 nt sequence consistent in genomic organization with known GBV and other species within the family Flaviviridae [16]. Whereas previous studies of bats have employed assays that test for known pathogens, ours is the first report of an unbiased molecular approach to pathogen discovery in this important reservoir of emerging infectious diseases. The modest yield of novel microbial sequences may reflect the choice of sample (e.g., serum vs feces, tissue or another specimen), competition between host and microbial template during unbiased amplification, or both. Efforts to address template competition are under way that include subtraction of host nucleic acids or the use of semi-random primers that do not amplify host sequences. Such efforts will likely enhance the sensitivity and throughput of unbiased sequencing technologies for pathogen discovery.
The discovery of this chiropteran flavivirus broadens both the taxonomical and geographical distribution of GB-like viruses. Three types of GB viruses have been described: GBV-A, -B and -C [18], [19], [24], [25], [54], [57]. GBV-B, which has never been found in humans and was only reported in captive tamarins after serial passage of the original human GB serum [58], is most closely related to HCV and was recently classified together with HCV into a new genus, Hepacivirus, within the family Flaviviridae [16]. GBV-A and -C remain unclassified members of the family. GBV-A have been isolated from several New World monkeys. Different genotypes appear to be associated with specific monkey species of the genera Saguinus, Callithrix (Callitrichidae family) and Aotus (Aotidae family), without any clinical signs associated with infection [24], [54], [57]. GBV-C have been isolated from humans with non-A-E hepatitis; however, its pathogenicity is unknown and the virus is widespread in the human population [21], [59]–[61]. Population studies showed that GB viruses are enzoonotic and species-specific within both Old and New World nonhuman primates as well as humans, and have likely co-evolved with their hosts over long periods of time [62]. Previously, the only GBV found in the Old world was GBV-C from chimpanzees (in Africa) and humans. Although GBV-C were found in humans, GB viruses have not been previously reported in primates or other animals on the Indian subcontinent.
GBV-C and -A are remarkable for a truncated or missing capsid (C) protein [18], [19]. Due to exhaustion of our samples we were unable to complete assessment of the 5′-terminal sequence; nonetheless, RACE experiments suggest that GBV-D likely codes for a short basic peptide, instead of a full-length C protein. The first methionine (M1) predicts a peptide of 57 aa (pI = 12); however, the more favorable Kozak context [63] of M3 indicates a 55 aa peptide. After signalase cleavage from the polyprotein precursor, this peptide may be functional, possibly influencing maturation of, or directly binding to, the E1 and/or E2 glycoproteins.
Phylogenetic analyses of NS5B, NS3 and complete polyprotein sequence place GBV-D at the root of the GBV-A and -C clades and are consistent with a model wherein GBV-D is ancestral to GBV-A and -C clades. Mixed relationships indicative of recombination events [64] were not evident (Figure 2, Figure 4). Both pteropid bats and chimpanzees are restricted to the Old World. While the range of chimpanzees (Africa) and P. giganteus (the Indian subcontinent) do not overlap, it is possible that other primate species in Bangladesh or India, such as macaques, or other fruit bats in Africa such as Eidelon spp., whose range overlaps that of chimpanzees, may carry related viruses. While GBV-A is only known from primates of the New World, an African origin has been suggested for GBV-C based on a 12-aa indel sequence in NS5A [65]. Although the NS5A sequence of GBV-D, similar to that of GBV-A, appears elongated in the indel region, compatible with their respective earlier phylogenetic branching compared to GBV-C, little sequence conservation is observed in that region.
The bats in this study, like primates infected with their associated GBV [66], all appeared to be healthy. The lack of chemical evidence of hepatic inflammation or dysfunction suggests that this virus may not target hepatic cells in bats. This is consistent with the behavior of GBV-A in its natural primate hosts [54]. In contrast, elevated alanine transferase levels and mild hepatitis are observed in experimental infections of macaques with GBV-C isolates from humans [67]. Five percent of the bats we studied were infected with one of at least two different strains of GBV-D, which suggests widespread viral circulation within this species. The observation that bats are asymptomatically infected with diverse strains that constitute a distinct phylogenetic clade is compatible with a co-evolutionary relationship between GBV and their hosts [57], [62], and supports the hypothesis that P. giganteus bats may be a natural reservoir for GBV-D. In one case we were able to detect GBV-D nucleic acid in saliva. This suggests a potential route for viral transmission via fighting or grooming behavior, or via food shared by bats.
Pteropus giganteus is a frugivorous bat species that carries NiV, a zoonotic paramyxovirus [10], [11]. This species lives in close association with humans in Bangladesh and bats have been observed drinking from (and urinating into) date palm sap collecting pots [14]. Human consumption of contaminated palm juice is proposed to be a major route of NiV transmission [68]. Although it is unclear whether infectious virus was present in bat saliva, the observation that saliva can contain GBV-D nucleic acids provides a biologically plausible mechanism for transmission from infected bats to other hosts. While it is currently unknown whether GBV-D virus occurs in humans, up to 20% of non-A-E hepatitis cases remain unexplained [19].
|
10.1371/journal.pgen.1007827 | Epigenetic inheritance of telomere length in wild birds | Telomere length (TL) predicts health and survival across taxa. Variation in TL between individuals is thought to be largely of genetic origin, but telomere inheritance is unusual, because zygotes already express a TL phenotype, the TL of the parental gametes. Offspring TL changes with paternal age in many species including humans, presumably through age-related TL changes in sperm, suggesting an epigenetic inheritance mechanism. However, present evidence is based on cross-sectional analyses, and age at reproduction is confounded with between-father variation in TL. Furthermore, the quantitative importance of epigenetic TL inheritance is unknown. Using longitudinal data of free-living jackdaws Corvus monedula, we show that erythrocyte TL of subsequent offspring decreases with parental age within individual fathers, but not mothers. By cross-fostering eggs, we confirmed the paternal age effect to be independent of paternal age dependent care. Epigenetic inheritance accounted for a minimum of 34% of the variance in offspring TL that was explained by paternal TL. This is a minimum estimate, because it ignores the epigenetic component in paternal TL variation and sperm TL heterogeneity within ejaculates. Our results indicate an important epigenetic component in the heritability of TL with potential consequences for offspring fitness prospects.
| Telomeres are DNA-protein structures at chromosome ends and a short telomere length predicts reduced survival in humans, birds and other organisms. Variation in telomere length between individuals is thought to be largely of genetic origin, but telomere inheritance may be unusual because not only genes regulating telomere length are inherited, but a fertilised cell already has a telomere length (from the parental gametes). Using long-term individual-based data of jackdaw families (a small corvid species), we found that as fathers aged, they produced chicks with shorter telomeres. This suggests that telomere length inheritance has an epigenetic component. To investigate to what extent telomere length in the fertilised cell affects telomere length after birth, we compared telomere length over years within fathers with the telomere length of their consecutive offspring. This epigenetic component explained a substantial part (≥ one third) of the telomere length inheritance; whereas there was no such effect of maternal telomere length. The sex difference fits the idea that lifelong sperm formation leads to change in telomere length of the sperm cells, whereas female gametes are all formed before birth and their telomere length does not change over time.
| Telomeres are evolutionarily conserved DNA sequence repeats, which form the ends of chromosomes together with associated proteins and contribute to genome stability [1]. Telomeres shorten due to incomplete replication during cell division, which can be accelerated by DNA and protein damaging factors and attenuated or counter-acted by maintenance processes, mainly based on telomerase activity, a telomere-elongating ribonucleoprotein [2]. On the organismal level, telomere length (TL) generally declines with age and short TL relates to ageing-associated disorders and reduced survival in humans [3,4] and other organisms [5,6]. Given this relationship of telomeres with health and lifespan it is of importance to understand how variation in TL among individuals arises, which is already present early in life [7–9].
TL has a genetic basis, but heritability estimates for TL are highly variable [10]. Compared with other traits, inheritance of TL is also unusual in that the TL phenotype is directly expressed in the zygote without any effect of its own genome. This is because the zygote’s set of chromosomes carries the telomeres of the two parental gametes. Subsequently, during development of the embryo, different telomere maintenance and restoration mechanisms, under the control of multiple genes, potentially regulate TL, but this process is poorly understood [11]. In the course of early development, such mechanisms can potentially compensate fully for gamete derived differences in TL (as suggested by e.g. [12,13]), in which case the effect of gamete TL is transient (Fig 1A). Alternatively, differences in gamete TL are carried over to later life (as suggested by e.g. [14]; Fig 1B). The latter case would imply the inheritance of parental TL, which is independent of DNA sequence variation (in vertebrates (TTAGGG)n [15]), but a change in telomere sequence length (n). We interpret this as a form of epigenetic inheritance component on TL [16,17]. Note that this epigenetic inheritance mechanism differs from better known epigenetic mechanisms such as DNA methylation in that it does not affect the phenotype (TL) by modulating gene expression, but instead through direct inheritance of the phenotype itself and therefore has also been referred to as “epigenetic-like” [17].
Strongest evidence for an epigenetic mechanism of TL inheritance comes from studies that show a relationship between parental (usually paternal) age and offspring TL [18–23] with a particularly interesting example showing a cumulative effect over generations in humans [24]. In humans, where offspring TL increases with paternal age, this trend parallels a qualitatively similar change in sperm TL with age, which is generally assumed to underlie the TL increase in offspring [25]. However, studies of parental age effects in other species show mixed results and trends differ in direction between and within taxa [20,26]. More importantly, some critical uncertainties remain unresolved in any species. Firstly, studies to date are all cross-sectional [18–23], thus, comparing offspring of different parents that reproduced at different ages. Such cross-sectional trends may differ from age related changes within individual parents if, for example, individuals with long TL are more likely to reproduce at older ages, which is not unlikely given the positive correlation between human TL and reproductive lifespan [27,28].
Secondly, parental age effects on offspring TL may arise from effects of parental age on pre- and postnatal conditions prior to sampling. Because telomere attrition is highest early in life e.g. [29,30], these effects can be substantial, as illustrated by parental age effects on TL dynamics during the nestling phase in European shags Phalacrocorax aristotelis [31] and Alpine swifts Apus melba [21]. Lastly, due to their cross-sectional character, studies to date could not test whether changes in TL within parents over their lifetimes are predictive of changes in TL of the offspring in relation to parental age at conception. These points need to be resolved to establish whether the correlations between parental age and offspring TL can be attributed to epigenetic inheritance of TL, and before we can begin to understand why parental age effects on offspring TL appear to differ between and within taxa [20,26].
To investigate whether offspring TL changes with parental age at conception over the lifetime of individual parents we used our long-term, individual-based dataset of free-living jackdaws Corvus monedula. Telomere length was measured in nucleated erythrocytes using terminal telomere restriction fragment analysis [32] from multiple chicks of the same parents that hatched up to 9 years apart. As telomere attrition is highest early in life, we took blood samples for telomere analysis shortly after hatching, when the oldest chick in a brood was 4 days old. To test if TL was influenced by age-dependent parental care prior to sampling, we cross-fostered clutches between nests immediately after laying and tested whether foster parent age affected offspring TL. To investigate if the rate of telomere attrition within parents predicts the change in TL of the offspring they produce over consecutive years, we measured TL of the parents repeatedly over their lifetimes.
For the first time, we here show that offspring TL declines as individual fathers age and that the change in TL over time in fathers is reflected in the TL of their offspring, which explains a substantial part of the telomere resemblance between fathers and offspring and can be interpreted as an epigenetic component in the inheritance of TL. Mother offspring resemblance on the other hand was independent of maternal age and within mother variation in TL was not associated with variation in the TL of her offspring.
Descriptive data of the study population are summarised in Table 1.
To be able to separately evaluate between- and within-individual patterns of parental age, we used within-subject centering [33]. Instead of using age in our models, we used the mean age per individual over multiple years as one variable, and delta age, the deviation from that mean as a second variable. Thus, the coefficient of mean age estimates the parental age effect compared between individual parents, while the coefficient of delta age estimates the age effect on offspring TL within parents. As fathers aged, they produced offspring with 56±20 bp shorter TL for each additional year (variable ‘delta age father’ in Table 2A, Fig 2A), showing that offspring TL declined with paternal age at conception within individual males. This effect was not apparent when comparing offspring of different fathers reproducing at different ages (cross-sectional component of the statistical model, variable ‘mean age father’ in Table 2A). In contrast, there was no effect of maternal age on offspring TL (Table 2B), neither when compared cross-sectionally, between offspring of different mothers over age (mean age mother, Table 2B), nor within mothers as they age (delta age mother). The negative, non-significant effect of maternal age on offspring telomere length we observed (Table 2B) we attribute to the age of their mates, because pair bonds in jackdaws are maintained over many years (pers. obs.) and hence maternal and paternal age are correlated. This interpretation is confirmed by the finding that the observed maternal age estimate (delta age mother) is close to what would be expected based on the estimate found in fathers and the observed correlation of r = 0.75 (n = 298) between maternal and paternal age (i.e. 0.75 * 56 bp = 42 bp, which is very close to the estimate ± s.e. for delta mother age, which was 38±23 bp; Table 2; see also [23]). Thus, we conclude a maternal age effect on offspring TL other than through the age of the females’ mates to be unlikely. The decline in offspring TL with fathers’ age was lower than the rate of TL attrition in the fathers themselves (-56±20 versus -87±15 bp/year, respectively). Individual variation in telomere attrition slopes was negligible both between individual fathers (additional variance explained by random slopes 1%) and in their offspring produced across the fathers’ lifetimes as well (variance explained by random slopes 0.3%).
The paternal age effect on offspring TL could potentially be caused by age-dependent paternal care (e.g. age-related feeding of the incubating partner, or the chicks prior to sampling), if this affects telomere dynamics between conception and the sampling age of 4 days. We tested this hypothesis by exchanging clutches between pairs shortly after clutch completion. Our analysis is based on telomere data of 61 chicks that hatched from 31 cross-fostered clutches. In a first test, we added the age of foster father or mother to the model in Table 2A, and neither parental age significantly affected offspring TL (age foster father: 5.6±28.5, p = 0.85; age foster mother: -9.7±17.4, p = 0.58). To avoid basing a conclusion solely on a negative statistical result, in a second analysis we compared the estimate of the age of the caring father (i.e. the genetic father if not cross-fostered) on offspring TL with the estimate of the age difference between genetic and foster father (which is 0 in case of no cross-fostering or matching ages between genetic and foster father) on offspring TL. Both estimates were negative and very similar (Table 3, Fig 2B). Because the age of the caring father and the age difference between the caring father and the genetic father add up to the age of the genetic father for the cross-fostered offspring, the similarity of the estimates implies that there was no effect of age-related care between conception and sampling on offspring TL. While the estimate of the age difference did not quite reach statistical significance in a two-tailed test (p<0.09), we consider the similarity of the estimates (10% difference) the more salient result. Thus, the older the father, the shorter the TL of his offspring, independent of the age of the male that cares for the eggs and offspring up to sampling. These results show that the paternal age effect on offspring TL is explained by the age of the genetic father and that the influence of the age of the foster fathers on offspring TL at age 4 days is negligible.
The paternal age effect on offspring TL raises the question whether changes in paternal TL with age predict the change in early life TL of the offspring produced over the fathers’ lifetimes. We tested this by replacing the two age terms in the model in Table 2 by TL at conception (i.e. mean and delta TL) of the father in the year the offspring hatched. Fathers’ mean TL as well as delta TL were strongly and positively correlated with offspring TL (Table 4A, Fig 3). The effect of father’s mean TL on offspring TL can be attributed to additive genetic inheritance, possibly augmented by effects of a shared environment [10]. The effect of fathers’ delta TL on offspring TL cannot be attributed to a genetic effect, because delta TL refers to variation of TL within fathers over their lifetime. We therefore consider an epigenetic effect the most likely explanation for the effect of fathers’ delta TL on offspring TL. The variance in offspring TL explained by mean and delta TL of the father was 1.87 and 0.96 respectively. This indicates that 34% (0.96 / 2.83) of the variance in offspring TL that was explained by paternal TL can be attributed to the paternal-age related epigenetic effect. In agreement with our finding that maternal age did not affect offspring TL, when we performed the same analysis for mother TL, we found that maternal TL shortening (delta TL mother) was not related to the TL of her subsequent offspring, with a slope of the variable delta maternal age that was more than 90% lower than the comparable slope in males (Table 4B). However, mean maternal TL, reflecting a similarity between maternal and offspring TL per se (independent of a maternal TL change over time), based on a combination of additive genetic and age-independent maternal effects, was highly significant (Table 4B).
Resemblance of TL between parents and offspring is potentially due to a dual inheritance mechanism, with on the one hand a ‘classic’ additive genetic effect and on the other hand an epigenetic effect of variation in TL in the gametes that at least in part carries through to later life (Fig 1). Suggestive evidence for an epigenetic contribution to the inheritance of TL comes from studies showing a paternal age effect on offspring TL, but available results are based on cross-sectional analyses [18–23]. Using a unique longitudinal dataset on free-living birds, and a high precision TL measurement technique (CV within individuals <3%), we show for the first time that offspring TL changes with age within individual fathers (i.e. longitudinally). We used a cross-foster experiment to test whether the paternal age effect may be due to paternal age-dependent parental care prior to offspring sampling. This showed that the paternal age effect is already present at laying. Mother age was not significantly associated with offspring TL, and the non-significant estimate of the maternal age effect matched almost exactly the expected estimate based on the observed paternal age effect in combination with the correlation between the ages of pair members. Thus, we conclude that offspring TL declined with parental age within individual fathers, but not mothers.
The parental sex dependent age effect on offspring TL is in agreement with most other studies [18–23,25,34], and is usually attributed to the different replicative history of the gametes of the two sexes. Male gametes are newly formed throughout reproductive life, while a female’s complete stock of gametes is formed before birth [35,36]. Hence TL of female gametes is less prone to changes with female age compared to TL of male gametes [37–39, but 40]. This is not to say that there is no epigenetic inheritance of TL through the female line, but only that its contribution to offspring TL does not depend on mothers’ age.
While we consider epigenetic inheritance of TL via a carry-over effect from paternal gamete TL the most parsimonious explanation of our findings, we acknowledge that we cannot yet fully exclude other mechanisms. There is some scope for females to modulate the contents of their eggs, which may affect TL dynamics [41]. Thus, it remains a possibility that females adjusted the content of their eggs in response to the age of their partner in a way that causes the paternal age effect on the TL of their offspring. However, if there were such an effect, one would perhaps also expect it to be expressed in egg volume (which varies considerably in jackdaws), but there was no evidence that females adjusted the volume of their eggs to the age of their partner (p = 0.35, n = 683 clutches, model including female identity and year as random effects). Another mechanism we cannot rule out is paternal age dependent expression of genes that control telomere dynamics of offspring. However, genetic influences on telomere dynamics are modest compared to environmental influences or heritability of TL itself [42], making it unlikely that this hypothetical mechanism explains a substantial part of the paternal age effect.
We tentatively estimated the relative contributions of additive genetic and epigenetic effects to the resemblance between males and their offspring using a statistical model in which we separated between- and within-individual variation in parental TL as predictors of offspring TL. In this model, the within-male component (‘delta TL father’, Table 4A) shows the strong epigenetic effect over the years within males on their offspring, while the between male component (‘mean TL father’) shows the putative additive genetic effect on offspring TL. When comparing the relative contributions of the two inheritance mechanisms, it appeared that 34% of the variance explained by paternal TL can be attributed to the epigenetic effect. Telomere loss within mothers (‘delta TL mother’) was unrelated to the TL of offspring produced over years (Table 4B). Estimates of the between-male effect (0.26±0.08, ‘mean TL father’, Table 4A) and the between-female effect (0.46±0.09, ‘mean TL mother’, Table 4B) together equate to a narrow sense heritability of jackdaw TL of 0.72, which is similar to results observed in humans [43] and within the range observed in other vertebrates [10] and is in line with other studies on birds estimating higher similarity between mothers and offspring [44,45]. We stress however that we measured telomere length in parental blood and not in sperm and that the estimates for the additive genetic and the epigenetic effects are tentative. Firstly, with respect to the additive genetic effect, it is of importance that shared environment effects are not controlled for in the present analysis. We note however that a more extensive analysis using multigenerational pedigree information and controlling for shared environmental effects [46] yielded a very similar estimate of the narrow sense heritability of TL in our study population (Bauch et al. in prep). Secondly, the variance in TL between males is not only of genetic origin, given that in addition there appears to be an epigenetic contribution to the between-male variance. Thus, the effect of ‘mean TL father’ (Table 4A) will to an unknown extent contribute to the epigenetic effect, as well as heterogeneity of sperm TL in ejaculates. Hence the epigenetic contribution to the resemblance between father and offspring TL will be more than the 34% we estimated based on parent-offspring regression over a single generation.
Narrow sense heritability of human TL has been estimated using monozygotic and dizygotic twins [e.g. 47], assuming that a weaker resemblance between dizygotic twins compared to monozygotic twins can be attributed to the difference in genetic relatedness. However, as monozygotic twins develop from a single zygote, and hence from a single sperm cell and oocyte, the difference in resemblance within a monozygotic versus a dizygotic twin pair may in part be due to an epigenetic effect of having developed from the same or different gametes [14]. This process would lead to an overestimation of the narrow sense heritability compared to techniques that do not depend on twins.
The direction of the paternal age effect in jackdaws (decreasing) is opposite to the direction of the paternal age effect in humans and chimpanzees (increasing) [20]. Assuming that paternal age effects in humans and chimpanzees [20] on the one hand and several bird species (including our study species) [20–23] and lab mice [34] on the other hand all reflect paternal age effects on sperm TL, this raises the question why these age effects on sperm TL are in opposite directions. Seasonality of reproduction may well play a role, with species that produce sperm for a small part of the year having less need to maintain sperm TL than species with year-round sperm production [20]. The lengthening of TL in human sperm with age has been interpreted as the result of an overshoot in telomere maintenance [25] that can be viewed as a safety margin in the maintenance process. Such a safety margin can be expected to be larger when the rate of sperm production and hence telomere attrition is higher. This may explain why chimpanzees, with a higher sperm production rate than humans, due to their promiscuous mating system, show a steeper paternal age effect on offspring TL compared to humans [48]. Information on the sign of the association between paternal age and offspring TL in strongly seasonal mammal species and / or continuously reproducing bird species would allow a test of this hypothesis.
The epigenetic inheritance of TL potentially has more general implications. Parental age at conception has previously been shown to have negative effects on offspring fitness prospects in diverse taxa, a phenomenon known as the Lansing effect [22,49–52]. The underlying mechanisms are likely to be diverse, but in taxa where the paternal age effect on offspring TL is negative, given that TL predicts survival in wild vertebrates [6], and TL early in life correlates strongly with TL in adulthood in jackdaws [7], offspring born to older fathers may have a shorter life expectancy due to their epigenetically inherited shorter TL. A further implication is that there may be cumulative changes in TL over multiple generations [24]. This could lead to population level changes in TL when the age structure of the population changes, as has for example been observed in birds in response to urbanisation [53]. A population level change in TL may in itself have further demographic consequences [54], providing a positive or negative feedback, depending on whether increasing paternal age has a positive or negative effect on offspring TL.
Data were collected under license of the animal experimentation committee of the University of Groningen (Dierexperimenten Commissie, DEC, license numbers: 4071, 5871, 6832A). License was awarded in accordance with the Dutch national law on animal experimentation (“Wet op de dierproeven”) and research was carried out following the guidelines of the Association for the Study of Animal Behaviour (ASAB) [55].
Life-history data and blood samples originate from an individual-based long-term project on free-living jackdaws Corvus monedula breeding in nest boxes south of Groningen, the Netherlands (53.14° N, 6.64° E). Jackdaws produce one brood per year with mostly 4 or 5 chicks. They are philopatric breeders and socially monogamous with close to zero extra-pair paternity as shown in different populations [56,57]. Females incubate the eggs, while males feed their female partners. Chick provisioning is shared by the sexes. Each year, during the breeding season around the hatching date nest boxes were checked daily for chicks. Freshly hatched chicks were marked by clipping the tips of the toenails in specific combinations and therefore the exact ages of offspring were known. Between 2005 and 2016, 715 jackdaw chicks were blood sampled when the oldest chick(s) was (were) 4 days (note that chicks hatch asynchronously). These chicks originated from 298 nests, of 197 different fathers, whereof 66 were blood sampled repeatedly over years (max. difference of age between offspring 8 years) and 194 different mothers, whereof 62 were blood sampled repeatedly over years (max. difference of age between chicks 9 years; see Table 1 for more information). 61 chicks (that contributed telomere data) hatched from 31 cross-fostered nests, i.e. eggs were exchanged between nest boxes (selected for equal clutch sizes and laying dates (or up to one day difference), but otherwise randomly) soon after clutch completion. 54 (89%) of those chicks were fostered by a father of different age. Jackdaws in this project are marked with a unique colour ring combination and a metal ring. Parents were identified by (camera) observation during incubation and also later during chick rearing when caught for blood sampling (by puncturing the vena brachialis). Unringed adults were caught, ringed and assigned a minimum age of 2 years, as this is the modal recruitment age of breeders that fledged in our study colony. All jackdaws were of known sex (molecular sexing [58]).
Blood samples were first stored in 2% EDTA buffer at 4–7°C and within 3 weeks snap frozen in a 40% glycerol buffer for permanent storage at -80°C. Terminally located telomere lengths were measured in DNA from erythrocytes performing telomere restriction fragment analysis under non-denaturing conditions [29]. In brief, we removed the glycerol buffer, washed the cells and isolated DNA from 5 μl of erythrocytes using CHEF Genomic DNA Plug kit (Bio-Rad, Hercules, CA, USA). Cells in the agarose plugs were digested overnight with Proteinase K at 50°C. Half of a plug per sample was restricted simultaneously with HindIII (60 U), HinfI (30 U) and MspI (60 U) for ~18 h in NEB2 buffer (New England Biolabs Inc., Beverly, MA, USA). The restricted DNA was then separated by pulsed-field gel electrophoresis in a 0.8% agarose gel (Pulsed Field Certified Agarose, Bio-Rad) at 14°C for 24h, 3V/cm, initial switch time 0.5 s, final switch time 7.0 s. For size calibration, we added 32P-labelled size ladders (1kb DNA ladder, New England Biolabs Inc., Ipswich, MA, USA; DNA Molecular Weight Marker XV, Roche Diagnostics, Basel, Switzerland). Gels were dried (gel dryer, Bio-Rad, model 538) at room temperature and hybridized overnight at 37°C with a 32P-endlabelled oligonucleotide (5’-CCCTAA-3’)4 that binds to the single-strand overhang of telomeres of non-denatured DNA. Subsequently, unbound oligonucleotides were removed by washing the gel for 30 min at 37°C with 0.25x saline-sodium citrate buffer. The radioactive signal of the sample specific TL distribution was detected by a phosphor screen (MS, Perkin-Elmer Inc., Waltham, MA, USA), exposed overnight, and visualized using a phosphor imager (Cyclone Storage Phosphor System, Perkin-Elmer Inc.). We calculated average TL using ImageJ (v. 1.38x) as described by Salomons et al. [29]. In short, for each sample the limit at the side of the short telomeres of the distribution was lane-specifically set at the point of the lowest signal (i.e. background intensity). The limit on the side of the long telomeres of the distribution was set lane-specifically where the signal dropped below Y, where Y is the sum of the background intensity plus 10% of the difference between peak intensity and background intensity. We used the individual mean of the TL distribution for further analyses. Samples were run on 92 gels. Repeated samples of adults were run on the same gel, chicks were spread over different gels. The coefficient of variation of one control sample of a 30-day old jackdaw chick run on 26 gels was 6% and of one control sample of a goose, with a TL distribution within a similar range, run on 31 other gels was 7%. The within-individual coefficient of variation for samples run on the same gel was <3% [7] and the within-individual repeatability of TL was estimated to be 97% [59].
The relationships between parental age or parental TL and early-life TL of offspring were investigated in a linear mixed effects model framework using a restricted maximum-likelihood method (testing specific predictions). To be able to separately evaluate between- and within-individual patterns of parental age or parental TL, we used within-subject centering [33]. Thus, instead of father age, mother age or father TL, mother TL per se we introduced the mean value per individual over (if available) multiple years and delta age or delta TL, the deviation from that mean, respectively. To account for (genetic and potential other) similarities in TL between offspring of the same father or mother, we included father ID or mother ID as random effect in the model. As the dataset contains also siblings raised in the same nest, we additionally added a random effect of nest ID as a nested term in father ID or mother ID to the models investigating paternal or maternal age effects on TL, respectively. The age of chicks at sampling differed slightly (2–4 days) and as TL shortens with age [7], we included their age (in days) as a covariate. Offspring sex was never significant and was therefore excluded from the final models. We added gel ID as random effect. Analyses were performed separately for fathers and mothers as their ages are correlated.
The cross-foster experiment was designed to test for potential effects of parental age on early-life telomere attrition between egg laying and sampling (age 2–4 days). First, we modified the linear mixed effect model with offspring TL as dependent variable testing for paternal age effects (see above) by adding the age of the foster father or mother as covariate. Second, in a linear mixed model with offspring TL as dependent variable, we included both the age of the father caring for the clutch after cross-fostering and the age difference between the genetic father and foster father as covariates (age genetic father-age foster father). When the paternal age effect is independent of age-dependent effects between conception and sampling, we predict the coefficients of the caring father’s age and the age difference between genetic father and foster father to be indistinguishable. This is so because the age of the male caring for the clutch, and the age difference between the genetic and the caring father add up to the age of the genetic father. In contrast, when the paternal age effect is entirely due to age-dependent paternal effects after laying, the coefficient will be the same, but opposite in sign. In case of a mixture of the two effects, the coefficient will be intermediate. In this analysis we used all offspring, i.e. also those that were not cross-fostered, and further included genetic father ID, nest ID, gel ID and year of telomere analysis as random effects, and offspring age at sampling as covariate.
Statistics were performed using packages lme4 [60], lmerTest [61], MuMIn [62] in R (version 3.3.3) [63]. In the results mean ± standard error is given unless stated otherwise.
|
10.1371/journal.pcbi.1004487 | The Shape of an Auxin Pulse, and What It Tells Us about the Transport Mechanism | Auxin underlies many processes in plant development and physiology, and this makes it of prime importance to understand its movements through plant tissues. In stems and coleoptiles, classic experiments showed that the peak region of a pulse of radio-labelled auxin moves at a roughly constant velocity down a stem or coleoptile segment. As the pulse moves it becomes broader, at a roughly constant rate. It is shown here that this ‘spreading rate’ is larger than can be accounted for by a single channel model, but can be explained by coupling of channels with differing polar transport rates. An extreme case is where strongly polar channels are coupled to completely apolar channels, in which case auxin in the apolar part is ‘dragged along’ by the polar part in a somewhat diffuse distribution. The behaviour of this model is explored, together with others that can account for the experimentally observed spreading rates. It is also shown that saturation of carriers involved in lateral transport can explain the characteristic shape of pulses that result from uptake of large amounts of auxin.
| Auxin is one of the most important signalling molecules in plants. It is a key player in the development of veins and plant organs, and in responses that the plant makes to light and gravity. Yet we have a rather limited understanding of how auxin moves around plant tissues. I show here that a classic experiment, first carried out almost 50 years ago, has more information hidden in it than one might suppose. In this experiment, one studies how a pulse of radioactive auxin moves down a stem. The first lesson is that the peak of the pulse moves with a well-defined velocity, usually around 1 cm per hour. It has also been observed that the pulse spreads out as it moves, which is perhaps unsurprising, since any source of noise would be expected to have this effect. I show, though, that it is not easy to account for the degree of spreading in terms of noise. Instead, I propose that auxin travels down stems via several, probably many, coupled channels with differing transport rates. This gives one some insight into how plants manage their auxin signalling and puts some constraints on the underlying parameters.
| Auxin is the key integrating signal in plants [1, 2], and the dynamics of its movement within the plant is crucial to understanding development [3–9] and a multitude of physiological responses [10–12]. A traditional way to observe these dynamics in stems or coleoptiles is to apply a pulse of radioactively-labelled auxin at one end of a segment of the tissue, allow it to be transported for some period of time, and then cut up the segment into small pieces and measure the amount of label in each piece [13–15]. The profiles one obtains this way allow one to make some inferences about the underlying mechanism [14, 16].
One parameter that is often measured is the velocity of the pulse. It provides information about the underlying permeabilities and diffusion constants of the transport mechanism [16, 17]. I show here that one can also measure the rate at which the pulse spreads out, and that the values one obtains imply further constraints on the underlying mechanism. These constraints lead one to reject single channel models and to consider a variety of ways in which multiple auxin channels may be coupled together. (Here the term ‘channel’ will be used for a file of cells specialised for auxin transport, though later we also consider intracellular compartments as potential channels. Proteins that move auxin in or out of cells will be called transporters, more specifically importers or exporters.)
Models of auxin transport with many channels have certainly been considered before; in particular, the flow and counterflow pattern of auxin in the root has been modelled and applied to gravitropism and growth control [18–20]. However, the aim here is somewhat different, namely, to use the shape of an auxin pulse to diagnose properties of the stem transport system.
The classic experiments on auxin transport in coleoptiles [13] show a pulse moving basipetally and broadening as it goes. One would like to measure both the velocity, v, of this pulse and what will be called the spreading rate, denoted by ρ, which is defined to be the rate at which the variance of the pulse increases. It is not always straightforward to measure the velocity, because there are typically several components to the auxin transport profile [14, 21]. In particular, there are fixed or slower-moving components that trail behind the main pulse and may obscure its shape. These features cause even greater difficulties when measuring the variance, as small components far from the peak can cause a large increase in the variance.
One strategy would be to try to fit the data to a composite model that attempts to explain all the features in the profile. An alternative strategy, taken here, is to try to isolate the main peak, separating it from trailing or fast moving components. There are reasons to expect the shape of a pulse due to a single channel to be gaussian in form (S1 Text). The data were therefore fitted to a gaussian using least-squares, in the hope that this would pick out the main peak. This works quite well for the data from Goldsmith [13]; see S1 Fig. Given these fits, one can plot the mean and variance against time to obtain estimates for the velocity and spreading rate; see Table 1 and Fig 1A.
Goldsmith’s data points represent averages of several coleoptile segments, four for the zero time point and two for all other time points. This introduces the danger that the spreading rate may be over-estimated, since averaging two pulses moving at slightly different velocities will give rise to an artifactual increase of variance with time as the pulses separate. The overestimate will probably not be very large, being of the order of the variance of the velocity distribution. However, one can entirely avoid this objection by using the counts from individual segments. Here we used the single-segment data (S1 Data) underlying the averaged curves in Fig 4 in Morris et al. [15]. These were kindly provided by Dr Morris.
The raw data are of course noisier, and the problem is to determine the shape of the main peak in a reliable manner. Least squares performs poorly in this noisier setting (S2A Fig), as it attempts to fit the entire profile. However, one can modify the fitting algorithm so that it more effectively selects the peak region and ignores flanking regions. We call this procedure maximal fit (S1 Text and S2B Fig). One can also select the peak region by eye, trying to err towards excess width at short times and narrower regions at long times, so as to give a conservative estimate of the spreading rate. There is good agreement between this procedure and the maximal fit algorithm at long time intervals (S3 Fig); at shorter times the match is less good (S4 Fig) because of edge effects. Nonetheless, the estimates of spreading rate obtained by the two methods agree well, Table 1, and the estimates of velocity and spreading rate obtained from Goldsmith’s data by least squares and maximal fit also agree reasonably well, within their error ranges.
The conclusion of this section is that it is possible to measure both the velocity and the spreading rate of the main peak of an auxin pulse, though the spreading rate is more vulnerable to noise and questions of interpretation.
Next we ask what the velocity and spreading rate can tell us about the underlying transport mechanism. Consider first a simple model consisting of a row of cells, where auxin is assumed to reach a uniform distribution rapidly inside cells and the flux ϕ between cell n−1 and cell n takes the form
ϕ = p a n - 1 + q ( a n - 1 - a n ) , (1)
where ai denotes the auxin concentration in cell i. Thus there is a polar component to auxin transport between cells given by the permeability p and a symmetric, diffusion-like component given by the permeability q. In the next section it will be shown how p and q may relate to known auxin carriers. Note that a non-zero q can be produced by combinations of polar carriers, and does not necessarily imply that there is an underlying physical diffusive coupling.
The simple model is appropriate where the cells are small enough for intracellular diffusion to be rapid (e.g. 10 microns or less in length). For instance, an implicit model of this kind underlies treatments of canalization or up-the-gradient mechanisms near to the apical meristem. It is probably not such a good model for the auxin pulse experiments we are considering, since the relevant cells, e.g. the xylem parenchyma, are long enough (e.g. 100 microns) for intracellular diffusion times to be significant. Nonetheless, it serves as a useful warm-up exercise.
One can show (S2 Text) that the velocity, v, and the spreading rate, ρ, are given by
v = p, (2) ρ = L(p+2q), (3)
where L is the cell length. Given an estimate of cell length L, knowing the velocity and the spreading rate fully characterises the model. In fact, the pulse at time t is well approximated by a gaussian of mean vt and variance ρt; see S2 Text. Fitting the data with a gaussian, as was done in the last section, therefore has some theoretical justification.
The data in Table 1 allow one to calculate p and q. The velocity gives the permeability p, and taking L = 100 microns and using Eq (3) gives q. Instead of giving q itself in the Table, we give the ratio q/p, which is small when the movement of auxin is essentially polar and large when the diffusion-like coupling between cells dominates. As can be seen, the inferred value of q/p is large in all the data sets.
A version of this model has been used by Renton et al. [22] (their model I). The units in their version are not individual cells but 2 mm segments of stem (the length of the pieces into which the stem is subdivided for counting labelled auxin). They assume that these segments are coupled by a completely polar flux; i.e. q = 0. They are able to fit the peak region in their data by taking p = 9 mm/hr. Putting L = 2 mm, p = 9 mm/hr and q = 0 in Eq (3) gives ρ = 18 mm2/hr. This is larger than our estimate in Table 1, which could be due to the fact that they use averaged data and different fitting criteria.
There is a caveat here. In the above model, and in other models where a piece of tissue is used as the computational unit, e.g. the ‘metamers’ in [23], this computational strategy may offer a gain in speed and simplicity. But care is needed in interpreting the results at a cellular level. In the above model, we can ask what parameter values would be needed to achieve the same velocity and spreading rate with cells as the computational units instead of 2 mm pieces. Taking L = 100 microns, with a velocity of 9 mm/hr and spreading rate of 18 mm2/hr (as calculated above), Eq (3) gives 18 = Lp(1 + 2q/p) = 0.1 × 9 × (1 + 2q/p), or q/p = 9.5. Thus the completely polar segment-based model translates into a model at the cell level where the diffusive component is much larger than the polar component.
We now turn to the key question: is this simple model plausible? We have already pointed out that one needs to take account of intracellular diffusion in a realistic model. Another potentially unrealistic feature of the model is the dominance of diffusive coupling over polar transport between cells, implied by the values of q/p in Table 1. Yet the strongly polar localization of the auxin transporter PIN1 [24] in what are presumed to be the principal auxin-transporting tissues of the stem suggests a high overall polarity of transport. We consider next whether this view is justified.
Several families of auxin importers and exporters assist the movement of auxin between plant cells. One family of exporters consists of the PIN-formed family (PINs) [25, 26]. Some PINs, e.g. PIN5 [27], PIN6 [28] and PIN8 [29] are involved in movement between compartments in cells, and will be considered later. The remaining PINs are associated with the plasma membrane, and often show polar localisation within cells, e.g. a basal location of PIN1 in the stem [24], or a more complex pattern of polarity at the shoot apex [30], or a basal localization of PIN2 in epidermal root cells [31], or a gravitropic movement of PIN3 to the bottom side of Arabidopsis hypocotyl endodermal cells [12].
Another family of exporters consists of the ABCB transporters (also known as P-glycoproteins, or PGPs) [32, 33]. These are generally uniformly distributed over the plasma membrane, but can sometimes be polar. Thus ABCB1 is known to be uniformly distributed at the shoot and root apices but polarly localized in the mature root cortex and endodermis [32], where it is found in the basal membrane, as are PINs [34].
Some ABCBs, e.g. ABCB21 [35] and ABCB4 [36], may act both as exporters and importers, according to the balance of internal and external auxin concentrations. However, the predominant class of importers is the AUX/LAX family [26, 37, 38]. In general, these too seem to be uniformly distributed in the plasma membrane, though in one special case, the protophloem of Arabidopsis roots, AUX1 is asymetrically distributed [38].
Auxin can also be imported passively, without a specific protein channel, because of the lipid solubility of the protonated acid and the fact that the apoplast is acidic (pH typically 4.5 to 5). As the pK of the carboxyl group of auxin is 4.7, about half the auxin in the apoplast will be protonated, and therefore able to enter a cell fairly easily. This is one of the twin pillars of the chemiosmotic theory [39–42], the other being that auxin inside a cell, where the pH is higher (cytoplasmic pH 7.2 [43]), will pass through the membrane far less readily, being about 99.7% in the charged anion form. Thus auxin is in effect trapped inside cells, and exporters are required for intercellular auxin movement, whereas importers would seem less necessary. However, recent measurements of the permeability of the cell membrane to protonated auxin are about two orders of magnitude lower than those available when the theory was formulated. For instance Rutschow et al. [44] find values for PIAAH close to those of Delbarre et al. [45], at around 4 × 10−5 cm/sec, whereas older measurements gave 10−3 cm/sec [41] or 3.3 × 10−3 cm/sec [46]. These high estimates may have come about because the contribution of AUX importers was not appreciated. It seems likely, therefore, that auxin influx is dominated by AUX importers [44].
Now consider two adjacent cells in a channel, and let a1 be the cytoplasmic concentrations of auxin at the basal end of the uppermost cell, and a2 the concentration at the apical end of the cell below it (see Fig 2). Let b1 and b2 be the concentrations at the apical and basal boundaries of the apoplast between the two cells, and let L0 be the vertical width of this apoplastic region. Thus we allow there to be an auxin gradient within the apoplast; let the diffusion constant of auxin in the apoplast be D0.
Let α1 be the total permeability for exporters, i.e. the sum of permeabilities for both PINs and ABCBs, in the basal cell membrane of the upper cell, and α2 the total permeability for the apical membrane of the lower cell. Let us assume that AUX/LAX importers are uniformly distributed, so the combined permeabilities for import via AUX/LAX and passive movement of protonated auxin are equal at the basal and apical membranes, and have a value β. The flux ϕ per unit area between the cells is given by
ϕ = α1a1−βb1, (4) = D0L0(b1 - b2), (5) =βb2 − α2a2. (6)
Eliminating b1 and b2 gives
ϕ = ( α 1 a 1 - α 2 a 2 ) / ( 2 + r ) , (7)
where r = βL0/D0. This can be put in the form of Eq (1), i.e. ϕ = pa1 + q(a1 − a2), by taking
p = ( α 1 - α 2 ) / ( 2 + r ), (8) q = α2/(2 + r) .(9)
From this we get
q / p = α 2 α 1 - α 2 . (10)
Since PIN1, which is probably the principal exporter in the xylem parenchyma in stems, is strongly localised at the basal end of cells [24], the contribution of PIN1 permeability to α1 will be much larger than that to α2. On the assumption that ABCBs are non-polarly distributed, the ABCB permeability contributions to α1 and α2 will be equal. However, they must surely be very small compared to the PIN1 contribution to α1, since otherwise the isotropic distribution of ABCBs implies that auxin would be pumped out at a high rate in all directions. Thus α1 must greatly exceed α2, and hence q/p must be small, highlighting the implausibility of the values of q/p in Table 1.
The simple model assumes complete mixing inside cells, which is unlikely to be a good approximation for the cells that transport auxin in stems and coleoptiles. Let us therefore assume that auxin diffuses through the cytoplasm with a diffusion constant D that is 60–90% [47] of its value in aqueous solution; the latter being 6.7 × 10−6 cm2/sec [48]. Suppose the cells transporting auxin have length L, which we take to be 100 microns. Then the analogues of Eqs (2) and (3), derived in S3 Text, are
1v = 1p + L2D( 1 + 2qp), (11) ρ ≤ L v ( 1 + 2q p ) , (12)
with the inequality in Eq (12) approximating an equality as D becomes large.
These two formulae give constraints on q/p and D. For instance, from the data of Morris et al. [15] we have v = 9 mm/hr and ρ = 10 mm2/hr, from which inequality Eq (12) implies q/p > 5.06, and Eq (11) gives D ≥ 1.4 × 10−5 cm2/sec. Thus we still have the problem of a large value of q/p, and the unrealistic requirement for instant mixing inside cells has been replaced by the requirement for a diffusion constant that is twice the measured value for diffusion in water [48]. Possibly some intracellular transport mechanism could faciliate diffusion. Cytoplasmic streaming seems not to play an important part [49], but there could be some other kind of active process, such as the movement of carriers along fibres postulated in model II in Renton et al. [22]. Against this, Kramer et al. [17] have shown that there is a broad agreement between auxin transport speeds and the bound 2D/L based on passive diffusion [16]. Moreover, even if there is enhanced diffusion, the problem of the large value of q/p still remains.
Fig 3 compares a plausible single-channel model, with q/p small and a realistic cytoplasmic diffusion constant (q/p = .05,D = 5 × 10−6 cm2/sec), with Goldsmith’s Fig 1A and 1D, for zero time and 30 minutes, respectively. The zero time curves match approximately, but the 30 minute single-channel distribution is much too narrow compared to the data, as expected from the arguments above.
Is there some obvious modification of the single channel model that would predict larger spreading rates? In general, any source of noise makes response curves broader, and the longer an auxin pulse runs, the more noise it would be expected to accumulate. However, incorporating various plausible sources of noise, such as random variation in cell length (so that neighbouring files are no longer in register), and randomised permeability p, has little effect on the spreading rate (see S4 Text).
One might also imagine that the loading of auxin from a source could have an effect on the spreading rate. Certainly, a sustained period of loading will lead to a broader pulse. However, the rate of broadening is unchanged, whatever the temporal pattern of loading. This is because the variance of a pulse is the sum of the variance due to an instantaneous loading event and the variance of the loading distribution (see S5 Text). The latter variance term, being a constant, disappears when we take the time-derivative to get the spreading rate.
Another possible explanation for a large spreading rate would be saturation of polar transporters. Saturation will cause the high concentration part of a pulse to be held back while the lower concentration part can move forwards, thus stretching out the pulse.
Consider the situation where only the PINs, usually considered to be the principal polar transporters, show saturation, all other import and export mechanisms being treated as linear. Thus the flux due to PINs obeys Michaelis-Menten kinetics of the form flux = Vmax a/(Km + a). We simplify the situation by neglecting ABCB efflux. Then instead of Eq (7) we obtain
ϕ = κ ( 1 ) a 1 K m + a 1 - κ ( 2 ) a 2 K m + a 2 , (13)
where κ(i)=Vmax(i)/(2+r), and r = βL0/D0.
This expression for flux will come in useful later. For now, to illustrate the basic qualitative behaviour, we simplify even further by assuming that PINs are exclusively present in the basal ends of cells, so κ(2) = 0. This model can easily generate the required large spreading rates, by the stretching process described above. However, it leads to a characteristically-shaped pulse, with an abrupt edge at the apical end and an extended basal tail; see Fig 4. This is quite unlike the symmetrical gaussian which, as we have seen, gives a good fit to the peak region. This point is discussed in Renton et al. [22] (see their Fig 7c and 7d). In their model II, saturation of the auxin carriers can produce broad enough peaks to match the data, but the shape of the peak is then incorrect.
This suggests that saturation does not explain the spreading rate of pulses in the experiments discussed so far. However, Brewer et al. [50] showed, by using sources with differing auxin concentrations, that it was possible to change the total uptake by a factor of ten in Arabidopsis and almost a hundred in pea. Under these conditions, effects of saturation are seen. However, as will be shown later, these effects point to saturation of lateral transport, which brings us to the topic of auxin movement between multiple channels.
It appears not to be possible to generate a plausible match to the data with a single channel, and the natural next step is to consider a number of channels. If one has a collection of independent channels with a scatter of velocities, one might expect them to combine to give a pulse that grows broader as its components drift apart. However, it is unlikely that the channels are truly independent. Various experiments, such as those of Sachs [8, 9] on the induction of vascular strands by external application of auxin, suggest that auxin can move fairly freely, at least through some tissues.
One can model coupling of neighbouring cells by expressing the lateral flux per unit area formally in the same way as the axial flux, using Eq (1), or Eq (13) if there is saturation. In this section we assume that there is no saturation and also that the coupling is symmetric, so one can write
ϕ = s ( a n - 1 - a n ) , (14)
where s is the symmetric lateral permeability and an−1 and an denote auxin concentrations in cells that lie side-by-side.
Consider first a minimal model that illustrates lateral coupling. It consists of two adjacent channels, one polar and the other apolar. The situation is depicted in Fig 5A. When s is large enough the two channels are synchronised, producing a single sharp pulse travelling at their average velocity; see the curve for log s = −3 in Fig 5B. As the lateral permeability decreases, the single peak eventually separates into two distinct peaks (curve for logs = −7). However, there is an intermediate region where the peak broadens while still keeping an essentially gaussian form (curve for log s = −5).
To explore this further, we observe how the velocity of the pulse in a channel varies with the lateral permeability (see Fig 5C). Starting with large s, the two channels have the same velocity. As s decreases, a critical point is reached where the velocities begin to separate, and with sufficiently small s the channels behave essentially separately, the apolar channel pulse having zero velocity.
Fig 5D shows the spreading rate, which rises as the lateral permeability decreases, and then eventually declines. The dotted region on the curve corresponds to a range of permeabilities where the variance does not depend linearly on time, so the spreading rate is not properly defined. This is due to a “streak” of auxin left behind the fast moving peak, which causes the variance to increase slightly faster than linearly. However, before this dotted segment is reached, the spreading rate is well-defined and can match those seen experimentally.
We use Goldsmith’s data as a test set for various models. The vertical line in Fig 5C and 5D marks the value s = 7.1 × 10−6 cm/sec, where the 2-channel model gives a reasonable fit (by eye) to her Fig 1A and 1D at zero time and 30 minutes, respectively. Fig 6A and 6B show the corresponding auxin distributions.
One can scale this two-channel model up to larger numbers of channels, where there is a group of polar channels adjacent to a group of apolar channels (see the thumbnail sketches in Fig 6). This might be the situation, for example, when there is a specialised transporting tissue adjacent to a tissue where auxin moves diffusively. In these scaled-up models, the lateral coupling is assumed to be symmetric and of equal strength, s, everywhere. Thus one can regard ws, where w is the width of a cell, as a diffusion coefficient for lateral movement. More precisely, diffusion within cells with rate D will combine with the intercellular coupling [51] so the effective diffusion constant, Deff, is given by:
1 D e f f = 1 D + 1 w s . (15)
Now one intuitive explanation for the broadening of a pulse by coupling of channels is that it is due to the lateral diffusion of auxin. The diffusion distance, or the average distance that a molecule diffuses in time t, is 2 D t, (e.g. expression (3.40) in [52]). If one is comparing models with varying numbers of channels, and hence varying total width L, the time needed to diffuse the distance L is given by L = 2 D t, or t = L 2 4 D. So if the amount of broadening depends on this time t, we expect models to give similar shapes of pulses when the diffusion constant, Deff, is proportional to L2, or the square of the number of the channels. One can test this by finding the best matches to our test set for various numbers of channels; see Fig 6. The resulting values of Deff do indeed scale in a quadratic manner, and the same is true for s when ws ≪ D, since the term 1/(ws) then dominates 1/D in Eq (15). However, the scaling only holds approximately (Table 2). It seems this intuition only partially captures the complicated dynamics.
Note that there is a limit to lateral movement set by diffusion within the cytoplasm. From an engineering point of view, to achieve rapid lateral movement without wasted resources, the permeability s for movement between cells should be matched to diffusion within them. This means that the equivalent diffusion constant, ws, (where w is the width of a cell) should be equal to the diffusion constant D for auxin in cytoplasm, or equivalently that the two terms in Eq (15) should be equal. Taking w = 20 microns and D = 5 × 10−6 cm2/sec, this gives s = 2.5 × 10−3 cm/sec, which is not far from the best-fitting value for the 21-channel model (Table 2). With this value, Deff = 2.5 × 10−6 cm2/sec, which means that auxin can move rapidly enough through tissues to register changes at distances of the order of a millimetre within an hour (since t = L2/Deff with L = 1 mm is 1.1 hrs), which seems biologically reasonable. The biological parallel for this engineering argument would be that evolutionary pressure to achieve rapid signalling would lead to increasing values of the permeability s. However, at the point where ws = D, increasing s gives ever smaller gains in Deff, and selection for larger values of s becomes weak.
There are many other ways of varying the minimal model that still allow a good fit to Goldsmith’s data. For instance, one can consider two channels that are both polar, but with strengths that are in some ratio α : 1−α (Fig 7A). One can plot the spreading rate against coupling strength (as in Fig 5D), and Fig 7B shows a number of such graphs for different values of the ratio α : 1−α. As can be seen, the maximum spreading rate decreases as the polarities of the two channels become more similar. This implies that one can only find a match to Goldsmith’s data when the polarities are sufficiently different. Fig 7C shows that there is a match for 0.8:0.2, but not for 0.7:0.3, where the peak splits as the coupling is weakened in an attempt to get a sufficiently large spreading rate.
The spreading of a pulse due to lateral movement through the cells of a tissue could in principle also be produced by local movement within a cell: instead of looking at coupling between cells, we can look at coupling between subcellular compartments. One obvious example is the vacuole, which takes up most of the volume of a plant cell, in mature tissues at least. We can represent the cytosol/vacuole pair by the minimal model, i.e. Fig 5A, with n = 2. The cytosol is the polar channel and the vacuole the apolar channel; they are coupled laterally via the tonoplast (Fig 8). One difference is that there is no coupling between successive cells in the vacuolar channel, i.e. p = q = 0 for apical and basal faces, whereas the apolar channel in the minimal model has the same permeability on all cell faces (See Models, Fig 6). However, setting the apical and basal permeabilities to zero has only a very small effect on the dynamics, especially for the case n = 2 where s is small (Table 2).
Another example is the endoplasmic reticulum (ER). Auxin is thought to be moved from the cytosol to the ER by PIN5 [27], and PIN6/PIN8 may move auxin in the reverse direction [28]. Again we can represent the cytosol/ER pair by the minimal model, with PINs 5/6/8 providing the lateral coupling. The question is whether the minimal model, applied to these compartments, can account for the observed spreading rate, or for some fraction of it (lateral movement between and within cells could both be operating at the same time).
Consider first the vacuole. Two new features have to be added to the minimal model: asymmetry in the size of the compartments, and also a possible asymmetry in the inward and outward permeabilities, so in place of Eq (14), or ϕ = s(a1−a2) for the flux ϕ from channel 1 (cytosol) to channel 2 (vacuole), we have ϕ = scyt a1−svac a2, where scyt is the permeability from cytosol to vacuole, and svac is the permeability in the opposite direction. In the minimal model it was assumed that each channel (cell file) was 20 microns wide. Let us assume that the cytosol and vacuole have widths dcyt and dvac microns, respectively. We can rescale the permeabilities to take account of these widths, so that the dynamics of the best-fitting minimal model will be reproduced if we take scyt = s(dcyt/20), svac = s(dvac/20), where s is given by Table 2, n = 2, as 7.1 × 10−6 cm/sec.
Let us pause to consider the geometry of the cell and its vacuole. A cell is represented computationally by a brick shape whose long axis measures 100 microns and the two sides 20 microns; thus it has a square cross-section. The cytoplasm is assumed to form a thin layer, 1 micron in depth, just inside the boundary of the cell. Thus the vacuole to cytoplasm volume ratio is 182 × 98:202 × 100−182 × 98, or about 4:1. This is low compared to the average of about 10 : 1 [53], but probably more typical for the elongated xylem parenchyma cells that transport auxin efficiently in the stem. In comparing dcyt and dvac, we need to take into account that there are two layers of cytoplasm on opposite sides of the cell, so we can pair each layer of cytoplasm with the adjacent half of the vacuole, whose width is half that of the whole vacuole, namely 9 microns. Thus we take dcyt = 1 and dvac = 9. With these values, and with the best-matching s we get scyt = s(dcyt/20) = 3.6 × 10−7 cm/sec and svac = s(9dvac/20) = 3.2 × 10−6 cm/sec.
Thus we obtain a good match to Goldsmith’s data if scyt and svac have the above values. We can compare them with the values expected from passive diffusion. With a pH for the cytoplasm of 7.2 [43]), and using pIAAH = 4 × 10−5 cm/sec [45], we get scyt = 1.2 × 10−7 cm/sec. Similarly, taking the pH of the vacuole to be 5.5 [42, 43, 54], we find svac = 5.6 × 10−6 cm/sec. As can be seen, these are of the same order as the best-fitting values. However, scyt is about 3 times smaller and svac almost twice as large as these best-fitting values, so there is a bias towards excluding auxin from the vacuole.
These calculations show that, given passive diffusion, only about a quarter of the auxin in a cell resides in the vacuole, despite its large relative volume. The effect of this exclusion is that coupling with the vacuole produces only a small amount of broadening of the pulse: about 2.6 mm2/hour with passive diffusion. In addition to passive diffusion there may also be auxin transporters in the tonoplast; indeed, there is evidence for an exporter from the vacuole, the WAT1 transporter [55] (though see also [56]). It seems possible that a combination of low pH and exporters ensures that auxin is largely excluded from the vacuole, in which case its contribution to the spreading rate will of course be negligible.
We take a final look at the geometry of the vacuole to explain why it is plausible, when looking at lateral movement between cells, to treat each cell as having a 20 micron width, despite the vacuole filling much of the space. The reason is that one can view lateral flux as taking place in sheets of cytoplasm 1 micron thick but 20 microns wide within each cell, and with corresponding sheets in adjacent cells (marked in red in Fig 9). The cytoplasmic sheets at right angles serve as an extended interface between adjacent cells. This is of course a gross oversimplication, but at least provides a tractable computational model.
Turning next to the ER, if auxin is both imported and exported by specific channels [28], then the permeabilities are likely to be much larger than for passive diffusion, so we will be in the left-hand part of the graph in Fig 5D where the spreading rate is low. Thus if we regard the ER simply as a space through which auxin diffuses, it is unlikely to explain the spreading rate. However, there is evidence that auxin undergoes metabolic processing once it enters the ER [27], and this could plausibly account for the frequently observed accumulation of label in the wake of a pulse (see for instance Fig 8 in [14]).
We now return to the question of saturation. The pea data of Fig 4A–4E in Brewer et al. [50], kindly supplied by Dr Brett Ferguson (S1 Data), are redrawn in Fig 10A. They show how the shape of a pulse changes when increasing amounts of auxin are applied and larger amounts of auxin taken up by plant tissues (determined by the total counts in the stem segments). There is a gradual build-up of auxin near to the apical end of the stem which gradually swamps the peak as the uptake increases.
This does not resemble the effect of saturation of the polar transporters in a single channel model described earlier, where the trailing edge of the pulse remains sharp as the uptake increases (Fig 4). A better match to the experimental data can be obtained by assuming that there is saturation of the lateral movement between channels. As a first try at modelling this, S6 Text, we add lateral saturation to the n = 6 version of the model shown in Fig 5 (with the minor modification that we make the two polar channels of different strengths, which gives a better eventual fit). We assume that auxin initially moves from the externally applied source into the apolar channels, from which it can move laterally into the polar channels. The idea is that, as the uptake of auxin increases, lateral transport will be limited by saturation, and a diffusion gradient of residual auxin in the apolar channels will be created, accounting for the apical build-up seen in the data. There is no saturation of the axial (polar) transporters, only of those that move auxin laterally.
S6A Fig shows how this model behaves. The 26.5 ng peak is retarded relative to the peaks with smaller loadings. There is also a marked apical build-up with the 95.9 ng pulse. However, there is little or no apical build-up with the lower uptakes, as there is in the pea data. Weakening the lateral coupling partially brings this about but distorts the curves in other ways. However, selectively weakening of the lateral coupling between the fifth and sixth channels, as shown in Fig 10B, does achieve a fairly good fit to the data. Thus a model of modest complexity can capture the qualitative features of the family of data curves.
If we take this model and keep all parameters unchanged except that we switch the saturation from lateral to axial transporters, the simulated pulses are as shown in Fig 10C. This is clearly a very poor fit to the data, with large shifts in the positions of peaks, like those seen in Fig 4, and no build-up at the apical end.
Finding an explanation for the large spreading rate of auxin pulses turns out to be an intriguing puzzle. One can rule out a single-channel model, and also variants of it that introduce noise, e.g. by randomising cell length or the degree of polarity in successive cells in the channel. These models fail by a large margin to achieve a large enough spreading rate.
I propose here that the explanation lies in the coupling together of several, perhaps many, channels of different degrees of polarity. When the coupling strength is large, the channels are essentially bound together into one “super-channel” that has too sharp a peak to match the data. When the coupling is sufficiently weak, pulses in the different channels move at different velocities and separate. There is, however, an intermediate range of lateral coupling strength where the pulses in all the channels move coherently but the resulting pulse spreads out.
It is this intermediate coupling regime that matches the data best. A good fit can only be obtained if the coupled channels have sufficiently disparate polarities. This might mean, for example, that some of them correspond to specialised polar transport tissue and others to apolar, or very weakly polar, tissues. Alternatively, the specialised transport tissues themselves might consist of channels with varying degrees of polarity, due to their stage of maturation for example. There is also a wide range of possible physical scales that can produce the observed phenomena, ranging from a minimal model of two channels (with a total width of 40 microns) to extended models with 20–30 parallel channels (0.4 to 0.6 mm).
This brings home the fact that there are many ways of modelling the same data, which is hardly surprising, given that one is only fitting two parameters: the velocity and spreading rate of a pulse. There are more constraints once one brings saturation into the picture and is required to explain not just the peak but also the build-up at the apical end when large amounts of auxin are taken up, as with the data from Brewer et al.; Fig 10A). The six-channel model of Fig 6 can be adapted to give a good match to the data, by introducing saturation of lateral permeabilities and weak coupling between some of the channels, which allows them to act as a slowly emptying reservoir of auxin that is waiting to enter the rapid transport stream (Fig 10B).
However, even with this more complex scenario, the model is not unique. An alternative model, S6 Text and S6B Fig, can achieve as good a match. In this model there is an asymmetry in the lateral permeabilities, which ensures that auxin can only enter, and not exit from, the polar channels. There is some evidence for this kind of polar lateral transport (e.g. [57], Fig 1), and it could in principle have a strong influence on the lateral distribution of auxin. Thus the difference between the two models bears on the problem of balancing basipetal flow with lateral propagation.
Although the channels have been envisaged as files of cells, it is also possible that they could be compartments within cells. However, it has been argued that two conspicuous compartments, the vacuole and the endoplasmic reticulum, are unlikely to play this role, the former because the coupling with the cytosol is too weak, the latter for the opposite reason. Thus the cell file interpretation seems more plausible, and one attractive conjecture is that the 21-channel expanded version of the minimal model, with a corresponding physical width of 0.4 mm, comes closest to reality for the coleoptile, as it fits not only Goldsmith’s data but also accords with the ‘engineering’ criterion of matching lateral diffusion between channels to diffusion within cells.
An intriguing final observation is that the best-fitting parameters for Goldsmith’s data imply a coupling strength that is close to the point where the velocities in the channels begin to separate; the vertical line in Fig 5C lies close the split. A similar conclusion applies to the data of Morris et al. [15]. This suggests that the polar transport system is only just able to carry along with it the auxin flow in the surrounding apolar tissues. This could be an indication of energetic economy, or perhaps a way of creating responsiveness to changes.
In all the models, cells are assumed to be 100 microns long and 20 microns wide. To simulate intracellular diffusion, cells are represented as rectangles divided up lengthwise into N compartments, where N was generally taken to be 5. If lateral coupling constants are large enough so that lateral intracellular diffusion needs to be taken into account, this is done by a further lengthwise subdivision of each compartment into two.
A permeability having a computational value of p′ is converted into a permeability p measured in cm/sec by
p = p ′ L N Δ ,
where L is the cell length in cm and Δ is the time step, generally taken to be 1/20 sec. The computational parameter D′ for intracellular diffusion is converted into a diffusion constant D in cm2/sec by
D = D ′ L 2 N ( N - 1 ) Δ .
This makes Eq (11) equivalent to
1 v ′ = 1 p ′ + N - 1 2 D ′ ( 1 + 2 q ′ p ′ ) ,
as observed in [16] (see the passage following Eq (40)).
The constants and simulation details for the models used for generating the figures are as follows:
Fig 3. For the single channel model in this figure, p = 4 × 10−4 cm/sec, q = 2.08 × 10−5 cm/sec, D = 5 × 10−6cm2/sec.
Fig 4A Here we take Vmax(i)=10−6 cm2/sec, Km = 10−3, q = 0, and D = 5 × 10−6 cm2/sec. The total amount of auxin uptake, treated as an instantaneous pulse, takes the values 0.3, 1 and 2 in arbitrary units. The simulation runs for the equivalent of 90 minutes. B The same data as in A normalised.
Fig 5 For the polar channel, p = 1.4 × 10−3cm/sec, q = 0; for the apolar channel p = q = 0. For all channels, D = 5 × 10−6 cm2/sec. It could be argued that q should be comparable to s, since an apolar cell should have uniformly distributed permeabilities. However, we wish to study the effect of changing the lateral coupling alone, so q is fixed at zero. And in fact, the picture does not change qualitatively if we put q = s rather than q = 0. Uptake of auxin was simulated by adding at each computational step a fixed amount to the concentration in all the channels at the apical end for the loading period. At the end of the loading period, the distribution was normalised to a total of 1. We used Goldsmith’s protocol for loading, with 15 minutes of uptake at a constant rate at the apical end of the segment. This was followed by 45 minutes of transport, corresponding to Goldsmith’s 30 minute data, since she includes a 15 minute wait after loading before “zero time”. In the dotted region of the graph of spreading rate in D, a trail of auxin behind the frontal peak makes the variance increase faster than linearly, so the spreading rate is not defined. Instead, what is plotted is (Var(0)-Var(t))/t, where Var(t) denotes the variance of the distribution at time t, and t is taken to be 30 mins. If one had chosen t to be 15 mins, for example, the solid-line parts of the curve would have been identical, and the dotted line part slightly lower.
Fig 6 For all polar channels, q = 0. For apolar channels we take p = 0 and put q = s, where s is the value of the lateral permeability needed to match Goldsmith’s data given the number of channels in the model. Thus permeabilities are the same on all cell faces in apolar cells. For the polar channels we take the following values of p: n = 2, p = 1.4 × 10−3 cm/sec, n = 6, p = 6 × 10−3 cm/sec, n = 21 p = 6.4 × 10−3cm/sec, and n = 30, p = 8.8 × 10−3cm/sec. In all cells D = 5 × 10−6 cm2/sec. Loading follows the protocol given above for Fig 5. Graphs in B were scaled so that the peak heights coincide.
Fig 7 To make the graph on the left, the two-channel model is used with p = 1.44 × 10−3, so pleft = αp, and pright = (1−α)p, where α:1−α takes the values 1 : 0, 0.9 : 0.1, etc., indicated on the graph. For both channels, q = 0, D = 5 × 10−6 cm2/sec. To make the right-hand graph, we again take q = 0, D = 5 × 10−6 cm2/sec in both channels, and now seek values of p and s that optimise the fit (or come as close as possible), for the indicated ratios α:1−α. For 1:0 we take p = 1.44 × 10−3 cm/sec, s = 1.4 × 10−5 cm/sec. For 0.8:0.2 we take p = 9.6 × 10−4 cm/sec, s = 9.6 × 10−6 cm/sec. For 0.7:0.3 we take p = 8.8 × 10−4 cm/sec, s = 4.0 × 10−7 cm/sec.
Fig 10B Saturation kinetics is used only in the lateral walls (marked by an ‘S’ in the sketch). The lateral coupling between the channels is symmetric, with κ1 = κ2 = Km s in Eq (13), where Km = 10−4 and s = 2.4 × 10−5 cm/sec for the green arrows and 4 × 10−7 cm/sec for the red arrows. The other (unsaturated) permeabilities are p = 3.2 × 10−3 cm/sec in channel 1, p = 4 × 10−4 cm/sec in channel 2, and for channels 3 to 6 p = 0 and q = 8 × 10−3 cm/sec. For all channels, D = 5 × 10−6cm2/sec. Loading is assumed to be instantaneous in the initial computational cycle and to be into channels 2 to 6 only. The computation runs for the equivalent of 4 hours. The parameters in C are as in B, but now saturation kinetics are used for the axial permeability, so in Eq (13) for channels 1 and 2 we have κ1 = Km p, κ2 = 0. There is no saturation in the lateral coupling.
|
10.1371/journal.ppat.1003298 | Systems Analysis of a RIG-I Agonist Inducing Broad Spectrum Inhibition of Virus Infectivity | The RIG-I like receptor pathway is stimulated during RNA virus infection by interaction between cytosolic RIG-I and viral RNA structures that contain short hairpin dsRNA and 5′ triphosphate (5′ppp) terminal structure. In the present study, an RNA agonist of RIG-I was synthesized in vitro and shown to stimulate RIG-I-dependent antiviral responses at concentrations in the picomolar range. In human lung epithelial A549 cells, 5′pppRNA specifically stimulated multiple parameters of the innate antiviral response, including IRF3, IRF7 and STAT1 activation, and induction of inflammatory and interferon stimulated genes - hallmarks of a fully functional antiviral response. Evaluation of the magnitude and duration of gene expression by transcriptional profiling identified a robust, sustained and diversified antiviral and inflammatory response characterized by enhanced pathogen recognition and interferon (IFN) signaling. Bioinformatics analysis further identified a transcriptional signature uniquely induced by 5′pppRNA, and not by IFNα-2b, that included a constellation of IRF7 and NF-kB target genes capable of mobilizing multiple arms of the innate and adaptive immune response. Treatment of primary PBMCs or lung epithelial A549 cells with 5′pppRNA provided significant protection against a spectrum of RNA and DNA viruses. In C57Bl/6 mice, intravenous administration of 5′pppRNA protected animals from a lethal challenge with H1N1 Influenza, reduced virus titers in mouse lungs and protected animals from virus-induced pneumonia. Strikingly, the RIG-I-specific transcriptional response afforded partial protection from influenza challenge, even in the absence of type I interferon signaling. This systems approach provides transcriptional, biochemical, and in vivo analysis of the antiviral efficacy of 5′pppRNA and highlights the therapeutic potential associated with the use of RIG-I agonists as broad spectrum antiviral agents.
| Development of safe and effective drugs that inhibit virus replication remains a challenge. Activation of natural host defense using interferon (IFN) therapy has proven an effective treatment of certain viral infections. As a distinct variation on this concept, we analyzed the capacity of small RNA molecules that mimic viral components to trigger the host antiviral response and to inhibit the replication of several pathogenic human viruses. Using gene expression profiling, we identified robust antiviral and inflammatory gene signatures after treatment with a 5′-triphosphate containing RNA (5′pppRNA), including an integrated set of genes that is not regulated by IFN treatment. Delivery of 5′pppRNA into lung epithelial cells in vitro stimulated a strong antiviral immune response that inhibited the multiplication of several viruses. In a murine model of influenza infection, inoculation of the agonist protected animals from a lethal challenge of H1N1 influenza and inhibited virus replication in mouse lungs during the first 24–48 h after infection. This report highlights the therapeutic potential of naturally derived RIG-I agonists as potent stimulators of the innate antiviral response, with the capacity to block the replication of diverse human pathogenic viruses.
| The innate immune system has evolved numerous molecular sensors and signaling pathways to detect, contain and clear viral infections [1]–[4]. Viruses are sensed by a subset of pattern recognition receptors (PRRs) that recognize evolutionarily conserved structures known as pathogen-associated molecular patterns (PAMPs). Classically, viral nucleic acids are the predominant PAMPs detected by these receptors during infection. These sensing steps contribute to the activation of signaling cascades that culminate in the early production of antiviral effector molecules, cytokines and chemokines responsible for the inhibition of viral replication and the induction of adaptive immune responses [2], [5]–[7]. In addition to the nucleic acid sensing by a subset of endosome-associated Toll-like receptors (TLR), viral RNA structures within the cytoplasm are recognized by members of the retinoic acid-inducible gene-I (RIG-I)-like receptors (RLRs) family, including the three DExD/H box RNA helicases RIG-I, Mda5 and LGP-2 [2], [3], [5], [8]–[12].
RIG-I is a cytosolic multidomain protein that detects viral RNA through its helicase domain [13], [14]. In addition to its RNA sensing domain, RIG-I also possesses an effector caspase activation and recruitment domain (CARD) that interacts with the mitochondrial adaptor MAVS, also known as VISA, IPS-1, Cardif [15], [16]. Viral RNA binding alters RIG-I conformation from an auto-inhibitory state to an open conformation exposing the CARD domain, resulting in the generation of an activated state characterized by ATP hydrolysis and ATP-driven translocation on RNA [17]–[19]. Activation of RIG-I also allows ubiquitination and/or binding to polyubiquitin. In recent studies, polyubiquitin binding has been shown to induce formation of RIG-I tetramers that activate downstream signaling by inducing the formation of prion-like fibrils composed of the MAVS adaptor [20]. MAVS then triggers the activation of IRF3 and NF-κB transcription proteins through the IKK-related serine kinases TBK1 and IKKε [10], [21]–[23], leading to the primary activation of the antiviral program, involving production of type I interferons (IFNβ and IFNα), as well as pro-inflammatory cytokines and antiviral factors [5], [24], [25]. A secondary response involving the induction of IFN stimulated genes (ISGs) is induced by the binding of IFN to its cognate receptor (IFNα/βR), which triggers the JAK-STAT pathway to amplify the antiviral immune response [6], [26]–[29].
The nature of the ligand recognized by RIG-I has been the subject of intense study given that these PAMPs are the initial triggers of the antiviral immune response. In vitro synthesized RNA carrying an exposed 5′ terminal triphosphate (5′ppp) moiety was first identified as RIG-I agonists [30]–[32]. The 5′ppp moiety is present at the end of viral and self RNA molecules generated by RNA polymerization; however, in eukaryotic cells, RNA processing in the nucleus cleaves the 5′ppp end and the RNA is capped prior to release into the cytoplasm. This mechanism distinguishes viral ‘non-self’ 5′pppRNA from cellular ‘self’ RNA, and renders it recognizable to the innate RIG-I sensor [30], [31], [33]. Further characterization of the RNA structure demonstrated that blunt base pairing at the 5′ end of the RNA, with a minimum double strand (ds) length of 20 nucleotides was also important for RIG-I signaling [17], [33], [34]. Furthermore, short dsRNA (<300 bp) triggered RIG-I, whereas long dsRNA (>2000 bp) such as poly I:C and lacking 5′ppp failed to trigger RIG-I, but was recognized by Mda5 [35].
Natural RNA extracted from virally infected cells, specifically the viral RNA genome or viral replicative intermediates, were also shown to activate RIG-I [30], [31], [36]–[38]. Interestingly, the highly conserved 5′ and 3′ untranslated regions (UTRs) of negative single strand RNA virus genomes display high base pair complementarity and the panhandle structure theoretically formed by the viral genome meets the requirements for RIG-I recognition [17]. The elucidation of the crystal structure of RIG-I highlighted the molecular interactions between RIG-I and 5′ppp dsRNA [18], [39], providing a structural basis for the conformational changes involved in exposing the CARD domain for effective downstream signaling [18].
Given the level of molecular understanding of the RIG-I ligand and subsequent signaling leading to induction of antiviral immune response, we sought to investigate the range of the protective innate immune response triggered by RIG-I agonists against viral infections. A short in vitro-synthesized 5′pppRNA derived from the 5′ and 3′ UTRs of the VSV genome activated the RIG-I signaling pathway and triggered a robust antiviral response that interfered with infection by several pathogenic viruses, including Dengue, HCV, H1N1 Influenza A/PR/8/34 and HIV-1. Furthermore, intravenous delivery of the RNA agonist stimulated an antiviral state in vivo that protected mice from lethal influenza virus challenge. This report highlights the therapeutic potential of naturally derived RIG-I agonists as potent stimulators of the innate antiviral response, with the capacity to mobilize genes essential for the generation of efficient immunity against multiple infections.
A 5′ triphosphate containing RNA derived from the 5′ and 3′ UTRs of the negative-strand RNA virus Vesicular Stomatitis Virus (VSV) [17] was generated by in vitro transcription using T7 polymerase, an enzymatic reaction that synthesizes RNA molecules with a 5′ppp terminus [17]. Predicted panhandle secondary structure of the 5′pppRNA is depicted in Fig. 1A; gel analysis and nuclease sensitivity confirmed the generation of a single RNA product of the expected size (67 nucleotides). Transfection of increasing amounts of 5′pppRNA resulted in Ser396 phosphorylation of IRF3 at 8 h – a hallmark of immediate early activation of the antiviral response (Fig. 1B, lane 2 to 6). Induction of apoptosis was detected following treatment with higher concentrations of 5′pppRNA; the pro-apoptotic BH3-only protein NOXA – a direct transcriptional target of IRF3 [40] – as well as cleavage products of caspase 3 and PARP were up-regulated in a dose dependent manner (Fig. 1B, lane 2–6). Optimal induction of antiviral signaling with limited cytotoxicity was achieved at a concentration of 10 ng/ml (∼500 pM) (Fig. 1B; lane 4). Importantly, the stimulation of immune signaling and apoptosis was dependent on the 5′ppp moiety; a homologous RNA without a 5′ppp terminus abrogated stimulation over a range of RNA concentrations (Fig. 1B, lane 8–12).
To characterize the antiviral response triggered by 5′pppRNA, the kinetics of downstream RIG-I signaling were measured at different times (0–48 h) after stimulation of A549 cells (Fig. 1C). IRF3 homodimerization (1st panel) and IRF3 phosphorylation at Ser396 (2nd panel) were detected as early as 2 h post treatment with 5′pppRNA, and sustained until 24 h. Using a newly characterized anti-IRF7 antibody, induction of endogenous IRF7 was detected with kinetics that was delayed compared to IRF3 activation (4th vs. 3rd panel). IκBα phosphorylation was likewise detected as early as 2 h post-treatment and was sustained in A549 cells (6th panel). Altogether, IRF3, IRF7 and NF-κB are required for optimal induction of the IFNβ promoter [27]. JAK-STAT signaling was detected at 4 h with Tyr701 phosphorylation of STAT1 (9th panel), as well as later accumulation at 24 h (10th panel). IFIT1 and RIG-I itself, were up-regulated 4 h post-treatment (11th and 12th panel) whereas a second group of ISGs (STAT1 and IRF7; 4th and 10th panel) was induced between 6 h and 8 h after agonist treatment. IFNβ was detectable in cell culture supernatant as early as 6 h after 5′pppRNA treatment with a substantial release (4000 pg/ml) that peaked at 12–24 h (Fig. 1D, top panel). IFNα release was detected later at 12 h, and remained high thereafter (400 pg/ml) (Figure 1D). Thus 5′pppRNA triggers a full antiviral response as demonstrated by the activation of transcription factors IRF3, IRF7 and NF-κB, release of interferons, JAK/STAT pathway activation and induction of ISGs.
To address whether 5′pppRNA exclusively activated the RIG-I sensor, wild type mouse embryonic fibroblasts (wt MEF) and RIG-I−/− MEF were co-transfected with 5′pppRNA and type 1 IFN reporter constructs to measure promoter activity. 5′pppRNA activated the IFNβ promoter 60-fold and the IFNα promoter 450-fold in wt MEF; promoter activity was dependent on RIG-I since these promoters were not stimulated in RIG-I−/− MEF. As positive control, a constitutively active, CARD domain-containing RIG-I mutant [41] was used to bypass the requirement for RIG-I (Figure 2A). Furthermore, induction of the IFN response was exclusively dependent on intact RIG-I signaling, since IFNβ promoter activity was not decreased in Mda5−/−, TLR3−/− or TLR7−/− MEFs (Figure 2B). In A549 cells treated with 5′pppRNA, knocking down RIG-I abolished IRF3 and STAT1 phosphorylation, as well as IFIT1 and RIG-I upregulation compared to control siRNA-treated cells (Fig. 2C; lane 6 vs. 4). Of note, generation of the knock down by transient transfection of short RNA did not activate immune signaling (Fig. 2C; lane 3 vs. 1). Hence, this 5′pppRNA signals specifically via RIG-I.
To evaluate the breadth of the host intrinsic response resulting from RNA agonist stimulation of RIG-I, modulation of the transcriptome of A549 cells stimulated with 5′pppRNA from 1 to 48 h was analysed by gene array using the Illumina platform. Figure 3A shows a waterfall plot of differentially expressed genes (DEG; selected based on fold change ≥±2, p-value ≤0.001) after 5′pppRNA stimulation. The number of genes up-regulated by 5′pppRNA administration steadily increased with time, while the majority of down-regulation occurred at 24–48 h (Fig. 3A). The heatmap presents DEG with emphasis on the most highly deregulated genes over time (Fig. 3B). Canonical pathway analysis using Ingenuity Pathway Analysis software identified IFN signaling, activation of IRFs by cytosolic PRRs, TNFR2 signaling and antigen presentation as the main up-regulated functional categories, while functions related to cell metabolism and cell cycle were down-regulated by RIG-I agonist treatment (Fig. 3C). Subsequent kinetic analysis revealed that RIG-I agonist induced distinct temporal patterns of gene expression (Fig. 3D, S1 and S2A). For example, some genes were highly expressed early at 6–12 h, including IFNB1 and the IFNλ family (IL29, IL-28A, IL28B), but the expression of these genes was not sustained throughout the time course and decreased at 24–48 h (Fig. 3D; lest panel). A second subset of genes associated with the antiviral response were induced early at 6–8 h, but expression was sustained and markedly augmented at 24–48 h, as exemplified by IFI family members, IRF7, and other ISGs (Fig. 3D; middle panel). A third subset of genes was induced primarily at later time points, as part of the secondary response to 5′pppRNA treatment, and included HLA and CCL3 families (Fig. 3D; right panel). Representative genes from different subsets were validated by quantitative real-time RT-PCR (Fig. S1). Overall, 5′pppRNA induced a robust bi-phasic transcriptional response, characterized by strong activation of antiviral and inflammatory gene signatures; the kinetics of the transcriptional profile mirrored the biochemical activation events detected in Figures 1.
In order to gain systems-wide insight into the RIG-I transcriptome, a functional clustering of 5′pppRNA-induced DEGs was performed. This functional clustering identified a variety of transcriptional sub-networks and biological processes regulated by RIG-I (Fig. 4). As expected, at 6 h (Fig. 4A), induction of antiviral and inflammatory response programs downstream of IRF, NF-κB, STAT signaling were identified (Fig. 4A and S2B); expression of several cytokine and chemokine genes were also up-regulated. Concomitantly, genes related to Fos and TGF-β signaling, as well as hypoxic signaling via HIF-1α were down-regulated. At 24 h (Fig. 4B), genes associated with pathogen recognition receptor signaling, the ubiquitin pathway, inflammation and apoptosis were also induced by RIG-I activation; interestingly, profiling of down-regulated genes identified functional clusters involved in cell cycle regulation, MYC signaling, and the heat shock response.
Although the contribution of type I IFN to the antiviral response stimulated by 5′pppRNA is unquestionable, other factors may also augment the antiviral state established by 5′pppRNA. To define gene uniquely induced by 5′pppRNA, gene expression profiles of A549 cells stimulated with 5′pppRNA or with IFNα-2b for 6 and 24 h were compared. The heatmap in Figure 5A displays the expression profile of genes differentially up- and down-regulated (fold change ≥2; p-value ≤0.001) by 5′pppRNA and IFN (black and blue genes) or exclusively by 5′pppRNA (red and green genes). Interestingly, 5′pppRNA induced a significantly broader gene expression program compared to IFNα-2b, especially at 24 h. To determine whether differences in gene expression observed between 5′pppRNA and IFNα-2b were due to sub-optimal stimulation by IFNα-2b, higher concentrations of IFNα-2b within the range reported for in vitro applications were tested. Treatment with 5′pppRNA at 10 ng/ml was equivalent to treatment with IFNα-2b at 100 IU/ml in terms of IFNα levels released into cell culture supernatant. Increasing the concentration of IFNα-2b to 1000 IU/ml corresponded to levels of IFNα that were 8-times greater than physiological secretion following 10 ng/ml 5′pppRNA treatment (2500 pg/ml vs 480 pg/ml; Fig. 5B). Regardless of the amount of IFNα-2b used to activate cells, gene expression levels remained relatively unchanged (Fig. 5A; IFNα-2b 100 IU/ml vs 1000 IU/ml), indicating that IFNα-2b treatment was saturating.
Remarkably, the spectrum of genes differentially expressed by IFNα-2b treatment were virtually all contained within the transcriptome induced by 5′pppRNA treatment - 57 at 6 h and 134 at 24 h, out of 139 genes (Fig. 5C) - demonstrating that 5′pppRNA induced a complete IFN response by 24 h. Some of the genes characteristically activated as part of the IFN signature and maximally induced by 5′pppRNA are MX1, IFIT1, ISG15 (Fig. 5A, black portion). This comparison also highlighted the fact that a surprisingly large number of genes are uniquely regulated by 5′pppRNA - 38 genes at 6 h and 730 genes at 24 h (Fig. 5C). Most notably, IFNB1 and all three members of the IFNλ family, as well as the cytokines CCL5, CXCL10, IL-6, and CCL3, were highly induced by 5′pppRNA but not IFNα-2b treatment (Fig. 5A; red genes). While IFN signaling was highly induced by both treatments, IFNα-2b strongly activated antigen presentation machinery, and 5′pppRNA preferentially stimulated dendritic cell maturation and crosstalk, linking innate and adaptive immunity as well as induction of a wider range of signaling pathways (Fig. 5D). Significantly, 5′pppRNA preferentially stimulated a more extensive induction of IRF7 and NF-κB signaling nodes compared to IFNα-2b treatment (Fig. S2C). Thus, 5′pppRNA treatment, besides inducing a complete IFN response, additionally stimulated the transcription of a large and unique set of inflammatory and antiviral genes.
To determine if the RIG-I agonist was capable of inducing a functional antiviral response, A549 cells were treated with 5′pppRNA, and 24 h later, challenged with VSV, Dengue (DENV), or Vaccinia viruses. All viruses established infection in untreated cells as assessed by flow cytometry (60%, 20% and 80%, respectively) but in 5′pppRNA-treated cells, VSV and DENV infectivity was reduced to <0.5%, while infection with vaccinia was reduced to 10% (Fig. 6A). Release of infectious VSV and DENV virus was completely blocked by 5′pppRNA treatment (1.7×109 and 4.3×106 PFU/mL in untreated cells, respectively vs. undetectable in treated cells; Fig. S3). Similarly, in primary human CD14+ monocytes, DENV infection decreased from 53.7% to 2.6% in the presence of 5′pppRNA; in the CD14− fraction, DENV infectivity was lower (3%), but was likewise inhibited by RNA agonist treatment (0.4%; Fig. 6B). To demonstrate the requirement for intracellular delivery of 5′pppRNA, primary CD14+ cells from three patients were treated with 5′pppRNA alone, transfection reagent alone or the combination; DENV infectivity was reduced from ∼30% to ∼0.5% only upon transfection of the RNA agonist (Fig. 6C). To evaluate the antiviral effect of 5′pppRNA against HIV infection, activated CD4+ T cells were pre-treated with supernatant isolated from 5′pppRNA-treated monocytes and then infected with HIV-GFP. In the absence of treatment, 24% of the activated CD4+ T cells were actively infected by HIV as determined by GFP expression by flow cytometry, whereas infection of CD4+ T cells treated with 5′pppRNA supernatants was reduced to 11% (Fig. 6D). 5′pppRNA also had an antiviral effect against HCV in hepatocellular carcinoma cell line Huh7; HCV NS3 expression was inhibited by 5′pppRNA treatment (Fig. 6E; lane 4 vs. 2 and 6). The antiviral effect was fully dependent on RIG-I, as demonstrated in Huh7.5 cells (that express a mutated inactive RIG-I) by the absence of IFIT1 up-regulation following 5′pppRNA treatment (Fig. 6E; lane 9) and NS3 expression comparable to untreated HCV-infected cells (Fig. 6E; lane 10 vs. 8 and 12). Thus, 5′pppRNA is a broad-spectrum antiviral agent able to trigger an efficient innate immune response in different cell types and prevent infection by RNA and DNA viruses.
To further explore the inhibitory potential of 5′pppRNA, A549 cells were pre-treated with 5′pppRNA for 24 h and then infected with H1N1 A/PR/8/34 Influenza virus at increasing MOI (0.02, 0.2, 2); influenza replication was monitored by NS1 protein expression (Fig. 7A) and plaque assay (Fig. 7B). Viral replication was blocked by 5′pppRNA pre-treatment, even at the highest MOI, as demonstrated by complete loss of NS1 expression and 40-fold decrease in viral titer at MOI 2. In A549 cells pre-treated with decreasing concentrations of 5′pppRNA (10–0.1 ng/ml) prior to influenza virus challenge (0.2 MOI), 5′pppRNA significantly blocked influenza replication at a concentration of 1 ng/ml, as demonstrated by a 3-fold reduction in NS1 protein expression (Fig. 7C; lane 7) and a 7-fold reduction in virus titer by plaque assay (Fig. 7D).
To demonstrate that the antiviral activity of 5′pppRNA against influenza relies on RIG-I signaling, A549 cells were knocked down for RIG-I and infected with influenza; in the knockdown, ISGs were not induced (Fig. 7E, lanes 3 vs. 6) and 5′pppRNA treatment failed to inhibit NS1 expression (Fig. 7E; lanes 5 vs. 6), indicating that the antiviral effect of 5′pppRNA is exclusively dependent on RIG-I. Next, to determine whether the RIG-I ‘unique’ gene expression profile characterized in Figure 5 could compensate for the IFN response, A549 cells were knocked down for the IFNα/βR; the knock down was efficient, as demonstrated by the absence of IFIT1 and RIG-I induction following IFNα-2b stimulation (Fig. 7F; lane 6). Interestingly, induction of ISGs was only partially reduced following 5′pppRNA treatment (2.2-fold reduction of IFIT1 vs. siCtrl; Fig. 7F; lane 5 vs. 2); this IFN-independent activation of innate signaling was sufficient to reduce viral NS1 expression by 2.4-fold (Fig. 7E; lane 9 vs. 8). Thus, in lung epithelial A549 cells, 5′pppRNA treatment can efficiently inhibit influenza H1N1 replication in a RIG-I-dependent manner and stimulate an antiviral and inflammatory response independently of IFN signaling to limit influenza infection in vitro.
To determine the potential of 5′pppRNA in vivo, C57Bl/6 mice were inoculated intravenously with 5′pppRNA (25 µg) in complex with the in vivo-jetPEI transfection reagent. 5′pppRNA stimulated a potent immune response in vivo characterized by IFNα and IFNβ secretion in the serum and lungs (Fig. S4A) as well as antiviral gene up-regulation (Fig. S4B). The response was potent and rapid with serum IFNβ levels increased ∼20-fold compared to basal levels, as early as 6 h post administration (Fig. S4A; top left panel). The immune activation observed in vivo correlated with an early and transient recruitment of neutrophils to the lungs along with a more sustained increase in macrophages and dendritic cells populations (Fig. S4C).
Next, to determine the antiviral potential of 5′pppRNA in vivo, mice were treated with 5′pppRNA 24 h before (day −1), and on the day of infection (day 0) with a lethal inoculum of H1N1 A/PR/8/34 Influenza. Whereas all untreated, infected mice succumbed to infection by day 11, all 5′pppRNA-treated mice fully recovered (100% survival) (Fig. 8A). Overall, weight loss was similar between the two groups (Fig. 8B), although a noticeable delay of 2–3 days in the onset of weight loss was observed in 5′pppRNA-treated animals; treated mice then fully recovered within 12–14 days (Fig. 8B). Influenza replication in the lungs was monitored by plaque assay over the course of infection with virus titers in the lungs of untreated mice reaching a maximum at day 3 post-infection (Fig. 8C). A decrease in virus titer was noted by day 9 post-infection, possibly correlating with the onset of adaptive immunity, and all animals succumbed to influenza infection by day 11 (Fig. 8A). Interestingly, 5′pppRNA treatment inhibited influenza virus replication in the lungs early after infection, within the first 24–48 h (Fig. 8C; Day 1); by day 3, virus titers in the lung had increased, although influenza titers were still ∼10-fold lower compared to titers in untreated mice (Fig. 8C; Day 3). By day 9, the 5′pppRNA-treated animals had controlled the infection, as demonstrated by the decrease in viral titers. Continuous administration of 5′pppRNA at 24 h intervals post-infection had an additive therapeutic effect that further delayed viral replication (Fig. 8D; 3 versus 2 doses of 5′pppRNA), indicating that antiviral immunity may be sustained with repetitive administration of 5′pppRNA. Furthermore, therapeutic administration of 5′pppRNA also controlled influenza viral replication; although prophylactic treatment was most effective at blocking influenza dissemination in the lung, administration of the RNA agonist on day 1 and day 2 after influenza infection also reduced viral lung titers by ∼10-fold (Fig. 8E).
The antiviral response triggered by 5′pppRNA in vivo was dependent on an intact RIG-I signaling; serum IFNβ release was abolished in MAVS−/− mice, whereas the absence of TLR3 did not affect 5′pppRNA-induced IFNβ release (Fig. 8F). In agreement, MAVS−/− mice treated with 5′pppRNA did not control influenza lung titers (5-fold increase vs. wt mice) and the titer was comparable to untreated wt mice (Fig. 8G). To determine whether 5′pppRNA treatment was sufficient to protect against influenza in the absence of IFN signaling, IFNα/βR−/− mice were treated or not with 5′pppRNA and challenged with influenza H1N1 virus. While untreated IFNα/βR−/− animals succumbed to infection, 40% of the animals that received 5′pppRNA treatment survived, suggesting that an IFN-independent effect of 5′pppRNA functioned in the absence of the IFN response. Thus, intravenous administration of 5′pppRNA stimulated a potent and rapid immune response in vivo that delayed influenza H1N1 virus replication in the lungs of infected animals and rescued mice from a lethal inoculation of influenza H1N1.
To further evaluate the effect of RNA agonist administration on influenza-mediated pathology, histological sections of lungs from untreated and treated mice were prepared and analysed. 5′pppRNA treatment alone was characterized by a modest and rare leukocyte-to-endothelium attachment; mixed leukocyte populations (mononuclear/polymorphonuclear) infiltrated the perivascular space at 24 h after injection (data not shown) but the infiltration resolved and was limited to endothelial cell attachment at 3 and 8 days after intravenous administration (Fig. 9A). Influenza virus infection induced severe and extensive inflammation and oedema in the perivascular space and the bronchial lumen at day 3 post-infection. In animals receiving the RNA agonist, influenza triggered a mild and infrequent inflammation that did not extend to the bronchial lumen at day 3 post-infection. Epithelial degeneration and loss of tissue integrity were more severe in the lungs of untreated, infected animals and correlated with epithelial hyperplasia observed at later times, compared to the lungs of animals treated with 5′pppRNA. Inflammation and epithelial damage progressed in untreated mice by day 8 (Fig. 9E), and correlated with increased virus titer in the lungs (Fig. 8C); inflammation and epithelial damage was consistently less apparent in agonist-treated mice. Strikingly, the surface area of the lungs affected by pneumonia was significantly reduced in 5′pppRNA-treated mice compared to non-treated mice – on day 3, 16% vs 35%; day 8, 41% vs 73% (Fig. 9C; bottom panel). Overall, influenza-mediated pneumonia was less severe in animals administered 5′pppRNA before influenza challenge, demonstrating that 5′pppRNA possesses an antiviral effect in vivo that limits influenza replication in the lung, limits lung damage and prevents influenza-mediated pneumonia and mortality.
RIG-I agonists are attractive potential antiviral agents, as triggering the innate cytosolic RIG-I pathway mimics the earliest steps of immune recognition and response to viral pathogens. In the present study, a short 5′pppRNA agonist of RIG-I derived from the 5′ and 3′ UTRs of the VSV genome stimulated an antiviral response that protected human lung epithelial A549 cells or human PBMCs from challenge with several viruses, including DENV, Influenza, HIV, VSV, HCV and Vaccinia virus. Intravenous administration of the 5′pppRNA agonist in mice stimulated an antiviral state in vivo that protected animals from lethal influenza virus challenge and controlled influenza virus-mediated pneumonia. Analysis of the dynamics of the host transcriptome following 5′pppRNA stimulation was characterized by antiviral and inflammation related gene expression patterns with transcriptional nodes of genes regulated by IRF, NF-κB, and STAT families. Virtually all of the genes activated by IFNα-2b were encompassed within the 5′pppRNA transcriptome; bioinformatics analysis also identified distinct gene patterns and functional processes that were uniquely induced or inhibited by 5′pppRNA. Because of its potency both in vitro and in vivo, 5′pppRNA represents a specific and powerful trigger of innate immunity and a novel approach to antiviral therapy.
For the first time, an RNA-based agonist of RIG-I was shown to block the replication of multiple viruses; this broad-spectrum antiviral activity of 5′pppRNA is attributable in part to a potent stimulation of the inflammatory and antiviral response driven by the early induction of IRF, NF-κB, STAT, chemokines and pro-inflammatory cytokine genes. In parallel, we also observed an inhibition of genes involved in TGF-β signaling. Because of the immunosuppressive nature of the TGF-β pathway [42], inhibition of this transcriptional node could further potentiate immune activation in response to 5′pppRNA agonist. The emergence of apoptosis and ubiquitin signaling nodes at later times (24 h) suggests a role for cell death and ubiquitin-based signal modification in the antiviral response.
5′pppRNA stimulation of RIG-I triggered a complete IFN response. At 24 h post treatment, 5′pppRNA induced the expression of 97% of the genes stimulated by IFNα-2b treatment and the magnitude of ISG induction by 5′pppRNA was enhanced compared to IFNα-2b profile. Among the ISGs, the tripartite motif containing (TRIM) proteins, the IFITM proteins, MX1 and viperin exemplify the range of ISGs induced by RIG-I and all have been implicated as inhibitors of HIV-1, Influenza, VSV, West Nile, Dengue, and HCV [43]–[53]. In a recent study, high throughput screening of antiviral effectors identified a panel of broadly acting antiviral molecules, with the combined expression of multiple ISGs providing additive inhibitory effects against HCV replication [29]. Of the 28 validated antiviral ISGs identified, 19 were induced by 5′pppRNA in A549, including IRF1, RIG-I, Mda5, IFITM3.
The transcriptome analysis also identified a distinct subset of 968 genes specifically induced by 5′pppRNA - and not IFNα-2b - that additively or synergistically enhanced the antiviral response stimulated by 5′pppRNA treatment. Bioinformatics analysis identified a unique functional signature with up-regulated genes involved in inducing a wider range of signaling pathways and bridging innate and adaptive immune responses. The importance of genes uniquely induced by 5′pppRNA is highlighted by the antiviral response that limited influenza infection in vitro and in vivo, even in the absence of functional type I IFN signaling. We speculate that the extended range of genes induced by 5′pppRNA, compared to IFNα-2b, reflect the activation of multiple signaling pathways downstream of RIG-I/MAVS [10], versus the more-limited transactivation potential of the IFN-regulated JAK-STAT axis [26].
Type III IFNs (IL29, IL28A, IL28B) were among the most highly stimulated genes uniquely up-regulated in response to 5′pppRNA. Recent studies have demonstrated that both type I and type III IFNs activate similar components of the JAK-STAT pathways, although type III IFNs were shown to prolong the activation of JAK-STAT signaling and induce a delayed and stronger induction of ISGs, compared to type I IFNs [54]. IFNλs have been increasingly implicated in antiviral therapies: 1) IFNλ administration in mice stimulated expression of Mx1 and protected IFNαR−/− mice from lethal influenza challenge [55]; 2) IL29 blocked HIV-1 replication by inhibiting virus integration and post-transcriptional events [56]; and 3) the combination of IL29 and IFNα or IL29 and IFNγ enhanced the induction of antiviral genes and effectively inhibited HCV and VSV replication [57]. In addition, polymorphism in or near IFNλ3 gene correlated with spontaneous or treatment-induced clearance of Hepatitis C infection and IFNλ therapy is now actively investigated for the treatment of HCV [58]. Altogether these results indicate that specific induction of IFNλ by 5′pppRNA may contribute to immunity at the site of viral infection.
Of note, negative regulators of the innate immune response were also detected after 5′pppRNA stimulation. In addition to the induction of SOCS1, USP18, and IFIT1 by both 5′pppRNA and IFNα-2b treatment, unique negative regulators activated exclusively by 5′pppRNA were also identified: SOCS3 contributes to the inhibition of the JAK/STAT signaling [59], and hence limits the amplification of the IFN response; A20 and IκBα inhibit the activation of the NFκB signaling complex [60], [61], which would prevent excessive inflammation. This observation suggests that targeting an upstream viral sensor may provide activation as well as negative feed-back regulation to terminate the immune response and prevent uncontrolled inflammation; as such, this approach may offer an advantage over IFN therapy in terms of limiting potential toxic side-effects.
Activation of the RIG-I signaling pathway using 5′pppRNA also induced an integrated set of genes and pathways that can efficiently bridge the innate and adaptive immune responses and utilize multiple arms of both systems. 5′pppRNA mobilized genes that enhance trafficking of immune cells such as neutrophils, monocytes, naïve and memory T cells and B cells, including CCL17, CCL20, CXCL10, CCL3, CCL5 and many others. We also observed the induction of genes important for the activation of the effector arm of the adaptive immune system such as IL-6, which has been shown to enhance CD8+ T cell survival and killing potential [62]. These cytokines and chemokines certainly play a role in initiating the innate and adaptive immune cell response, which is critical for the generation of efficient immunity against multiple viral infections in vivo.
Intravenous administration of the RIG-I agonist stimulated a potent immune response in vivo that reached the lungs and prevented mortality associated with virus challenge. Histopathology analysis revealed diminished influenza-mediated lung damage and recruitment of inflammatory cells in infected lungs following 5′pppRNA treatment. The rapid control of virus replication, as demonstrated by significantly reduced virus titers in the lungs within 3 days of virus inoculation, may have prevented excessive immune cell recruitment early after infection. Influenza infection generates a complex pathogenesis mediated in part by viral- and immune-mediated damage [63]; therefore, the activation and recruitment of limited numbers of specific immune cell types, such as neutrophils, alveolar macrophages and dendritic cells may generate a beneficial antiviral microenvironment that additionally favor the initiation of adaptive immune response, which would eventually contribute to the in vivo efficacy of the agonist.
Other groups have recently reported that 5′pppRNA induces a protective RIG-I mediated antiviral response that inhibits the replication of influenza virus [64]–[66]. The inhibition of influenza infection by 5′pppRNA was dependent on the 5′ppp moiety and the secondary IFN response was crucial for mounting an effective antiviral response [64], [65]. Recently, a short RNA molecule with dual functionality was developed - a siRNA against influenza NP gene and an agonist of the RIG-I pathway [66]. This 5′pppRNA inhibited influenza infection in vitro and in vivo, but the contribution of RIG-I activation to the inhibition of influenza was not demonstrated. The defective interfering RNA produced during Sendai virus life cycle is the best characterized natural RIG-I ligand and is known to induce strong inflammatory response [67]. Interestingly, this molecule has adjuvant potential and could stimulate an antibody-dependent response directed to influenza antigens. Compared to these earlier studies, we adopted a systems approach to provide biochemical, transcriptional and biological mechanistic explanations for the antiviral efficacy observed in vitro and in vivo.
Activating natural host defense to prevent establishment and dissemination of viral infection is a valuable alternative strategy to antiviral drugs that specifically target viral processes. Interferon therapy has been used in the clinic for over two decades and has proven effective in the treatment of certain viral infections, mainly Hepatitis B and Hepatitis C [58], as well as malignancies and autoimmune diseases [68], [69]. However, IFN therapy is also associated with significant side effects that limit its use [70]. PolyI:C, another dsRNA immune modulator, is also being tested in vitro and in vivo and has demonstrated efficacy against respiratory infections [71]–[73]. Along with antiviral drugs, vaccination is the primary approach to reduce morbidity and mortality associated with viral infection. Increasing the immunogenicity of vaccines with molecular adjuvants eliciting cytokines, co-stimulatory molecules, or immunomodulatory factors enhance the vaccine-elicited immune responses. 5′pppRNA has the advantage of mimicking viral recognition to trigger an immune response analogous to natural viral infection. Furthermore, the response stimulated by 5′pppRNA was reminiscent of the integrated and multipotent response elicited early following immunization by the most protective vaccine identified so far, the yellow fever YD17 vaccine [74]. Therefore, an immune modulator such as a RIG-I agonist may not only function as an antiviral therapeutic, but may also serve as a vaccine adjuvant to increase the magnitude of the antiviral immune response elicited by vaccine epitopes. Thus, the present study not only demonstrated the prophylactic and therapeutic antiviral potential of 5′pppRNA, but also opens the door to further investigation of the potential of RIG-I agonists as vaccine adjuvants.
The sequence of the 5′pppRNA was derived from the 5′ and 3′ UTRs of the VSV genome as previously described [17]. In vitro transcribed RNA was prepared using the Ambion MEGAscript T7 High Yield Transcription Kit according to the manufacturer′s instruction (Invitrogen, NY, USA). The template consisted of two complementary viral sequences containing T7 promoter that were annealed at 95°C for 5 minutes and cooled down gradually over night (5′-GAC GAA GAC AAA CAA ACC ATT ATT ATC ATT AAA ATT TTA TTT TTT ATC TGG TTT TGT GGT CTT CGT CTA TAG TGA GTC GTA TTA ATT TC-3′). The in vitro transcription reactions proceeded for 16 hours. 5′pppRNA was purified and isolated using the Qiagen miRNA Mini Kit (MD, USA). Homologous RNA without 5′ppp moiety was purchased from IDT (Integrated DNA Technologies Inc, Iowa, USA); dephosphorylation of the 5′pppRNA using CIAP (Invitrogen, NY, USA) generated identical results (data not shown). Secondary structure was predicted using the RNAfold WebServer (University of Vienna, Vienna, Austria). RNA was analysed on a denaturing 17% polyacrylamide, 7 M urea gel following digestion with 50 ng/ul of RNase A (Ambion, CA, USA) or 100 mU/ul of DNase I (Ambion, CA, USA) for 30 min.
A549 were grown in F12K (Invitrogen, NY, USA) supplemented with 10% FBS and antibiotics. MEFs were grown in DMEM supplemented with 10% FBS, non-essential amino acids, and L-Glutamine (Wisent, Quebec, Canada). WT and RIG-I −/− MEFS were kind gifts from Dr. Shizuo Akira (Osaka University, Osaka, Japan) [75]. WT, Mda5−/−, TLR3−/−, TLR7−/− MEFS were kind gifts from Dr. Michael Diamond (Washington University, St Louis, USA) [76], [77]. Lipofectamine RNAiMax (Invitrogen, NY, USA) was used for transfections in A549 according to manufacturer's instructions. For luciferase assays, transfections were performed in wt and RIG-I−/−; wt and Mda5−/−, TLR3−/−, TLR7−/− MEFs using Lipofectamine 2000 (Invitrogen, New York, USA) or jetPRIME (PolyPlus, France), respectively. Plasmids encoding GFP-ΔRIG-I, IRF-7, pRLTK, IFNα4/pGL3 and IFNβ/pGL3 were previously described [78]. MEFs were co-transfected with 200 ng pRLTK reporter (Renilla luciferase for internal control), 200 ng of reporter gene constructs, together with 5′pppRNA (500 ng/ml) or 100 ng of a plasmid encoding a constitutively active form of RIG-I (ΔRIG-I) [41]. IRF7 plasmid (100 ng) was added for transactivation of IFNα4 promoter. At 24 h after transfection, reporter gene activity was measured by Dual-Luciferase Reporter Assay, according to manufacturer's instructions (Promega, Wisconsin, USA). Relative luciferase activity was measured as fold induction (relative to the basal level of reporter gene). For siRNA knock down, A549 cells were transfected with 50 nM (30 pmol) of human RIG-I (sc-61480), IFN-α/βR α (sc-35637) and β (sc-40091) chain, or control siRNA (sc-37007) (Santa Cruz Biotechnologies, Dallas, USA) using Lipofectamine RNAi Max (Invitrogen, NY, USA) according to the manufacturer's guidelines. Treatment with 5′pppRNA was performed 48 hrs later.
Whole cell extracts were separated in 8% acrylamide gel by SDS-PAGE and were transferred to a nitrocellulose membrane (BioRad, Mississauga, Canada) at 4°C for 1 h at 100 V in a buffer containing 30 mM Tris, 200 mM glycine and 20% methanol. Membranes were blocked for 1 h at room temperature in 5% dried milk (wt/vol) in PBS and 0.1% Tween-20 (vol/vol) and then were probed with primary antibodies: anti-pIRF3 at Ser396 (EMD Millipore, Massachusetts, USA), anti-IRF3 (IBL, Japan), anti-RIG-I (EMD Millipore, Massachusetts, USA), anti-ISG56 (Thermo Fischer Scientific, Massachusetts, USA), anti-pSTAT1 at Tyr701 (Cell Signaling Technology, Inc, Massachusetts, USA), anti-STAT1 (Santa Cruz Biotechnology), anti-NS1(Santa Cruz Biotechnology), anti-pIkBα at Ser32 (Cell Signaling Technology, Inc, Massachusetts, USA), anti-IκBα (Cell Signaling Technology, Inc, Massachusetts, USA), anti-NOXA (EMD Millipore, Massachusetts, USA), anti-cleaved Caspase 3 (Cell Signaling Technology, Inc, Massachusetts, USA), anti-PARP (Cell Signaling Technology, Inc, Massachusetts, USA), anti-β-Actin (EMD Millipore, Massachusetts, USA). Antibody signals were detected by chemiluminescence using secondary antibodies conjugated to horseradish peroxidise and an ECL detection kit (Amersham Biosciences, Inc, NJ, USA)
Whole cell extracts were prepared in NP-40 lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 30 mM NaF, 5 mM EDTA, 10% glycerol, 1.0 mM Na3VO4, 40 mM β-glycerophosphate, 0.1 mM phenylmethylsulfonyl fluoride, 5 µg/ml of each leupeptin, pepstatin, and aproptinin, and 1% Nonidet P-40). WCE was then subjected to electrophoresis on 7.5% native acrylamide gel, which was pre-run for 30 min at 4°C. The electrophoresis buffers were composed of an upper chamber buffer (25 mM Tris, pH 8.4, 192 mM glycine, and 1% sodium deoxycholate) and a lower chamber buffer (25 mM Tris, pH 8.4, 192 mM glcine). Gels were soaked in SDS running buffer (25 mM Tris, pH 8.4, 192 mM glycine, 0.1% SDS) for 30 min at 25°C and were then transferred to nitrocellulose membrane (Amersham Biosciences). Membranes were blocked in PBS containing 5% milk (wt/vol) and 0.05% Tween-20 (vol/vol) for 1 h at 25°C and blotted with an antibody against IRF3 (IBL, Japan). Antibody signals were detected by chemiluminescence using secondary antibodies conjugated to horseradish peroxidise and an ECL detection kit (Amersham Biosciences, Inc, NJ, USA)
The release of human IFNα (multiple subunits) and IFNβ in culture supernatants of A549, and murine IFNα and IFNβ in serum or lung homogenate (20% w/v) from mice in response to 5′pppRNA were measured by ELISA according to manufacturer's instructions (PBL Biomedical Laboratories, Piscataway, NJ).
PBMCs were isolated from freshly collected blood using a Lymphocyte Separation Medium (Cellgro) as per manufacturer's instructions. After isolation, total PBMCs were frozen in heat-inactivated FBS with 10% DMSO. On experimental days, PBMCs were thawed, washed and placed at 37°C for 1 hr in RPMI with 10% FBS supplemented with Benzonaze™ nuclease (Novagen) to prevent cell clumping. The optimal concentration of 5′pppRNA to efficiently activate PBMC with minimal cytotoxicity was 100 ng/mL (data not shown). PBMCs were isolated from the blood of patients in a study both approved by IRB and by the VGTI Florida Institutional Biosafety Committee (2011-6-JH-1). Written informed consent approved by the Vaccine and Gene Therapy Institute Florida Inc. ethics review board (FWA#161) was provided and signed by study participants. Research was conformed to ethical guidelines established by the ethics committee of the OHSU VGTI and Martin Health System.
VSV-GFP, which harbors the methionine 51 deletion in the matrix protein-coding sequence [79], was kindly provided by J. Bell (Ottawa Health Research Institute, CA). Virus stock was grown in Vero cells, concentrated from cell-free supernatants by centrifugation, and titrated by standard plaque assay as described previously [80]. The recombinant vaccinia-GFP virus (VVΔE3L-REV), a revertant strain of the E3L deletion mutant, was kindly provided by Jingxin Cao (National Microbiology Laboratory, Public Health Agency of Canada, Winnipeg) [81], [82]. Dengue virus serotype 2 (DENV-2) strain New Guinea C was grown in C6/36 insect cells for 7 days. Briefly, cells were infected at a MOI of 0.5, and 7 days after infection, cell supernatants were collected, clarified and stored at −80°C. Titers of DENV stocks were determined by serial dilution on Vero cells, with intracellular immunofluorescent staining of DENV E protein at 24 h post-infection and denoted as infectious units per ml. Titers of Dengue virions were determined by standard plaque assay in Vero cells; plaques were fixed, stained and counted 5 days later. In infection experiments, both PBMCs and A549 cells were infected in a small volume of medium without FBS for 1 hour at 37°C and then incubated with complete medium for 24 h prior to analysis.
HIV-GFP virus is an NL4-3 based virus designed to co-express Nef and eGFP from a single bicistronic RNA. HIV-GFP particles were produced by transient transfection of pBR43IeG-nef+ plasmid into 293T cells as described previously [83], [84]. Briefly, 293T cells were transfected with 22.5 µg of pBR43IeG-nef+ plasmid by polyethylenimine precipitation. Media was replaced 14–16 h post-transfection, and viral supernatants were harvested 48 hrs later, cleared by low-speed centrifugation and filtered through a 0.45 µm low binding protein filter. High-titer viral stocks were prepared by concentrating viral supernatants 100-fold through filtration columns (Amicon), then aliquoted and stored at −80°C. Viral titers were determined by p24 level (ELISA) and TCID50; briefly, 10-fold serial dilutions of concentrated viral supernatants were used to infect PBMCs from two donors pre-activated for 3 days with 10 µg/ml of PHA. Half of the media was replaced on day 4, and 7 days after infection, supernatants were harvested and processed for p24 by ELISA. The Reed–Muench method was used to calculate the TCID50. For HIV infection, CD14+ monocytes were negatively selected using the EasySep Human Monocytes Enrichment Kit (Stem Cell) as per manufacturer's instructions. Isolated cells were transfected with 5′pppRNA (100 ng/ml) using Lyovec (Invitrogen) according to the manufacturer's protocol. Supernatants were harvested 24 h after stimulation and briefly centrifuged to remove cell debris. CD4+ T cells were isolated using EasySep™ Human CD4+ T cells Enrichment Kit (Stem Cell) according to the manufacturer's guidelines. Purified CD14+ monocytes and CD4+ T cells were allowed to recover 1 h in RPMI containing 10% FBS at 37°C with 5% CO2 before experiments. For HIV infection, anti-CD3 Ab (0.5 µg/ml) were immobilized for 2 hours in 24-well plate and CD4+ T cells were then added along with anti-CD28 Ab (1 µg/ml) to allow activation of T cells for 2 days. After activation, cells were incubated for 4 hours with supernatant of 5′pppRNA-stimulated monocytes and infected with HIV-GFP at an MOI of 0.1. Supernatant from the monocytes was left for another 4 h before adding complete medium.
HCV RNA was synthesized using the Ambion MEGAscript T7 High Yield Transcription Kit using linearized pJFH1 DNA (a generous gift Takaji Wakita; National Institute of Infectious Diseases, Shinjuku-ku) as template. Huh7 cells were electroporated with 10 mg of HCV RNA and at 5 days post-transfection, virus containing supernatant was collected, filtered (0.45 µm) and stored at −80°C (HCVcc). Huh7 or Huh7.5 cells were pre-treated with 5′pppRNA (10 ng/mL) for 24 h. Supernatants containing soluble factors induced following 5′pppRNA treatment was removed and kept aside during infection. Cells were washed once with PBS and infected with 0.5 ml of undiluted HCVcc for 4 h at 37°C; then, supernatant was added back. At 48 h post-infection, WCEs were prepared; expression of HCV NS3 protein was detected by Western blot (Abcam, Toronto, Ca)
Influenza H1N1 strain A/Puerto Rico/8/34 was kindly provided by Veronika von Messling (Duke-NUS, Singapore). Viral stock was amplified in Madin-Darby canine kidney (MDCK) cells and virus titer was determined by standard plaque assay [85]. Cells were infected in 1 ml medium without FBS for 1 hour at 37°C. Inoculum was aspirated and cells were incubated with complete medium for 24 hours, prior to analysis. For viral infections, supernatants containing soluble factors induced following 5′pppRNA treatment was removed and kept aside during infection. Cells were washed once with PBS and infected in a small volume of medium without FBS for 1 h at 37°C; then supernatant was added back for the indicated period of time.
The percentage of cells infected with VSV, Vaccinia and HIV was determined based on GFP expression. The percentage of cells infected with Dengue was determined by standard intra-cellular staining. Cells were stained with a mouse IgG2a mAb specific for DENV-E-protein (clone 4G2) followed by staining with a secondary anti-mouse antibody coupled to PE (Jackson Immuno Research). PBMCs infected with DENV were first stained with anti-human CD14 Alexa Fluor 700 Ab (BD Biosciences). Cells were analyzed on a LSRII flow cytometer (Becton Dickinson). Compensation calculations and cell population analysis were done using FACS Diva.
C57Bl/6 mice (8 weeks-old) were obtained from Charles River Laboratories. MAVS−/− and WT (mixed 129/SvEv-C57Bl/6 background) were obtained from Z. Chen (The Howard Hughes Medical Institute, US). TLR3−/− mice were obtained from Taconic. IFNα/βR−/− mice were bred on a C57Bl/6 background. For intra-cellular delivery, 25 ug of 5′pppRNA was complexed with in vivo-jetPEI (PolyPlus, France) at an N/P ratio of 8 as per manufacturer's instructions and administered intravenously via tail vein injection. Unless otherwise indicated, 5′pppRNA was administered on the day prior to infection (Day −1) and on the day of infection (Day 0). Mice under 4% isoflurane anesthesia were infected intra-nasally with 500 PFU of Influenza A/PR/8/34 (Day 0). For viral titers, lungs were homogenized (20% wt/vol) in DMEM and titers were determined by standard plaque assay as previously described [85]. Briefly, confluent Madin-Darby Canine Kidney Cells (MDCK) were incubated with 250 µL of serial Log10 dilutions for 30 minutes, the sample was aspirated, and cells overlaid with 3 ml of 1.6% agarose in DMEM. Plaques were fixed and counted 48 h later. All animal experimentations were performed according to the guidelines of the Canadian Council on Animal Care and approved by the McGill University Animal Care Committee. The IFNα/βR−/− animal experimentations were approved by the INRS Institutional Animal Care and Use Committee.
All five lobes of the lungs were collected and fixed in 10% neutral-buffered formalin for 24 h. The organs were paraffin-embedded and 4 µm sections were cut and stained with hematoxyline and eosin staining (H&E). The slides were analysed by a board-certified independent veterinary pathologist.
The kinetics and the comparison to IFNα-2b were performed as two separate experiments. A549 cells were stimulated with either 5′pppRNA (10 ng/ml) or IFNα-2b (100 IU/ml or 1000 IU/ml) for designated times. IFNα-2b (Intron A) was purchased from Schering Plough (Kenilworth, NJ). Cells were collected and lysed for RNA extraction (Qiagen, Valencia, CA, USA). Reverse transcription reactions were performed to obtain cDNAs which were hybridized to the Illumina Human HT-12 version 4 Expression BeadChip according to the manufacturer's instruction, and quantified using an Illumina iScan System. The data were collected with Illumina GenomeStudio software. First, arrays displaying unusually low median intensity, low variability, or low correlation relative to the bulk of the arrays were discarded from the rest of the analysis. Quantile normalization, followed by a log2 transformation using the Bioconductor package LIMMA was applied to process microarrays. To account for variability between batches, the data were adjusted using the ComBat procedure (http://dx.doi.org/10.1093/biostatistics/kxj037). Missing values were imputed with the R package (http://cran.r-project.org/web/packages/impute/index.html). In order to identify differentially expressed genes between treated and controls (untreated) samples, the LIMMA package [86] from Bioconductor was used. For data mining and functional analyses, genes that satisfied a p-value (<0.001) with ≥2 fold change (up or down) were selected. Probes that do not map to annotated RefSeq genes and control probes were removed. The expected proportions of false positives (FDR) were estimated from the unadjusted p-value using the Benjamini and Hochberg method [87]. All network analysis was done with Ingenuity Pathway Analysis (IPA: Ingenuity systems, Redwood City, CA). The differentially expressed genes selected based on above criteria were mapped to the ingenuity pathway knowledge base with different colors (red: up-regulated; green: down-regulated). The significance of the association between the dataset and the canonical pathway was measured in two ways: (1) A ratio of the number of genes from the dataset that map to the pathway divided by the total number of genes that map to the canonical pathway was displayed; (2) by over-representation analysis Fisher's exact test was used to calculate a p-value determining the probability that the association between the genes in the dataset and the canonical pathway is explained by chance alone. The pathways were ranked with −log p-values. Microarray data have been deposited in the NCBI Gene Expression Omnibus.
Total RNA was isolated from cells using RNeasy Kit (Qiagen, Valencia, CA, USA). Spleen and lungs were homogenized in RLT buffer and RNA isolated as per manufacturer's instruction. 1 ug of RNA was reverse transcribed using High-Capacity cDNA Reverse Transcription Kits from Applied Biosystems according to manufacturer's instructions. Parallel reactions without reverse transcriptase were included as negative controls. Relative amount of an intracellular RNA of interest was quantified by real-time PCR on a 7500 fast real-time PCR system and expressed as a fold change using SYBR Green (Roche) according to the manufacture's protocol. All data presented are relative quantification with efficiency correction based on the relative expression of target genes versus GAPDH as reference gene. Primers sets used for these studies are presented in Table S1.
|
10.1371/journal.pntd.0006386 | Prospective evaluation of a rapid diagnostic test for Trypanosoma brucei gambiense infection developed using recombinant antigens | Diagnosis and treatment are central elements of strategies to control Trypanosoma brucei gambiense human African trypanosomiasis (HAT). Serological screening is a key entry point in diagnostic algorithms. The Card Agglutination Test for Trypanosomiasis (CATT) has been the most widely used screening test for decades, despite a number of practical limitations that were partially addressed by the introduction of rapid diagnostic tests (RDTs). However, current RDTs are manufactured using native antigens, which are challenging to produce.
The objective of this study was to evaluate the accuracy of a new RDT developed using recombinant antigens (SD BIOLINE HAT 2.0), in comparison with an RDT produced using native antigens (SD BIOLINE HAT) and CATT. A total of 57,632 individuals were screened in the Democratic Republic of the Congo, either passively at 10 health centres, or actively by 5 mobile teams, and 260 HAT cases were confirmed by parasitology. The highest sensitivity was achieved with the SD BIOLINE HAT 2.0 (71.2%), followed by CATT (62.5%) and the SD BIOLINE HAT (59.0%). The most specific test was CATT (99.2%), while the specificity of the SD BIOLINE HAT and SD BIOLINE HAT 2.0 were 98.9% and 98.1%, respectively. Sensitivity of the tests was lower than previously reported, as they identified cases from partially overlapping sub-populations. All three tests were significantly more sensitive in passive than in active screening. Combining two or three tests resulted in a markedly increased sensitivity: When the SD BIOLINE HAT was combined with the SD BIOLINE HAT 2.0, sensitivity reached 98.4% in passive and 83.0% in active screening.
The recombinant antigen-based RDT was more sensitive than, and as specific as, the SD BIOLINE HAT. It was as sensitive as, but slightly less specific than CATT. While the practicality and cost-effectiveness of algorithms including several screening tests would need to be investigated, using two or more tests appears to enhance sensitivity of diagnostic algorithms, although some decrease in specificity is observed as well.
| Sleeping sickness, or human African trypanosomiasis (HAT), is a neglected tropical disease that represents a risk to more than seventy million people in Sub-Saharan Africa. Most cases are caused by infection with Trypanosoma brucei gambiense. Diagnosis of HAT relies on the identification of suspected cases by serological methods, which include recently developed rapid diagnostic tests (RDTs). Current RDTs are produced using native antigens that are purified from live parasites in a laborious and dangerous process. The objective of this study was to evaluate the performance of a new RDT made using recombinant antigens, by screening people in fifteen endemic sites in the Democratic Republic of the Congo. The new RDT was found to be more sensitive than, and as specific as, the reference RDT made using native antigens. It was also more sensitive than CATT, a serological test that has been widely used for decades. While one third of HAT cases were correctly diagnosed by all tests, the other cases were only identified by one or two of the tests. In order to enhance case detection and accelerate elimination of HAT, there may be a need to explore diagnostic strategies that combine two or more screening tests.
| Human African trypanosomiasis (HAT) is a vector-borne, neglected tropical disease, which puts 70 million people living in sub-Saharan African countries at risk [1]. The most common form of the disease is caused by infection with the protozoan parasite Trypanosoma brucei gambiense (g-HAT), which in 2015, accounted for more than 97% of all reported HAT cases [2]. Patients progress from an early disease stage that is characterized by the presence of trypanosomes in the blood and lymphatic system, to a late stage that is associated with the invasion of the central nervous system by parasites [3]. If left undiagnosed and untreated, the disease is generally fatal, although asymptomatic cases and others that progress spontaneously to apparently pathogen-free status have been reported [4].
Identification of serological suspects is the main entry point into diagnostic algorithms for g-HAT. The card agglutination test for trypanosomiasis (CATT/T.b. gambiense) has been the most commonly used screening test for g-HAT. It detects antibodies using a suspension of purified, fixed and stained bloodstream-form trypanosomes expressing LiTat 1.3 variant surface glycoprotein (VSG), a predominant variant antigen of T.b. gambiense [5]. While CATT has played a central role in the control of HAT, its large-scale implementation for passive screening in health facilities in remote locations has been limited due to operational challenges such as the need for an agitator, electricity and refrigeration. In some settings, the sensitivity and specificity of CATT have also been reported as being problematic [6].
In an effort to address the shortcomings of CATT, two rapid diagnostic tests (RDTs) that detect host antibodies have recently been developed, the HAT Sero-K-SeT manufactured by Coris BioConcept (Belgium), and the SD BIOLINE HAT, hereinafter referred to as “RDT1”, produced by Alere/Standard Diagnostics (SD, South Korea), which include the same two antigens, VSG LiTat 1.3 and VSG LiTat 1.5. These RDTs were evaluated in retrospective studies, with very promising performance results [7,8]. Evaluation of a prototype of the RDT1 in a prospective study in three endemic countries, Angola, the Democratic Republic of the Congo (DRC) and the Central African Republic, showed that the sensitivity of the RDT was not different from the sensitivity of CATT, while its specificity was 1.3% lower [9]. A prospective study using the HAT Sero-K-SeT also reported excellent performance [10]. A comparison of both RDTs in an independent study using stored plasma samples collected in Guinea and Côte d’Ivoire concluded that there was no difference in diagnostic accuracy between the two tests [11]. The RDTs have now been introduced in multiple HAT endemic countries, where they are being used in HAT elimination programmes. However, production of the native antigens used in the manufacture of the RDTs remains a challenge, as it relies on a labor-intensive, costly and risky process that involves inoculating rats with human-infective trypanosomes. To address this challenge, and to improve standardization and quality of manufacturing, a new RDT that is produced exclusively using recombinant antigens, the SD HAT BIOLINE 2.0 (“RDT2”), has been developed in a partnership facilitated by the Foundation for Innovative New Diagnostics (FIND).
The primary objective of this study was to evaluate the diagnostic accuracy of RDT2 in a multi-centric, prospective study in the DRC, and to demonstrate its non-inferiority to RDT1. As a secondary objective, the accuracy of RDT2 was compared to that of CATT.
Study participants were enrolled from 6 June 2015 to 5 January 2016 in the Bandundu Province of the DRC by passive screening in ten health facilities, and by active screening using five mobile teams of the Programme National de Lutte contre la Trypanosomiase Humaine Africaine (PNLTHA) of the DRC (Table 1). In the health facilities, participants were enrolled among patients presenting themselves or referred from other health facilities after suspicion of HAT, and among relatives who accompanied patients. During active screening, anybody who presented to the mobile team was eligible for enrolment in the study. Study sites were visited by an external monitor prior to commencement of the study to verify that they were adequately prepared and personnel properly trained, and during the study to verify that the protocol was being adhered to. HAT cases were defined as subjects in whom trypanosomes were demonstrated by microscopy in either lymph node aspirate, blood or cerebrospinal fluid (CSF). All positive parasitology results were verified by the site supervisor. Cases were classified as early stage when no trypanosomes were observed in their CSF, and the CSF white cell count was lower than or equal to 5 cells/μL, while those with trypanosomes in the CSF and/or a cell count above 5 cells/μL were classified as late stage [12]. Controls were subjects living in the same areas as cases, with no known history of HAT infection, and who were either negative with all three screening tests, or who were positive with one or several screening tests, but in whom no parasites were detected in any body fluid. Clinical signs and symptoms were not considered exclusion criteria for controls.
The RDT2 (SD, South Korea) is an immuno-chromatographic test for qualitative detection of antibodies of all isotypes (IgG, IgA and IgM). It includes a nitrocellulose membrane strip with two test regions (T1 and T2) that are pre-coated with two recombinant antigens. T1 is coated with Invariant Surface Glycoprotein 65–1 (ISG65) expressed in Escherichia coli [13] and T2 with the N-terminal domain of Variant Surface Glycoprotein LiTat 1.5 (VSG LiTat 1.5) produced using a Baculovirus expression system. A procedural control line (C) is also included. The test is stable for at least 24 months at 40°C, or 5 weeks at 55°C. The test is performed in the same way as RDT1, as described by Lumbala et al. [14]. In summary, a sample of 20 μl of whole blood is taken from a finger prick and transferred into a sample well using a disposable plastic capillary tube, and 4 drops (approximately 120 μl) of test diluent are then added. The sample flows along the membrane by capillarity, passing through the test regions T1 and T2. Results are read after 15 to 20 minutes by comparing the intensity of the test lines against a colour chart provided by the manufacturer. A result is considered positive when the control line C and either one or both T1 and T2 test lines are visible (regardless of their intensity), negative when only the C line is observed, and invalid if the C line is not observed. In active screening, all participants found positive with a HAT screening test were also tested for malaria using an RDT (SD BIOLINE Ag P.f), while in passive screening, all participants were tested with a malaria RDT (S3 Table). However, results of malaria RDTs were only recorded for subjects who were eligible for enrolment (see below). Those who tested positive for malaria were examined, and if necessary, treated in line with national guidelines.
Three screening tests (CATT, RDT1 and RDT2) were performed on finger-prick blood from each subject who presented to mobile teams, any subject who presented to a health facility with symptoms indicative of HAT, and accompanying individuals who consented to participate in the study. The results of screening tests were read by two independent laboratory technicians or nurses, and the results recorded separately. To avoid overburdening study teams and to keep the study design as simple as possible, CATT was only performed on whole blood, and not on diluted plasma. Similarly, the trypanolysis test was not performed during this study, as this would have required additional resources to collect and transport samples for analysis, which at that time could not be performed in DRC.
In both active and passive enrolment, any subject who was positive with at least one of the screening tests, or who showed symptoms highly suggestive of HAT, was eligible for immediate enrolment in the subsequent parasitological work-up. Written informed consent was sought from these subjects prior to enrolment. Any individual who declined to participate in the study was managed according to the standard procedures of the PNLTHA. Individuals who were negative to all three screening tests and who had no symptoms highly suggestive of HAT were not investigated further. Persons with palpable cervical lymph nodes had a lymph node aspirate taken and examined for motile parasites by bright field microscopy. A sample of 5 ml of venous blood was collected from each participant in a heparinized tube. Three hundred μL of blood was used to perform the capillary tube centrifugation (CTC) test (4 capillary tubes of approximately 65–70 μL) [15]. If the result of CTC was negative, 500 μL of whole blood was used to perform the mini anion exchange centrifugation technique (mAECT-wb) [16] and the remaining volume of blood (4.2 ml) was centrifuged to perform mAECT on buffy coat (mAECT-bc) as described by Camara et al. [17]. Since the mAECT-bc procedure had only been evaluated in one study in DRC, we took advantage of this study to collect some additional performance data to compare it with mAECT-wb, even though mAECT was only performed on a subset of cases. A lumbar puncture was performed on all HAT cases confirmed by any of the parasitological methods, as well as on other participants with clinical signs that were strongly suggestive of HAT, according to routine procedures. Parasitological examination of CSF was done using the modified single centrifugation technique [16]. The technicians who performed the tests were employees of the PNLTHA, with experience in performing routine parasitological tests for detection of trypanosomes. Training of personnel of mobile teams and fixed health facilities included how to perform, read and interpret results of the RDTs, the study protocol and related SOPs, completion of CRFs and data management. Any positive or doubtful parasitology result was verified and confirmed by the site supervisor. Participants with any missing screening test or parasitology results were excluded from the study. All the HAT cases that were identified during the study were treated according to national guidelines.
Two levels of blinding were adopted. During the initial screening of participants using blood from a finger prick, three health workers were each responsible for performing one of the three screening tests. The health workers operated independently (but used blood from the same finger prick), without exchanging results (first level of blinding), and did not have access to any clinical information. A supervisor was responsible for collecting results of the tests and deciding whether or not to collect venous blood for parasitological tests. Samples of venous blood were labelled with blinding codes by the supervisor (second level of blinding). The same codes were used to identify all samples collected from the participants (i.e. blood, lymph node aspirate, CSF) and constituted the anonymisation process that was maintained throughout the entire study.
Participant information and test results were recorded at study sites on paper case report forms, which were transferred to PNLTHA in Kinshasa for double data entry using a web-based clinical data management platform (VisionForm). Since two independent readings were available for each test and each sample, an approach based on bootstrapping resampling [18] was adopted: At each iteration, a random sequence of readings from the available data was generated (one reading per patient and per test) and used to calculate the performance metrics. This process was repeated (2,000 iterations per metric) to generate an empirical distribution of values for each metric, from which it was possible to derive values for the sample mean and 95% confidence intervals as bootstrapped percentiles.
Estimates of sensitivity and specificity were calculated for each screening test, on the overall data, and stratified by disease stage and by screening method (i.e. active and passive screening). The diagnostic performance of each antigen in the RDTs was also calculated. Sensitivity and specificity were defined as the percentage of HAT cases that were found positive and the percentage of controls that were found negative, respectively. Accuracy was assessed by calculating Youden’s index [19]. To evaluate the agreement between readers, Cohen’s Kappa factor was calculated. The statistical analysis was performed in the R statistical environment (version 3.2.3).
The sample size was calculated to demonstrate non-inferiority of the sensitivity and specificity of RDT2 in comparison to RDT1. Based on the sensitivity of RDT1 of 92.0% that was reported by Lumbala et al. [14], using a non-inferiority margin of 8%, a confidence level of 5% and a power of 80%, the required number of HAT cases was calculated to be at least 143. Based on the same report, the expected specificity of RDT1 was 97.1%. Using a non-inferiority margin of 1%, a confidence level of 5% and a power of 90%, it was calculated that a minimum of 4,775 controls would be needed [20]. Based on the expected prevalence of HAT in the study area, the minimum number of subjects estimated to be screened in order to enrol 143 cases was 44,700.
The study received ethical clearance from the School of Public Health of the University of Kinshasa (authorization number ESP/CE/012/2015). Participants provided written informed consent before being enrolled in the study. For children below 18 years, consent was provided by a parent or guardian. All individuals who presented at study sites during the period of enrolment and consented to being screened were eligible. Those who presented for screening but did not wish to participate in the study were screened according to the procedures of the PNLTHA. All participants’ samples were blinded and further analysed anonymously.
A total of 260 HAT cases and 56,269 controls were enrolled after screening 56,942 people. 413 individuals could not be included in the study because they did not provide informed consent (Fig 1). A total of 138 (53%) cases and 45,654 controls were enrolled by active screening, while 122 (47%) cases and 10,615 controls were enrolled by passive screening. Among cases, the early stage to late stage ratio was 4.3 in active screening and 0.53 in passive screening. The HAT prevalence was 0.30% in active and 1.13% in passive screening. On average, 255 persons were tested per day by each mobile team.
The estimates of sensitivity, specificity and accuracy of the RDT2, RDT1 and CATT tests in active screening, passive screening, and active and passive screening combined are shown in Fig 2. When the results of active and passive screening were combined, the sensitivity of the three screening tests was unexpectedly low. While RDT2 detected 71.2% [CI: 65.7%; 76.6%] of the HAT cases, CATT detected only 62.5% [CI: 56.2%; 68.4%] and RDT1 only 59.0% [CI: 53.0%; 64.6%] of the cases. Sensitivity was particularly low in active screening, with only 54.8% [CI: 46.8%; 63.5%], 51.8% [CI: 43.1%; 59.9%] and 49.2% [CI: 40.9%; 57.6%] of cases being detected by the RDT2, CATT and RDT1, respectively. In passive screening, the three tests were more sensitive, with RDT2 achieving the highest sensitivity (90.1% [CI: 84.7%; 95.3%]), followed by CATT (74.6% [CI: 66.7%; 82.3%]) and RDT1 (70.0% [CI: 61.5%; 77.9%]).
CATT had the best specificity (99.2% [CI: 99.1%; 99.2%]), followed closely by RDT1 (98.9% [CI: 98.8%; 99.0%]) and RDT2 (98.1% [CI: 98.0%; 98.2%]). With all the screening tests, specificity was significantly higher in active than in passive screening. In active screening, specificity was highest with CATT (99.5% [CI: 99.5%; 99.6%]), which was followed by RDT1 (99.4% [CI: 99.3%; 99.5%]) and RDT2 (99.1% [99.0%; 99.2%]). Similarly, in passive screening, specificity was highest with CATT (97.6% [CI: 97.3%; 97.9%]), while lower results were obtained with RDT1 (96.7% [CI: 96.3%; 97.0%]) and RDT2 (93.7% [CI: 93.2%; 94.2%]).
RDT2 had the highest accuracy (69.3% [CI: 63.5%; 74.5%]), followed by CATT (61.7% [CI: 55.7%; 67.4%]) and RDT1 (57.9% [CI: 51.8%-63.7%]). All tests had a higher accuracy in passive than in active screening.
The agreement between the two technicians who read the screening tests was excellent. Cohen’s Kappa factor was above 99.8% with all the tests, both in active and passive screening.
The differences in sensitivity and specificity between two screening tests are shown in Table 2 for each possible pair of tests. The RDT2 was 12.3% [CI: 3.8%; 20.7%] more sensitive than the RDT1 when the results of active and passive screening were considered together. The difference was particularly pronounced in passive screening, where the sensitivity of RDT2 was 20.1% [CI: 9.4%; 29.8%] higher than that of RDT1. By contrast, there was no evidence of a difference in sensitivity between RDT1 and RDT2 in active screening (+5.6% [CI: -7.4%; 18.7%]). The objective of non-inferiority using a margin of 8% was met in both active and passive screening. The RDT2 was also more sensitive than CATT (+8.7% [CI: 1.0%; 16.6%]) when the results of active and passive screening were combined, and again, this effect was stronger in passive than in active screening (+15.5% [CI: 7.5%; 24.3%]). There was no evidence of a difference in sensitivity between RDT1 and CATT in both active (-2.6% [CI: -15.2%; 9.5%]) and passive screening (-4.6% [CI: -13.3%; 5.0%]).
The RDT2 was 0.83% less specific [CI: -0.96%; -0.70%] than RDT1 when results of active and passive screening were combined, which was within the non-inferiority margin of 1%. While the difference in specificity was minimal in active screening (-0.33% [CI: -0.44%; -0.23%]), it was more pronounced in passive screening (-2.98% [CI: -3.50%; -2.48%]). The RDT2 was also less specific than CATT (-1.10% [CI: -1.22%; -0.98%]), and this difference was more pronounced in passive (-3.86% [CI: -4.37%; -3.39%]) than in active screening (-0.46% [CI: -0.56%; -0.37%]). The RDT1 was slightly less specific than CATT (-0.27% [CI: -0.37%; -0.17%]), and this difference was also more pronounced in passive (-0.88% [CI: -1.28%; -0.50%]) than in active screening (-0.13% [CI: -0.21%; -0.05%]).
Considering that RDT1 and RDT2 are each made using two different antigens, we calculated the sensitivity and specificity of individual antigens. Table 3 shows that for each RDT, individual antigens detected partially overlapping groups of HAT cases, since the sensitivity obtained with single antigens was lower than the result of the RDT. Therefore, having two antigens in these tests resulted in higher sensitivity than if only one antigen had been used. While each of the antigens in RDT2 detected almost the same number of cases and contributed almost equally to the sensitivity of this test, one of the antigens of RDT1 (native VSG LiTat 1.3) detected a larger number of cases than the other antigen (native VSG LiTat 1.5). Similarly, Table 3 shows that the individual antigens of RDT2 contributed almost equally to specificity, while in the case of RDT1, native VSG LiTat 1.3 gave a slightly greater number of false positive results than the other antigen. All antigens were significantly more sensitive in passive than in active screening. The strongest difference was observed with recombinant ISG65 and recombinant VSG LiTat 1.5, whose sensitivity was two times higher in passive than in active screening.
The three screening tests were significantly more sensitive in late stage than in early stage patients, as shown in Table 4. The strongest difference in sensitivity between stages was observed with CATT, whose sensitivity went up from 50.6% [CI: 42.7%; 58.8%] in early stage patients to 80.2% [CI: 72.3%; 87.8] in late stage patients. The RDT2 was the most sensitive in both early stage (59.8% [CI: 52.2%; 67.3%]) and late stage patients (87.9% [81.3%; 93.7%]). Similarly, all the individual RDT antigens were significantly more sensitive in late than in early stage patients. The largest difference between stages was observed with recombinant VSG LiTat 1.5 (42.1%), while the smallest difference was with native VSG LiTat 1.3 (27.0%). The most sensitive antigen in early stage patients was native VSG LiTat 1.3 (41.9% [34.0%; 49.4%]), while the most sensitive antigen in late stage patients was recombinant VSG LiTat 1.5 (80.4% [72.7%; 87.8%]).
We also calculated the diagnostic performance that would be achieved by combining two or three screening tests, with the goal of improving the overall sensitivity of screening, which is important in enhancing control of HAT, as humans are considered the main reservoirs of the disease [21]. The sensitivity and specificity of all possible combinations of two or three screening tests is shown in Fig 3. As expected, the highest sensitivity was achieved by combining all three tests (99.6% [CI: 98.7; 100.0]). This did not reach 100% because there were some differences between the two readers who interpreted test results. The most sensitive combination of two tests was RDT1 and RDT2, which detected 90.1% of cases [CI: 86.2%; 93.6%] and was markedly more sensitive than the individual tests. Lower sensitivity values were obtained by combining CATT and RDT2 (87.8% [CI: 83.7%; 91.6%]) and even more so by combining CATT and RDT1 (81.4% [CI: 76.4; 85.9]). Combining screening tests provided a greater increase in sensitivity in active than in passive screening. In active screening, sensitivity increased from 54.8% [CI: 46.8%; 63.5] with RDT2 to 83.0% [CI: 76.2%; 89.3%] when combining RDT1 and RDT2. In passive screening, this same combination achieved a remarkable sensitivity of 98.4% [CI: 95.6%; 100.0%], compared to 90.1% [CI: 84.7%; 95.3%] with RDT2 alone. In other words, combining these two RDTs would mean that only 1.6% of cases would have been missed in passive screening, while 9.9% of them would have remained undiagnosed using RDT2 only. However, using such combinations resulted in some trade-off in specificity, which went down to 96.9% [CI: 96.8%; 97.1%] when the three tests were taken together, or to 97.3% [CI: 97.2%; 97.4%] when combining RDT1 and RDT2.
The contribution of each screening test to the detection of cases and to false positive results is demonstrated using Venn diagrams in Figs 4 and 5. Fig 4 shows that for true positive results, the degree of overlap between the tests was much higher in passive than in active screening. Fig 5 shows that for false positive results, the degree of overlap between the tests was also higher in passive than in active screening, but this difference was much less pronounced than for true positive results.
Both mAECT tests were performed on 124 HAT cases. Ninety cases (72.6%) were positive by mAECT-bc, while only 64 cases (51.6%) were positive by mAECT-wb. There was a high degree of overlap between the two tests, with 58 cases detected using both methods.
The number of HAT cases identified using the different parasitological tests used in the study, as well as the corresponding positivity rates of each of the screening tests, are included as supporting information (S2 Table). This data indicates that among the three parasitological tests performed on blood samples, the positivity rates of screening tests were highest in cases identified by mAECT-bc and lowest in cases diagnosed by CTC. The positivity rates of screening tests in cases identified by examining lymph node aspirates were not significantly different from the positivity rates obtained in cases that were positive with parasitological tests performed on blood samples. The highest screening test positivity rates were obtained in cases with trypanosomes detected in the cerebrospinal fluid.
The main objective of this study, to demonstrate the non-inferiority of the sensitivity and specificity of the RDT2 in comparison to the RDT1, was successfully achieved. However, all three screening tests that were evaluated were unexpectedly insensitive, particularly in active screening, which is in contrast with earlier reports. While CATT has been extensively evaluated and used in clinical settings, and its sensitivity has been reported to range between 68.8% and 100% [22], in this study, the test missed almost half of the cases in active screening. Previous retrospective studies also reported the sensitivity of the RDT1 to be between 82% and 99.6% [8,11]. A sensitivity of 89% was reported in a prospective study of a prototype version of the RDT1 [9], while in another trial, the sensitivity of the commercialized RDT1 was 92% [14]. This apparent discrepancy could be explained by assuming that each of the three screening tests detected cases with different serological profiles, which were only partially overlapping, as evidenced by the results shown in Fig 4. The design of the study, which included three screening tests to identify suspects during enrolment, would be responsible for the low sensitivity of an individual screening test. By contrast, earlier studies only included one, or sometimes two screening tests during enrolment, and as a result, the sensitivity of screening tests could have been significantly overestimated, since cases with serological profiles that were different from the ones identified by the particular test could have been missed. Therefore, there is the need to explore the possibility of including two or more screening tests in diagnostic algorithms, in order to increase sensitivity and accelerate interruption of disease transmission, particularly by enhancing detection of patients in early stage disease. Based on the results presented here, strategies combining RDT2 with either RDT1 or CATT in active screening, and combining RDT2 with RDT1 in passive screening, could be considered to enhance case detection. In active screening, each test detected a particularly large number of cases that were missed by the other tests, and combining several screening tests would therefore result in a stronger gain in sensitivity than in passive screening. However, operational aspects would also need to be considered, and cost-effectiveness analyses may provide helpful information to select the most appropriate strategies that would ensure optimal detection of cases. In particular, there is the need to determine whether the extra complexity of the diagnostic algorithm and workload that would result from performing two or more screening tests and having more serological suspects to test by microscopy would cause a significant reduction in the number of people screened by a mobile team in a day, and balance it against the gain in detection of a larger proportion of cases among the people screened. Performing several screening tests would also be a logistical challenge in terms of transportation and storage of tests. Some patients could also refuse to have two or more tests performed on them, an unlikely possibility since blood is taken from the same finger prick. If only one screening test had to be used, the results presented here support using RDT2 in order to enhance case detection, as it was more sensitive than RDT1 and CATT, in both active and passive screening settings. RDT2 was also the most sensitive test in both early and late stage patients, which indicates that the test is able to detect patients with various clinical profiles. Maximizing sensitivity would be a sensible strategy in a disease elimination context, but the marginally lower specificity of RDT2 would also need to be considered, as it would result in an increase in workload to confirm suspects, decrease in confidence in test results and would also have a negative impact on patients, since a larger number of suspects would need to undergo confirmatory testing, which often requires travelling long distances. Such limitations will become increasingly relevant as progress is being made towards elimination of the disease, since the positive predictive value of screening tests will decline along with the disease prevalence.
Alternatively, investing in the development of a new screening test that would be more sensitive than the tests that were evaluated here, and which would include multiple antigens, could be considered. Such a test might be developed by combining three or more antigens, which could include those in the RDT2, as well as other promising candidates identified in previous studies [23–28]. Other RDTs being developed using recombinant antigens will also need to be considered once they are available and their performance has been evaluated [25]. With the increasing prospects of new, safer treatments for g-HAT that would be effective for both stages of the disease [29,30], a test with high sensitivity and specificity could make a “test and treat” approach possible, without requiring any parasitological confirmation.
A number of hypotheses could be formulated to try and explain the low sensitivity of individual screening tests observed in this study, which would require further investigations. African trypanosomes are notorious for having evolved a mechanism of escaping the host immune system by regularly changing the variant surface glycoprotein (VSG) that composes their cell coat, using a large repertoire of dedicated genes [31,32]. It is therefore likely that HAT patients who have been infected recently could have raised an immune response to only a limited number of VSG antigens, while patients who are in a more advanced disease stage could harbour antibodies against a larger panel of VSGs. This could explain why screening tests that include some specific VSGs, such as the three tests evaluated here, would detect different HAT cases, and why screening tests were more sensitive in late than in early disease stage patients. Similarly, this would provide an explanation for the lower sensitivity that was observed in active screening, since the disease is generally less advanced in most cases among the people screened. Alternatively, the difference in sensitivity between early and late-stage patients could be due to higher antibody titres in the latter because of a longer period of exposure to parasite antigens, and hence stronger immune response. This explanation would better support the finding that the sensitivity of an invariant antigen like ISG65, which is expressed throughout the infection, was higher in late-stage patients. These hypotheses could be tested using animal models infected with T.b. brucei [32]. Some patients could have also been infected with trypanosome strains lacking the genes encoding the VSG antigens present in these screening tests. In particular, deletions of the gene encoding VSG LiTat 1.3 have been reported in some T.b. gambiense isolates from Cameroon [33], and such deletions could be among the factors responsible for the low sensitivity of CATT and RDT1. Although there is currently no evidence to directly support this hypothesis, it is also conceivable that these deletions could have become increasingly frequent due to the selection pressure applied by the extensive use of CATT in HAT-endemic populations. This phenomenon could have remained unnoticed, since most studies conducted until recently only included CATT during enrolment. Finally, it cannot be excluded that some HAT cases could have corresponded to false positive parasitological test results, which would have been negative with screening tests. It is likely that several of the hypotheses described here could partially explain the observed low sensitivity of screening tests that was found in this study. Other studies comparing the performance of different screening tests in various settings will hopefully help clarify this point.
The fact that RDT1 detected 49 HAT cases that were missed by CATT (Fig 4) could be explained by the presence in RDT1 of the VSG LiTat 1.5 antigen, which is not included in CATT, and also possibly by differences in test formats. On the other hand, it is noteworthy that CATT also detected 58 HAT cases that were missed by RDT1, yet RDT1 contains VSG LiTat 1.3, the antigen that is predominantly expressed by the fixed trypanosomes in the CATT test. This could be due to the nature of the CATT reagents, which in addition to VSG LiTat 1.3, would include other trypanosome antigens that could react with corresponding antibodies in the blood of HAT patients. Another explanation might be the difference in test formats, which may be associated with different binding or exposure characteristics of antigens and epitopes. While in the RDT, antigens are printed on a nitrocellulose membrane, CATT is performed by mixing a suspension containing fixed parasites with the test sample on a plasticised card. In addition, although the exact composition of the RDT buffer is unknown, it is likely to be different from the CATT buffer (phosphate buffered saline, pH 7.2 with 0.1% sodium azide), which could have an impact on antigenic binding.
In an earlier prospective study that was conducted in the DRC to evaluate the performance of RDT1, the VSG LiTat 1.5 antigen was more sensitive than the VSG LiTat 1.3 antigen (83.6% [CI: 76.3%; 89.0%] and 76.0% [CI: 68.0%; 82.5%], respectively) [14], which is in contrast to what was found in the present study. In another multi-country study that evaluated the performance of the prototype RDT1, identical sensitivity values were reported for each antigen (85.9% [CI: 79.4%; 90.6%]) [9]. While these differences may be due to slightly different study designs, they do not appear to be statistically significant, and would therefore tend to support the view that both antigens contribute equally to the sensitivity of RDT1.
While the three screening tests were highly specific in active screening, they were significantly less specific in passive screening. This difference might be due to serological differences between the two populations, with the population presenting to fixed health facilities being more likely to be infected with other pathogens that could trigger immune responses cross-reacting with the tests. Alternatively, this difference might be explained by the relatively low sensitivity of routine parasitological methods [34]. Indeed, since the HAT prevalence was higher in passive than in active screening, this population was also more likely to have included HAT patients who could have been found positive by screening tests but missed by parasitology, which would have resulted in an underestimate of the specificity of screening tests. The difference in specificity between active and passive screening could thus be an artefact related to the imperfect parasitological reference standard, rather than reflect a real difference in test specificity.
RDT1 and CATT were previously evaluated in another prospective study that was conducted in the DRC [14], which reported that the sensitivity of RDT1 (92.0% [CI: 86.1%-95.5%]) was significantly higher than that of CATT (69.1% [CI: 60.7%-76.4%]). Surprisingly, there was no evidence of any difference in sensitivity between RDT1 and CATT in the present study. The reasons for this discrepancy are unclear, and several hypotheses could be drawn. First, although the studies shared some of the sites, it is possible that the two study populations may have had significantly distinct serological profiles, resulting in different degrees of overlap between the tests. According to this hypothesis, the degree of overlap between screening tests should not be viewed as constant and specific to the tests, but instead, considered as a dynamic phenomenon that may exhibit significant variability in time and in space, depending on the population that is sampled and the underlying immune responses of individual patients. Although this hypothesis seems rather unlikely since the studies were conducted in similar populations, it would be useful to conduct additional studies to establish the reproducibility of such differences. In spite of efforts to ensure compliance with the study protocol and procedures through training, supervision and monitoring, it is still possible that some of the sites could have performed less well, which could have had an impact on study results. Alternatively, differences between these studies could be due to operational or logistical factors causing some of the tests to have a lower performance than expected. While this seems unlikely, subtle changes during the production of antigens or other components of one of the tests could have occurred and gone undetected, resulting in the lower sensitivity of some test batches. No failure to follow storage procedures was observed during the study, screening tests were used according to manufacturers’ instructions and staff performing the tests ensured that positive and negative controls (for CATT) as well as procedural controls (for RDTs) reacted according to instructions. Yet it is possible that some tests could have deteriorated within the limits of the controls, thereby affecting performance.
The mAECT-bc method [17] may be considered as a replacement of mAECT-wb, which is routinely used in the DRC and other endemic countries. Although based on a subset of participants, the data presented here are in agreement with earlier results showing a significant increase in sensitivity using mAECT-bc. In a first study that was conducted in Guinea, the sensitivity of mAECT-bc was 96.5%, while the sensitivity of mAECT-wb was 78.9% [17]. Another study that was conducted in DRC reported a somehow smaller difference in sensitivity between these two methods (90.9% and 80.4%, respectively) [34]. The lower sensitivity values that were found here (72.6% and 51.6%, respectively) could be explained by the fact that the mAECT methods were only performed on a subset of participants who had been negative with other parasitological methods, and who were therefore likely to include cases with a lower parasitaemia than the other cases that were enrolled in the study. This selection bias could also explain why the difference in sensitivity was higher than in previous reports, since patients with a low parasitaemia could have provided a better dynamic range to evaluate subtle differences in sensitivity. While the difference in sensitivity could be an overestimate of the true difference that would be observed in an unbiased population, implementing the mAECT-bc protocol could be considered to enhance case finding, for a minimal additional workload. Since there was a high degree of overlap between the mAECT-wb and mAECT-bc results, performing both methods may not be justified, as it would increase costs without resulting in any significant increase in sensitivity. Although introducing mAECT-bc would require specific training to prepare buffy coat samples, it did not present a particular challenge during this study, and therefore, implementing it at other sites that are already equipped to perform mAECT-wb should be relatively straightforward.
While mAECT-bc has been shown to be more sensitive than mAECT-wb, mAECT-wb is known to be more sensitive than CTC [6,34]. Thus, the observation that the positivity rates of the screening tests were highest in cases found positive by mAECT-bc and lowest in cases that were positive by CTC could suggest that screening tests would be more sensitive in low-parasitaemia than in high-parasitaemia cases. Although this would need to be further investigated, it would be in agreement with the assumption that patients with a low parasitaemia would have a stronger immune response, which would facilitate their identification using antibody-detection screening tests. Conversely, patients unable to mount a strong immune response against trypanosomes and therefore more likely to have a high parasitaemia could be more difficult to identify using these screening tests.
This study confronted a number of challenges, which could have somehow impacted the quality of the results. Although study sites were carefully selected based on the available epidemiological data, the HAT prevalence was generally low, making it necessary to enroll patients at 15 different sites. This presented a significant challenge to the study team in terms of coordination, in particular when considering that most of the sites are located in remote, rural areas that were difficult to access. In addition, there was significant turn-over of personnel at some of the sites, requiring additional training. Enrolment was also interrupted at some sites due to stock-outs of supplies, such as mAECT kits. While the study was blinded, it is possible that technicians performing the tests could have been aware of the clinical status of some participants. This is probably more likely to be true in passive screening, since the number of patients presenting daily to health facilities was sometimes very low, making blinding more difficult. Although it is hypothetical, this imperfect blinding could be one of the factors leading to the high degree of overlap of true positive results, and to a lesser extent of false positive results, between the tests that was found in passive screening.
The results presented here have confirmed that the RDT2 would be a useful test for both active and passive screening, either as a single test or in combination with other screening tests. Since it is produced using recombinant antigens exclusively, it will also be easier and safer to manufacture than screening tests that are made with native antigens. The RDT2 is thus a welcome addition to the set of tools that are currently available to control and eventually eliminate HAT.
|
10.1371/journal.pgen.1006348 | Metatranscriptomic Study of Common and Host-Specific Patterns of Gene Expression between Pines and Their Symbiotic Ectomycorrhizal Fungi in the Genus Suillus | Ectomycorrhizal fungi (EMF) represent one of the major guilds of symbiotic fungi associated with roots of forest trees, where they function to improve plant nutrition and fitness in exchange for plant carbon. Many groups of EMF exhibit preference or specificity for different plant host genera; a good example is the genus Suillus, which grows in association with the conifer family Pinaceae. We investigated genetics of EMF host-specificity by cross-inoculating basidiospores of five species of Suillus onto ten species of Pinus, and screened them for their ability to form ectomycorrhizae. Several Suillus spp. including S. granulatus, S. spraguei, and S. americanus readily formed ectomycorrhizae (compatible reaction) with white pine hosts (subgenus Strobus), but were incompatible with other pine hosts (subgenus Pinus). Metatranscriptomic analysis of inoculated roots reveals that plant and fungus each express unique gene sets during incompatible vs. compatible pairings. The Suillus-Pinus metatranscriptomes utilize highly conserved gene regulatory pathways, including fungal G-protein signaling, secretory pathways, leucine-rich repeat and pathogen resistance proteins that are similar to those associated with host-pathogen interactions in other plant-fungal systems. Metatranscriptomic study of the combined Suillus-Pinus transcriptome has provided new insight into mechanisms of adaptation and coevolution of forest trees with their microbial community, and revealed that genetic regulation of ectomycorrhizal symbiosis utilizes universal gene regulatory pathways used by other types of fungal-plant interactions including pathogenic fungal-host interactions.
| Ectomycorrhizal fungi (EMF) comprise the dominant group of symbiotic fungi associated with plant roots in temperate and boreal forests. We examined host-specificity and gene-expression of five EMF Suillus species that exhibited strong patterns of mycorrhizal compatibility/incompatibility with either white pines (Pinus subg. Strobus) or hard pines (subg. Pinus). Using RNA-Seq, we identified conserved transcriptomic responses associated with compatible versus incompatible Pinus-Suillus species pairings. Comparative metatranscriptomic analysis of compatible vs. incompatible pairings allowed us to identify unique sets of fungal and plant genes associated with symbiont recognition and specificity. Comparativ transcriptomic study of the Suillus-Pinus system provides insight into the core functions involved in ectomycorrhizal symbiosis, and the mechanisms by which host-symbiont pairs recognize one another.
| Growing evidence has shown that many symbiotic plant-microbial associations including pathogenic as well as mutualistic symbioses are governed by similar genetic interaction mechanisms [1,2]. For example, in many groups of pathogenic fungi and oomycetes, coevolution with their plant hosts has resulted in typical 'arms-race' patterns of interactions, in which pathogens evolve batteries of effectors that suppress plant defense responses, while plants evolve modified receptors that sense microbial molecules and reactivate plant defense responses [3]. The molecular functions of several fungal and oomycete effectors involved in host-pathogen recognition have recently been elucidated. For instance, cysteine-rich avirulence genes (Avr) have been identified in several fungi including Cladosporium fulvum and Melampsora lini [4, 5], while Avr1b was isolated from the oomycete Phytophthora sojae [6]. Studying the functions of these effectors is a challenging task, because of the highly divergent nature of effectors in diverse taxa of pathogenic microbes and the lack of similarity of the sequences of these effectors to other proteins in public databases. Plant defense proteins that perceive microbial effectors include nucleotide-binding leucine-rich repeat (NB-LRR) proteins [1, 7, 8] and cell membrane receptors (e.g. phosphatidylinositol 3-P) [9]. These receptors can be activated by direct binding of effectors or modified by effector-associated proteins, leading to a plant-defense response.
Mutualistic plant-fungal interactions, including arbuscular mycorrhizae and ectomycorrhizae, also share similar conserved genetic interaction mechanisms with other symbiotic plant-fungal systems [10–12]. Over 30 plant families are known to form ectomycorrhizal associations with over 80 lineages (250 genera) of fungi [13]. A highly diverse community of EMF form the dominant guild of soil microbes in most of the world's forests [14,15], where they provide their plant hosts with essential resources (N, P, H2O) as well as protection from pathogens, in exchange for photosynthetically fixed carbon [16].
Details about molecular interactions between EMF and their plant hosts are emerging. Recent studies have identified differentially expressed genes associated with EMF symbiosis for several EMF-plant interactions including Pisolithus microcarpus with Eucalyptus [17], Paxillus involutus with Betula [18], and Laccaria bicolor with different Populus spp. [2]. One of these genes, a small secreted protein (MiSSP7) produced by the ectomycorrhizal basidiomycete Laccaria bicolor, functions as a critical effector for compatible mycorrhizal interaction with Populus. MiSSP7 was shown to be imported into plant nuclei where it suppresses plant host defenses, enabling mycorrhiza formation. Other recent studies also demonstrated that jasmonic acid (JA) and related plant defense-activated compounds are produced by Populus in response to signals from their symbiont [19,20]. These results suggest a general involvement of JA-mediated and other conserved plant signaling pathways for plant-fungal communication during EMF symbiosis. Similar to the mechanisms of EMF interaction in Laccaria [2], plant pathogenic fungi (e.g. M. larici) can also deliver SSPs to multiple cellular compartments in Populus [21]. These studies demonstrate that EMF are able to modulate plant defense system during symbiosis [2,10,21], and suggest that that most plant-microbial associations (including pathogenic and mutualistic interactions) may be governed by similar mechanisms. Unlike biotrophic/necrotrophic parasitisms, mutualistic fungal-plant interactions such as EMF must also establish stable long-term relationships with their living host cells, with benefits to both the fungus and its host. Thus, there is considerable potential for an array of distinct elements to regulate the host-specific communications of symbiosis compared to plant-pathogen interactions.
Many groups of EMF are known to exhibit preference or specificity for different plant host genera [22,23]. A good example of strong host-specificity is the bolete genus Suillus, which grows in association with the conifer family Pinaceae [24,25]. Most species of Suillus form ectomycorrhizae with specific Pinaceae host species (e.g., white pine, douglas fir, larch), suggesting a long history of plant-fungal coevolution in this genus [26–28]. Other examples of EMF with host-specific interactions include Laccaria bicolor, which shows differential host-compatibility with different species of Populus [29], and Paxillus involutus, which favors Betula as a host over Populus [30]. In order to study the molecular basis for host-specificity between different Pinus and Suillus species, we used pairwise plant-fungal bioassays to identify patterns of compatible and incompatible EMF interactions. Compatible EMF interactions are characterized by morphogenesis of plant and fungal tissues leading to development of modified plant short roots with bifurcated root tips that are sheathed by a hyphal mantle over the root epidermal surface, with hyphal ingrowth into the root cortex to form the Hartig-net [31]. In contrast, incompatible EMF interactions fail to induce root morphogenesis, resulting in little or no mycelial growth, and are morphologically indistinguishable from uninoculated (non-symbiotic) roots.
The pace of genetic studies of EMF-plant symbiosis has greatly accelerated by expanding numbers of genome sequencing for many EMF [10]. Though study of most EMF is still hindered by a lack of ‘finished’ genomes, we recently developed a procedure that employs RNA-Seq and de-novo assembly and annotation to characterize the metatranscriptome of EMF associated with Pinus taeda from field-collected mycorrhizal root clusters [32]. Here we apply metatranscriptomic profiling to study compatible versus incompatible mycorrhizal interactions from both plant and fungal perspectives. Our studies demonstrate that Suillus and Pinus each exhibit well-differentiated transcriptomic profiles during compatible and incompatible interactions. Comparison of expression patterns in compatible and incompatible pairings helped us to identify gene sets associated with plant-fungal recognition and establishment of EMF symbiosis.
To investigate occurrence of Suillus in natural Pinaceae forests, we first examined patterns of host specificity for Suillus operational taxonomic units (OTUs) detected by a recent survey of North American pine forest soils using next generation amplicon sequence analysis of the ribosomal RNA internal transcribed spacer (ITS) region [14]. Eleven Suillus OTUs detected by that survey (out of a total of >10,000 fungal OTUs detected across North America) exhibit distinct host range patterns corresponding with different Pinaceae hosts (S1 Fig): S. glandulosa with Picea glauca; S.hirtellus and S. cothurnatus with Pinus taeda; S. granulatus, S. spraguei (= S. pictus) and S. americanus with Pinus strobus; and an unidentified Suillus sp. with Pinus monticola. Several Suillus species were observed to be broadly associated with multiple Pinus species, including Suillus brevipes, which is associated with several Pinus spp. across North America (S1 Fig) but was restricted to hosts in the subgenus Pinus (P. ponderosa, P. contorta, P. banksiana, and P. taeda).
To study host specificity, a plant bioassay was developed using axenically grown pine seedlings inoculated with Suillus basidiospores to establish Suillus-Pinus mycorrhizae in vitro [31]. Seedlings of ten Pinus species were inoculated in all pairwise combinations with basidiospores of five Suillus species and scored for ectomycorrhiza formation after 8 weeks growth. In Pinus, successful formation of ectomycorrhizae (compatible interaction) results in a series of characteristic morphogenetic changes to young root tips that become swollen and bifurcated, and ensheathed by a mycelial mantle which penetrates into the root cortex to form a Hartig net [33] (Fig 1A and S2 Fig). In contrast, incompatible pairings are characterized by little or no colonization of roots by fungal mycelium (both fungal mantle and Hartig-net absent).
Basidiospore inoculations of two generalist species, S. hirtellus and S. decipiens, resulted in well-developed (compatible) ectomycorrhizae with most Pinus species (Fig 1B), when S. hirtellus had relatively lower rates of colonization on all hosts. Three white pine specialists (S. granulatus, S. americanus and S. spraguei) readily formed ectomycorrhizae with white pines (P. strobus and P. monticola), but had lower colonization rates on hard pines (e.g. P. banksiana), and failed to form visible ectomycorrhizae on P. taeda (incompatible pairing) (Fig 1B).
Variation in mycorrhizal compatibility between different Suillus and Pinus species suggests that genetic differences underlie host recognition and specificity during ectomycorrhizal symbiosis. To test this hypothesis, we compared transcriptomic activities across a panel of compatible and incompatible root tip samples formed by inoculation of three Pinus species (P. monticola, P. strobus, P. taeda) with four species of Suillus (S. americanus, S. granulatus, S. spraguei, and S. decipiens). Detailed descriptions of the individual Suillus-Pinus sample pairs, including strains used are listed in S1 Dataset. Transcriptomes from uninoculated pine roots were included as controls (to confirm that Suillus genes were not expressed by uninoculated roots) along with pure cultures of each fungal species (as references for transcriptome assembly). Comparative transcriptome profiling was used to identify candidate genes involved in Pinus-Suillus recognition (Table 1). The computational strategies included a) de novo transcriptome assembly to identify reads representing genes for different rRNA, Suillus, Pinus, and b) comparative transcriptomic analysis to identify common (core) and unique (host-specific) genes involved in symbiosis (see Materials and Methods, and SI text A1-A4; S3–S5 Figs). Unique genes were defined as upregulated genes detected in the RNA contig assembly of one Suillus species, but absent in other species examined. However, whether these genes are truly unique to different Suillus species still need to be determined through whole genome sequencing.
Up to 28 million (M) high quality reads were recovered from inoculated root tips using RNA-Seq (approx. 1 mg root tissue per sample, equal to about ten root tips) (S1 Dataset). Compatible Pinus-Suillus pairs resulted in roughly equal numbers of plant and fungal reads, while incompatible pairs resulted in much lower number of fungal reads compared to the corresponding plant reads (Fig 2). These differences of Suillus/Pinus reads recovered from compatible and incompatible interactions are also consistent with to the higher proportion of fungal biomass present in compatible versus incompatible mycorrhizal pairings. The Suillus transcriptome generated from de novo assembly of pooled data was used to identify 15M (51% of total reads) and 2M (6.1% of total reads) reads from compatible and incompatible reactions, respectively (Fig 2 and S1 Dataset). Approximately 3M (11% of total reads) and 21M (66% of total reads) Pinus transcriptome reads were also recovered from compatible and incompatible pairings, which could be matched to 44% and 69% of publicly available Pinus EST databases (~0.3M ESTs), respectively. In total, 11,029 and 5,947 Suillus contigs were obtained through de novo assembly from compatible and incompatible root samples respectively (S1 Dataset).
We hypothesized that pairings between different Suillus (and Pinus) species would share common gene expression patterns during compatible vs. incompatible pairings. Similarly, unique gene sets expressed by individual Suillus/Pinus pairings could also be identified (Fig 2). Here we defined “common genes” as the core sets of genes that were upregulated (> 2-fold) in response to compatible hosts; in contrast, “unique genes” were identified as those were only expressed in individual Suillus spp. in response to specific Pinus host species. To test these hypotheses, we used comparative transcriptomic analysis to identify Suillus and Pinus expressed genes during compatible and incompatible ECM interactions of four Suillus species grown with three different hosts, P. monticola, P. strobus, and P. taeda, (Fig 3). To compare gene expression patterns between interacting fungal and host genomes, sequencing reads aligned to either Suillus or Pinus contigs were normalized using DESeq package (ver. 1.14.0) [34]. (Details were provided in Support Information SI A2, S5 Fig). Gene expression biplots revealed strong differences between compatible and incompatible EMF pairings (S6A and S6B Fig). All of the compatible EMF pairings showed similar expression patterns of Suillus genes, even on different hosts (e.g. P. strobus and P. monticola) (S7 and S8 Figs), which suggests that different Suillus species all employ common regulatory pathways across different compatible host species. Significant differences were observed in gene expression between compatible and incompatible reactions (t-test, p-value < 0.01) (Fig 4). On average, 8,765 Suillus contigs were upregulated when they grew with compatible hosts, whereas fewer contigs (1,918 contigs in average) were upregulated from incompatible pairings (S1 Dataset).
Gene expression patterns were analyzed among all individual Suillus-Pinus species pairs to identify common genes involved in both compatible and incompatible interactions (SI text A2; S5 Fig). A majority of Suillus transcripts (~3,800 contigs) were similarly regulated in response to different compatible Pinus species. We compared the sequence identities of these genes across all four Suillus species and identified 231 “common genes” that were upregulated during the compatible mycorrhizal interactions (Fig 3; SI text A3; S1 Dataset). In contrast to common genes expressed during compatible interaction, a smaller number of genes (261–571 genes) were found to be upregulated during incompatible interactions in different Suillus species (Fig 4A). BLASTX search against all four Suillus species only identified seven common genes expressed during incompatible mycorrhizal interactions in all species (S1 Dataset). Functional annotations of these seven common genes identified two GHs (glucoside hydrolase), one F-box, one fatty acid desaturase, one signal transduction receptor, and two genes with unknown functions.
In contrast to sharing of 231 expressed genes in compatible mycorrhizal interactions, most genes associated with incompatibility were unique to individual Suillus-Pinus species pairs. These included a large number of SSPs, G-proteins, and other genes with little similarity/homology to each other or with other known genes, suggesting that these unique genes for host specificity are highly diverse at the genomic level (Fig 4A, for detailed analysis strategies see SI text A4 and S5 Fig). Unique genes varied among different plant-fungus combinations (from 68 to 571 genes for an individual pair Fig 4B), and were found to represent 14 functional groups (Fig 4B) with similar functions but very low sequence similarity to one another (S1 Dataset). Over two thirds of unique genes expressed by Suillus spp. were related to G-protein signaling, such as G-protein coupled receptor (GPCR), GTPase P-loop, Gβ WD40, and G-protein regulated kinases (Figs 4B and 5), which suggests a strong involvement of G-protein pathways in host-specific recognition. Other differentially expressed Suillus genes were those related to FAD/AND(P) binding, cytochrome P450-related, secretory, catalysis (proteinase/hydrolysis/reductase/terpene synthesis) and nucleus-associated genes.
Of the 261–571 contigs that were strongly upregulated in response to Suillus-Pinus incompatibility (Fig 5A), functional profiling revealed 22 to 28 contigs for shared functions related to tat signaling pathway for exporting small secreted proteins, GPI anchored proteins, fungal LRR-domain proteins, phosphatase, and pectin lyase (Figs 4B and 5B). Expression of these genes was not detected in most compatible pairings.
Fungal small-secreted proteins (SSPs) are predicted to be key mycorrhizal effectors for the recognition of EMF by their plant host system. Using domain analysis (SI text A4), SSPs were defined by several criteria including (a) size smaller than 300 amino acid, (b) signal peptide predicted at the N-terminal and extracellular localization activity; (c) absence of transmembrane domains; (d) absence of endoplasmic reticulum retention motifs [12]. 47 Suillus SSP's matching these criteria were upregulated in response to different Pinus hosts (Fig 6). More SSPs were upregulated during incompatible than compatible interactions. At the sequence level, most SSPs are highly diverse and do not share sequence similarity with other SSPs from currently available databases. Most Suillus SSPs were also observed to be highly diverse in their tertiary structure (S8A Fig).
Comparative transcriptional profiling of Pinus genes across the Suillus-root pairs also identified a large number of pine transcripts with similar expression in response to compatible vs. incompatible EMF pairings (~18,000 contigs; S10 Fig). Overall, a smaller number of Pinus genes (from 253–5452 contigs) were differentially expressed in response to pairings with different species of Suillus. The largest number of upregulated genes was observed for Pinus-S. spraguei interactions compared to other compatible pairs, suggesting the possibility of a greater Pinus response to S. spraguei.
Highly expressed Pinus genes with at least two-fold change (FDR<0.05) were further characterized as “pine unique genes” involved in fungal recognition (expressed by individual Pinus spp. in response to specific species of Suillus) (Fig 7). On average, 20 Pinus contigs were identified as unique genes for every Suillus-pair sample. BLASTX annotation identified sets of unique Pinus genes with common function involved in Suillus recognition, including genes for leucine rich (LRR)- proteins, UDP-glucosyl transferase, and cytochrome P450. Inoculation with S. spraguei also upregulated distinct Pinus genes encoding lipoxygenase 2, suggesting a potential effect on JA pathways for the Pinus-S. spraguei interaction.
Two different sets of P. taeda genes were found to be expressed during incompatible response including fungal species-specific (Fig 7) and species-nonspecific genes (S11 Fig). Comparative transcriptomic analysis also captured changes in expression patterns of 460 P. taeda genes associated with incompatibility, but these do not appear to be Suillus species-specific (S11A Fig). A number of Pinus genes known to be associated with defense responses were only weakly or not expressed in compatible pairings and uninoculated roots, including genes involved in plant resistance and water stress response including genes for salicylic acid acquired resistance (NDR1), ethylene-responsive transcription factor and RNA helicase, leucine rich proteins (e.g. Cf2.1, receptor kinase), thaumatin-like proteins, dehydrin and water deficit induced-LP3 (S11B Fig).
Comparative metatranscriptomic profiling of compatible vs. incompatible Pinus-Suillus interactions reveals several novel aspects of ectomycorrhizal symbiosis: (a) Suillus are transcriptionally active under both compatible and incompatible reactions; (b) Suillus spp. vary in their host specificity with different species of Pinus; (c) Suillus spp. share common sets of genes expressed during compatible and incompatible responses with different Pinus spp.; (d) Individual Pinus-Suillus species pairings induce expression of unique gene sets including genes for small secreted proteins (SSPs)/G-protein signaling pathway (Suillus genes) and LRR/PR proteins (Pinus genes).
We hypothesize that the shared functions among “common genes” contribute to a common role in core mechanisms of host-recognition. In contrast, “unique genes” may be involved in recognition between individual Suillus-Pinus species pairs. During incompatible interactions, these unique genes are largely associated with host recognition, specificity, and incompatibility.
We identified 231 common genes expressed during compatible mycorrhizal interactions (9 out of 12 Suillus-Pinus species pairings, Fig 3). In contrast, comparative analysis revealed only 7 common genes expressed during incompatible interactions between all three white-pine specialists when paired with loblolly pine (P. taeda). These findings suggest that different Suillus spp. share a common set of genes involved in compatible but not in incompatible responses. These estimates are likely to be higher, however, since our strategy employing de novo assembly and annotation could not detect less abundantly expressed genes without much deeper sequencing or access to a high quality reference genome. Further mapping of compatible/incompatible gene sets to fully-sequenced reference genomes of Suillus and Pinus is likely to reveal additional shared common genes involved in compatible/incompatible interactions.
To study the distribution of Suillus in natural Pinaceae forest soils, next generation sequencing was conducted to identify fungal operational taxonomic units (OTUs) of Suillus from the soils collected in Pinaceae forests across the North America. Technical details and data source to generate S1 Fig can be found in Talbot et al. [15]. For mycorrhizal plant bioassays, seeds of different Pinus species were purchased from Sheffield’s Seed Co., Inc. (Locke, NY) (see S3 Table for detailed description). The seeds were surface sterilized in 10% bleach for 10 min, suspended in sterilized water overnight and stratified at 4°C for different time periods prior to germination. Germinated seedlings were planted in sterilized sand and watered using sterile water. Basidiospores of different Suillus spp. were collected as spore deposits from field-collected fruit bodies by placing pilei overnight on wax paper or aluminum foil. Fruit body collection data are given in SI text A6.
A Suillus-Pinus pairwise bioassay was conducted using basidiospore inoculations with six-week old pine seedlings. Ten Pinus species were crossed with five Suillus species for a total of 50 pairwise combinations (replicated three times). Basidiospores (106 spores) were suspended in sterile water with 0.1% Tween-20, and added to sterilized 400 g of autoclaved sand to fill a four inch pot. Seedlings growing in sterile sand (without inoculum) were used as controls for all experiments (and also to check for airborne growth chamber contamination). Seedlings were grown in a growth chamber at 25°C, 80% humidity and fluorescent light at 200 μmol for 12 hours per day. At 180-d post-inoculation, EMF root tips were visualized under a dissection microscope, and percentage of EMF root tips were counted in comparison with bare (uninoculated) root tips.
Root tips were harvested from the bioassay pots at 90-d post-inoculation. From each plant, 10 root tips were collected using forceps, frozen in liquid N2 and stored in -80°C for RNA extraction. Four species of Suillus (S. americanus; S. granulatus; S. spraguei (= S. pictus); S. decipiens) and three species of Pinus (P. monticola; P. strobus; P. taeda) were grown in all 12 pairwise combinations (each replicated three times). Root tips collected from uninoculated Pinus species were also included as controls. The controls included six samples for a total of three species of Pinus that were replications for two different seedlings for each species (Table 1).
Total RNA was extracted using CTAB/chloroform extraction and LiCl precipitation method as described [32]. The mRNA samples for RNA-seq analysis were performed using a TruSeq RNA sample preparation kit (Illumina, San Diego, CA). The cDNA libraries were sequenced on the Illumina HiSeq 2000 (Illumina, San Diego, CA) instruments in Duke Center for Genomic and Computational Biology (GCB). Thirteen samples were sequenced using a single lane of Illumina run and generated 38Gb of data. The data generated from four lanes were applied for this study. The raw reads were deposited in the NCBI Short Read Archive (accession no. SRP057033).
We employed a genome-free assembly method to sort reads representing genes for different rRNA, Suillus, Pinus, and other genes (S3 and S4 Figs and SI Text A1). The computational workflow for sequence assembly (S3 Fig) was modified after Liao et al. [32]. First, Suillus sequence references were generated using the sequencing reads generated from Suillus fungal cultures, including S. americanus, S. granulatus, S. spraguei and S. decipiens. Next, de novo assembly was applied using Trinity [34]. The quality of the assembled contigs/unigenes for the four Suillus species are listed in S1 Table. The filtered reads (~28 million) were mapped onto four sets of reference sequences using bowtie with default settings (http://bowtie-bio.sourceforge.net/index.shtml), including references of fungal rRNA, 16S rRNA, contigs generated from Suillus cultures, and EST database of P. taeda. Remaining unmapped reads (approximately 3-million) were assembled de novo into contigs using Trinity followed by sorting into fungal and plant reads BlastX. Detailed descriptions of bioinformatics and databases used for three steps are included in SI text A1. The numbers of reads belonging to Suillus, Pinus, rRNA (and others) is shown in S1 Database. Comparative analysis of gene expression was used to evaluate their biological functions. The t-test (P<0.01) was used to identify the genes of Suillus in response to their compatible vs. in compatible hosts (Fig 3). A false discovery rate (FDR) of 5% was used to identify highly expressed transcripts with at least 2-fold change for the common and unique genes of Suillus and Pinus (Figs 4–7).
Transcriptome (EST) databases for S. americanus (19,123 contigs), S. granulatus (15,724 contigs), S. spraguei (18,898 contigs) and S. decipiens (16,871 contigs) were assembled de novo from fungal cultures using RNASeq. Besides the transcriptome references generated in our study (S1 Table), the other reference databases used in this study include: Fungal rRNA (NCBI, UNITE); Bacterial 16S (Ribosomal Database Project, http://rdp.cme.msu.edu); P. taeda EST database (NCBI). The databases were quality filtered using FASTA within the Galaxy web-based package. Detailed protocols for plant and fungal annotation databases are provided in SI text A2-A4.
|
10.1371/journal.pgen.1004320 | Null Mutation in PGAP1 Impairing Gpi-Anchor Maturation in Patients with Intellectual Disability and Encephalopathy | Many eukaryotic cell-surface proteins are anchored to the membrane via glycosylphosphatidylinositol (GPI). There are at least 26 genes involved in biosynthesis and remodeling of GPI anchors. Hypomorphic coding mutations in seven of these genes have been reported to cause decreased expression of GPI anchored proteins (GPI-APs) on the cell surface and to cause autosomal-recessive forms of intellectual disability (ARID). We performed homozygosity mapping and exome sequencing in a family with encephalopathy and non-specific ARID and identified a homozygous 3 bp deletion (p.Leu197del) in the GPI remodeling gene PGAP1. PGAP1 was not described in association with a human phenotype before. PGAP1 is a deacylase that removes an acyl-chain from the inositol of GPI anchors in the endoplasmic reticulum immediately after attachment of GPI to proteins. In silico prediction and molecular modeling strongly suggested a pathogenic effect of the identified deletion. The expression levels of GPI-APs on B lymphoblastoid cells derived from an affected person were normal. However, when those cells were incubated with phosphatidylinositol-specific phospholipase C (PI-PLC), GPI-APs were cleaved and released from B lymphoblastoid cells from healthy individuals whereas GPI-APs on the cells from the affected person were totally resistant. Transfection with wild type PGAP1 cDNA restored the PI-PLC sensitivity. These results indicate that GPI-APs were expressed with abnormal GPI structure due to a null mutation in the remodeling gene PGAP1. Our results add PGAP1 to the growing list of GPI abnormalities and indicate that not only the cell surface expression levels of GPI-APs but also the fine structure of GPI-anchors is important for the normal neurological development.
| Glycosylphosphatidylinositols (GPI) are glycolipid anchors that anchor various proteins to the cell surface. At least 26 genes are involved in biosynthesis and modification of the GPI anchors. Recently, mutations in eight of those genes have been described. Although those mutations do not fully abolish the functions of encoded enzymes, they lead to a decreased expression of surface GPI-anchored proteins and to different forms of intellectual disability. Here we report a mutation in PGAP1 that encodes a protein that modifies the GPI anchor. We found that the mutation leads to a full loss of PGAP1 enzyme activity, but that the patient cells still express normal levels of surface GPI-anchored proteins. However, the GPI anchors have an abnormal lipid structure that is resistant to cleavage by phosphatidylinositol-specific phospholipase C. Our results add PGAP1 to the growing list of GPI abnormalities that cause intellectual disability and indicate that the fine structure of GPI-anchors is also important for a normal neurological development.
| Many eukaryotic cell-surface proteins with various functions are anchored to the membrane via glycosylphosphatidylinositol (GPI) [1]–[3]. After biosynthesis in the endoplasmic reticulum (ER), GPI-anchors are transferred to the proteins by the GPI transamidase and the structure of the GPI-anchor is then remodeled, which is critical for sorting, regulating and trafficking of the GPI anchored proteins (GPI-APs) [3]. This remodeling starts in the ER by eliminating the acyl-chain linked to the inositol in the GPI-anchor by PGAP1 [4], then a side-chain of ethanolamine-phosphate on the second mannose of the GPI-anchor is removed by MPPE1 (PGAP5) [5]. GPI-APs are then transported from the ER to the plasma membrane through the Golgi apparatus, where further remodeling by PGAP3 and PGAP2 takes place [6], [7]. Germline mutations in eight genes that are involved in the GPI-anchor biosynthesis and remodeling have been described (Table 1) [8]–[22]. The mutations in all of those, PIGA, PIGL, PIGM, PIGV, PIGN, PIGO, PIGT and PGAP2, are hypomorphic and lead to partially decreased cell surface expression of various GPI-APs, thus causing a wide phenotypic spectrum ranging from syndromic disorders with various malformations to non-specific forms of intellectual disability. The reported mutations in genes of early steps of the GPI-anchor synthesis such as PIGA (MIM 311770), PIGL (MIM 605947), and PIGM (MIM *610273), or in a gene involved in GPI transfer to proteins such as PIGT (MIM *610272) are supposed to result in a degradation of precursor non-GPI-anchored proteins by ER associated degradation, whereas mutations in genes that are involved in later steps of the pathway, such as PIGV (MIM *610274), PIGO (MIM *614730), and PGAP2 (MIM *615187) result in partial secretion of non-GPI-anchored proteins such as alkaline phosphatase (in case of PIGV or PIGO deficiency) [23] or of proteins bearing cleaved GPI-anchor (in case of PGAP2 deficiency), and are therefore characterized by hyperphosphatasia. Here we report on the identification of a mutation in PGAP1 that encodes the GPI inositol-deacylase [4]. This leads to a new type of GPI-anchor deficiency manifesting non-specific autosomal recessive intellectual disability (ARID), in which cell surface levels of GPI-APs are not affected whereas the structure of GPI moiety is abnormal.
We undertook clinical characterization, mapping [24] and exome sequencing in a large cohort of families with non-specific ARID. We identified the PGAP1 mutation in the Syrian family MR079. The parents in family MR079 are the first-degree cousins and the family has one healthy girl and two affected children that carry the mutation in a homozygous status. The affected girl (III-2) was 4 years and 5 months old and the affected boy (III-3) was 2 years and 9 months old at the time of examination (Figure 1). Pregnancy, delivery, and birth parameters of both children were unremarkable. In the neonatal period, III-2 was hypotonic and III-3 was a floppy baby. Motor development was delayed; III-2 could sit at age of 18 months and at age of 45/12 years first tried to walk independently. At age of 29/12, III-3 could only roll from back to stomach and back. Both children did not finish potty training and were still partially fed with milk bottles. Both children have a developmental delay and severe intellectual disability with an estimated IQ below 35. III-2 could only babble a few syllables. While III-2 had major and absence epilepsy, III-3 did not yet have seizures. Sleeping patterns of both children were normal. They showed some stereotypic movements such as hitting on their own mouth and some washing movements of the hands. Both children seemed to see and hear properly, but specific tests could not be done. Brain CT scan of III-2 at age of one year revealed pronounced brain atrophy. At the time of examination, III-2 was 96 cm tall (25th percentile) with a head circumference of 46 cm (2 cm below the 5th percentile). III-3 was also of normal height and had a head circumference of 47 cm (1.5 cm below the 5th percentile). Their parents had head circumferences of 52 and 53 cm, also in the lower percentiles. Both children have large ears and a flattened nasal root. G-banding, cytogenetic examination and genome wide copy number variants analyses were unremarkable. We did not have information on the levels of alkaline phosphatase and it was not possible to obtain blood probes retrospectively.
Autozygosity mapping [24] in family MR079 led to the identification of six candidate regions of a total length of 64 Mb. Subsequently, exome sequencing using DNA from individual III-3 was performed as described in former studies [21], [25] resulting in an average coverage of 53.28. 66% of the target sequences were covered with a depth of at least 20×, and 80.51% were covered with a depth of at least 5×. A total of 42,352 SNVs and 2,529 indels were identified. 342 SNVs and 64 indels were neither annotated, nor reported in 1000Genomes and Exome Variant Server, nor in in-house controls, and may affect the protein sequence (non-synonymous, splicing, or UTR). Of those, only two, in PGAP1 and SLC40A, were located in a candidate region, conserved, and predicted to be pathogenic by in silico programs. To exclude further candidate mutations, we repeated the exome sequencing using DNAs of both affected siblings. We enriched the exome using a PCR based targeting method (Ion AmpliSeq Exome Kit) and sequenced on the Ion Proton. The average coverage of III-3 and III-2 was 149.6× and 94.6×, respectively. 91.1% and 85.0% of the target sequences were covered with a depth of at least 20×, 96.3% and 93.4% with a depth of at least 5×, respectively. A total of 49,455 and 47,693 SNVs as well as 3,343 and 3,167 indels were identified. When applying the above mentioned filtering steps, we were by both affected children once again left with the variants in PGAP1 and SLC40A. Since mutations in SLC40A cause hemochromatosis of type 4 and have no effect on cognition (MIM 606069) [26], [27], we focused on the variant in PGAP1, NM_024989.3:c.589_591delCTT, NP_079265.2:p.Leu197del. Genotyping the variant in PGAP1 in 372 healthy Syrian adults using Sanger sequencing revealed no further carriers. Taking the minor allele frequency of 0 in the Exome Sequencing Project (ESP) data set and in our control sample of 372 healthy Syrian individuals, it seems that the mutation has prevalence far less than 0.001.
Molecular modeling using the GeneSilico fold recognition metaserver [28] and Modeler9.9 [29] using the closest related hydrolase (PDB code: 3LP5) as template highlighted the detrimental effect of the deletion of leucine 197 on the structure of PGAP1. Leucine 197 is located in the central strand of a β-sheet and is oriented towards the hydrophobic core of the enzyme where it forms multiple stabilizing interactions with the adjacent helices (Figure 2A, B). Deletion of this amino acid would place Ile198 at the position originally occupied by Leu197 (Figure 2C). The Cβ-branched side-chain of isoleucine cannot be accommodated at this sequence position resulting in several clashes with adjacent amino acids (Leu184, Ile194) of the hydrophobic core (Figure 2C). This will disrupt the packing of the hydrophobic core and consequently of the entire β-sheet topology, thus leading to a loss of tertiary structure and enzymatic activity.
We then ran large scale homozygosity mapping using PLINK in our sample of over 100 consanguineous families [24] and over 600 sporadic cases of ID [30] and identified 7 index patients, 2 from consanguineous families with multiple affected children and 5 from outbreed families with single affected patients, that are homozygous at the PGAP1. Sequencing all seven individuals using Sanger did not reveal any mutations in PGAP1.
We then screened the exome variant server for functional variants in PGAP1. 149 variants are reported in this gene, of those 44 were coding or at splice sites. All of those are extremely rare (0.0077%–0.569%, i. e. 1–74 alleles out of ca. 13000 alleles). Based on the conservation of the variants and the prediction of in silico programs (Table S1), we roughly estimate that a maximum of 48 individuals may carry a mutation in PGAP1 (carrier rate of 48/6500 = 0.0073) and that the prevalence of the disease would be about 13 per million. If we take more conservative in silico prediction numbers, the prevalence of the disease would be 7 per million inhabitants (Table S1). The two most frequent variants in the ESP data were p.Lys111Glu and p.Gln585Glu and were observed in a heterozygous form 15 and 74 times out of 12992 and 12932 alleles, respectively. Both sites are well conserved in the mammalian. Molecular modeling showed that the most common variant Gln585Glu is located outside of catalytic active domains and it was not possible to make a prediction for this variant. Lys111Glu is at the C terminus of a helix of the deacylase domain. The charging pattern of the helix is highly conserved so that we expect that the change from Lys to Glu would change the charge of the protein and destabilize the helix.
To determine effects of p.Leu197del alteration on cellular GPI-APs, we investigated the surface expression of GPI-APs on B-lymphoblastoid cell lines (LCLs) derived from the homozygous individual III-3 (−/−), 2 heterozygous parents (+/−), and the healthy sister (+/+) (Figure 3), as well as 6 healthy volunteers with a confirmed wild type genotype (data not shown). Using flow cytometry analysis, the respective surface expressions of CD59, CD55/DAF, and CD48 were quantified. Surface expression of these GPI-APs on LCLs from an affected person, other family members or healthy volunteers showed no significant difference, indicating that the PGAP1 mutation did not affect the surface expression levels of various GPI-APs (Figure 3A, dotted lines). The surface expression of the GPI anchor itself was quantified using fluorochrome conjugated aerolysin (FLAER, Pinewood Scientific), a bacterial toxin that specifically binds GPI anchors, and did not show significant differences between the affected individual, the heterozygous individuals, and the controls (data not shown).
We then investigated the expected structural abnormality of GPI-anchors by testing sensitivity of GPI-APs to phosphatidylinositol-specific phospholipase C (PI-PLC) [31]. The LCLs were incubated with 10 unit/ml of PI-PLC for 1.5 h at 37°C and the remaining surface GPI-APs were determined by flow cytometry. Of GPI-APs, 61% to 90% were removed from the surface of LCLs of the healthy sister with a homozygous wildtype (Figure 3A, solid line) and healthy control individuals (data not shown). In contrast, no significant or only slight reduction of the surface GPI-APs was seen with LCLs from the affected person (Figure 3A), indicating that almost all GPI-APs on the affected LCLs had abnormal GPI anchors resistant to PI-PLC [4]. This is a strong indication that the p.Leu197del mutation causes null or almost null activity of the PGAP1 enzyme. GPI-APs on LCLs from heterozygous parents were only partially sensitive to PI-PLC (Figure 3A), indicating that the p.Leu197del mutation causes haplo-insufficiency. These defective sensitivities of affected the person's and parents' GPI-APs to PI-PLC were fully restored by transfection of wild-type PGAP1 cDNA (Figure 3B, solid lines).
Finally, the functional effect of the p.Leu197del mutation was tested in the PGAP1 deficient Chinese hamster ovary (CHO) cell system [4]. GPI-APs expressed on the PGAP1 deficient CHO cells are resistant to PI-PLC and the activity of PGAP1 cDNA can be assessed by its ability to make PI-PLC-sensitive GPI-APs after transfection. CHO cells defective for PGAP1 were transiently transfected with N-terminally-FLAG-tagged wild-type and p.Leu197del mutant human PGAP1 cDNA in an expression vector with a strong SRα promoter, or an empty vector. Four days after transfection, each transfectant was treated with or without PI-PLC, and the surface expression of CD59, DAF and urokinase plasminogen activator receptor (uPAR) were assessed by flow cytometry. The wild-type PGAP1 cDNA rescued PI-PLC sensitivity (Figure 4A, left panels). In contrast, the transfection of the mutant p.Leu197del cDNA did not increase the sensitivity to PI-PLC, thus indicating functional loss of the mutant PGAP1 cDNA (Figure 4A, center panels). To determine PGAP1 protein levels, lysates were prepared two days after transfection, immunoprecipitated with anti-FLAG beads and analyzed by SDS-PAGE/Western blotting. The p.Leu197del mutant protein was not detected at all, indicating that the deletion of Leu197 caused an unstable protein (Figure 4B).
In order to evaluate other known variants in PGAP1, we screened the public database of ESP (see above). Of listed variants, we chose the two most frequent variants: rs142320636: c.331A>G (p.Lys111Glu) and rs62185645: c.1753C>G (p.Gln585Glu), and tested the functional effect of these mutations in the PGAP1 deficient Chinese hamster ovary (CHO) cell system. Transfection of the mutant p.Lys111Glu cDNA did not increase the sensitivity to PI-PLC, indicating functional loss of the mutant PGAP1 cDNA. Mutant p.Gln585Glu showed an activity comparable to the wild type PGAP1 (Figure S1). Thus, it is possible that homozygosity of p.Lys111Glu leads to ARID.
Eight GPI deficiencies caused by hypomorphic mutations in the coding regions of GPI biosynthesis genes PIGM, PIGA, PIGL, PIGV, PIGN, PIGO, PIGT, and PGAP2 have been reported. Except PIGM, all lead to a decreased surface expression of GPI-APs and result in intellectual disability, often associated with epilepsy, distinct facial characteristics, and further organ malformations [9]–[22]. We showed here that complete PGAP1 deficiency did not affect the surface expression of GPI-APs but expressed structurally abnormal GPI-APs with the acylated inositol.
In previous works, we have reported that Pgap1 knock-out mice had otocephaly, male infertility, growth retardation, and often died right after birth [32]. Also further two mutant mouse strains, otoxray (oto for otocephaly) [33], [34] and beaker [35] were reported to have disrupted Pgap1. Both mice strains showed developmental abnormalities of the forebrain; the recessive lethal otoxray showed a truncation of the forbrain and the breaker mutant displayed a holoprosencephaly-like phenotype. Both Wnt signaling and Nodal signaling were reported to be affected in these mutant mice. These data emphasize the importance of PGAP1 for vital functions and for brain development. It was also indicated that the Pgap1 mutant mice phenotypes are dependent upon the genetic background since otocephaly and holoprosencephaly are not seen in some mouse strains [34], [35].
Based on our mapping results, exome sequencing data and functional experiments that proved pathogenicity of the mutation, the previous reports on intellectual disability caused by mutations in the GPI synthesis pathway, and the mouse models that clearly show an association between the disruption of Pgap1 and abnormalities of brain, we consider the deletion of leucine197 to be causative for the severe non-specific autosomal recessive intellectual disability in our examined patients of family MR079. PGAP1 is the ninth gene of the GPI synthesis pathway that is now associated to a human phenotype (Table 1). Further mutations in PGPA1 are needed to confirm our findings. Also, describing further patients with different mutations is necessary to delineate the phenotypes of the GPI deficiencies. For example, considering the defect in the modification of the GPI anchors, the alkaline phosphatase would not be elevated in patients with PGAP1 mutations, but this needs to be confirmed.
In conclusion, null mutations in PGAP1 lead to severe intellectual disability and encephalopathy with no obvious malformations; we add PGAP1 to the growing number of genes involved in GPI-anchor deficiencies with human phenotypes. PGAP1 deficiency causes a defect in the ER part of the GPI-AP biosynthesis that involves the remodeling of the anchors after attachment to proteins, and it leads to normal protein expression on the cell surface but to abnormal anchor structure.
The study was approved by the Ethic Committees of the Universities of Bonn and of Erlangen-Nürnberg in Germany, and Osaka University in Japan. Informed consent of all examined persons or of their guardians was obtained.
Genomic DNA was extracted from EDTA blood probes by standard methods and genotyped with the Affymetrix Mapping array 6.0 (Affymetrix, Santa Clara, CA, USA). Analysis did not reveal pathogenic deletions or duplications. Mendelian segregation was calculated using PedCheck software and was confirmed in all instances. Autozygosity mapping was performed using HomozygosityMapper [36]. DNA from individual III-3 was enriched using the SureSelect Human All Exon Kit, which targets approximately 50 Mb of human genome (Agilent, Santa Clara, Ca, USA) and paired-end sequenced on a SOLiD 5500 xl instrument (Life Sciences, Carlsbad, CA, U.S.A.). Image analysis and base calling was performed using the SOLiD instrument control software with default parameters. Read alignment was performed with LifeScope 2.5 using the default parameters with human genome assembly hg19 (GRCh37) as reference. Single-nucleotide variants and small insertions and deletions (indels) were detected using LifeScope, GATK 2 and samtools/bcftools [37], [38]. To replicate the results, DNA from individuals III-2 and III-3 was amplified using the Ion AmpliSeq Exome Kit (Life Technologies, Carlsbad, CA, U.S.A.) which targets approximately 58 Mb of the human genome. After quality control on the Bioanalyzer High Sensitivity Chip (Agilent, Santa Clara, Ca, USA) and emulsion PCR (Ion PI Template OT2 200 Kit v3, Life Technologies, Carlsbad, CA, U.S.A.) the samples were sequenced on a Proton PI chip Version 2 (Life Technologies, Carlsbad, CA, U.S.A.). Base calling, pre-processing of the reads, short read alignment and variant calling was performed using the Torrent Suite including the Torrent Variant Caller (TVC, Version 4.0) with default parameters recommended for the Ampliseq Exome panel (low stringency calling of germline variants, Version September 2013). Variant annotation was performed using Annovar, integrating data from a variety of public databases [39], [40]. Additionally, variants were compared to an in-house database containing more than 350 sequenced exomes to identify further common variants which are not present in public databases. Finally, the variants were validated by PCR and Sanger sequencing according to the standard protocols to exclude technical artifacts and to test for segregation.
Heparin blood samples were collected from one affected and from all unaffected siblings and parents. Lymphoblastoid Cell lines (LCLs) were generated and cultured in RPMI 1640 (Gibco, Life technologies, Darmstadt, Germany) that is supplemented with 10% FCS (PAA Biotech, Cölbe, Germany) and different other supplements. LCLs from one of the affected siblings (III-3) and the parents were transfected with empty pMEoriP vector or pMEoriP-FLAG-humanPGAP1. Cells from healthy sister were used without transfection. Cells (5×106) were suspended in 0.8 ml of Opti-MEM and electroporated with 20 µg each of the plasmids at 260 V and 960 µF using a Gene Pulser (Bio Rad, Hercules, CA). Four days after transfection, cells were treated with or without 10 unit/ml of PI-PLC (Molecular probes, Eugene, OR) for 1.5 h at 37°C. Surface expression of GPI-APs was determined by staining cells with mouse anti-human CD59 (5H8), -human DAF (IA10), -human CD48 (BJ40) antibodies and each isotype IgG followed by a PE-conjugated anti-mouse IgG antibody (BJ40, mouse IgG1 and IgG2a, and secondary antibody were purchased from BD Biosciences, Franklin Lakes, NJ) and analyzed by flow cytometer (Cant II; BD Biosciences) using Flowjo software (Tommy Digital Inc., Tokyo, Japan).
pMEFLAG-hPGAP1 mutant (L197del) bearing patient's mutation was generated by site directed mutagenesis. PGAP1 deficient CHO cell (C10) [4] were transiently transfected with wild type or mutant pMEFLAG-hPGAP1 by electroporation. Cells (107) were suspended in 0.4 ml of Opti-MEM and electroporated with 20 µg each of the plasmids at 260 V and 960 µF using a Gene Pulser. Four days after transfection, cells were treated with or without 10 unit/ml of PI-PLC for 1.5 h at 37°C. Surface expression of GPI-APs was determined by staining cells with mouse anti-human CD59 (5H8), -human DAF (IA10), -hamster uPAR (5D6) antibodies and each isotype IgG, followed by a PE-conjugated anti-mouse IgG antibody and analyzed by flow cytometer using Flowjo software. Two days after transfection of each PGAP1 construct, lysates were immunoprecipitated with anti-FLAG beads and analyzed by SDS-PAGE/Western blotting.
1000Genomes, http://www.1000genomes.org/
ABI, L.T. (2012). LifeScope.: http://www.lifetechnologies.com/lifescope.
ANNOVAR: http://www.openbioinformatics.org/annovar/
GeneTalk: http://www.gene-talk.de
BWA, Burrows-Wheeler Aligner; http://bio-bwa.sourceforge.net/
dbSNP, NCBI: http://www.ncbi.nlm.nih.gov/snp/
GATK 2, Genome Analysis Toolkit: http://www.broadinstitute.org/gatk/index.php
Kyoto Encyclopedia of Genes and Genomes, KEGG, http://www.genome.jp/kegg/
MutationTaster: http://www.mutationtaster.org/ELAND, alignment algorithm, Illumina.com
NHLBI Exome Sequencing Project (ESP): http://evs.gs.washington.edu/EVS/
Online Mendelian Inheritance in Man (OMIM): http://www.omim.org
PolyPhen2: http://genetics.bwh.harvard.edu/pph2/
SIFT: http://sift.jcvi.org/
UCSC Genome Browser: www.genome.ucsc.edu
|
10.1371/journal.pntd.0007340 | Spatio-temporal distribution of Spiroplasma infections in the tsetse fly (Glossina fuscipes fuscipes) in northern Uganda | Tsetse flies (Glossina spp.) are vectors of parasitic trypanosomes, which cause human (HAT) and animal African trypanosomiasis (AAT) in sub-Saharan Africa. In Uganda, Glossina fuscipes fuscipes (Gff) is the main vector of HAT, where it transmits Gambiense disease in the northwest and Rhodesiense disease in central, southeast and western regions. Endosymbionts can influence transmission efficiency of parasites through their insect vectors via conferring a protective effect against the parasite. It is known that the bacterium Spiroplasma is capable of protecting its Drosophila host from infection with a parasitic nematode. This endosymbiont can also impact its host’s population structure via altering host reproductive traits. Here, we used field collections across 26 different Gff sampling sites in northern and western Uganda to investigate the association of Spiroplasma with geographic origin, seasonal conditions, Gff genetic background and sex, and trypanosome infection status. We also investigated the influence of Spiroplasma on Gff vector competence to trypanosome infections under laboratory conditions. Generalized linear models (GLM) showed that Spiroplasma probability was correlated with the geographic origin of Gff host and with the season of collection, with higher prevalence found in flies within the Albert Nile (0.42 vs 0.16) and Achwa River (0.36 vs 0.08) watersheds and with higher prevalence detected in flies collected in the intermediate than wet season. In contrast, there was no significant correlation of Spiroplasma prevalence with Gff host genetic background or sex once geographic origin was accounted for in generalized linear models. Additionally, we found a potential negative correlation of Spiroplasma with trypanosome infection, with only 2% of Spiroplasma infected flies harboring trypanosome co-infections. We also found that in a laboratory line of Gff, parasitic trypanosomes are less likely to colonize the midgut in individuals that harbor Spiroplasma infection. These results indicate that Spiroplasma infections in tsetse may be maintained by not only maternal but also via horizontal transmission routes, and Spiroplasma infections may also have important effects on trypanosome transmission efficiency of the host tsetse. Potential functional effects of Spiroplasma infection in Gff could have impacts on vector control approaches to reduce trypanosome infections.
| We investigated the association of symbiotic Spiroplasma with the tsetse fly host Glossina fuscipes fuscipes (Gff) to assess if Spiroplasma infections are correlated with Gff genetic background, geography, or season and its interaction with trypanosome parasites. We analyzed distribution and prevalence of Spiroplasma infections across different Gff sampling sites in northern and western Uganda, and found that the symbiont is unevenly distributed and infections have not reached fixation within these sampling sites. We tested for associations with geographic origin of the collections, seasonal environmental conditions at the time of collection, Gff host genetic background and sex, plus trypanosome co-infections. Spiroplasma prevalence was strongly correlated with geographic origin and seasonal environmental conditions. Our parasite infection data suggested a negative correlation of Spiroplasma with trypanosome infection, with only 5 out of 243 flies harboring trypanosome co-infections. We further investigated the influence of Spiroplasma on trypanosome parasite infections in the laboratory. We found that trypanosomes were less likely to establish an infection in Gff individuals that carried Spiroplasma infections. Our results provide new information on host-endosymbiont dynamics in an important human disease vector, and provide evidence that Spiroplasma may confer partial resistance to Gff trypanosome infections. These findings provide preliminary evidence that a symbiont-based control method could be successful in combating tsetse trypanosome transmission to humans and livestock in sub-Saharan Africa.
| Tsetse flies (Glossina spp.) are vectors of parasitic African trypanosomes that cause human African trypanosomiasis (HAT, commonly referred to as sleeping sickness) and African animal trypanosomiasis (AAT, also known as Nagana in cattle) [1–3]. Several major HAT epidemics in sub-Saharan Africa have occurred during the last century, with the most recent one resulting in over half a million deaths in the 1980s [4–7]. An ambitious campaign led by WHO and international partners has now reduced the prevalence of HAT in west Africa [8] to a threshold considered irrelevant for epidemiological considerations, but millions continue to live at risk of contracting HAT in tsetse inhabited areas [9–11]. Despite calls for elimination of HAT by 2030, there is a lack of effective tools for long-term control of the disease (e.g., vaccines and field-ready diagnostic assays). Furthermore, the presence of animal reservoirs threatens disease elimination efforts going forward, particularly in East Africa (reviewed in [12]) and necessitates the inclusion of vector control applications. Practical interventions [13–15], as well as mathematical models [16–18], suggest that vector control can accelerate efforts for reaching the disease elimination phase. Thus, enhancing the vector-control tool box with effective and affordable methods is a desirable goal. Biological approaches that reduce vector reproduction as well as vector competency have emerged as promising means to reduce disease transmission [19].
Variation in the microbiota associated with tsetse flies can influence their pathogen transmission dynamics [20–24]. Several symbiotic microorganisms have been described from laboratory and field populations of tsetse, including the obligate Wigglesworthia glossinidia, commensal Sodalis glossinidius, parasitic Wolbachia and more recently Spiroplasma [25–31]. A survey of Ugandan Gff revealed that all individuals harbored Wigglesworthia, while Sodalis and Wolbachia associations were sporadic [28,32,33]. Spiroplasma was found in the tsetse species that belong to the Palpalis group (Gff, G. p. palpalis and G. tachinoides), while the tsetse species in the other two subgroups, Morsitans and Fusca, lacked associations with this microbe [31].
The bacterium Spiroplasma confers protection against nematodes, fungi and parasitoid wasps in several insects [34–36]. Spiroplasma also acts as a reproductive parasite that induces a male-killing effect in some arthropod hosts [37–40]. Based on phylogenetic analyses, the Spiroplasma species infecting Gff (both field-caught individuals and one laboratory line) is most closely related to the Citri-Chrysopicola-Mirum (S. insolitum) clade. Certain members of this clade confer protection against parasitoid wasps, nematodes and fungal pathogens in the fruit fly and aphid hosts [31]. In addition, among the Spiroplasma species within this clade are some that are pathogenic to plants and invertebrates and some that exhibit a male killing phenotype in ladybirds, fruit flies, and some butterfly species they infect (reviewed in [41]). Spiroplasma function(s) within the tsetse host remain unknown.
In Uganda, Gff is the main vector of HAT, where it transmits the chronic Gambiense disease caused by Trypanosoma brucei gambiense in the northwest and acute Rhodesiense disease caused by T. b. rhodesiense in the center, southeast and west (reviewed in [42]). It is possible that differences in Gff trypanosome susceptibility (vector competence) among varying geographic regions could be influenced by Spiroplasma, but patterns of Spiroplasma occurrence remain unexplored. Spiroplasma infection success may be influenced by seasonal fluctuations in host Gff population health (fitness) and density. Seasonal changes in the environment can dramatically alter host-endosymbiont dynamics through changes in Gff host physiological status (e.g. hemolymph lipid levels [43,44]) and tsetse population density [45,46]. Collections of Gff in Uganda during different seasons provides the opportunity to test for influence of seasonal fluctuations on Spiroplasma prevalence.
Additionally, patterns of Spiroplasma prevalence are likely influenced by Gff host genetic background, either because of vertical transmission that follows host inheritance patterns, or because of lineage-specific coevolutionary dynamics. Gff in Uganda has high genetic structuring at multiple spatial scales, which provides the opportunity to test for influence of Gff genetic background on Spiroplasma. There are three distinct Gff genetic clusters in the north, west and south of the country, and further population structure that separates the northwest from the northeast with ongoing gene flow between them in an admixture zone [47–49]. For this study, we focus our work on the north and west of Uganda in five watersheds. This sampling included four nuclear (biparentally inherited) genetic backgrounds: the northwest genetic unit (NWGU), the northeast genetic unit (NEGU), the west genetic unit (WGU) and the admixed (ADMX) genetic background intermediate to the NWGU and NEGU [47–49]. This sampling also included three mitochondrial (maternally inherited) genetic backgrounds: a group of related haplotypes associated with the NWGU known at “haplogroup A” (mtA), a group of related haplotypes associated with the NEGU known at “haplogroup B” (mtB), and a group of related haplotypes associated with the WGU known at “haplogroup C” (mtC) [47–49].
In this study, we (i) assessed the infection prevalence of Spiroplasma in Gff among five watersheds in northern and western Uganda, (ii) tested the effect of seasonal environmental variations, host Gff genetic background and sex, and trypanosome co-infections on Spiroplasma prevalence, and (iii) investigated the influence of Spiroplasma infections on Gff trypanosome transmission ability under laboratory conditions. We discuss how our results can elucidate potential functional associations of Spiroplasma with its tsetse host, and the potential applications of this knowledge to disease control.
A total of 1415 Gff individuals collected from northern and western Uganda during the period of 2014–2018 (pooled for analysis) across the wet (April—May and August—October), intermediate (June—July), and dry (December—March) seasons were assayed for Spiroplasma infection (S1 Table). Flies were collected on public land using biconical traps. Samples included in this study were chosen to represent a wide range of environmental conditions and Gff backgrounds (Fig 1). The location of each sampling site was placed on a map of Uganda using QGIS v2.12.1 (August 2017; http://qgis.osgeo.org) with free and publicly available data from DIVA-GIS (August 2017; http://www.diva-gis.org). We sampled 11 sites from the Albert Nile watershed (DUK, AIN, GAN, OSG, LEA, OLO, PAG, NGO, JIA, OKS, and GOR), where the majority of Gff belong to the NWGU/mtA genetic background, and a minority belong to the ADMX/mtB genetic background [47–49]. We sampled six sites from the Achwa River watershed (ORB, BOL, KTC, TUM, CHU, and KIL), where Gff belong to a mix of genetic backgrounds including NWGU/mtA, ADMX/mtA, ADMX/mtB, and NEGU/mtB [47–49]. We sampled six sites from the Okole River watershed (ACA, AKA, OCA, OD, APU, and UWA), where Gff hosts belong to a mix of genetic backgrounds including NWGU/mtA, ADMX/mtA, ADMX/mtB, ADMX/mtC, and NEGU/mtB [47–49]. We sampled two sites from the Lake Kyoga watershed (OCU and AMI), where the majority of Gff belong to the NEGU/mtB genetic background, and a minority belong to the ADMX/mtA genetic background [47–49]. Finally, we sampled a single site from the Kafu River watershed (KAF), where the majority of Gff belong to the WGU/mtC genetic background, and a minority belong to the WGU/mtA genetic background [47–49].
To assess the Spiroplasma strain infecting the field-collected samples, we cloned and sequenced Spiroplasma-16S rDNA and rpoB (RNA polymerase, subunit beta) from flies from two NGU sampling sites (GAN and GOR). The 16S rDNA touchdown PCR was performed in 20 μL reactions containing 1x GoTaq Green Mastermix (Promega, USA), 0.4 μM of each primer SpirRNAF and SpirRNAR (S2 Table) and 1 μL of template DNA (20-100ng). The PCR profile consisted of an initial denaturation for 3 mins. at 94°C, followed by 8 cycles of 1 min. at 94°C, 1 min. at 63°C, and 1 min. at 72 with a reduction in annealing temperature of 1°C/cycle (63–56°C). This step was followed by 30 cycles of 1 min. at 94°C, 1 min. at 55°C, and 1 min. at 72°C, with a final extension at 72°C for 10 min. Reaction set up for the rpoB-PCR was the same as for 16S rDNA (S2 Table), and the PCR profile consisted of a 3-min. initial denaturation at 94°C, followed by 34 cycles of 90 sec. at 94°C, 90 sec. at 55°C, and 90 sec. at 72°C, with a final extension at 72°C for 10 min. All amplicons were gel purified using the Monarch DNA Gel Extraction Kit (New England Biolabs, USA) and cloned into pGEM-T vector (Promega, USA). In total six 16S rDNA and two rpoB clones were sequenced and analysed using the BLASTN algorithm and finally aligned using Spiroplasma glossinidia data from NCBI as reference.
To test for potential Spiroplasma strain variation among the different Gff sampling sites, we cloned and sequenced the Multi Locus Sequence Analysis (MLST) genes dnaA, fruR, parE and rpoB from two individuals from ACA (Okole River) and OKS (Lake Kyoga) sampling sites. MLST-PCR reactions were performed with the same reaction set up as described above (S2 Table). The parE touchdown PCR was run as described above for 16S rDNA but with 61°C/52°C annealing temperature. A touchdown PCR approach was also used for the fruR locus with 30 sec steps in each cycle and 60°C/52°C annealing temperature. Cloning and sequencing was performed as described above. All sequences were manually edited using Bioedit v7.1.9 [50] and aligned with related sequences available at NCBI using MUSCLE [51].
At the time of collection, sex and wing fray information (all flies were wing fray 2–3, i.e. 4–8 weeks old) was recorded for each fly and midguts were dissected and microscopically analyzed for trypanosome infection status. The dissected guts and reproductive parts were stored in 95% ethanol for further analyses. Genomic DNA was extracted from female and male reproductive parts (RP, n = 1157) as well as from whole bodies (WB, n = 258) using DNeasy Blood and Tissue Kit (Qiagen, Germany). We used flies from Pagirinya (PAG) for which we had DNA available from both the WB and RP tissues to compare the Spiroplasma infection prevalence to rule out potential DNA source bias in the prevalence results. We noted Spiroplasma infection of 18% in WB versus 17% in RP, which is not significantly different (Fisher’s P > 0.999). Consequently, both tissue datasets were pooled for the final infection prevalence analyses. The final sample set was composed of 893 females and 522 males.
Spiroplasma infection prevalence was determined by 16S rDNA touchdown PCR as described above. The presence of Wolbachia was assessed by PCR targeting a single copy membrane protein encoding gene (Wolbachia Surface Protein gene, wsp). The PCR reactions were performed as previously described in [52–54]. All amplified fragments were analysed on 1% agarose gels using Gel Doc EQ quantification analysis software (Bio-Rad, Image Lab Software Version 4.1). We used generalized linear models (GLM) to test for the significance of difference in Spiroplasma infection between Wolbachia-infected and uninfected flies.
Potential Spiroplasma density differences between individuals and across seasons were tested via quantitative PCR (qPCR) using a CFX96 Real-Time PCR Detection System and iTaq Universal SYBR Green Supermix (Biorad). The relative amount of Spiroplasma was calculated with the 2-(ΔΔCt) method using primers targeting the Spiroplasma RNA polymerase beta gene (rpoB). Values were normalized against the single copy Glossina Peptidoglycan Recognition Protein-LA gene (pgrp-la). All primer sets are listed in S2 Table. The trypanosome infection status of each sample was assessed by microscopic analysis of dissected guts at the time of sample collection in the field.
In addition to the samples typed for a previous study [47], we sequenced an additional 161 flies for a 491 bp fragment of mtDNA, and genotyped an additional 131 flies at 16 microsatellite loci. To do this, we extracted total genomic DNA from two legs of individual flies using the Qiagen DNeasy Blood & Tissue kit. For mtDNA sequencing, a 491 bp fragment from the cytochrome oxidase I and II gene (COI, COII) was amplified as described in [47]. PCR amplicons were run on 1% agarose gels, purified and sequenced at the Yale Keck DNA Sequencing facility. New sequences were combined with existing data [47] for a total data set of 490 sequences (S1 Table). For microsatellite analysis, we genotyped flies at 16 microsatellite loci using methods described in [47]. New genotype calls were combined with existing data [47] for a total data set of 558 genotypes at 16 microsatellite loci (S1 Table).
To evaluate the potential association between Spiroplasma infection prevalence and the Gff genetic background, we assigned each individual to a single mtDNA haplogroup based on the phylogenetic relationships among haplotypes, and to a single nuclear genetic background based on clustering analysis of microsatellite genotypes.
For mtDNA haplogroup assignment, first evolutionary relationships between the mtDNA haplotypes were assessed by constructing a parsimony-based network using TCS 1.21 [55] as implemented in PopART ([56]; Population Analysis with Reticulate Trees: http://otago.ac.nz). These haplotypes were then grouped by phylogenetic relationship following [48] into three haplogroups, each imperfectly associated with the NWGU, NEGU, and WGU.
For nuclear genetic background assignment, we used a Bayesian clustering analysis in the program STRUCTURE v2.3.4 [57] to group individuals based on their microsatellite genotypes. STRUCTURE assigns individuals into a given number of clusters (K) to maximize Hardy-Weinberg and linkage equilibrium. The program calculates the posterior probability for a range of K and provides a membership coefficient (q-value) of each individual to each cluster. In this analysis, we assessed membership to just three clusters corresponding to the NWGU, NEGU, and WGU. The ADMX does not represent its own cluster, but instead represents a mixed assignment to the NWGU and NEGU. We performed 10 independent runs for K = 3 with a burn-in of 50.000 followed by 250.000 MCMC steps and summarized results across the 10 independent MCMC runs using the software CLUMPAK [58]. Individual flies were assigned to nuclear genetic units based on q-values. Individuals with q-values > 0.8 were assigned to one of the three distinct clusters (NWGU, NEGU or WGU), and individuals with mixed assignment (q-values ranging from 0.2 to 0.8) to the NWGU and NEGU were assigned as “admixed” (ADMX).
Predictive variables considered included watershed of origin, season of collection, Gff host genetic background (both nuclear and mitochondrial), Gff host sex, and trypanosome co-infection. A challenge in this analysis was that correlation between the geographic origin (specifically watershed) and environmental conditions, as well as Gff genetic background and trypanosome infection status is well established [47–49,59]. We confirmed these correlations among predictive variables and patterns of Spiroplasma infection, we performed a multiple correspondence analysis (MCA).
We took two approaches to control for correlation among predictive variables with watershed of origin. First, we fit generalized logistic mixed models (GLMM) with the predictive variables of interest (season, Gff host genetic background, sex, and co-infection) as fixed effects with and without watershed of origin as the random effect, and tested the improvement of the models with an analysis of variance (ANOVA) Chi square test. We followed these tests with more complex combinations of predictive variables using watershed as the random effect. GLMM was performed with the ‘glmer’ function in the R [60] package lme4 [61] and fitted using maximum likelihood.
Second, we fit generalized linear models (GLM) one watershed at a time for each predictive variable (season, Gff host nuclear and mtDNA genetic background, sex, and co-infection). Direction and significance of the effect of each predictive variable was assessed with Tukey’s contrasts with p-values (P) obtained from the z distribution and corrected for multiple comparisons using the unconstrained (“free”) adjustment. GLM was performed in R with the multcomp package [62].
Effect of Spiroplasma presence on trypanosome infection outcome was also tested in the laboratory by infecting the Gff colony (IAEA, Vienna, Austria), with bloodstream form Trypanosoma brucei brucei (RUMP503) parasites. Pupae from the colony were sent to Yale, and emerging flies were used for parasite infections. This Gff line has been shown to exhibit a heterogeneous Spiroplasma infection prevalence [31]. Following our established protocols [63,64], teneral (newly eclosed) flies were infected by supplementing their first blood meal with 5x106 parasites/ml. All flies that had successfully fed on the infectious blood meal were subsequently maintained on normal blood, which they received every other day. Fourteen days post infection (dpi), the presence of trypanosome infections in the midgut was assessed microscopically plus using a PCR assay. PCR was performed using primers trypalphatubF and trypalphatubR, which target the alpha chain of T. brucei tubulin (S2 Table). The reaction set up and the cycler profile were the same as used for Wolbachia ARM-PCR described above. Spiroplasma presence was assessed in the corresponding reproductive tissue of each fly via PCR assay using the Spiroplasma infection assay described above. We used generalized linear models (GLM) to test for the significance of difference in trypanosome infection between Spiroplasma-infected and uninfected flies.
To assess the Spiroplasma strain infecting Gff sampling sites, we employed 16S rDNA and rpoB sequencing analysis. In samples analyzed from the Okole River and Lake Kyoga watersheds, we found a single strain infection (S1 Fig), which belongs to the Citri-Chrysopicola-Mirum clade and which was also previously identified from a Gff colony [31].
We tested for Spiroplasma infection in 1415 Gff individuals (894 females and 522 males; S1 Table) collected from 26 sampling sites spanning five watersheds (Fig 1). In the northwest region, flies from the Albert Nile watershed (n = 487) had a mean infection rate of 34% (Fig 1 and S1 Table). In the northcentral region, flies from the Achwa River watershed (n = 234) had a mean infection rate of 20%, and the Okole River watershed (n = 389) had a mean infection rate of 5% (Fig 1 and S1 Table). In the northeast region, flies from the Lake Kyoga watershed had relatively low Spiroplasma infection rate, with one of the two sampling sites (AMI, n = 73) having an infection rate of 11%, and the other site (OCU, n = 90) lacking Spiroplasma infection altogether (Fig 1 and S1 Table). In the western region, flies from the Kafu River watershed (n = 142) had an infection rate of 3% (Fig 1 and S1 Table). These results indicate higher Spiroplasma infection in the northwest than in the northeast or west of Uganda, a conclusion that was further supported by tests for association of Spiroplasma with watershed of origin using generalized linear modeling (see discussion of MDS, GLMM, and GLM results below).
To assess whether potential differences in Spiroplasma infection density could influence our ability to detect the microorganism in the DNA source, we performed qPCR on individuals from the NEGU (AMI) sampled across multiple seasons. We detected varying densities, with the highest Spiroplasma levels observed in the intermediate season (S2 Fig).
We assigned Gff host nuclear (biparentally inherited) genetic background using STRUCTURE cluster assignments that were based on the microsatellite genotypes. STRUCTURE assignment indicated 195 flies had high (> 0.8) membership probability to the NWGU, 94 to the NEGU, and 109 to the WGU (S1 Table). We found 160 flies with mixed assignment, which we considered members of the ADMX genetic background (S1 Table). Flies assigned to the NWGU had a mean Spiroplasma infection rate of 25%, the ADMX genetic background had a mean infection rate of 15%, the NEGU (n = 94) had a mean infection rate of 2%, and the WGU had a mean infection rate of 4% (Fig 2 and S1 Table).
Although these results suggest higher Spiroplasma infection in the NWGU and ADMX nuclear genetic backgrounds, this pattern was found to driven by correlation between nuclear genetic background and watershed of origin (see discussion of MDS, GLMM, and GLM results below). We assigned Gff host mitochondrial (maternally inherited) genetic background by generating a TCS network of the mtDNA sequences (n = 490). We identified 29 unique mitochondrial haplotypes that grouped into the three previously described major haplogroups: mtA, mtB and mtC (Figs 2 and S4) [47]. 266 flies were assigned to haplogroup A, 209 to haplogroup B, and 15 to haplogroup C (S4 Fig). Flies assigned to mtA had a mean Spirplasma infection rate of 19%, mtB had a mean infection rate of 10%, and mtC had a mean infection rate of 13% (Fig 2 and S1 Table). Although these results suggest higher Spiroplasma infection in the mtA mitochondrial genetic background, this pattern was found to be driven by correlation between mitochondrial genetic background and watershed of origin (see discussion of MDS, GLMM, and GLM results below).
The MCA confirmed that there were strong correlations among predictive variables, especially with watershed of origin (S3 Fig). GLMM indicated that watershed of origin and season of collection (wet, intermediate and dry) were the two most important factors influencing Spiroplasma prevalence (Tables 1 and S3 and S4). GLMM of all predictive variables (considered one at a time) were significantly improved by adding watershed as a random effect (ANOVA P ranging from 2.20e-16 to 0.0022, S3 Table). Models exploring multiple predictive variables at a time indicated that season of collection was the only variable that positively influenced the fit of the model (ANOVA P = 3.57e-6, Table 1). Adding random slope or any of the other predictive variables (nuclear or mtDNA genetic background, sex, or trypanosome co-infection) did not significantly improve the model (ANOVA P ranging from 0.2870 to 1.0, S4 Table).
GLM by watershed indicated that flies from the Albert Nile had by far the highest probability of Spiroplasma infection [Pr(Spiro+) = 0.32], followed by flies from the neighboring Achwa River [Pr(Spiro+) = 0.22, Fig 3]. Tukey’s contrasts indicated that the Albert Nile had only somewhat higher Pr(Spiro+) than the Achwa River (Tukey’s contrast P = 0.0015), but that these two watersheds had significantly higher Pr(Spiro+) than any of the other watersheds (Tukey’s contrasts P ranging from 2.00e-16 to 0.0002, Tables 2 and S5).
In addition to watershed of origin, season of collection was strongly associated with Spiroplasma infection. The intermediate season had significantly higher probability of Spiroplasma infection than the wet season, which was especially apparent in the Albert Nile [Pr(Spiro+) = 0.42 vs 0.16)] and Achwa River watersheds [Pr(Spiro+) = 0.36 vs 0.08] (Tukey’s contrasts P = 1.16E-07 and 5.38E-06, respectively; Table 2). There were no other significant differences in Pr(Spiro+) among seasons in any of the other watersheds (Table 2). Additionally, none of the other predictive variables (nuclear or mtDNA genetic background, sex, or trypanosome co-infection) had significant effects on Pr(Spiro+) when analyzed by watershed (Tukey’s contrasts P ranging from 0.0962 to 1.0, S5 Table).
Previous analysis of Gff from southern and central regions of Uganda also indicated presence of distinct genetic backgrounds associated with these individuals [47,48] as well as the presence of heterogeneous and low-density infections with another endosymbiont, Wolbachia [27, 31]. To investigate for similar patterns in northern Uganda, we analyzed 106 Gff individuals from the Albert Nile (NWGU) and Achwa/Okole River (ADMX) watersheds for the presence of Wolbachia infections (S1 Table). Within this set, 92% of flies were not carrying Wolbachia (98/106). The remaining infected 8% (8/106) were all from Achwa river except one sample. Most Wolbachia-infected flies (6/8) were Spiroplasma-negative, and Wolbachia-negative flies were Spiroplasma-positive and -negative (S1 Table). The generalized mixed models (GLM) indicated no significant correlation between the presence of Wolbachia and the probability of Spiroplasma infection (Tukey’s P = 0.78).
Patterns of Spiroplasma infection from the field collections, although not significant in the GLMM or GLM, indicated a possible correlation between Spiroplasma and trypanosome co-infections. Of the 243 Spiroplasma infected flies identified in the study, only 2% (n = 5) had trypansome co-infection (infection rate among the Spiroplasma-negative flies was 10%; n = 115). To test this correlation under more controlled conditions where we could ensure statistical power, we used a laboratory line of Gff with heterogeneous infection of Spiroplasma [30] to complete a trypanosome infection challenge experiment. Microscopic examination and PCR analysis of challenged individuals (n = 123) revealed a negative correlation between the presence of Spiroplasma and trypanosome co-infection, with 18% of individuals co-infected with both microbes (S+T+; n = 22), 20% infected with only trypanosomes (S-T+; n = 25), 46% infected with only Spiroplasma (S+T-; n = 56), and 16% without infection with either microbe (S-T-; n = 20; Fig 4 and S6 Table). The generalized mixed models (GLM) indicated a significant negative correlation between the presence of Spiroplasma and the probability of trypanosome coinfection (Tukey’s P = 0.0010).
In this study, we performed a spatial and temporal analysis of the prevalence of the endosymbiont Spiroplasma in genetically distinct Gff populations across northern Uganda and assessed the factors that could impact the dynamics of infection with this symbiotic bacterium. The most influential factors that shaped the patchy distribution of Spiroplasma in Gff were the fly’s watershed of origin and the season of collection. The importance of watershed and season suggests that the prevalence of the symbiont is significantly affected by the geographic dispersal of the flies as well as by the changing environmental conditions. The low rate of Spiroplasma and trypanosome co-infections in the field collections, and the negative association of Spiroplasma and with trypanosome infection in a challenge experiment carried out in a laboratory Gff line suggests that Spiroplasma infections may negatively influence the success of parasite infection outcome.
We found that flies residing in geographically separated watersheds have significantly different Spiroplasma infection prevalences (Figs 1 and S3). Gff is a riparian species that inhabits low bushes or forests at the margins of rivers, lakes or temporarily flooded scrub land. These water connections may influence dispersal patterns by limiting dispersal between the different watersheds. Limited dispersal would minimize contact among flies from different watersheds. This could suggest that horizontal transfer occurs between flies within the same watersheds, or that there is environmental acquisition of Spiroplasma in Gff. However, it remains to be elucidated whether flies encounter Spiroplasma from the environment or during feeding. Interestingly, although tsetse are strict blood feeders on vertebrate hosts, a recent study has indicated that they are capable of feeding on water with or without sugar when deprived of a blood meal [65]. This feeding behavior would allow Gff to acquire Spiroplasma from the environment, and could account for the strong association of Spiroplasma with watershed of origin, and lack of correlation with other host genetic background. Finally, transfer of Spiroplasma via ectoparasites could account for different infection frequencies. Although there are no ectoparasites described yet, which are associated with tsetse flies, this route of transmission cannot be excluded.
In addition to geographic origin, the season of collection was an important factor shaping the Gff-Spiroplasma association in Uganda. We found that the intermediate season (June—July) was correlated with higher Spiroplasma infection prevalence than either the wet (April—May and August—October) or dry (December—March) seasons (Table 1 and Fig 3). This might be because the environmental conditions during the wet and dry seasons restrict the availability of animal hosts for tsetse blood feeding. This could indicate that the intermediate season represents optimal foraging conditions for Gff. In fact, Spiroplasma survival in Drosophila is dependent on the availability of hemolymph lipids [43,44]. Hence, during nutritionally optimal times, maintenance of the symbiont may be less costly for the host than during a period of compromised fitness, and thus higher prevalence and density is observed during these periods [43,66]. Other symbionts capable of triggering host phenotypes, such as male-killing, are also affected by host fitness. The persistence of Wolbachia, for example, is negatively affected by host fitness via temperature as well as by other stress factors (e.g. [66–68]). Thus, the more stressful conditions of the dry season may reduce the overall fitness of the fly and consequently the Spiroplasma infection densities. This scenario is supported by our finding of different Spiroplasma densities in flies analyzed across the three sampling seasons (Figs 2 and S2). Varying Spiroplasma densities can also influence the transmission efficiency of the symbiont from mother to tsetse’s intrauterine progeny. Furthermore, if Spiroplasma is horizontally transferred via the environment, the climatic conditions can also impact the abundance of Spiroplasma for acquisition by Gff. Higher Spiroplasma density in the environment during the intermediate season could facilitate their acquisition by the tsetse host. In support of this theory, free-living bacterial communities that reside in aquatic systems are strongly affected by environmental factors, such as pH [69,70] and seasonality [71]. A more recent study has highlighted the significant effect of seasonality-related changes in soil bacterial communities [72]. Hence, the changing environmental conditions that occur between the dry, wet, and intermediate season might impact Spiroplasma density and consequently the infection density. However, seasonality alone cannot explain the differences of the symbiont differences but other parameters such as e.g. temperature variations should be considered.
Genetically variable Gff populations residing within the Gambiense and Rhodesiense HAT belts could influence the inheritable microbiota composition, which in turn could have important implications in transmission dynamics and vector control outcomes [73–75]. In that context, we evaluated the potential correlation between host nuclear genetic differences and Spiroplasma infection prevalence across the 26 Gff sampling sites. However, we found that after accounting for watershed of origin in the GLMM model, Spiroplasma infection was not significantly influenced by Gff host nuclear or mitochondrial genetic background (Figs 2 and 3 and S5 Table). Lack of association of with host genetic background provides further support for the idea that Spiroplasma may be acquired by horizontal transfer between flies within the same watersheds, or through contact with other sources of the bacteria in the environment.
In Drosophila, Spiroplasma can either coexist with the endosymbiotic Wolbachia without little or no impact on each other [76], or negatively affect Wolbachia presence. Goto and coworkers showed that the presence of Spiroplasma suppresses Wolbachia infection density in D. melanogaster [77]. Interestingly, while the tsetse species in the Morsitans subgroup, such as G. morsitans, harbor Wolbachia infections, they lack Spiroplasma associations [31]. In contrast, the tsetse species in the Palpalis subgroup, such as Gff and G. palpalis, harbor Spiroplasma infections, but lack Wolbachia associations [31]. Spiroplasma might be negatively affected by the presence of Wolbachia infection in the species within the Morsitans subgroup [31]. As such, the high titer of Wolbachia noted in G. morsitans reproductive organs could suppress the establishment of Spiroplasma infections. Our prior and current studies with distinct Gff populations in Uganda also indicate a potential negative influence for the two endosymbiont infections [28]. Our previous finding reported Wolbachia infections across 18 Gff sampling sites from the Northcentral, West and South of Uganda [28]. In this study we detected 8% Wolbachia infections in samples analyzed from the Achwa River watershed, which is the geographic area with highest Spiroplasma infection prevalence (S1 Table). However, 6/8 Wolbachia-positive samples were negative for Spiroplasma, and hence the idea of reciprocal exclusion of both entities remains a possibility although the GLM did not suggest a significant correlation between the two bacterial entities. We had also noted that the Wolbachia density in the reproductive organs of Gff is unusually low, and that Spiroplasma densities can vary across seasons, thus rendering co-infection detections technically challenging, particularly when using whole body DNA, as was the case with a number of flies analyzed in this study.
It also remains to be elucidated if the presence of viral microorganisms plays a role for the infection dynamics of Spiroplasma in Ugandan Gff populations. The salivary gland hypertrophy virus (SGHV), a common virus of Glossina, is inversely correlated with Wolbachia infection prevalence in Gff in Uganda [28]. Hence, the correlation between Spiroplasma and SGHV remains to be tested.
Reproductive influences, such as male killing, have been associated with maternally inherited symbionts, including Spiroplasma, where the symbiont drives its own dispersal by selectively killing male embryos in ladybirds, fruit flies and certain butterfly species (reviewed in [41]). As we observed both Spiroplasma-infected males and females and there was also no sex bias in the offspring of the laboratory-reared Gff, this bacterium likely does not confer a male killing trait to Gff. Pairwise comparison suggested a slightly higher infection prevalence in females, but such a slight difference is unlikely to be the result of a male killing phenotype.
We also evaluated the potential role of Spiroplasma on tsetse’s immune physiology by measuring the correlation between trypanosome and Spiroplasma infection status. Of the 1415 samples analyzed, we found only five with trypanosome and Spiroplasma co-infections. Although not significant in the GLMM or GLM, this negative results could have been caused by lack of statistical power in the field collected data. To further address the question of a similar protective effect, we performed trypanosome infection experiments using a colonized Gff line that displays heterogeneous Spiroplasma infection prevalence [31]. Our finding that infections with Spiroplasma alone were more frequent than co-infections with Tbb (56% vs. 18%, respectively; S4 Table) indicates a negative correlation between the presence of the symbiont and the parasite, and suggests that both entities negatively impact each other’s fitness. In accordance with investigations in Drosophila [78,79], Spiroplasma infections may confer physiological traits that protect its tsetse host from being colonized by trypanosome infections. In different hosts, Spiroplasma induces a protective effect against nematodes, fungi and parasitoid wasps [35,36,78]. It remains to be elucidated, whether a potential protection in Gff against parasite infections results from niche competition of both microbes within the host, or by expression of certain Spiroplasma-derived molecules, which block trypanosomes. Alternatively, it has been shown that infections with certain Wolbachia strains confer an immune enhancement phenotype to their Drosophila and mosquito hosts, hence increasing the resistance to other pathogens [80; also reviewed in 81]. Such an immune enhancement, which affect the trypanosome transmission success in Spiroplasma infected Gffs however, is possible but rather unlikely given the low infection frequency of Wolbachia (12%) in the tested flies.
The heterogeneous Spiroplasma infection prevalence in the Gff line that we used in the parasite challenge experiment (Fig 4) might result from imperfect maternal transmission to progeny and may similarly be influencing the infection prevalence noted in natural populations. Tsetse are viviparous and the mother supports the development of her progeny in an intrauterine environment. Endosymbiotic Wigglesworthia and Sodalis are maternally acquired by the progeny in female milk secretions during the lactation process, while Wolbachia is transovarially transmitted. Spiroplasma infections in the gonads suggest that this endosymbiont is also transovarially transmitted, although transmission through milk secretions remains to be investigated. While tsetse females remain fecund throughout their entire life (and can produce 8–10 progeny), the transmission of Spiroplasma from mother to her intrauterine progeny however may be more efficient during the early gonotrophic cycles and decrease over the course of female’s reproductive lifespan. Such a scenario could explain why we observed that only 50–60% of colony flies are infected with Spiroplasma. We will further address this question by testing the efficiency of symbiont transmission from mother to each of her offspring in a follow-up study by developing single lines from each pregnant female. Such an imperfect transmission efficiency can also influence the heterogeneous infection prevalence we noted in field populations.
In conclusion, the infection prevalence of Spiroplasma in Gff populations in northern Uganda is significantly correlated with different watersheds of origin and seasonal environmental conditions. These associations indicate that seasonal fluctuations and other transmission modes than strictly vertically are drivers of Spiroplasma acquisition in Gff in Uganda.
We further demonstrate that colonized Gff are less likely to establish trypanosome parasite infections when carrying Spiroplasma infections, which is of particular interest in the context of alternative vector control approaches to control trypanosome infections.
|
10.1371/journal.pgen.1007559 | Transcriptome analysis of adult Caenorhabditis elegans cells reveals tissue-specific gene and isoform expression | The biology and behavior of adults differ substantially from those of developing animals, and cell-specific information is critical for deciphering the biology of multicellular animals. Thus, adult tissue-specific transcriptomic data are critical for understanding molecular mechanisms that control their phenotypes. We used adult cell-specific isolation to identify the transcriptomes of C. elegans’ four major tissues (or “tissue-ome”), identifying ubiquitously expressed and tissue-specific “enriched” genes. These data newly reveal the hypodermis’ metabolic character, suggest potential worm-human tissue orthologies, and identify tissue-specific changes in the Insulin/IGF-1 signaling pathway. Tissue-specific alternative splicing analysis identified a large set of collagen isoforms. Finally, we developed a machine learning-based prediction tool for 76 sub-tissue cell types, which we used to predict cellular expression differences in IIS/FOXO signaling, stage-specific TGF-β activity, and basal vs. memory-induced CREB transcription. Together, these data provide a rich resource for understanding the biology governing multicellular adult animals.
| C. elegans is the simplest multi-cellular model system, with only 959 somatic cells in the fully-developed adult. This work describes the isolation and RNA-seq analysis of the worm’s major adult tissues. Previously, the isolation of adult tissues has been hampered by the worm’s tough outer cuticle, but identification of the transcriptomes of adult tissues is necessary to understand the biology of adults, which differs substantially from that of embryonic and larval cells. We recently developed a method to isolate and RNA-sequence adult tissues, and applied it here to characterize the muscle, neuron, intestine, and epidermis adult transcriptomes and isoform profiles. The data reveal interesting new characteristics for adult tissues, particularly the hypodermis’ metabolic function, which we have functionally tested. The tissue transcriptomes were also used to identify relevant human tissue orthologs in an unbiased manner. Finally, we present a new prediction tool for gene expression in up to 76 tissues and cell types, and we demonstrate its utility not only in predicting cell-specific gene expression, but in diagnosing expression changes in different genetic pathways and contexts.
| Animals progress through many stages of development before reaching adulthood, and as adults, they exhibit metabolic and behavioral differences from developing animals. Studies in the nematode C. elegans demonstrate this phenomenon well: both biological responses and gene expression differ significantly in different stages [1,2]. Therefore, to understand the biology underlying tissue-specific adult behavior, it is critical to identify adult, tissue-specific transcriptomes.
The advent of whole-genome gene expression approaches allowed the identification of a cell’s full set of mRNA transcripts, ushering in a new era of understanding biological dynamics [3]. The ongoing development of new methods to isolate and sequence individual cells in order to approximate their metabolic and biochemical state has refined our understanding of single cells [4]. The next frontier in this work is the gene expression analysis of whole animals on a tissue-by-tissue and cell-by-cell basis. While tissue-specific expression has been measured in other organisms, the combination of extremely small tissue size and adult cuticle impermeability have previously prevented the analysis of adult worm tissue expression, which is necessary in order to understand adult processes, including systemic aging, tissue-specific aging, and cell non-autonomous control of aging. More broadly speaking, adult tissue-specific expression can be used to better understand signaling and cell autonomous processes and to compare expression to that in other adult organisms. The complexity of tissue autonomous and non-autonomous mechanisms of aging and disease requires the understanding of tissue-specific expression. The delineation of adult tissue expression presented here, combined with the genetic and molecular tools available in the worm, provide a unique chance to more directly model aging and disease compared to more complex organisms.
C. elegans is the simplest multicellular model system, with only 959 somatic (non-germline) cells in the fully developed adult animal. Four tissues—muscles, neurons, intestine, and the epidermis (or “hypodermis”)—comprise the bulk of the animal’s somatic cells and are largely responsible for the animal’s cell autonomous and non-autonomous biological regulation. Until recently, most transcriptional analyses of C. elegans adults utilized whole worms, but the need to identify tissue-specific transcripts in order to better understand both tissue-specific and non-autonomous signaling has become apparent. Several tissue profiling techniques that rely on PAB-mediated RNA immunoprecipitation have been widely used, but these methods often introduce very high non-specific background [5] and studies have not focused specifically on adult animals [1,6,7]. Recent spliced-leader RNA-tagging methods [8] that avoid this problem are also limited, since only 50–60% of C. elegans genes exhibit SL1-based trans-splicing [9]. Furthermore, tools used to isolate embryonic and larval stage C. elegans cells using cell sorting [1,10–13] have allowed the transcriptional profiling of specific tissues and cell types, shedding light on larval development processes, but lack information specific to adult tissues. Much of worm behavioral analysis, and all aging studies—for which C. elegans is a premier model system—[14] are, not is performed in adults, which are less amenable to standard isolation approaches due to their tough outer cuticle. Therefore, we developed a method to disrupt and isolate adult tissues [2]. That work revealed that the adult neuronal transcriptome differs significantly from earlier embryonic and larval stages, and that the adult neuronal transcriptome best reveals genes involved in behavior and neuronal function. The other major tissues—muscle, intestine, and hypodermis—are likely to provide insight into important adult-specific processes that are widely studied in C. elegans as models of human biology, such as pathogenesis, reproduction, age-related decline, and others.
Here we have performed cell-specific transcriptional analysis and characterization of the four major somatic tissues isolated from adult worms. As examples of the utility of these data, we used the highly enriched tissue gene sets to identify transcriptional parallels between worm and human tissues and to determine the tissue specificity of DAF-16 transcriptional targets. Additionally, our sequencing method allowed the identification of tissue-specific alternatively spliced gene isoforms, which we have used to explore tissue-specific collagen isoform expression. Finally, we present a tool that predicts gene expression in 76 different sub-tissue cell types, and demonstrate its utility in the characterization of individual genes, gene classes, and potential cellular differences in gene expression for several different signaling pathways. Together, these data provide a rich resource for the examination of adult gene expression in C. elegans.
To identify the transcriptomes of adult C. elegans tissues, it is necessary to break open the outer cuticle and release, filter, and sort cells while minimizing cell damage [2]. We collected 27 Day 1 adult tissue samples (7 neuron, 5 intestine, 7 hypodermis, 8 muscle), utilizing strains with fluorescently-marked neurons (Punc-119::gfp), muscle (Pmyo-3::mCherry), hypodermis (pY37A1B.5::gfp), and intestine (Pges-1::gfp; Fig 1A; see Methods for details).
Multidimensional scaling analysis (Fig 1B) suggests that the samples cluster best with their respective tissue types, and that muscle and hypodermis are most closely related, while neuronal and intestine samples are more distinct from one another. Subsampling analysis [15], which determines whether sequencing has been performed to sufficient depth, suggests that this estimate of gene expression is stable across multiple sequencing depths (S1A Fig), and thus gene expression differences represent true differences between tissues.
We obtained reads across the whole transcript length (rather than selecting the 3’ end of mRNA via the polyA tail) in order to analyze tissue-specific isoform expression (see below). To assess RNA degradation in each sample, we determined the gene body coverage for all 20,389 protein-coding genes [16]; the transcripts have consistent, uniform coverage, with best coverage within the gene bodies (S1B Fig).
“Expressed” genes are defined as those with both (1) an average log(rpkm) greater than 2, and (2) with each replicate of that tissue having a log(rpkm) greater than 1, resulting in the detection of 8437 neuron, 7691 muscle, 7191 hypodermis, and 9604 intestine protein-coding genes (Fig 1C, S1 Table); 5360 genes are expressed in all sampled tissues. Hierarchical clustering of the top 2000 differentially-expressed genes per sample across the four tissue types shows that intra-group tissue samples are most similar, specific genes characterize particular tissue types (especially neurons), and that there is a subgroup of genes expressed in all tissues (Fig 1D). As expected, Gene Ontology (GO) analysis of the ubiquitously-expressed gene set shows that basic cell biological and metabolic processes are shared, including such terms as intracellular transport, protein metabolism, catabolism, glucose metabolism, ribosome biogenesis, translation elongation, maintenance of cell polarity, and microtubule-based process (Fig 1E; S2 Table). Additionally, terms associated with protection of the cell, such as response to stress, autophagy, protein folding, gene silencing by RNAi, and determination of adult lifespan appear in the ubiquitous category.
OH441: otIs45[Punc-119::GFP], CQ163: wqEx34[Pmyo-3::mCherry], CQ171: [Py37a1b.5::GFP], BC12890: [dpy-5(e907)I; sIs11337(rCesY37A1B.5::GFP + pCeh361), SJ4144: zcIs18 (Pges-1::GFP), CQ236: Pcrh-1g::GFP + Pmyo-2::mcherry. Worm strains were maintained at 20°C on HGM plates using E. coli OP50. Strains were synchronized using hypochlorite treatment prior cell isolation and grown to day 1 of adulthood on HGM plates with E. coli OP50.
Synchronized day 1 adult worms with GFP-labeled neurons, muscle, hypodermis, and intestine (Punc119::GFP, Pmyo-3::mCherry, pY37A1B.5::GFP, and Pges-1::GFP) were prepared for cell isolation, as previously described [2].
Cells were briefly subjected to SDS-DTT treatment, proteolysis, mechanical disruption, cell filtering, FACS, RNA amplification, library preparation, and single-end (140 nt) Illumina sequencing, as previously described [2]. Neuron cell suspensions were passed over a 5 μm syringe filter (Millipore). Muscle and hypodermal samples were gently passed over a 20 mm nylon filter (Sefar Filtration). Intestinal cells were passed through a 35 mm filter and by spinning at 500 x g for 30s in a tabletop centrifuge. The filtered cells were diluted in PBS/2% FBS and sorted using a either a FACSVantage SE w/ DiVa (BD Biosciences; 488nm excitation, 530/30nm bandpass filter for GFP detection) or a Bio-Rad S3 Cell Sorter (Bio-Rad; 488nm excitation). Gates for detection were set by comparison to N2 cell suspensions prepared on the same day from a population of worms synchronized alongside the experimental samples. Positive fluorescent events were sorted directly into Eppendorf tubes containing Trizol LS for subsequent RNA extraction. For each sample, approximately 50,000–250,000 GFP or mCherry positive events were collected, yielding 5–25 ng total RNA. Both sorters were used for each tissue, and the type of sorter did not affect the distribution of samples by multidimensional scaling analysis (Fig 1B), suggesting that the sorter did not contribute to the variability between samples of a given tissue.
RNA was isolated from FACS-sorted samples as previously described [2]. Briefly, RNA was extracted using standard Trizol/ chloroform/ isopropanol method, DNase digested, and cleaned using Qiagen RNEasy Minelute columns. Agilent Bioanalyzer RNA Pico chips were used to assess quality and quantity of isolated RNA. 10 to 100 ng of the isolated quality assessed RNA was then amplified using the Nugen Ovation RNAseq v2 kit, as per manufacturer suggested practices. The resultant cDNA was then sheared to an average size of ~200 bp using Covaris E220. Libraries were prepared using either Nugen Encore NGS Library System or the Illumina TruSeq DNA Sample Prep, 1 μg of amplified cDNA was used as input. RNA from a subset of samples was amplified using the SMARTer Stranded Total RNA kit-pico input mammalian, as per manufacturer suggested practices. No differences were observed between the two methods, and samples amplified by different methods clustered well (Fig 1B). The resultant sequencing libraries were then submitted for sequencing on the Illumina HiSeq 2000 platform. 35–200 million reads (average of 107,674,388 reads) were obtained for each sample and mapped to the C. elegans genome. Sequences are deposited at NCBI BioProject PRJNA400796.
FASTQC was used to inspect the quality scores of the raw sequence data, and to look for biases. The first 10 bases of each read were trimmed before adapter trimming, followed by trimming the 3’ end of reads to remove the universal Illumina adapter and imposing a base quality score cutoff of 28 using Cutadapt v1.6 The trimmed reads were mapped to the C. elegans genome (Ensembl 84/WormBase 235) using STAR [72] with Ensembl gene model annotations (using default parameters). Count matrices were generated for the number of reads overlapping with the gene body of protein coding genes using featureCounts [73]. The per-gene count matrices were subjected to an expression detection threshold of 1 count per million reads per gene in at least 5 samples. EdgeR [74] was used for differential expression analysis and the multidimensional scaling (MDS) analysis. MDS is a method that aims to visualize proximity data in such a way that best preserves between-sample distances and is a commonly used technique (similar to PCA) to transform higher-dimension dissimilarity data into a two-dimensional plot. Here, we used the log-fold-change of expression between genes to compute distances. Genes at FDR = 0.05 were considered significantly differentially expressed. DEXSeq [75] was used for differential exon usage (splicing) analysis.
Count matrices of the aligned sequencing data were down-sampled using subSeq [15]. Reads were down-sampled at proportions using 10^x, starting at x = -5 and increasing at 0.25 increments to 0. The down-sampled count matrices were used to assess stability of number of expressed genes detected at multiple depths (S1A Fig). Because of minimum library sizes for tractable differential exon usage analysis, reads with down-sampled proportions using 10^x, from x = -2, increasing at 0.25 increments to 0 were used for assessment of power in detecting differential splicing (S1B Fig).
Hypergeometric tests of Gene Ontology terms were performed on tissue-enriched gene lists; GO terms reported are a significance of q-value < 0.05 unless otherwise noted. REVIGO was used to cluster and plot GO terms with q-value < 0.05.
RSAtools [76] was used to identify the -1000 to -1 promoter regions of the tissue enriched genes and perform motif analysis. Matrices identified from RSAtools were analyzed using footprintDB [77] to identify transcription factors predicted to bind to similar DNA motifs. Alternatively, motifs were analyzed using gProfiler [78].
Hypodermal genes appearing in metabolic GO terms were selected from the top of the tissue-enriched list (aldo-2, gpd-2, sams-1, cth-2, pmt-1, idh-1, and fat-2) or the expressed list (far-2 and gpd-3) and knocked down using RNAi and compared to a vector (L4440) control. On day 1 of adulthood, all worms were stained in Oil Red O for 6–24 hours and then imaged using a Nikon Eclipse Ti microscope at 20x magnification [79]. Images were analyzed for mean intensity in fat objects using CellProfiler [80]. Additional genes from the hypodermal unique list were also selected and tested for fat (Oil Red O) levels.
Human orthologs [30] of genes in our tissue-enriched gene lists were compared with curated tissue-specific gene annotations from the Human Protein Reference Database [31] for significant overlap (hypergeometric test).
‘Tissue-enriched’ genes are highly enriched relative to all other tissues, defined as genes that are highly expressed (logRPKM > 5) and significantly differentially expressed relative to the average expression across all of the other tissues (FDR ≤ 0.05, logFC > 2; S8 Table).
‘Unique’ tissue-specific genes are strongly expressed (logRPKM > 5) and significantly differentially expressed in comparison to the expression of each of the three other tissues (FDR ≤ 0.05, logFC > 2 for each comparison; S9 Table, S1E Fig).
The expression level (expressed defined as log(rpkm) >2) for previously published IIS/FOXO targets (Tepper et al., 2013, cut-off 5% FDR) were identified for each tissue. Tissue overlaps were graphed in Venn diagrams using the Venn diagram package in R.
To construct these models, we needed a large data compendium and high quality examples of tissue expression. We assembled 273 C. elegans expression datasets (comprised of both adult and developmental expression data), representing 4,372 microarray and RNA-seq samples, including our tissue-ome library. All other datasets were downloaded from the Gene Expression Omnibus (GEO) data repository, ArrayExpress Archive of Functional Expression Data, and WormBase. Samples from each dataset were processed together (duplicate samples were collapsed, background correction and missing value imputation were executed when appropriate). Within each dataset, gene expression values were normalized to the standard normal distribution. All datasets were then concatenated, and genes that were absent in only a subset of datasets received values of 0 (in datasets in which they were absent). The predictions that were used to analyze the tissue-ome dataset were generated using a data compendium that excluded the tissue-ome library.
Gene annotations to tissues and cell types were obtained from curated anatomy associations from WormBase [81] (WS264) and other small-scale expression analyses as curated by Chikina et al. (2009). Only annotations based on smaller scale experiments (e.g., single-gene GFP, in situ experiments) were considered for the gold standard, excluding annotations derived from SAGE, microarray, RNA-seq, etc. Annotations from both adult and developing worm were considered. Annotations were mapped and propagated (up to each of its ancestor terms, e.g., a gene directly annotated to dopaminergic neuron would thus be propagated up to ancestor terms such as neuron and nervous system and included in the corresponding gold standards) based on the WormBase C. elegans Cell and Anatomy Ontology, where a stringent cutoff was used for which tissues and cell types were retained (>50 direct annotations and >150 propagated annotations).
We defined a “tissue-slim” based on system-level anatomy terms in the WormBase anatomy ontology (immediate children of “organ system” and “sex specific entity,” under “functional system”). The nine resulting terms are: alimentary system, coelomic system, epithelial system, excretory secretory system, hermaphrodite-specific, male-specific, muscular system, nervous system, and reproductive system. For each of the 76 tissues that were retained, a tissue-gene expression gold standard was constructed in which genes annotated (directly or through propagation, i.e., the gene has been associated with either the particular tissue or a part of that tissue in a smaller scale experiment) to the tissue were considered as positive examples. Genes that were annotated to other tissues, but not in the same tissue system, were considered negative examples. Thus, genes were assigned as positive or negative examples of tissue expression while taking into account the tissue hierarchy represented in the Cell and Anatomy Ontology.
Our tissue-gene expression predictions and similarity profiles have all been made accessible at a dynamic, interactive website, http://worm.princeton.edu. From this interface, users can explore the predicted expression patterns of their gene(s) of interest. To facilitate this exploration, we have designed an interactive heatmap visualization that allows users to view hierarchically clustered expression patterns or sort by any gene or tissue model of interest. In addition, we also provide suggestions of genes with similar tissue expression profiles, which users can immediately visualize alongside their original query. All predictions and visualizations are available for direct download.
For each of the 76 tissues and cell types, we used the expression data compendium and corresponding gold standard as input into a linear support vector machine (SVM) to make predictions for every gene represented in our data. Specifically, given the vector of gene expression data (xi) and training label (yi:{-1,1}) for gene i, hyperplanes described by w and b, and constant c, the SVM’s objective function is:
minw,ξ12wTw+c∑iξi,subjecttotheconstraints:yi(w⋅xi−b)≥1−ξi,ξi≥0.
SVM parameters were optimized for precision at 10% recall under 5-fold cross validation. Resulting SVM scores were normalized to the standard normal distribution for any comparisons across tissues. Feature weights for each of the tissue SVM models were also retained for ranking and analysis of samples.
|
10.1371/journal.pcbi.1006961 | Physical constraints on accuracy and persistence during breast cancer cell chemotaxis | Directed cell motion in response to an external chemical gradient occurs in many biological phenomena such as wound healing, angiogenesis, and cancer metastasis. Chemotaxis is often characterized by the accuracy, persistence, and speed of cell motion, but whether any of these quantities is physically constrained by the others is poorly understood. Using a combination of theory, simulations, and 3D chemotaxis assays on single metastatic breast cancer cells, we investigate the links among these different aspects of chemotactic performance. In particular, we observe in both experiments and simulations that the chemotactic accuracy, but not the persistence or speed, increases with the gradient strength. We use a random walk model to explain this result and to propose that cells’ chemotactic accuracy and persistence are mutually constrained. Our results suggest that key aspects of chemotactic performance are inherently limited regardless of how favorable the environmental conditions are.
| One of the most ubiquitous and important cell behaviors is chemotaxis: the ability to move in the direction of a chemical gradient. Due to its importance, key aspects of chemotaxis have been quantified for a variety of cells, including the accuracy, persistence, and speed of cell motion. However, whether these aspects are mutually constrained is poorly understood. Can a cell be accurate but not persistent, or vice versa? Here we use theory, simulations, and experiments on cancer cells to uncover mutual constraints on the properties of chemotaxis. Our results suggest that accuracy and persistence are mutually constrained.
| Chemotaxis plays a crucial role in many biological phenomena such as organism development, immune system targeting, and cancer progression [1–4]. Specifically, recent studies indicate that chemotaxis occurs during metastasis in many different types of cancer [2, 5–9]. At the onset of metastasis, tumor cells invade the surrounding extracellular environment, and oftentimes chemical signals in the environment can direct the migration of invading tumor cells. Several recent experiments have quantified chemotaxis of tumor cells in the presence of different chemoattractants [3] and others have been devoted to the intracellular biochemical processes involved in cell motion [10]. Since the largest cause of death in cancer patients is due to the metastasis, it is important to understand and prevent the directed and chemotactic behavior of invading tumor cells.
Chemotaxis requires sensing, polarization, and motility [11]. A cell’s ability to execute these interrelated aspects of chemotaxis determines its performance. High chemotactic performance can be defined in terms of several properties. Cell motion should be accurate: cells should move in the actual gradient direction, not a different direction. Cell motion should be persistent: cells should not waste effort moving in random directions before ultimately drifting in the correct direction. Cell motion should be fast: cells should arrive at their destination in a timely manner.
Indeed, most studies of chemotaxis use one or more of these measures to quantify chemotactic performance. Accuracy is usually quantified by the so-called chemotactic index (CI), most often defined in terms of the angle made with the gradient direction [12–15] (Fig 1A); although occasionally it is defined in terms of the ratio of distances traveled [16] or number of motile cells [17–19] in the presence vs. absence of the gradient. Directional persistence [10] (DP) is usually quantified by the ratio of the magnitude of the cell’s displacement (in any direction) to the total distance traveled by the cell (Fig 1A; sometimes called the McCutcheon index [20], length ratio [21], or straightness index [22]), although recent work has pointed out advantages of using the directional autocorrelation time [21, 23]. Speed is usually quantified in terms of instantaneous speed along the trajectory or net speed over the entire assay.
However, the relationship among the accuracy, persistence, and speed in chemotaxis, and whether one quantity constrains the others, is not fully understood. Are there cells that are accurate but not very persistent, or persistent but not very accurate (Fig 1B)? If not, is it because such motion is possible but not fit, or is it because some aspect of cell motion fundamentally prohibits this combination of chemotactic properties?
Here we focus on how a cell’s intrinsic migration mechanism as well as properties of the external environment place constraints on its chemotactic performance. The physics of diffusion places inherent limits on a cell’s ability to sense chemical gradients [24]. These limits, along with the cell’s internal information processing and its motility mechanism, determine the accuracy, persistence, and speed of migration. Using a human breast cancer cell line (MDA-MB-231) embedded within a 3D collagen matrix inside a microfluidic device imposing a chemical gradient, we are able to quantify the chemotactic performance of invasive cancer cells in response to various chemical concentration profiles. Results from chemotaxis assays are then compared with simulations and theoretical predictions in order to probe the physical limits of cancer cells to chemotaxis.
We measure accuracy using the chemotactic index (CI) [12–15]
CI ≡ ⟨ cos θ ⟩ , (1)
where θ is the angle the cell’s displacement makes with the gradient direction (Fig 1A), and the average is taken over many cell trajectories. CI is bounded between −1 and 1. For chemotaxis in response to an attractant, as in this study, CI generally falls between 0 and 1; whereas in response to a repellent, CI usually falls between −1 and 0. CI = 1 represents perfectly accurate chemotaxis in which cell displacement is parallel to the gradient direction (Fig 1B, top two examples), and CI = 0 indicates that the cells’ migration is unbiased (Fig 1B, bottom two examples). The facts that CI is bounded and dimensionless make it easy to compare different values across different experimental conditions, and get an intuitive picture for the type of cell dynamics it represents.
We measure persistence using the directional persistence (DP), defined as the ratio of the magnitude of the cell’s displacement (in any direction) to the total distance traveled [20–22] (Fig 1A),
DP ≡ ⟨ | displacement | distance ⟩ . (2)
Note that this ratio goes by several names [20–22], and although the name we use here contains the word ‘chemotactic,’ the ratio is in fact independent of the gradient direction. Indeed, DP measures the tendency of a cell to move in a straight line, in any direction. DP is also dimensionless and bounded between 0 and 1, and once again intuitive sense can be made of either limit. If DP = 1, then the cells are moving in perfectly straight lines in any arbitrary direction (Fig 1B, right two examples). In contrast, a low DP is representative of a cell trajectory that starts and ends near the same location on average (Fig 1B, left two examples), with DP → 0 in the limit of an infinitely long non-persistent trajectory.
An alternative measure of persistence is the directional autocorrelation time τ AC = ∫ 0 ∞ d t ′ 〈 cos ( θ t + t ′ − θ t ) 〉, where t′ is the time difference between two points in a trajectory, and the average is taken over all starting times t [21, 23]. The advantage of the autocorrelation time is that, unlike the DP, it is largely independent of the measurement frequency and total observation time. The disadvantage is that, unlike the DP, it is not dimensionless or bounded. Although we use the DP here, we verify in S1 Fig that the autocorrelation time varies monotonically with the DP for our experimental assay.
We measure speed using the instantaneous speed along the trajectory. That is, we take the distance traveled in the measurement interval Δt (15 minutes in the experiments, see below), divide it by the interval, and average this quantity over all intervals that make up the trajectory.
We begin by investigating the above properties of chemotaxis in the context of metastasis, specifically the epithelial-mesenchymal transition and subsequent invasion of cancer cells. To this end, we perform experiments using a triple-negative human breast cancer cell line (MDA-MB-231). Invasion of tumor cells in vivo is aided by external cues including soluble factors that are thought to form gradients in the tumor microenvironment [2, 5–9]. Among these soluble factors, transforming growth factor-β (TGF-β) is a key environmental cue for the invasion process [2, 25–28]. Therefore, we use TGF-β as the chemoattractant.
The in vivo tumor microenvironment is highly complex. As a result, in vitro platforms have been developed and widely used to investigate the cancer response to a specific cue. In this study, a microfluidic platform is used to expose the TGF-β gradient to the cells in 3D culture condition (Fig 2A). The microfluidic device is designed with three different channels, a center, source, and sink channel (Fig 2B). The center channel is filled with a composition of MDA-MB-231 cells and type I collagen while the medium is perfused through the side source and sink channels. TGF-β is applied only through the source channel, not the sink channel, and therefore a graded profile develops over time in the center channel by diffusion. Consequently, the MDA-MB-231 cells surrounded by type I collagen are exposed to a chemical gradient of TGF-β.
To verify that a graded TGF-β profile is generated in the center channel, we utilize 10kDa FITC-dextran, whose hydrodynamic radius (2.3 nm) is similar to that of TGF-β (approximately 2.4 nm [29]). The fluorescence intensity is shown in Fig 2C. The profile approaches steady state within 3 hours, is approximately linear, and remains roughly stationary for more than 12 hours. Therefore, we record the MDA-MB-231 cells using time-lapse microscopy every 15 minutes from 3 to 12 hours after imposing the TGF-β. See Materials and methods for details.
First, we perform a control experiment with no TGF-β to characterize the baseline of the MDA-MB-231 cell migratory behavior. Representative trajectories are shown in Fig 3A, and we see that there is no apparent preferred direction. Indeed, as seen in Fig 3C (black), the CI is centered around zero, indicating no directional bias. Notably, the spread of the CI values is very broad, with many data points falling near the endpoints −1 and 1. This is a generic feature of the CI due to its definition as a cosine: when the distribution of angles θ is uniform, the distribution of cos θ is skewed toward −1 and 1 because of the cosine’s nonlinear shape. Nonetheless, we see that the median of the CI is very near zero as expected. The speed and DP are shown in Fig 3D and 3E, respectively (black). We see that the DP is significantly above zero, indicating that even in the absence of any chemoattractant, cells exhibit persistent motion. This result is consistent with previous works that showed that cells cultured in 3D tend to have directionally persistent movement unlike those in 2D [10].
Next, we expose cells to a TGF-β gradient of g = 50 nM/mm. Representative trajectories are shown in Fig 3B, and we see a possible bias in the gradient direction. Indeed, as seen in Fig 3C (red), the CI is centered above zero, indicating a directional bias, and the difference with the control distribution is statistically significant (p value < 0.05). We also see in Fig 3D (red) that the speed increases, although we will see below that the increase is relatively small and that the trend is non necessarily monotonic. Finally, we see in Fig 3E (red) that the DP decreases, although the difference with the control is not statistically significant. These results suggest that a TGF-β gradient causes a significant increase in directional bias (CI) but not necessarily a significant change in cell speed or persistence (DP).
To confirm the trends suggested above, we evaluate the response to four different TGF-β gradient strengths, g = 0, 1, 5, and 50 nM/mm, in three separate experiments each (Fig 4A–4C; the trajectories for all experiments and g values are shown in S2 Fig). We see in Fig 4A that, consistent with Fig 3, the CI is zero for the control and increases with gradient strength g. In fact, the CI appears to saturate beyond 5 nM/mm, such that its value at 50 nM/mm is not significantly larger than its value at 5 nM/mm. We also see in Fig 4B, consistent with Fig 3, the DP slightly decreases with the gradient strength although the decrease is roughly within error bars. Finally, we see in Fig 4C that the increase in the speed is small, achieving a statistically significant difference with the control only at the largest gradient strength, and that the trend is not monotonic.
A striking feature of Fig 4A is that the cells respond to a gradient as shallow as g = 5 nM/mm. To put this value in perspective, we estimate both the relative concentration change and the absolute molecule number difference across the cell body [4]. The microfluidic device is about 1 mm in the gradient direction, and therefore a cell in the middle experiences a background concentration of about c = 2.5 nM. Assuming the cell is on the order of a = 10 μm wide, the change in concentration across its body is ga = 0.05 nM, for a relative change of ga/c = 2%. The number of attractant molecules that would occupy half the cell body is on the order of ca3 = 1500. Two percent of this is ga4 = 30, meaning that cells experience about a thirty-molecule difference between their two halves. The same quantities are approximately ga/c = 1% and 6%, and ga4 = 60 and 300, for amoebae in cyclic adenosine monophosphate gradients [14] and epithelial cells in epidermal growth factor gradients [30], respectively [4]. This suggests that the response of MDA-MB-231 cells to TGF-β gradients is close to the physical detection limit for single cells.
To understand the experimental observation that the CI increases with gradient strength, but the DP and speed do not (Fig 4A–4C), we turn to computer simulations. The cells in the experiments are executing 3D migration through the collagen matrix (as opposed to crawling on top of a 2D substrate). Nevertheless, the imaging is acquired as a 2D projection of the 3D motion. We do not expect this projection to introduce much error into the analysis because the height of the microfluidic device is less than 100 μm, whereas its width in the gradient direction is about 1 mm, and its length is several millimeters. Indeed, from the experimental trajectories (Fig 3) we have estimated that if motility fluctuations in the height direction are equivalent to those in the length direction, then the error in the CI that we make by the fact that we only observe a 2D projection of cell motion is less than 1%. Consequently, for simplicity we use a 2D rather than 3D simulation of chemotaxis of a cell through an extracellular medium.
Specifically, we use the cellular Potts model (CPM) [31, 32], a lattice-based simulation that has been widely used to model cell migration [33–35] (note that whereas often the CPM is used to model collective migration, here we use it for single-cell migration). In the CPM, a cell is defined as a finite set of simply connected sites on a regular square lattice (Fig 5). The cell adheres to the surrounding collagen with an adhesion energy α and has a basal area A0 from which it can fluctuate at an energetic cost λ. This gives the energy function
u = α L + λ ( A − A 0 ) 2 , (3)
where L and A are the cell’s perimeter and area, respectively.
Cell motion is a consequence of minimizing the energy u subject to thermal noise and a bias term w that incorporates the response to the gradient [33]. Specifically, for a lattice with S total sites, one update step occurs in a fixed time τ and consists of S attempts to copy a random site’s label (cell or non-cell) to a randomly chosen neighboring site. Each attempt is accepted with probability
P = { e − ( Δ u − w ) Δ u − w > 0 1 Δ u − w ≤ 0 , (4)
where Δu is the change in energy associated with the attempt. The bias term is defined as
w = Δ x → · p → , (5)
where Δ x → is the change in the cell’s center of mass caused by the attempt, and p → is the cell’s polarization vector (Fig 5 inset, black arrow), described below. The dot product acts to bias cell motion because movement parallel to the polarization vector results in a more positive w, and thus a higher acceptance probability (Eq 4).
The polarization vector is updated every time step τ according to
Δ p → τ = r ( − p → + η Δ x ^ τ + ϵ q → ) . (6)
The first term in Eq 6 represents exponential decay of p → at a rate r. Thus, r−1 characterizes the polarization vector’s memory timescale. The second term causes alignment of p → with Δ x ^ τ according to a strength η, where Δ x ^ τ is a unit vector pointing in the direction of the displacement of the center of mass in the previous time step τ. Thus, this term promotes persistence because it aligns p → in the cell’s previous direction of motion. The third term causes alignment of p → with q → according to a strength ϵ, where q → contains the gradient sensing information, as defined below. Thus, this term promotes bias of motion in the gradient direction.
The sensing vector q → is an abstract representation of the cell’s internal gradient sensing network and is defined as
q → = ⟨ ( n i − n ¯ ) r ^ i ⟩ , (7)
where the average is taken over all lattice sites i that comprise the cell, and receptor saturation is incorporated as described below. The unit vector r ^ i points from the cell’s center of mass to site i, the integer ni represents the number of TGF-β molecules detected by receptors at site i, and n ¯ is the average of ni over all sites. The integer ni is the minimum of two quantities: (i) the number of TGF-β receptors at site i, which is sampled from a Poisson distribution whose mean is the total receptor number N divided by the number of sites; and (ii) the number of TGF-β molecules in the vicinity of site i, which is sampled from a Poisson distribution whose mean is (c + gxi)ℓ3, where ℓ is the lattice spacing, and xi is the position of site i along the gradient direction. Taking the minimum incorporates receptor saturation, since each site cannot detect more attractant molecules than its number of receptors. The subtraction in Eq 7 makes q → a representation of adaptive gradient sensing: if receptors on one side of the cell detect molecule numbers that are higher than those on the other side, then q → will point in that direction. Adaptive sensing has been observed in the TGF-β pathway [36] in the form of fold-change detection [37] (for shallow gradients, subtraction as in Eq 7 is similar to taking a ratio as in fold-change detection [30]).
The simulation is performed at a fixed background concentration c and gradient g for a total time T. The position of the cell’s center of mass is recorded at time intervals Δt, from which we compute the CI, DP, and speed.
The parameter values used in the simulation are listed in Table 1 and are set in the following way. The values T = 9 h, Δt = 15 min, c = 2.5 nM, and g = 5 nM/mm are taken from the experiments. We estimate A0 = 400 μm2 from the experiments, and we take ℓ = 2 μm, such that a cell typically comprises A0/ℓ2 = 100 lattice sites. We find that realistic cell motion is sensitive to α: when α is too small the cell is diffuse and unconnected, whereas when α is too large the cell does not move because the cost of perturbing the perimeter is too large. The crossover occurs around α ∼ ℓ−1 as expected, and therefore we set α on this order, to α = 2 μm−1. In contrast, we find that cell motion is not sensitive to λ (apart from λ = 0 for which the cell evaporates), and therefore we set λ = 0.01 μm−4 corresponding to typical area fluctuations of λ−1/2/A0 = 2.5%. In order for our Poisson sampling procedure to be valid, the time step τ must be much larger than the timescale ℓ2/D for an attractant molecule or receptor to diffuse with coefficient D across a lattice site. Taking D ∼ 10 μm2/s, we find τ ≫ 0.4 s. At the other end, we must have τ < Δt = 900 s for meaningful data collection. We find that within these bounds, results are not sensitive to τ, and therefore we set τ on the larger end at τ = 100 s to reduce computational run time.
The parameters N, η, and ϵ are calibrated from the experimental data in Fig 4A–4C. Specifically, N sets the gradient value above which the CI saturates (see Fig 4A) because if the gradient is large but N is small, the cell quickly migrates into a region in which there are more attractant molecules than receptors at all lattice sites, and gradient detection is not possible. We find that N = 10,000, which is a reasonable value for the number of TGF-β receptors per cell [38, 39], places the saturation level at roughly g = 50 nM/mm as in the experiments (Fig 4D). We set ϵ = 56 μm−1 and η = 107 μm−1 to calibrate their cognate observables, CI and DP, respectively, to the corresponding experimental values at g = 5 nM/mm (Fig 4D and 4E).
The final parameter is the memory timescale of the polarization vector, r−1. As seen in Fig 4E (gray), we find that the behavior of the DP depends sensitively on this timescale. When r−1 is large, the DP increases with gradient strength. In contrast, when r−1 is small (indeed, equal to the smallest timescale in the system, τ), the DP does not increase with gradient strength, and in fact slightly decreases (Fig 4E, blue). Because the latter behavior is consistent with the experiments (Fig 4B), we set r−1 = τ. We conclude that the memory timescale of MDA-MB-231 cells is very short when responding to TGF-β gradients.
We validate the simulation in two ways, using the speed. First, we find that the magnitude of the speed in the simulations is on the same order as the speed in the experiments (Fig 4C and 4F), i.e., tens of microns per hour. Second, we find that the speed shows little dependence on the gradient strength in both the simulations and the experiments: it slightly increases in Fig 4C and slightly decreases in Fig 4F. Considering that the speed is not calibrated directly in our simulations, these consistencies validate the CPM as a reasonable description of the cell migration in the experiments.
Our finding that the cell’s memory timescale r−1 takes its minimum value allows for the following interpretation: the parameter r couples the persistence term and the sensory term in the CPM (Eq 6). Thus, when the memory timescale r−1 is long, biased motion must be also persistent and vice versa. In contrast, when the memory timescale r−1 is short, it is possible for bias to increase without increasing persistence. Therefore, the simulations suggest that the reason that CI but not DP increases with gradient strength in the experiments, is that the drivers of sensory bias and migratory persistence in the cell’s internal network are decoupled from one another.
Our finding that bias and persistence are decoupled in the simulations allows us to appeal to a much more simplified theoretical model in order to understand and predict global constraints on chemotaxis performance. Specifically, we consider the biased persistence random walk (BPRW) model [40, 41], in which bias and persistence enter as explicitly independent terms controlled by separate parameters. The BPRW has been shown to be sufficient to capture random and directional, but not periodic, behaviors of 3D cell migration [42]. Because we do not observe periodic back-and-forth motion of cells in our experiments, we propose that the BPRW is sufficient to investigate chemotactic constraints here.
As in the simulations, we consider the BPRW model in 2D. In the BPRW model, a cell is idealized as a single point. Its trajectory consists of M steps whose lengths are drawn from an exponential distribution. We take M = T/Δt = 36 as in the experiments. The probability of a step making an angle θ with respect to the gradient direction is
P ( θ | θ ′ ) = b cos θ ︸ bias + e p cos ( θ − θ ′ ) 2 π I 0 ( p ) ︸ persistence , (8)
where θ′ is the angle corresponding to the previous step. The first term incorporates the bias, with strength b. It is maximal when the step points in the gradient direction (θ = 0) and therefore promotes bias in that direction. It integrates to zero over its range (−π < θ < π) because the bias term only reshapes the distribution without adding or subtracting net probability. The second term incorporates the persistence, with strength p. It is a von Mises distribution (similar to a Gaussian distribution, but normalized over the finite range −π < θ < π) whose sharpness grows with p. It is maximal at the previous angle θ′ and therefore promotes persistence. The normalization factor I0 is the zeroth-order modified Bessel function of the first kind.
The requirement that P(θ|θ′) be non-negative over the entire range of θ mutually constrains b and p. However, apart from this constraint, b and p can take any positive value. We sample many pairs of b and p, reject those that violate the constraint, and compute the CI and DP from a trajectory generated by each remaining pair. The results are shown in Fig 6 (colored circles). We see in Fig 6 that the BPRW model exists in a highly restricted ‘crescent’ shape within CI–DP space. As expected, the CI increases with the bias parameter b (color of circles, from blue to red). The top corner corresponds to maximal bias and no persistence; indeed, when p = 0 the persistence term in Eq 8 reduces to (2π)−1, and non-negativity requires b < (2π)−1 ≈ 0.16, which is consistent with the upper limit of the color bar. Also as expected, the DP increases with the persistence parameter p (size of circles, from small to large), although only in the lower portion where the CI is low.
The crescent shape of the allowed CI and DP values in Fig 6 can be understood quantitatively because several moments of the BPRW are known analytically [41]. Specifically, the mean squared displacement and the mean displacement in the gradient direction are, in units of the mean step length,
⟨ r 2 ⟩ = 1 ( 1 − ψ ) 2 [ z 2 M ˜ 2 + 2 ( 1 − 2 z 2 − z 2 e − M ˜ ) M ˜ + 2 ( 2 z 2 − 1 ) ( 1 − e − M ˜ ) + 2 z 2 ( 1 − e − M ˜) 2 ] , (9)
⟨ x ⟩ = z 1 − ψ ( M ˜ − 1 + e − M ˜ ) , (10)
respectively, where M ˜ = M ( 1 − ψ ) and z = χ/(1 − ψ), with χ = ∫ − π π d θ b cos 2 θ = π b and ψ = ∫ − π π d ϕ [ 2 π I 0 ( p ) ] − 1 e p cos ϕ cos ϕ = I 1 ( p ) / I 0 ( p ). We approximate the CI and DP in terms of these moments,
CI= ⟨ x r ⟩ ≈ ⟨ x ⟩ ⟨ r ⟩ ≈ ⟨ x ⟩ ⟨ r 2 ⟩ , (11)
DP= ⟨ r ⟩ M ≈ ⟨ r 2 ⟩ M , (12)
and evaluate these expressions in specific limits to approximate the edges of the shape. In the limit b = 0, Eq 11 reduces to CI = 0 (bottom black line in Fig 6). In the limit p = 0, Eqs 11 and 12 are functions of only b and M, and b can be eliminated to yield DP = [ 1 + M ( 1 − CI 2 ) / 2 ] − 1 / 2 (left black line in Fig 6), where we have used the approximation M ≫ 1 (see Materials and methods). Note here that when CI = 0 we have DP ≈ (M/2)−1/2 for large M, which makes sense because for a simple random walk (p = b = 0) the displacement goes like M1/2 while the distance goes like M, such that DP ∼ M−1/2. Finally, the right edge corresponds to the maximal value of p for a given b, for which we compute the approximation curve parametrically (right black line in Fig 6; see Materials and methods). We see in Fig 6 that these approximate expressions slightly underestimate the CI and overestimate the DP, but otherwise capture the crescent shape well. The under- and overestimation are due to the approximation 〈 r 〉 ≈ 〈 r 2 〉 in Eqs 11 and 12: because σ r 2 = 〈 r 2 〉 − 〈 r 〉 2 ≥ 0 for any statistical quantity, we have 〈 r 2 〉 ≥ 〈 r 〉, making Eq 11 an underestimate and Eq 12 an overestimate.
The crescent shape can also be understood intuitively. First, we see that the DP cannot be smaller than a minimum value (region I in Fig 6). This is because the trajectory length M is finite, and as discussed above, the DP only vanishes for infinitely long trajectories. If M were to increase, the crescent would extend further toward DP = 0. Second, we see that the top of the crescent bends away from the CI →1, DP →0 corner (region II in Fig 6). In other words, it is not possible to have high bias without any persistence. This is because if the bias is strong, then cells will track the gradient very well. Consequently, they will move in nearly straight lines in the gradient direction, and straight movement corresponds to high persistence. This is a bias-induced persistence, distinct from the bias-independent persistence in the lower-right corner of the crescent. Finally, we see that the bending shape of the crescent implies that no solutions exist at large DP and intermediate CI (region III in Fig 6). In other words, it is not possible to have high persistence with partial bias. This is because, as mentioned above, persistence is induced either (i) directly, as a result of a large persistence parameter p which is independent of the bias, in which case the CI is low; or (ii) indirectly, as a result of a large bias parameter b, in which case the CI is high. Neither of these mechanisms permits intermediate bias, and therefore high persistence can be accompanied only by low or high directionality. Together, these features of the crescent shape imply that specific modes of chemotaxis are prohibited under our simple model, as indicated by the regions I, II, and III.
Finally, the crescent shape provides a qualitative rationale for the data from the simulations and experiments, which are overlaid in the cyan and red squares in Fig 6, respectively. Specifically, the shape of the crescent is such that if a cell has a low CI and intermediate DP (bottom right corner of the crescent) and its CI increases, its DP must decrease (solid magenta arrow in Fig 6). In contrast, a simultaneous increase in CI and DP from this starting position is not possible according to the model (dashed magenta arrow in Fig 6). We see that the data are qualitatively consistent with this predicted trend, as an increase in the CI corresponds to a decrease in the DP in both the experiments and the simulations (Fig 6, squares). There is quantitative disagreement, in the sense that the data do not quite overlap with the crescent, but this is a reflection of the extreme simplicity of the BPRW model. Nonetheless, the qualitative features of the BPRW model are sufficient to explain the way in which accuracy and persistence are mutually constrained during the chemotaxis response of these cells.
By integrating experiments with theory and simulations, we have investigated mutual constraints on the accuracy (CI), persistence (DP), and speed of cancer cell motion in response to a chemical attractant. We have found that while the CI of breast cancer cells increases with the strength of a TGF-β gradient, the speed does not show a strong trend, and the DP slightly decreases. The simulations suggest that the decrease in DP is due to a decoupling between sensing and persistence in the migration dynamics. The theory confirms that the decrease in DP is due to a mutual constraint on accuracy and persistence for this type of decoupled dynamics, and more generally, it suggests that entire regions of the accuracy–persistence space are prohibited.
The present results provide some insights into TGF-β induced migration mechanisms. Multiple signaling pathways induced by TGF-β affect the dynamics of actin polymerization regulating cell migratory behaviors [27, 43–45]. Among these, phosphatidylinositol 3-kinase (PI3K) and the small GTPase-Rac1 signaling have been reported to promote actin organization of breast cancer cells in response to TGF-β [45, 46]. PI3K and the Rho-family GTPase networks (including Rac1, RhoA and Cdc42) have been widely studied in chemotaxis, which regulates cell polarity and directional sensing [47–50]. The PI3K activity, thus, can possibly explain the present chemotactic responses of the breast cancer cells to TGF-β gradient. Recent studies have shown that PI3K is relevant to the accuracy of the cell movement in shallow chemoattractants, whereas it does not induce the orientation of cell movement in steep gradients; rather, PI3K contributes the motility enhancement [51, 52]. These results can be correlated with the cell motility trend in the present experimental results. In addition, the PI3K signaling pathway has been reported not to mediate the persistence of cell protrusions which could be directly related to the DP [47, 48]. The directional persistence could be more relevant to the polarity stability which is hardly controlled by chemotaxis [47] as presented in the present results. In TGF-β molecular cascades, activation of SMAD proteins could also affect the actin dynamics. Since SMAD-cascades include negative feedback inhibiting Rho activity [43, 44], it may affect the cell responses highly promoted in CI but not in speed. However, the underlying molecular mechanisms need further research.
Our finding that sensing and persistence are largely decoupled in the migration dynamics is related to the view that directional sensing and polarity are separate but connected modules in chemotaxis [11]. Indeed, CI, DP, and speed in our study play the roles of the directional sensing, polarity, and motility modules, respectively, that have been shown to reproduce many of the observed behaviors of chemotaxing cells. Moreover, several of the the molecular signaling pathways discussed above, including those involving PI3K and Rho family GTPases, have been proposed as the potential networks corresponding to these modules [11].
Several predictions arise from our work that would be interesting to test in future experiments. First, our simulation scheme assumes that the saturation of the CI with gradient strength (Fig 4A) is due to limited receptor numbers. However, alternative explanations exist that are independent of the receptors, such as the fact that it is more difficult to detect a concentration difference on top of a large concentration background than on top of a small concentration background due to intrinsic fluctuations in molecule number [30, 53]. An interesting consequence of our mechanism of receptor saturation is that, at very large gradients (beyond those of Fig 4A), the CI would actually decrease because all receptors would be bound. It would be interesting to test this prediction in future experiments.
Second, our work suggests that not all quadrants of the accuracy–persistence plane are possible for cells to achieve (Fig 6). It would be interesting to measure the CI and DP of other cell types, in other chemical or mechanical environments, to see if the crescent shape seen in Fig 6 is a universal restriction, or if not, what new features of chemotaxis are therefore not captured by the modeling. In this respect, the work here can be seen as a null model, deviations from which would indicate new and unique types of cell motion.
Human breast adenocarcinoma cells (MDA-MB-231) were cultured in Dulbecco’s Modified Eagle Medium/Ham’s F-12 (Advanced DMEM/F-12, Lifetechnologies, CA, USA) supplemented by 5% v/v fetal bovin serum (FBS), 2 mM L-glutamine (L-glu), and 100 μg ml-1 penicillin/streptomycin(P/S) for less than 15 passages. MDA-MB-231 cells were regularly harvested by 0.05% trypsin and 0.53mM EDTA (Lifetechnologies, CA, USA) when grown up to around 80% confluency in 75 cm2 T-flasks at 37 °C with 5% CO2 incubation. Harvested cells were used for experiments or sub-cultured.
Cell-matrix composition was prepared in the microfluidic device. For the composition, MDA-MB-231 cells were mixed with 2 mg/ml of type I collagen (Corning Inc., NY, USA) mixture prepared with 10X PBS, NaOH, HEPE solution, FBS, Glu, P/S, and cell-culture level distilled water after centrifuged with 1000 rpm for 3 minutes. The cell mixture was filled in center-channel of the microfluidic devices and incubated in at 37 °C with 5% CO2. The cells in the collagen matrix were initially cultured in basic medium (DMEM/F12 supplemented by 5% v/v FBS, 2 mM L-glu, and 100 μg ml−1 p/s) for 24 hours. Then the cells were exposed by reduced serum medium for another 24 hours, which was advanced DMEM/F12 containing 1% v/v FBS, 2 mM L-glu, and 100 μg ml−1 p/s [54]. After 24 hour-serum starvations, cells were exposed by a gradient of transforming growth factor beta-1 (TGF-β1, Invitrogen, CA, USA).
The microfluidic device was designed to generate a linear gradient of soluble factors (Fig 2). The device is composed of three channels which are 100 μm in thickness as described previously [55]. A center channel that is 1 mm wide aims to culture tumor cells with ECM components. The center channel is connected to two side channels. The 300 μm-wide side channels are connected to large reservoirs at the end ports including culture medium. Since the side channels are in contact with the top and bottom sides of the center channel, the growth factor gradient can be generated by diffusing the soluble factor from one of the side channels, a source channel, to the other, a sink channel. Assuming there is neither pressure difference nor flow between the side channels, the concentration of a given factor can be described by the chemical species conservation equation as follows:
∂ c i ∂ t = D i · ∇ c i (13)
Once the concentration profile in the center channel reaches steady state, the linear profile persists for a while and can therefore be approximated by assuming the boundary conditions of concentration at the side channels are constants. To verify the diffusion behavior, the gradient formation was examined by using 10k Da FITC-fluorescence conjugated dextran (FITC-dextran). FITC-dextran solution was applied in the source channel while the sink channel was filled with normal culture medium. The FITC-dextran concentration profile was evaluated by the FITC fluorescent intensity in the center channel. To disregard the effect of photo-bleaching on the results, the intensity was normalized by the intensity of the source channel. The normalized intensity was reasonably considered since the fluorescence intensity of the source channel consistently remained as maximum due to the large reservoirs. The FITC dextran intensity profile (Fig 2C) showed that the linear profile was developed within 3 hours after applying the source and continued for more than 9 hours.
Cell behaviors were captured every 15 minutes for 9 hours using an inverted microscope (Olympus IX71, Japan) equipped with a stage top incubator as described previously [56–58], so that the microfluidic platform could be maintained at 37 °C in a 5% CO2 environment during imaging. The time-lapse imaging was started 3 hours after applying TGF-β1 solution in the source channel to have sufficient adjusting time. To analyze each cell behavior, a cell area in the bright field images were defined by a contrast difference between the cells and a background, and the images were converted to monochrome images by using ImageJ. Cell trajectories were demonstrated by tracking centroids of the cell area. In tracking the cell movements, cells undergoing division were excluded to avoid extra influences to affect cell polarity [59]. Moreover, stationary cells due to the presence of the matrix were excluded [26, 59–61]. The stationary cells were defined as the cells that moved less than their diameter. A migration trajectory was defined by connecting the centroids of a cell from each time point.
In examining the chemotactic characteristics of each group, more than 40 cell trajectories were evaluated per a group. A data point in Fig 3C–3E indicates each metric of a cell trajectory showing distribution characteristics with a box plot. The box plot includes boundaries as quadrants and a center as a median. The distribution of each metric was statistically analyzed by using Mann-Whitney U-test. This non-parametric method was used since the distribution was not consistently normal (the CI is a function of cosine). The significant change on the population lies on the biased distribution of each cell parameter when the p value < 0.05. Furthermore, the experiments were repeated at least 3 times and reported with means of medians ± standard estimated error (S.E.M.) in Fig 4A–4C. To evaluate physical limits on each metric, the data points were compared each other using a student t-test. The statistical significance between comparisons were examined when the p value < 0.05.
In the limit p = 0, Eqs 9 and 10 become
⟨ r 2 ⟩ = z 2 M 2 + 2 ( 1 − 2 z 2 ) M + 2 ( 3 z 2 − 1 ) , (14)
⟨ x ⟩ 2 = z 2 ( M − 1 ) 2 , (15)
where z = πb, and we have neglected the exponential terms in the limit M ≫ 1. Defining the small parameter ϵ = 1/M, these expressions become
⟨ r 2 ⟩ = z 2 M 2 ( 1 + c ϵ ) , (16)
⟨ x ⟩ 2 = z 2 M 2 ( 1 − 2 ϵ ) (17)
to first order in ϵ, where c ≡ 2(z−2 − 2). Inserting these expressions into Eqs 11 and 12, we obtain
CI 2 = 1 − ( c + 2 ) ϵ , (18)
DP 2 = z 2 ( 1 + c ϵ ) (19)
to first order in ϵ. Because z and c are both functions only of b, we eliminate b from Eqs 18 and 19 to obtain
CI 2 = 1 − 2 ϵ 1 − DP 2 DP 2 (20)
to first order in ϵ. This expression is equivalent to that given below Eq 12 and provides the left black line in Fig 6.
The right black line in Fig 6 corresponds to the maximal value of p for a given b that keeps Eq 8 non-negative. Non-negativity requires that the sum of the minimal values of each term in Eq 8 is zero: −b + e−p/[2πI0(p)] = 0. With this expression for b in terms of p, Eqs 11 and 12 become functions of only p and M. Therefore, by varying p, we compute the right black line parametrically.
|
10.1371/journal.ppat.0040031 | Salp15 Binding to DC-SIGN Inhibits Cytokine Expression by Impairing both Nucleosome Remodeling and mRNA Stabilization | Ixodes ticks are major vectors for human pathogens, such as Borrelia burgdorferi, the causative agent of Lyme disease. Tick saliva contains immunosuppressive molecules that facilitate tick feeding and B. burgdorferi infection. We here demonstrate, to our knowledge for the first time, that the Ixodes scapularis salivary protein Salp15 inhibits adaptive immune responses by suppressing human dendritic cell (DC) functions. Salp15 inhibits both Toll-like receptor- and B. burgdorferi–induced production of pro-inflammatory cytokines by DCs and DC-induced T cell activation. Salp15 interacts with DC-SIGN on DCs, which results in activation of the serine/threonine kinase Raf-1. Strikingly, Raf-1 activation by Salp15 leads to mitogen-activated protein kinase kinase (MEK)-dependent decrease of IL-6 and TNF-α mRNA stability and impaired nucleosome remodeling at the IL-12p35 promoter. These data demonstrate that Salp15 binding to DC-SIGN triggers a novel Raf-1/MEK-dependent signaling pathway acting at both cytokine transcriptional and post-transcriptional level to modulate Toll-like receptor–induced DC activation, which might be instrumental to tick feeding and B. burgdorferi infection, and an important factor in the pathogenesis of Lyme disease. Insight into the molecular mechanism of immunosuppression by tick salivary proteins might provide innovative strategies to combat Lyme disease and could lead to the development of novel anti-inflammatory or immunosuppressive agents.
| Upon attachment of the tick, the host elicits both innate and adaptive immune responses directed against the vector. In turn, ticks have developed countermeasures to withstand and evade host immune responses. In the current paper we demonstrate how a tick salivary protein induces immunosuppression of human dendritic cells and how this could facilitate infection with B. burgdorferi, the causative agent of Lyme disease. Insight into the molecular mechanism of immunosuppression by tick salivary proteins might provide innovative strategies to combat Lyme disease or other tick-borne illnesses and could lead to the development of novel anti-inflammatory or immunosuppressive drugs.
| Ixodes ticks are a major arthropod vector for human pathogens, such as Borrelia burgdorferi, the causative agent of Lyme disease [1]. Ixodes ticks require five to seven days to feed to repletion [2]. In order to secure attachment of the vector and to ensure susceptibility of reservoir hosts for future tick infestations, tick saliva contains modulators of host immune responses. Salp15, a 15-kDa salivary gland protein, is a major immunomodulatory protein in I. scapularis saliva [3]. Salp15 has been shown to bind to CD4, thereby inhibiting T cell receptor (TCR) ligation-induced signals, resulting in impaired interleukin (IL)-2 production and impaired CD4+ T cell activation and proliferation [4–6]. While feeding on a host, ticks can introduce B. burgdorferi into the host's skin. Local immunosuppression of the host by tick molecules assists B. burgdorferi in establishing an infection. In addition, it has been shown that Salp15 binds to B. burgdorferi outer surface protein (Osp) C [7]. B. burgdorferi expresses OspC in the tick salivary glands and during the early stages of mammalian infection. Binding of Salp15 to OspC protects the spirochete from antibody-mediated killing by the immune host [7], and silencing of Salp15 by RNA interference in I. scapularis ticks resulted in a dramatically impaired ability to transmit B. burgdorferi to an immune host [7]. Thus, Salp15 is an important immunomodulatory protein in I. scapularis saliva that targets the T cell arm of adaptive immunity.
Dendritic cells (DCs) are essential in initiating adaptive immune responses in naive hosts [8]. After sensing invading pathogens in peripheral tissues, DCs capture them for processing and presentation to activate T cells in draining lymph nodes [8]. Previously we have shown that Salp15 is secreted by the feeding tick and is locally introduced in the host skin [4], where Salp15 also provides B. burgdorferi a survival advantage in a naive murine host, but only when co-injected, ruling out a systemic immunosuppressive effect of Salp15 [7]. However, local inhibition of immune responses by Salp15 could be responsible for the observed effect. Under normal circumstances there are very few T lymphocytes present at the site of the tick-bite, whereas DCs are abundantly present. Therefore, we hypothesized that DCs are a major target for immunomodulation by Salp15, since these cells are essential in initiating adaptive immune responses to exposed tick (salivary gland) antigens and B. burgdorferi in a naive host.
Here we have investigated the interaction of the major immunomodulatory protein in Ixodes scapularis saliva, Salp15, with human DCs. Salp15 inhibits the production of the pro-inflammatory cytokines IL-12p70, IL-6, and TNF-α of DCs stimulated with the Toll-like receptor (TLR)-2 and −4 ligands, LTA and LPS, respectively. Salp15 interacts with the C-type lectin DC-SIGN, which results in activation of the kinases Raf-1 and mitogen-activated protein kinase kinase (MEK). This leads to the inhibition of pro-inflammatory cytokine production and suppresses the T cell–stimulatory role of DCs. Strikingly, the Salp15/DC-SIGN-induced signaling pathway regulates the inhibition of pro-inflammatory cytokines at different levels: decreased nucleosome remodeling at the IL-12p35 promoter impairs IL-12p70 production, whereas the inhibition of IL-6 and TNF-α is caused by an increased decay of their respective mRNAs. A similar suppression of pro-inflammatory cytokines is observed when DCs are activated with viable B. burgdorferi in the presence of Salp15, indicating that the spirochete uses Salp15 to induce immune suppression. Thus, local interaction of Salp15 and DCs will lead to immunosuppression, which potentially allows the tick to feed for a longer period of time, and B. burgdorferi to escape from human immune responses, and might therefore be an important factor in the pathogenesis of Lyme disease.
To investigate the effect of Salp15 on human DC function, we incubated immature DCs with different concentrations of recombinant Salp15 for 18 h and analyzed DC maturation and cytokine production. Salp15 alone did not induce upregulation of the maturation markers CD80, CD83, or CD86 (Figure 1A). In addition, DCs incubated with Salp15 did not secrete detectable levels of IL-10, IL-6, IL-8, IL-12p70, or TNF-α (Figure 1B). Thus, Salp15 by itself does not affect activation of DCs.
Next, we investigated the effect of Salp15 on DC maturation and activation induced by the TLR-4 ligand LPS. DCs incubated with LPS in combination with Salp15 showed an upregulation of the maturation markers CD80, CD83, and CD86 similar to DCs incubated with LPS alone (Figure 1A). LPS induced the secretion of the cytokines IL-6, IL-8, IL-10, IL-12p70, and TNF-α by DCs. Strikingly, production of the pro-inflammatory cytokines IL-6, IL-12p70, and TNF-α was inhibited by Salp15 in a dose-dependent manner (Figure 1B). This effect was observed at both protein and mRNA level (Figure 1B and 1C). IL-8 levels were increased at both the protein and mRNA level, whereas IL-10 production was only affected at the mRNA level (Figure 1B and 1C). The inhibition of pro-inflammatory cytokine production was not due to cytotoxicity of Salp15 as determined by 7-AAD staining (unpublished data). DCs incubated with the TLR-2 ligand LTA showed a similar downregulation of pro-inflammatory cytokines in the presence of Salp15 (Figure S1). Thus, Salp15 suppresses TLR-induced cytokine responses by DCs.
Whole tick saliva obtained from Ixodes ricinus did not induce maturation by itself or cytokine production, but there was a decrease in pro-inflammatory cytokine production when saliva was added to DCs in combination with LPS similar to Salp15 (Figure 1A and 1D). These data indicate that Salp15 in Ixodes saliva is responsible, although not necessarily exclusively, for reducing pro-inflammatory cytokine production by DCs. The increase in IL-8 caused by recombinant Salp15 was not observed as clearly in tick saliva, indicating that other tick molecules could hamper IL-8 production by human DCs. Similar to Salp15, DCs stimulated with LPS in the presence of tick saliva did not have an increased production of the cytokine IL-10 compared to DCs stimulated with LPS alone (Figure 1D).
To further investigate the interaction of Salp15 and DCs, we investigated the binding of Salp15 to DCs. Salp15 interacts with DCs, and this interaction could be blocked to background levels by pre-incubating DCs with the calcium and magnesium-chelator EDTA, and the mannose-specific C-type lectin inhibitor mannan (Figure 2A). This strongly indicates that C-type lectins are involved in binding of Salp15 to DCs. Analysis of carbohydrate structures on Salp15 demonstrated that recombinant Drosophila-expressed Salp15 contains mannose and galactose structures, since the mannose- and galactose-specific plant lectins Con A and Peanut agglutin (PNA) strongly bound to Salp15 (Figure 2B). Similar to Ixodes ticks, Drosophila belongs to the phylum of Arthropoda, suggesting that Salp15 glycosylation is similar to native Salp15 in tick saliva.
Salp15 binding to DCs could be blocked with mannan, N-acetyl-D-glucosamine (GlcNac), Lewis X (LeX), Lewis Y (LeY), and Fucose (Figure 2C). These carbohydrates all have high affinity for the C-type lectin receptor DC-SIGN, suggesting that DC-SIGN mediates binding of Salp15 to DCs. To investigate the cellular binding of Salp15 to DC-SIGN we used a Raji cell line transfected with DC-SIGN and demonstrated that Salp15 binds to DC-SIGN-transfected cells, but not to mock-transfected cells (Figure 2D). The interaction between Raji cells expressing DC-SIGN and Salp15 could be blocked with mannan, similar to the interaction of Salp15 to DCs. HIV-1 gp120 binding to DC-SIGN seems stronger than Salp15, although this could reflect differences in coating of the fluorescent beads. DC-SIGN binding to Salp15 was further assessed in a DC-SIGN-Fc binding ELISA. DC-SIGN-Fc interacted with plate-bound Salp15 and binding could be blocked with EDTA and mannan, suggesting that the binding is specific for DC-SIGN (Figure 2E). Furthermore, deglycosylation of Salp15 by N-glycosidase F (PGNaseF) abrogated the binding to DC-SIGN-Fc, demonstrating that the carbohydrate structures on Salp15 are essential to the interaction with DC-SIGN (Figure 2E). Due to interaction of murine anti-DC-SIGN blocking antibodies with goat-anti-mouse IgG antibodies used to generate the fluorescent Salp15-beads, we were unable to investigate whether these anti-DC-SIGN antibodies were able to block Salp15 binding to DC-SIGN in our bindings assays. Instead, we used Salp15-coupled Prot-A-Sepharose beads to immunoprecipitate the receptor for Salp15 from a cell-surface biotinylated DC lysate. The immunoprecipitate was analyzed by SDS-PAGE, transferred to a nitrocellulose membrane and the biotinylated cell-surface proteins were visualized by streptavidin-PO. The Salp15 immunoprecipitate showed only one distinct band with an apparent molecular weight of 45–50 kDa (Figure 2F). The size of the immunoprecipitate corresponded to the size of DC-SIGN, which was immunoprecipitated with beads coupled to anti-DC-SIGN antibodies. To confirm that the Salp15-coupled beads had selectively pulled down DC-SIGN, we also immunoprecipitated the receptor for Salp15 from non-biotinylated DC lysate. The immunoprecipitate was analyzed by SDS-PAGE, transferred to a nitrocellulose membrane, and visualized with antibodies against DC-SIGN showing a band at 45–50 kDa (Figure 2F) confirming that Salp15 binds to DC-SIGN on DCs.
Next, we investigated the intracellular signaling pathways that Salp15 induces to inhibit the expression of pro-inflammatory cytokines by DCs. Recently, we have demonstrated that binding of mycobacterial ManLAM to DC-SIGN leads to activation of the serine/threonine kinase Raf-1 [9]. We investigated whether binding of Salp15 to DCs also activates Raf-1 by assessing phosphorylation of the Serine (S)338 and Tyrosine (Y)340/341 phosphorylation sites of Raf-1. Indeed, Salp15 induced phosphorylation of both sites (Figure 3A), demonstrating that Salp15 activates Raf-1 similar to ManLAM. Phosphorylation was partially blocked by blocking antibodies against DC-SIGN (Figure 3A).
Next, we investigated whether Raf-1 silencing through RNA interference could prevent the Salp15-induced inhibition of DC cytokine production. Immature DCs were transfected with siRNA specifically targeting Raf-1, while expression of the closely related kinase B-Raf remained unaltered as determined by quantitative real-time PCR (Figure 3B). Raf-1 silencing at the protein level was complete since Raf-1 protein was not detected in silenced DCs as determined by immunofluorescence analysis (unpublished data). Silencing of Raf-1 in DCs completely abrogated the Salp15-induced inhibition of cytokines, since stimulation of Raf-1 silenced DCs with LPS in the presence of Salp15 restored IL-12p35, IL-6, TNF-α, IL-8, and IL-10 levels to those observed in DCs stimulated with LPS alone (Figures 3C and S2A). To confirm the Raf-1 silencing data, we also used the Raf inhibitor GW5074 and observed a similar restoration of pro-inflammatory cytokine levels in DCs treated with the Raf inhibitor in the presence of LPS and Salp15 (Figures 3D and S2B). IL-8 enhancement by Salp15 was abrogated by GW5704 mostly at the mRNA level, whereas IL-10 was not increased by Salp15 (Figure S2B). These data demonstrate that Raf-1 is essential for the inhibition of pro-inflammatory cytokine production by Salp15.
Next, we investigated the downstream effectors of Raf-1. Raf-1 activation by ligation of ManLAM to DC-SIGN leads to phosphorylation of the NF-kB subunit p65 at residue S276, which subsequently results in acetylation of p65 by the histone acetyltransferases (HATs) p300 and CREB binding protein (CBP) [9]. Acetylation of p65 leads to an increased IL-10 cytokine response by DCs, which is inhibited by the CBP/p300-inhibitor anacardic acid [9]. Strikingly, anacardic acid (AA) did not prevent Salp15-induced modulation of IL-12p35, IL-6, TNF-α, IL-8, and IL-10 transcription (Figures 3E and S2C), whereas it did abrogate ManLAM-induced DC cytokine production (unpublished data, [9]), indicating that Salp15 induces downstream signaling cascades different from ManLAM-induced DC-SIGN signaling. Raf kinases are well known for their ability to activate MEK 1 and 2, that subsequently activate extracellular signal-regulated kinase (ERK) 1 and 2 [10]. Therefore, we investigated whether the MEK1/2 inhibitor U0126 could block Salp15-induced inhibition of DC cytokine production. Inhibition of MEK completely blocked the Salp15-induced modulation of cytokine production (Figures 3E and S2C). Strikingly, despite a role for MEK kinases in Salp15-induced signaling, stimulation of immature DCs with Salp15 did not lead to ERK activation as assessed by the phosphorylation of ERK (Figure 3F and 3G), which is a prerequisite for ERK activation [11]. Activation was followed over time but no phosphorylation of ERK in the presence of Salp15 could be detected (Figure 3G). In addition, ERK activation by LPS was not altered in the presence of Salp15, demonstrating that Salp15 does not sequester MEK away from ERK and thereby inhibiting LPS-induced cytokine production. Thus, Salp15 binding to DC-SIGN modulates DC function through a Raf-1- and MEK-dependent, but ERK-independent pathway that is distinct from the recently identified ManLAM/DC-SIGN-induced signaling pathway that results in Raf-1-dependent but MEK-independent phosphorylation and acetylation of p65 [9].
Since we demonstrated that phosphorylation and acetylation of p65 was not responsible for the observed Salp15-induced modulation of cytokine responses, we set out to identify the mechanisms responsible for the Salp15-induced Raf-1/MEK-dependent decrease in TLR-induced pro-inflammatory cytokines. Gene expression is regulated by complex mechanisms at many stages, including chromatin accessibility, transcription activation, mRNA nuclear export, mRNA decay, and translation. Many cytokine genes are subject to regulation at the level of mRNA decay through the presence of mRNA-stability elements, particularly AU-rich elements, within their 3′ untranslated regions, as it provides the means to use transcripts with optimal efficiency and to respond rapidly to cellular signals [12,13]. Therefore we determined the effect of Salp15 on mRNA stability. DCs were stimulated with LPS and Salp15 in the presence or absence of Raf and MEK inhibitors. After 6 h, actinomycin D was added to block mRNA synthesis, and cells were harvested every 20 min for 2 h and mRNA decay was determined. Both TNF-α and IL-6 mRNA showed a strongly decreased half-life in the presence of Salp15 and LPS compared to LPS alone (Figure 4A). IL-6 mRNA half-life was completely restored by the Raf and MEK inhibitors, GW5074 and U0126, respectively. Inhibition of Raf-1 by GW5074 also completely abrogated the effects of Salp15 on TNF-α mRNA half-life. We could not assess whether the MEK inhibitor abrogated the effect of Salp15 on TNF-α mRNA half-life, since inhibition of MEK has been shown to affect nucleocytoplasmic transport of TNF-α mRNA [14]. Strikingly, IL-12p35 mRNA stability remained unchanged in LPS-activated DCs in the presence of Salp15 compared to DCs stimulated with LPS alone, indicating that the decrease in IL-12 is not regulated at the level of mRNA stability. In addition, there was no change in IL-8 and IL-10 mRNA stability in the presence of Salp15 (Figure S3).
Previously, it has been described that regulation at the level of chromatin accessibility is an important factor in the regulation of IL-12p35 transcription. IL-12p35 gene activation during DC maturation involves selective and rapid remodeling of nucleosome 2 in the IL-12p35 promoter region [15]. To investigate the effect of Salp15 on nucleosome remodeling at the IL-12p35 promoter we used a Chromatin accessibility by real time PCR (ChART) assay [16]. In the presence of LPS, rapid nucleosome remodeling at the IL-12p35 promoter occurs in DCs, allowing for efficient transcription initiation (Figure 4B). Strikingly, Salp15 severely impaired this nucleosome remodeling, whereas both Raf and MEK inhibitors restored the level of remodeling to that observed with LPS alone (Figure 4B). Thus, Salp15 impairs nucleosome remodeling at the IL-12p35 promoter through a Raf-1/MEK-dependent signaling pathway, which results in decreased IL-12p35 mRNA expression in DCs.
Therefore, Salp15-DC-SIGN signaling regulates cytokine production at both transcriptional and post-transcriptional levels; the decreased IL-12p70 production in the presence of Salp15 and LPS is dependent on nucleosome remodeling at the IL-12p35 promoter, while the downregulation of IL-6 and TNF-α is a result of increased mRNA decay.
T lymphocyte activation by mature DCs is essential to initiate an effective adaptive immune response against invading pathogens. Therefore we investigated whether Salp15 interfered with the T lymphocyte activation capacity of DCs by performing a mixed leukocyte reaction (MLR). DCs were incubated with Salp15 and LPS for 18 h. The DCs were washed extensively to remove remaining Salp15 before allogenic lymphocytes were added, because it has been shown that Salp15 can directly inhibit T cell activation [4,5]. DCs pretreated with Salp15 alone did not suppress lymphocyte proliferation. However, LPS-matured DCs pre-incubated with Salp15 or tick saliva were less capable of inducing lymphocyte proliferation compared to DCs matured with LPS alone (Figure 5A and 5B). These data demonstrate that Salp15 in tick saliva does not only alter cytokine responses of DCs, but also inhibits T cell proliferation induced by DCs.
A recent study demonstrated that B. burgforferi alone does not suppress DC functions [17]. Since B. burgdorferi has been shown to interact with Salp15 to evade host antibody-mediated killing [7], we investigated whether B. burgdorferi could also benefit from the inhibition of DC cytokine production by Salp15. Therefore, we determined DC cytokine production by stimulating DCs with viable B. burgdorferi alone or in the presence of Salp15. B. burgdorferi alone induced DC maturation and pro-inflammatory cytokine secretion, as has been shown before [17]. Preincubation with Salp15 did not affect the upregulation of DC maturation markers by B. burgdorferi (Figure 6A). However, Salp15 inhibited the pro-inflammatory cytokine production by B. burgdorferi-activated DCs, similar to DCs activated by LPS in the presence of Salp15 (Figure 6B). B. burgdorferi, in contrast to LPS, did induce IL-1β production by human DCs (Figure 6B). Similar to the other pro-inflammatory cytokines IL-1β was inhibited by Salp15. In addition, compared to B. burgdorferi alone, B. burgdorferi preincubated with Salp15 induced enhanced production of the immunomodulatory cytokine IL-10 by DCs; this was similar to what was observed for LTA (Figure S1), underscoring the fact that the set of cytokines produced by human DCs is dependent on the TLR-ligand. Thus, these data demonstrate that the interaction of Salp15 with B. burgdorferi assists B. burgdorferi in altering and potentially evading host adaptive immune responses during early human infection.
The main vectors for Borrelia burgdorferi, the causative agent of Lyme borreliosis, are ticks belonging to the Ixodes genus. In Europe, B. burgdorferi is transmitted by I. ricinus, whereas in the United States B. burgdorferi is predominantly transmitted by I. scapularis. Upon attachment of the tick, the host elicits both innate and adaptive immune responses directed against the vector. In turn, ticks have developed countermeasures to withstand and evade host immune responses. Therefore, tick saliva contains anticoagulant, vasodilatory, and immuno-suppressive molecules and is abundantly secreted into the host's skin during tick feeding. The balance between host immune responses and the vector's immunosuppressive countermeasures determines the duration of tick attachment, the degree of engorgement of the tick, but also the extent of B. burgdorferi transmission [18–20].
Salp15 is a protein in I. scapularis saliva that is induced during feeding and it is abundantly secreted into the host. I. ricinus contains a Salp15 homologue, Salp15 Iric-1 (GenBank accession number: ABU93613) with 82% similarity to Salp15 from I. scapularis [21]. Previously, we have shown that native Salp15 can be readily detected in host skin at the site of the tick-bite [4]. Salp15 has been shown to inhibit CD4+ T cell activation [4], and to interfere with T cell receptor signaling [6] through binding to CD4 [5], which results in the inhibition of IL-2 production upon recognition of cognate antigens. However, low numbers of T cells are present in the dermis of the skin, the site where Salp15 is secreted during tick feeding. In contrast, DCs, as sentinels and initiators of adaptive immune responses, are abundantly localized in the skin and might therefore be targeted by the tick to suppress the initiation of adaptive immune responses. Indeed, we demonstrate that I. scapularis Salp15 interacts with human DCs and that Salp15 impairs TLR-induced pro-inflammatory cytokine production by DCs and suppresses DC-induced T lymphocyte activation. DCs recognize antigens by TLRs and C-type lectins. DC-SIGN is a C-type lectin abundantly expressed by DCs. It is becoming clear that DC-SIGN is involved in pathogen recognition, but that several pathogens target this receptor to modulate DC functions [22]. We here demonstrate that Salp15 targets the C-type lectin DC-SIGN on DCs to activate a Raf-1/MEK-dependent signaling cascade that results in downregulation of pro-inflammatory cytokines induced not only by TLR-2 and TLR-4 ligands, but also by viable B. burgdorferi. This downregulation is a result of decreased IL-6 and TNF-α mRNA stability and impaired nucleosome remodeling at the IL-12p35 promoter. Moreover, our data demonstrate that Salp15 and tick saliva inhibit T cell activation. Therefore, B. burgdorferi could use Salp15 in saliva to suppress DC-mediated cytokine production, which might assist the spirochete in establishing an infection of the host.
I. scapularis Salp15, similar to tick saliva, inhibited the production of the pro-inflammatory cytokines IL-12p70, IL-6, and TNF-α by LPS-activated DCs. These data indicate that Salp15 present in tick saliva induced the observed cytokine suppression. Indeed, anti-Salp15 antibodies prevented the suppression of IL-12 by tick saliva (unpublished data). This is also supported by the Salp15 concentrations used in this study, since approximately 0.1% of tick saliva consists of Salp15 which equals a concentration of 1 μg/ml [4] and the concentration of Salp15 increases over time during tick feeding, reaching high concentrations locally. Our observations are in line with data from Cassavani et al. showing that saliva from the Rhipicephalus sanguineus tick inhibits IL-12 production by murine bone marrow–derived DCs [23]. Also, very recently, I. scapularis saliva was shown to dose-dependently inhibit IL-12 and TNF-α production by murine bone marrow–derived DCs [24]. Although the authors contribute this effect to Prostaglandin E2 (PGE2), they state that other DC modulators might be present in I. scapularis saliva. Importantly, both studies were performed with murine DCs, whereas our experiments are performed with human DCs. The fact that the addition of rabbit polyclonal anti-Salp15-antibodies abrogates the capacity of Ixodes saliva to inhibit IL-12, but not IL-6 and TNF-α, might indicate that saliva contains other molecules, such as PGE2, that are able to block IL-6 and TNF-α (unpublished data). However, silencing of Raf-1, and inhibitors of Raf and MEK restored Salp15-induced suppression of IL-12p35, IL-6, and TNF-α, suggesting that the anti-Salp15 antibodies are less effective than small molecule inhibitors or RNAi treatment. Possibly due to genetic variability [25], DCs derived from approximately half the donors failed to produce IL-10 upon stimulation with LPS. Moreover, the IL-10 production might be dependent on the TLR ligand used to activate DCs, since Salp15 did enhance IL-10 production by LTA- and B. burgdorferi-treated DCs (Figures 6 and S1). Neutralization of IL-10 using antibodies did not restore IL-12 levels, demonstrating that inhibition of pro-inflammatory cytokines is not due to IL-10 production (unpublished data).
B. burgdorferi by itself does not modulate DC function, since human DCs have been shown to react adequately to B. burgdorferi [17]. However, similar to the observed effects of Salp15 on LPS-stimulated DCs, preincubation of B. burgdorferi with Salp15 resulted in inhibited pro-inflammatory cytokine production (Figure 6). Immunosuppression of the host by Salp15 could be advantageous to the arthropod vector and the spirochete, since it could impair adaptive immune responses to both tick as well as B. burgdorferi antigens and could therefore be an important factor in the pathogenesis of Lyme diseases. Interestingly, Ramamoorthi et al. have shown that Salp15 mRNA levels were 13-fold higher in B. burgdorferi-infected engorged ticks than in uninfected controls [7]. This symbiosis might be the result of millions of years of co-evolution of the spirochete and the arthropod vector.
Pro-inflammatory cytokines, such as IL-12, are essential to T cell activation as well as differentiation [26], and indeed Salp15, as well as tick saliva, inhibits DC-mediated T lymphocyte activation (Figure 5). This inhibition was not due to a direct effect of Salp15 on lymphocyte proliferation, since, besides extensive washing of the DCs before addition of lymphocytes, simultaneous addition of lymphocytes and Salp15 to DCs did not result in impaired lymphocyte proliferation (unpublished data).
The inhibition of DC cytokine production by Salp15 we describe here and the previously described direct inhibition of T cell activation through binding to CD4 [4,5] could be complementary to each other. Salp15 could inhibit T cell activation by directly binding to CD4 on T cells entering the tick-host-pathogen interface. In addition, inhibition of cytokine production by DCs by Salp15 could inhibit these DCs from adequately activating T cells in regional lymph nodes or at the site of the feeding lesion. Both approaches will lead to decreased numbers of effector T cells and impaired adaptive immune responses. In contrast to T cells, DCs express only low levels of CD4 [27] suggesting that CD4 is not a major receptor for Salp15 on human DCs. Indeed, only DC-SIGN was immunoprecipitated by Salp15 from immature DC lysates (Figure 2), supporting our findings that DC-SIGN plays an important role in the observed immunosuppression. However, we can not exclude a low affinity/avidity interaction between Salp15 and another receptor on DCs, possibly CD4.
Several pathogens and parasites have been demonstrated to target DC-SIGN on DCs to modulate DC-mediated immune responses [22,28] leading to evasion of host immune responses and prolonged pathogen survival. Blocking antibodies against DC-SIGN were not effective in restoring cytokine responses inhibited by Salp15 (unpublished data) suggesting that even a decreased binding of Salp15 to DCs is sufficient to inhibit cytokine responses. Recently, we have demonstrated that binding of various pathogens such as Mycobacterium tuberculosis to DC-SIGN leads to activation of the serine/threonine kinase Raf-1 [9]. To further identify the mechanism by which Salp15 inhibits DC cytokine production, we also assessed the ability of Salp15 to activate Raf-1. Similar to the mycobacterial component ManLAM, Salp15 activates Raf-1 by inducing phosphorylation of Raf-1 at the residues S338 and Y340/341, which is partially blocked by antibodies against DC-SIGN (Figure 3), supporting an important role for DC-SIGN in Raf-1 activation. Furthermore, silencing of Raf-1 in human primary DCs by RNA interference completely restored IL-12p35, IL-6, and TNF-α transcription (Figure 3) and protein expression (unpublished data) after activation by LPS in the presence of Salp15. In addition, cytokine levels could be completely restored by blocking Raf-1 activation during stimulation of DCs with Salp15 and LPS using a Raf inhibitor (Figure 3). Thus, DC-SIGN-induced Raf-1 activation is essential to the observed immunosuppression by Salp15. Recently, we have demonstrated that Raf-1 activation by the mycobacterial component ManLAM leads to acetylation of the NF-κB subunit p65, but only after TLR-induced activation of NF-κB. Acetylation of p65 both prolonged and increased IL-10 transcription to enhance anti-inflammatory cytokine responses [9]. Strikingly, our data demonstrate that acetylation of the NF-κB subunit p65 is not involved in the suppression of pro-inflammatory cytokines by Salp15, since an inhibitor of the HATs p300/CBP, responsible for p65 acetylation, did not abrogate pro-inflammatory cytokine suppression by Salp15 (Figure 3). Moreover, we did not see any effect on IL-8 and L-10 cytokine levels (Figure S2C). These data indicate that Raf-1 activation by Salp15 leads to activation of different downstream effectors than ManLAM [9]. Indeed, Salp15-induced inhibition of cytokine production is completely blocked by the MEK1/2 inhibitor U0126, demonstrating that MEK kinases are essential downstream effectors of Raf-1 after DC-SIGN/Salp15 ligation, in sharp contrast to DC-SIGN/ManLAM ligation [9]. Raf kinases are known to activate MEK1 and MEK2 through phosphorylation of two serine residues [29]. MEK kinases are well-known to signal through ERK kinases and there is evidence that suggests that antibody ligation of DC-SIGN leads to ERK activation [30]. However, DC-SIGN ligation by its pathogen-derived ligands such as ManLAM and Salp15 does not result in ERK activation (Figure 3 and [9]), demonstrating that ERK is not involved in pathogen-induced DC-SIGN signaling. The mechanism of MEK-dependent and ERK-independent signaling induced by Salp15 binding to DC-SIGN might be dependent on the subcellular localization of the kinases [31–33]. Salp15 might interact with low avidity to other receptors on DCs such as CD4, which could result in an altered signaling cascade. A recent study has demonstrated that Salp15 binding to CD4 modulates actin polymerization [6]. This observation implies that co-ligation of DC-SIGN and CD4 by Salp15 might affect the subcellular localization of Raf-1 or its downstream effectors. This would result in the activation of different kinases compared to DC-SIGN ligands that do not affect actin polymerization, such as ManLAM. Thus, although Salp15 binding to DC-SIGN does induce Raf-1 similar to ManLAM, the downstream effectors are different. Indeed, Salp15 affects pro-inflammatory cytokine production by human DCs both at the transcriptional and the post-transcriptional level. We demonstrate that Salp15 increases IL-6 and TNF-α mRNA decay, and impairs nucleosome remodeling at the IL-12p35 promoter. This is in contrast to pathogens such as mycobacteria and HIV-1 that interact with DC-SIGN to increase the transcription rate and prolong transcription activity through acetylation of p65 [9]. These data further demonstrate that pathogen binding to DC-SIGN might lead to different ligand-specific signaling cascades that regulate distinct adaptive immune responses. Although our data suggest that Raf-1 activation might play a central role in these immune responses, the downstream effectors of Raf-1 regulate the subsequent cytokine responses. Future research on the molecular mechanism of Salp15-induced DC immunosuppression will have to elucidate how MEK and downstream effectors affect mRNA stability and nucleosome remodeling to inhibit pro-inflammatory cytokine expression. This might be due to post-translational modifications of the proteins involved in nucleosome remodeling and mRNA decay [34,35]. Further characterization of the Salp15/DC-SIGN-induced signaling pathway could lead to the identification of new anti-inflammatory or immunosuppressive agents. TLR-mediated immune responses play an important role in a variety of diseases including infectious diseases, autoimmune diseases, and atherosclerosis. Therefore manipulation of TLR-triggered signaling is of wide clinical interest and the Salp15/DC-SIGN-induced signaling pathway described in the current study might prove to be important in the development of novel immunotherapies [36]. In addition, interfering with Salp15-induced signaling could potentially enhance anti-Borrelia immune responses and might be an alternative novel intervention strategy to prevent or potentially even treat Lyme borreliosis. Furthermore, efforts are being made to target biologically important vector proteins to prevent pathogen transmission from the tick to the host [37,38].
In summary, the salivary gland protein Salp15 is a tick protein exerting various activities at the tick-host-pathogen interface. We here report that Salp15 modulates TLR-induced DC pro-inflammatory cytokine production and renders DCs less capable of activating T lymphocytes. Salp15 binds to the surface of immature DCs through the C-type lectin receptor DC-SIGN, which results in the phosphorylation of the serine/threonine kinase Raf-1 and subsequent MEK activation. Silencing or inhibition of Raf-1 as well as inhibition of MEK abrogates the effects of Salp15. Both mRNA destabilization and impairment of nucleosome remodeling are responsible for the decreased pro-inflammatory cytokine production observed after DC-SIGN/Salp15 ligation. Immunosuppression of the host by Salp15 could be advantageous for both the arthropod vector as well as the spirochete, since it could impair adaptive immune responses against tick and/or B. burgdorferi antigens.
Salp15 (GenBank accession number: AAK97817) was isolated from cultured Drosophila S2 cells as described previously [4]. Briefly, Drosophila S2 cells (Invitrogen), co-transfected with the recombinant pMT/BiP/V5-His A-salp15 vector (Invitrogen) and the hygromycin selection vector, pCOHYGRO, were grown as large cultures in DES serum-free medium and induced with copper sulphate. The supernatant was used to purify recombinant Salp15 containing the V5 epitope and HIS-tag the using 5-ml pre-packed nickel charged HisTrap FF columns (GE Healthcare). The protein was eluted using 100 mM imidazole, extensively dialyzed against PBS (pH 7.4), and concentrated by centrifugal filtration through a 5-kDa filter (Vivascience). Tick saliva was collected from fed ticks as described previously [4]. Briefly, adult ticks were allowed to feed on rabbits, removed ticks were immobilized, and a capillary tube was fitted over the mouthparts. 2 μl of 5% pilocarpine (Sigma-Aldrich) in methanol was applied topically to the dorsa, and saliva was collected and stored at −80 °C until use.
B. burgdorferi sensu stricto strain B31 clone 5A11 [39], referred to as B. burgdorferi throughout the paper, was cultured in Barbour-Stoenner-Kelly (BSK)-H medium (Sigma-Aldrich). Spirochetes were grown to approximately 5 ×10 7/ml (enumerated using a Petroff-Hausser counting chamber as described previously [40]), pelleted by centrifugation at 2,000 x g for 10 min, and resuspended in RPMI medium without antibiotics.
Immature DCs were cultured as described before [41]. Immature DCs were used for experiments at day 6. A total of 100,000 DCs were stimulated with Salp15 (10–50 μg/ml) or tick saliva (diluted 1:75–1:300) in the presence or absence of Samonella typhosa LPS (10 ng/ml, Sigma-Aldrich) or LTA (10 μg/ml). For isolation of mRNA, cells were lysed after 6 h of incubation. To determine cytokine production and expression of cell surface markers, cells were incubated for 18 h. Supernatants were harvested to determine cytokine production, whereas the cells were analysed by flow cytometry analysis (FACS) for surface expression of CD80, CD86, or CD83. Cells were incubated at 4 °C for 30 min with PE-labeled antibodies (CD80-PE, CD86-PE, (Pharmingen); and CD83-PE (Immunotech). In addition, a staining with 7-amino-actinomycin D (7-AAD, Molecular Probes) was performed. Spirochetes (1 × 105 or 1 × 106) were pre-incubated with Salp15 (25 μg/ml final concentration) for 30 min before addition to 100,000 DCs. After 18 h of incubation supernatant was harvested, and cells were fixed in 2% PFA before performing FACS analysis. For binding experiments, we used Raji cells and Raji transfectants expressing wild-type DC-SIGN (Raji-DC-SIGN). Raji-DC-SIGN was generated as previously described [27,42]. All cells used in these studies were cultured in RPMI containing 10% fetal calf serum.
After 18 h of stimulation, DCs were washed extensively with medium. Next, DCs were cultured with allogenic PBLs (1 × 105) at different ratios (1:50–1:500) for 4 d at 37 °C. T lymphocyte proliferation was assessed by measuring the overnight incorporation of [methyl-3H] Thymidine (Amersham Biosciences).
For detection of cytokines, supernatants were harvested 18 h after DC activation and stored at −20 °C until further analysis. TNF-α, IL-10, IL-6, IL-12 p70, IL-8, and IL-1 β were measured using cytometric bead array kits (BD Biosciences) according to the manufacturer's recommendations.
mRNA was specifically isolated with the mRNA capture kit (Roche) and cDNA was synthesized with the reverse transcriptase kit (Promega). For real-time PCR analysis, PCR amplification was performed in the presence of SYBR green, as previously described [43]. Specific primers for IL-12p35, IL-6, TNF-α, IL-8, IL-10, and GAPDH were designed by Primer Express 2.0 (Applied Biosystems) [9]. IL-12p35, IL6, TNFα, IL-8, and IL-10 transcription was adjusted for GAPDH transcription. For the determination of mRNA half-life, DCs were stimulated with LPS for 6 h prior to the addition of 10 μg/ml actinomycin D (Sigma-Aldrich) to block transcription; mRNA was then isolated at 20-min time periods. We calculated the half-life of the different mRNA transcripts by applying non-linear fitting according to the one phase exponential decay model (Graphpad Prism Software version 4.0).
Fluorescent bead adhesion assays were performed as described previously [27]. Briefly, streptavidin-coated TransFluorSpheres (488/645 nm, 1.0 μm; Molecular Probes) beads were incubated with biotinylated goat anti-mouse F(ab)2 fragments (6 μg/ml; Jackson Immunoresearch), followed by overnight incubation with mouse anti-V5 (Invitrogen) (3 μg), and overnight incubation with Salp15 (5 μg), or PBS with 1% BSA for the generation of Salp15-, or negative-control beads, respectively. As a positive control fluorescent beads coated with the HIV protein gp120 were used [27]. 50,000 cells were incubated with beads for 45 min at 37 °C. When indicated cells were pre-treated with mannan (1 mg/ml), EDTA (10mM), free sugars N-acetylgalactosamine (GalNac), GlcNac, galactose, fucose; all 50 mM) or biotinylated LeX/LeY- (20 μg /ml) for 15 min at 37 °C. Bead adhesion to the cells was measured by FACS analysis.
Salp15 (5 μg/ml) was coated onto maxisorb ELISA plates (NUNC) for 18 h at room temperature (RT). The plate was blocked by incubating with 1% BSA for 1 h at 37 °C. Biotinylated lectins Con A, GNA (Galanthus nivalis agglutinin), NPA (Narcissus pseudonarcissus agglutin), PSA (Pisum sativum agglutin), LTA (Lotus tetragonolobus agglutin), UEA (Ulex europaeus agglutin), or PNA (Sigma-Aldrich) were added for 2 h at a concentration of 5 μg/ml. Binding of biotinylated lectins was detected using peroxidase-labeled strepavidin and absorbance was read at 450 nm.
The soluble DC-SIGN-Fc binding ELISA was performed as previously described [44]. Briefly, 5 μg/ml Salp15 or mannan was coated onto maxisorb ELISA plates (NUNC) for 18 h at RT. Unspecific binding was blocked by incubating the plate with 1% BSA for 1 h at RT. Soluble DC-SIGN-Fc was added for 1 h at RT. Specificity was determined (unless indicated otherwise) by blocking with mannan (1 mg/ml). Unbound DC-SIGN-Fc was washed away and binding was determined using a peroxidase-conjugated goat anti-human Fc antibody (Jackson Immunoresearch). Peroxidase-labeled strepavidin was used to detect DC-SIGN-Fc binding. Absorbance was read at 450 nm. To assess whether carbohydrate structures on Salp15 were involved in binding to DC-SIGN 10 μg of Salp15 deglycosylated using 1,000 units of PGNaseF under non-denaturing conditions according to the manufacturer's instructions (New England Biolabs), was coated.
The surface of DCs was biotinylated for 30 min at 4 °C with 0.5 mg/ml of sulfo-NHS-biotin (Pierce) in PBS (pH 7,4) and cells were then lysed for 1 h at 4 °C in lysis buffer (10 mM tri-ethanolamine [pH 8.2], 150 mM NaCl, 1 mM MgCl2, 1 mM CaCl2, and 1% [volume/volume] Triton X-100, containing EDTA-free protease inhibitors) (Roche Diagnotics). Salp15 ligands were immunoprecipitated with Salp15- (via anti-V5) coupled protein A–Sepharose beads (CL-4B; Pharmacia). As a positive and negative control, we used anti-DC-SIGN (AZN-D2) [27], and anti-V5-coupled protein A–Sepharose beads respectively. Immunoprecipitation products were separated by SDS-PAGE and transferred to nitrocellulose membranes. A Page Ruler Protein ladder (Fermentas) was run adjacently. Blots were blocked with 5% BSA in PBS followed by immunoblot analysis with streptavidin-coupled peroxidase (Vector Laboratories). To immunoprecipitate Salp15 ligands from non-biotinylated DC lysate we performed a similar assay. However, the blot was stained with specific goat antibodies against DC-SIGN (Santa Cruz Biotechnology), followed by secondary peroxidase-conjugated swine anti-goat (Tago). Blots were developed by enhanced chemiluminescence.
DCs were stimulated for 15 min with Salp15 (25 μg /ml) or controls. When indicated, cells were stimulated with LPS (10 ng/ml) or PMA (150 ng/ml) plus ionomycin (5 μg /ml) (positive control for ERK phosphorylation, Sigma-Aldrich) for 15 min, or pre-incubated with blocking anti-DC-SIGN antibodies (AZN-D2) [27] for 30 min. Subsequently, cells were fixed in 3% para-formaldehyde for 10 min and permeabilized in 90% methanol at 4 °C for 10 (Raf-1) or 30 (ERK) min. To assess phosphorylation of Raf-1 we used a rabbit anti-phospho-c-Raf (Ser338) mAb (Cell Signaling) and a rabbit anti-c-Raf (pTyr340, Tyr341) pAb (Calbiochem) and to assess phosphorylation of ERK a rabbit-anti-phospho-p44/42 MAPK (Thr202/Tyr204) mAb (Cell Signaling). Phosphorylation of Raf-1 and ERK was measured by flow cytometry after incubation with PE-conjugated donkey anti-rabbit antibodies as described [9]. When indicated, DCs were pre-incubated for 2 h with 1 μM GW5074 (Calbiochem), a Raf inhibitor, 4 μM U0126 (LC Laboratories), an inhibitor of MEK1 and MEK2, or 30 μM anacardic acid (AA) (Calbiochem), an inhibitor of the histone acetyltransferases p300 and CBP, respectively. Thereafter, cells were stimulated for 6 h with LPS as previously described, and mRNA was isolated for the generation of cDNA and real time PCR analysis.
DCs were transfected with 100 nM siRNA using transfection reagens DF4 (Dharmacon), according to the manufacturer's protocol. The siRNAs used were: Raf-1 SMARTpool (M-003601–00) and non-targeting siRNA pool (D-001206–13) as a control (Dharmacon). This protocol resulted in a nearly 100% transfection efficiency as determined by flow cytometry of cells transfected with siGLO-RISC free-siRNA (D-001600–01). At 72 h after transfection, cells were used for experiments as described above. Silencing of Raf-1 transcription was confirmed by quantitative real-time PCR.
To quantify nucleosome remodeling at the IL-12p35 promoter, chromatin accessibility was measured by a real-time PCR (ChART) assay. Nuclei were prepared from unstimulated cells, or cells stimulated as indicated, with lysis buffer (10 mM Tris-HCl [pH 7.5], 15 mM NaCl, 3 mM MgCl2, 0.5 mM spermidine, 1 mM PMSF, 0.5% Nonidet P-40). Digestion reactions were performed with 50 U BstIXI or 50 U EcoRI for 1 h at 37 °C. After proteinase K and RNase A treatment, DNA was purified using the QIAamp DNA blood kit (Qiagen). Real-time PCR reactions were then performed as described above. Amplification with primer set A (encompassing BstXI site located at nucleotide −298) is sensitive to remodeling of nuc-2 [16]. Increased accessibility of the region results in reduced amplification in the real-time PCR. Amplification with primer set B (encompassing BstXI site located at nt 456) was performed as an internal control to test for the efficiency of BstXI digestion as the accessibility of this locus is not subject to changes in the chromatin structure [16]. To normalize for DNA input amounts, each sample was analyzed with primer set C for GAPDH. Results are expressed as a percentage of the remodeling observed in the EcoRI-digested sample for each cell treatment using the formula (NtEcoRI – NtBstXI/NtEcoRI) × 100%, with Nt = 2Ct(primer set C)-Ct(primer set A). The following primer sequences were used: set A (IL-12p35 promoter): forward 5′ GCGGGGTAGCTTAGACACG 3′, reverse 5′ CCCAAAATGAAAGCGAAATG 3′; set B (BstXI control): forward 5′ TCTAAAGTCAGGCTTGGCCG 3′, reverse 5′ GGTTTCACCATGTTGGTCAGG 3′; set C (GAPDH promoter): forward 5′ TACTAGCGGTTTTACGGGCG 3′, reverse 5′ TCGAACAGGAGGAGCAGAGAGCGA 3′.
Statistical analysis was performed using parametric tests (Graphpad Prism Software version). When multiple conditions were compared a Dunnett multiple comparisons test was performed. Statistical significance of the data was set at p < 0.05, with one asterisk (*) representing 0.01 < p < 0.05; two asterisks (**) 0.001 < p < 0.01. |
10.1371/journal.pcbi.1003788 | Defining the Estimated Core Genome of Bacterial Populations Using a Bayesian Decision Model | The bacterial core genome is of intense interest and the volume of whole genome sequence data in the public domain available to investigate it has increased dramatically. The aim of our study was to develop a model to estimate the bacterial core genome from next-generation whole genome sequencing data and use this model to identify novel genes associated with important biological functions. Five bacterial datasets were analysed, comprising 2096 genomes in total. We developed a Bayesian decision model to estimate the number of core genes, calculated pairwise evolutionary distances (p-distances) based on nucleotide sequence diversity, and plotted the median p-distance for each core gene relative to its genome location. We designed visually-informative genome diagrams to depict areas of interest in genomes. Case studies demonstrated how the model could identify areas for further study, e.g. 25% of the core genes with higher sequence diversity in the Campylobacter jejuni and Neisseria meningitidis genomes encoded hypothetical proteins. The core gene with the highest p-distance value in C. jejuni was annotated in the reference genome as a putative hydrolase, but further work revealed that it shared sequence homology with beta-lactamase/metallo-beta-lactamases (enzymes that provide resistance to a range of broad-spectrum antibiotics) and thioredoxin reductase genes (which reduce oxidative stress and are essential for DNA replication) in other C. jejuni genomes. Our Bayesian model of estimating the core genome is principled, easy to use and can be applied to large genome datasets. This study also highlighted the lack of knowledge currently available for many core genes in bacterial genomes of significant global public health importance.
| Whole genome sequencing has revolutionised the study of pathogenic microorganisms. It has also become so affordable that hundreds of samples can reasonably be sequenced in an individual project, creating a wealth of data. Estimating the bacterial core genome – traditionally defined as those genes present in all genomes – is an important initial step in population genomics analyses. We developed a simple statistical model to estimate the number of core genes in a bacterial genome dataset, calculated pairwise evolutionary distances (p-distances) based on differences among nucleotide sequences, and plotted the median p-distance for each core gene relative to its genome location. Low p-distance values indicate highly-conserved genes; high values suggest genes under selection and/or undergoing recombination. The genome diagrams depict areas of interest in genomes that can be explored in further detail. Using our method, we analysed five bacterial species comprising a total of 2096 genomes. This revealed new information related to antibiotic resistance and virulence for two bacterial species and demonstrated that the function of many core genes in bacteria is still unknown. Our model provides a highly-accessible, publicly-available tool to use on the vast quantities of genome sequence data now available.
| The advent of next-generation sequencing (NGS) has greatly increased the number of bacterial genomes sequenced and made available for study in public databases such as GenBank, the Sequence Read Archive and European Nucleotide Archive (ENA) [1]–[3]. Increasing computational power allows for comparative genomics studies involving hundreds or even thousands of sequences, but large scale computational resources are not available to all researchers. Developing methods for analysing large datasets that capitalise on the computational power of modern desktop computers will make comparative genomics analyses much more accessible to the wider research community, allowing this vast quantity of data to be analysed more extensively.
A bacterial species can be defined by its pan-genome, which consists of a core genome conventionally defined as those genes present in all isolates, and an accessory genome, which includes the genes absent from one or more isolates or unique to a given isolate (note that we use the term “gene” here to refer to a putative protein-coding sequence) [4]. Identifying the core complement of genes in a bacterial species is often the first step in population genomics studies and the core genome can be defined in different ways. The most conservative and most frequently employed method is to only include genes present in 100% of isolates within the study population; however, this presents problems related to both biological sampling and the sequencing technology. Any collection of isolates is a subset of the entire population for the species of interest, and if the subset of isolates has limited genetic diversity then the number of “core” genes shared by all isolates in that sample will be higher than in a dataset which is genetically more diverse. This is not necessarily a problem, unless the intention is to extrapolate the findings to the wider bacterial population. Another biological limitation to using a 100% cut-off for inclusion in the core genome is that there may be rare variant strains which are missing genes that would otherwise be considered core genes. These variant strains may survive long enough to be sampled, potentially skewing the analyses. More generally, the size of the core genome is dependent on the size of the data set, with the core genome decreasing in size as more genomes are added to the analysis [4].
A large proportion of the bacterial genome sequences available at the time of writing are produced using next-generation sequencing platforms such as Illumina or Roche 454, so that even high-quality assemblies remain as incomplete or “draft” genomes. This is acceptable for most studies, but analyses of these genomes may exclude a gene from a list of core genes simply because it contains a sequence gap or is otherwise incomplete at that locus in the assembly of one or a few genomes. This assumes that the sequences being compared are all full-length: if an analysis accepts less than full-length coding sequences then gaps may not be an issue, but there will be other challenges with using incomplete sequence data, e.g. calculating pairwise distance (p-distance) measures.
If a definition of core genes as those found complete in all isolates in the dataset is too conservative, then the problem becomes that of determining an acceptable limit to the number of isolates missing any particular gene. One approach is to plot a frequency distribution that indicates how many genes are present in all isolates, or are missing in one or more isolates within the study population (Figure S1). For some bacterial species, there is a reasonably clear delineation between genes present in a large proportion of the study population versus those that are infrequent or rare, but for other bacterial species it is not clear.
Rather than make an arbitrary decision, we developed a statistical model for estimating the core genome that can be applied to different bacterial species by formalising the decision in the language of probability. The aim was to develop a Bayesian decision model to identify the genes found in what we will call the “estimated core genome” and apply this decision model to several large bacterial genome datasets. We described the nucleotide sequence diversity for each gene in the estimated core genome and considered how core genome sequence diversity varied across unrelated bacterial species. Finally, we depicted the data in a way that allowed us to explore the sequence data in greater detail and generate testable hypotheses about the estimated core genome.
The five bacterial species chosen for inclusion in this study were disease-causing organisms responsible for a large proportion of the global bacterial disease burden: Streptococcus pneumoniae (respiratory disease, the most important cause of infectious disease mortality); Campylobacter jejuni (gastrointestinal disease); Neisseria meningitidis (meningitis); Staphylococcus aureus (skin and soft tissue infections); and Helicobacter pylori (gastrointestinal ulcers).
In total, 2096 genomes were analysed across the 5 different bacterial species (Table 1 and Datasets S1). A phylogenetic network for each dataset was derived using Neighbor-Net [5] as part of the initial Genome Comparator [6] analyses (see Methods for a description of the Genome Comparator program); these diagrams demonstrated the overall diversity of the genomes in each study dataset (Figure 1).
The S. pneumoniae (pneumococcal) dataset consisted of 336 genomes for isolates of 39 different serotypes collected over 90 years (1916–2008) from at least 32 countries around the world. The isolates were recovered from individuals of a wide range of ages, including isolates from patients with disease and isolates recovered from healthy individuals. The multilocus sequence typing (MLST) data revealed 163 sequence types (STs), which could be clustered into 74 different clonal complexes (CCs) indicative of isolates with shared ancestry (Tables 1 and S1).
The largest genome dataset analysed was that of C. jejuni (N = 601 genomes). Isolates were recovered from human stool samples collected from patients in Oxfordshire, United Kingdom (UK) with gastroenteritis during 2011. 134 STs from 29 CCs were characterised in this collection, which was representative of the broader C. jejuni population genetic diversity [7], [8].
The N. meningitidis (meningococcal) dataset was comprised of 518 genomes and these isolates were collected nearly exclusively from patients residing in England, Wales and Northern Ireland in the 2010/11 epidemiological year, apart from 4 historical isolates from Norway (1976), The Gambia and UK (1983) and UK (1986). The 2010/11 genomes are part of the Meningitis Research Foundation Meningococcus Genome Library, which contains genomes from all culture-confirmed cases of meningococcal disease submitted to the Meningococcal Reference Unit in 2010/11 and 2011/12. Isolates of seven serogroups were included, mostly serogroup B (n = 394), Y (n = 74) and W-135 (n = 27). 198 STs were represented by the isolates and the STs clustered into 24 CCs. Culture-confirmed cases of meningococcal disease are largely representative of the England and Wales disease-causing N. meningitidis population as described previously [9].
The S. aureus dataset was large (N = 534 genomes) but genetically less diverse (25 STs and 11 CCs; Tables 1 and S1) than other datasets, since the analyses were restricted to methicillin-resistant S. aureus (MRSA) only. Most of the publicly-available genomes that are already published are of a limited number of CCs, predominantly the MRSA CCs that are epidemiologically the most important. The MRSA isolates were recovered from patients in 27 countries, although 39% of isolates were recovered in the UK. The H. pylori dataset included 107 genomes and 82% of the collection was from the USA, Canada or Japan. Only limited additional metadata were available for these isolates.
The size of the reference genomes used in the Genome Comparator analyses for each dataset varied from 1.6 to 2.8 Mb, and the total number of genes in each reference genome ranged from 1566 to 2547 (Table 2). There were small numbers of unique loci, i.e. genes found only in the reference genome and/or present in only one genome: S. pneumoniae (n = 6); C. jejuni (n = 6); N. meningitidis (n = 7); S. aureus (n = 4); and H. pylori (45).
An initial BLASTN criteria of 70% identity was chosen, which allowed for the identification of variable sequences among conserved gene classes [10] and avoided bias towards reference-specific sequences, and 100% sequence alignment, which means that coding sequences occurring at the ends of contigs or with gaps were therefore not included. Lowering the BLASTN criteria to 70% identity and 90% alignment increases the number of genes in the estimated core, as partial gene sequences will be detected and included (Table S2), which may be more suitable for other user-specific analyses but is not ideal for the calculation of p-distances to estimate sequence diversity.
The smallest estimated core genomes were those of MRSA and H. pylori (242 and 244 genes, respectively; Table 2) and the S. pneumoniae, C. jejuni and N. meningitidis core genomes were similar in size, ranging from 744 to 866 genes. The percentage of genomes within a dataset that possessed each estimated core gene ranged from ≥99.1% to ≥99.8%. The number of putative paralogues identified in the initial Genome Comparator analyses varied from 1 in C. jejuni to 40 among the S. pneumoniae genomes, and these genes were removed from further analyses (lists of putative paralogues for each species are provided in Table S3). If putative paralogues had not been removed, one would have been included in each of the estimated core genomes of S. pneumoniae and N. meningitidis.
Within each genome dataset, median p-distance values were calculated for each of the estimated core genes and the estimated probability density function was plotted for each bacterial species (Figure 2). The estimated probability density function plotted a smoothed histogram of the median p-distances vs. the estimated probability (relative frequency) of each p-distance value. The shape of the graphs for S. pneumoniae and C. jejuni were similar, showing a large peak of very small p-distance values, i.e. highly conserved genes, but these were entirely different from the graphs depicting the data for the other bacterial species. Each graph is an indication of the overall sequence diversity of the set of estimated core genes for that particular genome dataset.
The median p-distance value for each estimated core gene was then plotted against its position in the reference genome and illustrated as a circular bacterial chromosome (Figure 3). The length of each line indicates the median p-distance value for that gene. The estimated core genes were distributed around each genome and accessory regions in the reference genomes (e.g. ICE elements or phage genes), were observed as gaps where no core genes clustered. Estimated core genes with p-distances above 0 but less than the 95th percentile (blue lines) and those above the 95th percentile (red lines) stood out in a pattern on each genome diagram and allowed for an evaluation of specific genes and gene clusters in the genome, as demonstrated below. Table S4 lists all the estimated core genes and p-distances for each bacterial species. All genes with a p-distance value greater than the 95th percentile for each bacterial species, and the Cluster of Orthologous Groups (COG) functional category for each of those genes, are listed in Table S5.
This study exploited the large volume of publicly-available whole genome sequence data to outline a method for analysing bacterial genomes in a straightforward way, using web-based tools and computer programmes that run on modestly-powered computers. The analyses described here do not require access to supercomputers. The resulting data can be explored in a biologically relevant manner and there is flexibility to change the analysis parameters to suit different datasets and different questions.
As an example, we defined the core genes from among the coding loci present at full sequence length so that complete gene sequence information was included and p-distances could be reliably calculated. The model is designed to allow for the inclusion of genes present in <100% of genomes, which adjusts for arbitrary contig assembly issues. It is important to note that by excluding partial genes, many of which will be incomplete due to breaks in gene sequences based on sequencing technicalities and/or the genome assembly, the core genome estimates generated using our model are a conservative estimate for each bacterial species. However, as we demonstrated, simply lowering the sequence alignment to <100% will increase the number of estimated core genes, which may be appropriate for some datasets and analyses. In other words, our unit of count was the complete gene, but if the unit of count was nucleotides or the aim was to generate a list of full plus incomplete estimated core genes, then including partial sequences would be appropriate and requires the user to simply lower the sequence identity threshold in the initial analysis.
Most importantly, the process described here is completely transparent and the assumptions are easily understood, e.g. a list of genes is exported that includes the locus name, putative product, sequence length and genome position, which allows for detailed user inspection. The initial Genome Comparator output includes information about which genes are truncated (found at the ends of contigs) in each query genome and the user should evaluate these data carefully in conjunction with an assessment of the overall quality of the genome assemblies and consider whether or not to explore partial genes as part of a separate analysis. Furthermore, if a user wished to analyse amino acid sequences as opposed to nucleotide sequences, this is possible by selecting the appropriate option at the start of the Genome Comparator analysis; however, this will also significantly increase the run time and memory requirements and will become an issue with large genome datasets. A simpler option would be to convert the aligned nucleotide sequences for genes of interest into protein sequences after the Genome Comparator run is completed.
The estimated core genome sizes we obtained using our Bayesian model were expected to be lower than previous estimates, since the number of genes common to all genomes in a dataset decreases as the number of genomes increases, and our datasets are much larger and more diverse than the great majority of those previously analysed (see Table S8 for a list of relevant references). That is not a criticism of the previous studies and the analyses of small numbers of whole genomes; it is simply an indication of what data were available at the time each study was undertaken and how much the genomics field has changed in a short span of time.
We elected to demonstrate the utility of using this Bayesian model to determine which core genes are more diverse than others, but other investigators may wish to focus on the core genes that are highly conserved in a particular dataset. It is important to note that there is no one definitive “core genome” – the estimates of core genes will vary from dataset to dataset and between different methodologies.
We chose to use a Bayesian model for calculating the core genome but a frequentist model, based on a hypothesis test, could also have been applied. The advantage to using a Bayesian model is that it allows us to formalise the decision rule by a particular choice of prior. Furthermore, the Bayesian model implemented here does not account for correlations between the genes and each gene is considered to be statistically independent of each other; however, many genes are known (or likely) to be linked and operate as part of an operon or cassette. A revision of the proposed model could assess correlations between specific genes and/or gene regions. Such modifications would be a significant computational and statistical challenge given the large volume of genome data one would potentially wish to analyse, but a correlation model could provide useful biological information from the sequence data. Moreover, we selected the median as the summary of the distribution of gene pairwise distances since the underlying distributions of p-distances are not Gaussian. A better representation of the underlying probability distributions could be achieved by considering a number of different percentiles, but for this study we restricted our analyses to just the median.
Finally, the case studies we highlighted demonstrated how the data outputs and initial analyses may be used to derive more focussed analyses on specific genes or gene regions in the genome and generate new hypotheses to test. The genome diagrams are a useful way of depicting areas of potential interest in the genome and the bacterial species we evaluated here presented a huge range of possibilities for further study, from which we elected to highlight just a few as examples.
The field of bacterial genomics is advancing rapidly and it is now possible to generate enormous quantities of sequence data (albeit currently incomplete) at a low cost; therefore, it is also essential to find and develop suitable, widely accessible and inexpensive methods of processing and analysing these data in order to maximise the utility and benefits of whole genome sequence data. This model formalises the estimation of the core bacterial genome as a Bayesian decision problem and the resulting outputs reveal many areas for further exploration of the bacterial core genome.
Complete lists of the bacterial genome data included in this study, with accession numbers and available metadata (obtained by consulting the relevant published papers or websites) are listed in Datasets S1. Publicly-available whole genome sequence data for H. pylori (N = 107 genomes) and S. aureus (N = 534 genomes) were collected in two ways: 1) raw sequence trace files from the ENA were downloaded via an in-house genome assembly pipeline, assembled using Velvet [28] and uploaded to the rMLST BIGSdb database [6], [29]; and 2) finished reference genomes for each species were downloaded from GenBank and uploaded to the rMLST BIGSdb database. Only genomes that were already published in the scientific literature were included in our analyses. By comparison to MRSA genomes, few methicillin-susceptible S. aureus (MSSA) genomes are currently available (27 MSSA genomes were available at the time of our analysis, 18 of which were ST398) and thus we restricted the analyses to MRSA genomes only. Approximately 1000 S. aureus genomes were publicly available and published at the time, and we aimed to select a diverse dataset of ∼600 genomes such that the final MRSA dataset was similar in size to the C. jejuni and N. meningitidis datasets. Many of the available S. aureus genomes were ST239 or ST22, thus selection proceeded as follows: i) any non-ST239/ST22 genomes were automatically included; ii) among the 456 ST239/ST22 genomes available, 167 ST239 and 186 ST22 were selected (duplicates or re-sequenced genomes were removed); and iii) any genomes that were MSSA or an unknown ST (n = 54) were subsequently removed.
The Global Historical S. pneumoniae dataset (N = 336 genomes) included 85 assembled genomes from our previously published study [30]; sequences for 25 published genomes [31] downloaded from the ENA and assembled using Velvet; 134 genomes downloaded from GenBank; and 92 isolates sequenced and assembled as described in Protocol S1. Raw sequence data for the 616 pneumococcal genomes comprising the comparison Massachusetts data set [19] were downloaded from the ENA, assembled and uploaded to the rMLST BIGSdb database as described above.
Data for N. meningitidis were collected largely as part of the Meningitis Research Foundation Meningococcus Genome Library database (MRF GL; N = 514 genomes) plus 4 additional historical isolates were included [32]. The C. jejuni isolates included in this study (N = 601 genomes), all of human origin, were collected at the John Radcliffe Hospital in Oxford and form part of the Oxfordshire Human Surveillance collection [7]. Sequence data for C. jejuni and N. meningitidis can be found on the PubMLST [33] and rMLST BIGSdb [34] databases.
STs were assigned to genomes either by retrieving the ST information from previously published papers or by extracting the sequences corresponding to the MLST loci and looking up the ST on the MLST website (S. pneumoniae and S. aureus) [35]. STs and CCs for N. meningitidis and C. jejuni were extracted from the relevant BIGSdb databases. All other CCs were defined using goeBURST [36] and the species-specific MLST databases downloaded from the MLST website. When goeBurst could not resolve the group founder, the group was assigned to ‘CC NoneX’ where X is the ST with the lowest numerical value in the group. When a lack of closely-related STs meant that a CC could not be assigned, such genomes were named ‘SingletonX’ where X corresponds to the isolate ST. Table S1 provides a summary of the STs and CCs included in this study for each bacterial species apart from H. pylori. Although an MLST scheme is available for H. pylori, the high genetic diversity of the species means that virtually every new strain has new alleles and new STs, making the interpretation of such data difficult and thus we have not defined STs and CCs for H. pylori.
Genome Comparator is a component of the BIGSdb genome analysis database and software suite [7]; BIGSdb facilitates whole genome analysis based on the allelic variation of individual genes. The BIGSdb Genome Comparator tool allows whole genome sequence data for one or more genomes to be compared against an annotated reference genome. The BLASTN parameters selected used a cut-off of 70% identity over a 100% alignment with a word size of 15. Potential paralogues were removed from the analyses by identifying which of the coding loci were found in more than one copy in any query genome and excluding these sequences from any further analyses. For each coding locus in the reference genome, ClustalW [37] sequence alignments were generated for all of the query genomes containing that particular sequence. Neighbor-Net diagrams were also created by Genome Comparator as part of its standard analysis and figures were created using SplitsTree [38].
For each of the collections of bacterial genomes, we have chosen a reference genome, consisting of a set of genes . Each gene is considered independently during the analysis. Let be the number of isolates under consideration. For each isolate, , let if the -th gene is present in isolate or zero if it is not present. Then, for the -th gene, is the number of times the gene is found in isolates. We model as a sequence of binomial random variables with the probability parameter . Letting denote a probability density, the probability of observing the -th gene times in isolates given the model parameter is:(1)The above equation, viewed as a function of , is the binomial likelihood function. We specify, for the parameter , a prior probability density . Then, using Bayes' rule, we can compute the posterior density of conditional on the observed frequency:(2)If we assume the prior density to be a beta density with parameters then we can combine the prior with the binomial likelihood (1) and use Bayes' rule (2) to find that the posterior density is also a beta distribution with parameters [39]. The parameters of the prior are related to the posterior parameters by and .
The posterior density represents our uncertainty of the parameter . If the density has a greater value near then we are inclined to believe that the gene is in the core genome. In light of this observation we introduce the following decision rule for each gene:(3)The set of genes not rejected according to (3) can be defined as the estimated core genome.
The selection of a prior amounts to specifying our prior belief of whether a gene is or is not in the core genome. We might also adopt the belief that we are equally unsure of whether a gene is present in the core or not. Priors that reflect this type of belief are known as near ignorance priors [40]. Rather than selecting a near ignorance prior we argue that a prior should be selected to reflect the nature of the decision process. Prior to analysing any of the isolates we have no reason to believe that any of is or is not in the core. As each strain is analysed we accumulate evidence to suggest that the -th gene is not a core gene. To reflect this process we adopt the prior belief that every gene is a core gene and then attempt to falsify this statement using the decision rule. Formally, this corresponds to selecting as our prior a beta density with parameters .
Custom Perl scripts were used to split the merged sequence alignment files generated by Genome Comparator into separate sets of nucleotide sequence alignment files by estimated core gene, and then the nucleotide distance between each pair of sequences for each estimated core gene was calculated. We counted the number of sites at which the nucleotides differed between each pair of genes. We let be the proportion of sites that differed between isolate and isolate for the -th gene in the estimated core. Then for each gene the matrix of pairwise evolutionary distances was calculated using the Jukes-Cantor model [41] where the entry of the matrix was given by:(4)The median p-distance value was then calculated as a summary statistic for each estimated core gene:(5)For each species we collected the median distances for each gene in the estimated core. This information was plotted against the genome position of each gene (relative to the position of that locus in the reference genome) and depicted in a circular diagram created using Circos [42]. In addition, the estimated probability density function of the median pairwise evolutionary distances for each species, or for individual genes within a species, was plotted using ksdensity (Kernel smoothing function estimate) [43]. Finally, the COG functional groups of the genes with p-distance values greater than the 95th percentile were determined using eggNOG [44].
The computationally intensive part of the analysis is the Genome Comparator run (because it creates sequence alignments for every gene), but this runs via a publicly-available web interface on a cluster of servers hosted at the University of Oxford. Output files are stored for one week on the server. The Bayesian model for estimating the number of core genes, the calculation of p-distance values for all reference genes, creation of the genome diagrams (Figure 3) and the generation of estimated probability graphs (Figure 4) can be implemented using freely available scripts written in the open source R software package [45]. Along with wrapper scripts that prepare the Genome Comparator outputs and a detailed manual, the relevant code is available at: https://sourceforge.net/projects/bayesianestimatedcoregenome/. As a frame of reference, the Genome Comparator analysis of the largest dataset, C. jejuni (601 genomes) took 90 hours to run, including all sequence alignments and generation of the Neighbor-Net diagram, but the subsequent steps took approximately 2 hours and can be run on any modestly-powered computer.
For case studies 2 and 3 the names of the core genes for each of the datasets included in the comparisons were compared to each other using the VLOOKUP function in Microsoft Excel, which matches cells containing the same text (the same reference genomes were used for each dataset so the gene names were the same) in order to generate numbers of shared and unique genes. Functional groups for each of the sets of unique genes were assigned using eggNOG.
|
10.1371/journal.pgen.1002657 | Deep Sequencing of Plant and Animal DNA Contained within Traditional Chinese Medicines Reveals Legality Issues and Health Safety Concerns | Traditional Chinese medicine (TCM) has been practiced for thousands of years, but only within the last few decades has its use become more widespread outside of Asia. Concerns continue to be raised about the efficacy, legality, and safety of many popular complementary alternative medicines, including TCMs. Ingredients of some TCMs are known to include derivatives of endangered, trade-restricted species of plants and animals, and therefore contravene the Convention on International Trade in Endangered Species (CITES) legislation. Chromatographic studies have detected the presence of heavy metals and plant toxins within some TCMs, and there are numerous cases of adverse reactions. It is in the interests of both biodiversity conservation and public safety that techniques are developed to screen medicinals like TCMs. Targeting both the p-loop region of the plastid trnL gene and the mitochondrial 16S ribosomal RNA gene, over 49,000 amplicon sequence reads were generated from 15 TCM samples presented in the form of powders, tablets, capsules, bile flakes, and herbal teas. Here we show that second-generation, high-throughput sequencing (HTS) of DNA represents an effective means to genetically audit organic ingredients within complex TCMs. Comparison of DNA sequence data to reference databases revealed the presence of 68 different plant families and included genera, such as Ephedra and Asarum, that are potentially toxic. Similarly, animal families were identified that include genera that are classified as vulnerable, endangered, or critically endangered, including Asiatic black bear (Ursus thibetanus) and Saiga antelope (Saiga tatarica). Bovidae, Cervidae, and Bufonidae DNA were also detected in many of the TCM samples and were rarely declared on the product packaging. This study demonstrates that deep sequencing via HTS is an efficient and cost-effective way to audit highly processed TCM products and will assist in monitoring their legality and safety especially when plant reference databases become better established.
| Chemicals derived from plants and animals are widely used in traditional Chinese medicine (TCM), and it is commonplace for remedies to contain a complex list of ingredients. Due to their heterogeneous origins, and subsequent processing into pills and powders, it can be difficult for the biological origin of ingredients within each remedy to be reliably determined. In this study, we have, for the first time, used a second-generation DNA sequencing method to analyse TCM remedies and determine their animal and plant composition. Using this deep-sequencing approach we identified plant species that are known to contain toxic chemicals and identified animal DNA from species that are currently endangered and protected by international laws. Consumers need to be made aware of legal and health safety issues that surround TCMs before adopting them as a treatment option. More widespread testing of complementary medicines using the DNA methods developed herein represents an efficient and cost-effective way to audit their composition.
| Traditional Chinese medicines (TCMs) have been an integral part of Chinese culture and the primary medicinal treatment for a large portion of the population for more than 3000 years [1], [2]. Outside of Asia there has been, in recent decades, a growing use of TCMs where they are being taken in conjunction with, or as an alternative to, conventional Western medicine [3], [4]. The increasing popularity of TCM products has seen the monetary value of the industry increase to hundreds of millions of dollars per annum [5], its growth paralleled by the global increase in the use of complementary and alternative medicines. Despite its increased uptake, the therapeutic benefits of only a small number of TCM products have been scientifically validated [6], with their perceived efficacy being based largely on long-standing beliefs [7].
Chinese herbal medicines often contain numerous different plant and animal-derived products that combine to act synergistically to affect a desired outcome [8], [9]. However, due to the proprietary nature of TCM manufacture, coupled with a lack of industry regulation, the biological origin of contents can be difficult to determine with confidence, leading to questions regarding TCM quality, efficacy and safety [10], [11]. Undeclared or misidentified TCM ingredients and adulterants can pose serious health risks to consumers [10], [12], [13]. These include: allergenic substances [14], plant toxins [7], heavy metals such as mercury, lead, copper and arsenic [15], and pharmaceutically active compounds of undetermined concentration [5]. In the early 1990s the misidentification of the toxic herb Aristolochia fangchi for the anti-inflammatory agent Stephania tetrandra led more than a hundred women to suffer kidney failure, with many later developing cancer of the urinary system [13].
In addition to safety concerns, issues of legality also surround TCMs. These concerns fall into three main categories: matters relating to the trade of endangered species; issues pertaining to honesty of food labelling; and adulteration of samples with drugs. Some TCMs contain plant and animal species [16]–[18] that fall under the jurisdiction of the Convention on International Trade in Endangered Species (CITES). CITES-listed species (see appendicies at www.cites.org) that have had long-standing associations and use within TCM include: Asiatic black bear (Ursus thibetanus, Appendix I listed), Saiga antelope (Saiga tatarica, Appendix II listed), rhinoceros (all species, Appendix I listed), and non-cultivated varieties of the plant genus Panax; P. ginseng and P. quinquefolius, (Appendix II listed) [19]–[23]. The CITES appendices include lists of species afforded different levels or types of protection from over-exploitation. Appendix I species are deemed the most endangered and threatened with extinction, with Appendix II and III listed species regarded to be at lower, but still significant, threat levels [24]. With an increased international demand for TCMs, ascertaining the biological origins, and hence the CITES status, of ingredients contained variously in capsules, powders, liquids, and tablets represents a complex problem for customs officials. The second issue of legality concerns the mislabelling of TCMs. This might be done intentionally in order to reduce manufacturing costs, or to circumvent customs' scrutiny, or inadvertently if the TCM practitioner unwittingly uses a misidentified product [25]. For CITES member states to enforce legislation and to prosecute cases of illegal trade, reliable methods of species identification are needed [26]. Lastly, a number of TCM products appear to have been intentionally adulterated with drugs of known pharmacological activity such as anti-hyperglycaemic agents (anti-diabetic medication) and corticosteroids [5], presumably as a means to increase their efficacy.
To date, many of the analyses and identification of botanical components in TCM products have employed chromatographic methods [9], [27]. However, these methods may not be able to identify animal species, or be able to uncover all of the ingredients within heterogeneous samples. DNA technology has the potential to provide information about species composition and the honesty of ingredient declarations. For the identification of botanical constituents used in TCMs, the genetic techniques employed include fragment length polymorphism analysis, dot-blot hybridization, micro-arrays, and sequencing of plastid DNA genes [25], [28]–[33]. Likewise, genetic identification of animal species commonly involves DNA sequencing and characterisation of mitochondrial DNA (mtDNA) genes [1], [32], [34]. Despite the variety of genetic work that has been conducted to date, investigative research seems to have focused on detecting the DNA of specific targets within TCMs [22], [28], [30], [35]–[38] or herbal teas [39] rather than investigating all of the contributing species within a sample simultaneously.
The advent of Second Generation, high-throughput sequencing (HTS) platforms have enabled the rapid sequencing of genes, genomes and metagenomes [40]. The ability of these technologies to deep-sequence both PCR amplified plastid and mtDNA markers (using molecular identifier [MID] tags) has allowed the species composition of a variety of complex substrates including faecal material [41], sediments [42] and even, in a forensic context, microbial communities on computer keyboards [43], to be determined. The application of HTS technologies to analyse complementary medicines has not been previously attempted, but is likely to prove to be the best approach by which to genetically audit the species composition of multiple TCM samples in parallel.
Given the worldwide popularity, growing use and increasing financial significance of TCMs, an effective means of evaluating these medicines is urgently needed – a sentiment echoed by strategy reports from the World Health Organization (WHO) [11]. This study sets out to explore the probative value of HTS approaches by generating species audits from 15 TCMs (Figure 1; Table 1) seized by border protection officials upon entry into Australia.
An in-depth genetic audit of the species constituents contained within 15 TCM samples (Figure 1, Table 1) was determined by using amplification of trnL (p-loop, plastid) and 16S rRNA (mtDNA) genes, followed by deep sequencing via HTS (see methods). More than 49,000 sequence reads were obtained from the HTS approach using both trnL c/h and 16S primers, with the analysis of the plant and animal constituents discussed separately below. The DNA isolated from the various TCM samples was highly variable in quality. Using trnL and 16S primers in qPCR assays, DNA of sufficient quality was obtained from 15 of 28 (54%) samples attempted. Some of the TCMs failed to amplify due to severe PCR inhibition, while others yielded little, or no DNA. As with many other degraded/processed substrates it may be necessary to optimise DNA extraction methodologies depending on the physical and chemical properties of the TCM.
To our knowledge, this is the first study to apply an HTS approach to ascertain the species composition of medicinal products. Until recently, to dissect the molecular components of heterogeneous biological samples (such as TCMs) it has been necessary to clone amplicons into plasmid vectors and then sequence the insert. In direct contrast to previous cloning based methodologies HTS provides deeper coverage of more samples in a shorter time period, and represents a cost effective way to audit DNA in heterogeneous samples. The sequencing of indexed (MID-tagged) PCR amplicons [44] allows for the sequencing of multiple samples in parallel, with the GS Junior or Ion Torrent conservatively generating ∼50,000 reads for c. US$1000 [45]. DNA isolation and quantification of 15 TCM samples followed by a single HTS run of the pooled and tagged PCR products, was estimated, in this case, to cost less than $35 per sample (excluding labour). This demonstrates that after an initial outlay for MID-tagged primers this approach is extremely cost-effective. As such, the approach described here is both cost-effective, accessible, and can be easily adapted to profile the molecular constituents of other biologically derived complementary and alternative medicines. One of the aims of this study was to determine the efficacy of HTS auditing approaches specifically with the goal of screening additional samples whose constituents might need to be identified in cases involving illegal imports, food fraud, medicine fraud and forensics.
Taxonomic assignment of DNA sequences to a family, genus or species represents a complex problem, the accuracy of which largely depends on the level of coverage afforded by reference databases, the analytic method used [46] and the accuracy of the underlying taxonomic framework. In the TCM data generated here the vertebrate assignments were relatively straight forward, in contrast to the plant assignments, which were particularly challenging. The detection and identification to the family level, of genetically well-characterised plants and animals is generally uncomplicated. In contrast, if species-level assignments (without uncertainties) are required for each trnL sequence, the task is largely unachievable with current databases. While the MEtaGenome ANalyzer (MEGAN) [47] based assignment approach is not without problems, it is currently the best way to parse thousands of sequence reads. Alternative methods for assigning sequences are also available such as SAP [48] and QIIME [49] although all of these methods are computationally intensive when challenged with large volumes of data. Irrespective of the species assignment methodology used, whether it be phenetic or character-based, all are ultimately dependent on good reference database coverage and a robust taxonomy.
There are a number of caveats with regards to HTS technology that need to be considered when analysing data. Firstly the error rate of 454 Titanium chemistry is estimated to be ∼0.5–1% [50]. On top of this there is the possibility that recombination might occur, albeit at a low (∼0.3% on an Illumina platform) frequency [51]. The likelihood of error and recombination should at least be acknowledged, but with respect to the plastid trnL data presented here it is debatable how significant an impact this is going to have on species assignments due to the presence of both sequence and length polymorphisms in the p-loop region. Lastly, caution also needs to be exercised with drawing correlations between the genetic profiles detected by HTS approaches and the actual composition of the TCM. No genetic audit can detect DNA when it has been completely degraded (for example by processing procedures) and there will always be variation in the DNA concentrations between ingredients. The results should therefore be regarded as a qualitative, and potentially incomplete assessment of composition rather than a quantitative measure of each ingredient.
Within the confines of a manuscript it is impossible to document the significance of each of the ∼50,000 reads in this audit, instead, a summary of the data is presented (Table 2 and Table 3, and Figure S1A–S1N) and the discussion will focus on some of the more common, illegal or hazardous ingredients.
A total of 68 plant families were identified in this study with 48,682 DNA sequence reads (on average 3,745 per TCM sample) generated using the trnL c/h primer set [52] for the 13 analysed samples (Table 2). Six of the most common plant families that were identified included Fabaceae, Asteraceae, Poaceae, Lamiaceae, Solanaceae, and Apiaceae, with 70% of the samples containing at least three of these families (Table 2). Some of the most common plant genera identified in the TCM samples were Glycyrrhiza (liquorice root, Family Fabaceae), found in 62% of samples, Mentha (mint, Family Lamiaceae), found in 46% of samples and Asarum (wild ginger, Family Aristolochiaceae) found in 31% of samples. Mint is commonly included in medicines and is used in TCM to treat gastrointestinal upset, gallbladder problems and upper respiratory symptoms [53]. Likewise Glycyrrhiza uralensis, or Chinese liquorice root, is a common component of TCM remedies and is classified as one of the Chinese 50 fundamental herbs [54]. Containing glycyrrhizin, G. uralensis can be processed by microbes into 18β-glycyrrhetic acid — effective in the treatment of peptic ulcers, as well as having antiviral and antifungal activities [55]. Heavy harvesting of G. uralensis from the wild for TCM products, has resulted in the threat of species extirpation in Chinese provinces such as Gansu [56].
The results of the trnL audit on four samples, Yatong Yili Wan (TCM-016), Laryngitis pills (TCM-006, TCM-021), and Lingxin Mingmu Shangging Wan (TCM-013), indicated they contained DNA with close (>98%) similarity to the genera Ephedra and/or Asarum (Table 2). These TCMs could potentially pose a risk, as compounds from these genera can be poisonous or toxic at high dosages. Ephedra is classed as a poisonous herb, with Ephedra-containing products having been banned by the U.S. Food and Drug Administration (FDA) since 2004 [57]. Remedies that contain Ephedra should only be prescribed by experienced practitioners, as the therapeutic dose range is narrow [8]. Aristolochic acid, the same compound as contained in Aristolochia species, a known nephrotoxin, hepatotoxin, and carcinogen [27], [58], may be contained in certain species of Asarum. Further compound specific testing (via metabolomics) of TCMs from which Asarum DNA was detected (TCM-006; TCM-013; TCM-016; TCM-021, Figure 2, Table 2) would be required to determine whether this acid is actually present in the TCMs analysed here.
One trade-restricted plant species commonly found in TCM preparation is Panax ginseng (CITES Appendix II). Non-cultivated P. ginseng is subject to CITES regulation only when in the form of a whole root, or sliced parts of the root, and not after processing and manufacture [23]. It was not possible using the conservative assignment criteria implemented in MEGAN to definitively identify the genus Panax, this is primarily because the bit-score match was equally good with the genus Hedera (ivies). Both Panax and Hedera are in the family Araliaceae and further molecular characterisation is required to distinguish if one or both of these genera are present in the TCM-001, TCM-011, TCM-018 and TCM-027. Even if Panax is confirmed, the fact that all the TCMs containing Araliaceae sequences are in powdered form render them technically not subject to CITES legislation.
Additional plant taxa with purported medicinal activity identified in the samples include Xanthorhiza simplicissima (Ranunculeae), and Sophora flavescens (Fabaceae). Xanthorhiza simplicissima (Yellowroot) is a native American medicinal containing berberine which is anti-inflammatory, astringent, hemostatic, antimicrobial, anticonvulsant, immunostimulant, uterotonic and can temporarily lower blood pressure [59]: the roots of Sophora flavescens contain alkaloids such as oxymatrine and is commonly used to treat fever, asthma, cancer and viral myocarditis [60], [61]. Plant DNA assigning to the families Cannabaceae, Ranunculaceae, and Solanacea, which are known to contain medicinally important species, were also recovered. However to resolve these sequences beyond the family level another gene region would need to be targeted, and this might reveal, for example, whether the Solanaceae (Nightshade family) identified in four of the TCM samples comprised S. chrysotrichum (Giant Devil's Fig) which has known pharmacological activity [62], or perhaps the less exotic species such as potato or tomato.
The complexity and risk of possible drug interactions for consumers using TCMs in combination with conventional medicines could be heightened when there are poisonous or toxic ingredients of unknown concentrations in herbal remedies that may not be listed on the packaging (Table 1). Further to potential adverse drug interactions is the possibility of ingesting allergenic substances within herbal remedies, such as nuts, which can cause anaphylaxis in those with severe allergy. DNA from the Anacardiaceae (the cashew or sumac family) was detected in two TCMs - nut proteins from this family are know allergens [63]. Likewise, Glycine (soybean) was detected in four TCMs and is known to contain at least 16 potential protein allergens with the potential to cause adverse reactions ranging from mild rashes to life threatening systemic anaphylaxis [64]. However, our results were unable to determine whether the recovered DNA is derived directly from the nut/bean, or originates from other plant tissue.
The variety of species that the HTS technique can reveal when analysing TCMs, is demonstrated by the results obtained for the Yatong Yili Wan pills (TCM-016). This sample was one of the most botanically complex, containing 16 identifiable plant families. 2,124 DNA sequence reads, were assigned to a GenBank reference database sequence (Table 2; Figure 2), based on cut-offs in MEGAN (see methods). SAP analysis was also conducted on representative sequences from each of the terminal nodes. Results generated by SAP were in close accordance with the MEGAN assignments with high posterior support. The two cases where no assignment was made was the result of insufficient database coverage – the method is reliant upon sufficient sequence coverage to construct a phylogeny. A third assignment method was also implemented, QIIME, the results of which were also in broad agreement with the MEGAN and SAP assignments (Figure 2).
What is clear from the plant assignments of the HTS data is that better reference databases involving multiple genes (such as: trnL, rbcL, ITS and matK) are required to improve the species assignment. A medicinal materials DNA barcode database (MMDBD) is currently being generated and compiled to include thousands of DNA reference sequences for these and other genes covering species of plants, animals, insects and fungi that are commonly used in TCM (available at; http://www.cuhk.edu.hk/icm/mmdbd.htm) [31]. The recent work of the China barcode of life group [65] which has sequenced ∼6000 species may soon rectify inadequacies in the plant databases. Despite the constantly improving databases, the taxonomic framework under which the DNA assignments operate also needs to be scrutinised. What is reassuring about HTS data is that while the resolution may not currently be available, efforts to improve databases and the underpinning taxonomies are continually improving and hence the accuracy of assignments can only get better.
With the potentially enormous volumes of plant data produced (over 7,662 reads in the case of TCM-006), it is tempting to look for quantitative signals in results, but owing to various factors including differential preservation of DNA in the raw ingredients, different processing techniques, variation in PCR efficiency (due to amplicon length variation and primer binding site polymorphisms), a universal primer approach should be viewed as semi-quantitative at best. In the worst-case scenario a constituent may be entirely undetected, especially if it occurs at a very low abundance.
With the exception of human-derived sequences (which were excluded), vertebrate genetic signatures were detected in nine samples tested using two universal 16S rRNA primer pairs [66], [67]. A total of eight animal genera were identified from 539 DNA sequences (Table 3). The taxonomic assignments of the vertebrate sequences were simpler in comparison to the plant assignments, due to substantially better GenBank coverage, but as with other forensic studies caution still needs to be exercised when assigning a species in casework [68], [69]. This study identified four TCM samples - Saiga Antelope Horn powder (TCM-011), Bear Bile powder (TCM-015), powder in box with bear outline (TCM-024) and Chu Pak Hou Tsao San powder (TCM-027) – that were found to contain DNA from known CITES listed species. Two of these CITES species are classified by the IUCN Red List as vulnerable (Ursus thibetanus) and one as critically endangered (Saiga tatarica) (Table 3). The threat posed to these and other animal species' survival caused by the demand for TCM products is high [7], [18]. This highlights a serious concern for the conservation of these species and it is evident that illegal hunting still persists despite a high level of legal protection [70]. One hundred and seventy five countries are signatories to CITES, including China (member party since 1981) [24], yet penalties for illegal trafficking are relatively minor and penalties are rarely enforced [18]. DNA testing of highly processed medicines may assist in the successful prosecution of individuals who seek to profit from the illegal trade in endangered taxa. Likewise, such genetic screens will help to legitimise those medicines that contain components that are not trade restricted, but may still be confiscated on grounds of suspicion (e.g. TCM-003, 006 and 021).
Of the samples analysed using the 16S rRNA primers, 44% contained two or more animal species within the same sample (Table 3). Some of these species, such as water buffalo (Bubalus bubalis), Asiatic toad (of the genus Bufo), and domestic cow (Bos taurus), are known for their use in medicinal products [27], [71], whereas use of goat (Capra hircus) is less well represented in the literature and may be used as a substitute for traditionally used animal species. As with all animal-containing products the consumer needs to be aware of the possibility of zoonotic pathogens, such concerns have been raised previously in the context of TCM [7].
Consumers of TCMs need to be wary of honesty of food labelling [72], as in 78% of samples, animal DNA was identified that had not been clearly labelled on the packaging (in either English or Chinese). This adulteration of medicine occurred in the Saiga Antelope Horn powder (TCM-011; Table 1) which claimed to be 100% pure, yet was found to also contain significant quantities of goat (Caprine) and sheep (Ovine) DNA (Table 3). In some TCMs, undeclared ingredients are used to reduce the cost of manufacture of the medicine by increasing the bulk of the powder, but it is impossible to determine why Caprine and Ovine appeared in this product. Water buffalo (Bubalus bubalis), domestic cow (Bos taurus) and deer species were also not listed on the packaging of samples in which they were genetically identified (Table 1 and 3). The inadvertent consumption of undeclared animal products found in 78% of the medicines, such as bovid, risk violating certain religious and/or cultural strictures.
The results of this study demonstrate that high-throughput DNA sequencing methods are an invaluable tool for analysing constituents within complex TCMs. The techniques used enabled the identification of a larger number of animal and plant taxa than would have been possible through morphological and/or biochemical means. HTS methodology is well suited to the analysis of highly processed and degraded DNA from TCMs, including powders, crystals, capsules, tablets, and herbal tea. It is manifestly obvious that if there are trade-restricted biological materials in TCMs, or if they contain DNA from species known to synthesise toxic compounds, that better methods of detection are urgently required. Even in the 15 TCMs tested here, the occurrence of CITES-listed species, potentially toxic/allergenic plants and non-declared constituents was all too common. However, it should also be noted that the detection of DNA from a pharmaceutically active species does not necessarily indicate the presence of bioactive compounds: metabolomic analyses can be used in addition for the detection of specific compounds. For example, the bear-bile powder (TCM-015; Table 1 and Table 3) containing Asiatic black bear DNA was analysed using Gas Chromatography Mass Spectrometry and yielded a mass spectra consistent with ursodeoxycholic acid (data not shown), an active component of bile that has been reported to reduce pain and inflammation [73].
In the future, TCM screening approaches that involve both genetic (for species composition) and metabolomic (for compound detection) approaches could represent the best way to audit complementary medicines. With regard to TCMs and complementary medicines as a whole, controls need to be implemented to ensure consumer safety and to minimise impacts on protected biota. It is also important that consumers are made fully aware of legal and health safety concerns that surround TCMs before adopting them as a treatment option. A recent opinion piece [74] stated “if TCM is to take its place in the modern medicine cabinet, then it must develop ways to prove itself” – we endorse this view and note that it applies equally to safety as it does to medical efficacy.
Twenty-eight TCM samples were obtained from the Wildlife trade section of the Department of Sustainability, Environment, Water, Population and Communities after being seized by Australian Customs and Border Protection Service at airports and seaports across Australia. The samples were seized because they contravened Australia's international wildlife trade laws as outlined under Part 13A of the Environment Protection and Biodiversity Conservation Act 1999 (EPBC Act). The samples were stored in a quarantine-approved facility within the laboratory after being catalogued. TCM sample types included: powders, bile flakes, capsules, tablets, and herbal tea. Small amounts of each sample (between 70–290 mg) were dispensed into 2.0 mL Eppendorf tubes and digested overnight, on a shaking heat block at 55°C, in 700 µl–1500 µl of tissue digest buffer consisting of; 1 mg per mL proteinase K (Amresco, OH, USA), 20 mM Tris pH 8.0 (Sigma, MO, USA), 2.5 mM EDTA (Invitrogen, CA, USA), 5 mM CaCl2 (Sigma), 20 mM DTT solution (Thermo Fisher Scientific, MA, USA), 1% SDS (Invitrogen), and milliQ water.
All samples were centrifuged after digestion for 3 minutes at 16,813×g. 200 µL of supernatant was mixed with 1 mL of Qiagen (CA, USA) PB buffer and transferred to a Qiagen (PCR cleanup) spin column and centrifuged for 1 minute at 16,813×g. Two wash steps followed (Qiagen AWI then AWII buffer) prior to elution of DNA from the spin column membrane with 50 µL of 10 mM Tris pH 8.0. The DNA extracts were then quantified via real-time quantitative polymerase chain reaction (qPCR; Applied Biosystems [ABI], USA) using trnL g/h [52] and 16S ribosomal RNA (rRNA) [66], [67] primers (Integrated DNA Technologies [IDT], USA) (Primer sequences displayed in Table S1). Samples were assessed for quality and quantity of DNA using qPCR at three DNA dilutions (undiluted, 1/10, 1/100) to determine if successful isolation of DNA was achieved, and to investigate the presence of PCR inhibition. The trnL g/h qPCR assay was conducted in 25 µL reactions using ABI Power SYBR master mix together with 0.8 µM of trnL g and trnL h primers and cycled at 95°C for 5 minutes followed by 40 cycles of 95°C for 30 s, 50°C for 30 s, 72°C for 30 s, with a 1°C melt curve stage and a 10 minute final extension at 72°C. The 16S qPCR was conducted using the same conditions, except for the primer concentration used, which was 0.4 µM and an annealing temperature of 57°C. An optimal DNA concentration, free of inhibition was selected and used for further analysis. Samples with low template amounts and/or severe inhibition were not processed further.
Fusion primers with unique 6 bp MID tags were designed [74] for both the 16S rRNA [65], [66] (∼150 bp product for 16Smam, ∼250 bp product for 16S1/2 degenerate primers [Table S1]) and the p-loop region of trnL [52] (c/h primers generating a size variable product averaging ∼250 bp product [Table S1]) (IDT, Australia). The trnL c/h primer sets were used to generate a longer PCR amplicon for future HTS, instead of the trnL g/h primer set (∼100 bp) which were only used for initial quantification. For the most part, when we used qPCR on the c/h and g/h primers, there were no significant drops in detected copy number. For this reason we selected the longer c/h set as it affords greater taxonomic resolution. Ten samples were PCR amplified using both the trnL c/h and 16S fusion primers, three samples were PCR amplified using trnL c/h fusion primers only, and two samples were PCR amplified with 16S fusion primers only. Amplicons were generated via PCR for each sample in triplicate (Corbett Research, NSW, Australia) and pooled in an attempt to reduce the effect of PCR stochasticity. The trnL c/h PCR was carried out in a 25 µL total volume including 4 µL of template DNA, with the following reagents: 2 mM MgCl2 (Fisher Biotec, Aus), 1× Taq polymerase buffer (Fisher Biotec, Australia), 0.4 µM dNTPs (Astral Scientific, Australia), 0.1 mg BSA (Fisher Biotec, Australia), 0.4 µM of each primer, and 0.25 µL of Taq DNA polymerase (Fisher Biotec, Australia). The PCR conditions were as follows: initial denaturation at 95°C for 5 minutes, followed by 50 cycles of 95°C for 30 s, 50°C for 30 s, 72°C for 30 s, and a final extension at 72°C for 10 minutes (Corbett Research, NSW, Aus). The 16S PCR was carried out in 25 µL total volume including 4 µL of template DNA, and the same dNTP, primer and buffer concentrations, but with 2.5 mM MgCl2, 0.4 mg BSA, and 0.25 µL of AmpliTaq Gold DNA polymerase (ABI) instead. The PCR conditions included: initial denaturation at 95°C for 5 minutes, followed by 40 cycles of 95°C for 30 s, 54°C 30 s, 72°C for 30 s, and a final extension at 72°C for 10 minutes (Corbett Research, NSW, Aus).
All PCR amplicons were double purified using the Agencourt AMPure XP Bead PCR Purification protocol (Beckman Coulter Genomics, MA, USA). The purified PCR amplicons were then electrophoresed together on the same 2% agarose gel to confirm the presence of the amplicons and to allow estimates of DNA concentration to be made based on comparisons between band intensity, prior to approximate equimolar amplicon pooling for emulsion PCR.
To achieve the desired bead∶template ratio, pooled PCR amplicons were quantified using a synthetic 200 bp oligonucleotide standard (of known molarity) with the Roche A and B primers engineered at either end [75]. Quantitative PCR on both the standard and the pooled library was required to approximate the optimal bead∶template ratio. The Roche GS Junior run set up included an emulsion PCR step, bead recovery, and the sequencing run. All of these procedures were carried out according to the Roche GS Junior protocols (http://www.454.com).
The sequencing output Fasta (.fna) and Quality (.qual) formatted files were processed using the following applications. Reads were quality trimmed using BARTAB [76] with a minimum acceptable quality score of 20, averaged over a window size of five bases, then separated into sample batches using a map file containing sample and primer-MID tag information. A non-redundant data set was also generated for each sample. The batched sample read primer and MID tag sequences were masked with the cross_match application [77], for minimum match length of 12 and minimum score of 20, then trimmed using trimseq [78]. An alternative means of data sorting was also employed and involved using the “separate by barcode” and primer trim feature in Geneious (v5.5). Once deconvoluted (based on MID tags) each batch of reads was searched using BLASTn version 2.2.23 [79] with a gap penalties existence of five and extension of two. The low complexity filter option was set to false, and the number of hits was limited to 100 and an expected alignment value less than 1e-10. The BLASTn search was against the National Centre for Biotechnology Information (NCBI) GenBank nucleotide NR database [80], containing all GenBank, EMBL, DDBJ and PDB sequences, a total number of 13,504,325 database sequence entries. This dataset contained no EST, STS, GSS, environmental samples or phase 0, 1 or 2 HTGS sequences, database posted date was Oct 6, 2010 5:44 PM. This pipeline was automated in an Internet-based bioinformatics workflow environment, YABI (https://ccg.murdoch.edu.au/yabi/). The resultant BLAST files were imported into the program MEtaGenome ANalyzer (MEGAN version 4.62.1) [47] for taxonomic analysis and assignment of amplicon plant and animal sequence data, using the following lowest common ancestor parameters: min score of 65, top percent of 5, and min support of 1. To compare the MEGAN assignments with other algorithms we conducted a SAP analysis [48] on a subset of data from TCM-016 where Bayesian trees were constructed from an alignment of at least 30 homologous sequences. QIIME [49] analysis was also implemented. However establishing a valid reference alignment file proved difficult for the trnL of some of the TCM taxa.
Data described herein is available in a processed and annotated form from Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.8ps58rp2. Alternatively in its raw form from the short read archive – accession number SRA047476.
|
10.1371/journal.pgen.1004300 | Heterogeneity in the Frequency and Characteristics of Homologous Recombination in Pneumococcal Evolution | The bacterium Streptococcus pneumoniae (pneumococcus) is one of the most important human bacterial pathogens, and a leading cause of morbidity and mortality worldwide. The pneumococcus is also known for undergoing extensive homologous recombination via transformation with exogenous DNA. It has been shown that recombination has a major impact on the evolution of the pathogen, including acquisition of antibiotic resistance and serotype-switching. Nevertheless, the mechanism and the rates of recombination in an epidemiological context remain poorly understood. Here, we proposed several mathematical models to describe the rate and size of recombination in the evolutionary history of two very distinct pneumococcal lineages, PMEN1 and CC180. We found that, in both lineages, the process of homologous recombination was best described by a heterogeneous model of recombination with single, short, frequent replacements, which we call micro-recombinations, and rarer, multi-fragment, saltational replacements, which we call macro-recombinations. Macro-recombination was associated with major phenotypic changes, including serotype-switching events, and thus was a major driver of the diversification of the pathogen. We critically evaluate biological and epidemiological processes that could give rise to the micro-recombination and macro-recombination processes.
| Streptococcus pneumoniae, a bacterium commonly carried asymptomatically by children, is a major cause of diseases such as pneumonia and meningitis. The species is genetically diverse and is known to frequently undergo the remarkable process of transformation via homologous recombination. In this process, the bacterial cell incorporates DNA from other, closely related bacteria into its own genome, which can result in the development of antibiotic resistance or allow cells to evade vaccines. Therefore it is important to quantify the impact of this process on the evolution of S. pneumoniae to understand how quickly the species can respond to the introduction of such clinical interventions. In this study we followed the recombination process by studying the evolution of two important and very different lineages of S. pneumoniae, PMEN1 and CC180, using newly available population genomic data. We found that pneumococcus evolves via two distinct processes that we term micro- and macro-recombination. Micro-recombination led to acquisition of single, short DNA fragments, while macro-recombination tended to incorporate multiple, long DNA fragments. Interestingly, macro-recombination was associated with major phenotypic changes. We argue that greater insight into the adaptive role of recombination in pneumococcus requires a good understanding of both rates of homologous recombination and population dynamics of the bacterium in natural populations.
| The evolution of many bacterial species is largely driven by horizontal exchange of sequence. Often, this can be attributed to the movement of autonomously mobile genetic elements (MGEs). Many of those are able to insert into the host chromosome through site-specific recombination mediated by an integrase. However, in ‘naturally’ transformable species that possess a competence system, exogenous DNA can be imported from the environment and integrated into the chromosome through homologous recombination (HR). This process was first discovered in Streptococcus pneumoniae (the pneumococcus), representing some of the earliest work on molecular genetics [1]. Initially, recombination was considered by many microbiologists to be interesting but rare. However, later population-based studies demonstrated that it can have a quantifiable impact on population genetic structure of many bacteria, including S. pneumoniae [2]–[4]. Additionally, as this mechanism only requires that the acquired DNA is homologous at the ends, recombination allows for the cassette-like transfer of highly variable genes, such as those that encode for the pneumococcal capsule [5], [6], in a process originally defined as ‘homology-directed illegitimate recombination’ [7]. This has important clinical consequences, as this exchange of sequence has played a crucial role in the development of pneumococcal antibiotic resistance [8], as well as the ‘switching’ of capsule types that can result in vaccine escape [9], [10].
The rate at which the recombination process occurs is of importance when considering the adaptation of the bacterium to clinical interventions. The simplest null expectation is that HR is a homogeneous process across the species. However, recent findings suggest that homogeneity of recombination is unlikely to capture the dynamics of horizontal sequence exchange in pneumococci. In particular, heterogeneity has been observed in the rates at which different genotypes accumulate sequence diversity through HR. Analysis of multilocus sequence typing data identified a subset of ‘hyper-recombinant’ pneumococci that were more likely to be resistant to a number of antibiotics [11]; similarly, comparison of lineages within a single population found significant variation in the observed rate of HR [12]. Second, in vitro work has found that the frequency of recombination events occurring across the genome in isogenic recipient bacteria varies with the concentration of donor DNA, suggesting the environment is likely to influence the process of sequence transfer [13]. Similarly, extensive exchanges between pneumococci over short time periods have also been observed in clinical isolates, sometimes with important phenotypic consequences [14]–[16]. Third, variation has been observed in the rate at which pneumococci undergo transformation in experimental systems [17], [18]. Therefore more detailed quantification of the observed contribution of HR will be invaluable in defining and understanding the behaviour of distinct lineages under different conditions. This in turn should help us understand how recombination contributes to the overall rate of diversification, and how it drives adaptive changes in pneumococcal populations.
The opportunity for such an analysis is presented by the recent whole genome sequencing of two international collections representing contrasting pneumococcal genotypes. The first is a set of 241 pneumococcal genomes of the recently emerged pandemic multidrug resistant lineage, PMEN1 [19]. This lineage appears to have originated in Europe in 1970s, and in the following decades spread rapidly across the world. The ancestral serotype of this lineage, serotype 23F, has switched to new capsules by HR which have resulted in its evasion of the 7-valent vaccine introduced in the early 2000s. The second lineage is a set of serotype 3 isolates belonging to clonal complex 180 (CC180) [20]. Serotype 3, which causes disease associated with high levels of mortality, has been recently included in the expanded 13-valent conjugate vaccine formulation. The CC180 lineage appears to be older than PMEN1, yet there is little evidence of it having undergone homologous recombination in recent decades, with the consequence that it is generally susceptible to antibiotics and has not altered its serotype. Hence these two genotypes, PMEN1 and CC180, are highly distinct both in terms of their phenotypes and evolutionary dynamics.
This work describes the fitting of different mathematical models of sequence exchange to the HR identified in the PMEN1 and CC180 datasets in order to identify and characterise and heterogeneity evident in the process. This resulted in the identification of two different classes of HR in both lineages: micro-recombination and macro-recombination. Potential underlying mechanistic explanations for this observation, and the implications for bacterial evolution, are discussed.
In this section we give a short summary of the methods used here, including the datasets used, the approach and mathematical models. The full description, including the notation used and the derivation of the models, is given in supplementary Text S1.
The analysis presented here is based on the inference of individual HR events, as previously described by Croucher et al. [19]. Briefly, this approach identifies independent HR events as clusters of SNPs in a genealogy reconstructed from whole genome alignments. Removal of those events allows to establish a clonal tree based on vertical transmission of SNPs. The inference for the PMEN1 lineage was based on an alignment of sequences, resulting in a genealogy with branches and homologous recombinations, whereas the inference for the CC180 lineage was based on an alignment of sequences, resulting in a genealogy with branches and homologous recombinations.
Let label the branches, and let be the number of HR events assigned to branch , such that . For a given branch , let label the recombination events, and let be the length of genetic tract, in DNA base pairs, replaced by the HR event. We define the recombination rates in our models as rates per unit of branch length. Thus, their interpretation depends on the chosen measure of branch length. Since our model structure is generic with respect to this choice, by default the branch length is measured by years estimated using a dated genealogy based on a relaxed molecular clocked estimated using Bayesian methods. (The results for alternative branch lengths are given in Tables 4–5, Figures 8–9 and Text S2.) We thus use a statistical modelling approach to explain the number and size of HR events on a branch of length given the genealogy of a lineage.
We use a modelling approach to test whether recombination in S. pneumoniae is heterogeneous with regard to its rate or length distribution. Four models were devised to account for patterns observed in the data: (i) recombination is homogeneous in frequency and in size (Model 1); (ii) recombination is heterogeneous in frequency or in size, with heterogeneity modelled as deviation from the null model 1 (Model 2); (iii) recombination is heterogeneous in frequency and size, and is modelled by two independent and homogeneous processes of recombination with different frequency and size: micro-recombination and macro-recombination (Model 3); and (iv) recombination is heterogeneous in frequency and size, as in model 3, but the heterogeneity in frequency is independent from the heterogeneity in size (Model 4).
The models were fitted by the maximum likelihood method, namely maximising the log-likelihood function given in Text S1. This was done using optimization functions NMaximize or FindMaxiumum in Mathematica 8.0. The comparison between four different models was performed using the Akaike's Information Criterion, adjusted for finite degrees of freedom (AICc). We considered one model to be a better fit than another when the difference in AICc was less than 10 (). The best model was chosen as the one with the lowest value of . If multiple models were the best fit to the data, the model with the smallest number of parameters was chosen as the best by the rule of maximum parsimony. Goodness of fit was determined by verifying the ability of the model to replicate the data under re-simulation. To that end, marginal distributions of frequency and size of the simulations were compared to the equivalent marginal distributions of the data (see Results).
The details of the simulations are described in Text S3. In brief, an ancestral sequence of S. pneumoniae was chosen as the earliest isolate of PMEN1 known [19], [22]. A forward, discrete-time simulation was designed to simulate the evolution of the lineage, including diversification through recombination simulated through incorporating homologous sequence from other publically available pneumococcal genomes. We assumed that at every time step the sequence acquired a single base substitution, and could diversify into two independently diversifying lineages with a constant probability . Each sequence also had a probability of being sampled at each timestep, after which it stopped evolving. The simulation was stopped when the population reached a maximal number of sequences, . At each timestep, recombination occurred as prespecified by one of the four models: A, B, C or D. In Model A, recombination occurred homogeneously across the genome, with lengths of recombinations following a geometric distribution. In Model B, heterogeneity (micro/macro-recombination) was introduced in frequency but not the size. In Model C, heterogeneities in both frequency and size were correlated, as described in Model 3 above. In Model D, heterogeneity was also introduced in both frequency and size but the two were treated as independent variables for each recombination. Each model was run three times, giving 12 simulations overall.
To study the process of HR in the evolutionary history of the two lineages, PMEN1 and CC180, we fitted mathematical models which describe how recombination events are distributed along the branches of the evolutionary tree of each lineage of S. pneumoniae. The procedure of model fitting is described in detail in Text S1. The phylogenies of both lineages have been constructed as described previously in [19], [20] based on vertically inherited point mutations, and were shown to be highly consistent with a molecular clock. Recombination events were reconstructed such that they were associated with particular branches of the phylogeny [19]. To remove events that may have been introduced through the movement of MGEs in PMEN1, rather than being mediated by HR, any events affecting the prophage remnant, prophage MM1-2008 or ICE Sp23FST81 were not considered in this analysis [22]. Likewise, for CC180, these MGEs included the OXC141 prophage locus and a single putative integrative and conjugative element (ICE) [20]. The distribution of recombination events on the phylogenetic trees of both lineages is summarised in Fig. 2.
The simplest model considered is that recombination events occur as a homogeneous point Poisson process through time with rate , so that the number of events occurring on a genealogical branch of length is Poisson distributed with mean , and that event sizes are geometrically distributed, with the mean length of genetic tract replaced by recombination for each event being base pairs of DNA (see Fig. 1 and Methods). This model failed to capture clear heterogeneities in both the rate and size of events in PMEN1 (Fig. 3A–C & Table 1), and the same was true for the CC180 lineage (Fig. 4A–C & Table 2).
A standard way to empirically describe heterogeneity is to quantify over-dispersion of the distribution of interest. To quantify heterogeneity in frequency and size in both lineages, we extended the approach in model 1. The extension of Poisson and geometric distribution is in both cases a negative binomial distribution with parameter , which reduces to a geometric distribution for and to Poisson for very large values of (see Fig. 1). A model based on a negative binomial distribution of events per branch with mean and dispersion coefficient , and a negative binomial distribution of event sizes with mean bp and dispersion coefficient fit the data much better than the homogeneous, Poisson-based model for the PMEN1 dataset (; Fig. 3D–F & Table 1) and also for the CC180 dataset (; Fig. 4D–F & Table 2). This demonstrates that both the recombination rate and recombination event size are heterogeneous processes, but gives little insight into the potential mechanisms generating heterogeneity.
Heterogeneity in the recombination rate suggests that recombination sometimes occurs in discrete saltations rather than at a homogeneous rate. We further observed a correlation between the frequency of recombination events and their size (Fig. 2C and 2F). We thus modelled the recombination process by a mixture of two, homogeneous recombination processes. The first process, which we refer to as micro-recombination, leads to single small replacements. The second process, which we refer to as macro-recombination, leads to multiple synchronous or near-synchronous larger replacements. We assumed that the micro-recombination process is described by the same parameters and as in the null model; the macro-recombination process occurs at rate , in which multiple tracts of DNA are incorporated into the genome by HR simultaneously (or at least in a short period of time compared to the genealogical branching process, so that these end up assigned to a single phylogenetic branch). We model the number of gene segments incorporated per macro-recombination event by a Poisson distribution with mean , and the event sizes are geometrically distributed with mean length of genetic tract replaced by recombination for each event being bp (see Fig. 1). In this model, the heterogeneity in rates is generated dynamically through the process of near-simultaneous recombination events, but this model alone does not generate excess heterogeneity in the size distribution of recombination event. The mixture model 3 provided a much better fit than the homogeneous model 1 for both PMEN1 lineage and CC180 lineage ( and , respectively). It also provided a better fit than the heterogeneous model 2 ( and ), although results of comparing non-mechanistic descriptions of heterogeneity (Model 2) to mechanistic models (Model 3) should be interpreted with caution, since mechanistic models are likely to be more useful even for equivalent goodness of fit. (See also Figures 3G–I and 4G–I, Tables 1, 2 and 3.)
A key property of the mixture model (Model 3) is that it generates correlation between the rate of recombination and the size of recombination events, since macro-recombination events, when they occur, are simultaneously larger and more numerous. To test whether this correlation was supported by the data, we compared the mixture model to a model identical in every respect, except for this correlation between rate and size (the uncorrelated mixture model 4). The resulting model fitted the data less well than the mixture model, with for PMEN1 data (Fig. 3J–L & Table 1) and for CC180 data (Fig. 4J–L & Table 2).
In summary, the mechanistic mixture model 3 fit to the data well and generated novel mechanistic insight. These results were not dependent on the units used to measure branch length (see Methods and Text S2). Maximum likelihood estimates of the parameters and univariate 95% confidence intervals are given in Table 3. We then used this best fit model to determine the probability that each of the recombination events was generated either by micro-recombination or by macro-recombination. We found that of 615 events detected in PMEN1 lineage, 136 were likely to have been generated by micro-recombination, and 389 were likely to have been generated by macro-recombination, with the remainder indeterminate. In CC180 lineage, of 79 events, 14 were likely to have been generated by micro-recombination, and 64 were likely to have been generated by macro-recombination, with only one event indeterminate. The location of each event along the pneumococcal genome as well as in the inferred phylogeny of PMEN1 and CC180 lineage is shown in Figure 5. This figure shows the heterogeneity of recombination in the phylogenies of both lineages, where certain branches exhibit multiple, long macro-recombinations, whereas short, micro-recombinations tend to be more randomly distributed. This can also be seen in supplementary Figures 10 and 11 in Text S2, where an alternative distribution of recombination events in both lineages (i.e., all independent recombination events along the genome sorted by branch length) is shown. Finally, the distribution of micro- and macro-recombination events as a function of their length and the inferred number of SNPs is given in Figure 6. The figure shows that the inferred SNP density of micro- and macro-recombinations varies by approximately one order of magnitude, suggesting that the actual rate of micro-recombination may be considerably higher than that detectable through these data (but see Discussion).
In PMEN1, 10 serotype-switching events were observed [19] (i.e., those which induced a change from the serotype 23F to a different one), and all those events were found to be with 100% posterior probability likely to have been the result of macro-recombination. More generally, to examine whether recombinations at major antigen loci are likely macro-recombinations, we counted the number of recombinations spanning or overlapping five major antigen loci in PMEN1 (pspA, capsule biosynthesis locus, or cps, pclA, psrP and pspC) and three major antigen loci in CC180 (pspA, cps, and pspC). Of 171 such detected recombinations in PMEN1, 93 were likely to have been generated by macro-recombination. By contrast, in CC180 only 4 recombinations at major antigens were found, however all 4 of them were likely to have been generated by macro-recombination.
To assess our method of detecting heterogeneity of recombination in the genetic data we designed a simulation framework where we evolved a pneumococcal lineage over time with four prespecified mechanisms of recombination, and examined how well we can distinguish between those mechanisms (see Methods and Text S3). Specifically, we designed analyses in which the PMEN1 reference genome diversified into a sample of related sequences through discrete time-steps as specified by one of four different simulation frameworks (Models A–D). We then reconstructed the evolutionary history of the lineage, with recombination events mapped onto the phylogeny, as described above and in [19]. We next fitted our four models of recombination (Fig. 1) to assess which of them best explains the underlying mechanism of diversification (see Tables 6–7 in Text S3). In the first simulation (A), recombination was simulated as a homogeneous process, and the homogeneous model 1 was the best fit. In the second simulation (B), the distinction between micro-recombination and macro-recombination was introduced but only based on frequency and not size, and in these cases model 3 was the best fit to the data. However, there was no significant difference in the size distributions between the two modes of recombination, contrasting with the fits to the genomic data. In the third simulation (C), a full mixture model of micro- and macro recombination was considered, and again model 3 was the best fit, with the likelihood of each model fits being of the same order of magnitude as in PMEN1 and CC180 data. Finally, in the fourth simulation (D), an uncorrelated mixture model was assumed with independent heterogeneity in frequency and size. In this case, in two runs there was no significant difference in the fit of model 3 and 4, while in the third model 4 was a much better fit to data than model 3. These simulations thus demonstrate that the observation of model 3 fitting the genomic data best, with a dramatic difference in lengths between the micro- and macro-recombinations, is unlikely to be an artefact of the method used to detect recombination, or the models' formulation
We next investigated whether the obtained results can explain recent observations of recombination in the pneumococcus using whole genome data. The near-simultaneous import of multiple fragments through transformation has previously been observed between a donor and recipient during a chronic infection in vivo in one patient [14], and also inferred through reconstructing the history of another lineage, sequence type 695 [15]. In the study by Hiller and colleagues [14], 16 recombination events varying in size from 0.4 kb to 235 kb (mean of 15 kb) were unidirectionally transferred from one donor strain into a recipient strain during an infection followed over a period of seven months. The observation that, in each case, multiple long recombinations had occurred over a defined short period suggested these examples might represent clear examples of the macro-recombination process. We found the size distribution of macro-recombinations to be in accordance with the one observed by Hiller et al. for both PMEN1 (see Fig. 7A) and CC180 lineage (see Fig. 7B).
On the other hand, the study by Golubchik et al. identified 53 recombination fragments in 5 vaccine escape recombinant lineages, ranging in size from 0.4 kb to 90 kb (mean of 10 kb). Although the distribution of recombination sizes inferred by this analysis of re-sequencing data did not resemble any of the distributions defined by the models of recombination presented here, it nevertheless suggests a strikingly heterogeneous recombination process (see Fig. 7C and 7D). A more formal approach would be needed to determine whether this is due to an actual recombination heterogeneity or due to another factor like the method used to infer recombination, or vaccine-induced selection (see also Discussion).
Finally, it has been demonstrated that multiple fragments of DNA can be imported by a member of the PMEN1 lineage during a single period of competence for transformation under controlled conditions [13]. While the overall distribution of sizes observed was similar to that reconstructed as happening during the lineage's diversification, there was less variation in the range of detected sizes. The discrepancy between the size distributions from the transformation experiment and the one observed in the PMEN1 lineage (see Fig. 7E) points to some interesting questions about varying conditions under which pneumococci undergo recombination during their evolution (see Discussion). Perhaps unsurprisingly, the predicted size distribution of the CC180 lineage was even less consistent with the distribution of recombinations from the in vitro experiment (see Fig. 7F).
One hypothesis that could explain the observed difference between micro- and macro-recombination could be the effect of mismatch repair (MMR; see also Discussion). MMR inhibits the acquisition of polymorphisms through transformation, but in the pneumococcus becomes saturated upon the import of around 150 SNPs [23], [24]. Thus micro-recombinations could be acquired under the constraint of this system, whereas macro-recombination could represent the acquisition of sequence unlimited by MMR. In accordance with this hypothesis, when we divided branches of the phylogeny on the basis of the most common mechanism of recombination occurring on them, those on which micro-recombination predominated generally imported fewer than 150 substitutions in total, while those on which macro-recombination was more common typically acquired many more than this (see Figures 12–13 and Text S2). We also examined whether there were differences in the types of substitutions introduced by micro- and macro-recombination, as MMR varies in the efficiency with which is repairs different mutations. We found that macro-recombinations were enriched for ‘low efficiency' markers, which are repaired most effectively by MMR both in PMEN1 (), and in CC180 (). Interestingly, no association between the type of marker and the type of recombination was observed in the simulated pneumococcal sequences with preassumed micro- and macro-recombination mechanism (see Table 8 and Text S2).
Our analysis shows that both analysed lineages of Streptococcus pneumoniae, the multi-drug resistant PMEN1 and the older but less diverse CC180, have likely evolved under two distinct homologous recombination processes. The first process, which we call micro-recombination, occurred at a homogeneous clock-like rate and gave rise to isolated small genetic replacements. The second process, which we call macro-recombination, was more erratic, giving rise to large, multiple synchronous (or near-synchronous) replacements. While in PMEN1 we found both micro- and macro-recombinations to have occurred at a similar rate (every 17 years), in the less rapidly diversifying CC180 lineage micro-recombination was more frequent than macro-recombination (once in 340 years vs. once in 770 years). Overall, recombination was much more heterogeneous in CC180. Furthermore, the difference in sizes between micro- and macro-recombination was found to be greater in CC180 (0.03 kb vs. 14 kb) than in PMEN1 (0.6 kb vs. 9 kb). Finally, the number of simultaneous recombinations imported during macro-event was smaller in PMEN1 than in CC180 (2.3 vs. 15). The best fit parameters, together with the 95% confidence intervals, are summarised in Table 3.
The principal caveat in this analysis is that it is dependent on the correct identification of both the genealogy and the recombinations in the original analysis of the PMEN1 and CC180 lineages [19], [20]. The main evidence given for the correct identification of the recombinations is that their removal from the set of base substitutions used to construct the phylogeny results an improved ability to detect evidence of a molecular clock at a rate similar to other bacteria that do not undergo frequent homologous recombination [19], [25], the length distribution of putative events is similar to that detected experimentally [13], and that recombination events that can be inferred from phenotypic data (e.g., serotype switches) are predicted at the correct locus on the expected branch of the tree [12], [19]. However, we note that there is an inherent bias in the method described by Croucher et al., shared with other methods that use SNP density to detect recombination (e.g., maximum Chi-square method, ClonalFrame [21]), in that it is prone to missing short recombination events that happen to bring in few SNPs into the genome. Nonetheless, such events have a relatively small effect on estimates of branch length, and therefore estimates of the molecular clock rate. However, such bias means that we have likely under-estimated the rate of micro-recombination. This is best illustrated by comparing SNP density to the observed size of the recombination (Figure 6). The observed negative correlation between SNP density and recombination size (Spearman's rank correlation: , for PMEN1 and , for CC180) is likely the result of the detection bias described above, and this suggests that we may lack the sensitivity to accurately quantify the rate of micro-recombination events. Simulations of the heterogeneity suggest that the actual rate of micro-recombination is likely to be roughly three times the estimated rate. Correspondingly, we found that the methods employed in this study were able to correctly identify the underlying model of evolution when simulations were performed under different models of diversification. This suggests that our observations are unlikely to be an artefact of the method used to detect recombination.
The presented analysis provides a quantitative model that could potentially explain other observations of recombination in the pneumococcus using whole genome data. The near-simultaneous import of multiple fragments through transformation has been observed previously in in vivo [14], [15] and in vitro studies [13]. We found that the micro/macro-recombination process could be consistent with size distributions of recombinations in some patient-derived sequences (cf. Fig. 7). However, there is weak evidence that this happens in the case of transformation in vitro. Therefore the observation of these two different types of recombination requires an explanation that can link the differences in properties and kinetics. It could be that genetic transformation through the competence system is only responsible for recombination through one of the modes, like micro-recombination, while other forms of bacterial “sex”, like conjugation or transduction, would lead to the acquisition of long stretches of DNA associated with macro-recombination. Conjugation has been observed to cause extensive sequence transfer in other streptococci, which would be consistent with this hypothesis conjugative transfer can result in multiple events if multiple conjugative origins are involved [26]. However, these exchanges are associated with ori sequences from conjugative elements, and therefore result in more regular recombination boundaries than are observed for the macro recombination events in this analysis [27]. Similarly, general transduction of sequence can import large DNA fragments of variable lengths, but typically only one can be packaged into a virion. As such mispackaging events are rare, this does not provide a likely explanation for the near-simultaneous import of multiple fragments [28].
Another potential explanation of the difference between micro- and macro-recombination may be how stretches of DNA are processed within the cell. For example, the recently identified competence-specific DNA-binding protein SsbB has been found capable of storing about 1.15 Mb of DNA imported by the competence system [27]. As the expression of this protein varies according to regulatory processes, it could play an important role in controlling the properties of recombination. However, given the comparatively homogeneous length distribution of recombinations observed in experimental transformation of the pneumococcus, it seems likely that extracellular degradation or intracellular processing are not the best candidates to explain the observed heterogeneity.
Hence it seems more likely that the observed dynamics represent transformation behaving in two distinct modes. One known threshold that could explain the variation is saturation of repair systems. MMR inhibits the acquisition of polymorphisms through transformation, but in the pneumococcus becomes saturated upon the import of around 150 SNPs [23], [24]. Here we found moderate but significant evidence for this hypothesis, which would suggest that it is the extent and type of DNA imported that triggers the switch between the two types of exchange. In the PMEN1 dataset, each homologous recombination imports a mean of 70 substitutions (116 substitutions for CC180), and in vitro experiments have demonstrated that multiple fragments can be imported simultaneously. Therefore the availability of high concentrations of divergent DNA, as observed in pneumococcal biofilms [29], or a state of ‘hyper-competence’, in which cells imported DNA more readily than normal, would seem likely to saturate the MMR system and potentially trigger the conditions required for macro-recombination.
The idea of the emergence of micro-recombination and macro-recombination via saturation of the MMR has the advantage that it is consistent with the observed positive correlation between frequency and size of recombinations (cf. Fig. 2C and 2F). Many macro-recombinations found in this study are considerably larger than any individual segment of donated sequence acquired by S. pneumoniae in vitro. This is likely to reflect the algorithm employed in the analysis of pneumococcal genomes, which clusters together nearby transformation events that originate from the same imported strand of DNA [13]. Therefore, integrating a larger number of imported sequence segments into the chromosome can both result in a greater number of distinct recombinations, and generate more extensive ‘mosaic’ events that would be reflected by an increase in the length of the overall transformation event in this analysis. Hence if a mechanism like MMR becomes saturated, it might not only result in more acquired recombinations but also in transformation of larger mosaic segments, resulting in a simple mechanistic link between frequency and size of recombinations. Interestingly, in vitro transformation experiments of pneumococcus, despite investigating transformation at two very different concentrations of exogenous DNA, did not find strong evidence for two distinct mechanisms of recombination [13]. This indicates that the observed difference may represent other environmental factors that affect the regulation of systems such as MMR.
It is also important to consider that the observed distribution of sequence is also the consequence of selection, which could be an alternative explanation for the observed heterogeneity. However, such a selection pressure would have to be highly generic to account for such a genome-wide phenomenon. One potential pressure that affects multiple loci, in particular several affected by a high density of recombinations, is immune-driven selection. Loci which are most likely to be under selective pressure of the immune system have been shown to be recombination hotspots [19]. As this selection is likely to be diversifying, it is conceivable that longer recombinations at these loci, inducing greater phenotypic changes, are under positive selection, and are thus more frequently observed. However, the mixture model 3 remains the best fit even after those events have been removed from the dataset (see Table 9–10 and Text S2). Therefore, we conclude that, even though immune selection is likely to play a role in shaping the distribution of recombination events in the pneumococcal genome, it is unlikely to explain the observed heterogeneity of homologous recombination in S. pneumoniae.
Another process that may skew the pattern of observed recombinations is the non-systematic nature of the isolate collections used in the original analyses. Two analyses were performed to assess the potential for biased sampling to affect the conclusions: the first excluded all isolates from the extensively sampled South African collection, while the second excluded all isolates serotyped as 19A to rule out potential vaccine induced selective pressure. In both cases, the results were qualitatively the same (Table 11 and Text S2).
In summary, we have firmly demonstrated that homologous recombination is heterogeneous, and found that the heterogeneity shows evidence of two modes of action, which we term micro- and macro-recombination. We have also found that saturation of the mismatch repair system is the most likely mechanism for inducing macro-recombination.
From a whole population survey, it has been observed that total homologous recombination rates vary substantially between pneumococcal lineages [12], and that an increased propensity for recombination is associated with increased antibiotic resistance [11]. Given this observation, it is particularly interesting that the two lineages studied here, that are at the opposite extremes in terms of their phenotype and evolutionary history, are both characterised by a highly heterogeneous recombination process. Furthermore, the aggregate recombination distribution sizes appear quite relatively consistent across different pneumococcal genotypes [12]. This all suggests that the micro- and macro-recombination are likely to play a role across the entire pneumococcal species. Based on the results presented here, it seems that micro-recombination is the more frequent process, whereas macro-recombination is likely to be the main driver of the bacterium's diversification.
How generally applicable these models are to the evolution of other species, and their relevance to wider questions about the evolution of homologous recombination itself [30], can be addressed as more genomic datasets become available.
|
10.1371/journal.ppat.1005663 | Sequential Dysfunction and Progressive Depletion of Candida albicans-Specific CD4 T Cell Response in HIV-1 Infection | Loss of immune control over opportunistic infections can occur at different stages of HIV-1 (HIV) disease, among which mucosal candidiasis caused by the fungal pathogen Candida albicans (C. albicans) is one of the early and common manifestations in HIV-infected human subjects. The underlying immunological basis is not well defined. We have previously shown that compared to cytomegalovirus (CMV)-specific CD4 cells, C. albicans-specific CD4 T cells are highly permissive to HIV in vitro. Here, based on an antiretroviral treatment (ART) naïve HIV infection cohort (RV21), we investigated longitudinally the impact of HIV on C. albicans- and CMV-specific CD4 T-cell immunity in vivo. We found a sequential dysfunction and preferential depletion for C. albicans-specific CD4 T cell response during progressive HIV infection. Compared to Th1 (IFN-γ, MIP-1β) functional subsets, the Th17 functional subsets (IL-17, IL-22) of C. albicans-specific CD4 T cells were more permissive to HIV in vitro and impaired earlier in HIV-infected subjects. Infection history analysis showed that C. albicans-specific CD4 T cells were more susceptible to HIV in vivo, harboring modestly but significantly higher levels of HIV DNA, than CMV-specific CD4 T cells. Longitudinal analysis of HIV-infected individuals with ongoing CD4 depletion demonstrated that C. albicans-specific CD4 T-cell response was preferentially and progressively depleted. Taken together, these data suggest a potential mechanism for earlier loss of immune control over mucosal candidiasis in HIV-infected patients and provide new insights into pathogen-specific immune failure in AIDS pathogenesis.
| HIV infection is closely associated with enhanced host susceptibility to various opportunistic infections (OIs), among which mucosal candidiasis caused by the fungal pathogen Candida albicans (C. albicans) is an early and common manifestation. Even in the era of effective ART, mucosal candidiasis is still a clinically relevant presentation in HIV-infected patients. The underlying mechanisms are not well defined. CD4-mediated immunity is the major host defense mechanism against C. albicans. We here investigated a group of ART naïve, HIV-infected human subjects and examined longitudinally the impact of HIV on C. albicans-specific CD4 T-cell immunity as compared to CD4 T-cell immunity specific for CMV, another opportunistic pathogen that usually does not cause active disease in early HIV infection. We found that C. albicans-specific CD4 T cells were more susceptible to HIV in vivo and were preferentially depleted in progressive HIV-infected individuals as compared to CMV-specific CD4 T cells. Of importance, we also found that in these HIV-infected subjects C. albicans-specific CD4 T cell response manifested a sequential dysfunction with earlier impairment of Th17, but not Th1, functions. Our study suggests an immunological basis that helps explain the earlier and more common onsets of mucosal candidiasis in progressive HIV-infected patients.
| Untreated HIV infection causes progressive depletion of human CD4 T cells, leading to impaired cellular immunity, enhanced susceptibility to opportunistic infections (OIs) and development of acquired immunodeficiency syndrome (AIDS) [1–3]. Although the loss of immune control over OIs is known to be generally associated with overall reduction in CD4 T cells, HIV cohort studies have found that OI reactivation can occur at different stages of HIV disease and is not strictly associated with total CD4 loss [4–6]. For instance, while the opportunistic pathogen Mycobacterium tuberculosis (MTB) can cause active disease relatively early during HIV infection [7], cytomegalovirus (CMV) infection rarely causes evident diseases at early stage [8, 9]. These observations have suggested that host immunity specific for opportunistic pathogens may be impaired or lost at different stages of HIV disease [10–12]. In support, an important study by Geldmacher et al. demonstrated that compared to CMV, MTB-specific CD4 T cells are preferentially infected and depleted in HIV-infected human subjects [10, 13].
Mucosal candidiasis, predominantly caused by the commensal fungal organism Candida albicans (C. albicans), is one of the most common and earliest manifestations in HIV-infected subjects [14, 15]. In immune competent humans, C. albicans can be readily detected without overt signs of clinical disease [16]. However, under immune compromised conditions such as in AIDS patients, C. albicans can quickly cause active infections in multiple tissues, including oral mucosa [17]. Evidence has shown that about 50–90% of HIV-infected individuals could manifest an episode of oral candidiasis during their progression to AIDS [18, 19]. Even with the introduction of potent antiretroviral treatment (ART), oropharyngeal and esophageal candidiasis are still the two clinically relevant presentations in HIV-infected patients [20]. The underlying immunological basis for early and profound onsets of pathogenic C. albicans infections in HIV-infected individuals is not fully defined.
C. albicans exposure induces strong cellular immunity, as evidenced by the skin-test reactivity and in vitro lymphocyte proliferative response [21, 22]. Majority of evidence obtained so far from animal models and human studies has suggested CD4-mediated cellular immunity as the predominant host defense mechanism against C. albicans infection [23–30], although involvement of specific functional facets of CD4 T-cell immunity, for instance, Th1 vs. Th17 response, has been obscure. It was initially suggested that Th1 response was the key mediator of immunity [31]. More recently, increasing evidence has indicated that Th17, but not Th1, response is critical for immune protection against mucosal candidiasis [25, 32, 33]. Importantly, in the setting of HIV infection, limited information is currently available regarding the longitudinal impact of HIV on different functional facets of anti-C. albicans CD4 T-cell immunity in HIV-infected individuals.
To explore the effect of HIV on different antigen-specific CD4 T cells, we have previously described an in vitro system, where HIV susceptibility and the associated phenotypes of antigen-specific CD4 cells can be examined [12, 34]. We have found that human C. albicans-specific CD4 T cells are highly permissive to HIV infection in vitro compared to CMV-specific CD4 T cells [12]. It remains to be determined as to how HIV affects these two groups of pathogen-specific CD4 T-cell immunity in vivo in HIV-infected subjects. RV21 is an antiretroviral treatment (ART) naïve, longitudinal HIV-infection cohort established by the U.S. Military HIV Research (MHRP) and the HIV-infected subjects enrolled in this cohort were followed up for 2 to 6 years. In the current study, we studied HIV-infected subjects in the RV21 cohort who manifested ongoing CD4 depletion. Using PBMC samples from these individuals, we comparatively examined the longitudinal impact of HIV on functional profiles and magnitudes of C. albicans- and CMV-specific CD4 T cell responses in vivo during HIV disease progression. Our data showed that there was a sequential dysfunction for C. albicans-specific CD4 T cell response with an earlier and more profound impairment of Th17-associated functions (IL-17, IL-22) in HIV infection. Further analyses identified that compared to CMV-specific CD4 T cells, C. albicans-specific CD4 T cells were more susceptible to HIV in vivo and preferentially depleted in these HIV-infected subjects.
Antigen-specific T cell responses elicited by different pathogens can be qualitatively distinct. In our previous studies [12, 34], we have reported an in vitro system for examining the susceptibility of antigen-specific human CD4 T cells to HIV infection and the associated phenotypic and functional characteristics (Fig A in S1 Appendix). We here utilized this system and first determined the functional profiles of C. albicans-specific CD4 T cells as compared to CMV-specific CD4 T cells in healthy human subjects. PBMC samples from healthy donors were labeled with CFSE, a fluorescent dye to track T cell division, and then stimulated with C. albicans or CMV antigen for 6 days, during which memory CD4 T cells underwent Ag-specific proliferation in response to stimulation. Cells were re-stimulated on day 6 for de novo cytokine synthesis. Functional profiles (IL-17, IL-22, IL-2, IFN-γ and MIP-1β) of C. albicans- or CMV-specific CD4 T cells in PBMCs were examined in CFSE-low CD4 T cells by multi-color flow cytometry (Fig A in S1 Appendix). Verification of the system has been described in previous reports [12, 34].
We found that C. albicans-specific CD4 T cells displayed a distinct functional profile from CMV-specific CD4 T cells in healthy donors (Fig 1A). Compared to CMV-specific CD4 T cells, which predominantly expressed Th1-associated cytokine IFN-γ (75.6%) and MIP-1β (67%), C. albicans-specific CD4 T cells expressed high levels of IL-17 (20.4%), IL-22 (15%) and IL-2 (63.7%), in addition to expression of IFN-γ and MIP-1β, suggesting a Th17/Th1-like phenotype for C. albicans-specific CD4 T cells in human subjects (Fig 1A). Analysis of PBMCs from multiple donors (n = 6) showed that expression of IL-17 (p<0.0001), IL-22 (p<0.001), IFN-γ (p<0.001) and MIP-1β (p<0.01) was statistically different between C. albicans- and CMV-specific CD4 T cells (Fig 1B). Poly-functional analysis showed that C. albicans-specific CD4 T cells demonstrate a more poly-functional profile and can co-express multiple cytokines compared to CMV-specific CD4 T cells (Fig B in S1 Appendix).
We also measured gene expression of Th17 and Th1 lineage-specific transcription factors, including RORC (Th17), T-bet and EOMES (Th1), in C. albicans- and CMV-specific CD4 T cells from the same donor PBMCs (Fig 1C). CFSE-low, CD4 T cells were sorted from PBMC and subjected to real-time PCR quantification. We found that while gene expression of Th1 transcription factors T-bet and EOMES was comparable between C. albicans- and CMV-specific CD4 T cells, the Th17 transcription factor RORC was expressed at significantly higher levels in C. albicans-specific CD4 T cells, further suggesting the mixed Th17/Th1-like phenotype of C. albicans-specific CD4 T cells in human subjects (Fig 1C).
Since no significant difference in T-bet and EOMES expression was observed at mRNA level between C. albicans- and CMV-specific CD4 T cells, we measured protein expression of these two transcription factors using flow cytometry and found that compared to CFSE-Hi non-specific CD4 T cells, both C. albicans- and CMV-specific CD4 T cells expressed higher levels of T-bet and EOMES, although the expression levels in CMV-specific CD4 T cells appeared to be slightly higher than those in C. albicans-specific CD4 T cells (Fig C in S1 Appendix). The results suggest that both Ag-specific CD4 T cell populations in this system manifest increased expression of Th1 transcription factors than non-specific CD4 T cells, which is in line with previous reports showing that T-bet and EOMES were readily detectable in CMV-specific CD4 T cells albeit at lower level than in their CD8 counterparts [35–37].
Based on this system, we examined HIV susceptibility of C. albicans- and CMV-specific CD4 T cells from healthy donor PBMCs and found that C. albicans-specific CD4 T cells were substantially more permissive to HIV than CMV-specific CD4 T cells in vitro (Fig 2A), a finding that was consistent with our previous report [12]. To explore whether the significant difference in HIV susceptibility between C. albicans- and CMV-specific CD4 T cells is due to higher permissiveness of C. albicans-specific CD4 T cells or enhanced protection of CMV-specific CD4 T cells, we compared their HIV susceptibility with that of non-specific total CD4 T cells that were globally stimulated with anti-CD3/CD28 (Fig D in S1 Appendix), and found that HIV infectivity in globally stimulated CD4 T cells fell into the range between C. albicans- and CMV-specific CD4 T cells (Fig D in S1 Appendix), implying that the difference in HIV susceptibility between C. albicans- and CMV-specific CD4 T cells might attribute to the combination of both. This will be further investigated subsequently.
Functional characteristics of CD4 T cells have been shown to associate with their susceptibility to HIV [10, 13]. In order to define the relationship between HIV infectivity and functional characteristics for C. albicans-specific CD4 T cells, we performed the HIV susceptibility assay as described in Fig A in S1 Appendix. Healthy donor PBMCs were CFSE-labeled and stimulated with C. albicans antigen, followed by exposure to HIV. Three days after HIV infection, cells were re-stimulated with PMA/ionomycin and subjected to comprehensive flow cytometric analysis (Fig A in S1 Appendix). Cytokine expression in activated T cells is transient and the CFSE-low, Ag-specific CD4 T cells in this system undergo days of proliferation. In order to simultaneously measure functional characteristics (cytokine production) and HIV infectivity (intracellular p24), cells were re-stimulated with the global PMA/ionomycin stimulus on day 6 for cytokine re-synthesis in T cells.
As shown in Fig 2B, by gating on CFSE-low CD4 T-cell population, we determined HIV infectivity in each functional subset of C. albicans-specific CD4 T cells by measuring co-expression of intracellular HIV p24, as an indication of productive HIV infection, with individual cytokines (Fig 2B). Number in each plot showed intracellular p24+ rate in cytokine-producing, C. albicans-specific CD4 subset. We found that the IL-22, IL-17 or IL-2 functional subsets of C. albicans-specific CD4 T cells were more susceptible to HIV as compared to those subsets expressing MIP-1β and IFN-γ (Fig 2B). We measured expression of CCR5, an important co-receptor for HIV entry, on these different functional CD4 subsets and no significant difference was observed (Fig E in S1 Appendix). Instead, we found that the higher HIV infectivity in IL-22+, IL-17+ and IL-2+ subsets appeared to be associated with their lower levels of MIP-1β co-expression (Fig F in S1 Appendix). This observation was consistent with an earlier report showing that in vitro differentiated IL-17+ CD4 T cells are more susceptible to HIV than IFN-γ+ CD4 T cells due to reduced expression of MIP-1β [38].
We noted that HIV infectivity in IFN-γ+ and MIP-1β+ subsets was also fairly high (63% and 30%, respectively) (Fig 2B). Next, we gated on the IFN-γ+ or MIP-1β+, CFSE-low CD4 subsets (Fig 2C) and found that significant fractions of IFN-γ+ (or MIP-1β+) subset co-express with IL-2 and IL-17 or IL-22 (Fig 2C and Fig G in S1 Appendix). When we further performed intracellular p24 (red dots) and cytokine (blue background) overlaying analysis, as shown in Fig 2C, we identified that HIV predominantly infected IFN-γ+ or MIP-1β+ CD4 subsets that co-expressed IL-2 and IL-17 or IL-22; the single IFN-γ- or MIP-1β-producing CD4 subsets (bottom left quadrant) demonstrated very low HIV infectivity (Fig 2C). In order to better differentiate HIV infectivity in all different functional subsets (combination of cytokine+), we performed comprehensive Boolean gating and spice analysis. As shown in Fig 2D, we identified a consistent trend that HIV infectivity (p24+%) was substantially higher in populations that express of IL-17, IL-2 and IL-22, but lower in populations that only express MIP-1β and/or IFN-γ.
CD25, the high-affinity chain of IL-2 receptor, was shown to be important for HIV infection of CD4 T cells in vitro [13, 39]. We examined expression of CD25 on C. albicans- and CMV-specific CD4 T cells and evaluated its relationship with HIV infectivity and cytokine expression in our system (Fig 2E–2G). Interestingly, the data showed that C. albicans-specific CD4 T cells expressed substantially higher level of CD25 than CMV-specific CD4 T cells (Fig 2E). Importantly, productive HIV infection was predominantly observed in CD25+ subset both C. albicans- and CMV-specific CD4 T cells (Fig 2E, top panels; Fig 2F). We also examined the impact of exogenous IL-2 on CD25 expression and HIV infectivity (Fig 2E, bottom panels). Despite recombinant IL-2 (rIL-2) induced significant increase in CD25 expression and CD4 T-cell proliferation, HIV infectivity in exogenous IL-2-treated cells were not enhanced (Fig 2E, bottom; Fig 2F), indicating that HIV infectivity is associated with endogenous IL-2-CD25 signaling. In addition, we investigated relationship between CD25 and cytokine expression in C. albicans-specific CD4 T cells. As shown in Fig 2G, CD25 predominantly co-expressed with IL-17, IL-22 and IL-2, but not IFN-γ- or MIP-1β, which is consistent with the observation about HIV infectivity in different CD4 T-cell subsets. Taken together, these data suggest that compared to CMV, C. albicans-specific CD4 T cells manifest distinct phenotypic and functional characteristics that favor productive HIV infection in these cells.
As described earlier, CD4-mediated cellular immunity is a predominant host defense mechanism for immune control of pathogenic C. albicans infection [23–27]. While Th1 response was initially thought to be the key mediator of immunity, more recent studies have supported critical role of IL-17- and IL-22-producing Th17, but not IFN-γ-producing Th1, response in protection against candidiasis [31, 32] [33]. After showing that IL-17, IL-22 and IL-2 functional subsets of C. albicans-specific CD4 T cells were more permissive to HIV in vitro, we next aimed to determine the in vivo impact of HIV on C. albicans-specific CD4 T-cell immunity and the associated functional subsets in HIV-infected individuals. To do so, we selected HIV-infected subjects in RV21 cohort who manifested ongoing CD4 depletion, which permitted us to longitudinally examine the in vivo effect of HIV on pathogen-specific CD4 cells. We identified 20 HIV-infected subjects with positive responses to both C. albicans and CMV antigens and the PBMC samples from these subjects were accordingly investigated (Table 1). To better explore the impact of HIV on pathogen-specific CD4 T-cell immunity, multiple assays were performed. Due to limited cell number for each HIV-infected subject, we appropriately allocated cell samples from these 20 subjects for different assays as detailed below.
We first used a similar method as described in Fig 1 and measured functional profiles of proliferating C. albicans-specific CD4 T cells in the HIV-infected subjects as compared to healthy donors (Fig 3). PBMCs of HIV+ subjects measured here were collected at early HIV infection when profound CD4 depletion had not occurred and C. albicans-specific response remained detectable. As shown in Fig 3A, only the CFSE-low, C. albicans-specific CD4 T cell populations were gated for analysis. Interestingly, we found that compared to C. albicans-specific CD4 T cells in control PBMC of healthy donors, which manifested strong proliferative response and normal production of all cytokines tested (IL-17, IL-22, IL-2, IFN-γ and MIP-1β), C. albicans-specific CD4 T cells from HIV-infected subjects, despite being able to proliferate at comparable levels, demonstrated a preferential impairment in Th17 functions with substantial decrease in IL-17, IL-22 and IL-2 production, while their Th1 response (expression of IFN-γ or MIP-1β) was not significantly affected (Fig 3A). We examined PBMC samples from multiple subjects (n = 7) and observed statistically significant differences for expression of IL-17 (p<0.01), IL-22 (p<0.01) and IL-2 (p<0.01), but not IFN-γ (N.S.) or MIP-1β (N.S.) between healthy donors and HIV-infected subjects (Fig 3A). Importantly, we also measured functional profile of CMV-specific CD4 T cells in these HIV-infected subjects (Fig 3B) and found that CMV-specific CD4 T cells at early HIV infection manifest a comparable functional profile with that in uninfected healthy donors; no significant reduction in expression of Th1 cytokines IFN-γ (N.S.) and MIP-1β (N.S.) was observed, although the IL-2 expression in CMV-specific CD4 T cells appeared to be impaired in HIV-infected subjects as compared to healthy subjects (p<0.01) (Fig 3B). Taken together, these data from HIV-infected subjects were consistent with the in vitro observations in Fig 2 and imply that there was a sequential dysfunction for C. albicans-specific CD4 T-cell response with earlier impairment of Th17 functions during HIV infection. Given the critical role of Th17, but not Th1, response in protection for mucosal candidiasis, this finding is potentially significant and suggests that early following HIV infection, the anti-C. albicans specific immunity might become rapidly ineffective, despite their proliferative and Th1-type responses are still readily detectable.
We have shown in Fig 2A that C. albicans-specific CD4 T cells are more susceptible to HIV infection than CMV-specific CD4 T cells in vitro [12]. In order to determine if this occurs in vivo, we examined the levels of cell-associated HIV DNA using quantitative PCR in C. albicans- and CMV-specific CD4 T cells from HIV-infected subjects [13, 40]. To do so, we used PBMC samples of HIV-infected subjects collected at early HIV infection when profound CD4 depletion had not occurred and both C. albicans- and CMV-specific CD4 T cell responses were detectable. PBMCs were CFSE-labeled and stimulated with C. albicans- or CMV-antigen for 5 days, during which HIV replication inhibitor AZT was added to cell culture to prevent potential de novo HIV replication. No HIV virus was detected in the culture supernatants after antigen stimulation and T-cell division, supporting the effectiveness of AZT in blocking possible de novo viral replication (Fig H in S1 Appendix). After stimulation, C. albicans- and CMV-specific CD4 T-cell populations from the same PBMCs were sorted based on CFSE-low (Fig 4A) and then subjected to quantification of HIV DNA. Plasmids encoding HIV Gag or GAPDH were used to generate standard curves for quantifying real copy numbers of HIV DNA in the sorted Ag-specific CD4 T cells (Fig I in S1 Appendix). Quantification of HIV DNA in the sorted cells was normalized to GAPDH and expressed as copy number/106 CD4 T cells.
As shown in Fig 4B, C. albicans-specific CD4 T cells harbored modestly but significantly higher levels of HIV DNA than CMV-specific CD4 T cells (p = 0.004). We measured PBMCs from 5 HIV-infected individuals and observed a consistent trend, although the difference for some subjects was modest (Fig 4B). Since predominant sites of HIV infection and replication are lymphoid or mucosal tissues, not peripheral blood, we speculate that differences for HIV DNA content between C. albicans- and CMV-specific CD4 T cells, when isolated from these effector sites, might be more profound.
Quantification of cell-associated HIV DNA to evaluate in vivo HIV susceptibility has been previously reported for MTB- and HIV-specific CD4 T cells [13, 40]. To gain better insights into relative HIV susceptibility of different Ag-specific CD4 T cells in vivo in infected individuals in the absence of viral suppression, we further measured additional antigens for comparison with C. albicans. Since the HIV-infected subjects examined in this study resided in the US and demonstrated very low MTB response, we were unable to directly compare C. albicans with MTB. Instead, we measured CD4 T cells specific for varicella zoster virus (VZV), another herpes virus similar to CMV, as well as CD4 T cells specific for HIV Env protein; HIV DNA in CFSE-Hi non-specific CD4 T cells from the same individuals was also compared (Fig 4C). Since cell number for each subject was limited and some subjects only responded to certain antigens, not all antigens were compared for each subject. Interestingly, we found that among the subjects investigated, C. albicans-specific CD4 T cells appeared to also harbor higher levels of HIV DNA as compared to VZV-specific (subject 1–3) and HIV Env-specific (subject 6–9) CD4 T cells (Fig 4C). Of note, HIV DNA copies in C. albicans-specific CD4 T cells varied fairly substantially among different subjects, an observation that was also reported for MTB-specific CD4 T cells [13]. Taken together, these results suggest that C. albicans-specific CD4 T cells are highly susceptible to HIV in vivo when compared to multiple other antigens.
As discussed above, compared to peripheral blood, mucosal tissues represent a preferential site for HIV infection and manifest most remarkable CD4 depletion at all stages of HIV disease [41, 42]. Integrin α4β7 is an important mucosal homing receptor, directing migration of CD4 T cells from blood to gut, and CCR6 is a marker associated with Th17 cells that contributes to their migration to mucosal tissues [43]. We examined expression of these mucosal homing receptors on C. albicans- and CMV-specific CD4 T cells from the HIV-infected subjects. Our data showed that significant fraction of C. albicans-specific CD4 T cells expressed high levels of α4β7 and CCR6, while CMV-specific CD4 T cells rarely expressed these two receptors (C. albicans vs. CMV: 70.6% to 7.7% for α4β7; 26.4% vs. 3.1% for CCR6) (Fig 5A).
We also examined gene expression of CCL-20 and CCL-25, two important mucosal homing chemokines [43, 44], in C. albicans- and CMV-specific CD4 T cells. As described earlier (Fig A in S1 Appendix), Ag-specific CD4 T cells were sorted from PBMCs based on CFSE-low and expression of these two genes was quantified by real-time PCR. We found that consistent with expression of mucosal homing receptors, C. albicans-specific CD4 T cells also expressed significantly higher levels of CCL-20 and CCL-25 than CMV-specific CD4 T cells (Fig 5B). These data altogether indicate that in accordance with their Th17-like phenotype, C. albicans-specific CD4 T cells demonstrate a strong mucosal homing potential and may be more likely to migrate to mucosal tissues in HIV-infected subjects.
α4β7 integrin can directly interact with HIV surface protein gp120 [45]. Although the role of α4β7 in HIV pathogenesis is not fully clear, it has been suggested that strong α4β7 reactivity may provide an increased fitness for mucosal HIV transmission [45, 46]. We next investigated potential impact of α4β7 on cytokine expression and HIV infectivity in C. albicans-specific CD4 T cells. We found that IL-17, IL-2, IFN-γ and MIP-1β were expressed at comparable levels between α4β7+ and α4β7- subsets (Fig 5C). Analysis of α4β7 expression and HIV infectivity showed no significant difference in HIV infection between α4β7+ and α4β7- subsets as well (Fig 5D). In addition, we used ACT-1, the anti-human α4β7 antibody known to efficiently block binding of HIV gp120 to α4β7 [47], to block the interaction between HIV and α4β7 during HIV infection. The data showed that pre-inculcation of PBMCs with ACT-1 led to reduced staining of C. albicans-specific CD4 T cells for α4β7 (Fig 5E); however, blocking α4β7 could not reduce HIV infection of C. albicans-specific CD4 T cells (Fig 5E). Taken together, our results suggest that unlike CD4 and CCR5, α4β7 may not be required for in vitro HIV replication in our system, which is consistent with some previous reports [48].
In order to investigate longitudinal impact of HIV on C. albicans- and CMV-specific CD4 T-cell immunity in vivo, we first measured the proliferative responses of C. albicans- and CMV-specific CD4 T cells in PBMCs that were collected at early (mean CD4 count: 797) and chronic (mean CD4 count: 251) stages of HIV infection with time intervals of 2–6 years from the same HIV-infected individuals (Table 1). Ag-specific CD4 T-cell proliferative response in PBMCs was measured using the similar method as described in Fig 1. Since PBMCs were stimulated with whole C. albicans or CMV antigens, predominantly CD4, but not CD8, T-cell proliferative response was induced (Fig 6A). Since in vitro antigen stimulation can lead to significant down-regulation of CD4 receptor, we used CD3+CD8- phenotype to identify CD4 T-cell population following antigen stimulation (Fig 6A). Interestingly, among the subjects examined, we found that while the CMV-specific CD4 T-cell proliferative response was persistent and comparable between early and chronic stages (early vs. chronic: 14.2% to 13%; subject 1), the C. albicans-specific CD4 T-cell proliferative response from the same subjects, which was readily detectable at high magnitudes at early HIV infection, was preferentially lost at late HIV infection (early vs. chronic: 42% to 1.1%; subject 1) (Fig 6A). Significant difference for magnitudes of C. albicans-specific CD4 T-cell proliferative responses between early and chronic stages was observed (n = 4) (p = 0.005) (Fig 6B).
In order to compare the de novo frequencies of C. albicans- and CMV-specific CD4 T cells in PBMCs of HIV-infected subjects, we performed short-term antigen stimulation (overnight), where PBMCs were stimulated with peptide pools derived from C. albicans (MP65) or CMV (pp65), followed by intracellular cytokine staining. Expression of cytokines (IL-17 and IL-2 for C. albicans; IFN-γ and MIP-1β for CMV) was used to determine the frequencies of Ag-specific CD4 T cells in PBMCs. As shown in Fig 6C, while the C. albicans-specific CD4 T cells were readily detectable at fairly high levels early after HIV infection, they were lost or greatly reduced at late stage of HIV infection (p = 0.01); in contrast, the CMV-specific CD4 T cells were well maintained at comparable levels in both early and late stage of HIV infection from the same HIV-infected individuals (Fig 6C and 6D). The results are very consistent with the proliferation data and altogether provide strong evidence for preferential and progressive depletion of C. albicans-specific CD4 T-cell response in progressive HIV-infected subjects.
A better understanding of how pathogen-specific CD4 T cells are infected and/or depleted during HIV infection can provide important clinical insights into host susceptibility to opportunistic infections in AIDS patients. In this study, we used PBMC samples from an ART naïve, longitudinal HIV-infection cohort and reported a sequential dysfunction and preferential depletion of C. albicans-specific CD4 T-cell response, as compared to CMV-specific CD4 T-cell response, in HIV-infected individuals. Our results showed that C. albicans-specific CD4 T cells harbored higher levels of HIV DNA, which supports our in vitro findings and provides in vivo evidence for higher susceptibility of C. albicans-specific CD4 T cells to HIV. Such difference in HIV susceptibility may significantly contribute to their differential depletion rates in vivo. Also importantly, we identified an earlier impairment of Th17-associated functions (IL-22, IL-17 and IL-2) of C. albicans-specific CD4 T cells at early HIV infection when their proliferative and Th1 responses remain detectable, suggesting that anti-C. albicans cellular immunity may rapidly become inefficient early following HIV infection.
C. albicans commensalism in healthy individuals stimulates robust cellular immune responses. Many studies have shown that C. albicans-specific CD4 T-cell response serves as the predominant host defense mechanism for protection [20, 26, 49]. However, specific functional facets of anti-C. albicans CD4 T-cell responses responsible for immune control have been obscure. Earlier studies reported that Th1 response was the key mediator of immunity, based on the observation that deficiency of IL-12 p40 subunit in mice was associated with susceptibility to C. albicans; however, studies also showed that mice deficient in IFN-γ were still resistant to candidiasis [31]. It was later recognized that IL-12 shares the p40 subunit with IL-23, which promotes the differentiation of Th17 subset of CD4 T cells [32], suggesting a role for Th17, but not Th1, response in protection against candidiasis [25]. In support, a recent study by Santos et al. showed that Th17 CD4 cells confer the long-term adaptive immunity to oral C. albicans infections in a murine model [33]. Th17 cells produce two major cytokines, IL-17 and IL-22, which function to mobilize neutrophils and to enhance mucosal epithelial integrity respectively, and are shown to play important roles in host defense against mucosal candidiasis [50–53]. In our study, we identified that in the setting of HIV infection, C. albicans-specific CD4 T-cell responses manifest a sequential dysfunction with Th17-like functions (IL-17, IL-22 and IL-2) being impaired earlier following HIV infection as compared to Th1 functions (IFN-γ and MIP-1β) (Fig 3). In support, in vitro HIV susceptibility analysis showed that the IL-17, IL-22 and IL-2 functional subsets of C. albicans-specific CD4 T cells are more permissive to HIV than the IFN-γ and MIP-1β subsets (Fig 2). These data suggest that during HIV infection anti-C. albicans CD4 T-cell immunity, even in the presence of detectable proliferative and Th1 responses, might quickly become less efficient due to preferential impairment of Th17 functions. However, it is interesting to note that unlike mucosal candidiasis, disseminated candidiasis is remarkably uncommon in HIV-infected subjects. The mechanisms are not entirely clear but likely due to the relative normal functions of neutrophils in HIV-infected individuals [54]. Neutrophil activation requires IL-17 signaling and recent evidence has suggested that innate lymphoid cells (ILC) represent an important source of IL-17 to support neutrophil activation in the absence of Th17 CD4 T cells [55], which may help explain why disseminated candidiasis remains uncommon even in HIV-infected patients with severe CD4 T cell depletion.
Mechanisms for in vivo depletion of CD4 T cells in HIV-infected subjects might be highly complex. However, direct HIV infection and ongoing viral replication is thought to be a major driving factor for CD4 depletion at both acute and chronic stages of the disease [13, 56, 57]. The in vitro system established in our group provides a method to examine differential HIV susceptibility of antigen-specific CD4 T cells and the associated functional profile (Fig A in S1 Appendix). Based on this system, we have demonstrated that C. albicans-specific CD4 T cells are substantially more susceptible to HIV than CMV-specific CD4 T cells in vitro (Fig 2A) [12]. In the current study, we further investigated HIV susceptibility of these two Ag-specific CD4 T-cell populations in vivo. Like previously reported [13, 40], we used quantitative PCR to quantify cell-associated HIV load in sorted Ag-specific CD4 cells as an indication of their natural infection history. Ag-specific CD4 T cells were sorted from PBMCs based on CFSE-low, which provided an advantage in that we could obtain Ag-specific cells at relatively higher numbers for subsequent PCR. We showed that peripheral C. albicans-specific CD4 T cells harbored modestly but significantly higher levels of HIV DNA than CMV-specific CD4 T cells in HIV-infected individuals (Fig 4B). We noted that the differences for HIV load between C. albicans- and CMV-specific CD4 T cells in some subjects were modest. Considering that lymphoid and mucosal tissues such as GI tract are major sites of HIV infection, we speculate that differences might be more profound for Ag-specific CD4 T cells isolated from lymphoid or mucosal effector sites.
Previous studies have investigated in vivo HIV susceptibility of CD4 T cells specific for other important antigens such as MTB and HIV, in addition to CMV [13, 40]. We here included additional antigens for comparison wit C. albicans (Fig 4C). Since the HIV-infected subjects in RV21 demonstrated very low response to MTB, we were unable to directly compare between MTB and C. albicans in these subjects. Instead, we compared C. albicans antigen with varicella zoster virus (VZV) and HIV Env as non-CMV antigen controls. Based on the subjects investigated, we found that C. albicans-specific CD4 T cells also harbored higher levels of HIV DNA than these two groups of antigen-specific CD4 T cells (Fig 4C), suggesting that the higher HIV susceptibility for C. albicans-specific CD4 T cells may not be simply due to enhanced protection of CMV-specific CD4 T cells. An important previous study [40] reported that HIV-specific CD4 T cells are highly susceptible to HIV in vivo, where CD4 T cells specific for all HIV antigens (gag, env, nef, etc) were measured together. We here also showed that HIV Env-specific cells are susceptible to HIV, albeit at lower levels than C. albicans-specific CD4 T cells, which might attribute to variations between different cohorts. In addition, it is possible that HIV susceptibility of CD4 T cell subsets specific to different HIV antigens (e.g. Env vs. Gag or Nef) may also significantly vary, which is under investigation in our group. Nevertheless, our results altogether suggest that C. albicans-specific CD4 T cells are highly susceptible to HIV both in vitro and in vivo, which may contribute to their rapid depletion during HIV infection.
Phenotypic and functional characteristics of CD4 T cells are shown to be closely associated with their susceptibility to HIV infection, such as expression of IL-2/CD25 [10, 13], MIP-1β [13, 58, 59] and IL-17 [60, 61]. Based on the in vitro system (Fig A in S1 Appendix), we showed that compared to MIP-1β or IFN-γ, the IL-17, IL-22 or IL-2-producing subsets of C. albicans-specific CD4 T cells are more permissive to HIV (Fig 2). We noted that although total IFN-γ+ CD4 T cells are also fairly susceptible to HIV, single IFN-γ+ CD4 T cells in the absence of other cytokine expression (IL-17, IL-2 or IL-22) are substantially more resistant to HIV (Fig 2C), which is consistent with and possibly provides an explanation for earlier reports that IFN-γ+IL-2+ double-producing CD4 T cells are preferentially lost, while the IFN-γ single-producing CD4 T cells are frequently detected in HIV non-controllers [62]. While definitive molecular mechanisms for high HIV susceptibility of C. albicans-specific CD4 T cells remain not fully clear, data from the current study and our previous reports have suggested that multiple layers of mechanisms may contribute to this observation: 1) less protection of C. albicans-specific CD4 T cells from HIV at entry level due to limited production of beta-chemokines (Fig F in S1 Appendix and [12]); 2) permissive post-entry environment that favors productive HIV replication, including low expression of antiviral restriction factors [12] and high expression of pro-inflammatory cytokines such as IL-17, IL-22, and IL-2 (Fig 2B and 2C); 3) higher expression of important mucosal homing receptors (α4β7 and CCR6) that enhances exposure of C. albicans-specific CD4 T cells to HIV at mucosal sites (Fig 5). These multiple layers of mechanisms may act together to render C. albicans-specific CD4 T cells susceptible to HIV infection and depletion. α4β7 is a key gut mucosal homing receptor that can directly interact with HIV gp120 [45]. It has been suggested that strong α4β7 reactivity may provide an increased fitness for mucosal HIV transmission [45, 46]. Our results showed that expression of α4β7 does not appear to directly correlate with HIV infectivity in CD4 T cells (Fig 5D), which is further supported by the α4β7 blocking experiment (Fig 5E). In agreement with this result, further analysis also showed no difference in cytokine expression (IL-17, IL-2, IFN-γ and MIP-1β) between α4β7+ and α4β7- subsets of C. albicans-specific CD4 T cells (Fig 5C). These results imply that unlike CD4 and CCR5, α4β7 may not be critically important for in vitro HIV replication in our system, where both cells and virus were present at high concentrations and HIV may be able to efficiently bind to CD4 and CCR5 even in the absence of α4β7. This is consistent with some previous reports [48] and results from other groups (personal communication). However, impact of α4β7 on HIV susceptibility of Ag-specific CD4 T cells in vivo deserves further investigation.
Mechanisms for how expression of cytokines is associated with HIV susceptibility for Ag-specific CD4 T cells are not fully known. We examined CCR5 expression on different functional CD4 subsets and found no significant difference (Fig E in S1 Appendix); instead, we found that IL-2+, IL-22+ or IL-17+ CD4 T cells express lower levels of MIP-1β compared to IFN-γ+ CD4 T cells (Fig 3A), which might explain the differential HIV infectivity between these cytokine-producing CD4 subsets. This is consistent with an earlier study showing that in vitro differentiated IL-17-producing CD4 T cells express comparable levels of CCR5 with IFN-γ-producing CD4 T cells, but are more susceptible to HIV due to lack of beta-chemokine production [38]. Another potentially important correlating factor for HIV susceptibility in different cytokine+ Ag-specific CD4 T cell subsets is CD25. In this study, we showed that productive HIV infection was predominantly observed in CD25+ Ag-specific CD4 T cells and that CD25 co-expressed with IL-2, IL-22 or IL-17, but not IFN-γ+ or MIP-1β (Fig 2E–2G). Lastly, post-entry mechanisms might also be involved in regulating HIV susceptibility in different cytokine-producing CD4 T cells. For instance, polarized Th1-like, IFN-γ-producing CD4 T cells, such as CMV-specific CD4 T cells, which can acquire direct antiviral functions [58], were shown to be able to activate a broad array of innate antiviral factors and manifest strong post-entry inhibition of HIV replication [12]. Molecular mechanisms for how different cytokine signaling may affect HIV infectivity in target CD4 T cells remain largely unknown and further investigation is warranted.
In summary, in the present study, based on an ART naïve HIV infection cohort, we comparatively investigated the longitudinal impact of HIV on C. albicans- and CMV-specific CD4 T-cell immunity in HIV non-controllers. We identified a sequential dysfunction and preferential depletion of C. albicans-specific CD4 T cell response during progressive HIV infection. These findings may provide an immunological basis for early loss of immune control over mucosal candidiasis in HIV-infected individuals and also suggest a potential mechanism for pathogen-specific immune failure in AIDS.
The study involves use of PBMC samples from healthy human donors as well as from HIV-infected subjects enrolled in RV21 cohort, an ART naïve longitudinal HIV infection cohort established by US MHRP. Healthy donor PBMCs were obtained from University of Texas Medical Branch (UTMB) blood bank and the RV21 PBMC samples were obtained from MHRP. Characteristics of HIV-infected subjects were summarized in Table 1. All samples were analyzed anonymously and investigators of this study have no access to any subject identification information. The study was determined as non-human subject research and approved by both UTMB and MHRP IRBs. Written informed consents were obtained from study participants.
Antigens used for long-term (6 days) or short-term (overnight) stimulation of PBMCs include: C. albicans extracts (Greer Laboratories), C. albicans MP65 peptides (JPT), varicella zoster virus lysates (Advanced Biotechnologies), CMV lysates (Advanced Biotechnologies), CMV pp65 peptides and HIV Env peptides (NIH AIDS reagent program). HIV-1 US1 (GS004), an R5 subtype B isolate, was obtained through the NIH AIDS reagent program and used for in vitro HIV infection.
CFSE labeling and antigen stimulation of PBMCs were performed as previously described [12, 34]. Briefly, PBMCs were washed twice with staining buffer (RPMI-1640 medium containing 1% FBS) (Life Technologies, USA), followed by labeling with 1.0 μM CFSE (Life Technologies, USA) at a cell concentration of 2×107 cells/ml for 8 minutes at room temperature (RT) in dark. Equal volume of pre-warmed FBS was then added to cells for incubation at RT for 4 minutes to quench CFSE. CFSE-labeled PBMCs were subjected to antigen stimulation and various subsequent assays. For antigen stimulation, cells were first pulsed with whole antigens (C. albicans: 1:200; CMV: 5 μg/ml) at high cell concentrations (1×107 cells/ml) for 3–4 hours in tubes, and then diluted to 2×106 cells/ml for normal cell culture in culture plate for several days to stimulate antigen-specific T cell activation and proliferation. For experiments involving cell sorting and HIV DNA quantification for PBMCs of HIV-infected subjects (RV21), cells were also stimulated with VZV antigen (final concentration: 5μg/ml) and HIV Env peptides (final concentration: 1μg/ml). Different assays were subsequently performed as detailed below.
For normal PBMCs, stimulated cells were cultured for ~6 days and the proliferating CD4 T cells, identified by CFSE dilution (CFSE-low), in stimulated PBMCs were subjected to multiple analyses: 1) cytokine expression by flow cytometer; 2) Ag-specific CD4 cell sorting using FACS Aria for subsequent gene-expression analysis. In addition, the Ag-stimulated normal PBMCs were also subjected to in vitro HIV infection (day 3 after initial Ag stimulation) for the examining the HIV susceptibility of Ag-specific CD4 T cells. For PBMCs from HIV-infected subjects, CFSE-labeled and antigen-stimulated cells were cultured for ~6 days and subjected to: 1) analysis for cytokine expression and proliferative response (CFSE-low); 2) Ag-specific CD4 cell sorting for the in vivo HIV infectivity analysis. Methods for each assay were detailed below.
Three days after CFSE labeling and initial antigen stimulation, PBMCs of healthy donors were infected with pre-titrated HIV R5 (US1; 50ng/ml p24). In some experiments, prior to HIV infection, antigen-stimulated cells were pre-incubated with anti-human α4β7 antibody (ACT-1) (5 μg/ml) to block the interaction between HIV (envelope protein) and α4β7 present on antigen-specific CD4 T cells. Twenty-four hours after HIV exposure, free HIV virions were washed away from the cell culture. The infection was maintained for additional 2 days and cells were re-stimulated with PMA (500 ng/ml) and ionomycin (1μg/m) for de novo cytokine synthesis (3 days after HIV exposure). Productive HIV infection of antigen-specific CD4 T cells in PBMCs and the associated functional (cytokine) or phenotypic parameters were examined by multi-color flow cytometry based on intracellular HIV p24 expression in CFSE-low proliferating CD4 T cells as previously described [12, 34].
In addition to 6-day stimulation, PBMCs from RV21 HIV-infected subjects (early and chronic time points) were also stimulated with C. albicans and CMV antigens for overnight in the presence of anti-CD28/CD49d antibody cocktail and protein transport inhibitors (BD Bioscience). For short-term stimulation, C. albicans MP65 and CMV pp65 peptides were used. Intracellular cytokine staining and flow cytometric analysis were performed, as detailed below, to measure de nova frequencies of antigen-specific CD4 T cells in PBMCs of HIV-infected subjects.
Sorting antigen-specific CD4 T cells from PBMCs was performed either for quantification of cell-associated HIV DNA (HIV-infected subjects in RV21) or for gene-expression analysis (healthy PBMC). CFSE-labeled, Ag-stimulated PBMCs were first stained with Fixable LIVE/DEAD Violet Dead Stain Kit (Life Technologies), followed by staining for surface markers including CD3-APC-H7, CD4-PE-Cy5 and CD8-PE (BD Bioscience). Cells (from HIV+ subjects) were then fixed and subjected to sorting of antigen-specific CD4 T cells, based on CFSE-low CD3+ CD8-, as well as the CFSE-Hi non-specific CD4 T cells using FACS Aria (BD). Sorted cells were subsequently subjected to quantification of HIV DNA (below). Cells from healthy donors were live sorted for antigen-specific CD4 T cells, followed by RNA extraction and gene-expression analysis (below).
Total RNA was extracted from live-sorted, antigen-specific CD4 T cells using Quick-RNA MicroPrep kit (Zymo) according to the manufacturer's protocol. Gene expression was quantified using iTaq Universal SYBR Green Supermix (Bio-rad) and the CFX Connect Real-Time PCR Detection System (Bio-rad) after reverse transcription from RNA into cDNA using iScript Reverse Transcription Supermix for RT-qPCR (Bio-rad). Primers were designed to amplify the target genes (human T-bet, EOMES and RORC). Primer sequences for gene expression analysis were shown in Table 2. The relative quantity of gene expression was calculated using the 2-ΔΔCt method.
Genomic DNA was extracted from the fixed, sorted antigen-specific CD4 T cells in PBMCs of HIV+ subjects. After washing once in PBS, sorted cells were lysed in lysis buffer (10mM Tris, 5mM EDTA, 1% SDS pH8.0) for 1 hour at room temperature and then digested with 32 U/ml of Protease K (New England Biolabs) for 2 hours at 56°C. After Protease K inactivation at 95°C for 30 min, genomic DNA was purified and solved in Tris-Cl (10 mM, PH 8.0). HIV DNA was quantified using iTaq Universal SYBR Green Supermix (Bio-rad) and the CFX Connect Real-Time PCR Detection System (Bio-rad) according to the manufacturer's protocol. Primers used to amplify HIV Gag and the control GAPDH genes were shown in Table 3. pNL4-3 and recombinant plasmid encoding GAPDH gene were used to generate standard curves (Fig I in S1 Appendix). The absolute quantity of HIV DNA copies was calculated based on standard curves.
Statistical analysis was performed using Prism 6.0 (GraphPad). Statistical comparison between groups was performed using paired or non-paired t test. Two-tailed p values were denoted, and p values < 0.05 were considered significant.
|
10.1371/journal.pgen.1002103 | Chk2 and p53 Are Haploinsufficient with Dependent and Independent Functions to Eliminate Cells after Telomere Loss | The mechanisms that cells use to monitor telomere integrity, and the array of responses that may be induced, are not fully defined. To date there have been no studies in animals describing the ability of cells to survive and contribute to adult organs following telomere loss. We developed assays to monitor the ability of somatic cells to proliferate and differentiate after telomere loss. Here we show that p53 and Chk2 limit the growth and differentiation of cells that lose a telomere. Furthermore, our results show that two copies of the genes encoding p53 and Chk2 are required for the cell to mount a rapid wildtype response to a missing telomere. Finally, our results show that, while Chk2 functions by activating the p53-dependent apoptotic cascade, Chk2 also functions independently of p53 to limit survival. In spite of these mechanisms to eliminate cells that have lost a telomere, we find that such cells can make a substantial contribution to differentiated adult tissues.
| In this work, we describe two simple assays for examining the fate of cells that lose a telomere. We applied these assays to examine the role of DNA damage response genes in controlling the fate of such cells. The checkpoint kinase Chk2 is known to activate the p53 tumor suppressor to promote apoptosis of cells with DNA damage, including the loss of a telomere. In work described here, we discovered that Chk2 can also act independently of p53 to eliminate cells that have lost a telomere. We also show for the first time in Drosophila that the genes encoding Chk2 and p53 are haplo-insufficient, as they are in humans. These critical discoveries demonstrate that the response to DNA damage, in the form of telomere loss, has an unexpectedly high degree of functional conservation from Drosophila to humans. This greatly strengthens the utility of Drosophila as a model to characterize the mechanisms that cells use to respond to telomere loss and, most critically, the mechanisms by which such cells can escape apoptosis. The original assay we describe in this work provides a basis for high-throughput genome-wide genetic screens to identify these mechanisms.
| In the 1930s, seminal work from Hermann Muller and Barbara McClintock showed that the normal termini of linear chromosomes can be distinguished from ends produced by chromosome breakage [1], [2]. Muller showed that normal ends did not participate in chromosome rearrangements induced by irradiation, and conversely, that broken ends created by ionizing radiation could not substitute for normal termini. McClintock demonstrated that broken chromosome ends undergo end-to-end fusion, leading to anaphase bridges during mitosis, followed by breakage which then led this process to repeat. This Breakage-Fusion-Bridge (BFB) cycle could continue for several rounds of mitosis. Evidence for telomere dysfunction and BFB cycles is seen in human tumors and may represent a precipitating early step in carcinogenesis [3]. However, the importance of telomere integrity to ongoing cellular viability is made clear by the discoveries that even cancer cells possess a mechanism for telomere maintenance, either by upregulation of telomerase or through the Alternative Lengthening of Telomeres pathway [4], [5]. If such maintenance mechanisms are lost, the cancer cells undergo apoptosis.
Previously, we showed that telomere loss in somatic cells of flies results in robust activation of caspase-3 mediated apoptosis [6]. This apoptosis is regulated by two p53-dependent pathways, with the majority mediated through loki (lok), which encodes the Drosophila ortholog of the Chk2 checkpoint kinase, and a much smaller fraction mediated through mei41 and grapes (grp), which encode the fly orthologs of mammalian DNA damage response proteins ATM and Rad3 related protein (ATR) and the Chk1 checkpoint kinase, respectively. When telomere loss is accompanied by the generation of aneuploidy, a p53-independent pathway to apoptosis is also activated, but is delayed by many hours [6]–[8]. However, despite the two-pronged robust apoptotic response, a karyotype analysis of neuroblasts demonstrated that a fraction of cells (up to 20%) are able to survive and divide repeatedly for up to 96 hours, until pupariation. Furthermore, a small subpopulation of these surviving cells had experienced repeated BFB cycles, showing that some cells can divide multiple times even though they carry a chromosome lacking a telomere [6].
In contrast, in both the male and female germlines there is clear evidence that chromosomes that have lost telomere can become healed by de novo telomere addition. This healing occurs efficiently in wildtype males [9], [10] or in females that carry the mu2 mutation [11]. These data suggest that different cell types have varying responses to the same genetic lesion, a missing telomere, and studies in model organisms will be pivotal to elucidate new targets for cancer therapy.
Although previous work has shown that some cells that have lost a telomere are able to differentiate [12], [13], the degree to which they participate in forming adult structures remains unclear, nor is it known whether escape from apoptosis is sufficient to allow a cell to fully differentiate after telomere loss. In the work reported here we quantitate the ability of cells to contribute to adult structures after telomere loss and we show that mutation of the DNA Damage Response (DDR) genes p53 and lok greatly enhances the survival and differentiation of such cells. Our results show that the genes encoding these proteins are haplo-insufficient. Furthermore, we find that Chk2 functions independently of p53 to limit cell survival.
To determine the extent to which cells that have lost a telomere are capable of contributing to adult tissue we developed a highly sensitive assay called the Bar and Telomere Loss (BARTL) assay (Figure 1A). Cells that lose a single telomere normally suffer a high rate of apoptosis and, although some do survive and differentiate [12], [13], their ability to contribute to the adult is poorly defined. One drawback to the interpretation of those experiments is that telomere loss was accompanied by some degree of aneuploidy. We designed the BARTL assay so that, in somatic cells of the eye, a single telomere is lost from a dispensable Y chromosome and this coincides with loss of the dominant BarStone (BS) mutation. BS causes caspase-3-dependent cell death in the developing eye starting at least as early as second instar and continuing until only a small posterior segment of the eye imaginal disc remains, resulting in adults with bar-shaped eyes ([14], [15] and Figure S1). We reasoned that loss of BS could give cells a growth advantage and provide a favorable environment to assess their potential for growth and differentiation after telomere loss.
To induce telomere loss the FLP site-specific recombinase was used to mediate sister chromatid fusion by recombination between inverted FRTs on sister chromatids. This produces an acentric chromosome and a dicentric chromosome, which breaks in more than 90% of mitoses and delivers a chromosome with a non-telomeric end to each daughter cell (Figure 1A; [6]). BARTL uses the dicentric-inducible Y chromosome, DcY(H1), which carries inverted FRTs flanking the 3′ coding region of whs inserted proximal to BS (Figure 1B).
Cells that experience loss of a telomere in this assay have an advantage because they have lost BS, but are still subject to the telomere loss-induced DNA damage response that frequently results in apoptosis. To ascertain how effectively such cells would proliferate and differentiate in competition with unaltered BS cells, we measured the eyes of flies that carried DcY(H1) and an eyFLP transgene, which expresses FLP in the eye throughout development. Ten eyFLP lines were tested with DcY(H1): every combination produced eyes that, although rough and irregularly shaped, were significantly larger than BS (representative results shown in Figure 2). For further experiments we chose to use the P{eyFLP.N, ry+}2 line because the average eye size following telomere loss is ∼50% of wild type, permitting the identification of mutations that either limit or promote cell survival following telomere loss.
It is a formal possibility that the larger eye size seen following FLP induction results from loss of the entire Y chromosome, including BS. To test whether dicentric production precipitated chromosome loss we induced FLP expression by heat shock in flies carrying the heat-inducible 70FLP3F transgene and the DcY(H1) chromosome, then examined mitotic figures from larval neuroblasts 24 and 48 hours after FLP induction for loss of the Y chromosome. There was no increase in chromosome loss in flies that expressed FLP, compared to flies that did not (Table 1), confirming that the larger eye phenotypes seen in the BARTL assay were not the result of complete loss of the Y, but instead reflect extensive growth and differentiation of cells that lost a telomere.
The DNA damage response, acting primarily through Chk2 and p53, mediates apoptosis in response to telomere loss [6]. To determine whether reducing or eliminating the apoptotic response would allow such cells to proliferate and differentiate we used the BARTL assay to examine the influence of mutations in these genes. The lokp6/+ heterozygous flies had eyes that were much larger than the lok+ control, and the lokp6/p6 homozygotes had eyes of nearly wildtype size and morphology (Figure 3A). When we tested two alleles of p53 (p535A-1-4 and p5311-1B-1) the heterozygotes had eyes that were also much larger than the p53+ control, with p53−/− homozygotes having eyes that were near wild type in size (Figure 4). As expected, the addition of a p53+ transgene reduced eye size significantly. We also tested a hemizygous deletion of 276 kb that removes p53, and found that it had a similar effect as the heterozygous p53 mutations. The lok and p53 mutations used in these studies had no effect on the BS phenotype in the absence of FLP expression and telomere loss (mean sizes ±1 SD, normalized to wildtype eye: y w/H1 = 0.090±2.8 e-17, N = 20; y w/H1; p535A-1-4 = 0.099±0.028, N = 10; y w/H1; lokp6 = 0.090±2.8 e-17, N = 24; P values are 0.6 and 0.9 respectively).
To determine if our BARTL results were an anomaly of the extreme selection conferred by the context of surrounding BS cells, we induced telomere loss from an independently derived Y chromosome, DcY(FrTrYS)4B1A, that contains a P element with a white+ (whs) gene and inverted FRTs on the short arm of Y. Cells that lose whs must have experienced dicentric formation (and breakage), allowing positive identification of at least some of the cells that lose a telomere and survive to contribute to the adult eye. The eye shown in Figure 5B is typical of flies recovered in this experiment. The white sectors that predominate indicate that the majority of surviving cells have lost a telomere and the whs transgene.
Since we do not know the orientation of the P element on this chromosome it is possible that some cells experiencing dicentric formation and breakage may retain whs. If, for instance, whs lies proximal to the FRTs, then the long chromosomes produced by asymmetrical breakage of a dicentric bridge will retain whs. However, since eyFLP expresses continuously through eye development and such cells retain inverted FRTs (Figure 1A) they may experience further rounds of dicentric formation, giving added opportunity to lose the whs gene. It is also possible that some cells may escape recombination entirely and retain whs. Notwithstanding this uncertainty about whether pigmented cells have lost a telomere (or not), our conclusion that most surviving cells experienced telomere loss remains valid since it is based on the predominant occurrence of cells that lack pigment.
In a wildtype background, flies with DcY(FrTrYS)4B1A and eyFLP had small and rough eyes that were on average 71% as large as normal eyes (Figure 5B and 5E). Although cells are capable of surviving telomere loss and contributing to the adult eye in the absence of BS selection, this observation that the eyes were smaller than wild type indicates that many of the cells that lost a telomere succumbed to apoptosis.
We then asked how lok and p53 mutants would alter the outcome in this context. We found that p53−/−, p53+/− and lok+/− had eyes that were significantly larger than the non-mutant controls (Figure 5C, 5D, 5E; lok−/− were not tested), demonstrating that the results from the BARTL assay are not an artifact imposed by the BS allele.
We previously showed that mutations of mei-41 and grp, the genes that encode the Drosophila orthologs of mammalian ATR and Chk1, have a detectable effect on reducing apoptosis after telomere loss only in a lok background [6]. Since the BARTL assay is sensitive enough to distinguish a heterozygous effect with p53 and lok, we also tested mei-41 and grp with this assay. BARTL flies homozygous for grp, or hemizygous for either of two alleles of mei-41, showed no significant change in eye size when compared to wildtype flies (Figure 3B). We further tested a role for grp by analyzing lok grp double mutants and found that the effect was not different from lok−/− single mutant flies (Figure 3A), confirming that these genes play a minor role in the elimination of cells that have lost a telomere.
A number of genes that are essential for telomere protection have been identified. These include cav which encodes HOAP, Su(var)205 which encodes HP1a, tefu which codes for the ATM homolog, nbs, mre11, rad50 and hiphop [16]–[24]. The genes required for telomere protection are also required for cell viability, as loss of any one of these genes leads to global telomere dysfunction, multiple end-to-end fusions and ultimately to cell death. Use of the eyGAL4 transgene in conjunction with UAS-RNAi lines to knock down expression of cav, hiphop, or Su(var)205 in the eye, resulted in most flies dying as pharate adults with very small or no heads, indicating extensive cell death even in the absence of FLP-induced dicentric chromosome formation. The few that did survive had small rough eyes consistent with extensive cell death. Since RNAi-mediated knockdown of these genes strongly reduces cell viability, and homozygous mutants fail to develop past the early pupal stage, we were unable to assess their influence using the BARTL assay. However, we did test several genes as heterozygotes in the BARTL assay. We tested two components of the MRN complex, which consists of Mre11, Rad50 and Nbs, and is required for telomere protection and the DNA damage response [18], [24], [25]. BARTL flies heterozygous for nbs or rad50 mutations were not significantly different than controls (Figure 6). We also examined BARTL flies that were heterozygous for mutations of Su(var)205 or cav and saw no significant difference from controls (Figure 6).
Finally, we tested whether cell survival in the BARTL assay was affected by mu2 function. mu2 mutant females allow a high rate of recovery of broken-and-healed chromosomes through their germline, and examination of somatic cells suggests that Mu2 has a checkpoint function [26], [27]. If mu2 permits healing of broken chromosomes in the soma then we might expect to see increased survival of cells that have lost a telomere. We assayed two mu2 mutant genotypes: mu21/1 and an RNAi knock down construct, mu2P{GD12728}v28342, in the BARTL assay, and did not see a significant difference in eye size compared to controls (Figure 6 and data not shown). If de novo telomere addition does occur in the soma (of males at least), it does not appear to be controlled by mu2.
We also investigated the effect of telomere loss on cell survival using an autosome instead of the Y. The Dc3(FrTr61A5)1A chromosome 3 has inverted FRTs inserted very near the tip of the left arm, with whs located just distal to these FRTs. Therefore whs will be located on the acentric chromosome produced by FLP-mediated recombination and is frequently lost after dicentric/acentric formation. Similar to the results with DcY(FrTr4B1A), flies carrying Dc3(FrTr61A5)1A and eyFLP had predominantly white eyes, indicating that the vast majority of cells that contribute to the eye have lost a telomere, and the eyes were rough and smaller than wild type indicating frequent cell death (Figure 5E). Also consistent with previous results, flies that were homozygous for mutations in p53 or lok had eyes that were significantly larger than p53+ or lok+ flies.
In contrast to results obtained with the BARTL assay or with DcY(FrTrYS)4B1A, eyes from p53+/− or lok+/− heterozygotes had eyes that were not significantly larger than p53+ or lok+ flies. We hypothesize that the semi-dominant effect of p53 and lok was not seen because in this case, where aneuploidy is produced, the p53-independent aneuploidy-triggered cell death pathway plays an additional role in elimination of many of the cells that have lost a telomere [6]–[8].
One limitation of the BARTL assay is that, because the eyeless promoter is used to drive FLP expression continuously, telomere loss may occur throughout development of the eye. We wished to assay the ability of cells that have lost a telomere to proliferate and differentiate for the full length of development, so we developed an assay, using Dc3(FrTr61A5)1A, that provided this capability. This system is similar to the often-used SMART (Somatic Mutation And Recombination Test [28]) assay, but since it is based on catalyzed telomere loss we call it SMARTL, for Somatic Multiplication After Recombinase-mediated Telomere Loss. Flies carry a normal chromosome 3 marked with the recessive multiple wing hairs (mwh) mutation, heterozygous with the Dc3(FrTr61A5)1A chromosome that carries mwh+ (Figure 1C). These flies also carry the heat-shock-inducible hsFLP1 transgene on the X so that FLP can be induced at any point during development by application of a heat pulse. When the chromosome 3 dicentric bridge that is produced breaks asymmetrically, such that the break point is proximal to the mwh+ allele, then one daughter cell will become hemizygous for mwh. Some of the cells that lose a telomere are then identifiable by their mwh phenotype, and are easily recognized in the adult wing. In our experiments we scored mwh clone number and clone size in at least 3 and up to 46 wings per time point, from flies heat-shocked at different times throughout development.
In wildtype flies that eclosed five days after heat shock (d.a.h.s), which corresponds to heat shock applied at approximately the time of pupariation, mwh cells in the wing were so frequent that it was not possible to distinguish separate clones. In wildtype flies collected six d.a.h.s, which translates to a pulse of FLP ∼24 hours before pupariation, the average number of mwh cells/wing was 6.11 (Figure 7A), suggesting that within 24 hours the majority of cells that lost a telomere were eliminated from the viable cell population, likely by apoptosis. However, the average number of mwh cells was still ∼11-fold greater than the number of mwh cells generated spontaneously. The number of mwh cells continued to be elevated in flies collected seven and eight d.a.h.s. (heat-shocked ∼48 and ∼72 hours before pupariation, respectively), at ∼2–4 times the spontaneous level. Taken together these data indicate that although most cells that experienced telomere loss were eliminated within 3–4 days of induction of telomere loss (8–9 d.a.h.s), some of these cells do survive for this period, and are capable of differentiation. Heat shocks given at even earlier developmental stages did not produce an increase in mwh cells compared to the no heat shock control, indicating that nearly all cells that lose a telomere are eliminated after 4–5 days of normal growth in a wildtype background.
Mutation of p53 or lok greatly improved the survival of cells that had lost a telomere at all time points tested (Figure 7A; all statisitical results shown in Table S1). However, the effects of these two mutations were significantly different from each other. When telomere loss was induced early in development (flies eclosing 8–10 d.a.h.s), cells that lost a telomere survived much better in lok−/− mutants (∼130× wild type) than in p53−/− mutants (∼5× wild type). When telomere loss was induced later in development the survival of cells that lost a telomere improved in p53−/−, but stayed about the same in lok−/−, so that with flies eclosing seven d.a.h.s, survival of such cells was similar in both genotypes (∼160× wild type), and with flies that eclosed six d.a.h.s survival was better in the p53−/− flies (∼100× wild type) than in lok−/− (∼40× wild type). The large number of mwh cells produced at early developmental stages in lok flies clearly depends on telomere loss, since without heat shock to induce FLP we observed only an average of 3.5 mwh cells per wing in 34 wings.
Both lok and p53 mutants exhibited haplo-insufficiency in these experiments (Figure 7B and 7C), as they did in the BARTL assay, but it was observed only in flies that differentiated within a day or two of the hsFLP induction. With early heat shocks, the heterozygotes were able to eliminate cells that lost a telomere as well as wild type. For p53, the addition of a wildtype transgene to the homozygous mutant produced results similar to the heterozygous mutant (Figure 7C).
As was the case in the BARTL assay, loss of Chk1 (grp−/−) did not produce a significant effect (Table S1). We also tested lokp6 grpfs1 double mutants and they were not consistently different from lokp6 single mutants (Figure 7A).
One of the major roles of the complex of nucleic acids and proteins that form a telomere is to hide the chromosome terminus from machinery that mediates the DNA damage response [29], [30]. This response typically leads to the activation of p53, predominantly through phosphorylation by ATM and Chk2, major transducers of the DNA damage response [31], [32]. p53 is known to have a number of transcriptional targets, both in mammals and in flies [33]–[36]. The outcome of p53 activation ranges from cell cycle arrest and DNA repair to apoptosis, depending on the type and quantity of DNA lesions and the cellular context [37]–[41]. We previously showed that the primary cellular response to telomere loss in flies is rapid activation of apoptosis [6]. Nevertheless, by examining neuroblast karyotypes in otherwise wildtype flies we found that ∼20% of cells that lose a telomere survive and proliferate for at least 96 hours, even when they accumulate significant aneuploidy [6].
In the work reported here we investigated the capacity of cells that lose a telomere to survive through most of the life cycle of the developing fruit fly and differentiate into adult structures. The BARTL assay, based on the simultaneous loss of a telomere and the dominant BS mutation in the eye, is particularly useful because the phenotype can be readily scored in a semi-quantitative fashion, facilitating a rapid genetic screen. In this assay cells that have lost a telomere are conferred a selective advantage because the neighboring cells they must compete with are crippled by BS. Even so, the alternative tests we employed which do not confer a selective advantage, such as the SMARTL assay, confirmed the BARTL results.
We found that elimination of the apoptotic DNA damage response, either through mutation of lok, the gene encoding Chk2, or mutation of p53, greatly increased the ability of cells that had lost a telomere to proliferate and differentiate into adult tissues. It is striking that both lok and p53 mutants act semi-dominantly; in other words, the genes are haplo-insufficient for normal elimination of cells that have lost a telomere. This is highly reminiscent of Li-Fraumeni syndrome in humans, a cancer prone disorder that results from mutations in p53 [42], [43]. Human Chek2 mutants also confer a similar predisposition to tumors in multiple tissues [32], [42], and both p53 and Chek2 mutants are inherited as autosomal dominant diseases. Although the majority of p53 mutations that result in Li-Fraumeni syndrome are probably mis-sense, null alleles are also known. Moreover, many tumors arising in a mouse model of Li-Fraumeni do not show loss of heterozygosity for p53 [44]–[46], indicating that in mammals it is also likely that p53+ is haplo-insufficient. This high degree of conservation of function certainly makes Drosophila an appealing model for examining the role of p53 and other genes in the response to telomere loss.
In the SMARTL assay haplo-insufficiency of p53 and lok was only seen when telomere loss was induced during the last one to two days of larval development. When telomere loss was induced at earlier stages, in heterozygotes the mechanisms for eliminating such cells worked sufficiently so that there was no significant difference from wild type. Since most cancer cells show evidence of a period of early genomic instability that appears to stem from telomere dysfunction [47], a short lapse in the normal elimination of such cells may be sufficient to allow genomic reshuffling that can facilitate carcinogenesis, partially accounting for the increased incidence of cancer in humans that are heterozygous for p53 and Chek2 mutant alleles.
The results of the SMARTL experiment also indicate the existence of a Chk2-controlled pathway that functions independently of p53. When telomere loss was induced during the first few days of development, the lok−/− mutants allowed cells that had lost a telomere to proliferate and differentiate to a much greater extent than p53−/− mutants. This was not due to maternal deficiency of Chk2 because the female parents were lok+/− heterozygotes. Furthermore, the capability of p53−/− flies to eliminate cells that have lost a telomere early in development was not a result of maternal supply of p53 product, since the mothers in this experiment were homozygous for a p53 null allele. We previously showed that a p53-independent cell death pathway is triggered by aneuploidy [6]. Because the aneuploidy-triggered pathway is also independent of Chk2 [6], [8], that pathway cannot account for the difference. We conclude that loss of Chk2 in lok−/− mutants abrogates the apoptotic response that acts through p53 and also abolishes a second pathway to eliminate cells that have lost a telomere.
Chk2 has been shown to mediate several p53-independent processes [48], [49], including pathways that lead to cell cycle arrest [33], [50], [51] and apoptosis [52]. Because clones of cells that lost a telomere were both larger and more numerous in lok−/− mutants than in wild type or p53−/− (2), it appears that Chk2 may direct both cell cycle delay/arrest and apoptosis in response to telomere loss in addition to its role in activating p53. We propose that Chk2-mediated cell cycle delay or arrest could mimic the phenotype of a slow-growing Minute cell, and thus induce the Jnk pathway and apoptotic caspase cascade that eliminates poorly growing cells [7], [53] providing a route to cell death that does not require p53 (Figure 7).
One question that arises is why are cells that have lost a telomere in the SMARTL assay not eliminated by the aneuploidy-triggered pathway? It is likely that some cells are eliminated by this pathway. However, if aneuploidy-triggered death is a specific response to loss of Minute genes [7], then since mwh+ lies distal to all Minute genes [54] it is possible for dicentric breakage to produce a mwh cell that is not Minute. Such cells need not be subject to the aneuploidy-triggered death pathway and could proliferate extensively unless eliminated by other mechanisms.
Our results might be partially explained by an alternative hypothesis: that loss of Chk2 allows chromosomes that have lost a telomere access to a repair pathway which uses the homolog to restore the end of the chromosome. Break-induced replication is one such mechanism, and is similar to the ALT mechanism used for telomere maintenance in ∼15% of human cancers [55], [56]. Exchange with a normal chromosome is another possibility, analogous to telomere-sister chromatid exchange, except in this case the homolog would be used [57]. Although such mechanisms could operate in the SMARTL assay, where a genuine homolog is present, it is more difficult to imagine their employment in the experiments of the BARTL assay, where the flies carry no chromosome that is a DNA sequence homolog of the Y. So, even though it may occur, this is not likely to be the only mechanism by which cells can escape the apoptotic pathway.
The telomere-loss induced p53-dependent apoptotic response is also activated through a secondary pathway, albeit it to a much lesser extent, via the mei41 and grp gene products Atr and Chk1 [6]. However, we detected no significant effects of these mutants using either the BARTL or SMARTL assays. Taken together these data confirm that the contribution of this pathway to the elimination of cells that have lost a telomere is small.
The picture that emerges is of a complex multi-pronged response to telomere loss (Figure 8). Two pathways recognize damaged ends and invoke the DNA damage response to produce p53-mediated apoptosis. A third pathway is activated in the context of substantial aneuploidy. The p53-mediated and aneuploidy-mediated pathways converge in the activation of caspases. The Chk2 branch of the response to telomere loss bifurcates, leading to activation of the p53 apoptotic response and an alternative pathway that can also eliminate cells that have lost a telomere.
In our view, the most pressing issue is to fully identify and understand the mechanisms that allow a cell to escape the apoptotic responses to telomere loss, and then continue to divide. Such division may amplify and spread the existing genomic damage and, in humans, place cells in peril of becoming cancerous [3], [6], [47]. One possibility is that detection and response to this type of DNA damage is not 100% efficient. Certainly, our observation that p53 and lok are haplo-insufficient is consistent with the interpretation that the system is not overly sensitive to DNA damage. Cells could also escape telomere-loss-induced apoptosis by a process termed adaptation [58]–[60]. When yeast cells adapt to persistent DNA damage, the normal checkpoint is attenuated, but the DNA damage remains. The resumption of cell division leads to chromosome instability and chromosome loss [60], [61]. Cell division in the presence of DNA damage has been taken as de facto evidence of adaptation in mammalian cells [62]. By this definition, adaptation does occur in Drosophila. We previously showed that some cells could undergo several rounds of division after telomere loss. In these cells there was evidence of further chromosome rearrangement, indicating that they had unrepaired damage [6]. Thus, one may conclude that adaptation had occurred. Finally, the broken chromosome might become healed by construction of a new telomere on the broken end. Such a mechanism clearly exists for chromosomes that have lost a telomere in the germline [9]–[11]. Whether this process also occurs in the soma is still an unanswered question. However our experiments showed no effect of the mu2 mutant, which allows healing to occur on chromosomes broken in the female germline. If de novo telomere addition does occur in the soma, it is not controlled by mu2.
To fully understand the complex responses to telomere loss, it will be necessary to identify downstream mediators of the response and link them with specific upstream activators. The combination of powerful genetic and cytological tools in concert with multicellular development makes Drosophila an ideal system to examine the genetic regulation of the responses to telomere loss. The BARTL assay provides a facile method to screen for genes that are involved in this response. In conjunction with the SMARTL assay, the examination of cellular apoptosis, and the observation of karyotypes of surviving cells it should be possible to thoroughly characterize the roles of genes in long-term cell survival following telomere loss. The examination of germline effects, where chromosome healing is readily assayed, should help to distinguish the roles of such genes.
All flies were maintained and mated at 25°C on standard cornmeal food. Heat shocks were carried out in a circulating water bath at 38°C for 1 hour. The fly lines yd2 w1118 P{ey-FLPN}2, yd2 w1118; P{ey-FLPN}5, w P{ey4x-FLP.Exel}1, sn3 mei-41D9, mei-41D5, Df(3R)Exel6193, nbs, P{EP}rad50EP1, mu2, Su(var)2055, P{UAS-FLP1.D}JD1, and {GAL4-ey.H}4-8 were obtained from the Bloomington stock center. Several eyFLP lines, including eyFLP16D, were generated by mobilizing a P insertion, P{ey4x-FLP.Exel}1, located on the X and selecting lines with multiple insertions, by screening for stronger w+ expression, in order to generate lines with stronger eyFLP expression. The lokp6 stock was obtained from William Theurkauf; the grpfs1 lokp6 double mutant was obtained from Michael Brodsky; the grpfs1 mutant was obtained from William Sullivan; the cav mutant stock was obtained from Maurizio Gatti. Construction of the p53+ transgene was described previously [6]. Apoptosis in BS eye discs was visualized using the cleaved caspase-3 antibody (protocol adapted from [63]) from Cell Signaling Technologies (cat. no. 9661) and Alexa-Fluor 568 from Invitrogen (cat. no. A11036).
Crosses were carried out using yd2 w1118 P{ey-FLPN}2 and DcY(H1) in mutant or wildtype backgrounds. P{ey-FLPN}5, which is located on chromosome 2, was used to evaluate the effect of multiple mei-41 alleles with DcY(H1). BARTL results were secondarily confirmed using multiple different insertions of P{ey4x-FLP.Exel}1 or the combination of P{UAS-FLP1.D}JD1 and P{GAL4-ey.H}4-8, as a FLP source.
To determine eye size, each eye was measured along the anterior-posterior axis and the dorsal-ventral axis using a digital filar micrometer and the two measurements were used to calculate the area of an ellipse. Statistical analysis used the Mann-Whitney test with Instat software for Macintosh. The area of the eyes was divided by the mean of P{eyFLPN}2/Y controls and size is presented as a fraction of wildtype size. Whisker plots were generated using Prism software for Macintosh. Eyes were photographed using a Nikon D200 digital camera and processed in Adobe Photoshop.
Neuroblast figures were generated as described [67], stained with DAPI and visualized with a Zeiss Axioplan. A single brain was mounted per slide. Karyotypes were scored by scanning the entire brain and scoring every metaphase nucleus for the presence or absence of the Y chromosome.
For the SMARTL assay flies were crossed and allowed to lay eggs for 5 days. The adults were then transferred to a new vial and the larvae were heat shocked for 1 hour at 38°C. Flies were immediately placed back at 25°C after heat shock and flies were collected every 24 hours for 10 days after heat shock. Wings were mounted on slides in isopropanol and mounting media, Cytoseal 60 Richard-Allan Scientific. Wing hairs were counted using a Zeiss Axioskop.
|
10.1371/journal.pntd.0004178 | The Immunology of a Healing Response in Cutaneous Leishmaniasis Treated with Localized Heat or Systemic Antimonial Therapy | The effectiveness of systemic antimonial (sodium stibogluconate, Pentostam, SSG) treatment versus local heat therapy (Thermomed) for cutaneous leishmaniasis was studied previously and showed similar healing rates. We hypothesized that different curative immune responses might develop with systemic and local treatment modalities.
We studied the peripheral blood immune cells in a cohort of 54 cutaneous Leishmania major subjects treated with SSG or TM. Multiparameter flow cytometry, lymphoproliferative assays and cytokine production were analyzed in order to investigate the differences in the immune responses of subjects before, on and after treatment.
Healing cutaneous leishmaniasis lead to a significant decline in circulating T cells and NKT-like cells, accompanied by an expansion in NK cells, regardless of treatment modality. Functional changes involved decreased antigen specific CD4+ T cell proliferation (hyporesponsiveness) seen with CD8+ T cell depletion. Moreover, the healing (or healed) state was characterized by fewer circulating regulatory T cells, reduced IFN-γ production and an overall contraction in polyfunctional CD4+ T cells.
Healing from cutaneous Leishmaniasis is a dynamic process that alters circulating lymphocyte populations and subsets of T, NK and NKT-like cells. Immunology of healing, through local or systemic treatments, culminated in similar changes in frequency, quality, and antigen specific responsiveness with immunomodulation possibly via a CD8+ T cell dependent mechanism. Understanding the evolving immunologic changes during healing of human leishmaniasis informs protective immune mechanisms.
| Globally, leishmaniasis treatment relies on the use of antimonial drugs (i.e. SSG). In an earlier study we showed that skin lesions due to L. major treated by the ThermoMed (TM) device healed at a similar rate and with less associated systemic toxicity than lesions treated with intravenous SSG. The current study compared the immune responses of these two therapeutic groups before, during and after therapy which may be relevant to resistance to reinfection and also in consideration for the development of local (versus systemic) therapy. Antimonials have immune effects on both the host and parasite while heat treatment locally kills the parasite and induces inflammation from a secondary burn. We demonstrated that healing from cutaneous leishmaniasis is a dynamic process associated with a modulation of immune responses independent of treatment modalities.
| Leishmaniasis, a vector-borne parasitic disease, remains a pressing global health concern with 12 million persons infected, 2 million new infections each year, limited therapeutic options and no effective vaccine [1].
Healing cutaneous leishmaniasis (CL) relies on the development of an effective and balanced protective immune response. The intracellular parasite needs to be contained, while the pathologic immune response needs to be controlled. The murine model for L. major substantially contributed to our understanding of protective immunity and helped establish the T helper 1 (Th1)/Th2 paradigm that explained resistance and susceptibility to Leishmania infection [2,3]. This model demonstrated that T lymphocytes are key for the generation of this protective response through their IFN-γ production which activates macrophages to produce toxic nitrogen and oxygen metabolites to kill the intracellular amastigotes [4]. The Th1 cytokine profile, i.e. IFN-γ, TNF-α and IL-12, is crucial to eliminate Leishmania [5], while the development of a Th2 immune response with the production of IL-4, TGF-β and IL-10 favors parasite multiplication and fails to control the infection [6]. The quality of a T cell response, defined by the pattern of cytokine production at the single cell level, underscores the importance of polyfunctional CD4+T cells specifically producing IFN-γ, TNF-α and IL-2 for protection [7,8]. Additionally, immunoregulatory mechanisms involving regulatory and memory T cells can significantly influence leishmaniasis outcome [9].
The precise role of human CD4+T cell subsets, their cytokine patterns and the immune response pathways engaged during and after effective leishmaniasis therapy are incompletely understood. While pentavalent antimonial drugs (i.e. SSG, meglumine antimoniate) have been used to treat CL for decades [10], they are toxic, require extended duration of treatment, and drug resistant parasites have emerged as a problem [11,12]. The mechanism of action of SSG includes effects on both the host macrophage and parasite [13]. Thermotherapy is an alternative treatment for CL [14,15], delivering localized radiofrequency waves into skin lesions to physically destroy the temperature sensitive parasites. Thermomed (TM, Thermosurgery Technologies, Phoenix AZ), cleared by the Food and Drug Administration, is one of the most studied devices in use [15]. Clinical trials comparing local heat to systemic antimonial therapy showed similar CL cure rates [14,16–20].
We previously reported that subjects treated with the TM device showed similar healing by 2 and 12 months follow-up, with less associated systemic toxicity than those treated with intravenous SSG [21]. We hypothesized that an immunomodulatory systemic therapy would induce a different immune response compared to a locally applied physical treatment, though both methods were ultimately curative. This work comparatively evaluated the immune response profile over time in the participants treated with SSG or TM. We showed a modulation of immune response occurs during healing from cutaneous leishmaniasis independent of either treatment modality.
All participants provided written informed consent and study protocols were approved by Institutional Review Boards at both WRAMC and the Walter Reed Army Institute of Research.
All participants were U.S. military personnel referred to the Walter Reed Army Medical Center (WRAMC) for treatment of parasitologically confirmed L. major infection (Table 1). Details of the clinical trial are published [21]. Seven healthy uninfected control subjects were recruited under a separate protocol.
Whole blood subjects were drawn at time points designated “pre-treatment” (PRE), “on-treatment” (ON) and “post-treatment” (POST) (Days 0, 9±1 and 219±68 following treatment initiation, respectively). For pre- and on-treatment subjects, blood was drawn at WRAMC and processed fresh. At POST, blood was drawn at alternate medical facilities and shipped via overnight carrier for processing. Peripheral blood mononuclear cells (PBMC) were isolated from whole blood as previously described [22].
The following fluorescence-conjugated antibodies were used for multiparameter flow cytometry: CD3 (SK7), CD4 (SK3), CD8 (SK1), CD14 (M5E2), CD19 (HIB19) CD25 (2A3), IL-10 (JES3-19F1), TNF-α (Mab11), IL-2 (5344.111), γδ TCR (B1) (BD Biosciences, San Jose, CA); CD4 (SFCI12T4D11) (Beckman Coulter, Fullerton, CA); IL-17 (eBio64DEC17) and αβ TCR (IP26) (BioLegend, San Diego, CA); IFN-γ (4S.B3) (eBioscience, San Diego, CA). All antibodies were titrated prior to use to determine optimal staining concentrations. Flow cytometry data was acquired either on a FACS Calibur or LSR-II flow cytometer (BD Biosciences) and data analyzed using FlowJo software (TreeStar, Ashland OR).
Prior to cryopreservation, a PBMC aliquot was stained for cell surface markers and analyzed by flow cytometry. Markers included the BD SimulTEST (CD45, CD14) and BD MultiTEST (CD3, CD16, CD56, CD45, CD19) reagents. T cell populations were further analyzed by staining with CD3, CD4, CD8, and CD25. Following staining, cells were fixed in 2% paraformaldehyde, data collected with a FACS Calibur flow cytometer (BD Biosciences) and analyzed using FlowJo software (TreeStar, Ashland OR).
Cryopreserved PBMC were thawed in complete media. A portion of the PBMC was depleted of CD8+T cells (CD8depl PBMC) using the Dynal CD8 Positive Isolation Kit (Invitrogen, Carlsbad CA). Total PBMC or CD8depl PBMC were plated in the presence of soluble Leishmania antigens from L. major parasites (SLA, 2.5 μg/mL, generous gift of Dr. Frank Neva) for 6 days at 37°C, 5% CO2. Pokeweed mitogen (PWM, 5 μg/mL, Sigma) was used as a positive control. Cell-free supernatant was collected from each well, triplicate subjects pooled, and used to quantify cytokines using the Q-Plex Human Cytokine–IR Array (Quansys Biosciences, Logan, UT) according to manufacturer’s protocol [23]. For LPA, cells were pulsed as previously reported [24].
Cryopreserved PBMC were thawed and labeled with carboxyfluorescein diacetate succinimidyl ester (CFDA-SE, Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions [25].
Cryopreserved PBMC were thawed and incubated overnight at 37°C, 5% CO2. Cells were plated at 1 x 106 per well and stimulated with L. major whole lysate (1μg/mL, generous gift from David Sacks) for 24 hours at 37°C, 5% CO2. Brefeldin A (10μg/mL, Sigma) was added to all wells at 18 hours. All cells were costimulated with 1 μg/mL CD28 and CD49d antibodies (BD Biosciences). Following stimulation, cells were stained for population identification markers (CD3, CD4, CD8, CD14 and CD19) and intracellular cytokine expression (TNF-α, IFN-γ, and IL-2). T cell receptor (TCR) phenotyping antibodies were included for the αβ TCR and γδ TCR.
All statistics were performed using GraphPad Prism 4.0 (GraphPad Software, San Diego, CA).
Fifty-four U.S soldiers (98% male) with CL were enrolled and randomized to either local heat therapy (TM) or 10 days of intravenous SSG (Table 1). To evaluate the immune response profiles in these subjects, PBMC were isolated from whole blood at three time points. Pre-treatment (PRE) cells were collected upon enrollment into the study (n = 54, 100%). The on-treatment cells were collected on 9±1 treatment day (n = 54, 100%), and post-treatment (POST) subjects collected at a mean of 7 months (range 4.7–9.2 months), after treatment (n = 39, 72%). Because 39/54 participants provided cells at all time points, the majority of our analysis is restricted to this subcohort (Table 1). No significant differences were noted between treatment arms or subcohort and cohort regarding demographic characteristics, disease burden and therapy outcome.
Freshly isolated cells were stained and analyzed by flow cytometry to characterize the circulating lymphocyte populations. Data from 30 subjects for which there were adequate numbers of cells for all time points is shown (Fig 1). The distribution of lymphocyte populations, including T cells, B cells, NKT-like and NK cells, was unchanged from pre-treatment through the first ten days of treatment (Fig 1A). At POST we observed a significant decrease in circulating T cells (pre, 73%; post, 63%; p< 0.0001), and a concomitant increase in circulating NK cells (pre, 8%; post, 12%; p = 0.0005). The proportion of B cells was unchanged while NKT-like cells showed a modest yet significant decrease (p = 0.036). Results were not affected by removing the few treatment failures from each group (S1 Fig). The observed changes did not correlate with the severity of disease in terms of lesion size (S2 Fig). Analysis of NK subsets based on CD56 and CD16 markers showed a significant decrease in CD16+CD56+ cells at POST in the SSG group (Fig 1B).
The subjects were stratified and reanalyzed to determine if the observed changes in cell populations in POST correlated with treatment arm. Similar declines in circulating T cells were seen in both the SSG and TM subjects. Surprisingly, there was no difference when comparing the percentage of T cells in POST between treatment groups (Fig 1C). Similar population changes for NK cells and NKT-like cells were observed in both treatment arms (S3 Fig). We next investigated CD4+ and CD8+T cells subsets before and after treatment. There was a marked decrease in the median percentage of CD4+T cells (pre, 62.3; post, 57.9; p = 0.0089) and a proportionate increase in CD8+T cells (pre, 30.3; post, 34.8; p = 0.0128) post-treatment, with no changes in the CD4-CD8- (double negative, DN) population (Fig 1D). We determined the TCR distribution in CL caused by L. major, using flow cytometry to profile the TCR repertoire of each of the four subsets of T cells (based on CD4 and/or CD8 expression) in our subjects and in healthy donors (n = 7). Here the αβ TCR was exclusively expressed on single-positive CD4+T cells and double-positive CD4+CD8+T cells, and predominantly on the single-positive CD8+T cells (representative donor shown, S4 Fig). The DNT cells, on the other hand, were a mixture of αβ expressors, γδ expressors and a population that was negative for both of these TCR. Surprisingly, our results for the αβ and γδ TCR align in healthy and L. major infected subjects. A decrease in αβ expression (p = 0.052) (Fig 1E) and trend in increase of γδ was observed in POST (Fig 1F) while the overall percentage of DNT cells remained unchanged during the course of the study.
The lymphoproliferative response in 34 evaluable subjects was analyzed at different time points with concurrent cytokine production. Interestingly, a significant decrease in Leishmania antigen-specific T cell proliferation against SLA (p = 0.0005) was seen in POST subjects of total PBMC (Fig 2A). These differences persisted when analyzed without the few treatment failures in each group (S5 Fig). However, when analyzed by treatment arm, this decrease in proliferation after therapy was only observed in the SSG but not TM treatment (Fig 2B).
Recent reports suggest that CD8+T cells play a regulatory role in immunity to leishmaniasis [26]. In testing the role of CD8+ cells in proliferation responses PRE and POST, we depleted CD8+ cells from the bulk PBMC prior to stimulation. The proliferation differences between PRE and POST responses were abrogated with CD8+T cell depletion pointing to a potential immunomodulatory or regulatory role for CD8+T cells (Fig 2A). Cytokines were quantified to determine if the suppressive effect of the CD8+T cells involved soluble mediators. Interestingly, IFN-γ, IL-10 and TNF-α were produced at significantly lower levels in POST, whether the CD8+T cells were present or not (Fig 2C, 2D and 2E) which restricts the CD8+T cell effects to modulation of lymphocyte proliferation independent of cytokines tested here.
We next used CFSE labeling to identify antigen-specific proliferating cell subsets in both bulk and CD8+T cell depleted PBMC. Aggregate data is shown in Fig 3Aand 3B and a representative gating example in S6 Fig. While the predominant proliferative fraction consisted of CD4+T cells (68%), there was a modest expansion of CD8+T cells (7%) and CD4-CD8- DNT cells (15%) (Fig 3A). As expected, the vast majority (>90%) of responding cells were activated, as assessed by CD25 expression (Fig 3B).
Based on CD25 expression and the observed modulation of proliferative immune response, we investigated the role of T regulatory (Treg) cells in the healing process. PBMC were analyzed by flow cytometry to determine the levels of activated T cells, identified by CD25 expression. At POST, we observed a decrease in the percentages of circulating activated T cells in both the CD4 and CD8 compartments (Fig 4A). We identified Treg as those cells within the CD4+T cell compartment that expressed the highest levels of CD25 (CD25+ bright) and FoxP3 (S7 Fig). Aggregate data from n = 20 sets of subjects shows that while there was no effect on the Treg population during treatment, there was a marked reduction in circulating CD4+ Treg cells in POST (pre, 3.1%; on, 3.3%; post, 2.3%; p-values = 0.0007 and 0.0036) (Fig 4B).
The degree of protection against various infections including leishmaniasis [7] is predicted by the frequency of polyfunctional CD4+ memory T cells that produce IFN-γ, TNF-α, and IL-2. We assessed intracellular cytokine production by CD4+T cells PRE and POST using multiparameter flow cytometry. First, we were able to independently quantify production of IFN-γ, TNF-α and IL-2 by the CD4+ cells, and observed a significant decrease in production of IFN-γ at POST (Fig 5A). Next, we used Boolean gating to analyze the polyfunctionality of these SLA-specific CD4+T cell responses and found a significant decrease in the frequency of triple positive CD4+ T cells expressing IFN-γ, IL-2 and TNF-α at POST also (Fig 5B). For the subjects that failed to meet the validation criteria in the Boolean gating (minimum 50 events), no values are reported which explains the fewer numbers of points in certain subsets.
Little is known about the cellular phenotypic profile and immune response of humans prior-to and following treatment with different leishmaniasis therapeutic regimens. In this study, we compared the immune response profile in a cohort of L. major infected subjects treated with intravenous SSG or locally applied heat therapy (TM) [21]. The mechanism of actions of these two treatment modalities and the nature, location and distribution of therapy are markedly different. Although both treatments resulted in clinical healing, we hypothesized that an immunomodulating systemic therapy might act through different immune mechanisms compared to a localized, physical, direct parasite-killing therapy.
In this study, we report two important findings with functional immunologic underpinnings. First, downmodulation of Leishmania antigen-specific CD4+T cell proliferative responses possibly through a CD8+T-cell dependent mechanism was observed after therapy. Second, we report that Leishmania-specific polyfunctional CD4+T cells also decrease after therapy.
Since clinical cure from leishmaniasis is classically and primarily dependent on T cell subtypes and relevant cytokine production profiles [27,28], cells were phenotyped from subjects before and after treatment. After treatment and independent of the treatment modality, circulating T cells and NKT-like cells were decreased with a concomitant increase in circulating NK cells highlighting the relevance of the innate immune system for Leishmania control. NK and T cells seemed to have reciprocal effects; wherein NK cell-produced IFN-γ which resulted in T cell activation and the T cell derived IL-2 lead to NK triggering [29]. Similarly, an association between the increased frequency of NK cells and lesion healing is reported after immunotherapy with BCG/Leishmania antigens [30].
NKT-like cells share several characteristics with NK cells [31] and serve as frontline innate immune effectors and potential regulators of adaptive immune responses against microorganisms [32]. Although only a trend, the increase of NKT-like cells observed during treatment could be explained by their ability to serve as an early source of regulatory cytokines and their degranulation-related killing function.
In our T cell subset analysis, we showed a high percentage of CD4+T cells in the early treatment phase, suggesting their association with disease progression [33]; while the percentage of CD8+T cells increased post treatment. This could reflect the down-modulation of the immune response, as a means to mitigate immunopathology, consistent with other studies linking CD8+T cell subset induction with the healing process [26] and lesion resolution during antimonial therapy [34]. Contraction of CD4+T cells and expansion of CD8+T cells during healing suggests CD4 modulation after cure [35]. CD8+T cells were also increased in healed Brazilian CL subjects suggesting potential modulation of the activity of CD4+ cells by direct cytolytic effect of infected macrophages, or by other regulatory effects [33]. Our results confirm that a balance between the proportion of CD4+ and CD8+T cells is important for leishmaniasis healing [33,36–38]. We also analyzed DN T cells, and in particular the αβ subpopulation, a highly activated T cell subset producing cytokines to activate monocytes and macrophages [39]. DN lymphocytes are the second most prevalent cell type producing IFN-γ in human CL [40] and contribute to a leishmanicidal immune environment [39]. DN T cells were recently described as important players in effective and protective primary and secondary anti L. major immunity in experimental cutaneous leishmaniasis [41]. Leishmania-reactive DN T cells express predominantly αβ TCR, are restricted by MHC class II molecules, lack immunoregulatory properties and display transcriptional profile distinct from conventional CD4+ T cells. Current dogma that DN T cells are CD4 and CD8 T cells that have lost their co-receptors is being challenged by the emerging theory that Fas-mediated apoptosis actively removes normally existing DN T cells from the periphery. Impaired Fas-mediated apoptosis may lead to accumulation of these cells rather than de novo generation of DN T cells from activated CD4 or CD8 T cells [42].
In our study, both αβ and γδ subpopulations were similarly represented in the L. major and uninfected control subjects and remained stable during the course of treatment. DN T cell population changes were previously described in human infection with L.(V) braziliensis. In that study, 75% of DN T cells from subjects expressed the αβ TCR compared to uninfected persons where 80% of DN T cells express the γδ TCR [39]. This discordance was not observed here and this may be attributed to different Leishmania species with differing disease patterns and/or genetic backgrounds of the individuals studied.
Leishmania induced immunity is based upon the generation of memory T cells that recognize cognate Leishmania antigens and proliferate after exposure thus activating the effector cells [43]. In our study, responses to SLA were consistently diminished in the post treatment phase. Surprisingly, the proliferative responses were significantly decreased only for subjects receiving systemic treatment but not subjects receiving local treatment. This could in part be explained by the higher numbers of treatment failures at 6 months in TM (4/19 in TM group versus 2/20 in SSG group) causing LPA due to parasite persistence. Similarly to our findings, others also report a decline of the lymphoproliferative response after therapy [28,36–38].
The CD8+T cell-dependent decrease in CD4+T cell proliferation suggested a post treatment, curative type counter-regulatory mechanism. In contrast, in a BALB/c mouse model, CD8 T cell depletion did not interfere with the proliferative ability of draining lymph node CD4 T cells and was associated with an increase in parasite load [44]. As demonstrated for CD4 T cells [45], CD8 immunomodulation maybe due, for example, to up- regulation of Fas expression on CD4 to induce their apoptotic death. We know that CD8 T cells play a role in the healing process and resistance to reinfection in New World human CL. Conversely, other studies associate CD8+ to tissue injury [46]. Recently, it was hypothesized that changes in the frequency of effector CD8+ T cells, during and after antimonial therapy is a critical step to generate an efficient immune response either for by triggering or resolving the lesion [34]. In vivo experiments with human cells showed that CD8 T cells produce IFN-γ and drive Th1 differentiation [47]. However in our study, after treatment, all subjects showed decreased IFN-γ, IL-10 and TNF-α levels, with or without CD8+ depletion. This indicated that CD8+T cell mediated regulation of the CD4+T cell response was not attributable to the soluble mediators studied here. The high IFN-γ production observed pre-treatment suggests that the subjects have initiated an immune response to eliminate the parasite [48]. Additionally, during effective treatment, gradual parasite destruction by macrophages is expected with a diminishing parasite load. Overall, our results add evidence that local heat therapy of CL elicits a systemic cytokine response similar to that of systemic pentavalent antimony. In fact, a decrease in IFN-γ, IL-5 and TNF-α in both groups was seen at day 28 post treatment with meglumine antimoniate in a previous study [49]. These results indicate that proinflammatory responses were progressively downmodulated after therapy and that the cytokine profile produced after cure is shaped during the active phase of disease [50].
Our results were contrary to our hypothesis, as the subjects in the both treatment arms generally exhibited similar cellular immune response profiles. This may be explained, in part, by the tendency of CL to eventually self-heal so cure processes may have occurred despite therapy [51]. Another potential limitation of our study is that there were fewer subjects collected at the 6 month time point, however this was similar between treatment arms. A local immune analysis in the skin may have provided additional clues to immune response alterations induced by different treatments, as might an earlier post timepoint.
Taken together, our findings highlight the existence of regulatory mechanisms that counterbalance early immune responses without altering the CL healing outcome. The magnitude of effector T cell responses can be controlled by regulatory T cells at the lesion site by suppressing lymphocyte proliferation [52]. These mechanisms are important to maintain the host tissue integrity against a subsequent or persistent inflammatory response. Induction of Tregs during chronic infections results from antigen presentation in a particular cytokine environment [53,54]. Interestingly, we found that the percentage of CD25hiCD4+Foxp3+ cells decreased after treatment suggesting that Tregs may be responsible for the suppression that was associated with healing and that their drop is not an artifact of CD4 decrease demonstrated earlier. Tregs have been shown to substantially contribute to tissue repair by providing regulation at sites of healing [55].
To gain a better understanding of the complex immunopathogenesis of CL, study of the quality of a Th1 response, not solely its magnitude, was recently adopted [7,8]. Our analysis evaluated polyfunctional CD4+T cells in response to treatment. Overall, we observed a contraction in polyfunctional CD4+T cells in the post-treatment group, both in terms of number of responding cells and production of multiple cytokines.
In conclusion, healing of CL is a dynamic but consistent process. Similar changes in frequency, quality, and antigen specific responses were observed in both treatment arms and may represent a signature for curative responses.
|
10.1371/journal.pmed.1002871 | An association between maternal weight change in the year before pregnancy and infant birth weight: ELFE, a French national birth cohort study | Weight-control interventions in pregnant women with overweight or obesity have limited effectiveness for fetal growth and birth outcomes. Interventions or prevention programs aiming at the pre-pregnancy period should be considered. However, how the woman’s weight change before pregnancy affects fetal growth is not known. We investigated the association between weight change over the year before pregnancy and birth weight.
We used the inclusion data of 16,395 women from the ELFE French national birth cohort, a nationally representative cohort in which infants were enrolled at birth with their families in 2011. Maternal weight change was self-reported and classified into 3 groups: moderate weight variation or stable weight, weight loss > 5 kg, and weight gain > 5 kg or both weight loss and gain > 5 kg. Multiple linear regression models were used to investigate the association between pre-pregnancy weight change and a birth weight z-score calculated according to the French Audipog reference, adjusted for a large set of maternal characteristics. The analyses were stratified by maternal body mass index (BMI) at conception (<25 versus ≥25 kg/m2) and adjusted for BMI within these categories. We used the MacKinnon method to test the mediating effect of gestational weight gain (GWG) on these associations. Mother’s mean age was 30.5 years, 87% were born in France, and 26% had overweight or obesity. For women in either BMI category at conception, GWG was more than 2 kg higher, on average, for women with weight loss before pregnancy than for women with stable weight or moderate weight variation. For women with BMI < 25 kg/m2 at conception, birth weight was significantly higher with weight loss than stable weight before pregnancy (β = 0.08 [95% CI 0.02; 0.14], p = 0.01), and this total effect was explained by a significant mediating effect through GWG. For women with BMI ≥ 25 kg/m2 at conception, birth weight was not associated with pre-pregnancy weight loss during the year before pregnancy. Mediation analysis revealed that in these women, the direct effect of pre-pregnancy weight loss that would have resulted in a smaller birth weight z-score (β = −0.11 [95% CI −0.19; −0.03], p = 0.01) was cancelled out by the GWG. The mediating effect of GWG was even higher when weight loss resulted from a restrictive diet in the year before pregnancy. Weight gain before pregnancy was not associated with birth weight. Although we included a large number of women and had extensive data, the only potential cause of pre-pregnancy weight loss that was investigated was dieting for intentional weight loss. We have no information on other potential causes but did however exclude women with a history of pre-pregnancy chronic disease. Another limitation is declaration bias due to self-reported data.
Health professionals should be aware that GWG may offset the expected effect of weight loss before conception on fetal growth in overweight and obese women. Further studies are required to understand the underlying mechanisms in order to develop weight-control interventions and improve maternal periconceptional health and developmental conditions for the fetus.
| Maternal obesity is associated with macrosomia at birth and long-term health consequences for the offspring.
Results from weight-control interventions during pregnancy suggest that it is too late to address obesity consequences once the pregnancy has already started and that more attention should be paid to the preconception period.
The impact of maternal weight changes before pregnancy on infant birth weight has not been thoroughly investigated.
We assessed the association between maternal reported weight changes during the year before pregnancy and infant birth weight in the ELFE French national birth cohort.
For women with overweight and obesity, weight loss before pregnancy was associated with increased gestational weight gain compared to women with stable weight before pregnancy, which seemed to cancel out an expected reduction of infant birth weight.
For women with overweight and obesity, weight-control interventions before pregnancy could be beneficial for fetal growth. Health professionals should be aware of a potential weight gain rebound during pregnancy after a weight loss before pregnancy.
Further studies are required to understand the role of weight changes in the preconception period in order to refine nutritional prevention messages for women of childbearing age.
| Reducing adverse pregnancy and fetal outcomes for women with overweight and obesity is a public health priority. Maternal obesity is a risk factor for maternal complications during pregnancy and for infants being large for gestational age (LGA) [1–3]. Also, maternal obesity during pregnancy has been associated with long-term health consequences for the offspring, such as increased body mass index (BMI) during infancy, childhood, and later life and increased risk of type 2 diabetes in adulthood [4,5]. Excessive gestational weight gain (GWG) can also contribute to increased risk of poor maternal and birth outcomes [1,6]. The Institute of Medicine recommends GWG ranges during pregnancy, according to pre-pregnancy BMI category, that are associated with good maternal and infant outcomes [7,8].
Some interventions to prevent or reduce obesity and its consequences have been implemented during the pregnancy period [9]. Lifestyle interventions during pregnancy could reduce GWG [10]. However, further studies have suggested that for women with overweight and obesity, diet and lifestyle interventions during pregnancy have very limited impact on other pregnancy outcomes, birth weight, and overweight risk in offspring [11–13]. These results are consistent with those from observational studies showing that high BMI before pregnancy was a stronger predictor of the risk of LGA than was excessive GWG [14]. Altogether, these results suggest that it is too late to address obesity consequences once the pregnancy has already started and that more attention should be paid to the preconception period.
The periconception period may represent a critical window during which nutritional exposure can influence embryo development and risk of obesity in the offspring. Similar maternal weight status at the start of pregnancy may result from distinct weight trajectories before pregnancy, which reflect particular nutritional and metabolic states. Different dynamics in preconception weight could specifically influence fetal growth and play a distinct role from the effect of nutritional stores during pregnancy. However, few studies have evaluated the association of maternal weight changes before pregnancy and fetal growth. Most human studies on this topic have addressed the impact of inter-pregnancy weight changes [15–17]. Epidemiological data have shown that in obese women, weight gain before pregnancy is associated with an increased risk of complications during pregnancy and macrosomia at birth [15–19]. Further studies have suggested decreased risk of LGA with weight loss between 2 pregnancies or after bariatric surgery for obese women [15,16,20]. However, increased risk of small for gestational age (SGA) has also been observed in association with bariatric surgery before pregnancy [21,22]. Weight loss before pregnancy could help reduce pregnancy and perinatal complications in overweight and obese women, but studies are needed to ensure that it has no harmful side effects. Conversely, women who were underweight at conception showed increased risk of preterm delivery; another study showed, for those who lost weight before pregnancy, a risk of fetal growth restriction and low infant birth weight [23,24]. Hence, the consequences for pregnancy outcomes of weight change before pregnancy may differ according to maternal BMI status at conception.
This study aimed to investigate the association between maternal weight variation in the year before pregnancy and birth weight in a national birth cohort study in France. We hypothesized that maternal weight variation before pregnancy could be involved in the mechanisms programming fetal growth. Weight loss before pregnancy could be clinically relevant in overweight and obese women but not in normal-weight women, so we a priori stratified our analysis according to weight status at the beginning of pregnancy.
This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 STROBE guideline checklist). A brief analysis plan was written and approved before starting statistical analyses (S1 Protocol).
The ELFE study (Étude Longitudinale Française depuis l’Enfance) is a French national longitudinal birth cohort with more than 18,000 children included at birth. The rationale and design of the ELFE cohort were previously detailed [25]. Recruitment took place on 25 selected days during 4 periods in 2011. The inclusion criteria were birth at 33 weeks’ amenorrhea or more, singleton or twin birth, and mother > 18 years, who gave informed consent and did not plan to leave metropolitan France within 3 years. Participation in the cohort was proposed to women who gave birth in 349 maternity hospitals randomly selected among the 544 public and private maternity hospitals in metropolitan France. Among eligible mothers, 51% agreed to participate (N = 18,040). The ELFE study was approved by an ethics committee (Comité de Protection des Personnes), the national committee on information concerning health research (Comité Consultatif sur le Traitement de l’Information en Matière de Recherche dans le domaine de la Santé), and the data protection authority (Commission Nationale de l’Informatique et des Libertés).
We used data collected from the following sources in the maternity wards after birth: medical records (S1 Questionnaire), face-to-face interviews (S2 Questionnaire), and self-administered questionnaires (S3 Questionnaire). A telephone questionnaire answered by the parents at 2 months after birth was also used to complete information on sociodemographic characteristics.
Among the 18,040 mothers, who gave birth to 18,328 infants included in the study, 56 withdrew from the study and asked for data deletion. We excluded 574 twins from the analyses, as well as 175 and 59 mothers with missing medical records and face-to-face interviews, respectively. Finally, we also excluded 842 women with a medical history before pregnancy that may result in weight changes (history of metabolic or endocrine disease, thyroid disease, autoimmune disease, depression, psychological disorder such as anorexia, epilepsy, digestive disease, Crohn disease, bariatric surgery, infectious disease, cancer or congenital anomaly, chronic hypertension, or type 1 or 2 diabetes). Finally, 227 records with missing data on BMI, a key stratification variable, were not considered in the analysis. Fig 1 summarizes the steps of the population selection, which resulted in data for 16,395 women included in the analysis.
Most variables had less than 3% missing data, except for weight changes before pregnancy (collected by the self-administered questionnaire), which had about 13% missing data. Missing data for exposures, outcomes, and confounders were imputed using the SAS “MI” procedure. We compared women with and without a completed self-administered questionnaire and included all variables with significant difference in the imputation process (sociodemographic factors, birth characteristics, and maternal medical history). We generated 5 imputed datasets using the fully conditional specification method (S1 Table). The results of different imputed datasets were combined using the SAS “MI analyse” procedure, and standard errors were calculated using Rubin’s rules, which take into account the variability between the multiple regressions in imputed datasets [28,29].
Characteristics of mothers and their children are described with mean ± SD and frequency (%) before imputation. We compared sociodemographic characteristics between women with different weight variations before pregnancy by chi-squared test for categorical variables and ANOVA for continuous variables, before imputation. Linear regression analyses were used to investigate the association between weight changes in the year before pregnancy and birth weight z-score (hereafter called “birth weight”). Logistic regression was used for analyzing the risk of SGA (SGA/no SGA) and LGA (LGA/no LGA). Women in the weight loss or gain group were compared to women with stable weight. We adjusted our analysis for the following confounders after careful selection based on directed acyclic graphs [30]: level of education, maternal age, smoking before and during pregnancy, place of birth, parity, health insurance coverage, and activity status. We stratified our analysis on weight status (i.e., maternal BMI at conception: <25 versus ≥25 kg/m2) for clinical considerations because weight loss before pregnancy could be advised and beneficial in overweight and obese women but not in normal-weight or underweight women. The interaction between weight variation before pregnancy and weight status on birth weight was significant (complete-case analysis, p = 0.003). We additionally adjusted for BMI before pregnancy as a continuous variable within the 2 BMI categories. All analyses were performed with SAS version 9.3. p < 0.05 was considered statistically significant.
The characteristics of the 16,395 women included in the analysis are summarized in Table 1 for the whole sample and by weight status, before multiple imputation of missing data. The women were mainly born in France (87%) and were employed or students (81%). More than 42% smoked before pregnancy, and 26% were overweight or obese.
For women with BMI < 25 kg/m2 before pregnancy, increased GWG predicted increased birth weight (β = 0.041 [95% CI 0.037; 0.044], p < 0.01). The positive association between weight loss before pregnancy and birth weight (total effect) for a given pre-pregnancy BMI was fully mediated by the indirect effect of GWG, the direct effect being close to 0 (β = −0.008 [95% CI −0.07; 0.05], p = 0.79) (Fig 3).
For women with BMI ≥ 25 kg/m2, increased GWG was also associated with increased birth weight (β = 0.030 [95% CI 0.026; 0.036], p < 0.01). The association between pre-pregnancy weight loss and birth weight was not significant, but when the model was adjusted for GWG, weight loss before pregnancy had a significant negative direct effect on birth weight (β = −0.11 [95% CI −0.19; −0.03], p = 0.01) for a given pre-pregnancy BMI (Fig 4). We found no significant association between weight gain before pregnancy and birth weight in the 2 BMI groups. However, we observed a significant mediating effect of GWG.
The results were even stronger when we restricted the analysis to women with BMI ≥ 30 kg/m2 (N = 1,546), for a direct effect (β = −0.19 [95% CI −0.33; −0.06], p < 0.01) and negative total effect (β = −0.12 [95% CI −0.26; 0.014], p = 0.07). The direct effect was lower for overweight than obese women (β = −0.06 [95% CI −0.17; 0.04], p = 0.24), but we found a persistent and higher effect of GWG in this group (β = 0.10 [95% CI 0.07; 0.13], p < 0.001).
For women who were not overweight or obese at conception, we found a significant positive association between weight loss before pregnancy and birth weight that was totally mediated and explained by increased GWG. For women with overweight and obesity, we did not find an association between weight loss before pregnancy and birth weight. However, after taking into account GWG, weight loss before pregnancy had a negative direct effect on birth weight. GWG seemed to cancel out the expected effect on birth weight reduction of weight loss before pregnancy. For women with weight gain before pregnancy, GWG was increased during pregnancy and had a significant indirect effect on birth weight, but it did not translate into a significant total effect, and there was no direct association between weight gain before pregnancy and birth weight.
In contrast to our results, a previous French study evaluated weight change from age 20 years to pregnancy and showed that weight loss before pregnancy was associated with reduced fetal growth and with risk of SGA for women with BMI < 25 kg/m2 [24]. In this study, pre-pregnancy weight loss reflected a much longer period before pregnancy and less directly reflected the consequence of a recent weight loss, especially in relation to increased GWG. In addition, the analyses were adjusted for GWG, and therefore only tested a direct effect of the previous weight trajectory without considering GWG as a mediator. An American study including more than 10,000 obese women found an increased risk of macrosomia for women with inter-pregnancy weight retention, but weight loss (≥2 BMI units [kg/m2]) between pregnancies was associated with reduced risk of LGA. However, excessive weight loss (>8 BMI units) was related to increased SGA risk [16]. This finding highlights the need to also evaluate the risk of excessive weight loss before pregnancy for fetal development. Another study from a nationwide Swedish cohort also showed a dose-dependent increase in LGA risk with inter-pregnancy weight gain in overweight and obese women. The authors recommended weight loss after a pregnancy in overweight and obese women as well as prevention of weight gain before pregnancy for women with normal BMI [15]. After controlling for BMI at conception, we did not find an association between weight gain before pregnancy and birth weight. Women who self-reported weight gain in the year before conception showed a dynamic of excessive weight gain persisting over the pregnancy. However, their GWG was lower, on average, than that of women who lost weight before pregnancy. Obese women who gain weight before pregnancy are at risk of perinatal complications [15], and they might benefit from increased medical attention and monitoring during pregnancy. The differences between our study and results in the literature could also be explained by different time periods before pregnancy for considering weight change [35].
We have shown that a pre-pregnancy weight loss of >5 kg is associated with reduced birth weight, after taking into account GWG. For a given pre-pregnancy BMI, overweight and obese women with a decreasing weight trajectory may have a better metabolic health and lipid profile at the start of pregnancy [36,37]. Lower frequency of glucose intolerance, hyperinsulinemia, or lipidemia may explain our observed direct effect on decreased birth weight, because all of these factors contribute to fetal growth [38,39]. Moreover, animal studies have shown an association of weight loss and low-fat diet in obese animals with better offspring outcomes, such as reduced fat mass and improved metabolic and hepatic function in offspring [40–42].
However, weight loss is accompanied by several physiological changes such as altered storage of energy and modified nutritional status and hormone pathways involved in the regulation of appetite, which can predispose to weight regain [43]. GWG is a unique and complex mechanism that can be influenced especially by maternal metabolism and placental function [44]. Smoking cessation is also a well-known risk factor for weight gain. In our study, 26% of women who smoked before pregnancy stopped smoking during pregnancy. When we restricted our analyses to overweight and obese women with no history of smoking, our main result was similar: We observed a negative direct effect of weight loss before pregnancy on birth weight and a positive indirect effect of GWG.
These physiological conditions may facilitate the mechanisms of weight regain after caloric restriction. Other factors related to pregnancy such as a decrease in physical activity may also elicit weight regain.
The strengths of our study are the large number of women included and the availability of sociodemographic and medical data. We had information on the medical history of women and excluded women with diseases, psychological disorders, bariatric surgery, and metabolic status that could have affected weight change before pregnancy. The use of multiple imputation techniques limited bias due to missing data in variables of interest and confounders.
Among the limitations is the questionable accuracy of data reported by the women on their weight history before pregnancy. However, our analysis focused on large weight changes (± 5 kg) in a recent period (1 year before pregnancy). The accuracy would be similar for information collected by a heath professional at the start of pregnancy, and we feel that our results are relevant to a clinical situation. Another limitation is that our information on the potential cause of weight loss is limited to the practice of a restrictive diet, and information on the type of diet and dieting period in the year before pregnancy was lacking. We could not study whether our observed direct effect of weight loss on birth weight was due to the weight change itself or the modification of the maternal lifestyle driving it.
Using total GWG, including the baby’s weight, is also a study limitation. When adjusting for total GWG, we took out part of the association we were trying to assess between pre-pregnancy weight variation and birth weight. To circumvent this drawback, in a complementary analysis, we approached the maternal component of GWG by subtracting the child’s birth weight from the total GWG. We assumed that although GWG also includes placenta weight and amniotic fluid, the main source of variability of GWG minus birth weight was the maternal component. The results from this complementary analysis were similar to those of the main analysis, which supports our interpretation of the results.
Finally, we did not have information on intermediate weight measurements during pregnancy, which prevented us from assessing weight gain for specific periods or trimesters. GWG is not linear across gestation, and the rate of weight gain is slower in the first trimester [45]. Distinct kinetic GWG may have differential impact on fetal growth [46,47].
Only live births are included in the ELFE cohort. If weight change in the year before pregnancy affects the probability of miscarriage or stillbirth, our results could be biased, probably by underestimating the effect of extreme weight changes on fetal growth. A selection bias could also be possible if there were associations between, on the one hand, both fetal growth and weight change before pregnancy and, on the other, the probability of inclusion in the ELFE cohort. Except for low maternal education level, which was controlled for, we could not see any obvious reason why weight change before pregnancy would influence the probability of inclusion.
Few studies have evaluated the association between pre-pregnancy weight variation and fetal development. Our results suggest that increased GWG after weight loss before pregnancy may obscure any beneficial effect on fetal growth. These results call for increased vigilance on GWG in women who lost weight or dieted before pregnancy.
|
10.1371/journal.ppat.1000756 | Exacerbated Innate Host Response to SARS-CoV in Aged Non-Human Primates | The emergence of viral respiratory pathogens with pandemic potential, such as severe acute respiratory syndrome coronavirus (SARS-CoV) and influenza A H5N1, urges the need for deciphering their pathogenesis to develop new intervention strategies. SARS-CoV infection causes acute lung injury (ALI) that may develop into life-threatening acute respiratory distress syndrome (ARDS) with advanced age correlating positively with adverse disease outcome. The molecular pathways, however, that cause virus-induced ALI/ARDS in aged individuals are ill-defined. Here, we show that SARS-CoV-infected aged macaques develop more severe pathology than young adult animals, even though viral replication levels are similar. Comprehensive genomic analyses indicate that aged macaques have a stronger host response to virus infection than young adult macaques, with an increase in differential expression of genes associated with inflammation, with NF-κB as central player, whereas expression of type I interferon (IFN)-β is reduced. Therapeutic treatment of SARS-CoV-infected aged macaques with type I IFN reduces pathology and diminishes pro-inflammatory gene expression, including interleukin-8 (IL-8) levels, without affecting virus replication in the lungs. Thus, ALI in SARS-CoV-infected aged macaques developed as a result of an exacerbated innate host response. The anti-inflammatory action of type I IFN reveals a potential intervention strategy for virus-induced ALI.
| Severe acute respiratory syndrome coronavirus (SARS-CoV) infection causes acute lung injury that may develop into the life-threatening acute respiratory distress syndrome (ARDS) in mostly elderly individuals. Although SARS-CoV infection can be fatal, most patients recover, suggesting that protective host responses are operational to combat the viral infection. Therefore, we used age as predisposing factor to obtain insight into the pathogenesis of SARS-CoV. In this study, we show that SARS-CoV-infected aged macaques developed significantly more pathology than young adult animals, which could not be contributed to differences in viral replication. Using comparative microarray analyses, it was shown that although the nature of the host response to SARS-CoV infection was similar in aged and young adult macaques, the severity was significantly different, with aged macaques displaying an increase in differential expression of genes associated with inflammation. Interestingly, type I IFN-β mRNA levels correlated negatively with gross pathology. Therapeutic treatment of aged macaques with type I IFN reduced pathology without affecting virus replication. However, pro-inflammatory gene expression was significantly diminished. Thus, modulation of the host response by type I IFNs provides a promising outlook for novel intervention strategies.
| The zoonotic transmission of severe acute respiratory syndrome coronavirus (SARS-CoV) caused pneumonic disease in humans with an overall mortality rate of ∼10%. The exact reasons why some individuals succumbed to the infection while others remained relatively unaffected have not been clarified. Aging, an important risk factor in SARS-CoV-associated disease, is associated with changes in immunity [1],[2],[3]. Consequently, elderly individuals are at greater risk of contracting more severe and longer lasting infections with increased morbidity and mortality, exemplified by respiratory tract infections caused by influenza A virus and severe acute respiratory syndrome (SARS) coronavirus [4],[5],[6]. The clinical course of SARS-CoV-induced disease follows a triphasic pattern [5]. The first phase is characterized by fever, myalgia and other systemic symptoms that are likely caused by the increase in viral replication and cytolysis. The second phase of the disease is characterized by a decrease in viral replication that correlates with the onset of IgG conversion. Interestingly, it is also in this phase that severe clinical worsening is seen, which can not be explained by uncontrolled viral replication. It has been hypothesized that the diffuse alveolar lung damage in this phase is caused by an over exuberant host response [5],[7],[8]. The majority of patients recovers after 1–2 weeks, but up to one-third of the patients progress to the third phase and develop severe inflammation of the lung, characterized by acute respiratory distress syndrome (ARDS) [9]. The clinical course and outcome of SARS-CoV disease are more favorable in children younger than 12 years of age as compared to adolescents and adults [10],[11],[12]; elderly patients have a poor prognosis, with mortality rates of up to ∼50% [5],[6].
For SARS-CoV-associated disease in humans, it has been hypothesized that seemingly excessive pro-inflammatory responses, illustrated by elevated levels of inflammatory cytokines and chemokines, mediate immune-pathology resulting in acute lung injury (ALI) and ARDS [5],[13],[14],[15],[16]. Direct support for this concept, however, is scarce. ALI and ARDS are typified by inflammation, with increased permeability of the alveolar-capillary barrier, resulting in pulmonary edema, hypoxia, and accumulation of polymorphonuclear leukocytes and macrophages. Inflammatory cytokines, among which IL-1β and IL-8, play a major role in mediating and amplifying ALI/ARDS [9] and are elevated in SARS-CoV-infected patients as well [13],[14]. In vitro experiments confirm that SARS-CoV infection induces expression of cytokines/chemokines in a range of cell types [15],[17],[18]. Moreover, infection of cynomolgus macaques with SARS-CoV leads to a strong immune response, with expression of various cytokines/chemokines, resembling the host response seen in human SARS patients [19]. Nevertheless, the determinants that lead to severe virus-associated ALI/ARDS and that cause people to succumb to infection remain largely obscure, restraining development of appropriate treatments.
As advanced age is a predictor of adverse clinical outcome in both ARDS and SARS-CoV infections [5],[20], we used age as predisposing factor to study the pathogenesis of SARS-CoV in a macaque model. By performing comparative analyses of young adult and aged SARS-CoV-infected macaques regarding pathology, virus replication and host response, insight into the pathogenesis of SARS-CoV is obtained and a potential therapeutic intervention strategy for virus-induced ALI is revealed.
To obtain further insights in the pathogenesis of SARS-CoV, six aged (10–19 years old) and six young adult (3–5 years old) cynomolgus macaques were infected with SARS-CoV HKU39849 and euthanized four days after infection. Four young adult and four aged PBS-infected cynomolgus macaques were used as negative controls. During the 4-day experiment, some of the SARS-CoV-infected aged animals displayed decreased activity and mildly labored breathing. All aged infected macaques showed an increase in body temperature either during the night or during the day one to two days after infection (Figure 1A). The lungs of aged macaques showed large (multi)focal pulmonary consolidation that was severe (∼40–60% of affected lung tissue) in two macaques (Figure 1B and Figure S1). Microscopic examination revealed typical ALI-associated lesions, similar to what has been seen in SARS-CoV-infected humans that progress to ARDS [6]. Lesions involved the alveoli and terminal bronchioli, showing areas with acute or more advanced phases of diffuse alveolar damage (Figure 2). Lumina of alveoli were variably filled with protein rich edema fluid, cellular debris, alveolar macrophages and neutrophils, eosinophils, and lymphocytes (Figure 2 and Figure S2A–B). Moderately thickened alveolar walls were lined by cuboidal epithelial cells (type 2 pneumocyte hyperplasia; Figure 2 and Figure S2C). The epithelial origin of these enlarged type 2 pneumocytes with large vacuolated nuclei, prominent nucleoli and abundant vesicular cytoplasm was confirmed by keratin staining (Figure S2D). Hyaline membranes and multinucleated giant cells were occasionally observed in the alveoli (Figure S2E–F). In contrast, all young adult animals remained free of clinical symptoms and had no or less extensive pulmonary consolidation (Figure 1A–C). Hyaline membranes were not observed in SARS-CoV-infected young adult macaques. A multifocal mild chronic lymphoplasmacytic tracheo-bronchoadenitis, characterized by moderate numbers of lymphocytes, plasma cells, macrophages, less neutrophils and occasional eosinophils in the lamina propria of the bronchi, focally surrounding and infiltrating the submucosal glands, was observed in all young adult macaques, but not in aged macaques (Figure 2). Our data were confirmed by retrospective analysis of earlier experiments in which aged animals were used [19],[21],[22],[23]. Overall, aged macaques develop more severe SARS-CoV-associated ALI than young adults.
Because viral replication is important for disease pathogenesis, we determined virus titers in aged and young adult animals. Virus excretion in the throat (Figure 3A) and nose (Figure 3B) of aged and young adult macaques at days 2 and 4 post infection was not significantly different. Moreover, no significant difference in quantity of SARS-CoV mRNA in the lungs of young adult and aged animals was observed (Figure 3C). Differences in the nature and percentage of SARS-CoV-infected cells in the lungs of aged and young adult macaques were not seen either (Figure 3D). Apparently, augmented pathology in aged macaques cannot be rationalized by increased viral replication.
To understand why SARS-CoV-infection in aged macaques results in more severe pathology than in young adult macaques, we determined global gene expression profiles by analyzing total RNA isolated from the lungs using microarray analysis. Hierarchical clustering methods were used to order gene transcripts and individual aged and young adult animals to identify groups of animals with similar expression patterns. These data were plotted as a heat map in which each entry represents a gene expression value (Figure 4). As expected in an animal experiment with outbred animals, the inter-animal variation was relatively high (Figure S3). There were two major roots to the hierarchical dendogram, with one root containing the PBS-infected control animals, and the second root containing the SARS-CoV-infected animals. The root of the PBS-infected control animals was divided in two minor roots, clustering young adults together and aged animals as a group. These data suggest that the baseline expression patterns are different in young adult and aged macaques. The root of the SARS-CoV-infected animals was also divided in two minor roots, largely clustering young adult animals together and grouping aged infected macaques. The hierarchical clustering heat map suggests that both age and SARS-CoV infection are key factors involved in determining transcription of cellular genes.
To determine whether aged and young adult animals respond differently to SARS-CoV-infection, their gene expression profiles were compared. In a direct comparison of aged (n = 6) versus young adult (n = 6) SARS-CoV-infected animals using an ANOVA-based analysis called LIMMA, 202 gene transcripts were differentially expressed (fold change ≥2; p<0.05; Table S1). Upon analysis of these gene transcripts within the context of biological processes and pathways using Ingenuity Pathways Knowledge Base, this subset of genes showed indications for an innate host response to viral infection. Among the top significantly differentially regulated (p<0.005) functional categories were immune response, inflammatory response and hematological system development and function, which included genes like F3, IL1RL1, IL1RN, IL6, IL8, S100A8, SERPINA1, SERPINA3, NP, ACPP, TFPI2, SPP1, IGF1, EDN3, DEFB1, and SOCS3 (Figure 5A) most of which were upregulated in SARS-CoV-infected aged animals compared to young adult infected animals. In addition, three of the most significantly regulated molecular/cellular functions (p<0.005) were associated with a pro-inflammatory response and included cell death, cell movement, and cell-to-cell signalling (Figure 5A). The top gene interaction network, showing the interplay between genes during the host response to viral infection, contained NF-κB as central node (Figure 5B). NF-κB is a redox-sensitive transcription factor implicated to play a major role in pro-inflammatory host responses and the development of ALI/ARDS [24],[25]. Several of the 202 differentially expressed gene transcripts, among which IL1RN, SERPINA1, IL8, F3 and TFPI2, are target genes for NF-κB. Thus, significant differences exist in the host response to SARS-CoV infection, corresponding with age.
To obtain a more in-depth view of the host response to infection, global gene expression profiles were determined in lungs of SARS-CoV-infected aged (n = 6) or young adult (n = 6) macaques in comparison to aged or young adult PBS-infected macaques (n = 4), respectively. Aged macaques differentially expressed 1577 gene transcripts (Figure 6A). Gene ontology analysis revealed that the majority of genes in the aged macaque group compared to aged PBS-infected animals were associated with a pro-inflammatory response and included cellular growth and proliferation, cell death, cell movement, and cell-to-cell signalling (Figure 6B). Although SARS-CoV-infected young adult macaques differentially expressed much less gene transcripts compared to young adult PBS-infected animals (Figure 6A), the most significantly regulated molecular/cellular functions also included cellular growth and proliferation, cell death, cell movement, and cell-to-cell signalling (Figure 6B). This suggested that the nature of the host response to infection in aged and young adult animals was strikingly similar, even though the severity was different.
Because the above described gene ontology molecular/cellular functions are very broad, genes were further subdivided based on available annotations to gain insight in differences in the host response to infection in aged and young adult macaques compared to aged and young adult PBS-infected animals, respectively. Heat maps were generated for differentially regulated genes with pro-inflammatory functions such as cell adhesion (Figure 6C), apoptosis (Figure 6D), and cytokine/chemokine signalling (Figure 6E). The greater number of differentially expressed genes, as well as the brighter intensities (fold changes in transcripts) included in the heat map for aged macaques, suggested that aged macaques show a more zealous response to virus infection than young adult macaques. This assumption was corroborated using Goeman's global test [26] on the defined gene subsets cell adhesion, cytokine/chemokine signalling, and apoptosis. When macaques were grouped according to severity of pathology instead of age and compared to their respective PBS-infected controls, increased numbers of differentially expressed gene transcripts and increased fold changes for differentially expressed genes in inflammatory pathways correlated positively with gross pathology scores as well (Figure S4). Our data show that the innate host response to SARS-CoV infection changes during aging in macaques; age, pathology, and pro-inflammatory host response go hand-in-hand.
In order to understand the host responses in the context of senescence, we directly compared lung samples from PBS-infected aged (n = 4) and young adult (n = 4) macaques. LIMMA analysis revealed that 518 gene transcripts were differentially expressed (fold change ≥2; p<0.05), with categories such as immunological disease, haematological system and development, cell death, cell movement, and cellular growth and proliferation among the top significantly differentially regulated functions (p<0.005). Only 14 out of the 518 differentially expressed gene transcripts were also differentially expressed in the direct contrast of SARS-CoV-infected aged and young adult macaques. Our data indicate that significant differences exist in the basal gene expression levels of aged and young adult macaques, which may partly explain why differences in pathology were observed after SARS-CoV infection.
As NF-κB target genes were differentially regulated in the direct comparison of SARS-CoV-infected aged and young adult macaques (Figure 5), we focussed on NF-κB in the indirect comparison of aged and young adult SARS-CoV-infected macaques compared to aged and young adult PBS-infected animals, respectively. A gene interaction network, showing the interplay between “immune response”-type genes with NF-κB as central node, revealed that aged SARS-CoV-infected macaques showed a much more robust regulation of these genes than young adult infected animals (Figure 7A) compared to their respective PBS-infected animals, which was corroborated by an analysis of differentially expressed target genes of NF-κB (Figure 7B). Several of these genes, among which VCAM1, F3, PTX3, and IL-8, have also been implicated in development of ARDS (Figure 7C) [9],[27],[28].
In order to visualize NF-κB-signalling in the lungs of SARS-CoV-infected aged and young adult macaques, translocation of NF-κB was studied using immunohistochemistry with antibodies against phosphorylated NF-κB on day 4 after infection. As shown in Figure 7D, hardly any phosphorylated NF-κB could be detected in the nuclei of cells of PBS-infected macaques, while in the lungs of SARS-CoV-infected animals, cells with phosphorylated NF-κB in their nuclei were abundantly present. Phosphorylated NF-κB was detected primarily in the nuclei of non-infected cells (Figure 7D). No obvious differences in the translocation of NF-κB in the lungs of aged and young adult macaques were observed.
Overall, our data indicate that SARS-CoV-infected aged macaques display a stronger pro-inflammatory host response to infection than young adult macaques. For example, mRNA levels for IL8, a key player in ALI/ARDS and a potent chemotactic factor essential in acute inflammation that is induced by a wide range of stimuli among which IL1β, viral products, and oxidative stress, were strongly upregulated in SARS-CoV-infected aged macaques as compared to young adult animals (Figure 5B, 7B, 8A). Despite the overall stronger activation of innate host gene responses in SARS-CoV-infected aged animals, microarray analyses revealed that IFN-β, well-known for its antiviral activities, was not differentially expressed in aged macaques compared to PBS-infected animals, in contrast to young adults (Figure 6E). RT-PCR analysis confirmed differential expression of IFN-β mRNA between young adult and aged macaques (Figure 8B). As shown in Figure 3, this difference in IFN-β levels in aged and young adult macaques did not affect viral replication efficiency. IFN-β mRNA levels, however, negatively correlated with gross pathology (Figure 8C).
The observation of a reverse correlation of IFN-β and IL-8 mRNA levels with age after SARS-CoV infection may reflect a physiological cross-regulation in which type I interferon and/or its respective signalling pathways modulate pro-inflammatory host responses [29],[30]. To corroborate this hypothesis, we treated uninfected human PBMC with IL-1β, which is known to rapidly activate NF-κB-signalling [31],[32], and observed the induction of pro-inflammatory cytokines in uninfected human PBMC, such as IL-1β and IL-8 (Figure 8A–B). An anti-inflammatory effect of pegylated IFN-α on IL-1β-induced responses was confirmed in vitro, as a dose-dependent inhibition of IFN-α on recombinant IL-1β-induced IL-1β and IL-8 mRNA levels in human PBMC was observed (Figure 9A–B).
Because type I IFNs can inhibit pro-inflammatory signalling pathways, among which NF-κB signalling pathways [29],[30], we examined whether exogenous administration of type I IFN in SARS-CoV-infected aged macaques could influence SARS-CoV pathogenesis. Retrospective analyses of the lungs of SARS-CoV-infected aged animals treated therapeutically with type I IFN [22] showed that SARS-CoV-infected IFN-treated aged animals remained free of clinical symptoms and had no or less extensive pulmonary consolidation than untreated aged macaques (Figure 10A). Virus titers in the lungs, however, were similar between IFN-treated and untreated aged macaques (Figure 10B) and viral antigen expression in the lungs was not significantly different [22].
In a direct comparison of the host response to infection in aged (n = 6) versus IFN-treated aged (n = 3) macaques using LIMMA, 961 gene transcripts were differentially expressed (fold change ≥2; p<0.05). Upon analysis of these gene transcripts within the context of genetic pathways, four of the most significantly regulated molecular/cellular functions (p<0.005) were associated with a pro-inflammatory response and included cellular growth and proliferation, cell death, cell movement, and cell-to-cell signalling, indicating that significant differences exist in the host response to SARS-CoV infection in animals treated with type I IFN compared to untreated aged macaques.
To obtain a broader view of the host response to infection, global gene expression profiles were determined in lungs of SARS-CoV-infected aged (n = 6) or IFN-treated aged macaques (n = 3) in comparison to aged PBS-infected macaques (n = 4). IFN-treated macaques differentially expressed (fold change ≥2; p<0.05) approximately four-fold less gene transcripts than untreated aged macaques (Figure 10C) as compared to PBS-infected animals. The most significantly regulated molecular/cellular functions in the IFN-treated macaque group compared to PBS-infected animals were associated with a pro-inflammatory response and included cellular growth and proliferation, cell death, cell movement, and cell-to-cell signalling, similar to what was observed for the aged macaque group, although less genes per function were differentially expressed (Figure S5A). These data suggested a common nature of the host response to infection in aged and IFN-treated aged animals, although the severity seemed different.
To gain more insight in differences in the host response to infection in aged and IFN-treated macaques compared to PBS-infected animals, heat maps were generated for differentially regulated genes involved in pro-inflammatory pathways apoptosis (Figure S5B) and cell adhesion (Figure S5C). Using Goeman's global test [26] on the defined gene subsets cell adhesion and apoptosis, significant differences between aged and IFN-treated animals in these pro-inflammatory pathways were obtained, providing statistical evidence for a difference in host response of aged and IFN-treated animals to SARS-CoV infection. Moreover, a decrease in differentially expressed target genes of NF-κB was observed (Figure S5D). Most notably, a dramatic decrease in the expression of cytokine/chemokine mRNA levels was observed, among which IL-8 (Figure 10D, Figure S5B–C). These data show that therapeutic treatment of SARS-CoV-infected aged macaques with type I IFN primarily results in downregulation of pro-inflammatory host responses.
The present study aimed at gaining insight into the pathogenesis of SARS-CoV by studying the relationship between age, pathology, virus replication, and host response in a macaque model. In humans, SARS-CoV infection progresses from an atypical pneumonia to acute diffuse alveolar damage and ARDS [5]. The overall human fatality rate reached ∼10% and up to 50% in elderly [5],[6]. The acute lung injury observed after SARS-CoV infection in aged macaques is similar to what has been seen in humans that progress to ARDS [6]. This disease process includes an acute exudative phase, consisting of severe leukocyte infiltration, edema, the formation of hyaline membranes, and proliferation characterized by type II pneumocyte hyperplasia [33]. SARS-CoV-infected aged macaques develop more severe pathology than young adult animals, even though viral replication levels are similar. The chronic phase, which is characterized by persistent intra-alveolar and interstitial fibrosis and mortality was not observed because animals were sacrificed early after infection.
Comparative analyses of gene expression in aged and young adult SARS-CoV-infected macaques revealed that the host response to SARS-CoV infection is similar in nature, but differs significantly in severity in pro-inflammatory responses. Aged macaques had a stronger host response to virus infection than young adult macaques, with an increase in differential expression of genes associated with inflammation that center around the transcription factor NF-κB. Comparative analysis of PBS-infected aged and young adult macaques revealed significant differences in gene expression as a result of aging only. These observations are in line with earlier hypotheses that age-related accumulated oxidative damage and a weakened antioxidative defense system cause a disturbance in the redox balance, resulting in increased reactive oxygen species. Subsequently, the oxidative stress-induced redox imbalance activates redox-sensitive transcription factors, such as NF-κB, followed by the induction of pro-inflammatory genes including IL1β, IL6, TNFα and adhesion molecules, key players in the inflammatory process [34]. Oxidative stress may also potentiate the cellular responses to IL-1β [35], an early mediator of inflammation [36]. Thus, aging is associated not only with alterations in the adaptive immune responses, but also with a pro-inflammatory state in the host [34],[37],[38],[39]. Oxidative stress and toll-like receptor-4 signaling via NF-κB triggered by viral lung pathogens, such as SARS-CoV, may further amplify the host response ultimately resulting in ALI [25]. Taking the host gene expression profiles of PBS-infected aged and young adult macaques into account, we also observed a stronger activation of the pro-inflammatory pathways in SARS-CoV-infected aged macaques than in young adults. The finding that genes activated by NF-κB are significantly differentially upregulated in aged macaques infected with SARS-CoV is in line with the role of NF-κB as a redox-sensitive transcription factor in pro-inflammatory host responses and the development of ALI/ARDS [24],[25]. Given the fact that several SARS-CoV proteins block NF-κB signaling [40],[41],[42], we hypothesize that NF-κB-signaling in non-infected cells is largely responsible for the upregulated expression of NF-κB target genes, such as IL8, in aged compared to young adult macaques.
These observations are largely in line with transcriptome analyses in mice and SARS patients. In severe SARS patients, cytokines/chemokine involvement as the illness progresses may lead to widespread immune dysregulation and serious pathogenic events [13]. Aged mice show more pathology than young adult mice and the transcriptional profile in aged mice generally indicates a more robust pro-inflammatory response to virus infection than in young mice [43],[44].
Previously, we demonstrated IFN induction and signalling in SARS-CoV-infected macaques early after infection [19]. Based on the observation that plasmacytoid dendritic cells are able to produce type I IFNs after SARS-CoV infection in vitro [45], it was speculated that these cells are the IFN-producing cells in lungs of SARS-CoV-infected macaques. In addition, phosphorylated STAT-1 was observed in the nuclei of numerous cells in the lungs of SARS-CoV-infected macaques, indicating that these cells had been activated by IFNs or other agonists produced in the lung [19]. In SARS-CoV-infected cells, however, STAT-1 signalling was blocked [19], consistent with the fact that a range of SARS-CoV proteins can function as interferon antagonists that inhibit IFN production and signalling [41],[42]. Therefore, a large part of the genes activated downstream of STAT-1, observed in genomics analyses, is likely due to signalling in non-infected cells [19]. In the current study, we observed that aged macaques expressed significantly lower levels of IFN-β mRNA than young adult macaques and that IFN-β mRNA levels correlated negatively with severity of pathology. Interestingly, aged and young adult SARS-CoV-infected macaques showed opposite expression patterns for type I IFN-β and certain pro-inflammatory cytokines, such as IL-8. These data are corroborated by previous observations showing that higher amounts of pro-inflammatory cytokines, such as IL-1β and IL-8, are produced upon stimulation of leukocytes of the elderly, whereas induction of type I IFNs is decreased compared to young adults [46],[47],[48].
The observation of a reverse correlation of IFN-β and IL-8 mRNA levels with age after SARS-CoV infection may reflect a physiological cross-regulation between antiviral STAT-1 and proinflammatory NF-κB pathways. Evidence for such a cross-regulation between type I IFN/STAT-1 and pro-inflammatory/NF-κB signaling pathways exists. Type I interferons exert significant anti-inflammatory effects and provide at least partial protection from disease in collagen-induced arthritis, auto-immune encephalitis, and multiple sclerosis [49],[50],[51],[52],[53]. Not only inhibits IFN-beta expression of the IL8 gene at the transcriptional level [54], type I IFNs can also activate TAM receptor tyrosine kinases that inhibit toll-like receptor-induced cytokine-receptor cascades [55],[56] and induce the immunosuppresive cytokine IL-10 [57]. Direct NF-κB/STAT-1 protein-protein interactions [58] and modification of STAT-1 by acetylation, may be involved in this process [30]. A loss of type I IFN/STAT-1 signaling in aged macaques may negatively regulate interferon-induced gene expression and type I IFN signaling, which may lead to enhanced inflammatory responses. On the other hand, increased activation of NF-κB signaling pathways in aged macaques may negatively regulate interferon-induced gene expression and type I IFN signaling [29],[59],[60], which may enhance pro-inflammatory responses even further.
We have integrated our data and other findings on cross-regulation in a model (Figure 11). The model depicts the innate immune response to SARS-CoV infection as a coordinated series of signaling pathways aimed at clearing the virus while not harming host cells. Upon SARS-CoV infection, infected cells, depicted in the model as pneumocytes, produce inflammatory mediators that activate NF-κB, resulting in the production of pro-inflammatory cytokines and chemokines, such as IL-8. IL-1 is one of the cytokines highly upregulated on day 1 after infection upon SARS-CoV infection of macaques [19] and capable of activating NF-κB. At the same time, the virus is recognized by sentinel cells, such as pDCs, that produce type I IFNs to signal that a foreign invader has entered the host. The production of IFN induces neighboring non-infected cells to remodel the intracellular environment by producing a range of antiviral proteins, aiding in a block of viral replication. A cross-regulation between the “antiviral” and “pro-inflammatory” pathways occurs, which is a critical requirement to allow fine-tuning of the host response to infection and return to homeostasis. Disease outcome may be determined by the relative contribution of “antiviral” and “pro-inflammatory” pathways and apparently aging influences this intricate balance significantly.
Causal relationships between “antiviral” and “pro-inflammatory” pathways in macaques are difficult to prove and future studies in specific gene knock-out mice should therefore further clarify the complex interactions in the response to SARS-CoV. Our own in vitro experiments and the type I IFN intervention in SARS-CoV-infected aged macaques indicate that type I IFNs can play a role in mitigating pro-inflammatory host responses and severity of pathology. Therapeutic treatment of SARS-CoV-infected aged macaques with type I IFN reduces pathology and diminishes pro-inflammatory gene expression, including IL-8 levels, without affecting virus replication in the lungs. Antiviral effects of type I IFNs were not obvious, probably due to the fact that SARS-CoV infected cells inhibit STAT-1 signalling and viral replication peaks early after infection when treatment with pegylated IFN-α started. Given the fact that phosphorylated NF-κB was present mainly in the nuclei of non-infected cells in the lungs of SARS-CoV-infected macaques, these cells are potential targets for the action of IFN and subsequent STAT-1 signalling. It remains uncertain whether endogenously produced IFNs in young adult macaques are essential in the control of inflammatory responses or that enhanced activation of inflammatory pathways simply does not occur. Our data are in line with the observation that treatment of SARS-CoV-infected aged mice with type II IFN-γ, which like type I IFN also signals via STAT-1, protected against lethal respiratory illness, seemingly without an effect on viral replication [61]. Moreover, in humans with SARS, use of type I IFNs was associated with reduced disease-associated hypoxia and a more rapid resolution of radiographic lung abnormalities [62]. Whether the anti-inflammatory action of type I/II IFNs in macaques, mice and humans occurs via common pathways and is interchangeable between host species remains to be determined. Assuming that there is a conserved pathway in ALI/ARDS induced by multiple pathogens, including pandemic viruses that may emerge from avian influenza, modulation of the host response by type I IFNs provides a promising outlook for novel intervention strategies.
Six young adult cynomolgus macaques (Macaca fascicularis), 3–5 years old, four of which carried active temperature transponders in the peritoneal cavity, and four aged cynomolgus macaques, 10–18 years old, which all carried active temperature transponders, were inoculated with SARS-CoV strain HKU39849, as described previously [19],[21],[22],[23]. Two additional aged animals (17 and 19 years old), previously infected with SARS-CoV strain HKU39849 [22], were enrolled in this study as well. Four young adult mock (PBS) infected animals from a previous study [19] and four aged macaques were taken as controls. Lung tissues stored in RNA-later from three cynomolgus macaques, 13 years old, previously inoculated with SARS-CoV strain HKU39849 and treated with pegylated IFN-α at a dose of 3 µg/kg intramuscularly on days 1 and 3 after infection, were taken along for molecular analyses [22]. All animals were infected with the same dose of virus, using the same inoculation procedure, by the same person to minimize inter-experiment variation. All animals were checked daily for clinical signs and anaesthetised with ketamine on days 0, 2 and 4 after infection to collect oral, nasal, and rectal swabs [22]. All animals were euthanized on day 4 post infection. Necropsies and sampling for histology/immunohistochemistry were performed as described [22]. The percentage of affected lung tissue from each lung lobe was determined at necropsy, recorded on a schematic diagram of the lung and the area of affected lung tissue was subsequently calculated (gross pathology score).
Approval for animal experiments was obtained from the Institutional Animal Welfare Committee and performed according to Dutch guidelines for animal experimentation.
Serial 3 µm lung sections were stained using mouse-anti-SARS-nucleocapsid IgG2a (clone Ncap4; Imgenex) 1∶1600, mouse-anti-human neutrophil elastase (clone NP-57; DAKO) 1∶10, mouse-anti-human CD68 (clone KP1; DAKO) 1∶200, mouse-anti-human pankeratin (clone AE1/AE3; Neomarkers) 1∶100, rabbit anti p-NF-κB p65 (Santa Cruz) or rabbit control and isotype antibodies (clones 11711 and 20102; R&D), according to standard protocols [22],[23]. Quantitative assessment of SARS-CoV infection in the lungs was performed as described previously [22].
RNA from 200 µl of swabs was isolated with the Magnapure LC total nucleic acid isolation kit (Roche) external lysis protocol and eluted in 100 µl. SARS-CoV RNA was quantified on the ABI prism 7700, with use of the Taqman Reverse Transcription Reagents and Taqman PCR Core Reagent kit (Applied Biosystems), using 20 µl isolated RNA, 1× Taqman buffer, 5.5 mM MgCl2, 1.2 mM dNTPs, 0.25 U Amplitaq gold DNA polymerase, 0.25 U Multiscribe reverse transcriptase, 0.4 U RNAse-inhibitor, 200 nM primers, and 100 nM probe [23]. Amplification parameters were 30 min at 48°C, 10 min at 95°C, and 40 cycles of 15 s at 95°C, and 1 min at 60°C. RNA dilutions isolated from a SARS-CoV stock were used as a standard. Average results (±s.e.m.) for young adult (n = 6) and aged macaque (n = 4) groups were expressed as SARS-CoV equivalents per ml swab medium.
Lung tissue samples (0.3–0.5 gram) were taken for RT-PCR and microarray analysis in RNA-later (Ambion, Inc.). RNA was isolated from homogenized post mortem tissue samples using Trizol Reagent (Invitrogen) and the RNeasy mini kit (Qiagen). cDNA synthesis was performed with 1 µg total RNA and Superscript III RT (Invitrogen) with oligo(dT), according to the manufacturer's instructions. Semi-quantitative RT-PCR was performed to detect SARS-CoV mRNA and to validate cellular gene expression changes as detected with microarrays [19]. Differences in gene expression are represented as the fold change in gene expression relative to a calibrator and normalized to a reference, using the 2−ΔΔCt method [63]. GAPDH (glyceraldehydes-3-phosphate dehydrogenase) was used as endogenous control to normalize quantification of the target gene. The samples from the young adult PBS-infected macaques were used as a calibrator. Average results (±s.e.m.) for young adult (n = 6), aged (n = 6), and IFN-α-treated aged (n = 3) macaque groups were expressed as fold change compared to young adult PBS-infected animals, respectively [63]. In addition, groups were based on severity of pathology: young adult macaques (n = 6), aged macaques with pathology (n = 4), and aged macaques with severe pathology with >40% of lungs affected (n = 2) (Supplementary Figure 4). As titration of lung homogenates gave inconsistent results in our hands and because the effects of endogenous and exogenous IFN may influence titration outcomes, we chose taqman and immunohistochemistry to determine viral replication levels in the lung.
PBMC from healthy blood donors were isolated from heparinized venous blood using Lymphoprep (Axis-Shield). PBMC were resuspended at 2×106/ml in RPMI 1640 medium (Biowhittaker) supplemented with L-glutamine (2 mM), penicillin (100 U/ml), streptomycin (100 µg/ml), and 10% fetal calf serum. Freshly isolated PBMC were incubated with IL-1β (5 ng/ml; eBioscience), IFN-α 2a (1000 U/ml, 100U/ml, or 10U/ml; Roferon-A; Roche) or both for 24 hours in duplo or triplo per donor. Total RNA from stimulated PBMC was isolated using Trizol Reagent (Invitrogen) and the RNeasy mini kit (Qiagen). cDNA synthesis was performed with 100 ng total RNA and Superscript III RT (Invitrogen) with oligo(dT), according to the manufacturer's instructions. Semi-quantitative RT-PCR was performed for IL-8 [19] and IL-1β (Taqman gene expression assays; Applied Biosystems) as described previously using the 2−ΔΔCt method [63]. Average results (±s.e.m.) were expressed as fold change compared to untreated (mock) cells [63].
Data (RT-PCR and gross pathology scores for SARS-CoV-infected young adult versus aged and aged versus aged animals treated with IFN) were compared using Student's t-test with Welch's correction. Differences were considered significant at P<0.05. One-way ANOVA and Bonferroni's multiple comparison test were used for the comparison of data in groups based on severity of pathology (low, medium, high) and in vitro IFN inhibition experiments. Correlation coefficients were determined using Spearman's correlation test.
Pooled total RNA (2.4 µg) from one-three separate lung pieces of all animals (including previously infected animals), with substantial SARS-CoV replication (>105 fold change), was labeled using the One-Cycle Target Labeling Assay (Affymetrix) and hybridized onto Affymetrix GeneChip Rhesus Macaque Genome Arrays (Affymetrix), according to the manufacturer's recommendations. Image analysis was performed using Gene Chip Operating Software (Affymetrix). Microarray Suite version 5.0 software (Affymetrix) was used to generate .dat and .cel files for each experiment. All data were normalized using a variance stabilization algorithm (VSN) [64]. Transformed probe values were summarized into one value per probe set by the median polish method [65]. Primary data is available at http://www.virgo.nl in accordance with proposed MIAME standards.
Probe set (gene) wise comparisons between the experimental conditions (aged, young adult and IFN-treated animals versus young adult or aged PBS-infected animals and directly compared to each other) were performed by LIMMA (version 2.12.0) [66]. Correction for multiple testing was achieved by requiring a false discovery rate (FDR) of 0.05, calculated with the Benjamini-Hochberg procedure [67]. To understand the gene functions and the biological processes represented in the data and obtain differentially expressed molecular and cellular functions, Ingenuity Pathways Knowledge Base (http://www.ingenuity.com/) was used. Heat maps of pro-inflammatory pathways were produced using complete linkage and Euclidian distance in Spotfire DecisionSite for Functional Genomics version 9.1 (http://www.spotfire.com/) and Ingenuity Pathways Knowledge Base (http://www.ingenuity.com/), using log (base 2) transformed expression values with minimum and maximum values of the color range being −4 and 4, respectively. Differences between conditions in expression of specific pro-inflammatory pathways, e.g. direct comparison of defined gene sets (aged versus young adult and aged versus aged IFN-treated animals), were tested by Goeman's global test procedure [26]. Hierarchical clustering analysis of normalized log-2 based hybridization signals of individual young adult and aged macaques of a set of gene transcripts that were identified as being differentially regulated (fold change ≥2; FDR<0.05) in at least one of the comparisons of young adult versus young adult PBS or aged versus aged PBS animals were created using Spotfire DecisionSite for Functional Genomics version 9.1 (http://www.spotfire.com/) with complete linkage and Eucledian distance parameters.
|
10.1371/journal.pntd.0000974 | Prevalence and Risk Factors of Intestinal Parasitism in Rural and Remote West Malaysia | Intestinal parasitic infections (IPIs) have a worldwide distribution and have been identified as one of the most significant causes of illnesses and diseases among the disadvantaged population. In Malaysia, IPIs still persist in some rural areas, and this study was conducted to determine the current epidemiological status and to identify risk factors associated with IPIs among communities residing in rural and remote areas of West Malaysia.
A total of 716 participants from 8 villages were involved, comprising those from 1 to 83 years old, 550 (76.8%) participants aged ≤12 years and 166 (23.2%) aged ≥13 years, and 304 (42.5%) male and 412 (57.5%) female. The overall prevalence of IPIs was high (73.2%). Soil-transmitted helminth (STH) infections (73.2%) were significantly more common compared to protozoa infections (21.4%) (p<0.001). The prevalence of IPIs showed an age dependency relationship, with significantly higher rates observed among those aged ≤12 years. Multivariate analysis demonstrated that participants aged ≤12 years (OR = 2.23; 95% CI = 1.45–3.45), low household income (OR = 4.93; 95% CI = 3.15–7.73), using untreated water supply (OR = 2.08; 95% CI = 1.36–3.21), and indiscriminate defecation (OR = 5.01; 95% CI = 3.30–7.62) were identified as significant predictors of IPIs among these communities.
Essentially, these findings highlighted that IPIs are highly prevalent among the poor rural communities in West Malaysia. Poverty and low socioeconomic with poor environmental sanitation were indicated as important predictors of IPIs. Effective poverty reduction programmes, promotion of deworming, and mass campaigns to heighten awareness on health and hygiene are urgently needed to reduce IPIs.
| Intestinal parasitic infections (IPIs) are among the most prevalent human afflictions; these infections still have major impact on the socioeconomic and public health of the bottom billion of the world's poorest people. Although Malaysia has a thriving economy, IPIs are still very much prevalent and causing major health problems among the poor and in deprived communities especially in rural and remote areas. A comprehensive study is paramount to determine the current prevalent and factors closely linked to IPIs so that effective control measures can be instituted. In view of this, we conducted this study to provide detailed data of the existing status of IPIs among 716 participants living in rural and remote areas in Peninsular Malaysia. The establishment of such data is beneficial for the public health service to justify and facilitate the reassessment of control strategies and policies in terms of reducing intestinal parasitism. With effective control measures in place, these communities (especially children) will have a greater opportunity for a better future in terms of health and educational achievement and eventually will be at par socially and economically with urban communities in Malaysia.
| Globally, the neglected intestinal parasitic infections (IPIs) such as soil-transmitted helminth (STH) and protozoa infections have been recognized as one of the most significant causes of illnesses and diseases especially among disadvantaged communities. With an average prevalence rate of 50% in developed world, and almost 95% in developing countries, it is estimated that IPIs result in 450 million illnesses [1], [2], [3]. These infections are ubiquitous with high prevalence among the poor and socioeconomically deprived communities where overcrowding, poor environmental sanitation, low level of education and lack of access to safe water are prevalent [4], trapping them in a perennial cycle of poverty and destitution [5]. These parasitic diseases contribute to economic instability and social marginalization; and the poor people of under developed nations experience a vicious cycle of under nutrition and repeated infections leading to excess morbidity with children being the worst affected [2], [6].
Of these illnesses, infections by STH have been increasingly recognized as an important public health problem and most prevalent of IPIs [7]. STH infections caused by Ascaris lumbricoides, Trichuris trichiura and hookworm (Necator americanus and Ancylostoma duodenale) are most significant in the bottom billion of the world's poorest people (i.e., <US$1.25 per day) [8]. To date, approximately one third of the world's population is infected with at least one species of STH, with A. lumbricoides infecting 800 million people, T. trichiura 600 million, hookworm 600 million and resulting in up to 135,000 deaths annually [5].
With regards to intestinal protozoan infections, giardiasis caused by Giardia duodenalis, is the most predominant protozoa infection with an estimated prevalence rates ranging from 2.0 to 7.0% in developed countries and 20.0 to 30.0% in most developing countries, affecting approximately 200 million people worldwide [9]. Amoebiasis caused by Entamoeba histolytica is another important pathogenic protozoa affecting approximately 180 million people, of whom 40,000 to 110,000 succumbed to death annually [10]. The opportunistic protozoa, Cryptosporidium sp. has also emerged as an important cause of diarrhoeal illnesses worldwide particularly in young children and immunocompromised patients with a prevalence of 4% in developed countries and three to four times more frequent in developing countries [11].
Since the colonial era (i.e., 1930s) in Malaysia, many surveys and studies have been conducted on IPIs, in particular STH infections as they are deemed to be of great medical importance among Malaysian population. While vector-borne diseases such as malaria and filariasis have declined significantly over the years, IPIs which are closely associated with environmental and personal hygiene practices are still causing major health problems among the poor in rural and remote communities in Malaysia [12]. Within this context, we conducted this study to provide a comprehensive data of the current status of IPIs among rural communities residing in remote areas of West Malaysia. The establishment of such data will be beneficial for the public health service to justify and facilitate the reassessment of control strategies and policies.
A cross-sectional study was carried out from November 2007 to July 2009 among 8 villages from 5 different states in rural and remote areas of West Malaysia without being discriminatory towards age or gender. Villages include Betau (101.78°E longitude, 4.10°N latitude), Kuala Betis (101.79°E longitude, 4.90°N latitude), Sungai Bumbun (101.42°E longitude, 2.85°N latitude), Sungai Perah (100.92°E longitude, 4.48°N latitude), Gurney (101.44°E longitude, 3.43°N latitude), Pos Iskandar (102.65°E longitude, 3.06°N latitude), Bukit Serok (102.82°E longitude, 2.91°N latitude) and finally Sungai Layau (104.10°E longitude, 1.53°N latitude) (Figure 1). The villages were selected based on (i) village entry approval by the Ministry of Rural and Regional Development Malaysia and (ii) willingness to participate by the head and community members of the villages.
All villages were located at lowland altitude at the jungle fringes surrounded by rubber and palm oil estates. In general, although these communities have the provision of basic infrastructure (i.e., treated water and electricity) with concrete houses, these facilities are either not fully utilized or evenly distributed. Even if these provisions were given, most of them could not afford to pay their monthly utility bills due to extreme poverty leading to the termination of water and electricity supplies. Therefore, rivers located adjacent to the village remained their main source of water for domestic needs (i.e., drinking, cooking, bathing and washing clothes). Some households still lived in traditional structures of bamboo, wood, brick or a mixture of both with attap roof (i.e., thatched roof made from leaves of Nipah palm tree). In addition, although there are pour flush toilets, this facility is not consistently used as the villagers prefer to use nearby bushes and river for defecation. Rearing of animals such as pigs, chickens, ducks, dogs, cats and monkeys are common practices. Most of the residents were employed as unskilled laborers in construction sites, factories, vegetable farms, oil palm and rubber plantations.
Before the commencement of the study, an oral briefing explaining the objectives of the study was given to the participants and a voluntary written informed consent was taken from each participant. The participant was then asked by a trained field assistant to answer a pre-tested questionnaire developed to elicit information on the demographic data (i.e., age, gender and education attainment), socioeconomic (i.e., occupation, household income), behavioral (i.e., personal hygiene such as wearing shoes and food consumption), medical treatment (i.e., whether the participant has taken anthelminthic drugs and iron supplement), environmental sanitation and living condition characteristics (i.e., type of water supply, latrine system, garbage disposal system and presence of domestic animals) which will be used to assess the potential risk factors for IPIs. The questionnaire was designed in Bahasa Malaysia, which is the national language for Malaysia and well understood by the participants. For children and very old participants, the questionnaire was completed by interviewing their parents and guardians or the relevant adult (normally head of the family) who signed the informed consent.
After consent was obtained and questionnaire answered, a wide mouth screw capped containers pre-labeled with their names and coded were distributed to each participant. Their ability to recognize their names was counter-checked. The participant was instructed to scoop a thumb size fecal sample using a provided scoop into the container, making sure that the sample was not contaminated with urine. Parents and guardians were instructed to monitor their children during the sample collection to ensure that they placed their fecal samples into the right container. Participants who turned up with their fecal samples the following day were honored with a small token of appreciation. The collected fecal samples were processed and examined for the presence of parasites by the formalin ether concentration technique [13]. Briefly, 1 to 2g of fecal sample was mixed with 7 ml of formalin and 3 ml ethyl acetate, centrifuged, stained with 0.85% iodine and examined under light microscope. For Cryptosporidium sp., all fecal samples were examined using modified Ziehl Neelsen stain which includes usage of strong carbol fuchsin, 1% acid alcohol and 0.4% malachite green [14]. In addition, Kato-Katz technique was employed to determine the intensity of STH infections, as estimated by egg counts per gram of feces (EPG) as described by Martin and Beaver for A. lumbricoides, T. trichiura and hookworm [15]. The total number of eggs observed was multiplied with an appropriate exchange number (i.e., number of eggs X 22.2) to give the number of eggs per gram of feces. The worm burden was categorized as light, moderate or heavy intensity based on the threshold proposed by World Health Organization (WHO) Expert Committee in 1987 [16]. Dysenteric or inadequate samples, which were unsuitable for egg counts were used only for examination of intestinal parasites ova by formalin ether concentration technique.
Statistical analysis was carried out using the SPSS software (Statistical Package for the Social Sciences) programme for windows version 13 (SPSS, Chicago, IL, USA). Initial data entry was cross-checked by two independent individuals in order to be sure that data were entered correctly. Before each analysis, data were again checked for consistency. Prevalence of IPIs was determined on the basis of combined results from the different diagnostic methods. For descriptive data, rate (percentage) was used to describe the characteristics of the studied population, including the prevalence of IPIs according to villages, age and gender. The intensity of STH infections (worm burden) was quantitatively estimated as eggs per gram of feces (EPG) and was divided into three main categories: light, moderate or heavy infections and expressed as means. A Pearson's Chi-square (X2) on proportion was used to test the associations between each variable. In univariate statistical model, the dependent variable was prevalence of IPIs, while the independent variables were sociodemographic, behavior, medical treatment, environmental sanitation and living condition characteristics. All variables that were significantly associated with prevalence of IPIs in univariate model were included in a logistic multivariate analysis using forward elimination model to identify the predictors of IPIs. The level of statistical significance was set as p<0.05 and for each statistically significant factor, an odd ratio (OR) and 95% confidence interval (CI) was computed by the univariate and multivariate logistic regression analysis.
The study protocol (Reference Number: 638.36) was approved by the Ethics Committee of the University Malaya Medical Centre (UMMC), Malaysia before the commencement of the study. The participants were informed that the procedure used did not pose any potential risk and their identities and personal particulars will be kept strictly confidential. During the meetings, parents and their children were informed that their participation was voluntarily and they could withdraw from the study at any time without giving any reason. Consent of those who agreed to participate were taken either in written form (signed) or verbally followed by their thumb prints (for those who are illiterate) and from parents or guardians (on behalf of their children).
A total of 716 villagers participated in this study. With regards to age groups, there were a total of 550 (76.8%) participants aged ≤12 years and 166 (23.2%) aged ≥13 years ranging from 1 to 83 years with a median age of 11 years and a proportion of 1.1%, 2.4%, 73.3%, 2.0% and 21.2% for the age groups 1 to 4, 5 to 6, 7 to 12, 13 to 17 and above 18 years, respectively. These participants consisted of 304 (42.5%) male and 412 (57.5%) female.
The overall prevalence of IPIs among 716 participants was 73.2% with STH infections (73.2%; 524) being significantly more common compared to protozoa infections (21.4%; 153) (p<0.001). In addition, there were also 2 (0.3%) cases of Fasciolopsis/Fasciola sp. infection detected (legend indication “other infection”) (Table 1). Prevalence of IPIs were very high in most of the surveyed villages, ranging from 66.7% to 97.8%. Interestingly, infections were very low in Sungai Layau village (4.5%) (Table 1). There was no significant difference of the IPIs between male and female, although female (73.3%) had slightly higher overall prevalence rate compared to male (73.0%) (Table 2). The prevalence of IPIs showed an age dependency relationship, with significantly higher prevalence seen among participants aged ≤12 years compared to those aged ≥13 years (76.7% versus 61.4%, p<0.001).With regards to specific age groups, prevalence was highest in the 5 to 6 aged group (94.1%) and lowest (59.2%) among those aged 18 years and above (Table 2).
The overall prevalence of STH infections was 73.2% with T. trichiura (66.8%) being the most predominant, followed by A. lumbricoides (38.5%) while only 12.8% had hookworm infections. In general, participants from Betau village had the highest prevalence of STH infections (97.8%) whilst those from Sungai Layau village (4.5%) had the least. Based on the total sample size (n = 716), double infections (35.6%) were most common, followed by single infections (31.6%) and triple infections (6.0%). T. trichiura (28.4%) was the most dominant cause of single infections. The combination of T. trichiura and A. lumbricoides were the most predominant in the double infections, accounting for 30.6% of the infection rates. With regards to the intensity of infections, all three species of STH showed light to heavy infections. In general, both T. trichiura and A. lumbricoides infections had similar pattern of worm burden with moderate infection being the most common followed by light and heavy infections. In contrast, most of hookworm infections were light (Table 3).
With regards to the protozoa infections, the overall prevalence was 21.4%. The highest prevalence rate was due to G. duodenalis (10.4%), followed by E. histolytica/dispar (10.2%) and Cryptosporidium sp. (2.1%). For protozoa infections, participants from Kuala Betis village (39.0%) had the highest prevalence, whilst those from Sungai Layau village had the least (4.5%) (Table 1). Based on the total sample size (n = 716), single infections (21.4%) were most predominant, followed by double infections (1%). There were no triple infections recorded. As for mix infections with both STH and protozoa, the most common combinations were T. trichiura, hookworm and G. duodenalis (17.4%) followed by T. trichiura either with G. duodenalis or E. histolytica/dispar (12.4%) and finally a combination of four species which included T. trichiura, A. lumbricoides, G. duodenalis and E. histolytica/dispar (1.4%) infections.
The risk factors associated with IPIs in relation to sociodemographic and lifestyles among rural communities were examined by univariate analysis. There were eight risk factors identified which included those less than 12 years old (OR = 2.10; 95% CI = 1.43-2.98; p<0.001), low household income (OR = 7.60; 95% CI = 5.30–11.13; p<0.001), using untreated water supply for daily chores (OR = 2.84; 95% CI = 2.08–3.86; p<0.001), the lack of proper latrine system (OR = 2.19; 95% CI = 1.54–3.10; p<0.001), non-existence of pour flush toilet (OR = 3.29; 95% CI = 2.62–4.12; p<0.001), indiscriminate defecation (OR = 3.45; 95% CI = 2.76–4.32; p<0.001) and indiscriminate garbage disposal (OR = 2.06; 95% CI = 1.45–2.94; p<0.001), and finally not taking any anthelminthic drugs in the last 12 months (OR = 1.29; 95% CI = 1.01–1.65; p = 0.038). Although being a female (73.3%; p = 0.935), jobless (74.1%; p = 0.088), have had no close contact with animals (80%; p = 0.193) and not taking any iron supplement (73.5%; p = 0.801) were factors which had higher infection rates, nonetheless these variables were not statistically significant (Table 4). Multivariate analysis using forward logistic regression model further confirmed that participant less than 12 years old were 2.2 times (95% CI = 1.45–3.45; p<0.001), low household income had 4.9 times (95% CI = 3.15–7.73; p<0.001), using untreated water supply had 2.1 times (95% CI = 1.36–3.21; p<0.001) and indiscriminate defecation were 5 times (95% CI = 3.30–7.62; p<0.001) more likely to suffer from an IPIs, respectively.
As shown by the results of the present study, intestinal parasitic infections (IPIs) are still a major public health problem (i.e., overall prevalence of 73.2%) among the impoverished and underprivileged communities in rural and remote West Malaysia. However, this study also observed some very encouraging trends. In Sungai Layau village where each family was provided with a concrete house and basic amenities like treated water supply, prevalence of IPIs was shown to be significantly lower (4.5%). This proved that proper provision of basic infrastructure and education are effective tools to reduce the prevalence of these infections. On the contrary, in Betau, Kuala Betis, Sungai Bumbun, Sungai Perah, Gurney, Pos Iskandar and Bukit Serok villages where some villagers still lived in traditional-built houses and using water from wells or rivers, prevalence of IPIs were very high. This was evident in the present findings whereby Betau village which was less provided or developed had the highest rate of infection (97.8%).
Results also showed that STH infections (73.2%) were more common compared to protozoa infections (20.1%). T. trichiura infection is the most common (66.8%) followed by A. lumbricoides (38.5%) and hookworm (12.8%). These findings were in agreement with other previous local studies where T. trichiura infection was found to be the most prevalent (range: 26.0% to 98.2%), followed by A. lumbricoides infections (range: 19.0% to 67.8%) and lastly hookworm infections (range: 3.0% to 37.0%) [17], [18], [19], [20], [21]. However, global data has indicated that A. lumbricoides infections were the most prevalent among the three STH infections. The higher rate of T. trichiura infection has been reported to be due to the ineffective dosage and choice of anthelminthic used. Currently, the recommended treatment regime for STH infection is broad spectrum anthelminthics such as albendazole and mebendazole. Important therapeutic differences do exist between these drugs which affect their uses in clinical practice [22]. Both drugs are effective against ascariasis in single dose, whereas single doses of either albendazole or mebendazole have been found to be ineffective in many cases of trichuriasis [22]. Furthermore, potential resistance of T. trichiura to anthelminthic drugs has been highlighted in two intervention studies in Malaysian communities [23], [24].
It has been noted that unscheduled deworming without proper monitoring system was common among the children of these communities. Since the mass deworming program of schoolchildren has been discontinued in 1983 [25], some of the children received anthelminthic drug during visits to health clinic or from their school medical health team. Some parents have also bought anthelminthic drug for their children without following the recommended treatment intervals (i.e., periodic deworming) and this could have resulted in the inefficacy of the drug and subsequently lead to drug resistance [24]. Another important problem encountered in treatment is the high rate of re-infection especially in highly endemic areas. Local studies among rural communities have found that re-infection can occur as early as 2 months post treatment, by 4 months almost half of the treated population had been re-infected [24] and by 6 months the intensity of infections had returned to pre-treatment levels [26]. Similar findings have also been reported in other parts of the world indicating that by 6 months, the intensity of infections of T. trichiura and A. lumbricoides were similar to pre-treatment levels [27]. WHO has recommended that mass deworming programme should be carried out in communities when the cumulative STH prevalence is more than 50% or the cumulative percentage of moderately or heavily infected individuals is more than 10% [28]. As the present findings have indicated that the overall prevalence was 73.2%, it is strongly recommended that mass deworming programmes are restored and a systematic evaluation of treatment regime must be put in place to reduce the rates of re-infection.
As for protozoa infections, the overall prevalence was 21.4%. However, in contrast with the latest local study in rural area, Noor Azian and colleagues reported very high rates of protozoa infection (72.3%) [29]. The present study found that G. duodenalis (10.4%) was the most predominant protozoa, followed by E. histolytica/dispar (10.2%) and lastly Cryptosporidium sp. (2.1%). In Malaysia, the prevalence of G. duodenalis infections varied from 2.0% to 29.2% while the prevalence of E. histolytica/dispar infections was reported to range from 1.0% to 18.5% among rural community [23], [29]. Although amoebic liver abscess (65% of 34) has been documented in patients admitted to an urban hospital in Malaysia [30], information from rural communities is not available as this infection can only be confirmed in a hospital setup. Two previous studies have indicated that Cryptosporidium sp. infections in rural areas ranged from 5.5% to 20.1% [31], [32].
The present study also reported 2 cases (0.3%) of Fasciolopsis/Fasciola sp. infection in Gurney village. This infection is most probably spurious due to consumption of infected animal liver. To date, there has not been any published data on intestinal fluke infection in West Malaysia, however, a case report of fasciolopsiasis by Fasciolopsis buski has been reported among rural community in East Malaysia [33]. In addition, two reported cases of food-borne diphyllobothriasis after consuming sushi and sashimi have also been reported in urban West Malaysia [34], [35].
Previous local studies indicated that there was a web of risk factors associated with the high prevalence of IPIs which included age, low family income, inadequate sanitation, presence and close contact with livestock or pets, untreated water supply, low level of parental education, poor geographical and personal hygiene [17], [22], [23]. Using multivariate analysis, the present study confirmed that children, low household income, untreated water supply, indiscriminate defecation were significant risk predictors of IPIs. This finding is further confirmed with the significantly lower prevalence in Sungai Layau village where household incomes are much higher and basic amenities provided by the government are fully utilized by the villagers.
Intestinal parasitic infections are highly prevalent and are major public health concerns among the poor and socioeconomically deprived rural and remote communities in West Malaysia. Given that IPIs are intimately associated with poverty, poor environmental sanitation and lack of clean water supply, it is crucial that these factors are addressed effectively. Improvement of socioeconomic status, sanitation, health education to promote awareness about health and hygiene together with periodic mass deworming are better strategies to control these infections. With effective control measures in place, these communities (especially children) will have a greater opportunity for a better future in terms of health and educational achievement.
|
10.1371/journal.pcbi.1004618 | SAAS-CNV: A Joint Segmentation Approach on Aggregated and Allele Specific Signals for the Identification of Somatic Copy Number Alterations with Next-Generation Sequencing Data | Cancer genomes exhibit profound somatic copy number alterations (SCNAs). Studying tumor SCNAs using massively parallel sequencing provides unprecedented resolution and meanwhile gives rise to new challenges in data analysis, complicated by tumor aneuploidy and heterogeneity as well as normal cell contamination. While the majority of read depth based methods utilize total sequencing depth alone for SCNA inference, the allele specific signals are undervalued. We proposed a joint segmentation and inference approach using both signals to meet some of the challenges. Our method consists of four major steps: 1) extracting read depth supporting reference and alternative alleles at each SNP/Indel locus and comparing the total read depth and alternative allele proportion between tumor and matched normal sample; 2) performing joint segmentation on the two signal dimensions; 3) correcting the copy number baseline from which the SCNA state is determined; 4) calling SCNA state for each segment based on both signal dimensions. The method is applicable to whole exome/genome sequencing (WES/WGS) as well as SNP array data in a tumor-control study. We applied the method to a dataset containing no SCNAs to test the specificity, created by pairing sequencing replicates of a single HapMap sample as normal/tumor pairs, as well as a large-scale WGS dataset consisting of 88 liver tumors along with adjacent normal tissues. Compared with representative methods, our method demonstrated improved accuracy, scalability to large cancer studies, capability in handling both sequencing and SNP array data, and the potential to improve the estimation of tumor ploidy and purity.
| Somatic copy number alterations (SCNAs) are essential in oncogensis and progression of a variety of cancers. Accurate identification and quatification of SCNAs are fundamental in the effort of cataloging different variants in cancer genome. This task has its own challenges due to complex nature of tumor SCNA profile and is further complicated by the heterogeneity of the cells collected from a tumor tissue and the contamination from adjacent normal cells, making it difficult for the methods well tailored for the detection of germline copy number variation (CNV) to fit in tumor SCNA detection. Next generation sequencing provides an opportunity to comprehensively characterize SCNA at unprecedent resolution. While total read depth information is commonly used in SCNA detection methods, the allele-specific read depth is less often considered, leading to sub-optimal solution. By incorparating both pieces of information, we developed a segmentation-based pipeline to address aforementioned issues in SCNA detection. This tool is applicable on both deep sequencing data as well as SNP array data and enables accurate and efficient characterization of genome-wide SCNA profile to facilitate large-scale cancer studies.
| Profound somatic copy number alternations (SCNAs) are present in many types of tumors [1–3], where they affect a larger fraction of the genome than other types of somatic variations [3,4]. The roles of SCNAs in promoting oncogenesis and tumor progression are under intensive study. In contrast to germline copy number variations (CNVs), which are sparsely distributed along the genome and of small to moderate size, tumor SCNAs are large in size and have a much wider range of magnitudes in copy number. Accurate detection and characterization of genome-wide SCNA profile are further complicated by aneuploidy and heterogeneity of tumor cells and contamination of normal cells [5].
Array comparative genomic hybridization (array-CGH) [6] and single nucleotide polymorphism (SNP) array [7] were widely used in surveying genome-wide SCNAs in the past decade. More recently, next-generation sequencing (NGS) technology provides unprecedented resolution to comprehensively characterize SCNAs at rapidly decreasing cost [8,9]. A long list of tools has been developed and successfully applied in analyzing CNV data harvested from NGS; see [2,10,11] for reviews. The available methods mainly fall into four categories: 1) read depth (RD), 2) pair-end mapping (PEM), 3) split read (SR) and 4) Assembling (AS). While they take advantage of complementary information, featuring different perspectives of CNVs, each of them encounters important limitations, due to the complications from the high dynamics of cancer genome. With whole genome sequencing (WGS) data, the latter three methods are good at detecting CNVs of small to moderate sizes, locating CNV break points at higher resolution, and discovering copy neutral rearrangements, but they are less capable in characterizing large-size and wide-range copy number changes at the genome-wide scale. Moreover, their applicability in whole exome sequencing (WES) data is limited by the fact that short reads from WES are concentrated in interspersed genomic regions. Therefore, the RD-based methods are used more widely in the study of tumor CNVs with both WGS and WES data [10]. Basically, the majority of RD-based methods, such as CNV-seq [12], SegSeq [9], ExomeCNV [13] and PatternCNV [14], follows a “bottom-up” procedure: short-reads mapping, normalization of read depth, copy number estimation in a local region (usually in a window of certain size or in an exonic region) and segmentation to merge regions with the same copy number status [10]. This strategy is sensitive in the detection of germline CNVs, but for tumor CNVs (i.e. SCNAs), it is difficult for the local inference to correctly decide the baseline of ploidy and accurately discern weak signals of copy number change in the presence of aneuploidy and normal cell contamination, so that the genome-wide inference drawn from the later segmentation step is inclined to accumulate false positive findings from the earlier local inference step. Further, these methods only consider the aggregated depth of sequencing reads carrying paternal and maternal alleles, aimed at estimating the total copy number, but largely ignore allele specific read depth, while the latter contains critical information of copy number change, copy-neutral loss of heterozygosity (CN-LOH), and, importantly, genome ploidy.
In analogy to Illumina SNP array [7], the allele specific signal from NGS can be quantified at heterozygote sites (e.g. SNP loci) and converted to the so-called B allele frequency (BAF), which is defined as the proportion of the reads carrying “B” allele (i.e. non-reference allele) among all the mappable reads at that site. At an SCNA-free locus, the expectation of BAF is close to 1/2, indicating equal amount of paternal and maternal alleles. With an SCNA event, the expectation of BAF might deviate from 1/2, reflecting differential copy number changes in the two alleles.
The two data dimensions, total read depth and allele-specific signal (i.e., BAF), carry complementary and consensual information of SCNA events. Incorporating both quantities could theoretically improve power and accuracy in identifying SCNA segment boundaries and characterizing alteration types (i.e., gains, losses, and CN-LOHs). Control-FREEC [15] is a representative method that incorporates both types of information. However, it uses them separately rather than jointly in the segmentation and SCNA calling, nor does it make good use of data from match normal sample to generate reliable allele-specific signal. Herein, we propose a tumor SCNA analysis method, SAAS-CNV, based on a joint segmentation algorithm [16], to accommodate both total read depth and BAF. Using both data dimensions simultaneously, the method first takes a “top-down” strategy to partition the genome into segments with different alterations by joint segmentation, and subsequently, determines their alteration types. The method was designed for paired normal-tumor settings, in which the spatial variability from non-uniform distribution of short reads along the genome can be alleviated [13]. It is able to accommodate WGS and WES data as well as SNP array data.
In this paper, we first characterized the type-I-error properties of SAAS-CNV using a carefully designed “null” dataset containing no SCNAs (H0), created by pairing sequencing replicates of a single HapMap sample as normal-tumor pairs [17] (Materials and Methods, Dataset I). Further, we demonstrated its ability to analyze a real-world WGS study of 88 hepatocellular carcinoma (HCC) samples (Ha), some of which exhibit highly complicated SCNA profiles [18,19] (Materials and Methods, Dataset II). In both settings (Ho and Ha), we compared the performance of SAAS-CNV with other existing methods. We also explored its potential extension in the estimation of ploidy and purity of tumor cells.
We used GATK variant analysis pipeline [21–23] to process raw fastq files in this study [17] (more details in S1B Text). As a standard output of the pipeline, single nucleotide variants (SNVs) and small insertion and deletions (Indels), called from mapped high-quality reads, were saved in the variant call format (VCF) files [24]. At each variant locus, we extracted the genotype and the depth of reads carrying reference allele and alternative allele, all informative for SCNA inference. Fig 1 overviews the workflow of our algorithm in the paired normal-tumor setting.
For both datasets, a standard implementation of NGS analysis pipeline following the GATK best practices for variant detection [17,21,22] was applied to the raw FASTQ files to generate recalibrated and deduplicated high-quality BAM files [27], as well as VCF files containing detected SNVs and Indels. Information retrieved from biallelic heterozygous sites were processed by our SCNA analysis pipeline (Fig 1). Illumina SNP array data was processed to generate log2ratio and log2mBAF signals in a similar way for downstream analysis. Detailed data analysis steps are described in S1B Text.
First, we present a visualization of the processed signals and the results from SAAS-CNV. Fig 2 demonstrates a typical example taken from the analysis of Dataset II. The 2-dimensional profiles (log2ratio and log2mBAF) from SNP array (Fig 2A and 2B) and WGS (Fig 2C and 2D) were processed for genome-wide SCNA detection. The signals in the two dimensions are altered by copy number gains and losses, showing consistent patterns of changes, for both platforms. The alteration status assigned to each segment is highly consistent between the two platforms (Fig 2A and 2C). We projected the medians of log2ratio and log2mBAF of each segment onto a 2-D space for SNP array and WGS respectively, confirming the consistency of the inference drawn from the two platforms (Fig 2B and 2D). Segments with the same alteration status are clustered together. Segments with less difference in log2ratio dimension may be distinguished in log2mBAF dimension, suggesting the valuable information added from BAF for SCNA inference. It is noticed that the original baselines in both dimensions deviate substantially from zero (Fig 2B and 2D), and simply using zero as baseline for SCNA inference would result in many false positive calls, underscoring the necessity of baseline adjustment.
We processed the HKU HCC WGS data, consisting of 88 pairs of tumor and adjacent normal tissues, with the standard implementation of GATK pipeline [17]. The average read depths of 86 pairs are moderate, ranging from 24.8x to 71.5x, with two pairs sequenced at higher depth (>120x). The average effective read depths measured at heterozygous sites range from 18.2x to 99.8x, composing 68%~90% of available read depth (S1 Fig). We applied SAAS-CNV, CNVnorm and Control-FREEC on the 88 pairs of normal-HCC samples. Further, we synthesized WES data by retrieving reads located in exonic regions from WGS data. The synthesized WES data was used to test the performance of SAAS-CNV and ExomeCNV. We also applied SAAS-CNV and GAP on the SNP array data from the same samples. As stated above, the GAP results on SNP array data were used as “truth” in benchmarking other SCNA methods.
Some basic metrics from these analyses are summarized in Fig 6. WGS provides millions of data points (loci) per sample while SNP array and WES provides hundreds of thousands of data points (loci) (Fig 6A). Since SAAS-CNV takes heterozygous loci as input, it utilizes less number of loci than competing methods. For example, only tens of thousands of loci were used by SAAS-SNV in WES data. On SNP array data, SAAS-CNV and GAP produced comparable number of segments per sample (Fig 6B), as well as comparable segment size in terms of locus number per segment (Fig 6C) and physical length (Fig 6D). On WES and WGS data, the competing methods tended to chop the genome into smaller segments (Fig 6B) than SAAS-CNV (Fig 6C and 6D).
We highlighted the potential usage of BAF information in improving the estimation of absolute copy number and tumor purity on the HKU HCC sample PT116 in Dataset II (Materials and Methods). This sample exhibits complicated SCNA profile (Fig 9). We applied SAAS-CNV and CNAnorm together to analyze this sample. In CNAnorm analysis, the ratio of tumor versus normal read depth was calculated per 1kb window along the genome (gray dots in Fig 10A). A smoothing approach was then employed by CNAnorm [5] to reduce the random error variability (fitted black curve in Fig 10A). The distribution of smoothed ratio signal clearly showed seven major peaks corresponding to different copy numbers (Fig 10B). CNAnorm attempted to fit a Gaussian mixture model along with Akaike’s information criterion (AIC) onto the distribution in order to identify the number of components (peaks) and their locations. However, CNAnorm was only able to identify five out of the seven peaks (Fig 10C). It then searched different configurations, where plausible copy numbers and the five identified peak centers were aligned in different ways, and chose the most likely configuration that resulted in the best correspondance. Here, the correspondance was measured by the goodness-of-fit (R2 value) of the linear regression model of peak centers on estimated copy numbers (Fig 10C). In this case, CNAnorm assigned the most common component (the highest peak in Fig 10A) to copy number 2 (i.e. tumor-normal ratio 1). Finally, CNAnorm estimated tumor purity (ρ) based on the relationship:
μiμCN=2−1=(CNi2−1)⋅ρ
(5)
where μi is the identified peak center associated with copy number CNi and μCN = 2 corresponds to peak center associated with copy number 2. With the estimated tumor purity, the absolute copy number of each segment was obtained (S13A Fig).
By a visual check of Fig 10B and 10C, it is obvious that CNAnorm failed to assign copy number correctly. Without accounting for the allelic imbalance pattern information (Fig 9), CNAnorm was not able to correctly infer absolute copy number in such a complicated cancer genome. For example, Chromosomes 3, 9, 11, 12, 14 and 15 were estimated to be diploid (S13A Fig), contradictory to the strong pattern of allelic imbalance manifested in log2mBAF dimension (Fig 9); Chromosomes 2, 6 and 18 were estimated to be approximately triploid, also contradictory to the log2mBAF pattern showing allelic balance (Fig 9). These observations motivated us to manually correct the inference from CNAnorm. We adjusted the correspondence between peak centers and copy numbers and as a result of the correction, the R2 was substantially improved (Fig 10D). The tumor purity was re-estimated with the corrected correspondence between μi and CNi in Eq (5), and revised from 64.92% to 80.88%. The re-estimated copy number profile was improved in terms of better fit to integer copy numbers (horizontal black segments overlap more with the horizontal gray lines in S13B Fig). The correction also lead to strikingly better fit of the theoretical mBAF to the observed mBAF values (Fig 11). The theoretical mBAF was calculated based on the estimated copy number and tumor purity (detailed in S1K Text). Interestingly, this tumor genome likely underwent a whole-genome doubling event [31] and was estimated to be tetraploid, based on which the baseline of “normal” copy number was anchored. In summary, this case study illustrated the joint inference from BAF along with total copy number could result in more reliable estimation of tumor ploidy and purity.
We have developed a joint segmentation and inference approach for SCNA analysis using both total and allele-specific sequencing depth and investigated its performance with a “null” dataset and a real-world large-scale dataset. Compared with existing methods: ExomeCNV, PatternCNV, CNAnorm and Control-FREEC, our method demonstrated improved accuracy, scalability to large cancer sequencing studies, and the potential in improving the estimation of tumor ploidy and purity. Our approach exhibits flexibility and applicability in a wide range of platforms, including both deep sequencing and SNP array. These good properties can facilitate integrative cancer genomics study using multiple platforms. An R package called saasCNV, which implements our proposed appoach, is avaiable at https://zhangz05.u.hpc.mssm.edu/saasCNV/index.htm.
In contrast with germline CNV detection, characterization of SCNA in cancer genome gives rise to particular challenges, including ambiguous baseline due to aneuploidy and diluted signal pattern due to heterogeneity and normal cell contamination. For this regard, allelic specific information adds valuable input. Our method takes advantage of this information in both deep sequencing and SNP array data and achieves better capability in the correct identification of copy number baseline. We have also demonstrated that incorporating BAF could substantially improve the inference of tumor ploidy and purity. An important future work is the development of an integrated statistical model based on segment-level total sequencing depth and BAF to infer absolute copy number and tumor purity simultaneously.
SAAS-CNV features good efficiency and scalability by adopting the “top-down” strategy, which reduces the number of statistical inferences from millions to hundreds, and by taking advantage of condensed information from VCF file, which is about 1% of the size of BAM file. On Dataset II, we showed WES of moderate sequencing depth can still provide comparable specificity and sensitivity as WGS in large SCNA detection. At the same time, we acknowledge that quantifying SCNA at heterozygous loci has limitations in that not all information from mapped reads is fully utilized. While this limitation has less influence in WGS, where SNPs and indels are densely distributed on the genome, it affects the resolution in a greater degree on WES data. A possible improvement is to leverage BAMs to boost the SNR of log2ratio signal and the resolution by averaging over read depths within the windows surrounding or nearby heterozygous sites. However, it is quite time-comsumig to manipulate large BAM files. An alternative option is to utilize genomic VCF (GVCF) file, which is in the similar format as VCF, but record both variant sites and non-variant blocks. Inspired by the normalization procedure of Illumina SNP array data [7], the data processing step in our method can be further improved by incorparating multiple samples simultaneously, especially when manipulating GVCF makes it computationally feasible.
The matched tumor-normal design provides several desired features for the identification of SCNA: 1) it is biologically sensible to take matched normal genome as reference to define somatic alterations in tumor genome; 2) it is helpful to reduce the bias induced from the spatially non-uniform distribution of short reads across the genome, due to variability in GC content, exon capture efficiency, mappability of complex regions and so forth (also see S1L Text); 3) it is able to improve the normality of signals (S2 Fig), which is desirable for joint segmentation step; 4) it can help alleviate allelic signal bias commonly observed in SNP array data (S10 Fig) [30].
Lastly, our method can be used in accompany with paired-end mapping or split read methods, for example CREST [28], to refine the resolution of break points up to base pair level and verify other types of genomic rearrangements, such as intra-chromosomal and inter-chromosomal translocations, which are commonly associated with SCNAs [32].
|
10.1371/journal.pntd.0002280 | Housefly Population Density Correlates with Shigellosis among Children in Mirzapur, Bangladesh: A Time Series Analysis | Shigella infections are a public health problem in developing and transitional countries because of high transmissibility, severity of clinical disease, widespread antibiotic resistance and lack of a licensed vaccine. Whereas Shigellae are known to be transmitted primarily by direct fecal-oral contact and less commonly by contaminated food and water, the role of the housefly Musca domestica as a mechanical vector of transmission is less appreciated. We sought to assess the contribution of houseflies to Shigella-associated moderate-to-severe diarrhea (MSD) among children less than five years old in Mirzapur, Bangladesh, a site where shigellosis is hyperendemic, and to model the potential impact of a housefly control intervention.
Stool samples from 843 children presenting to Kumudini Hospital during 2009–2010 with new episodes of MSD (diarrhea accompanied by dehydration, dysentery or hospitalization) were analyzed. Housefly density was measured twice weekly in six randomly selected sentinel households. Poisson time series regression was performed and autoregression-adjusted attributable fractions (AFs) were calculated using the Bruzzi method, with standard errors via jackknife procedure.
Dramatic springtime peaks in housefly density in 2009 and 2010 were followed one to two months later by peaks of Shigella-associated MSD among toddlers and pre-school children. Poisson time series regression showed that housefly density was associated with Shigella cases at three lags (six weeks) (Incidence Rate Ratio = 1.39 [95% CI: 1.23 to 1.58] for each log increase in fly count), an association that was not confounded by ambient air temperature. Autocorrelation-adjusted AF calculations showed that a housefly control intervention could have prevented approximately 37% of the Shigella cases over the study period.
Houseflies may play an important role in the seasonal transmission of Shigella in some developing country ecologies. Interventions to control houseflies should be evaluated as possible additions to the public health arsenal to diminish Shigella (and perhaps other causes of) diarrheal infection.
| Whereas previous researchers have noted that seasonal peaks in the numbers of houseflies and patients suffering from Shigella diarrheal infection seemed to coincide, this is the first research to quantify the association using time-series statistical methods. The results show that houseflies could account for approximately 37% of all cases of shigellosis in an area in rural Bangladesh. This research adds to the existing published experimental and observational evidence from other parts of the world implicating houseflies as mechanical transmission vectors for Shigella. The results can be used to advocate for cluster-randomized intervention trials that can demonstrate how much control of housefly density can diminish Shigella disease incidence. This question should be answered because there are currently no licensed Shigella vaccines, and rising antibiotic resistance is limiting treatment options. Control of houseflies using methods such as baited fly traps could be an affordable, effective intervention to add to the public health arsenal for routine use and in the context of disaster response.
| Shigella, a human host-restricted pathogen that invades and damages gut mucosa, persists as a public health problem in developing and transitional countries because of its high transmissibility via direct fecal-oral contact, the severe clinical disease it causes, widespread drug resistance that limits the utility of previously effective antibiotics and the absence of licensed vaccines. The minute inoculum (ten Shigella organisms) capable of causing full blown dysentery enables direct person-to-person transmission [1], [2], even where environmental sanitation is otherwise adequate and safe water is available [3], [4]. Less commonly, Shigella is transmitted by contaminated food [5] or water vehicles [5]. Least appreciated is the observational and robust experimental evidence that demonstrates that the housefly, Musca domestica, can serve as a mechanical vector that also fosters transmission of Shigella [6], [7].
Houseflies breed in human feces [8], Shigella can be cultured from flies trapped in endemic areas [6], [9], [10], and observational studies have shown increased incidence of dysentery or diarrhea during periods of high fly density [11]–[13]. Most importantly, controlled intervention studies have shown that reducing housefly density is accompanied by reduced incidence of diarrhea [6], [11], [14], [15], dysentery [6], culture-confirmed shigellosis [6], [14], [15] and serological evidence of Shigella infection [6]. To gather evidence of the association of housefly population density with Shigella-associated illness among children <five years of age in a developing country setting, we systematically enumerated houseflies in sentinel households in Mirzapur, Bangladesh, a site characterized by an unusually high prevalence of Shigella among children with acute moderate-to-severe diarrhea (MSD), and few apparent risk factors for transmission of diarrheal disease pathogens, when compared with the other six sites in the Global Enteric Multicenter Study (GEMS) [16]. To our knowledge, this is the first study to attempt to correlate the density of houseflies in environs of typical households with the occurrence of laboratory-confirmed Shigella-associated illness in young children in the community.
A cross-sectional study examining the association between site-wide housefly population density and Shigella-associated MSD among children <five years of age was carried out from December 3rd, 2008 to December 1st, 2010 in Mirzapur, Bangladesh.
The study was nested within the three-year GEMS, which included a matched case-control study of the burden and etiology of MSD. Informed consent was sought from parents or caretakers of the research subjects, all of whom were children <5 years of age. Study purpose, risks and benefits were first explained to caretakers of children invited to participate in GEMS before the consent form was read aloud, while the caretaker, if literate, read his or her own copy of the consent form. Ample time was allowed for questions and discussion. If the parent/caretaker consented, he or she was then asked to provide written consent by signing the consent form. If the caretaker was illiterate, a person not employed by the study was asked to witness the informed consent process; upon consent, both the caretaker and witness were asked to sign their names to the consent form (illiterate caretakers unable to provide a written signature were asked to apply an ink fingerprint impression instead). The presence of a witness signature indicated that consent was oral rather than written. Permission was obtained from the head of household for placement within the household compound of devices (Scudder grills) to quantify fly density. The consent forms and protocol, including the provision for oral consent, were approved by the ICDDR,B Ethical Review Committee and the University of Maryland Human Research Protections Office.
Mirzapur is a mainly Muslim rural community 70 km northwest of Dhaka with a population of approximately 254,751 (∼24,077 children <five years of age) under a Demographic Surveillance System (DSS). Most men are engaged in agriculture or daily wage labor and women typically work in the home. Many households have one or more family members working long-term abroad (mainly in Persian Gulf States and Saudi Arabia) who send home financial supplements that substantially improve the household's economic situation. “Winter” generally lasts from December to mid-February, while the monsoon rains and flooding occur during the hot months of June to October. The months of March to May are warm and dry. Children 0–59 months of age living within the Mirzapur DSS area and presenting for care at Kumudini Hospital were registered, and those with diarrhea (≥three abnormally loose stools within the previous 24 hours) were screened for disease severity. MSD is defined as diarrheal illness of <seven days duration accompanied by clinical signs of moderate or severe dehydration (sunken eyes, loss of skin turgor) or administration of intravenous fluids based on clinical assessment, dysentery (blood visible in loose stools), or hospitalization based on clinical judgment [17]. Caretakers of children with MSD were invited to enroll their children in GEMS. Up to approximately nine MSD cases were enrolled per fortnight (though more may have presented) in each of three age groups: 0–11, 12–23 and 24–59 months [17]. Stool samples were examined for a wide array of bacterial, viral and protozoal pathogens [18].
Shigella was identified by culture on differential and selective media [18].
Houseflies were counted using a Scudder grill device (slats of wood screwed onto a Z-shaped wooden template to create a lattice), allowing counting in a standardized manner as houseflies typically alight on edges (Figure 1) [19]. One fourth of the Scudder grill was painted yellow to allow the flies to stand out visually; the restricted area allowed more practical counting when fly densities were high. The number of flies on the yellow area was multiplied by 4 to obtain a count for the entire grill. The Scudder grills were placed twice–weekly between 11 am and 2 pm in six sentinel household compounds selected at random from the DSS; grills were put near the household's latrine(s) or in cooking/eating areas where people and flies congregate and where there might be opportunities for mechanical contamination of food and eating utensils. Because the households were selected randomly, they tended to be clustered among the most densely populated area of the DSS (Figure 2). After field workers placed the Scudder grills on the ground or another flat surface, they waited for 30 minutes for flies to settle before counting.
Daily mean, maximum and minimum temperatures were obtained for the study period from the Dhaka weather station, approximately 70km away [20].
Twice-weekly fly counts at all six sites were pooled to provide a Mirzapur-wide weekly count. For exploratory data analysis, weekly counts were summed and divided by the number of weeks falling primarily in a month to provide a mean weekly fly count for each month. For time series analysis, mean weekly fly counts that corresponded to the GEMS biweekly periods were summed and then divided by two. To calculate the estimated total number of children with Shigella-associated MSD presenting to Kumudini Hospital, the proportion of enrolled children testing positive was multiplied by the total number of eligible children presenting during that period. The biweekly average was calculated for each of the three daily temperature statistics, yielding an average mean, average maximum and average minimum temperature for each biweekly period, henceforth referred to simply as mean, maximum and minimum temperature.
A transitional regression model (TRM) for autocorrelated count data was used for the primary analysis, with housefly population density as the explanatory variable and Shigella-associated case counts as the outcome [21]. The TRM is a generalized linear model (GLM) of the Poisson family, with a log link, in which autocorrelation is accounted for by including one or more lagged values of the outcome among the explanatory variables. The scale (whether untransformed or logarithmic) of the fly counts and the number of lags to include was determined by regressing all combinations of lags and minimizing Akaike Information Criterion (AIC) [22] and Bayesian Information Criterion (BIC) values [23] calculated using the estat ic command in Stata 12 (StataCorp, College Station, TX). To assess the possibility of a lagged effect of housefly population density on presentation of Shigella-positive cases, the housefly counts were lagged by one to seven biweekly periods, and AIC and BIC values were calculated to determine whether each lag (or combinations thereof) improved the model fit. To assess for the the possibility that temperature may be confounding the association between the housefly population density exposure and Shigella case count outcome, mean temperature was added to the model at one to seven lags, and the beta for log housefly population density was observed for a change >10% that would suggest confounding. Scatterplots of log Shigella-positive case counts on all lags of fly values and temperature in both untransformed and logarithmic scales were used to determine the appropriate scale.
All statistical analyses were performed using Stata 12.
To estimate the number of Shigella cases that could have been prevented by a public health intervention if flies were reduced to the level observed in the lowest 10% of biweekly periods, we used the Poisson regression output to calculate an attributable fraction (AF) that was adjusted for autocorrelation using the method originally developed by Bruzzi for adjusting for confounders [24], [25]. To enable the calculation, the fly count variable was converted into a decile, then regressed against Shigella-associated MSD case counts. A separate incidence rate ratio (IRR) was calculated for each decile (using the lowest decile as the referent), and was then used to estimate the percentage of infected cases that was attributable to flies. For each decile, this percentage was then multiplied by the total number of Shigella-associated cases to estimate the number of cases attributable to flies. These numbers of attributable cases were summed over the upper nine deciles, then divided by the total number of cases to estimate the AF. The standard error was calculated using a jackknife procedure [25], [26]. This procedure was repeated to estimate the AF of reducing flies to the level observed in the lowest 30% and 50% of periods by setting the referent to the lowest 3 deciles and lowest 5 deciles, respectively. Only the 50 periods on which the lagged effect might operate were counted in the denominator. The total number of Shigella cases observed during the 50 periods was 362.6.
The study covered 53 biweekly periods of GEMS study enrollment. Housefly population density was stable, with the exception of two dramatic peaks that occurred in the late winter to early spring of 2009, and again in 2010 (Figure 3). In February 2009, fly density more than doubled from the previous month, rising to 174 flies/week, and climbed to 238 flies/week in March, before decreasing again to 70 flies/week. The following year, fly density more than tripled, rising from 42 to 143 flies/week in February and 135 flies/week in March before decreasing to 53 flies/week the following month.
Table 1 shows for each age group the number of children with MSD presenting to Kumudini Hospital, the number enrolled into the study, the number and percent positive for Shigella, and the estimated total number of Shigella cases (calculated by multiplying the percent Shigella-positive among enrolled by the total MSD cases). Among 391 children 0–11 months of age, there were relatively few Shigella-associated MSD cases (N = 40) and no obvious seasonal pattern (Figure 3A). By contrast, among the 343 toddlers 12–23 months of age, there was a large number of Shigella-associated MSD cases (N = 194) (Table 1) and a clear pattern showing spikes during the months of April, May and June of both 2009 and 2010 (Figure 3B). Similarly, among the 282 pre-school children 24–59 months of age, there was also a high number of both confirmed and estimated Shigella-associated MSD cases (N = 190) (Table 1) and an obvious pattern showing spikes of shigellosis in March, April and May of 2009 and April and May of 2010 (Figure 3C). The distribution of Shigella species included S. flexneri isolated from 224 children, S. sonnei from 108, S. boydii from 16 and S. dysenteriae from nine; there were five instances of dual infections between S. flexneri and species. There was no apparent association between a particular Shigella species and houseflies.
For the time series analysis, the 12–23 and 24–59 month age groups were combined to enable analysis of biweekly data, thus optimizing the sample size (number of time periods) for the Poisson model, while ensuring that there were enough cases in each period to avoid a zero-inflated data situation. This analysis showed a similar pattern compared with the monthly data and revealed that Shigella cases can vary by large amounts on a biweekly basis (Figure 4).
There was no apparent association between housefly population density and Shigella-associated MSD presentations among infants 0–11 months of age (Figure 3A). However, among toddlers 12–23 months of age, each spike in housefly population density was followed approximately two months later by a surge in Shigella-associated MSD (Figure 3B). Among children 24–59 months of age, the housefly population density spike in 2009 was followed by a surge in Shigella-associated MSD cases approximately one month later, while the housefly spike in 2010 was followed by a surge in Shigella-associated MSD cases about two months later (Figure 3C).
The log scale was found to be more appropriate than the untransformed scale for the lagged fly counts (Figure 5). The best fitting Poisson model used Shigella case counts in the log scale at a lag of one biweekly period to account for autocorrelation (Tables 2–3). Log housefly population density was positively associated with Shigella case counts at a three-period temporal lag, (Table 3). Each log increase of houseflies was associated with an IRR of 1.39 three periods later (95%CI: 1.23 to 1.58).
As the air warmed in springtime to temperatures favorable for housefly reproduction (minimum temperatures above 20C [27]), housefly population density increased both in 2009 and 2010 (Figure 4). As the air continued to warm into the summertime to temperatures that favor growth of Shigella spp. (maximum temperatures approaching 37°C) [28], [29], Shigella-positive case counts were observed to increase in 2009 and 2010 (Figure 4). Because temperature may have been responsible for the association between housefly population density and Shigella (either completely, or in part), we explored average temperature as a potential confounder. As with fly counts, the log scale was found to be more appropriate than the untransformed scale for mean temperature. A GLM Poisson model with Shigella case counts as the outcome, accounting for autocorrelation by including a variable for Shigella case counts in the logarithmic scale at one lag, showed that each log increase in mean temperature was associated with an IRR of 4.09 (95% CI: 1.70 to 9.87) four periods later. When added to the model that included log housefly population density at three lags, log average temperature at four lags resulted in the best model fit. However, the association between log housefly population density and Shigella case counts was essentially unchanged (IRR = 1.37, 95% CI: 1.21 to 1.56) (Table 3). As there was no evidence of confounding, mean temperature was not included in the final model.
Among children <5 years of age, if housefly population density were diminished to the average level of fly count in the lowest decile, an intervention might have prevented 37.4% (95% CI: 16.9 to 57.9) of the total Shigella-associated MSD cases (Table 4). If housefly population density were diminished to the level observed in the lowest 3 deciles, an intervention might have prevented 29.7% (95% CI: 12.9 to 46.6). Reducing housefly population density to the level observed in the lowest 5 deciles might have prevented 26.1% (95%CI: 14.1 to 38.1) of Shigella-associated MSD cases.
The epidemiologic behavior of Shigella infections has fascinated and perplexed epidemiologists and microbiologists for many years. Recognition of the minuscule infectious dose of Shigella (ten colony forming units) [1], [2] that can cause full blown clinical disease explains its transmission by direct fecal-oral contact, its propensity to be spread in sub-populations even in industrialized countries if personal hygiene is compromised, and underlies the propagated epidemic pattern observed in shigellosis outbreaks [30]. Two notable features of Shigella disease in developing countries are its seasonality and its temporal association with houseflies [7]. It has long been recognized that a marked increase in Shigella dysentery cases accompanies or follows shortly after the annual seasonal increase in the density of houseflies. This association has been noted in tropical [12], sub-tropical [11] and temperate [31] regions of the world. The Mirzapur GEMS site offered an opportunity to investigate in depth the association of shigellosis in relation to housefly density. This paper reports results of applying the appropriate time-series analysis to these unique entomological, clinical and microbiologic datasets.
Housefly population density in Mirzapur peaked in February and March of 2009 and 2010 (Figures 3A–C), indicating an annual “fly season”. Housefly densities vary with temperature (20–25°C is most favorable), number of sunshine hours, humidity and availability of breeding sites [27]. In tropical and subtropical climes, fly density increases as mean daily temperature rises following the end of the cool season; however, as mean daily temperatures approach their peak in the hot season, housefly density then decreases. Reports from elsewhere in South and Southeast Asia have also identified marked fly seasons in the springtime before the full heat of summer, as in Uttar Pradesh, India (fly density peak in February and March) [32], North West Frontier Province, Pakistan (peak in March–June) [11], and central Thailand (March–June) [12].
Shigella-positive acute MSD cases also showed a marked seasonality in Mirzapur, with surges occurring in the summer months of March–June 2009 and March–July 2010 (Figures 3 and 4), when air temperature nears the 37°C optimum for growth of Shigella bacteria is [28], [29]. Once again, multiple reports from Asia have similarly noted an April–May surge in Shigella infections, as in Dhaka [33], central Thailand [34] and Jakarta, Indonesia, indicating a regional phenomenon in areas with similar climates.
It is well-recognized that the incidence of shigellosis is much higher in children 12–48 months of age than in infants 0–11 months of age [30], [35]. Accordingly, in Mirzapur, the peak of housefly density that was followed six weeks later by a surge in Shigella-associated MSD cases was seen among children 12–59 months of age (Figure 4), showing a strong, statistically significant association (Tables 2–3).
The shape of the spikes in fly density also corresponded well with the subsequent surges in Shigella-associated case presentations, further suggesting a causal association. During World War 1, Dudgeon observed in Macedonia that a spike in housefly density in April–May was followed one month later by a spike in Shigella incidence in British Army field hospitals [36]. In Mesopotamia between July 1916 and December 1918, Ledingham also noted April–May surges in fly density that were followed two weeks to one month later by an increased incidence of dysentery [37]. Ledingham proposed an explanation for this delay that could also apply to the young children in Mirzapur. He suggested that the springtime surge in fly density leads to an abundance of mechanical vectors capable of contaminating food and cooking and eating utensils with Shigella. Subsequent ingestion of the contaminated food or handling of the contaminated fomites (eating utensils) by susceptibles thereupon establishes many new Shigella infections. This initial burst of Shigella infections that shortly follows the peak fly density results in a temporary surge in the magnitude of the human reservoir of Shigella from which transmission then ensues by more usual modes during the hot summer months, in particular by direct contact transmission.
Because a housefly's habitat can range over a two-mile radius [7], [38], from a few foci where the flies encounter human feces containing Shigella, they can thereupon effectively “seed” a much broader and more dispersed human population with Shigella, as the flies alight on human food and eating utensils. The highly transmissible Shigella can then continue to spread through person-to-person (and occasional foodborne) transmission within families [39] and across wider geographic areas [40]. Indeed, our GLM Poisson time-series model showed precisely this effect – i.e., housefly population density was associated with Shigella MSD three periods (six weeks) later (Table 3), an association that was not confounded by mean ambient temperature. This suggests that houseflies may be seeding the population with Shigella infections, resulting in many small outbreaks at about a six-week lag. The noise inherent in these data does not preclude the possibility of associations occurring at multiple lags, simultaneously. Indeed,Shigella-associated MSD cases appeared to be associated with fly density at several lags, but only a lag of three periods was retained in a model when multiple lags were included together (Table 3). We note also that the logarithmic nature of the association between fly density and log Shigella case counts suggests a biological process (Figure 5).
The AF calculation allowed us to estimate the potential effect of a public health intervention that was highly successful, eliminating fly density peaks by reducing housefly density to a very low level (the average in the lowest decile) (Table 4). We also estimated the effect of a less highly successful intervention (reducing fly density to the average in the 3 lowest deciles) and moderately successful intervention (reducing fly density to the average in the lowest 5 deciles). If a highly successful intervention could be instituted in a setting such as Mirzapur, it might prevent approximately 37% of the Shigella cases observed over the study period, assuming a causal association. More rigorous interventions that decreased fly density to an even lower level presumably might achieve even greater efficacy. A less highly successful intervention might prevent 30% of the Shigella cases observed, while a moderately successful intervention might prevent 26%, showing that an intervention might produce robust results even for moderate reductions in housefly density. Assuming the association between housefly density and Shigella infection is causal, this means fly control could potentially rank highly among other public health interventions as a means of preventing shigellosis (and perhaps other diarrheal infections such as those caused by enterotoxigenic Escherichia coli) [6].
Several limitations should be taken into account when interpreting the results of this study: 1) We assume that children seen at sentinel health centers are representative of all children in the DSS population. However, children seen at the health centers may be subtly different from children in the community whose families do not take them to health centers when they have diarrhea. 2) The use of a limited number of sentinel households where fly density was measured that were clustered in the most highly populated area of Mirzapur may not have been optimal for measuring a site-wide fly density value, and certainly it did not enable analysis by geographic area. However, one may argue that the wide housefly flight radius [7], [38] means that a limited number of surveillance sites may be used to represent flies as if they are a site-wide environmental exposure, as with studies of particulate pollution that often use a single site for their exposure measurements [41]. 3) Lastly, though we found that temperature was not a confounder, the presence of other unknown confounding factors could have resulted in some bias in our estimates.
Baited fly trap technology constitutes one inexpensive, effective tool for reducing housefly density, when implemented as part of a well-designed fly mitigation strategy [6], [7]. Moreover, manufacture of simple fly traps could become a local cottage industry [42], [43]. Whereas the importance of fly control in reducing the incidence of pediatric diarrhea and dysentery was recognized in the past [14], [15], [19], [31], [36], [37], the modern public health community has not generally embraced fly control efforts as a public health imperative. Our experience instructs that this is largely based on the lack of familiarity with information about the role of flies in the transmission of Shigella (and perhaps other enteric pathogens) and a lack of knowledge of of baited fly traps as an effective, affordable, environmentally-friendly measure to reduce housefly density. The time is ripe for a modern, cluster-randomized trial that can not only establish unequivocally whether a causal relationship exists between houseflies and Shigella transmission but can also quantify the effectiveness of baited fly traps (alone or in conjunction with other interventions that decrease fly density) on diminishing Shigella disease.
|
10.1371/journal.ppat.1001060 | A Rapid Change in Virulence Gene Expression during the Transition from the Intestinal Lumen into Tissue Promotes Systemic Dissemination of Salmonella | Bacterial pathogens causing systemic disease commonly evolve from organisms associated with localized infections but differ from their close relatives in their ability to overcome mucosal barriers by mechanisms that remain incompletely understood. Here we investigated whether acquisition of a regulatory gene, tviA, contributed to the ability of Salmonella enterica serotype Typhi to disseminate from the intestine to systemic sites of infection during typhoid fever. To study the consequences of acquiring a new regulator by horizontal gene transfer, tviA was introduced into the chromosome of S. enterica serotype Typhimurium, a closely related pathogen causing a localized gastrointestinal infection in immunocompetent individuals. TviA repressed expression of flagellin, a pathogen associated molecular pattern (PAMP), when bacteria were grown at osmotic conditions encountered in tissue, but not at higher osmolarity present in the intestinal lumen. TviA-mediated flagellin repression enabled bacteria to evade sentinel functions of human model epithelia and resulted in increased bacterial dissemination to the spleen in a chicken model. Collectively, our data point to PAMP repression as a novel pathogenic mechanism to overcome the mucosal barrier through innate immune evasion.
| Some bacterial species contain pathogenic strains that are closely related genetically, but cause diseases that differ dramatically in their clinical presentation. One such species is Salmonella enterica, which contains non-typhoidal serotypes associated with a localized gastroenteritis and serotype Typhi (S. Typhi), the causative agent of a severe systemic infection termed typhoid fever. Conventional wisdom holds, that the ability of S. Typhi to overcome mucosal barriers and spread systemically in immunocompetent individuals evolved through acquisition of new virulence factors, which are absent from non-typhoidal Salmonella serotypes. Here, we demonstrate that acquisition of a regulatory gene, tviA, by S. Typhi alters expression of existing virulence factors (the flagellar regulon) such that molecular structures that are detected by the host innate immune are repressed after entering tissue. We propose that this mechanism contributes to innate immune evasion by S. Typhi, thereby promoting systemic dissemination.
| Epithelial barriers form a first line of defense against microbial invasion. However, the ability to cross this physical barrier does not automatically result in systemic dissemination of the invading microbe. For example, non-typhoidal Salmonella serotypes, such as Salmonella enterica serotype Typhimurium (S. Typhimurium), invade the intestinal epithelium using the invasion associated type III secretion system (T3SS-1) [1] and employ a second type III secretion system (T3SS-2) to survive within tissue macrophages [2]. Despite the ability of non-typhoidal Salmonella serotypes to penetrate the epithelium and survive in macrophages, the infection remains localized to the terminal ileum, colon and mesenteric lymph node in immunocompetent individuals [3]. S. enterica serotype Typhi (S. Typhi) differs from non-typhoidal serotypes by its ability to cause a severe systemic infection in immunocompetent individuals termed typhoid fever [4]. However, little is known about the virulence mechanisms that enabled S. Typhi to overcome mucosal barrier functions and spread systemically, which is at least in part due to the lack of animal models for this strictly human adapted pathogen.
The chromosomes of Salmonella serotypes exhibit a high degree of synteny, which is interrupted by small insertions or deletions. One such insertion in S. Typhi is a 134 kb DNA region, termed Salmonella pathogenicity island (SPI) 7, which is absent from the S. Typhimurium genome and likely originates from a horizontal gene transfer event, as indicated by the presence of flanking tRNA genes [5]. Within SPI 7 lies a 14 kb DNA region, termed the viaB locus [6], which contains genes required for the regulation (tviA), the biosynthesis (tviBCDE), and the export (vexABCDE) of the Vi capsular antigen [7]. In addition to activating expression of the S. Typhi-specific Vi capsular antigen, the TviA protein represses important virulence factors that are highly conserved within the genus Salmonella. These include genes encoding flagella and T3SS-1, whose expression in S. Typhi is reduced by a TviA-mediated repression of the master regulator FlhDC [8]. However, the consequences of these changes in gene regulation for host pathogen interaction remain unclear.
Here we addressed the biological significance of TviA-mediated gene regulation. To explore how acquisition of a new regulatory protein impacted host microbe interaction, we determined whether introduction of the tviA gene into S. Typhimurium resulted in similar changes in gene expression as observed in S. Typhi. We then investigated how these TviA-mediated changes in gene expression affected the outcome of host interaction in an animal model, the chicken, in which S. Typhimurium causes a localized enteric infection.
In S. Typhi, TviA-regulated genes have been identified and encompass the flagella regulon and genes encoding T3SS-1 [8]. To determine how TviA affects gene expression in a non-typhoidal serotype, the tviA gene was introduced into the S. Typhimurium chromosome and the gene expression profile compared to a published gene expression profile of TviA-regulated genes in S. Typhi [8]. Cluster analysis of gene expression profiles revealed that TviA influenced the transcription of similar regulatory circuits in S. Typhimurium and S. Typhi (Figure S1), including genes encoding regulatory, structural and effector components of the T3SS-1, and genes involved in chemotaxis, flagellar regulation and flagellar biosynthesis. To validate results obtained from gene expression profiling, relative transcription levels of genes encoding the flagellar regulator FlhD, the flagellar basal body protein FlgB, the flagellin FliC, and the T3SS-1 regulator HilA were determined in both serotypes by real-time qRT-PCR (Figure 1). Strains lacking the tviA gene (i.e. the S. Typhimurium wild-type strain, the S. Typhimurium ΔphoN mutant and the S. Typhi ΔviaB mutant) contained significantly higher mRNA levels of hilA, flhD, flgB, and fliC than observed in strains carrying the tviA gene (i.e. the S. Typhi wild-type strain, the S. Typhi ΔtviB-vexE mutant and the S. Typhimurium ΔphoN::tviA mutant, a strain in which the phoN gene had been replaced by the tviA gene).
Expression of the flagellum is controlled by the master regulator FlhDC (reviewed in [9]) and is reduced under low osmolarity in S. Typhi compared to S. Typhimurium [10]. Osmoregulation is mediated through the EnvZ/OmpR system in S. Typhi, which controls the availability of TviA. Under conditions of low osmolarity, TviA is expressed and represses flhDC transcription, thereby negatively regulating flagella biosynthesis [8], [11]. To understand the consequences of acquiring tviA by horizontal gene transfer, we determined whether differences in flhDC transcription between S. Typhi and S. Typhimurium could be fully accounted for by TviA-mediated gene regulation. Therefore, expression of flhC in S. Typhi and S. Typhimurium was monitored using transcriptional fusions to the Escherichia coli lacZYA reporter genes (Figure 2).
In the S. Typhi wild-type strain, flhC expression increased with increasing salt concentrations present in the culture medium (Figure 2A, dark gray bars). The S. Typhimurium wild-type strain exhibited a strikingly different flhC gene expression pattern, which peaked at medium salt concentrations (between 0.1 and 0.2 M NaCl) (Figure 2A, light gray bars). Removal of the tviA gene in the S. Typhi ΔviaB mutant resulted in an flhC gene expression pattern (Figure 2A, open bars) that was similar to that of the S. Typhimurium wild-type strain. Similarly, introduction of tviA into S. Typhimurium resulted in a flhC gene expression pattern (Figure 2A, closed bars) resembling that of the S. Typhi wild-type strain. TviA repressed motility under conditions of low osmolarity. Under conditions of high osmolarity (0.3 M NaCl), the presence or absence of the tviA gene did not alter motility in S. Typhi or S. Typhimurium, suggesting that TviA-mediated repression is relieved under this growth condition [8](Figure S2).
These observations suggested that the tviA gene is responsible for differences between S. Typhi and S. Typhimurium in expressing the master regulator of flagella expression and that the tviA gene product can be fully incorporated into the regulatory network existing in S. Typhimurium. Furthermore, these data supported the idea that TviA does not affect flagella expression under conditions of high osmolarity (Figure 2A), which are encountered in the intestinal lumen. In contrast, TviA repressed flagella expression under conditions that closely resembled the osmolarity encountered in human tissue.
We next wanted to investigate whether TviA-mediated changes in gene transcription altered the amount of flagellin protein produced when S. Typhi strains were grown at an osmolarity encountered in tissue (i.e. after growth in DMEM tissue culture medium) (Figure 2B). Expression of the S. Typhi flagellin, FliC (also known as the S. Typhi Hd antigen), was monitored by Western blot (using anti Hd serum). Expression of the heat shock protein GroEL remained constant and was used as a loading control. In the presence of the tviA gene (i.e. in the S. Typhi wild-type strain or the S. Typhi ΔtviB-vexE mutant), a low level of FliC expression was detected when bacteria were grown under conditions mimicking tissue osmolarity (Figure 2B) or under conditions of low osmolarity (Figure S3). Deletion of tviA in S. Typhi (ΔviaB mutant) resulted in increased expression of FliC and introducing the cloned tviA gene (pTVIA1) restored FliC expression to wild-type levels.
Introduction of the tviA gene into the S. Typhimurium chromosome (ΔphoN::tviA mutant) reduced FliC (also known as the S. Typhimurium H1 or Hi antigen) protein levels when bacteria were grown in DMEM tissue culture medium (Figure 2C) or under conditions of low osmolarity (Figure S3). Expression of FljB, the H2 flagellin antigen of S. Typhimurium, was not detected by Western blot under conditions used in this study (data not shown). Collectively, these data suggested that TviA reduced the amount of FliC production in S. Typhi and S. Typhimurium under conditions of tissue osmolarity.
To further test this idea, we mimicked osmotic conditions encountered in the intestinal lumen or in tissue by suspending green fluorescent protein (GFP)-labeled bacteria in medium with high osmolarity or in serum, respectively. After a two-hour incubation, flagella expression was detected on the bacterial surface by flow cytometry. This analysis revealed that flagella were expressed by S. Typhimurium strains under osmotic conditions encountered in intestinal contents, regardless of the presence of tviA (Figure 3A). In contrast, TviA repressed flagellin expression under osmotic conditions encountered in serum, as indicated by a reduction of FliC on the surface of the strain carrying the tviA gene (i.e. the S. Typhimurium ΔphoN::tviA mutant) (Figure 3B).
Invasion of epithelial cells allows Salmonella to gain access to the lamina propria of the small intestine, a process that is accomplished in as little as two hours [12]. To test, whether tviA can repress flagellin expression within this time frame, the S. Typhimurium ΔphoN mutant and the ΔphoN::tviA mutant were grown under conditions of high osmolarity and subsequently shifted to osmolarity encountered in the tissue. Expression of FliC was determined at different time points by Western blot (Figure 3C). In comparison to the wild-type strain, the tviA gene product reduced the amount of flagellin expression as early as two hours after decreasing the osmolarity of the culture medium.
These data were consistent with the hypothesis that TviA does not alter gene expression in the intestinal lumen but rapidly (within two hours) represses flagellin expression upon bacterial entry into tissue.
To mount responses that are appropriate to the threat, the innate immune system in the intestine needs to distinguish between harmless commensal bacteria that are present in the lumen and pathogenic microbes that invade tissue. One player in this process is the intestinal epithelium, which can discriminate between luminal commensals and invasive pathogens by a functional compartmentalization of Toll-like receptor (TLR) 5 expression. TLR5 is a pathogen recognition receptor specific for bacterial flagellin [13]. TLR5 is only expressed on the basolateral surface of the intestinal epithelium [14], [15]. Human colonic epithelial (T84) cells can be polarized to form a model epithelium that recapitulates the sentinel function of TLR5 in detecting bacterial translocation from the lumen [15], [16], [17]. We used this model to investigate whether TviA-mediated repression of flagellin expression in tissue is a mechanism to evade sentinel functions of model epithelia. The expression of CCL20 (encoding the chemokine MIP-3α) and CXCL1 (encoding the chemokine GROα) in polarized T84 cells was flagellin-dependent, as indicated by an absence of responses elicited by non-flagellated S. Typhi and S. Typhimurium mutants (Figure 4, S4, and Table S1). Furthermore, T84 model epithelia responded to basolateral, but not to apical stimulation with purified flagellin (Figure 4A), which was consistent with a functional compartmentalization of TLR5 expression reported previously [15]. Model epithelia were stimulated basolaterally with S. Typhi strains grown under conditions mimicking tissue osmolarity. The presence of tviA in the S. Typhi wild-type strain and the S. Typhi ΔtviB-vexE mutant resulted in a dramatic reduction in the relative transcript levels of CXCL1 and CCL20 (Figure 4A and B) compared to levels elicited by the S. Typhi ΔviaB mutant, which lacked the tviA gene.
To determine whether introduction of the tviA gene into S. Typhimurium would confer the ability to evade detection by model epithelia, polarized T84 cells were stimulated basolaterally with S. Typhimurium strains grown under conditions mimicking tissue osmolarity (Figure 4C). The absence of tviA in the S. Typhimurium wild-type strain and the S. Typhimurium ΔphoN mutant resulted in considerable higher mRNA levels of CXCL1 in T84 cells compared to levels elicited by strains in which flagellin expression was repressed (S. Typhimurium ΔphoN::tviA mutant) or abrogated (S. Typhimurium ΔphoN ΔfliC fljB mutant). In summary, these data suggested that sentinel functions of the intestinal epithelium could be evaded by a TviA-mediated repression of flagellin expression in tissue.
By evading detection through sentinels of the intestinal immune system, TviA-mediated flagellin repression might prevent induction of mucosal barrier functions orchestrated by proinflammatory signals. Since our data pointed to a high degree of similarity between S. Typhi and S. Typhimurium in the mechanisms and consequences of TviA-mediated gene regulation, we reasoned that the relevance of TviA-mediated flagellin repression in vivo could be assessed using animal models of S. Typhimurium infection. The mouse model is not suited for this purpose, because S. Typhimurium rapidly disseminates to the liver and spleen of mice, suggesting that the pathogen can overcome mucosal barrier functions in this host species. In contrast, S. Typhimurium causes a localized gastroenteritis in immunocompetent individuals and is therefore susceptible to mucosal barrier functions encountered in humans. These barrier functions, which are present in humans but absent from mice, are specifically overcome by S. Typhi, as indicated by its ability of to cause typhoid fever. We thus reasoned that the consequences of TviA-mediated flagellin repression should be investigated in an animal, whose mucosal barrier functions, like the ones in humans, are sufficient for preventing systemic dissemination of S. Typhimurium. S. Typhimurium causes a localized enteric infection in chickens, an animal detecting flagellin expression through TLR5 [18], resulting in the activation of mucosal barrier functions [19]. This host was chosen for our analysis.
Groups of four-day-old chickens were infected orally with the S. Typhimurium ΔphoN mutant, the S. Typhimurium ΔphoN::tviA mutant or the S. Typhimurium ΔphoN ΔfliC fljB mutant and bacterial translocation to the spleen was monitored at 8 hours after infection. The presence of the flagellin repressor TviA (ΔphoN::tviA mutant) or the absence of flagellin (ΔphoN ΔfliC fljB mutant) resulted in markedly increased systemic dissemination of S. Typhimurium compared to that observed with flagellated S. Typhimurium (ΔphoN mutant) (Figure 5). In contrast, no significant differences were detected between numbers of the S. Typhimurium ΔphoN mutant, the S. Typhimurium ΔphoN::tviA mutant or the S. Typhimurium ΔphoN ΔfliC fljB mutant recovered from intestinal contents. Since the flagellin proteins are among the most abundant proteins expressed by S. Typhimurium it was conceivable that TviA increased the growth rate by repressing the flagella regulon. However, the tviA-expressing strain (ΔphoN::tviA mutant) and the ΔphoN mutant were recovered in comparable numbers from the spleen of intraperitoneally infected mice 8 h after infection (Figure S5), indicating that TviA did not alter the growth rate of S. Typhimurium in tissue.
Taken together, these data were consistent with the idea that TviA-mediated repression of flagellin expression is a mechanism to overcome mucosal barrier functions, thereby promoting increased bacterial dissemination to the spleen.
The ability to cross epithelial linings is not sufficient for causing systemic bacterial dissemination in immunocompetent individuals, suggesting that additional barrier functions encountered in tissue successfully limit bacterial spread. At least some of these barrier functions are inducible by proinflammatory signals generated during bacterial translocation from the gut [20]. Here we provide support for the idea that evasion of inducible barrier functions by repressing a bacterial PAMP (i.e. flagellin) is a mechanism enhancing systemic bacterial dissemination from the intestine.
S. Typhimurium expresses flagellin during growth in the intestinal lumen as well as in Payers patch tissue, but flagellin expression ceases once bacteria disseminate to internal organs of mice, such as the spleen [21], [22]. Our data suggest that TviA-mediated flagellin repression is not operational in the intestinal lumen, but is rapidly initiated once bacteria encounter tissue osmolarity. The presence of TviA might therefore enable S. Typhi to more rapidly repress flagellin expression upon invasion of the intestinal mucosa (Figure 6) compared to S. Typhimurium, which still expresses flagellin in intestinal tissue [21]. Bacterial translocation across the epithelial barrier into the underlying tissue is observed within 2 hours after infection of ligated ileal loops with S. Typhimurium [12], [23]. TviA markedly reduced flagellin repression within 2 hours of bacterial growth at an osmolarity encountered in tissue. TviA-mediated flagellin repression thus occurred within the time frame required for bacterial translocation across an epithelial barrier in vivo. Similarly, TviA activates expression of the Vi capsular antigen when S. Typhi transits from the intestinal lumen into tissue in a ligated ileal loop model [24].
Expression of flagellin by bacteria arriving in tissue is of consequence, because sentinels monitoring microbial translocation from the gut can detect this PAMP. One of the mechanisms by which the intestinal mucosa distinguishes luminal bacteria from bacteria in tissue can be recapitulated using polarized T84 intestinal epithelial cells, which express TLR5 only on their basolateral surface [15], [17]. Here we show that TviA-mediated flagellin repression enabled bacteria to evade this sentinel function of epithelial cells. It is possible that other cell types may contribute to detecting flagella in vivo. However, regardless of the mechanism(s) by which flagellin stimulates innate immunity in the intestine, our results demonstrate that TviA-mediated flagellin repression resulted in increased bacterial dissemination to the spleen of chickens. The idea that detection of flagella contributes to barrier function is also consistent with the finding that a non-flagellated S. Typhimurium fliM mutant exhibits an enhanced ability to establish systemic infection in chickens compared to the wild-type strain [19]. It may therefore not be a coincidence that S. enterica serotype Gallinarum (S. Gallinarum), the only serotype associated with a severe systemic infection in chickens [25], does not express flagella. Similarly, tight regulation of flagellin expression is required for virulence of Yersinia enterocolitica in mice [26].
It should be pointed out, however, that evading detection of flagella by the innate immune system, although necessary, might not be sufficient for causing systemic disease. For example, Shigella species cause a localized colitis in humans, despite the fact that these pathogens do not express flagellin. A possible explanation for the lower propensity of Shigella species to cause systemic infection is the absence of a Salmonella T3SS-2 equivalent. T3SS-2 is a Salmonella virulence factor important for macrophage survival [2], [27], and its absence in Shigella species may render these pathogens more vulnerable to phagocyte attack. In turn, T3SS-2 may be necessary, but it is not sufficient for systemic dissemination, because S. Typhimurium, which carries this virulence factor, causes a localized infection in immunocompetent individuals. Thus, the ability of S. Typhi to cause systemic disease in humans likely evolved by combining virulence factors conserved among Salmonella serotypes (e.g. T3SS-2 and others) with newly acquired virulence traits (e.g. TviA-mediated flagellin repression and others).
The picture emerging from these studies is that the presence in S. Typhi of a regulator, TviA, which senses the transition of bacteria from the intestinal lumen into tissue, enables the pathogen to rapidly cease flagellin expression when crossing the epithelial lining, thereby preventing the induction of barrier functions that limit bacterial dissemination (Figure 6). At the same time, TviA induces expression of the Vi capsular antigen [24], a virulence factor preventing detection of the pathogen through TLR4 [28]. Collectively, these mechanisms interfere with innate immune surveillance at the mucosal surface [17], [29], [30], [31], resulting in reduced intestinal inflammation [32], [33] and contributing to increased dissemination. It should be pointed out that overcoming barrier functions through TviA-mediated regulation is not sufficient for causing typhoid fever, because subsequent to its initial systemic spread, S. Typhi requires additional virulence mechanisms to establish residence in internal organs, persist and, after a two-week incubation period, cause disease.
Bacterial strains and plasmids used in this study are listed in table 1. Salmonella strains were routinely grown aerobically at 37°C in Luria Bertani (LB) broth (10 g/l tryptone, 5 g/l yeast extract, 10 g/l NaCl) or on LB agar plates. To induce optimal expression of TviA, strains were grown overnight in LB, diluted in either Super Optimal Broth (SOB) (20 g/liter tryptone, 5 g/liter yeast extract, 10 mM NaCl, 2.5 mM KCl, 10 mM MgCl2) [29] or tryptone yeast extract broth (10 g/l tryptone, 5 g/l yeast extract) and aerobically grown to mid-log phase at 37°C. When appropriate, antibiotics were added at the following concentrations: chloramphenicol 0.03 mg/ml, carbenicillin 0.1 mg/ml, and kanamycin 0.05 mg/ml.
Phage P22 HT int-105 was used for transduction as described previously[34], [35]. To construct strain SW335, a P22 lysate of strain TH4054 was used to transduce the flhC5456::MudJ mutation into IR715. SW681 was constructed by transducing the ΔphoN::Kanr mutation of the strain AJB715 into SPN313.
Bacterial RNA was isolated as described previously [8]. Briefly, Salmonella strains were statically grown in 5 ml SOB broth for 2 h. 0.8 ml of a 5% phenol solution (in ethanol) was added and the bacterial cells collected by centrifugation. The pellet was resuspended in 0.4 ml 0.1 mg/ml lysozyme, 1 mM ethylenediaminetetraacetic acid (EDTA) 10 mM, Tris/Cl pH 8.0 and incubated at room temperature for 30 min. Cells were lysed by adding 40 µl 10% sodium dodecyl sulfate (SDS). 0.44 ml 1 M sodium acetate as well as 0.9 ml hot (65°C) phenol was added to the sample and the emulsion was incubated at 65°C for 6 min, incubated on ice for 10 min and centrifuged at 20,000 g for 10 min at 4°C. The upper phase was extracted with 0.9 ml chloroform. After centrifugation at 20,000 g for 5 min at 4°C, the RNA was precipitated by adding 80 µl 1 mM EDTA 3 M sodium acetate pH 5.2 and 1 ml isopropanol. Samples were centrifuged for 30 min at 20,000 g at 4°C and the RNA pellet was washed with 1 ml 80% Ethanol. The air-dried RNA was resuspended in RNase-free water and traces of genomic DNA were removed by rigorous DNase treatment according to the recommendation of the manufacturer (DNA-free DNase treatment, Applied Biosystems).
Gene expression profiling experiments of the S. Typhimurium strains SW124 and SW125 were conducted identically to experiments described previously [8]. Briefly, RNA was extracted from one bacterial culture grown statically in 5 ml SOB broth until the turbidity reached an optical density of OD600 = 0.4−0.5. Microarray hybridization and scanning steps were performed by the UC Davis ArrayCore Microarry facility as described previously [36] with the modifications described in [8]. The TM4 Microarray Software Suite [37] was used for data processing and analysis as described previously [8]. Data from the reference data set (S. Typhi, [8]) was averaged and a cluster analysis of the gene expression profile of S. Typhi and S. Typhimurium was performed by the Clustering Affinity Search Technique (CAST) algorithm [38], [39] (initial threshold parameter of 0.85). Genes identified to be regulated by TviA in S. Typhi and S. Typhimurium are listed in supplementary table S1. Microarray data have been deposited at the Gene Expression Omnibus database under the accession number GSE20321.
Expression of flagellin was determined by Western blot as described previously [8]. In brief, Salmonella strains were grown aerobically for 2 h at 37°C in Dulbecco's Modified Eagle Medium (DMEM) (Invitrogen). For time course experiments, Salmonella strains were grown for 16 h in tryptone yeast extract broth containing 0.3 M NaCl and diluted in Minimum Essential Medium Eagle (MEM) medium (Invitrogen). Culture turbidity (OD600) was measured and bacterial cells were lysed in loading buffer (50 mM Tris/HCl, 100 mM dithiothreitol, 2% SDS, 0.1% bromophenol blue, 10% glyerol). A portion of the lysate corresponding to approximately 5×107 colony forming units (CFU) was resolved by SDS-polyacrylamide gel electrophoresis (PAGE) [40]. Proteins were transferred onto a polyvinylidene fluoride membrane (Millipore) using a semi-dry transfer system (Bio-Rad laboratories). To detect FliC and GroEL expression, rabbit Salmonella H antiserum d (Difco), Salmonella H antiserum i (Difco), and anti-GroEL antiserum (Sigma), respectively, as well as a horse radish peroxidase-conjugated goat anti-rabbit secondary antibody (Bio-Rad laboratories) were used. Chemiluminescence (SuperSignal West Pico Chemiluminescent Substrate, Thermo Scientific) was detected by a BioSpectrum Imaging System (UVP) and images were processed in Photoshop CS2 (Adobe) to adjust brightness levels.
Salmonella strains were grown overnight in tryptone yeast extract broth, diluted 1∶50 in 5 ml tryptone yeast extract broth and incubated for 3 h at 37°C. To adjust the osmolarity, NaCl was added to the media of the subculture as indicated. β-Galactosidase activity was measured as described previously [8], [41]. The experiment was performed in triplicate.
Strains were grown overnight in LB broth, diluted 1∶50 in fresh LB and incubated at 37°C until log phase. 5×108 CFU were re-suspended in either 0.05 ml of mouse serum or in 0.05 ml of tryptone yeast extract broth containing 0.3 M NaCl and incubated for 2 hours at 37°C. Bacteria were cellected by centrifugation at 6000 g for 5 min at room temperature. Pellets were washed twice in fluorescence activated cell sorting (FACS) buffer (1% Bovine serum albumin in phosphate buffered saline [PBS]) and re-suspended in 0.1 ml of FACS buffer. Polyclonal rabbit anti-FliC was added and incubated on ice for 30 minutes. A secondary R-PE conjugated goat-anti rabbit (Jackson ImmunoResearch) was added and incubated on ice for 30 minutes. Bacteria were fixed in 4% Formalin for 1 hour and analyzed using an LSR II flow cytometer (Beckton-Dickinson). Results were analyzed using FlowJo software (Treestar).
The colonic carcinoma cell line T84 was obtained from the American Type Culture Collection (ATCC, CCL-248). T84 cells were routinely maintained in DMEM-F12 medium containing 1.2 g/l sodium bicarbonate, 2.5 mM L-glutamine, 15 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), 0.5 mM sodium pyruvate (Invitrogen), and 10% fetal bovine serum (FBS; Invitrogen). To polarize T84 cells, cells were seeded at a density of 1×106 cells per well in the apical compartment of transwell plates (12 mm diameter, pore size 0.4 µm) (Corning) and incubated for 5 to 10 days until the transepithelial electrical resistance exceeded a value of 1.5 kΩcm2. Media in both compartments was replaced every second day.
Salmonella strains were grown over night at 37°C in LB, diluted 1∶50 in yeast extract broth or MEM medium (Invitrogen) and incubated for 2 h 30 min at 37°C with aeration. T84 cells were activated by adding 2×106 CFU into the basolateral compartment containing 1 ml of media. Purified Salmonella flagellin (InvivoGen) was added into the indicated compartment at a concentration of 1 µg/ml. After 3 h, eukaryotic RNA was isolated as described previously [11] using TRI reagent (Molecular Research Center) In brief, cells were lysed in 0.5 ml TRI reagent and this homogenate extracted with 0.1 ml chloroform (Sigma). The suspension was centrifuged at 12,000 g for 15 min. Nucleic acids were precipitated from the aqueous phase by adding 0.25 ml isopropanol (Sigma) and by centrifugation at 12,000 g for 8 min. The RNA pellet was washed with 75% Ethanol, air-dried and resuspended in water. Traces of DNA were removed by DNase treatment according to the recommendation of the manufacturer (DNA-free DNase treatment, Applied Biosystems).
Real-time quantitative (q) reverse transcriptase (RT)-polymerase chain reaction (PCR) was performed as described previously [11]. 1 µg of DNase treated bacterial or eukaryotic RNA served as a template for RT-PCR in a 50 µl volume. Random hexamer dependent amplification was performed according to the recommendations of the manufacturer (TaqMan reverse transcription reagents; Applied Biosystems). SYBR Green (Applied Biosystems) based real-time PCR was performed in an 11 µl volume employing 4 µl of cDNA as a template. Primers are listed in table 2 and were added at a final concentration of 250 nM. Primers used to detect expression of bacterial genes were designed to amplify targets from both Salmonella serotypes with equal efficiency. Data was acquired by a GeneAmp 7900 HT Sequence Detection System (Applied Biosystems) and analyzed using the comparative Ct method (Applied Biosystems). Bacterial gene transcription in each sample was normalized to the respective levels of guanylate kinase mRNA, encoded by the gmk gene. Eukaryotic gene expression was normalized to the respective levels of GAPDH mRNA.
All procedures described in this study were conducted as described previously [42]. Briefly, specific pathogen free eggs were obtained from Charles River (North Franklin, CT). Eggs were kept in an egg incubator at 38°C and a humidity of 58–65% for 21 days and were periodically rolled for the first 18 days. Chickens were housed in a poultry brooder (Alternative Design Manufacturing, Siloam Springs, AR) at a temperature of 32°C to 35°C. Tap water and irradiated lab chick diet (Harlan Teklad, Madison, WI) was provided ad libitum. S. Typhimurium strains were grown aerobically at 42°C for 16 h in LB broth. Fifteen 4-day-old, unsexed White Leghorn chicks were orally inoculated in groups of five with either 1×109 CFU of the S. Typhimurium strains AJB715, SW474, or SW681 in 0.1 ml LB broth. Animals were euthanized by asphyxiation with CO2 8 h after inoculation. The spleen and a sample of the cecal content were homogenized in sterile PBS and serial ten-fold dilutions spread on LB agar plates containing the appropriate antibiotics.
C57BL/6 mice were obtained from The Jackson Laboratory. Animals were housed under specific-pathogen-free conditions and provided with water and food ad libitum. S. Typhimurium strains were grown aerobically for 16 h at 37°C. Groups of 4 female mice (10 to 11 weeks of age) were injected intraperitoneally with 1×106 CFU of the S. Typhimurium strains IR715, AJB715, or SW474 suspended in PBS. 8 h after infection, animals were euthanized and the spleen collected. Serial 10-fold dilutions of the splenic homogenate were spread on LB agar plates containing nalidixic acid.
For the statistical analysis of ratios (i.e. increases in gene expression), values were transformed logarithmically for further statistical analysis. Data presented in bar graphs are geometric means +/− standard error. A parametric test (Student's t-test) was used to determine whether differences between treatment groups were statistically significant (P<0.05). For data from tissue culture experiments and gene expression analysis, paired statistical analysis was used.
All animal experiments were performed according to Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) guidelines. Experimental procedures with chickens were approved by the Texas A&M University Institutional Animal Care and Use Committee (IACUC). All experimental procedures with mice were approved by the UC Davis IACUC.
|
10.1371/journal.pgen.1000590 | A Computational Screen for Regulators of Oxidative Phosphorylation Implicates SLIRP in Mitochondrial RNA Homeostasis | The human oxidative phosphorylation (OxPhos) system consists of approximately 90 proteins encoded by nuclear and mitochondrial genomes and serves as the primary cellular pathway for ATP biosynthesis. While the core protein machinery for OxPhos is well characterized, many of its assembly, maturation, and regulatory factors remain unknown. We exploited the tight transcriptional control of the genes encoding the core OxPhos machinery to identify novel regulators. We developed a computational procedure, which we call expression screening, which integrates information from thousands of microarray data sets in a principled manner to identify genes that are consistently co-expressed with a target pathway across biological contexts. We applied expression screening to predict dozens of novel regulators of OxPhos. For two candidate genes, CHCHD2 and SLIRP, we show that silencing with RNAi results in destabilization of OxPhos complexes and a marked loss of OxPhos enzymatic activity. Moreover, we show that SLIRP plays an essential role in maintaining mitochondrial-localized mRNA transcripts that encode OxPhos protein subunits. Our findings provide a catalogue of potential novel OxPhos regulators that advance our understanding of the coordination between nuclear and mitochondrial genomes for the regulation of cellular energy metabolism.
| Respiratory chain disorders represent the largest class of inborn errors in metabolism affecting 1 in every 5,000 individuals. Biochemically, these disorders are characterized by a breakdown in the cellular process called oxidative phosphorylation (OxPhos), which is responsible for generating most of the cell's energy in the form of ATP. Sadly, for approximately 50% of patients diagnosed, we do not know the molecular cause behind these disorders. One possible reason for our limited diagnostic capability is that these patients harbor a mutation in a gene that is not known to act in the OxPhos pathway. We therefore designed a computational strategy called expression screening that integrates publicly available genome-wide gene expression data to predict new genes that may play a role in OxPhos biology. We identified several uncharacterized genes that were strongly predicted by our procedure to function in the OxPhos pathway and experimentally validated two genes, SLIRP and CHCHD2, as being essential for OxPhos function. These genes, as well as others predicted by expression screening to regulate OxPhos, represent a valuable resource for identifying the molecular underpinnings of respiratory chain disorders.
| Mitochondrial oxidative phosphorylation (OxPhos) is central to energy homeostasis and human health by serving as the cell's primary generator of ATP. The core machinery underlying OxPhos consists of approximately 90 distinct protein subunits that form five complexes residing in the inner mitochondrial membrane. Complexes I through IV comprise the oxygen-dependent electron transport chain responsible for driving the generation of ATP by complex V. OxPhos is the only process in the mammalian cell under dual genetic control: thirteen essential structural subunits are encoded by mitochondrial DNA (mtDNA) while remaining subunits are encoded by nuclear genes, and are imported into mitochondria [1]. The biogenesis of OxPhos requires many accessory factors responsible for replicating mtDNA as well as transcribing and translating the mitochondrial mRNAs (mtRNA) [2],[3]. Furthermore, the mtDNA-encoded subunits must be coordinately assembled with the nuclear-encoded subunits and metal co-factors to form functional complexes, a process likely requiring far more assembly factors than are currently known [4]. Dysfunction in any of these processes or in the OxPhos machinery itself may result in a respiratory chain disorder, a large class of inborn errors of metabolism [5]. For approximately 50% of patients with respiratory chain disorders, the underlying genetic defect remains unknown, despite excluding obvious members of the OxPhos pathway [4], [6]–[8]. Many of these disorders are likely due to genetic defects in currently uncharacterized OxPhos assembly or regulatory factors.
The OxPhos structural subunits exhibit tight transcriptional regulation that offers a strategy for identifying its non-structural regulators based upon shared patterns of co-expression in microarray experiments [9],[10]. In fact, our laboratory used this approach to identify the gene LRPPRC, which encodes a critical regulator of mtRNA and when mutated is the underlying cause of a respiratory chain disorder called Leigh Syndrome French-Canadian variant [11]. However, while successful in identifying LRPPRC, this previous analysis used only one data set interrogating tissue-specific gene expression [12],[13]. Such co-expression analyses that rely upon individual contexts are not ideal for functional prediction because they are subject to inherent limitations of microarray experiments including technical artifacts, experimental bias and real but confounding correlations with functionally distinct pathways [14].
To overcome these limitations and to generalize our previous approach, we reasoned that large-scale integration across many independent microarray experiments, each surveying a different biological context, would help distinguish genuine co-regulation from random co-expression by identifying genes that consistently co-express with OxPhos. In the yeast Saccharomyces cerevisiae, several groups have performed such expression data integration studies to predict protein function [15]–[17]. With the recent availability of large repositories of mammalian microarray data, it is now possible to apply similar approaches to functionally classify uncharacterized human proteins [18],[19]. Studying mammalian data is especially important for OxPhos given that the mammalian OxPhos pathway differs significantly from the yeast counterpart. For example, S. cerevisiae lacks a proton pumping complex I, the largest OxPhos complex in human cells consisting of forty-five distinct protein subunits [20] and a common target of respiratory chain disease [21],[22]. Furthermore, mammalian mtDNA is circular, whereas yeast mtDNA can form linear concatemers [23]. Moreover, mammalian mtRNA processing differs markedly from S. cerevisaie as mammalian mtRNA does not contain introns and is polyadenylated [24].
In the present paper, we introduce a computational methodology, called “expression screening”, that takes advantage of the growing wealth of freely available mammalian microarray data to search for genes that exhibit consistent co-expression with a given “query” gene set. Applying this procedure to the mammalian OxPhos pathway revealed a number of putative regulators that now emerge as attractive candidate genes for OxPhos disorders. We experimentally validated two genes, CHCHD2 (coiled-coil-helix-coiled-coil-helix domain containing 2) and SLIRP (SRA-stem loop interacting RNA-binding protein; also known as C14orf156) as essential for OxPhos function. We further characterized SLIRP as a RNA-binding domain containing protein necessary for the maintenance of mtRNA protein-encoding transcripts and whose robust co-expression with the nuclear OxPhos subunits provides a putative regulatory link between nuclear and mitochondrial gene expression.
We developed a computational procedure called “expression screening” (Figure 1), which applies large-scale co-expression analysis to a compendium of microarray experiments to predict genes with a functional role in a given pathway, such as OxPhos. We first assembled a compendium of microarray data sets by downloading publicly available expression data from the NCBI Gene Expression Omnibus [18]. We focused on mammalian biology by selecting human and mouse data, and avoided cross-platform discrepancies by limiting our analysis to data from Affymetrix oligonucleotide arrays. To ensure high quality data was used in downstream co-expression analyses, we removed small (n<6) data sets and duplicated experiments. Since previous studies have largely focused on tissue-specific gene expression [10],[11],[13] we decided to instead focus on datasets that measure changes in gene expression within individual tissues or cell types in response to various stimuli. We therefore excluded data sets containing multiple tissues. These filtering steps resulted in a final compendium of 1,427 microarray data sets each surveying transcriptional changes resulting from a different biological context (Table S1).
Expression screening accepts as input this compendium of microarray data as well as a given a query gene set. It then examines each data set in the compendium and calculates for each gene, g, the expression correlation between g and all other genes. The method uses these correlations to produce a rank ordered list of g's expression neighbors and assesses whether the query gene set is significantly over-represented near the top or bottom of this list using an enrichment statistic (see Materials and Methods). This enrichment statistic, following correction for multiple hypothesis testing, serves as a co-expression metric between each gene and the query gene set in that dataset. The procedure is repeated for all datasets in the compendium to generate a co-expression matrix whose values represent each gene's co-expression to the query gene set within a dataset (Figure 1).
Genes that consistently co-express with the query gene set in many independent microarrray datasets likely have a functional role in the query pathway. We therefore sought to generate a measure of consistent coexpression by integrating the co-expression scores for each gene across all data sets. A key feature of the integration scheme is that it offers a principled means of weighting the evidence from each of the data sets. Since the query gene set may itself not be co-expressed in all data sets, we weight data sets according to the intra-correlation of the query gene set to ensure that experiments where the query pathway is itself regulated have greater influence upon the final result. Finally, we apply a data integration procedure that incorporates these weights to arrive at an integrated probability for each gene summarizing its overall co-expression with the query gene set in the microarray compendium (see Materials and Methods). Our data integration procedure is based on the naïve Bayes scheme, which allows independent co-expression evidence from different data sets to strengthen each other, but is modified to be robust against outliers [25],[26]. Importantly, this procedure avoids direct comparison between expression signals from separate data sets, which can introduce artifacts and distort co-expression measures [27],[28].
To validate the expression screening methodology, we first applied it to the well-studied and transcriptionally-regulated cholesterol biosynthesis pathway [29]. We manually curated a set of 19 genes encoding established cholesterol biosynthesis enzymes (Table S2) and applied expression screening to this set. We were able to reconstruct the entire cholesterol biosynthesis pathway within the top 41 high-scoring genes, a substantial improvement over co-expression scores obtained from the best microarray experiment alone (Figure 1B). Among the top 41 co-expressed genes we also recovered the LDL receptor, SREBF2 and INSIG1, three well-known regulators of this pathway (Table S3). A key feature of expression screening is the weighting of each data set according to the intra-correlation of the input pathway. In the case of cholesterol biosynthesis, a variety of data sets representing many distinct biological conditions were given high weights, consistent with the pathway's central role in cellular metabolism (Table S1). Performing data integration without these weights resulted in a substantial loss of specificity (Figure S1). Thus, expression screening is capable of identifying informative datasets in a microarray compendium and reconstructing transcriptionally co-regulated pathways with high precision.
We next applied expression screening to the OxPhos pathway using the 1427 microarray dataset compendium, and a manually curated gene set of nuclear-encoded structural OxPhos subunits (Figure 2A, Table S4). We excluded the mtDNA-encoded subunits from the query set since these were not well measured by the Affymetrix platforms. The resulting co-expression matrix (Figure 2B) reveals the robust coordination of OxPhos gene expression in a large variety of biological contexts. The OxPhos gene set exhibits robust intra-correlation (weight wd>0.75) in nearly 10% of microarray datasets present in the compendium (Table S1). The data set weights enable us to spotlight biological contexts in the compendium for which the modulation of OxPhos gene expression may play an important role. Experiments with large weights include expected conditions such as exercise (GSE1659), Alzheimer's disease (GSE5281) and Pgc1α over-expression (GSE4330) as well as lesser-studied contexts including down-regulation of OxPhos followed by recovery during time-courses of skeletal muscle regeneration (GSE469, GSE5413).
We applied the data integration procedure to identify genes that are consistently co-expressed with OxPhos in the microarray compendium (Table S5). As with the cholesterol biosynthesis pathway, data integration better predicts known genes involved in the OxPhos pathway when compared to the most predictive data set alone (Figure 2C). At a specificity of 99.4%, we were able to recover 85% of the OxPhos pathway (Figure 2C). The integration procedure also lessens confounding correlations with functionally distinct pathways. For example, OxPhos is frequently co-expressed with other genes encoding mitochondrial proteins during mitochondrial biogenesis and turnover, regardless of their specific role in oxidative phosphorylation [30]–[32]. Additionally, OxPhos gene expression may correlate with the expression of other functionally distinct “house-keeping” pathways, especially the cytosolic ribosome, since genes involved in both pathways share a similar set of conserved promoter elements and are controlled by an over-lapping set of transcriptional regulators [33],[34]. In agreement with these findings, we observed significant co-expression (median integrated probability pg = 0.70) of the cytosolic ribosome with the OxPhos subunits (Figure 2D). However, integrating co-expression across all data sets in the microarray compendium clearly distinguished the OxPhos pathway from other mitochondrial genes and components of the cytosolic ribosome, demonstrating the specificity of expression screening (Figure 2D).
We next examined the non-OxPhos genes exhibiting the highest co-expression scores. To ensure that co-expression is conserved among mammals, we required that a gene is co-expressed with OxPhos when analyzing human and mouse microarray datasets independently (pg>0.70 in both species). The top 20 non-OxPhos genes meeting this criterion are shown in Figure 3. Several of the non-OxPhos genes listed in Figure 3 have known metabolic roles in oxidative metabolism such as genes encoding Kreb's cycle enzymes, MDH2 and SUCLG1, as well as several mitochondrial ribosomal subunits necessary for translation of the OxPhos subunits encoded by mtDNA. Other high-scoring genes have never been functionally associated with OxPhos and most lack orthologues in S. cerevisiae. It is notable that recent mass spectrometry studies of highly purified mammalian mitochondria have localized every protein present in Figure 3 to the mitochondria with the exception of HINT1, TCEB2 and MDH1, which are primarily cytosolic proteins [35]–[37].
Recently, two candidates identified by our expression screen, C14orf2 and USMG5, have been co-purified with complex V, having been previously missed in purifications of OxPhos due to their small size and biochemical properties (∼7 kD) [38],[39]. While the functions of these two proteins are still unknown, their physical association with complex V further supports the specificity of the expression screening results for identifying OxPhos-related genes. Interestingly, two other uncharacterized proteins presented in Figure 3, C1orf151 and C12orf62, are also less than 10 kD in size (8.6 kD and 6.4 kD, respectively) and contain a single-pass transmembrane domain similar to C14orf2 and USMG5. These molecular similarities suggest that C1orf151 and C12orf62 may also physically associate with OxPhos.
To validate the results from the OxPhos expression screen, we selected five functionally uncharacterized mitochondrial candidates from Figure 3 for which we could obtain reliable shRNA reagents to experimentally test their role in OxPhos function (C14orf2, USMG5, CHCHD2, SLIRP and PARK7). For each of the five candidate genes, we identified at least two independent, non-toxic shRNAs that deplete mRNA abundance by more than 85% (Figure S2). We were unable to obtain high quality shRNA reagents for other candidates including C1orf151, C19orf20 and C12orf62.
We first silenced each candidate gene in immortalized human fibroblasts and measured the live-cell oxygen consumption rate (OCR) as a general parameter of basal OxPhos activity. Silencing of two candidates, CHCHD2 and SLIRP, significantly reduced cellular OCR by approximately 40% compared to control cells (P<.05; Figure 4A). Additionally, a single shRNA targeting C14orf2 reduced OCR by 35% (P<.05); however, this result may be due to an off-target effect since a second hairpin targeting C14orf2 did not substantially affect OCR (Figure 4A).
Inherited or acquired mutations causing OxPhos dysfunction often destabilize or cause the misassembly of one or more of the five complexes comprising OxPhos. We therefore assessed whether knock-down of any of the five candidates affected complex stability by blotting for “labile” OxPhos subunits whose stability depends on their respective complex being properly assembled in the mitochondrial inner membrane (Figure 4B). Again, we noted that knock-down of SLIRP and CHCHD2 clearly affected OxPhos as both dramatically reduced the abundance of the complex IV subunit, COX2, and to a lesser extent, NDUFB8, a component of complex I. To ensure that CHCHD2 and SLIRP are responsible for maintaining the activities of OxPhos complexes I and IV in native form, we measured the activity of immuno-captured preparations of these complexes (Figure 4C and 4D) [40]. Reducing the expression of both candidates reduced cellular CIV activity (P<.05) while only SLIRP significantly affected CI (P<.05).
The SLIRP protein contains an RNA-binding domain and was previously reported to associate with steroid receptor RNA activator (SRA), a nuclear non-coding RNA, and thereby repress the ability of SRA to activate nuclear receptors [41]. However, SLIRP is predominantly mitochondrial [35],[41]. Since the protein is localized to the mitochondria and is able to bind RNA, we hypothesized that it might affect OxPhos activity by directly modulating the level of mtRNA, either through expression, processing or stability of the mitochondrial transcripts. mtRNA is transcribed from mtDNA in two continuous poly-cistronic transcripts (one from each mtDNA strand), which are subsequently processed to produce eleven OxPhos protein-encoding mRNAs, two ribosomal RNAs (rRNA) and a full complement of tRNAs. The processed mtRNAs are individually regulated by mtRNA stability factors, many of which remain to be identified [42].
To determine whether SLIRP acts in the mtRNA processing pathway, we designed a full panel of qPCR assays to measure the abundance of each protein-coding and ribosomal mtRNA transcript (Table S5). We again used shRNA to reduce SLIRP expression and measured the resulting effect on each mtRNA transcript. Knock-down of SLIRP significantly reduced the abundance of all eleven protein-encoding mtRNA transcripts (Figure 5A), while the mtDNA copy-number was unaffected (Figure 5B). The most pronounced mtRNA reduction was seen for transcripts encoding complex IV subunits as well as the bi-cistronic transcript encoding the ND4 and ND4L subunits of complex I, which is concordant with the specific complex I and IV biochemical defects shown in Figure 4. The effect of SLIRP depletion upon mtRNA appears specific to the protein-encoding mtRNA transcripts since it did not affect the expression of the 12S or 16S mitochondrial rRNAs (Figure 5A), even though these rRNAs are encoded on the same primary poly-cistronic transcript that contains all but one of the mitochondrial mRNAs. To assess whether this regulation of mtRNA by SLIRP is conserved among mammals, we also silenced the gene encoding the mouse ortholog of SLIRP in C2C12 myoblasts. We again observed down-regulation of all three complex IV-encoding mtRNAs (Figure S3).
Since SLIRP is proposed to be alternatively localized to the nucleus, we wondered whether it might affect mtRNA expression indirectly by regulating the nuclear expression of known mtDNA transcription factors or mtRNA regulators. However, shRNA targeting SLIRP did not significantly alter the expression of known nuclear-encoded mtRNA regulators TFAM, TFB1M, and TFB2M, nor did it affect the expression of the nuclear-encoded OxPhos subunit UQCRC1, further suggesting that SLIRP acts within the mitochondria to regulate mtRNA abundance (Figure S4). Finally, we investigated whether over-expression of SLIRP would be sufficient to boost mtRNA abundance in the cell. Over-expressing SLIRP for 48 hours did not alter mtRNA abundance, but over-expression did rescue the down-regulation of mtRNA resulting from knock-down of SLIRP in human cells. Besides demonstrating that the over-expression construct is functional and that the shRNA is on-target, this indicates that SLIRP is not a limiting factor for mtRNA abundance in wild-type cells (Figure 5C).
Since adequate expression of SLIRP is essential for maintaining mtRNA levels, we asked whether SLIRP is transcriptionally regulated in response to a depletion of mtRNA. We used ethidium bromide (EtBr), a DNA-intercalating agent that is selectively absorbed by mitochondria and reduces mtDNA copy-number and mtRNA expression in the cell [43]. Following treatment with EtBr for four days, we did not observe any compensatory increase in SLIRP expression as mtDNA and mtRNA were depleted in a concentration-dependent manner (Figure 6A). Surprisingly, however, we did observe a dramatic reduction in SLIRP at the protein level, in a manner depending on the concentration of EtBr (Figure 6B). This suggests that the stability of SLIRP depends upon either mtDNA copy-number or mtRNA abundance. A similar phenomenon has been previously reported for TFAM, a critical regulator of both mitochondrial DNA and RNA [30]. TFAM coats the mtDNA to protect it from degradation but TFAM is also dependent upon mtDNA for its own protein stability [44]. We wondered whether SLIRP, being an RNA-binding protein, depends exclusively upon mtRNA for its stability. To assess whether mtRNA rather than mtDNA quantity is important for stabilizing SLIRP we used shRNA to deplete cells of LRPPRC, a mitochondrial protein necessary for maintaining mtRNA expression but not mtDNA copy-number [45] (Figure 6C). We again observed a substantial drop in SLIRP protein abundance, likely indicating a mutual partnership between SLIRP and mtRNA where each is responsible for the other's stability within the mitochondria (Figure 6D).
To predict novel OxPhos regulators we have developed a method called expression screening, which utilizes the inherent strong transcriptional co-expression of the known OxPhos structural subunits to identify other genes sharing similar expression profiles in a compendium of microarray data. While co-expression analysis alone cannot fully distinguish functionally relevant co-regulation from mere correlation, we have demonstrated that the integration of evidence from hundreds of biological contexts significantly enhances predictive power. In this manner, we were able to build a reliable classifier for membership in the OxPhos pathway that will be a useful resource for prioritizing candidate genes in patients with respiratory chain disorders.
Expression screening predicted several functionally uncharacterized genes as novel regulators of the OxPhos pathway. Of these, CHCHD2 and SLIRP resulted in clear OxPhos deficits when targeted by shRNA. CHCHD2 is a member of a family of proteins containing the CHCH domain (coiled-coil-helix-coiled-coil-helix; PFAM#06747). Conserved cysteines within this motif have previously been implicated in metal coordination or transport suggesting that CHCHD2's role in stabilizing complex IV may be related to regulation of the complex IV copper centers [46]. Interestingly, the CHCH domain is also found in an OxPhos complex I subunit, NDUFA8, and in another Complex IV assembly factor, COX19 [46],[47]. Further study of this important domain should lend insight to the assembly and function of OxPhos.
Silencing of some high-scoring expression screening candidates including USMG5, C14orf2 and PARK7 did not result in an oxidative phenotype. These outcomes may reflect common caveats with shRNA experiments including insufficient protein knock-down, functional redundancy or lack of the proper experimental context. For example, our expression screen implicates the gene PARK7, named for causing Parkinson's disease when mutated, as a key player in OxPhos biology. Currently, there is no established role for PARK7 in the OxPhos pathway [48]. While we did not observe an effect on basal oxygen consumption when perturbing PARK7 expression in our cell line, PARK7 protein may still be an important OxPhos regulator that acts in a context-dependent manner. Others have reported PARK7 to act as an antioxidant that scavenges mitochondrial radical oxygen species, a harmful by-product of an active OxPhos system [49]. Additionally, cells depleted of PARK7 are hyper-sensitive to rotenone treatment, a potent complex I inhibitor [50].
SLIRP is consistently co-expressed with the nuclear OxPhos machinery and regulates the abundance of the mitochondrial protein-encoding transcripts. These properties raise the interesting possibility that SLIRP is co-regulated with the nuclear OxPhos genes in order to coordinate nuclear and mitochondrial OxPhos gene expression. This phenomenon has also been previously reported for genes encoding the mtDNA transcription factors: TFAM, TFB1M and TFB2M [51],[52]. In certain biological contexts, these genes have been noted to be co-expressed with nuclear OxPhos genes [51],[52]. In our expression screen, we did observe co-expression of these factors with OxPhos in certain microarray experiments, but this co-expression was not frequent enough to generate a high score in the overall data integration. In contrast, SLIRP scored among the top 20 genes in the genome for its co-expression with OxPhos (Figure 3), strongly implicating a role for SLIRP in synchronizing nuclear and mitochondrial gene expression.
The precise molecular mechanism by which SLIRP maintains mtRNA is not yet clear. To date, most studies of mtRNA maintenance has focused upon the core transcriptional machinery responsible for transcribing the primary poly-cistronic transcripts. This essential machinery includes the mitochondrial polymerase, POLRMT and its partner MRPL12, as well as the transcription factors TFAM, TFB1M and TFB2M [3]. However, key factors in mammalian mtRNA post-transcriptional processing and stability remain unknown [24]. For example, a human mitochondrial poly(A) polymerase (mtPAP) has been recently identified [53], but this protein does not contain an obvious RNA-binding domain, suggesting that it requires one or more currently unidentified RNA-binding partners [24]. Additionally, in S. cerevisiae, several mitochondrial RNA-binding proteins stabilize mtRNA transcripts, but proteins with similar functions have not been found in mammals [54]–[56]. Given its involvement in maintaining mtRNA it is tempting to speculate that SLIRP fulfills one or more of these roles in mammalian mtRNA biology. SLIRP joins a small cast of RNA-binding mitochondrial proteins that are responsible for maintaining steady-state mtRNA in human cells: LRPPRC, SUPV3L1 and PTCD2 [45],[57],[58]. LRPPRC is of central importance for OxPhos function, and patients harboring mutations in LRPPRC develop the French-Canadian variant of Leigh syndrome, a devastating hepatocerebral metabolic disorder [11]. In our study, shRNA targeting SLIRP phenocopies mutations in LRPPRC by causing loss of mtRNA and a significant reduction in complex IV activity. Given these similarities, SLIRP should be considered a candidate gene for respiratory chain disorders.
Many of the proteins encoded by the human genome are still functionally uncharacterized and methods such as expression screening will be useful in closing the gaps in our knowledge of cellular pathways [59]. While we have applied expression screening to the cholesterol biosynthesis and OxPhos pathways, this technique is readily extendible to any gene set exhibiting transcriptional co-regulation. In a separate study we show the utility of this method in identifying new mitochondrial proteins important for heme biosynthesis [60]. Expression screening is not the only tool that should be considered for functional prediction. Network-based integration of protein-protein interaction data and other data integration methods have also been successfully applied to predict the functions of uncharacterized proteins [19],[61],[62]. Combining expression screening with these or other methods could possibly yield more accurate predictions, especially in cases where transcriptional regulation may not be the dominant mode of regulatory control. Still, microarray gene expression data has several advantages: it is by far the most abundant data source available, offers a more unbiased approach than most techniques, and permits investigating gene function in specific biological contexts. As public repositories of microarray data continue to grow at an accelerating pace, we anticipate that expression screening will become an increasingly important tool for discovering gene function.
A total of 2,052 mouse and human microarray data sets were downloaded from the NCBI Gene Expression Omnibus [18] in March 2008 for the Affymetrix platforms HG-U133A (GEO GPL96), HG-U133+ (GEO GPL570), MG-U75Av2 (GEO GPL81), Mouse430A (GEO GPL339) and Mouse430 (GEO GPL1261). We discarded data sets with less than 6 arrays as well as data sets containing multiple tissues. We then merged overlapping data so that no two data sets shared identical arrays, resulting in a final compendium of D = 1,427 data sets. For a given query gene set S, each data set d and each gene g, we calculated the vector of Pearson correlation coefficients rgj between g and all other genes j. We then define the correlation between g and S as the GSEA-P enrichment score (ES) statistic [63] with g as the “phenotype” variable and N representing the total number of genes.
We next calculated randomized enrichment scores ES0 by randomly permuting values (arrays) for gene g, re-calculating rgj and applying the above formula. We pooled N0 = 100,000 permuted ES0 values from all N genes to estimate the marginal null distribution of enrichment scores. From this we estimated the global false discovery rate (FDR) of each actual ES value [63],[64] as the ratio of tail probabilities:
We take qgd = 1−FDRgd to represent the probability of co-expression of gene g with the query gene set S in data set d. The data set weights wd were defined as the average of qgd across the query genes. We then integrated these probabilities using a robust Bayesian formula [25] to obtain a final probability pg of co-expression of gene g with the query gene set,where is the average of qgd in data set d. This method of data integration assumes conditional independence between data sets given the co-expression hypothesis, which allows concurring evidence from multiple data sets to reinforce each other in calculating the integrated probability. Incorporating the prior p0 affords some robustness to outliers in terms of qgd values close to 0 or 1, which can arise from the permutation-based FDR estimation. For the OxPhos expression screen the prior p0 was set to 5%, roughly corresponding to the fraction of mitochondrial genes in the genome [35]. Since the query genes are used to calculate the weights wd, the sensitivity and specificity was estimated using leave-one-out cross-validation, with one gene withheld from the weights calculation in each iteration.
We mapped Affymetrix probesets to NCBI Homologene identifiers using a previously described method [65]. For the query gene sets (OxPhos and cholesterol pathways) we validated each gene's mapping by Blast. It is important that each query gene is represented only once in the gene set. For cases in which multiple Affymetrix probesets map to a gene in the query gene set, we chose the probeset with the least potential for cross-hybridization according to Affymetrix probeset annotations. Specifically, we used the following Affymetrix probeset suffix hierarchy (‘at’>‘a_at’>‘s_at’>‘x_at’). In cases where there were ties, we chose the lower numbered probeset to represent a gene.
We performed expression screening as described above separately for each microarray platform, using Affymetrix probeset-level data. When integrating across all array platforms, we chose for each Homologene identifier the probeset with maximal pg in the platform-specific screen. We then re-integrated across all data sets, ignoring missing values, to produce the final list of probabilities, pg (Figure 3, Table S3, Table S5).
MCH58 immortalized human fibroblasts were kindly donated by Eric Shoubridge [66]. 293T and C2C12 cells were received from the American Type Culture Collection (CRL1772 & CRL11268). Unless otherwise indicated, all experiments were carried out in DMEM, 4.5 g/L glucose, 10% FBS (Sigma #2442) supplemented with 2 mM glutamine, 100 I.U Penicillin and 100 ug/ml Streptomycin.
Lentiviral vectors for expressing shRNA (pLKO.1) were received from the Broad Institute's RNAi Consortium or from Open Biosystems. Unique identifiers of each shRNA construct can be found in Figure S1. Procedures and reagents for virus production are adapted from the Broad Institute's RNAi Consortium protocols [67]. Briefly, 400,000 293T cells were seeded in a 24-well dish and 12 hr later triple-transfected with pLKO.1-shRNA, a packaging plasmid (pCMV-d8.91) and a pseudotyping plasmid (pMD2-VSVg) using Fugene6 reagent at 3∶1 (reagent∶DNA) (Roche #11815091001). Media was refreshed 18 hr post-transfection, supplemented with 1% BSA and virus collected 24 h later. For infection, 30,000 cells were seeded in a 24-well dish. 30 ul viral supernatant was added to cells for a final volume of 500 ul media containing 8 ug/ml polybrene (Sigma #H9268). The plates were spun 800rcf for 30 min at 32°C, returned to 37°C and 24 h post-infection were selected for infection with 2 ug/ml puromycin (Sigma #P9620). RNA for assessing knock-down efficiency was isolated 7–10 days post-infection.
RNA was isolated using the RNeasy system (Qiagen #74106) with two repetitions of DNAse digestion to remove mtDNA and genomic DNA from the sample. 1 ug of RNA was used for 1st-strand cDNA synthesis using a mix of poly(dT) and random hexamer primers (SuperScript III, Invitrogen, #18080). Genomic DNA used for the analysis of mtDNA quantity per cell was isolated using Qiagen DNeasy system. 1 ng genomic DNA was used for multiplex qPCR analysis to simultaneously measure nuclear DNA and mtDNA (See Table S6 and [68]). qPCR of cDNA and genomic DNA was performed using the 96-well ABI7500 qPCR system in 20 ul reactions prepared with 2× master-mix (ABI #4369510), the appropriate 20× ABI taqman assay (Table S6) and diluted cDNA sample.
A pDONR-221 Gateway clone for SLIRP, used previously by our laboratory for cellular localization studies [35], was sequence verified and cloned into a pcDNA destination vector in-frame with a C-terminal V5-His tag (pDEST40, Invitrogen #12274-015). 293T cells were infected with validated shRNAs targeting either SLIRP or GFP as described above. Seven days post-infection, 400,000 cells were seeded in a 24-well dish and transfected with either pcDNA-LacZ-V5-His or pcDNA-SLIRP-V5-His. 48 hrs post-transfection, cells were harvested for downstream analyses.
Live-cell oxygen consumption readings (OCR) were performed using a 24-well Seahorse XF24 Bioflux analyzer. MCH58 cells were seeded at a density of 30,000/well in unbuffered media (4.5 g/L glucose, 4 mM glutamine, 1 mM pyruvate in DMEM, pH 7.4). The XF24 analyzer was set to read OCR/well as an average of four 3 min measurements with a 1 min mixing interval between measurements. Each plate contained four different samples of five replicates each with one sample always being the shRNA vector control (pLKO.1) that was used for normalization and comparison between experiments. Each “batch” of four samples was measured on four different plates summing to 20 replicates per sample. Samples were randomized within each plate to avoid plate-position effects and experiments were repeated multiple times to ensure reproducibility. To correct for cell seeding errors, we measured the total protein content per well after each experiment (BCA method) and normalized the OCR per well by dividing by the corresponding protein concentration.
5 ug of cleared whole cell lysate isolated in RIPA buffer was used per lane on a 4–12% Bis-Tris gel (Invitrogen, #NP0321) and blotted on a PVDF membrane (Invitrogen, #LC2005) using a semi-dry transfer apparatus (Bio-Rad), 15 V, 20 min. Membranes were blocked for 2 hr at room temperature in tris-buffered-saline solution (Boston BioProducts #BM300) with .1% Tween-20 and 5% BSA (TBS-T-BSA). Primary antibodies were incubated with the membrane overnight at 4°C in TBS-T-BSA at the dilutions reported in Table S7. Secondary sheep-anti-mouse (GE-HealthCare, #na931v) or sheep-anti-rabbit (GE-HealthCare, #na934v) antibodies were incubated with the membrane at a 1∶5,000 dilution in TBS-T-BSA for 45 min at room temperature. The membrane was developed using Super Signal West Pico (Pierce, #34077).
Complex I and Complex IV Dipstick activity assays were performed on 20 ug and 25 ug cleared whole cell lysate, respectively, according to the manufacturer's protocol (Mitosciences #MS130 and #MS430). 30 ul of lysis buffer A was used per 500,000 cells for lysis and solubilization.
|
10.1371/journal.ppat.1002675 | The Lipopolysaccharide Core of Brucella abortus Acts as a Shield Against Innate Immunity Recognition | Innate immunity recognizes bacterial molecules bearing pathogen-associated molecular patterns to launch inflammatory responses leading to the activation of adaptive immunity. However, the lipopolysaccharide (LPS) of the gram-negative bacterium Brucella lacks a marked pathogen-associated molecular pattern, and it has been postulated that this delays the development of immunity, creating a gap that is critical for the bacterium to reach the intracellular replicative niche. We found that a B. abortus mutant in the wadC gene displayed a disrupted LPS core while keeping both the LPS O-polysaccharide and lipid A. In mice, the wadC mutant induced proinflammatory responses and was attenuated. In addition, it was sensitive to killing by non-immune serum and bactericidal peptides and did not multiply in dendritic cells being targeted to lysosomal compartments. In contrast to wild type B. abortus, the wadC mutant induced dendritic cell maturation and secretion of pro-inflammatory cytokines. All these properties were reproduced by the wadC mutant purified LPS in a TLR4-dependent manner. Moreover, the core-mutated LPS displayed an increased binding to MD-2, the TLR4 co-receptor leading to subsequent increase in intracellular signaling. Here we show that Brucella escapes recognition in early stages of infection by expressing a shield against recognition by innate immunity in its LPS core and identify a novel virulence mechanism in intracellular pathogenic gram-negative bacteria. These results also encourage for an improvement in the generation of novel bacterial vaccines.
| Brucellosis is one of the most extended bacterial zoonosis in the world and an important cause of economic losses and human suffering. The causative agents belong to the genus Brucella, a group of highly infectious gram-negative bacteria characterized by their ability to escape early detection by innate immunity. This stealthy behavior effectively delays the development of immunity, creating a gap that is used by the bacterium to penetrate into a variety of cells and to activate complementary virulence mechanisms such as the type IV secretion system. By this manner, the brucellae divert intracellular trafficking to reach a safe multiplication niche and establish chronic infections. Our results show that an inner section of the Brucella LPS (a molecule that in most bacteria is detected by innate immunity), effectively contributes to block recognition by soluble molecules and cellular receptors of the host innate immune system. Accordingly, a mutation disrupting the inner but no other lipopolysaccharide sections generates attenuation by impairing the stealthiness characteristics of this pathogen. This is the first Brucella mutant in which attenuation is specifically linked to the bolstering of immunity against this pathogen. Therefore, this new virulence mechanism opens the way for the development of improved bacterial vaccines.
| Innate immunity plays a fundamental role in the defense against microorganisms. In addition to the passive action of physical and physicochemical barriers, the effectiveness of innate immunity relies on pathogen recognition receptors that quickly perceive the presence of invaders. Upon binding to microbial molecules bearing pathogen-associated molecular patterns (PAMP), pathogen recognition receptors trigger a cascade of signals that include the release of proinflammatory mediators, which in turn may activate adaptive immunity. Cells like macrophages and dendritic cells are equipped with a variety of pathogen recognition receptors, which can be activated by bacterial PAMP such as lipoproteins, glycolipids, peptidoglycan or DNA. However, some bacteria are able to generate chronic infections by residing and multiplying in these host cells. A relevant model of this kind of pathogens is represented by the genus Brucella [1], a group of α-Proteobacteria that have a great impact on animal and human health worldwide, and whose virulence relies in part upon the failure of pathogen recognition receptors to sense Brucella during the initial stages of infection [2], [3].
Brucella surface lipoproteins, ornithine lipids, flagellum-like structures and the LPS do not bear a marked PAMP [2], [3], [4], [5]. The most conspicuous PAMP bearing component of the surface of gram-negative bacteria is LPS, also known as endotoxin, a molecule made of three sections: lipid A, core oligosaccharide and O-polysaccharide (O-PS). Typically, LPS express a lipid A made of a glucosamine disaccharide linked predominantly to C12 to C14 acyl chains in ester, amide and acyl-oxyacyl bonds. This structure carries a characteristic PAMP that is recognized by the TLR4-MD2 receptor-coreceptor complex, triggering potent proinflammatory responses that may lead to endotoxic shock. Since Brucella lipid A (a diaminoglucose disaccharide substituted with C16, C18, C28 and other very long acyl chains [6]) structurally departs from the canonical lipid A recognized by TLR4-MD2 [7], it is postulated to play a key role in the stealthy behavior of this pathogen; indeed Brucella LPS is poorly endotoxic [2], [4], [8], [9]. In addition, the O-PS characteristic of smooth brucellae like B. abortus, B. melitensis or B. suis confers serum and complement resistance, a property not uncommon in the O-PS of gram-negative pathogens, and also modulates the entry into cells [1]. It is not known whether the LPS core sugar structure of Brucella or any other gram-negative intracellular pathogen has a direct role in intracellular virulence. Indeed, mutants of smooth Brucella affected in the LPS core show different degrees of attenuation, but these results cannot be unambiguously interpreted because all core mutants described so far simultaneously lack the O-chain. (i.e., are rough [R] mutants) and thus are attenuated [10]. Here, we report that mutation of a hitherto unidentified Brucella LPS core glycosyltransferase gene generates attenuation without affecting the assembly and linkage of the O-PS or the lipid A section. This attenuation is not caused by a physiological defect associated with a damage of the envelope properties but rather by the removal of a core section that hampers recognition by complement, bactericidal peptides and TLR4-MD2, thus representing a novel virulence mechanism.
Up to now, only one Brucella core glycosyltransferase has been identified [20]. Since LPS core structures are often conserved in phylogenetically related organisms, we scanned the genomes of α-Proteobacteria looking for orthologues of glycosyltransferases not involved in O-PS synthesis. We identified B. abortus ORF BAB1_1522 as encoding an orthologue (78% similarity) of the Rhizobium leguminosarum core mannosyltransferase LpcC [11] and named it wadC following accepted nomenclature [12]. We then constructed a non-polar mutant (BaΔwadC; Figure S1) of virulent B. abortus 2308 NalR (Ba-parental) [9]. This mutant showed the same dye and phage sensitivity pattern as the parental strain, and its growth rate in bacteriological media was similarly unaffected (Figure S2). In addition, the mutant reacted normally with polyclonal antibodies to the O-PS and was smooth by the crystal violet-exclusion and acriflavine tests, suggesting the presence of a typical smooth LPS (S-LPS). Thus, for a better analysis of possible LPS defects, we extracted the Ba-parental and BaΔwadC LPSs using the phenol-water protocol [13], [14]. SDS-PAGE and Western-blots with anti-O-PS and anti-core monoclonal antibodies showed that the wild type LPS of the parental Ba-parental strain consisted of both S and R fractions, as expected (Figure 1). However, the BaΔwadC LPS extracts showed a different migration pattern suggesting a core defect. This peculiarity was confirmed by its lack of reactivity of the anti-core monoclonal antibodies A68/24D08/G09 (Figure 1), A68/24G12/A08 and A68/3F03/D5 (Figure S3). The implication of wadC was confirmed by complementation with plasmid pwadC (strain BaΔwadC-comp) (Figure 1). In addition, the lipids A of both Ba-parental and BaΔwadC were dominated by molecules carrying the very long chain fatty acids typical of Brucella LPS, with minor and not consistent differences in the intensity of some peaks as detected by mass spectrometry (Figure S4).
BaΔwadC displayed attenuation in mice (Figure 2A, left panel) with an estimated spleen clearance time of 27 weeks (66 weeks for Ba-parental). This attenuation, however, was less marked than that of a R mutant (BaTn5::per) blocked in the synthesis of the only sugar (N-formylperosamine) of the O-PS but with a complete core (Figure 2A, left panel). BaΔwadC induced a transient splenomegaly, in contrast to the increasing splenomegaly observed in Ba-parental and the almost absence of splenomegaly observed in the BaTn5::per inoculated mice (Figure 2A, right panel). Complementation with plasmid pwadC restored the replication, persistence and splenomegaly of BaΔwadC back to levels observed with the virulent Ba-parental (Figure S5).
Upon infection, BaΔwadC triggered a more intense leukocyte recruitment in the peritoneum and blood than Ba-parental (both at 106 CFU/mouse) but less than the highly endotoxic Salmonella enterica subspecies enterica serotype Typhimurium (S. Typhimurium) (at 105 CFU/mouse) (Figure 2B). Concurrently, cytokine levels (TNFα, IL-6, IL-12 p40/p70 and IL-10) measured in the serum of BaΔwadC infected mice were always higher than those induced by the Ba-parental (Figure 2C).
We then investigated whether the attenuation and inflammatory responses in mice were reproduced in target cells (dendritic cells and macrophages). BaΔwadC but not the complemented strain BaΔwadC-comp was killed in bone marrow-derived dendritic cells (BMDC), although less rapidly than the virB9 type IV secretion mutant used as a reference of attenuation (Figure 3, upper panel). Moreover, in contrast to the Ba-parental strain, which segregated at 24 h post-infection within the calnexin-positive endoplasmic reticulum, the BaΔwadC mutant-containing vacuoles colocalized with the lysosomal LAMP-1 marker, indicating its failure to reach the endoplasmic reticulum replication niche (Figure 3, lower panel). This attenuation was paralleled by an increased number of Dendritic Cells Aggresome-Like Structures (DALIS) in infected BMDC (Figure S6) indicating that dendritic cells were programmed to undergo a maturation process. Consequently, secretion of IL-12 and TNF-α proinflammatory cytokines in BaΔwadC-infected BMDC at 24 h post-infection reached intermediate levels between those of Ba-parental and S. Typhimurium (Figure 4A, left panel). These observations were fully consistent with the results obtained in mice. Interestingly, BaΔwadC multiplied as Ba-parental in bone marrow-derived macrophages (BMDM; Figure 3, upper panel), Raw 264.7 macrophages or HeLa cells (not shown) (see above).
In the absence of antibodies, Smooth Brucella cells are poor activators of complement and are thus markedly resistant to the bactericidal action of normal serum, a property that has been attributed to the O-chain [15]. BaΔwadC was more sensitive than Ba-parental to the bactericidal action of serum and comparison with the O-PS defective BaTn5::per mutant suggested that the LPS core may be as important as the O-PS (Figure 5A). Brucellae are also resistant to bactericidal polycationic peptides [16], [17], a property linked to a steric hindrance by the O-PS [17] as well as to the low negative charge in the core and lipid A LPS sections, as assessed by physicochemical methods [14]. We tested the sensitivity of BaΔwadC to two potent polycationic lipopeptides, polymyxin B and colistin and demonstrated a greater sensitivity of BaΔwadC (Figure 5B) to these two agents. These results were confirmed using synthetic poly-L-lysine and poly-L-ornithine (not shown).
We then carried out studies with purified LPSs. In experiments performed to stimulate BMDC with purified LPS, the saturating concentration for BaΔwadC LPS was 10 µg/mL. This is one hundred times more than the corresponding concentration of E. coli LPS, a result that illustrates the importance of the expression of an endotoxic lipid A for efficient cell activation. However, at 10 µg/mL, BaΔwadC LPS induced the secretion of IL-12 and TNF-α whereas Ba-parental LPS showed very low levels of cytokine secretion (Figure 4A, right panel) also showing that the presence of the core oligosaccharide in addition to the expression of a long acyl-chain lipid A prevent Brucella LPS to show a marked endotoxicity. These results closely matched those obtained in infected BMDC (Figure 4A, left panel). Cytokine secretion was TLR4-dependent since no pro-inflammatory cytokines were detected in BMDC from TLR4 but not TLR2 (Figure 4B), TLR6 (not shown) or TLR9 (Figure 4B) knockout mice stimulated with BaΔwadC LPS. Finally, BMDC showed an intermediate matured phenotype as judged by the expression of CD40 and CD86 co-stimulatory molecules and surface MHCII (Figure 4C) that led to efficient cytokine secretion. Although BaΔwadC multiplied in macrophages (BMDM) (Figure 3, upper panel), its core-defective purified LPS was capable of triggering a cytokine response higher than that of the parental Ba-LPS (Figure S7B). These results suggest that signaling by the mutated LPS during BMDM infection and subsequent cell activation did not occur early enough to prevent the mutant from reaching the replicative niche.
Purified S-LPS forms supramolecular aggregates in which the lipid A acyl chains display a characteristic and temperature-dependent fluidity that increases upon the disturbance of the aggregate caused by binding of bactericidal peptides and complement molecules [18]. Accordingly, we measured this parameter in the absence or the presence of serum or polymyxin B. In the absence of any agent, the β↔α transition that marks the shift from crystalline to fluid phase took place in the 30 to 40°C range for B. abortus wild type LPS, with a transition temperature of 37°C (Figure 5C). The LPS of the BaΔwadC mutant showed a very different acyl chain fluidity profile with a transition temperature between 45 and 55°C, and with a markedly more restricted fluidity below transition temperature than the wild type LPS. Therefore, the mutant LPS aggregates were in the crystalline phase at physiological temperatures and, since the acyl-chain composition of lipid A was not significantly affected (Figure S4), we attributed this to the disruption induced in the core structure by mutation of wadC. Despite this greater rigidity, the LPS aggregates of the BaΔwadC mutant were clearly affected by normal serum whereas those of Ba-parental LPS were not (Figure 5C). Moreover, when we measured the effect of polymyxin B on acyl chain fluidity, we found a much less marked effect on wild type LPS than on the LPS of the BaΔwadC mutant (Figure 5B). These results show that the core defect is uncovering the complement and polycations targets (Kdo and lipid A phosphate groups) that exist in the innermost sections of LPS [19], and are in agreement with the serum complement and bactericidal peptide sensitivity of the mutant.
Most endotoxic effects of enterobacterial LPS depend on the interaction with the TLR4 co-receptor MD-2, an event that triggers a cascade of signals leading to the NF-kB-dependent expression of immune response genes and a subsequent proinflammatory response [7]. Therefore, we explored the interaction of BaΔwadC LPS with MD-2 using a competitive ELISA using an antibody that recognizes free- but not LPS-bound human MD-2 (hMD-2) [20]. Whereas Ba-parental LPS did not interact with MD-2 in the range of concentrations tested, BaΔwadC LPS was capable of binding to MD2, although at concentrations higher than the reference endotoxic Salmonella LPS (Figure 6A) or E. coli LPS (not shown). These results were confirmed by testing the displacement of bis-ANS from MD-2. Whereas Ba-parental LPS did not cause displacement in the range of the concentrations tested (1.25–10 µg/mL), BaΔwadC displaced this probe from MD-2. Consistent with the results of the competitive ELISA, the interaction of the mutated LPS with MD-2 observed using this protocol did not reach the Salmonella LPS levels (Figure S8). Interestingly, the LPS of Ochrobactrum anthropi (which carries a Brucella-type lipid A but differs markedly in the core structure [8], [14]) showed a higher binding to MD2 than either the mutant or the wild-type B. abortus LPS (Figure S9). Moreover, the percentage of BMDC showing NF-κB translocation to the nucleus (Figure S10) after 1 h of stimulation with BaΔwadC LPS was clearly above that of Ba-parental LPS (Figure 6B). Clearly, both sets of results are in agreement with the cytokine profiles observed in mice and target cells. In addition, since activation of the mTOR pathway has been implicated in dendritic cell maturation and cytokine production, we determined the phosphorylation of S6, one of the downstream elements of this pathway [21]. When BaΔwadC LPS was compared to Ba-parental LPS, the former induced an earlier, more intense and more sustained S6 phosphorylation (Figure 6C) as detected from 30 min up to 6 h of LPS stimulation.
All this evidence suggests that the core of B. abortus LPS negatively modulates recognition by MD-2/TLR4 and the subsequent intracellular signaling leading to pro-inflammatory response and dendritic cell maturation. In support of this interpretation, the TLR4 dysfunction did not affect the multiplication of the wild type but allowed a better replication of the mutant in the spleens of mice (Figure S11). The replication of the wadC mutant in TLR4−/− did not reach the levels of the wild type, consistent with the existence of additional factors causing attenuation such as the complement and polycation sensitivity observed in vitro.
We have identified a B. abortus LPS gene (wadC) whose disruption does not result in the loss of the O-PS but in an altered core, which is in contrast to all Brucella LPS genes described [10], [22], [23], [24]. This allowed us to discriminate the role of the LPS core oligosaccharide from that of the O-polysaccharide in Brucella virulence. The wadC LPS core mutant induced a strong proinflammatory response and was attenuated in mice and in dendritic cells. These properties were reproduced by the purified wadC mutant LPS but not by the wild type Ba-parental LPS. Activation was TLR4-dependent and the core-mutated LPS displayed increased binding to MD-2, the TLR4 co-receptor, which paralleled an increased intracellular signaling. This is the first description of an LPS core hampering complement activation, cationic peptide binding and detection by TLR4-MD-2. Therefore, B. abortus LPS core acts as a shield against innate immunity recognition and represents a novel virulence-related mechanism. Studies in progress with the B. melitensis, B. suis and B. ovis mutants in the wadC orthologues confirm that this is a general mechanism of this group of bacteria. Indeed, all our results also provide experimental proof that the stealthy behavior of this pathogen towards innate immunity is essential for its virulence, as proposed before [2], [8].
Consistent with the key role of dendritic cells in brucellosis [25], [26], [27], the attenuation and proinflammatory responses of the wadC mutant in mice were reproduced in BMDC. It is noteworthy that the attenuation was not observed in macrophages, and this could be attributed to functional differences between these two types of cells in the ability to process and present antigens. Nevertheless, the results obtained in mice showed that the observations made in dendritic cells are more relevant for the course of the infection than those in macrophages in vitro. Interference with dendritic cell maturation is a strategy that prevents the development of efficient immunity and there are several examples of intracellular pathogens that target dendritic cells. Mycobacterium tuberculosis interferes with TLR signaling in dendritic cells blocking their maturation and IL-12 generation and directing the immunoresponse towards IL-10 production [28]. S. Typhimurium, on the other hand, does not block maturation but prevents MHC II antigen presentation in some subtypes of dendritic cells [29], [30]. Francisella tularensis, another gram-negative pathogen that does not block dendritic cell maturation, seems to be capable of multiplying in these cells and to inhibit the secretion of pro-inflammatory cytokines [31]. However, only B. abortus, B. melitensis and B. suis, the three classical smooth Brucella spp., have been reported to be simultaneously able to multiply in dendritic cells and to prevent their maturation, thereby thwarting the efficient presentation of proteins in either the MHC I or MHC II context [27]. The properties that enable the brucellae to display these abilities are beginning to be understood. TLR2, TLR4 and TLR9 seem to be the most important TLRs in dendritic cells, and they recognize lipoproteins, LPS and CpG DNA, respectively. Salcedo et al. [27] have recently shown that protein Btp1 of B. abortus interferes with the TLR2 signaling pathway and down modulates dendritic cell activation. Billard et al. [26] found that the response of dendritic cells derived from peripheral blood monocytes to wild type B. abortus LPS was low as compared to that obtained with E. coli LPS. Our results extend these observations and demonstrate the connection between the LPS core oligosaccharide and the ability of B. abortus to thrive in dendritic cells and to circumvent their maturation.
Chemical analyses of the B. abortus LPS core reveal the presence of Kdo, glucosamine, glucose, mannose and quinovosamine [6]. Our results suggest that these sugars must be arranged in such a way that some of them probably play no role in the section linking the O-PS, as predicted before by studies with monoclonal antibodies [23] and on the inability of polymyxin B to neutralize the charge of B. abortus LPS [14]. Moreover, we have identified a second glycosyltransferase gene (wadB) that, upon disruption, generates an LPS phenotype close to the one described here for BaΔwadC. These genetic data indicate that at least two core sugars must be arranged in a structure whose damage does not affect the link to the O-PS. Research in progress shows the existence of a mannose containing branch in the Brucella LPS core. WadC is a putative mannosyltransferase and it seems likely that it could be involved in the transfer of a mannose unit to such a branch. It is tempting to speculate that such a structure hinders the access to the Kdo and lipid A phosphates targeted by bactericidal peptides as well as complement C1q [19], and could thus account for the results of the transition measurements performed in the presence of PMB and serum. We also propose that the full core structure is one of the factors contributing to a defective MD-2 recognition, the other one being the peculiar acyl chain composition of Brucella lipid A. Two not mutually exclusive hypotheses could account for the role of the complete core. Its absence in the mutant could favor the dissociation of individual molecules from aggregates and make them more readily available for binding to MD2, and the anionic molecules and Kdo in the inner sections could be more accessible for binding to MD2. In addition, it is known that C12-C14 hexaacylated lipids A like that of E. coli interact with a large hydrophobic groove in MD-2, with five acyl chains deep inside, the remaining chain in a hydrophobic interaction with TLR4 and the bisphosphorylated glucosamine disaccharide tilted outwards [7]. Therefore, the lipid A phosphate groups contribute to receptor multimerization by interacting with positively charged residues in TLR4 and MD-2 [7]. Previously, we proposed that the very long chain fatty acids in Brucella lipid A are critical in preventing effective recognition by TLR4-MD-2 [32] and here we show that the interaction of both the wild type and the BaΔwadC LPS with MD-2 is significantly reduced as compared to that of Salmonella LPS. This suggests that the MD-2 hydrophobic pocket does not allow for efficient interaction with the bulky lipid A of Brucella. Moreover, removal of part of the B. abortus LPS core increases the binding to MD-2 indicating that interaction is hampered by virulent B. abortus intact core. Since O. anthropi and B. abortus LPS have similar a lipid A but a markedly different core structure [8], [14], the results obtained with the former in the MD2 assay also suggest that core structures are important. The work in progress on the structure of the B. abortus LPS core shows that the O-PS stems from a few sugars linked to Kdo I, suggesting Kdo II as the section linked to the sugar(s) removed by the wadC mutation. Thus, these sugars should be close to the negatively charged groups in the lipid A backbone and Kdo and could hinder cationic peptide, C1q and MD-2 interactions. This structure may not be unique to B. abortus. In fact, wadC homologues are found not only in the genomes of all Brucella species but also in Bartonella spp., suggesting structures acting as shields against innate immunity recognition.
Finally, current classical smooth brucellosis vaccines (B. abortus S19 and B. melitensis Rev 1) are doubtlessly useful tools in animal vaccination and, therefore, in the eradication of this zoonotic disease, but do not afford 100% protection in the natural host. Hence, their successful use requires complementary measures (tagging and control of animal movement, efficient veterinary services, repeated animal testing) that make brucellosis eradication a cumbersome and long process [33]. We found that the cytokine profile, with IL-12 being released in large amounts, the transitory splenomegaly and the eventual clearance of BaΔwadC could make mutation of wadC a tool to improve existing or future vaccines.
Animal experimentation was conducted in strict accordance with good animal practice as defined by the French animal welfare bodies (Law 87–848 dated 19 October 1987 modified by Decree 2001-464 and Decree 2001-131 relative to European Convention, EEC Directive 86/609). All animal work was approved by the Direction Départmentale des Services Vétérinaires des Bouches du Rhônes (authorization number 13.118).
All animals were handled and sacrificed according to the approval and guidelines established by the “Comité Institucional para el Cuido y Uso de los Animales” of the Universidad de Costa Rica, and in agreement with the corresponding law “Ley de Bienestar de los Animales No 7451” of Costa Rica (http://www.micit.go.cr/index.php/docman/doc_details/101-ley-no-7451-leyde-bienestar-de-los-animales.html).
Mice (Charles River, Elbeuf, France) were accommodated in the animal building of the CITA of Aragón (ID registration number ES-502970012005) in cages with water and food ad libitum and under biosafety containment conditions, for 2 weeks before the start and all along the experiment. The animal handling and procedures were in accordance with the current European legislation (directive 86/609/EEC) supervised by the Animal Welfare Committee of the institution (protocol numberR102/2007).
The bacterial strains and plasmids used are listed in Table S1. Moreover, the strain Salmonella enterica subespecies enterica serotype typhimurium (abbreviated as S. Typhimurium) reference ATCC SL1344, Salmonella abortus equi strain HL83 and E. coli strain MG1655 were used as controls in some assays. Bacteria were routinely grown in standard tryptic soy broth or agar either plain or supplemented with kanamycin at 50 µg/mL, or/and nalidixic acid at 25 µg/mL, or/and 5% sucrose. All strains were stored in skim milk at −80°C.
Plasmid and chromosomal DNA were extracted with Qiaprep spin Miniprep (Qiagen GmbH, Hilden, Germany), and Ultraclean Microbial DNA Isolation kit (Mo Bio Laboratories) respectively. When needed, DNA was purified from agarose gels using Qiack Gel extraction kit (Qiagen) and sequenced by the Servicio de Secuenciación de CIMA (Centro de Investigación Médica Aplicada, Pamplona, Spain). Primers were synthesized by Sigma-Genosys Ltd. (Haverhill, United Kingdom). Searches for DNA and protein homologies were carried out using the NCBI (National Center for Biotechnology Information; http://www.ncbi.nlm.nih.gov) and the EMBL-European Bioinformatics Institute server (http://www.ebi.ac.UK/ebi_home.html). In addition, sequence data were obtained from The Institute for Genomic Research website at http://www.tigr.org. Genomic sequences of B. melitensis, B. abortus and B. suis were analyzed using the database of the URBM bioinformatic group (http://urbm-cluster.urbm.fundp.ac.be/~apage).
In-frame deletion mutant BaΔwadC was constructed by PCR overlap using genomic DNA of Ba-parental as DNA template (Figure S1). Primers were designed based on the available sequence of the corresponding genes in B. abortus 2308. For the construction of the wadC mutant, we first generated two PCR fragments: oligonucleotides wadC-F1 (5′-CTGGCGTCAGCAATCAGAG-3′) and wadC-R2 (5′- GTGCAACGACCTCAACTTCC-3′) were used to amplified a 476-bp fragment including codons 1 to 16 of the wadC ORF, as well as 424 bp upstream of the wadC start codon, and oligonucleotides wadC-F3 (5′-GGAAGTTGAGGTCGTTGCACACGCCATC GAACCTTATCTG-3′) and wadC-R4 (5′-CGGCTATCGTGCGATTCT-3′) were used to amplify a 453-bp fragment including codons 308 to 354 of the wadC ORF and 320-bp downstream of the wadC stop codon (see S-1). Both fragments were ligated by overlapping PCR using oligonucleotides wadC-F1 and wadC-R4 for amplification, and the complementary regions between wadC-R2 and wadC-F3 for overlapping. The resulting fragment, containing the wadC deletion allele, was cloned into pCR2.1 (Invitrogen), to generate plasmid pRCI-23, sequenced to ensure the maintenance of the reading frame, and subsequently subcloned into the BamHI and the XbaI sites of the suicide plasmid pJQK. The resulting mutator plasmid (pRCI-26) was introduced in Ba-parental by conjugation. The first recombination (integration of the suicide vector in the chromosome) was selected by Nal and Kan resistance, and the second recombination (excision of the mutator plasmid leading to construction of the mutant by allelic exchange), was selected by Nal and sucrose resistance and Kan sensitivity. The resulting colonies were screened by PCR with primers wadC-F1 and wadC-R4 which amplify a fragment of 929 bp in the mutant and a fragment of 1805 bp in the parental strain. The mutation generated results in the loss of the 82% of the wadC ORF, and the mutant strain was called BaΔwadC.
Taking into account that the WadC sequences of B. melitensis and B. abortus are identical, we used the B. melitensis ORFeome constructed with the Gateway cloning Technology (Invitrogen) for complementation [34]. The clone carrying B. melitensis wadC was extracted, and the DNA containing the corresponding ORF was subcloned in pRH001 [35] to produce plasmid pwadC. To complement the wadC mutation, plasmid pwadC was introduced into the BaΔwadC mutant by mating with E. coli S17-1 and the conjugants harboring pwadC (designated as BaΔwadC-comp) were selected by plating the mating mixture onto TSA-Nal-Kan plates which were incubated at 37°C for 3 days.
Seven-week-old female BALB/c mice (Charles River, Elbeuf, France) were kept in cages with water and food ad libitum and accommodated under biosafety containment conditions 2 weeks before the start of the experiments. Inocula were prepared in sterile 10 mM PBS (pH 6.85). For each strain, 30 mice were inoculated intraperitoneally with 0.1 mL of inoculum containing 5.8×104 (Ba-parental) or 4.9×104 (BaΔwadC) CFU/mouse and the number of CFU in spleens (n = 5) was determined at 1, 2, 4, 6, 8, and 12 weeks after inoculation. BaTn5::per was used as control of representative R mutant with complete LPS-core, at inoculation dose of 1×108 CFU intraperitoneally and viable counts in spleens at 1, 2, 3, 6 and 9 weeks post-infection. An additional experiment was performed under the same conditions but including BaΔwadC-comp and the number of CFU in spleens was determined 8 weeks after inoculation. The identity of the spleen isolates was confirmed by PCR at each time-point during the infection process. The individual data were normalized by logarithmic transformation, and the mean and standard deviation (SD) of log10 CFU/spleen were calculated.
BALB/c mice from 20 and 24 g were intraperitoneally injected with 106 CFU of Ba-parental, 105 of S. Typhimurium or 0.1 mL pyrogen-free sterile PBS. Blood was collected from the retro-orbital sinus and subjected to analysis. Alternatively, 5 mL of ice cold PBS were injected in the peritoneal cavity of killed the mice, and the fluids collected with a syringe (from 3.8 to 4.5 mL) from exposed peritoneal cavity. Then fluids were centrifuged and the peritoneal cells resuspended in 0.2 mL of PBS and counted in Neubauer chambers. Giemsa-Wright staining smears were performed to distinguish between leukocytes.
Bone marrow cells were isolated from femurs of 7–8-week-old C57Bl/6 female, TLR2−/−, TLR4−/− or TLR9−/− [36], [37] mice and differentiated into either dendritic cells (BMDCs) or macrophages (BMDMs) as described previously [38], [39] in presence of decomplemented fetal bovine serum. Infections were performed by centrifuging the bacteria onto the differentiated cells (400×g for 10 min at 4°C; bacteria: cells ratio of 20∶1 for BMDCs or 50∶1 for BMDMs) followed by incubation at 37°C for either 15 min (BMDMs) or 30 min (BMDCs) under a 5% CO2 atmosphere. Cells were either extensively washed (BMDMs) or gently washed (BMDCs) with medium to remove extracellular bacteria and incubated in medium supplemented with 100 µg/mL gentamicin for 1 h to kill extracellular bacteria. Thereafter, the antibiotic concentration was decreased to 20 µg/mL. To monitor Brucella intracellular survival, infected cells were lysed with 0.1% (vol/vol) Triton X-100 in H2O (BMDCs) or after PBS washing (BMDMs) and serial dilutions of lysates were rapidly plated onto tryptic soy agar plates to enumerate CFU.
The levels of TNF- α, IL-6, IL-10 and IL-12 p40/p70 were estimated at different time points by enzyme-linked immunosorbent assays (ELISA) in the sera of BALB/c mice infected intraperitoneally, and in the supernatants of BMDC or BMDM at 24 hours after infection (see above) or after stimulation with 10 µg/mL of the appropriate LPS from different Brucella strains or 100 ng/mL from E. coli ATCC 35218 obtained by the phenol-water procedure and purified further by the phenol-water-deoxycholate method [40]. For the latter purpose, a stock of 1 mg/mL in pyrogen free sterile water was prepared, sonicated briefly and sterilized by autoclaving. Prior to use, the stock was heated at 56°C for 15 min and then cooled to room temperature.
BMDCs were grown on glass coverslips and inoculated with bacteria as described above or stimulated with the appropriate LPS. At different times after inoculation (see Results), coverslips were fixed with 3% paraformaldehyde pH 7.4 at 37°C for 15 min and washed three times with PBS. Coverslips were processed for immunofluorescence staining as previously described [39]. Briefly, cells were permeabilized with 0.1% saponin and incubated with primary antibodies. After several washes, the primary antibodies were revealed with the appropriate secondary antibodies. The primary antibodies used for immunofluorescence microscopy were: cow anti-B. abortus; rat anti-mouse LAMP1 ID4B (Developmental Studies Hybridoma Bank, National Institute of Child Health and Human Development, University of Iowa); mouse anti FK2 (Biomol); Moab anti-calnexin (kindly provided by Dr. D. Williams, University of Toronto) and NF-κB subunit p65/RelA (Santa Cruz). In all experiments, BMDCs were labeled using an antibody against a conserved cytoplasmic epitope found on MHC-II I-A ß subunits or MHC II [29] which does not produce significant labeling with BMDMs. In addition, BMDCs were labeled with an anti–CD11c antibody (Biolegend) confirming that they are dendritic cells [27]. Samples were analyzed under a Leica DMRBE epifluorescence microscope for quantitative analysis, or a Zeiss LSM 510 laser scanning confocal microscope for image acquisition. Images were then assembled using Adobe Photoshop 7.0. Quantifications were done by counting at least 300 cells in 3 independent experiments.
30 µg of cell lysates were subjected to SDS-PAGE and, after transfer to nitrocellulose, the membrane was probed with a polyclonal antibody against phospho-S6 (Cell Signaling Technology) that detects phosphorylation on Ser235/236 or anti-actin antibody. Blots were subjected to enhanced chemiluminescence detection (ECL, PIERCE).
BMDCs treated with different types of LPS's were collected and stained immediately before fixation with paraformaldehyde. Isotype controls were included as well as BMDCs treated with the different secondary antibodies for control of autofluorescence. Cells were always gated on CD11c and a minimum of 12,000 CD11c-positive events were obtained for analysis. A FACScalibur cytometer (Beckton Dickinson) was used and data were analysed using FlowJo software (Tree Star). Allophycocyanin (APC) conjugated-anti-CD11c antibody (HL3) from Pharmigen was used in all experiments along with either phycoerythrin (PE) conjugated anti-CD86, anti-IA-IE (MHC class II) or fluorescein isothiocyanate (FITC) conjugated anti-CD40, all from BioLegend.
Extraction of whole-cell LPS by SDS-proteinase K protocol was performed as described previously [41]. In addition, LPS was obtained by methanol precipitation of the phenol phase of a phenol-water extract [13]. This fraction (10 mg/mL in 175 mM NaCl, 0.05% NaN3, 0.1 M Tris-HCl [pH 7.0]) was then purified by digestion with nucleases (50 µg/mL each of DNase-II type V, and RNase-A [Sigma, St. Louis, Missouri, U.S.A.], 30 min at 37°C) and three times with proteinase K (50 µg/mL, 3 hours at 55°C), and ultracentrifuged (6 h, 100,000× g). When we applied this method to BaΔwadC, we found that the supernatant of the ultracentrifugation step contained the major fraction of the LPS. The studies were performed with the major LPS fractions of each bacteria. Free lipids (ornithine lipids and phospholipids) were then removed by a fourfold extraction with chloroform-methanol. (2∶1 [vol/vol]) [14].
LPS were analyzed in 15 or 18% polyacrylamide gels (37.5∶1 acrylamide/methylene-bisacrylamide ratio) in Tris-HCl-glycine and stained by the periodate-alkaline silver method [42]. For Western blots, gels were electrotransferred onto nitrocellulose sheets (Schleicher & Schuell GmbH, Dassel, Germany), blocked with 3% skim milk in PBS with 0.05% Tween 20 overnight, and washed with PBS–0.05% Tween 20. Immune sera were diluted in this same buffer and, after incubation overnight at room temperature, the membranes were washed again. Bound immunoglobulins were detected with peroxidase-conjugated goat anti-mouse immunoglobulin (Nordic,Tilburg, Netherlands) and 4-chloro-1-naphthol- H2O2. Monoclonal antibodies (Moabs) used in this study were Cby-33H8 (Ingenasa, Madrid, Spain), which recognizes the C/Y O-chain epitope, and A68/24D08/G09, A68/24G12/A08, and A68/3F03/D5 which recognize core epitopes [43]. The inner core LPS marker Kdo was determined colorimetrically by the thiobarbituric acid method using pure Kdo and deoxyribose as the standards, with the modifications described previously [44]. Kdo contents were 1.6 and 2.4% for the wild type and the mutated LPS, respectively.
Lipid A fractions were extracted using an ammonium hydroxide/isobutyric acid method and subjected to negative ion matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectrometry analysis [45], [46]. Briefly, lyophilized crude cells (20 mg) were resuspended in 400 µL isobutyric acid/1M ammonium hydroxide (5∶3, v/v) and were incubated in a screw-cap test tube at 100°C for 2 h, with occasional vortexing. Samples were cooled in ice water and centrifuged (2,000×g for 15 min). The supernatant was transferred to a new tube, diluted with an equal volume of water, and lyophilized. The sample was then washed twice with 400 µL methanol and centrifuged (2,000×g for 15 min). The insoluble lipid A was solubilized in 100 µL chloroform/methanol/water (3∶1.5∶0.25, v/v/v). Analyses were performed on a Bruker Autoflex II MALDI-TOF mass spectrometer (Bruker Daltonics, Incorporated) in negative reflective mode with delayed extraction. Each spectrum was an average of 300 shots. The ion-accelerating voltage was set at 20 kV. Dihydroxybenzoic acid (Sigma Chemical Co., St.Louis, MO) was used as a matrix. A few microliters of lipid A suspension (1 mg/mL) was desalted with a few grains of ion-exchange resin (Dowex 50W-X8; H+) in an Eppendorf tube. A 1 µL aliquot of the suspension (50–100 µL) was deposited on the target and covered with the same amount of the matrix suspended at 10 mg/mL in a 0.1 M solution of citric acid. Different ratios between the samples and dihydroxybenzoic acid were used when necessary. A peptide calibration standard (Bruker Daltonics) was used to calibrate the MALDI-TOF. Further calibration for lipid A analysis was performed externally using lipid A extracted from E. coli strain MG1655 grown in LB at 37°C.
Binding of LPS to MD-2 was assayed by two different methods: binding to hMD-2 by competitive ELISA and displacement of bis-ANS (4,4′-Dianilino-1,1′-binaphthyl-5,5′-disulfonic acid dipotassium salt) from MD-2 by LPS. In both assays, LPS was sonicated before use and subjected to three cycles of heating at 56°C followed by cooling to 4°C.
The ELISA for determination of LPS binding to hMD-2 was performed in 96-well plates (NUNC immunoplate F96 cert. Maxi-sorp, Roskilde, Denmark).Chicken anti-hMD-2 (GenTel, Madison, WI, U.S.A) (5 µg/mL) in 50 mM Na2CO3 (pH 9.6) was used to coat the microtiter plate at 4°C overnight. Excess binding sites were blocked with 1% BSA in PBS buffer (pH 7.2) for 1 h at room temperature, and rinsed three times with the same buffer. During the blocking step, hMD-2 (0.75 µM) was preincubated with 0 µM to 8 µM LPS at 37°C and, as a negative control, LPS was also preincubated in absence of hMD-2. This preincubated solutions were added to the plate, which was then incubated for 1 h at 37°C. After rinsing, hMD-2 not bound to LPS was detected by incubation with 0.1 µg/mL of mouse anti h-MD-2 (clone 9B4 e-Bioscience San Diego, CA., U.S.A.) in PBS buffer at 37°C for 1 h, followed by incubation with 0.1 µg/mL peroxidase-conjugated goat anti-mouse IgG (Santa Cruz, CA., U.S.A.), also in PBS buffer at 37°C for 1 h. After plate washing, ABTS (Sigma) was added, the reaction was stopped with 1% SDS after 15 min, and the absorbance at 420 nm measured using a Mithras LB940 apparatus (Berthold Technologies). Salmonella abortus equi (strainHL83) LPS, used as a control, was prepared by a phenol extraction procedure and was kindly provided by Dr. Brandenburg (Forschungszentrum Borstel, Germany).
Since the binding site of bis-ANS overlaps with the MD-2 binding site of LPS we measured the displacement of the probe by the LPS of Ba-parental and BaΔwadC using Salmonella LPS as a control [47].Binding of bis-ANS to hMD-2 was measured at 20°C using excitation at 385 nm and measuring the emission fluorescence spectra between 420 and 550 nm. Then, increasing amounts of LPS were added to preincubated bis-ANS/hMD-2 complex (200 nM/200 nM). The F0 value was the fluorescence intensity of bis-ANS/hMD-2 complexes at 490 nm after 30 min of incubation (to reach stable fluorescence).The FLPS value was the fluorescence at 490 nm after LPS addition to the complex. Fluorescence was measured on Perkin Elmer fluorimeter LS 55. Quartz glass cuvette (5×5 mm optical path, Hellma Suprasil) was used and bis-ANS was obtained from Sigma (St. Louis, Missouri, U.S.A.).
The transition of the acyl chains of LPS from a well-ordered state (gel phase) to a fluid state (liquid crystalline phase) at a lipid-specific temperature (Tc) was determined by Fourier transform infrared spectroscopy. A specific vibrational band, the symmetric stretching vibration of the methylene groups νs(CH2) around 2,850 cm−1, was analyzed since its peak position is a measure of the state of order (fluidity) of the acyl chains. Measurements were performed in a Bruker IFS 55 apparatus (Bruker, Karlsruhe, Germany) as described previously [48]. To ensure homogeneity, LPS suspensions were prepared in 2.5 mM HEPES (pH 7.2) at room temperature, incubated at 56°C for 15 min, stirred, and cooled to 4°C. This heating/cooling step was repeated three times, and the suspensions were stored at 4°C for several hours before analysis. LPS suspensions (water content, 90%) were analyzed in CaF2 cuvettes with 12.5-µm Teflon spacers, and for each measurement, 50 interferograms were accumulated, Fourier transformed, and converted to absorbance spectra. The measurements were obtained in continuous heating scans (2°C/min) between 10°C and 60°C. To test the effect of complement, the experiments were performed in the presence of normal human serum. The effect of polymyxin B was assessed similarly at different LPS∶polymyxin B molar ratios (see Results), and using an average MW of 11800 for Ba-2308 LPS (determined by SDS-PAGE with Yersinia enterocolitica O:8 LPS as a standard).
The minimal inhibitory concentrations (MIC) of polymyxin B, poly-L-ornithine, poly-L-lysine, colistin, penicillin, doxycycline, clarithromycin, erythromycin, rifampicin, basic fuchsin, safranine and thionin was determined in Müller-Hinton medium by standard procedures. Sensitivity to the S (Tb, Wb, Iz) and rough (R/C) -specific brucellaphages was measured by testing the lysis of bacteria exposed to serial 10-fold dilutions made from a routine test dilution phage stock [49].
Exponentially growing bacteria were adjusted to 104 CFU/mL in saline and dispensed in triplicate in microtiter plates (45 µL per well) containing fresh normal bovine serum (90 µL/well). After 90 min of incubation at 37°C, brain heart infusion broth was dispensed (200 µL/well), mixed with the bacterial suspension and 100 µL was plated on TSA. Results were expressed as the percentage of the average CFU with respect to the inoculum.
The Kolmogorov-Smirnov test was applied to assess the normal distribution of data obtained in each experiment. Thereafter, means were statistically compared by an unpaired Student's t test. Kruskal-Wallis and Mann-Whitney tests were used for experiments with non-normal data distribution. The StatviewGraphics 5.0 for Windows (SAS Institute Inc) statistical package was used in all cases.
|
10.1371/journal.pgen.1003291 | miR-199a-5p Is Upregulated during Fibrogenic Response to Tissue Injury and Mediates TGFbeta-Induced Lung Fibroblast Activation by Targeting Caveolin-1 | As miRNAs are associated with normal cellular processes, deregulation of miRNAs is thought to play a causative role in many complex diseases. Nevertheless, the precise contribution of miRNAs in fibrotic lung diseases, especially the idiopathic form (IPF), remains poorly understood. Given the poor response rate of IPF patients to current therapy, new insights into the pathogenic mechanisms controlling lung fibroblasts activation, the key cell type driving the fibrogenic process, are essential to develop new therapeutic strategies for this devastating disease. To identify miRNAs with potential roles in lung fibrogenesis, we performed a genome-wide assessment of miRNA expression in lungs from two different mouse strains known for their distinct susceptibility to develop lung fibrosis after bleomycin exposure. This led to the identification of miR-199a-5p as the best miRNA candidate associated with bleomycin response. Importantly, miR-199a-5p pulmonary expression was also significantly increased in IPF patients (94 IPF versus 83 controls). In particular, levels of miR-199a-5p were selectively increased in myofibroblasts from injured mouse lungs and fibroblastic foci, a histologic feature associated with IPF. Therefore, miR-199a-5p profibrotic effects were further investigated in cultured lung fibroblasts: miR-199a-5p expression was induced upon TGFβ exposure, and ectopic expression of miR-199a-5p was sufficient to promote the pathogenic activation of pulmonary fibroblasts including proliferation, migration, invasion, and differentiation into myofibroblasts. In addition, we demonstrated that miR-199a-5p is a key effector of TGFβ signaling in lung fibroblasts by regulating CAV1, a critical mediator of pulmonary fibrosis. Remarkably, aberrant expression of miR-199a-5p was also found in unilateral ureteral obstruction mouse model of kidney fibrosis, as well as in both bile duct ligation and CCl4-induced mouse models of liver fibrosis, suggesting that dysregulation of miR-199a-5p represents a general mechanism contributing to the fibrotic process. MiR-199a-5p thus behaves as a major regulator of tissue fibrosis with therapeutic potency to treat fibroproliferative diseases.
| Fibrosis is the final common pathway in virtually all forms of chronic organ failure, including lung, liver, and kidney, and is a leading cause of morbidity and mortality worldwide. Fibrosis results from the excessive activity of fibroblasts, in particular a differentiated form known as myofibroblast that is responsible for the excessive and persistent accumulation of scar tissue and ultimately organ failure. Idiopathic Lung Fibrosis (IPF) is a chronic and often rapidly fatal pulmonary disorder of unknown origin characterized by fibrosis of the supporting framework (interstitium) of the lungs. Given the poor prognosis of IPF patients, new insights into the biology of (myo)fibroblasts is of major interest to develop new therapeutics aimed at reducing (myo)fibroblast activity to slow or even reverse disease progression, thereby preserving organ function and prolonging life. MicroRNAs (miRNAs), a class of non-coding RNA recently identified, are associated with normal cellular processes; and deregulation of miRNAs plays a causative role in a vast array of complex diseases. In this study, we identified a particular miRNA: miR-199a-5p that governs lung fibroblast activation and ultimately lung fibrosis. Overall we showed that miR-199a-5p is a major regulator of fibrosis with strong therapeutic potency to treat fibroproliferative diseases such as IPF.
| Tissue fibrosis, defined as the excessive and persistent formation of non functional scar tissue in response to repeated injury and insult, is a leading cause of morbidity and mortality associated with organ failure in various chronic diseases such as those affecting the lung interstitium [1]. Among the interstitial lung diseases of unknown etiology, Idiopathic Pulmonary Fibrosis (IPF) is the most common and lethal with a median survival of 3 to 5 years after diagnosis [2]. The pathogenesis of IPF is complex and largely unknown [2], but observations based on both animal models of pulmonary fibrosis and lung sections from patients with IPF suggest a dynamic pathobiological process involving excessive wound healing with chronic inflammation, apoptosis of epithelial and endothelial cells, mesenchymal cell proliferation and activation with the formation of fibroblasts/myofibroblasts foci, and finally excessive deposition of extracellular matrix resulting in the destruction of the lung architecture and the loss of lung functions [2]. In particular, myofibroblasts play a substantial role in IPF by secreting important amount of ECM components and by promoting lung tissue stiffening [3]. Given the poor response rate of IPF patients to current therapy, a detailed understanding of the underlying pathogenic mechanisms is of major interest to develop new effective therapeutic strategies targeting the cellular and molecular events involved in the fibrotic response.
MicroRNAs (miRNAs) are a class of noncoding small RNA, which most often bind to the 3′ UTR of target genes mRNAs and thereby repress their translation and/or induce their degradation. Since the first miRNA identification in Caenorhabditis elegans in a context of larval development [4], [5], thousands miRNAs have now been characterized including about 2000 in human (miRbase v19) [6]. MiRNAs are now recognized as major regulators of gene expression with crucial functions in numerous biological processes including development, proliferation, differentiation, apoptosis and stress response. Importantly, recent studies have identified specific miRNA expression patterns related to the initiation and progression of various diseases including cancer as well as inflammatory, infectious and autoimmune diseases [7]–[9]. Additionally, gain and loss of function miRNA studies have further established their functional impact in various in vivo models [10]–[15]. Nevertheless, the precise contribution of miRNAs in fibrotic diseases, especially lung fibrosis, is still poorly understood [16], [17]. Our rationale was therefore to test whether miRNAs may provide new perspectives on disease mechanisms, diagnosis as well as new therapeutic opportunities in the specific context of fibrosis.
In an effort to identify miRNAs with potential roles in the development of lung fibrosis (strategy detailed in Figure S1), we aimed to identify miRNAs of interest in two mouse strains showing different susceptibility to develop lung fibrosis after bleomycin exposure. This led to the identification of a panel of miRNAs specifically dysregulated in the lungs of fibrosis prone mouse strain in response to bleomycin. Among these miRNAs, miR-199a-5p was found to be selectively up-regulated in myofibroblasts of the injured lung in bleomycin-treated mice and fibroblastic foci of IPF patients. In lung fibroblasts, miR-199a-5p acts as an effector of TGFβ signaling, regulates CAV1 expression, a critical mediator of the lung fibrosis process [18]–[21] and participates to multiple fibrogenic associated-processes including cell proliferation, migration, invasion and differentiation into myofibroblasts. Finally, dysregulation of miR-199a-5p was also found in two other mouse models of tissue fibrosis, namely kidney fibrosis and liver fibrosis, suggesting therefore that miR-199a-5p is likely to be a common mediator of fibrosis.
Previous studies based on mice have demonstrated a genetic susceptibility to bleomycin-induced pulmonary fibrosis [22], [23]. Indeed, C57BL/6 mice are considered to be fibrosis prone, whereas BALB/C mice are less prone to fibrosis. To identify miRNAs that may contribute to the lung fibrosis process, miRNA expression profile in response to bleomycin was assessed 7 days and 14 days following bleomycin administration (i.e. when active fibrogenesis occurs) on both strains using a microarray based platform (Data set 1, GEO accession number GSE34812) described elsewhere [24]–[26]. We identified 22 differentially expressed miRNAs between lungs from bleomycin- and control-treated animals in at least one strain, the majority being upregulated in bleomycin-instillated lungs (Figure 1A). We focused our analysis on miRNAs that exhibited an enhanced expression in response to bleomycin during disease progression in the C57BL/6 sensitive mice only. Among several miRNAs candidates with such a profile, miR-199a-5p displayed the highest statistical score (Figure 1B). This was further established using an independent set of mice at day 14 following bleomycin treatment (Figure S2A). These findings strongly suggested that miR-199a-5p may play an important role during the lung fibrosis process. To investigate the regulatory mechanisms underlying miR-199a-5p production, we assessed the expression status of the 2 mouse genes, miR-199a-1 (on chromosome 9) and miR-199a-2 (on chromosome 1) in response to bleomycin using a Taqman assay designed to discriminate between pri-miR-199a-1 and pri-miR-199a-2. Our results showed that, 14 days after bleomycin instillation, both pri-miR-199a transcripts were up-regulated in the lungs of C57BL/6 mice (Figure S2B) and thus, contributed to miR-199a-5p production. In addition, in situ hybridization experiments performed in the injured lungs from C57BL/6 mice 14 days after bleomycin instillation revealed a selective expression of miR-199a-5p in myofibroblasts (Figure 1C).
Of note, consistent with previous findings [14], we also found a significant upregulation of miR-21 (now referenced in miRbase as mmu-miR-21a-5p) in response to bleomycin (Figure 1A and Figure S3). Nevertheless, miR-21 induction did not differ between bleomycin-sensitive and bleomycin-resistant strains of mice.
We next sought to determine the mechanism by which miR-199a-5p dysregulation may lead to tissue fibrosis. To address this question, we first attempted to identify gene targets and cellular pathways regulated by miR-199a-5p using the methodology described earlier [25], [26]. The influence of miR-199a-5p on human pulmonary hFL1 fibroblast transcriptome was compared with that of miR-21, which has been previously associated with the development of fibrotic diseases including lung fibrosis [14], [15], [27] (Data set 2, GEO accession number GSE34815). Forty-eight hours after ectopic overexpression of each miRNA, a significant alteration (defined by an absolute log2ratio above 0.7 and an adjusted p-value below 0.05) of 1261 and 753 transcripts was detected in the miR-199a-5p and miR-21 conditions, respectively. While these 2 miRNAs induced very distinct gene expression patterns (Figure 2A), a functional annotation of these signatures, using Ingenuity Pathway software, indicated an overlap for “canonical pathways” including “Cell Cycle regulation” and “TGFβ Signaling” (Table S1). Consistent with previous findings [28], highly significant pathways associated with miR-21 were related to “Cyclins and Cell Cycle Regulation” as well as “Cell Cycle Control of Chromosomal Replication”, “Mismatch Repair in Eukaryotes” and “ATM signalling”. While the highest scoring pathway for miR-199a-5p corresponded to the metabolic pathways “Biosynthesis of Steroids”, we also noticed enrichment for pathways related to “Integrin Signaling” and “Caveolar-mediated Endocytosis Signaling”. We next looked for an enrichment of putative direct targets in the population of down-regulated transcripts, as described in [29]. A specific overrepresentation of predicted targets for miR-199a-5p and miR-21 in the population of down-regulated transcripts was noticed after heterologous expression of either miR-199a-5p or miR-21, respectively. This enrichment was independent of the prediction tool used to define the targets (Figure 2B and not shown). We then focussed our analysis on a subset of 21 transcripts containing miR-199a-5p complementary hexamers in their 3′UTR, showing the largest inhibition of expression, and identified by TargetScan, PicTar and miRanda (Figure 2C and Table 1). The gene list of interest was further narrowed by focussing on targets also associated with the most significant canonical pathways described above. Interestingly, the expression levels of 4 out of 21 mouse orthologs were also significantly down-regulated in C57BL/6 mice 14 days after instillation of bleomycin (Data set 3, GEO accession number GSE34814, Table S2). These targets, highlighted in Table 1, are ARHGAP12, CAV1, MAP3K11 and MPP5. Based on previous studies that demonstrated a significant link between the downregulation of caveolin-1 (CAV1) in lung fibroblasts and the deleterious effects mediated by TGFβ [19], [30], CAV1 represented a particularly relevant putative miR-199a-5p target gene.
Alignment of miR-199a-5p with human CAV1 3′UTR sequence revealed one potential conserved seed site (Figure 3A). We then fused part of the human CAV1 3′UTR to a luciferase reporter using the psiCHEK-2 vector and transfected it into HEK293 cells in the presence of either a pre-miR-199a-5p mimic or a pre-miR-control (Figure 3B). As a control, we also used a CAV1 3′UTR construct mutated on the predicted miR-199a-5p site. Human pre-miR-199a-5p induced a significant decrease in the normalized luciferase activity relative to control in the presence of the wild type construction only, confirming that it represents a functional site. Moreover, this inhibition was also repeated using the whole 3′-UTR of human CAV1 (Figure S4), demonstrating that CAV1 is indeed a direct target of miR-199a-5p. Finally, transfection of pre-miR-199a-5p into MRC-5 and hFL1 lung fibroblasts led to a significant and specific decrease of CAV1 at both mRNA and protein levels while miR-21 had no significant effect (Figure 3C–3E and Figure S5).
As TGFβ is known to downregulate CAV1 in pulmonary fibroblasts [19], we then investigated whether decreased expression of CAV1 upon TGFβ stimulation was associated with an increase in miR-199a-5p expression. We exposed the MRC-5 cell line to TGFβ, and analyzed the expression levels of CAV1 and miR-199a-5p. As detected by Taqman RT-PCR, TGFβ treatment of human fibroblasts for 24 h or 48 h caused a marked decrease of CAV1 mRNA, whereas miR-199a-5p expression was significantly upregulated (Figure 4A and 4B). Decrease of CAV1 protein levels after TGFβ treatment was time dependent (Figure 4C). To further investigate whether miR-199a-5p is involved in TGFβ-induced downregulation of CAV1, we performed additional experiments using a LNA-based inhibitor of miR-199a-5p as well as a CAV1 target site blocker to specifically interfere with miR-199a-5p binding on CAV1 3′UTR. As depicted in Figure 4D and S6, both LNA-mediated silencing of miR-199a-5p and blocking miR-199a-5p binding on CAV1 3′UTR inhibit TGFβ-induced downregulation of CAV1. Altogether, these experiments demonstrate that, in lung fibroblasts, induction of miR-199a-5p in response to TGFβ mediates CAV1 downregulation through binding on a unique site located in CAV1 3′UTR.
We then assessed the expression of CAV1 in the fibrotic lungs of mice. Consistent with previous studies [19], [31], our data showed a significant decrease in both CAV1 mRNA and protein expression levels in C57Bl/6 mice 14 days after bleomycin administration (Figure 5A–5C). Additionally, immunohistochemistry staining of CAV1 on lung tissue sections from C57Bl/6 mice 14 days after bleomycin treatment revealed a marked reduction of CAV1 in fibrotic area of the lungs (Figure 5D). Taken together, these experiments show that the observed up-regulation of miR-199a-5p expression in the fibrotic lungs of mice is correlated with a downregulation of CAV1. Of note, BALB/c mice, for which pulmonary expression of miR-199a-5p was not upregulated in response to bleomycin, did not display a significant decrease in CAV1 mRNA expression level 14 days after bleomycin treatment (Figure S7).
Expression of miR-199a-5p expression was increased in lungs of IPF patients (GEO accession number GSE13316 from [13]; dataset consisting of ten IPF samples and ten control samples; two different probes for miR-199a-5p with a p-value of p = 0.005 and p = 0.006, wilcoxon rank sum test, Table S3). This result was confirmed with an independent dataset composed of 94 IPF and 83 control lungs (p<0.001) (Figure 6A) and in an additional cohort using qPCR (Figure S8). As observed in mice, IPF samples also exhibited a significant decrease in CAV1 expression (p<0.001) (Figure 6B). The linear fold ratio for CAV1 between IPF and control was 0.54 (FDR<0.05) and the linear fold ratio for miR-199a-5p for the same subjects was 1.35 (p<0.05). Finally, examination of IPF lung sections revealed a specific expression of miR-199a-5p in fibroblastic foci of the injured lung as well as a decreased CAV1 expression (Figure 6C and 6D).
Given that loss of CAV1 expression represents a critical factor involved in the fibrogenic activation of pulmonary fibroblasts [19], we assessed whether overexpression of miR-199a-5p in lung fibroblasts was sufficient to recapitulate known profibrotic effects associated with a decrease in CAV1 expression (i.e. ECM synthesis, fibroblasts proliferation, migration, invasion and differentiation into myofibroblasts) [30], [32], [33]. Transfection of miR-199a-5p precursors resulted in a significant induction of migration (Figure 7A and 7B) and invasion (Figure 7C). In addition, cell cycle analysis (percent cells in S phase) showed that proliferation rate of pulmonary fibroblasts overexpressing miR-199a-5p was significantly enhanced (Figure 7D). Finally, heterologous expression of miR-199a-5p also led to a strong increase in α smooth mucle actin (αSMA) expression (Figure 7E and Figure S9), a hallmark of myofibroblast differentiation as well as to a significant potentiation of COL1A1 induction in response to TGFβ (Figure 7F).
Comparison of the gene expression profiles obtained in lung fibroblasts transfected with miR-199a-5p precursors or with a siRNA specifically directed against CAV1 revealed an overlap between the 2 signatures, mainly among the down-regulated transcripts (Figure S10A, group 2): 34% of miR-199a-5p downregulated transcripts were also repressed by a siCAV1 (Figure S10B). To gain insights into the pathways modulated by miR-199a-5p, Ingenuity Pathways canonical pathways associated to miR-199a-5p were analyzed and compared to those of miR-21 and siCAV1 conditions. This analysis revealed some proximity between miR-199a-5p and siCAV1 based on the existence of shared regulated pathways (Figure 8A). Pathways that were specific to miR-199a-5p were related to inflammation, such as “IL-1 Signaling”, “Acute Phase Response Signaling” and “P38 MAPK Signaling”, i.e. all typical of fibrotic processes. Importantly, several profibrotic genes were specifically regulated by miR-199a-5p and their altered expression was confirmed in vivo (Figure S11 and Table S4). MiR-199a-5p thus regulates multiple signaling pathways involved in lung fibrogenesis. In particular, compared to siCAV1 transfected cells, overexpression of miR-199a-5p significantly increased CCL2, TGFBRI and MMP3 expression and significantly decreased CAV2 and PLAU expression (Figure S12). Of note, these two downregulated genes were predicted to be direct targets of miR-199a-5p by Pictar.
We next investigated whether miR-199a-5p is associated with TGFβ signaling. For this, we experimentally defined a TGFβ signaling signature in lung fibroblasts (Dataset 2, GSE34815) and compared it to miR-199a-5p signature using gene set enrichment analysis (GSEA) [34]. This analysis revealed a significant overlap between these two signatures, as assessed by normalized enrichment scores above 1 (1.4 and 2.17 for up- and down-regulated genes respectively, with nominal p-value and FDR q-value being <0.05), suggesting therefore, that miR-199a-5p is involved in the TGFβ response of lung fibroblasts (Figure 8B). To further demonstrate the importance of miR-199a-5p in TGFβ response, silencing of miR-199a-5p was performed in lung fibroblasts using LNA-based inhibitors. In particular, we showed that LNA-mediated silencing of miR-199a-5p strongly inhibited TGFβ-induced differentiation of lung fibroblasts into myofibroblasts (Figure 8C and Figure S6), SMAD signaling (Figure 8D) and stimulation of wound repair (Figure 8E and 8F). Remarkably, similar results were obtained using CAV1 protector, demonstrating therefore that miR-199a-5p is a key effector of TGFβ response through CAV1 regulation (Figure 8C, 8D, 8E, 8F and Figure S6).
A growing body of evidence suggests that miRNAs contribute to the fibrotic process in various organs such as heart, kidneys, liver or lungs. For example, previous reports have shown that miR-21 has an important role in both pulmonary and heart fibrosis experimental mouse models. Thus, we investigated whether miR-199a-5p was also dysregulated in other fibrotic tissues, namely kidney fibrosis and liver fibrosis. To this end, we assessed the overlap between the miRNA expression profiles corresponding to three experimental models of fibrosis. Measurements were made using the same miRNA based platform. We identified 5 miRNAs commonly dysregulated at a p-value of less than 0.01 (Figure 9A). Among these miRNAs, 3 were downregulated (miR-193, miR-30b and miR-29c) and 2 were upregulated (miR-199a-3p and miR-199a-5p) (Figure 9B).
The enhanced expression of miR-199a-5p was confirmed in two independent experimental models of liver fibrosis (Figure 10A–10C) and was correlated with the severity of liver fibrosis, as BALB/C mice have a more pronounced liver fibrosis than C57BL/6 mice, following administration of CCL4 (Figure 10A and 10B). In addition, miR-199a-5p was significantly decreased during regression of experimental CCL4-induced liver fibrosis (Figure 10D). Furthermore, we showed that TGFβ exposure of stellate cells was associated with an increase of miR-199a-5p expression and a decrease of CAV1 expression level (Figure 10E and 10F). Interestingly, enhanced expression of miR-199a-5p was also observed in clinical samples from patients with liver fibrosis (Figure S13).
Similarly, data obtained from the unilateral ureteral obstruction model of kidney fibrosis showed an enhanced expression of miR-199a-5p in the injured kidney compared to sham operated mice (Figure 11A). Interestingly, as for lung fibrosis, kidney expression of miR-199a-5p was correlated with disease progression. As depicted in Figure 11B, in situ hybridization performed 28 days after surgery (i.e. when the fibrosis is established) showed no detectable signal for miR-199a-5p in normal kidney, whereas the hybridization signal was greatly enhanced throughout the injured kidney in area consistent with (myo)fibroblasts. Furthermore, immunohistochemistry of CAV1 performed on fibrotic kidney from mice 28 days after surgery showed a marked reduction of CAV1 in fibrotic area of the kidney (Figure 11C).
MiRNA expression profiling using high-throughput genomic approaches has provided important new insights into the pathogenesis, classification, diagnosis, stratification, and prognosis of many human diseases including tissue fibrosis [15], [35], [36]. In particular, such approaches have been previously successfully applied to IPF, revealing miR-21 and let-7d as important contributors to the lung fibrosis process [13], [14]. Our work represents however, to our knowledge, the first analysis of miRNAs involved into the differential susceptibility of two murine strains to bleomycin-induced lung fibrosis. The identification of a specific miRNA profile associated with bleomycin-sensitive animals suggests the functional importance of these dysregulated miRNAs during the pathogenic processes leading to lung fibrosis. MiR-199a-5p appeared as the most statistically significant and was well correlated to IPF progression. Thus, altered expression of miR-199a-5p is likely to represent a primary pathogenic mechanism in the development of lung fibrosis rather than a secondary effect of the long-standing disease process. Other miRNAs candidates including miR-214, clustered with miR-199a-2 on mouse chromosome 1 as well as other miRNAs that have been previously associated to fibrosis, including miR-221-222 and miR-449a [37], [38] also showed an enhanced expression in the lung fibrosis-susceptible mice. These miRNAs need to be further analyzed in IPF samples, as previous studies have shown their implication in the regulation of the stress response or the cell cycle/apoptosis balance in the epithelial or fibroblast compartment [38]–[41].
MiR-199a is an evolutionary conserved small RNA initially identified in the context of inner ear hair cells development and chondrogenesis [42]–[44] and numerous reports have now shown its implication in various tumor types [45]–[47]. In the context of tissue fibrosis, both mature forms of miR-199a (i.e., miR-199a-5p and miR-199a-3p) have been associated with the progression of liver fibrosis in both humans and mice [48], [49]. While our data also showed an enhanced pulmonary expression of these two miRNAs in the bleomycin-induced mouse model, expression of miR-199a-5p was more significant in IPF samples (Table S3 and Figure S14A). Moreover, our data indicated that miR-199a-3p had distinct effects on lung fibroblasts differentiation than miR-199-5p, as assessed by their different impact on αSMA (Figure S14B and data not shown). This led us to investigate in depth miR-199a-5p profibrotic effects.
In a recent report describing the miRNA expression profile of lung fibroblasts, miR-199a-5p was found to be highly expressed [25]. Our present data establish stromal cells as the primary source of miR-199a-5p in the injured lungs and also suggests that miR-199a-5p is involved in the profibrotic effects mediated by pulmonary fibroblasts. A combination of in silico and experimental data, described in [25], [26], [40], identified the transcripts affected by miR-199a-5p in lung fibroblasts. Functional annotations of the miR-199a-5p experiments highlighted terms such as “Integrin Signaling” and “Caveolar-mediated Endocytosis Signaling”. Among the set of transcripts that were down-regulated after ectopic expression of miR-199a-5p, we then restricted our work to a group of 21 miR-199a-5p target genes predicted by 3 independent algorithms, showing the largest modulation factors and smallest statistical p-values. This short list included CAV1, a structural component of caveolae, previously associated with lung fibrosis [14], [18], [19].
Caveolae refer to 50–100 nanometers small bulb-shaped invaginations of the plasma membrane. They exert major biological functions in numerous cellular processes such as membrane trafficking or cell signaling [50]. CAV1 and CAV2, the main coat proteins of caveolae, are relatively highly expressed in endothelial cells and fibroblasts of pulmonary origin [51]. Caveolae role is particularly important in the context of TGFβ signaling. Whereas TGFβ receptor endocytosis via clathrin-coated pit-dependent internalization promotes TGFβ signaling, the lipid raft-caveolar internalization pathway facilitates the degradation of TGFβ receptors, therefore decreasing TGFβ signaling [52]. Previous studies have shown that a reduced CAV1 expression in lung fibroblasts contributes to IPF pathogenesis by promoting TGFβ profibrotic effects [19]. In line with this, we provide evidence that miR-199a-5p can directly repress CAV1 in lung fibroblasts, thereby stimulating their proliferation, migration, invasion and differentiation into myofibroblasts (Figure 12). Additionally, we showed in a large cohort of IPF patients an enhanced expression of miR-199a-5p that was reproduced in three independent mouse models of fibrosis as well as a decreased expression of CAV1. Finally, in contrast to a recent report showing that miR-199a-5p, by targeting SMAD4, inhibited TGFβ-induced gastric cancer cell growth [53], we found that lung fibroblasts overexpressing miR-199a-5p have an increased SMAD4 expression (Figure S14B), suggesting thus a potential opposite function of this miRNA between epithelial and mesenchymal cells.
MiRNAs, by affecting the expression of multiple genes, can act as master regulators of complex biological processes and aberrant expression of miRNA is known to have a profound impact on various distinct biological pathways. Thus, the elucidation of the critical genes and relevant pathways/networks modulated by miRNAs is important to understand the mechanisms by which miRNAs exert their pathogenic effects. Our systematic analysis of the gene expression profiles of lung fibroblasts overexpressing miR-199a-5p led to the identification of a large number of transcripts that were significantly modulated by this miRNA. These experiments have established that miR-199a-5p is directly or indirectly involved in the regulation of genes previously associated with lung fibrosis: CCL2, a potent mononuclear cell chemoattractant, PLAU [54], a component of the fibrinolysis system, TGFBRI, the TGFβ receptor type I [55], MMP3 [56] and CAV2 [57]. Interestingly, these regulations were independent of CAV1 targeting, suggesting therefore that miR-199a-5p modulates the expression of several components of various distinct biological pathways to elicit, in lung fibroblasts, a profibrotic response.
Before this study, miR-21 was clearly established as an effector of TGFβ signaling, able to promote fibroblast proliferation and differentiation into myofibroblasts [58]. In the context of lung fibrosis, miR-21 has been described to mediate lung fibroblast activation and fibrosis [14]. MiR-199a-5p and miR-21 exert indeed similar pro-fibrotic effects on lung fibroblasts. This is further demonstrated by overexpression of miR-21 and miR-199a-5p, which induce lung fibroblast migration to a similar extent (Figure S15). Interestingly, while both miRNAs appear as TGFβ effectors, the comparison of their associated gene expression signature indicated a limited overlap (Figure 2A). Moreover, CAV1 expression is unaffected by overexpression of miR-21 in lung fibroblasts, suggesting that both miRNAs, in response to TGFβ, modulate distinct signaling pathways to produce cooperative effects involved in fibroblast activation.
The mechanisms involved in the TGFβ-dependent modulation of miR-21 and miR-199a-5p are also of particular interest. While both miR-21 and miR-199a-5p have been shown to be regulated by TGFβ, their expression may be primarily regulated through a Smad-dependent post-transcriptional mechanism promoting miRNA maturation by Drosha [59], [60]. Our data showing that both pri-miRNA-199a1 and pri-miRNA-199a2 are significantly upregulated in bleomycin-treated mice (Figure S2B) and TGFβ-stimulated fibroblasts (Figure S16) suggest that additional TGFβ-dependent transcriptional regulations occur that will need to be more fully analyzed.
Finally, our observation that miR-199a-5p is also dysregulated in two additional experimental models of tissue fibrosis (i.e. kidney fibrosis and liver fibrosis) establishes miR-199a-5p as a ubiquitous factor associated with tissue fibrogenesis. The recently reported association of CAV1 with kidney fibrosis [61], [62], together with the exclusive distributions of miR-199a-5p and CAV1 in the injured kidney, leads us to hypothesize that miR-199a-5p also controls CAV1 expression in kidney, thus contributing to kidney fibrosis. Further information came from the liver fibrosis model. As liver fibrosis can regress after cessation of the triggering injury, even at advanced fibrotic stages [63], the decrease of miR-199a-5p observed during resolution of liver fibrosis sets for the first time a specific miRNA as an important player for orchestrating the molecular events occurring during regression of liver fibrosis. Importantly, this implies that therapeutic strategies based on modulation of miRNAs have a potential to prevent liver fibrosis progression but also to resolve liver fibrosis.
In conclusion, the results of this study further underline the pivotal roles played by specific miRNAs in mediating changes in gene expression and cell functions occurring during pulmonary fibrosis. In particular, our results identified miR-199a-5p as a new determinant of tissue fibrosis. We thus anticipate that strategies preventing the up-regulation of miR-199a-5p may represent a new effective therapeutic option to treat fibroproliferative diseases.
Human normal pulmonary fibroblasts MRC-5 (CCL-171) and hFL1 (CCL-153), human lung cancer cell line A549 (CCL-185) and HEK-293 (CRL-1573) cells were purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA), frozen at an early passage and each vial used for experiments was cultured for a limited number of passages (<8). For maintenance, cells were cultured in the appropriate medium (MEM for MRC-5, F12-K for hFL1 and A549, DMEM for HEK-293) containing 10% fetal calf serum (FCS), at 37°C with 5% v/v CO2. Recombinant TGFβ was purchased from Sigma-Aldrich. The following monoclonal (mAbs) and polyclonal (pAbs) Antibodies were used: rabbit anti-CAV1 pAbs (sc-894, Santa Cruz Biotechnology Inc.), rabbit anti-SMAD4 (9515) and anti- β-Actin (13E5) mAbs (Cell Signalling), mouse anti- αSMA mAbs (1A4, Dako) for immunohistochemistry and (4A8-2H3, Abnova) for Western Blot and immunocytofluorescence.
All animal care and experimental protocols were conducted according to European, national and institutional regulations. Personnel from the laboratory carried out all experimental protocols under strict guidelines to insure careful and consistent handling of the mice.
Primary stellate cells were isolated from C57BL/6 mice at the age of 40 to 55 weeks and stimulated with 20 ng/ml of TGF-ß (Sigma Aldrich) for 48 h as previously described [65].
Flash frozen lung tissue from 94 human subjects with IPF and 83 control subjects with no chronic lung disease were obtained from the Lung Tissue Research Consortium (LTRC). These diagnoses were made using ATS/ERS guidelines [66], [67] from review of clinical history, pathology, and radiology. All experiments were approved by the local Institutional Review Board at the University of Pittsburgh (IRB# 0411036). Clinical data were made entirely available to the investigators for review.
Paraffin lung sections from patients with IPF were obtained from Lille's Hospital. Experiments were approved by the institutional review board of Lille's Hospital.
Kidneys and lungs were fixed overnight with neutral buffered formalin and then embedded in paraffin. Five-micrometer-thick sections were mounted and stained with hematoxylin and eosin as well as Masson's trichrome to assess the degree of fibrosis. Histologic sections were reviewed by an experienced pathologist.
Total RNA were extracted from lung tissue and cell samples with TRIzol solution (Invitrogen). Integrity of RNA was assessed by using an Agilent BioAnalyser 2100 (Agilent Technologies) (RIN above 7).
Cells or tissues were lysed in lysis buffer (M-PER protein extraction reagent for cells, T-PER protein extraction reagent for tissues) and protease inhibitors cocktail (Pierce). The lysates were quantified for protein concentrations using the Bradford assay (Biorad). Proteins (10 µg per sample) were separated by SDS-polyacrylamide gel and transferred onto nitrocellulose membranes (GE Healthcare). The membranes were blocked with 5% fat free milk in Tris-buffered saline (TBS) containing 0.1% Tween-20 (TBS-T) and subsequently incubated with CAV1, α-SMA or β-actin primary antibodies overnight at 4°C. After washing with TBS-T for 30 minutes at room temperature, the membrane was further incubated with horseradish peroxidase–conjugated secondary antibodies for 1.5 hours, followed by 30 minutes of washing with TBS-T. Protein bands were visualized with Amersham ECL substrates (GE Healthcare).
Five-µm paraffin-embedded sections were sequentially incubated in xylene (5 minutes twice), 100% ethyl alcohol (5 minutes twice), 95% ethyl alcohol (5 minutes twice), and 80% ethyl alcohol (5 minutes). After washing with water, the sections were antigen-retrieved using citrate buffer (pH 6.0; DAKO) in a steamer for 20 minutes and cooled to ambient temperature. Sections were then washed with TBS-T and quenched with 3% hydrogen peroxide in TBS for 10 minutes, blocked for avidin/biotin activity, blocked with serum-free blocking reagent, and incubated with primary antibody as follows: for CAV1 staining, sections were incubated with antibody for 1 hour at ambient temperature; for alpha-SMA staining, sections were incubated with antibody overnight at 4°C. Immunohistochemical staining was developed using the DAB substrate system (DAKO).
In situ hybridization of miR-199a-5p was performed using double DIG-labeled LNA probes (Exiqon, Woburn, MA). Paraffin-embedded mouse tissues were dewaxed in xylene and rehydrated in descending grades of alcohol. The slides were then washed in PBS (pH 7.5) and permeabilized by incubating for 15 min in proteinase K (Ambion) for 15 min at 37°C. The slides were again washed in PBS, and prehybridized in hybridization buffer (50% formamide, 5× SSC, 0.1% Tween-20, 9.2 mM citric acid, 50 µg/ml heparin, and 500 µg/ml yeast RNA, pH 6) in a humidified chamber. The double DIG-labeled LNA probes were then added to the sections at a 80 nM concentration and incubated 2 hours at 50°C in a humidified chamber. The slides were rinsed in 5× SSC, 1× SSC and 0.2× SSC solutions at the same hybridization temperature. This step was followed by blocking with 2% sheep serum, 2 mg/ml BSA in PBS+0.1% Tween-20 (PBST) and incubation with anti-DIG-AP Fab fragments antibody (1∶800) (Roche Applied Sciences) for 2 hours at room temperature. After washing in PBST, the color reaction was carried out by incubation in 5-bromo-4-chloro-3-indolyl phosphate (BCIP)/nitro blue tetrazolium (NBT) color solution (Roche Applied Sciences) with 1 mM levamisole overnight at room temperature. The color reaction was stopped after observation of sufficient development of blue precipitate by washing with PBST. The slides were then counterstained with Fastred (Sigma Aldrich), mounted and coverslipped.
MRC5 cells were grown on a Round Glass Coverslips Ø 16 mm (thermo scientific) placed inside a 12 Multiwell Plate. Coverslips slides were washed in phosphate-buffered saline and fixed in 4% paraformaldehyde for 15 min, cells were then permeated using 0.1% Triton X-102 (Agilent Technologies) for 10 min. and blocked with PBS solution containing BSA (3%) for 30 min. Incubation with primary antibodies was performed in a bloking solution BSA (1%) at 37°C for 1 h at the following dilutions; α-SMA (1∶1000), CAV1 (1∶50),. After three washes with PBS, cells were incubated with secondary Alexa Fluor 488 goat anti-Mouse IgG (Invitrogen) (1∶500), Alexa Fluor 647 goat anti-rabbit IgG (Invitrogen) (1∶500) and Alexa Fluor 647 Phalloidin (A22287 - Life technologies) (1Unit/slide). Fourty five min later, Coverslips slides were fixed on microscope slides using ProLong Gold Antifade Reagent with DAPI (Invitrogen). Fluorescence was viewed with an FV10i Olympus confocal scanning microscope.
MRC5 cells (150,000/well) were seeded in duplicate in DMEM supplemented with 10% FBS on 60-mm cell culture dishes. Cells were serum starved the next day and transfected with pre-miR-199a-5p at 10 nM. Cell proliferation was assessed 48 h after transfection by flow cytometry using the click-iT EdU cell proliferation assay (Invitrogen) according to the manufacturer's instructions.
hFL1 cells were seeded on Type-I collagen coated 12-well plates and transfected as described above. Forty eight hours after transfection, confluent cells were (FCS) starved 3 h before adding 10 ng/ml TGFβ and wounded using a pipet tip. The in vitro wound-healing process was then recorded by videomicroscopy for 24 h from then scratching on an Axiovert 200 M inverted microscope (Carl Zeiss) equipped with 37°C and 5% CO2 regulated insert (Pecon GmbH). Brightfied images were taken each 30 min through a 10× phase contrast objective with a CoolSNAPHQ CCD Camera managed by Metamorph Software (Roper Scientific). The motility of the cells was assessed by evaluating the repaired area percentage using ImageJ sotware.
Invasion of MRC5 fibroblast overexpressing miR-199a-5p was assessed using commercially available 24-well BioCoat Matrigel Invasion Chamber (BD Biosciences). In brief, pulmonary fibroblasts were transfected either with pre-miR-199a-5p or negative control as described above. Twenty four hours after transfection, cells were harvested with trypsin-EDTA, centrifuged, and resuspended in DMEM medium. Cell suspensions (1×105 cells/well) were added to the upper chamber. Bottom wells of the chamber were filled with DMEM medium containing 10% FBS as chemoattractant, whereas the upper chamber was filled with DMEM only. After incubation for 48 h at 37°C, the non-invading cells on the top of the membrane were removed with a cotton swab. Membrane containing invading-cells were fixed with methanol, washed three times with PBS and mounted with DAPI hard set (Vector Laboratories) onto glass slides for fluorescent microscopy.
Results are given as mean±S.E.M. Statistical analyses were performed by using Student's t-test as provided by Microsoft Excel.
|
10.1371/journal.pntd.0000367 | A New PCR-Based Approach Indicates the Range of Clonorchis sinensis Now Extends to Central Thailand | Differentiation of the fish-borne trematodes belonging to the Opisthorchiidae, Heterophyidae and Lecithodendriidae is important from a clinical and epidemiological perspective, yet it is impossible to do using conventional coprological techniques, as the eggs are morphologically similar. Epidemiological investigation therefore currently relies on morphological examination of adult worms following expulsion chemotherapy. A PCR test capable of amplifying a segment of the internal transcribed spacer region of ribosomal DNA for the opisthorchiid and heterophyid flukes eggs taken directly from faeces was developed and evaluated in a rural community in central Thailand. The lowest quantity of DNA that could be amplified from individual adults of Opisthorchis viverrini, Clonorchis sinensis and Haplorchis taichui was estimated at 0.6 pg, 0.8 pg and 3 pg, respectively. The PCR was capable of detecting mixed infection with the aforementioned species of flukes under experimental conditions. A total of 11.6% of individuals in rural communities in Sanamchaikaet district, central Thailand, were positive for ‘Opisthorchis-like’ eggs in their faeces using conventional parasitological detection techniques. In comparison to microscopy, the PCR yielded a sensitivity and specificity of 71.0% and 76.7%, respectively. Analysis of the microscopy-positive PCR products revealed 64% and 23% of individuals to be infected with O. viverrini and C. sinensis, respectively. The remaining 13% (three individuals) were identified as eggs of Didymozoidae, presumably being passed mechanically in the faeces following the ingestion of infected fishes. An immediate finding of this study is the identification and first report of a C. sinensis–endemic community in central Thailand. This extends the known range of this liver fluke in Southeast Asia. The PCR developed herein provides an important tool for the specific identification of liver and intestinal fluke species for future epidemiological surveys.
| It is estimated that approximately 17 million people are currently infected with fish-borne flukes worldwide. The fish-borne liver flukes Opisthrochis viverrini and Clonorchis sinensis cause hepatic and biliary disease in humans. The minute intestinal flukes are widely distributed in southeast Asia and are increasingly recognised as an emerging pathogen associated with diarrhoea and gastritis. The most significant finding of this study is the discovery and first report of a C. sinensis–endemic community in Thailand. This finding was aided by the development and application of a new PCR-based technique capable of specifically detecting and characterising O. viverrini, C. sinensis and the minute intestinal flukes, directly from eggs in faeces. Since the eggs are morphologically similar, the fish-borne flukes cannot be differentiated on basis of microscopic examination of stool. This publication also questions the presumption that the distribution of fish-borne liver fluke species in Asia closely parallels the distribution of the snail intermediate hosts. The PCR provides a useful diagnostic tool for further large-scale epidemiological surveys to be carried out in Southeast Asia, which will shed further light on the distribution of these liver flukes in human and snail intermediate hosts with the advantage that targets for more arduous anthelmintic flushing confirmations can be carried out.
| It is estimated that approximately 17 million people are currently infected with fish-borne trematodes worldwide [1]. In Asia, Opisthorchis viverrini is known to occur in Thailand, Laos, Cambodia and southern Vietnam and Clonorchis sinensis in Korea, China, Taiwan and northern Vietnam [2],[3].
Liver fluke infection in Thailand is unevenly distributed with a highly endemic focus of infection in the northeast region [4]. Previous parasite surveys have mostly focussed on these communities and frequently found infection with O. viverrini mixed with minute intestinal flukes of the Heterophyidae and Lecithodendriidae [5],[6]. The heterophyids Haplorchis taichui and less frequently H. pumilio are the most common minute intestinal flukes recovered.
Microscopic examination of faecal samples for the presence of eggs using the formalin-ether concentration technique (FECT) is currently considered the most sensitive and reliable method for screening liver and intestinal flukes and is therefore the most widely employed technique for fluke parasite surveys [7]. This technique is limited by its capacity to differentiate between the Opisthorchiidae, Heterophyidae and Lecithodendriidae, which have similar egg morphologies. Eggs can therefore only be characterised as ‘Opisthorchis/Clonorchis- like’ [5],[6], but no further. A definitive diagnosis to species level requires morphological identification of adult flukes following expulsion chemotherapy [8],[9]. The ability to differentiate the species of liver and minute intestinal flukes is important from both a clinical and epidemiological perspective. Heavy infections with the minute intestinal flukes are associated with diarrhoea, mucus-rich faeces, dyspepsia, nausea and vomiting [10], whereas infections with the liver flukes result in mostly biliary and hepatic disease. The frequency and types of pathology and clinical disease among C. sinensis and O. viverrini also seem to differ [7]. For example, cholelithiasis is one of the more serious complications of clonorchiasis, but a rare complication of opisthorchiasis. Although both flukes are implicated as predisposing factors for cholangiocarcinoma, this is more frequent with O. viverrini. From an epidemiological perspective, C. sinensis has a wider definitive host range than O. viverrini [11],[12] which makes control more challenging.
To overcome the diagnostic limitations associated with conventional parasitological methods, a number of PCR-based techniques capable of amplifying species of flukes directly from eggs in faeces have been developed [13]–[16]. An O. viverrini-specific PCR test capable of detecting O. viverrini eggs directly from human faeces was shown to have an analytical sensitivity of 100%, 68.2% and 50% compared to the Stoll's egg count containing >1000, 200 to 1000 and <200 eggs per g of faeces respectively and an analytical sensitivity of 97.8% under experimental conditions [16]. This PCR assay proved less reliable however, once evaluated under field conditions with an overall diagnostic sensitivity of 45% compared to the FECT [17]. PCR tests based on amplification of the mitochondrial gene for the identification and discrimination of C. sinensis and O. viverrini [13] and C. sinensis, O. viverrini and H. taichui [15] have been developed and shown to be analytically sensitive under experimental conditions. Amplicons for the targeted fluke species could be obtained in reactions containing 0.78 ng of genomic DNA [13] and 10−4 ng of genomic DNA [15], however the assays have yet to be evaluated and compared to conventional parasitological methods in the field.
Here we developed a PCR test capable of specifically amplifying a segment of the internal transcribed spacer (ITS-2) region of ribosomal DNA (rDNA) from opisthorchiid and heterophyid flukes directly from eggs in faeces. The ITS-2 has successfully discriminated species from many digenean families and has become the default region of choice for distinguishing species of trematodes [18]. This PCR test is evaluated in terms of both analytical and diagnostic sensitivity and specificity against conventional parasitological methods in a community endemic for liver fluke infection in central Thailand. This study is also the first to demonstrate the occurrence of a C. sinensis endemic community in Thailand.
A rural community consisting of a total population of approximately 5465 people in Nayao village, Sanamchaikaet District, Chachoengsao Province, 150 km east of Bangkok was chosen for this cross-sectional study. The area lies in a low basin of land which is suitable for cultivation of rice which provides the principle income for the province. The dietary habit of eating raw and fermented fish dishes such as ‘koi pla’, ‘pla som’ and ‘pla ra’ from fish sourced at local ponds is popular among residents of this community. Houses were chosen at random and household members informed of the study by medical students from the Phramongkutklao College of Medicine. After signing human ethics consent forms, single stool samples were collected from a total of 335 individuals of all ages, gender and backgrounds during a 10-day period in mid November 2004. Participants found positive for gastrointestinal parasites received free anthelmintic treatment from the medical doctors on the research team in order to increase compliance. All samples were qualitatively evaluated and run in parallel for the presence of Opisthorchis-like eggs using a direct faecal smear (DFS), Kato Katz (KK) technique and the FECT by experienced parasitology technicians from the Phramongkutklao College of Medicine in the field. Any remaining faecal material was fixed in 20% dimethylsulfoxide (DMSO) saturated with salt for transport to the University of Queensland for molecular testing.
A single individual found positive for ‘Opisthorchis-like’ eggs in their faeces was treated with a single dose of praziquantel (40 mg per kg) and was then given 30 g magnesium sulfate with as much water as possible to facilitate expulsion of adult flukes. Whole diarrhoetic stools were collected and washed several times before isolating the adult flukes [9]. Adult flukes that had been expelled by this individual were fixed in 70% ethanol for molecular and morphological identification at the University of Queensland. Morphological identification was performed by staining the adult fluke with haematoxylin, dehydrating it in alcohol and clearing it in methyl salicylate before mounting in Canada balsam.
This study was approved by the Murdoch University Human and Animal Ethics Committees of Western Australia and the Ethical Committee, the Medical Department Royal Thai Army.
Adult flukes of O. viverrini, C. sinensis and H. taichui were extracted using the Qiagen DNeasy Blood and Tissue Kit according to manufacturer's instructions. Those faecal samples found microscopically positive for ‘Opisthrochis-like’ eggs using at least one parasitological test and where sufficient quantities of stool remained, were subjected to DNA extraction and PCR (n = 31). In addition, 30 faecal samples negative for ‘Opisthorchis-like’ eggs by microscopy were also randomly selected and subjected to DNA extraction and PCR. All PCR reactions were conducted by a single experienced molecular biologist that was blind to the results of the parasitological test results for each sample. It was observed that subjecting ‘Opisthorchis-like’ eggs purified by a saturated salt and glucose gradient to freezing in liquid nitrogen followed by thawing them at 98–100°C resulted in the eggs ‘disintegrating’ to release genomic DNA. Two hundred milligrams of faeces were suspended in 1.4 ml ATL tissue lysis buffer (Qiagen, Hilden, Germany) and this suspension subjected to 5 cycles of freezing-thawing at liquid nitrogen temperatures. DNA was then isolated from the supernatant using the QIAamp DNA Mini Stool Kit according to manufacturer's instructions. Final elutions of DNA were made in 50 µl of elution buffer instead of 200 µl as recommended by the manufacturer.
Sequences of the ITS-2 region of C. sinensis, O. viverrini, O. felineus, H. taichui, H. pumilio and Centrocestus sp. (GenBank accession nos. EF688144, EF688143, AY584735, DQ513403, DQ513405, AY245705, AY245706, AY245699) were aligned using Clustal W (http://align.genome.jp/) and the primer pair: RTFlukeFa 5′CTTGAACGCACATTGCGGCC-3′ and RTFlukeRa 5′-CACGTTTGAGCCGAGGTCAG-3′ were designed to amplify a 375 bp, 381 bp and 526 bp region of O. viverrini, C. sinensis and H. taichui, respectively, The PCR primers, were also designed with the potential to amplify other species of opisthorchiid and heterophyid flukes. The PCR assay was carried out in a volume of 20 µl containing 1×PCR buffer from Qiagen (Tris-HCl, KCl, (NH4)2SO4, 1.5 mM MgCl2; pH 8.7) additional MgCl2 to give a final 2.0 mM concentration, 200 µM of each dNTP, 0.25 µM of each primer, and 1 unit Hot Star Taq DNA polymerase (Qiagen). The PCR cycle consisted of an initial stage: 94°C for 15 min, 60°C for 1 min and 72°C for 2 min followed by 35 cycles of 94°C for 30 sec, 60°C for 30 sec, 72°C for 30 sec, a final extension at 72°C for 7 min and a holding temperature of 12°C.
PCR products were run on 1.5% agarose in 1×TAE buffer at 150V in a Biorad electrophoresis system and were purified using Qiagen spin columns (Qiagen) prior to sequencing. Where a multi-banded product was obtained, target bands were excised, frozen and cleaned up with a Quantum Prep Freeze ‘N Squeeze DNA Gel Extraction spin column (Biorad) or a Qiaquick Gel Extraction kit (Qiagen). Sequencing was done using an ABI 3130xl Genetic Analyzer (Applied Biosystems) using Big Dye 3.0 chemistry, after which sequences were edited and assembled using Chromas Pro (Technelysium Pty Ltd).
Titration experiments were conducted to determine the analytical sensitivity of the PCR for the detection of C. sinensis, O. viverrini and H. taichui DNA. The assay's ability to detect artificially mixed infections with varying ratios of C. sinensis and O. viverrini with H. taichui were also assessed.
Assuming microscopy as the ‘gold standard’, the diagnostic sensitivity, and specificity together with their 95% confidence intervals were calculated for the PCR using the Wilson method.
The assay's ability to detect artificially mixed infections of O. viverrini and C. sinensis was assessed by development of a PCR-RFLP as both species produced PCR products that could not be differentiated by size. Amplified ITS-2 products of RTFlukeFa – RTFlukeRa for C. sinensis and O. viverrini were digested with AcuI (New England Biolabs). According to the restriction profile generated by Nebcutter V2.0 (New England Biolabs), O. viverrini does not possess a restriction site for AcuI and remains uncut (375 bp), whereas C. sinensis has a single AcuI site and gives rise to two bands at 286 bp and 95 bp. Ten microlitres of PCR product were digested with 2.5 units of the restriction endonuclease AcuI (New England Biolabs) at 37°C for 3 hours in a volume of 20 µl.
Using the primer pair RTFlukeFa and RTFlukeRb, DNA from morphologically identified adults of O. viverrini, C. sinensis and H. taichui gave specific products of 375 bp, 381 bp and 526 bp respectively. The lowest quantity of DNA that could be amplified from individual adults of O. viverrini, C. sinensis and H. taichui was estimated at 0.6 pg, 0.8 pg and 3 pg respectively. Appropriate sized amplicons were produced in reactions artificially mixing DNA of C. sinensis and O. viverrini separately, with H. taichui, in ratios of 1∶1, 1∶2, 1∶3, 3∶1 and 2∶1 (Fig. 1A and 1B). The PCR however, preferentially amplified O. viverrini when artificially mixed with H. taichui. Weak to negligible bands of H. taichui were produced when mixed in ratios of less than 1∶1 with O. viverrini (Fig. 1B). The PCR-RFLP patterns for differentiating and detecting mixed infections of O. viverrini and C. sinensis are displayed (Fig. 1C). The PCR-RFLP was successful at detecting artificially mixed infections of O. viverrini and C. sinensis in ratios of 1∶1, 1∶2. 1∶3, 3∶1 and 2∶1.
For a diagrammatic guide to the study design and summary of diagnostic results refer to Fig. 2. A total of 39 (prevalence 11.6%, 95% CI, 8.6%, 14.92%) individuals were found positive for ‘Opisthrochis-like’ eggs in their faeces using a combination of all three microscopic techniques (DFS, KK and FECT). The FECT detected ‘Opisthorchis-like’ eggs in more faecal samples (25/31) than the DFS (9/31) and KK (10/31) methods. Using primer pair RTFlukeFa and RTFlukeRb, PCR-positive samples derived from DNA extracted directly from faeces produced a single product corresponding to the expected amplicon size for O. viverrini and C. sinensis (approximately 380 bp). In three cases, non-specific amplicons were produced in addition to the target PCR product, however these amplicons were too weak (faint) to subject to DNA sequencing. The results of the PCR analysis of 31 microscopy positive and 30 microscopy negative samples are presented in Table 1. The PCR test, when compared to the combined microscopy results yielded a sensitivity of 71.0% (95% CI, 53.4%, 83.9%), and specificity of 76.7% (95% CI, 59.1%, 88.2%). PCR detected an additional seven samples positive for liver fluke that were negative by microscopy. Mixed infections of O. viverrini and C. sinensis were detected in a single individual by PCR-RFLP.
Morphological and genetic characterisation of the fluke species expelled by the human participant
Adult fluke specimens isolated from a single human participant in the community were identified by morphology as Clonorchis sinensis [19]. Two adult fluke specimens subjected to PCR demonstrated 100% DNA sequence homology to the ITS-2 region of C. sinensis isolates from Japan and Russia (GenBank accession nos. EF688144 and EF688143).
Phenogram construction of the ITS-2 region of the flukes using the neighbour-joining algorithm and maximum parsimony (Fig. 3), produced strong bootstrap support for the placement of 15 PCR-positive samples within a single clade corresponding to O. viverrini (GenBank accession number AY584735) and 11 PCR-positive samples corresponding to C. sinensis (GenBank accession nos. EF688144, EF688143). Mixed infections with fluke species were not observed by sequencing of the PCR product. Of the 22 individuals found positive for ‘Opisthorchis-like’ eggs by both microscopy and by PCR, 14 (64%) were characterised as O. viverrini and five (23%) as C. sinensis. In addition, three samples (13%) microscopy positive for ‘Opisthorchis-like’ eggs produced amplicon sizes of approximately 410 bp each and upon sequencing, were genetically similar to the didymozoids (parasites of fishes), Rhopalotrema elusiva (GenBank accession no. AJ224759) and Indodidymozoon sp. (GenBank accession no. AJ224754). Six of the microscopy negative but PCR positive samples were genetically characterised as C. sinensis and a single sample as O. viverrini. No intra-species variation was observed for the O. viverrini isolates obtained from this community relative to those obtained from northeast Thailand (positive control and published GenBank isolate AY584735). Apart from three isolates of C. sinensis obtained from this community that differed by a transition at a single base, all other isolates of C. sinensis were identical to published ITS-2 sequences of C. sinensis from Japan (GenBank accession no. EF688144) and Russia (GenBank accession no. EF688143).
A significant finding of this present study was the identification and first report of a community endemic for C. sinensis in central Thailand. It is possible that the humans in this community were infected by imported fish or visited C. sinensis endemic areas, however when questioned about this, villagers reported only eating fish caught in local ponds or from local village markets and had no history of travelling outside Thailand. Previous studies have only reported C. sinensis in Korea, China, Taiwan, Japan, northern Vietnam and the far eastern part of Russia [20]. It is hypothesised that the geographical distribution of clonorchiasis closely parallels the distribution of the snail intermediate host [20], however this assumption may not be as simple as previously thought. Species of Parafossarulus and Bithynia are most commonly reported to act as first intermediate hosts for C. sinensis. The important species in China, Korea and Japan is Parafossarulus manchouricus and P. anamalospiralis [7]. Other susceptible snails in China are reported as Bithynia fuchsiana, B. longicornis, Melanoides tuberculata and Assiminea lutea [20]. In Thailand, Bithynia siamensis goniomphalos, B. s. funiculate and B. s. siamensis act as hosts for O. viverrini [19]. In a recent survey of freshwater mollusks in Thailand, an intermediate host of C. sinensis, Melanoides tuberculata was isolated in the provinces to the south (Chanthaburi) and north (Nakhon Ratchasima Province) of our study area [21]. It is possible that this species of snail may be acting as the natural intermediate host of C. sinensis in Thailand. If this is true, then C. sinensis may be as geographically widespread as O. viverrini in Thailand, reflecting the geographical distribution of M. tuberculata, which was isolated from 9/15 districts sampled in the north, east and central regions of Thailand [21]. In saying this however, M tuberculata has been shown to harbour both C. sinensis and O. viverrini in both northern and southern regions of Vietnam, yet surveys to date have found C. sinensis to be restricted to the northern provinces and O. viverrini to the southern provinces [2]. The distribution of potential snail intermediate hosts therefore does not necessarily reflect the distribution of the liver flukes in Southeast Asia. The PCR test developed in this study provides a useful diagnostic tool for further epidemiological surveys to determine the distribution of these liver flukes in human and intermediate hosts.
The PCR test developed in this study is capable of amplifying O. viverrini, C. sinensis and potentially the minute intestinal flukes, directly from eggs in faeces. In terms of test parameters, this assay demonstrated a superior sensitivity (Se) to the PCR developed by Stensvold et al. (2006) in the field (Se = 70.9% compared to Se of 45.0%). It also has the added advantage of being able to amplify fluke species other than O. viverrini. It may be likely that the presence of faecal inhibitors and/or the unsuccessful ‘cracking open’ of these highly resistant eggs during DNA extraction accounted for the false negative results produced by the PCR in this study. The overall specificity (Sp) of the PCR evaluated using microscopy negative field samples were inferior to those reported by Stensvold et al. (2006) (Sp = 76.7%, compared to Sp: 90.0%), however these assumed ‘false positive’ samples were being compared to the microscopy results (DFS, KK, FECT) which are in themselves not ‘gold standards’. DNA sequences generated from the PCR products of these samples were characterised as either O. viverrini or C. sinensis and therefore the specificity of this PCR may be under-estimated.
Three faecal samples microscopy positive for ‘Opisthorchis-like’ eggs were sequenced and identified as being close to the didymozoids Rhopalotrema elusiva and Indodidymozoon sp. This is not the first time that eggs of didymozoid flukes have been recovered in human faecal samples [22]. Flukes belonging to the Didymozoidae parasitize a wide range of species of marine fish and ingestion of these adult flukes by humans during the consumption of fish results in the mechanical passage of the relatively thick-shelled eggs into the faeces. Because of their dimensions (35–43×12–28 µm) and morphology (oval, operculate) of the eggs of didymozoid flukes, they can easily be confused with eggs of the Opisthorchiidae, Heterophyidae and Lecithodendriidae. This added confusion may result in further inaccuracies when estimating the prevalence of liver and intestinal flukes in a community using conventional parasitological procedures alone.
The apparent absence of H. taichui in the Sanamchaikaet district community was surprising given it is reported commonly in the northeast region of Thailand. It is possible that the PCR failed to amplify eggs from faecal samples with mixed infections of H. taichui and O. viverrini. Under experimental conditions, the PCR showed good analytical sensitivity for detecting H. taichui as a single infection and also when artificially mixed with C. sinensis, but failed to amplify a strong band when artificially mixed with O. viverrini. Since small intestinal flukes have commonly been found as mixed infections with liver fluke species [5],[23], the PCR developed in this study may not be successful at detecting these infections.
In conclusion, we present data to demonstrate for the first time in Thailand a community endemic for C. sinensis infection. This significant finding undoubtedly opens a new chapter for further research into investigating the distribution, and prevalence of C. sinensis in Thailand and determining the natural intermediate host species capable of supporting its life cycle. Furthermore, the PCR described herein provides a valuable tool for screening and determining the species of liver and intestinal flukes in epidemiological surveys.
|
10.1371/journal.ppat.1002291 | A Genetic Screen Reveals Arabidopsis Stomatal and/or Apoplastic Defenses against Pseudomonas syringae pv. tomato DC3000 | Bacterial infection of plants often begins with colonization of the plant surface, followed by entry into the plant through wounds and natural openings (such as stomata), multiplication in the intercellular space (apoplast) of the infected tissues, and dissemination of bacteria to other plants. Historically, most studies assess bacterial infection based on final outcomes of disease and/or pathogen growth using whole infected tissues; few studies have genetically distinguished the contribution of different host cell types in response to an infection. The phytotoxin coronatine (COR) is produced by several pathovars of Pseudomonas syringae. COR-deficient mutants of P. s. tomato (Pst) DC3000 are severely compromised in virulence, especially when inoculated onto the plant surface. We report here a genetic screen to identify Arabidopsis mutants that could rescue the virulence of COR-deficient mutant bacteria. Among the susceptible to coronatine-deficient Pst DC3000 (scord) mutants were two that were defective in stomatal closure response, two that were defective in apoplast defense, and four that were defective in both stomatal and apoplast defense. Isolation of these three classes of mutants suggests that stomatal and apoplastic defenses are integrated in plants, but are genetically separable, and that COR is important for Pst DC3000 to overcome both stomatal guard cell- and apoplastic mesophyll cell-based defenses. Of the six mutants defective in bacterium-triggered stomatal closure, three are defective in salicylic acid (SA)-induced stomatal closure, but exhibit normal stomatal closure in response to abscisic acid (ABA), and scord7 is compromised in both SA- and ABA-induced stomatal closure. We have cloned SCORD3, which is required for salicylic acid (SA) biosynthesis, and SCORD5, which encodes an ATP-binding cassette (ABC) protein, AtGCN20/AtABCF3, predicted to be involved in stress-associated protein translation control. Identification of SCORD5 begins to implicate an important role of stress-associated protein translation in stomatal guard cell signaling in response to microbe-associated molecular patterns and bacterial infection.
| Pathogen entry into host tissue is a critical first step in causing infection. For foliar bacterial plant pathogens, natural surface openings, such as stomata, are important entry sites into the leaf apoplast (internal intercellular spaces). Recent studies have shown that plants respond to surface-inoculated bacterial pathogens by reducing stomatal aperture as part of the innate immune response to restrict bacterial invasion. Once inside plant tissue, bacteria encounter defenses in the apoplast. To counter host defenses during invasion and in the apoplast, bacterial pathogens produce a variety of virulence factors, such as the polyketide toxin coronatine produced by Pseudomonas syringae pv. tomato (Pst) DC3000. Coronatine-deficient Pst DC3000 mutants are compromised in virulence, especially when inoculated onto the plant surface. In this study, we conducted a random genetic screen to identify Arabidopsis mutants that could rescue the virulence of coronatine-deficient mutant bacteria and obtained three classes of Arabidopsis mutants: those that are defective in stomatal closure only, those defective in apoplastic defense only, and those compromised in both stomatal closure and apoplastic defenses. The isolation of these host mutants highlight the important role of COR, a molecular mimic of the plant hormone jasmonate, in overcoming both stomatal and apoplastic defenses during Pst DC3000 infection.
| Pseudomonas syringae pv. tomato (Pst) strain DC3000 is a Gram-negative bacterium that infects tomato and Arabidopsis and is a model pathogen used to investigate molecular mechanisms underlying plant-pathogen interactions [1]–[3]. In nature, P. syringae strains often exhibit an epiphytic phase (i.e., surviving and/or multiplying on the plant surface) before entering the intercellular space of the host through wounds or natural openings such as stomata. Once inside the apoplastic space of a susceptible host plant, P. syringae can multiply to high levels within days, a process aided by many virulence factors including secreted phytotoxins and effector proteins delivered into the host cells through the type III secretion system (T3SS). Successful colonization of P. syringae in the apoplastic space eventually results in the development of disease symptoms, which usually include localized tissue necrosis and discoloration.
Coronatine (COR) is a non-host-specific phytotoxin produced by several pathovars of P. syringae, including Pst DC3000 [4]–[8]. COR has been implicated in the inhibition of stomatal closure to facilitate bacterial invasion, promotion of bacterial multiplication and persistence in planta, induction of disease symptoms, and enhancement of disease susceptibility in uninfected parts of the plant [7], [9]–[19]. COR shows remarkable structural similarity to the active form of the plant hormone jasmonoyl isoleucine (JA-Ile), and targets the JA receptor complex directly [20]–[22]. Indeed, COR serves as a potent inducer of JA-associated responses and expression of JA-responsive genes in plants [7], [12], [23], [24].
The molecular mechanism(s) by which COR-mediated activation of JA signaling results in various virulence roles of COR throughout the bacterial infection cycle remains to be determined. COR-deficient mutants of Pst DC3000 are able to multiply to high levels in salicylic acid (SA)-deficient Arabidopsis plants nahG and sid2, suggesting that COR may be required to overcome SA-mediated defenses [12], [14], [19]. Indeed, SA-dependent expression of pathogenesis-related (PR) genes is suppressed in a COR-dependent manner during Pst DC3000 infection in tomato plants [15], [25]. However, such COR-dependent suppression of PR gene expression has not yet been observed in Pst DC3000 infection of Arabidopsis plants [12]. Interestingly, recent studies show that COR can suppress plant defense responses triggered by microbe-associated molecular patterns (MAMPs). For example, COR inhibits MAMP- and bacterium-triggered stomatal closure, which is dependent on SA signaling [14], [19], and COR and JA suppress MAMP-induced callose deposition in leaf mesophyll cells and root cells [26], [27] in Arabidopsis. Taken together, these results collectively indicate that a major function of COR is to suppress interconnected MAMP- and SA-activated defense responses during bacterial infection.
Thus far, attempts to elucidate the role of COR in virulence have relied heavily on known plant defense signaling mutants. We reasoned that an unbiased genetic screen for plant mutants that affect the function of COR in the context of bacterial infection would be useful. Accordingly, we conducted a genetic screen aimed at isolating Arabidopsis mutants that would rescue the virulence of COR-deficient mutant bacterium Pst DC3118. This screen allowed us to identify eight susceptible to coronatine-deficient Pst DC3000 (scord) mutants. Six of them are defective in stomatal closure responses to Pst DC3118; the other two have a normal stomatal closure response to bacteria, but show weakened apoplastic resistance to both Pst DC3000 and Pst DC3118. Further characterization of the six stomatal closure mutants revealed defects in salicylic acid (SA) accumulation, SA-triggered stomatal closure, and/or abscisic acid (ABA)-induced stomatal closure. We cloned SCORD3, which is required for SA biosynthesis, and SCORD5, which encodes an ABC-transporter-family protein (AtGCN20/AtABCF3), predicted to be involved in stress-associated protein translation control. The identification of SCORD5 revealed potentially a new layer of regulation of MAMP/bacterium-triggered stomatal closure in Arabidopsis.
When inoculated onto leaves of wild-type Col-0 plants by dipping at 1×108 CFU/ml, COR-deficient mutants of Pst DC3000, such as Pst DC3118, do not cause significant disease symptoms or multiply to a high level [14], [19]. Pst DC3118 carries a Tn5 insertion in the cfa6 gene (Figure S1), which is essential for the biosynthesis of COR [5], [7] (Figure S1). However, Pst DC3118 exhibits aggressive growth and causes severe disease symptoms in Arabidopsis mutants that are defective in MAMP signaling, or in biosynthesis or signaling of ABA and SA [12], [14], [19]. For example, the ost1-2 mutant, defective in a guard cell/ABA signaling kinase, OST1 [28], [29], allows Pst DC3118 to multiply to high levels and cause prominent disease symptoms when surface-inoculated [14]. On the other hand, ost1-2 plants remained as resistant as wild-type Ler plants to the nonpathogenic hrcC mutant (defective in the T3SS) of Pst DC3000 (Figure S2), indicating that the greatly enhanced susceptibility of ost1-2 plants was specific to Pst DC3118.
For large-scale mutant screens we needed to grow plants in high density (approximately 20 to 30 plants per pot and 15 pots per flat). Accordingly, we tested whether ost1-2 mutant plants would still exhibit enhanced susceptibility to Pst DC3118 under this growth condition. Indeed, when dip-inoculated with Pst DC3118, ost1-2 plants allowed much higher bacterial multiplication and exhibited much more severe disease symptoms than wild-type Ler plants (Figure S2). In subsequent mutant screening, ost1-2 plants were included in each flat as a positive control.
About 14,000 activation-tagged T-DNA lines were screened by dip-inoculation with 1×108 CFU/ml of Pst DC3118. Plants that showed disease symptoms similar to ost1-2 plants were selected as putative mutants and allowed to set seeds. The progeny of each putative mutant were screened at least four more times in the following two generations. Here we report eight mutants that reproducibly showed enhanced disease symptoms, and are named susceptible to coronatine-deficient Pst DC3000 (scord). Pst DC3118 populations in these mutants were almost 100 times higher than in wild-type plants when dip-inoculated, except for the scord7 mutant, in which Pst DC3118 population was about 10 times higher (Figure 1A, Table 1).
As the scord mutants may be affected in different aspects of COR action during Pst DC3000 infection, we first determined whether some of them are defective in the bacterium-induced stomatal closure response. When subjected to leaf peel assays with suspensions of Pst DC3118 [14], [19], most stomata of wild-type Col-7 plants closed and resulted in a significant reduction of the mean aperture (Figure 1B). However, the stomata of six scord mutants—scord1, -3, -5, -6, -7, and -8—did not show significant closure, indicating loss of normal stomatal closure response to Pst DC3118 (Figure 1B). The other two mutants, scord2 and scord4, maintained normal stomatal closure responses to bacteria, at least within the time frame of our assays (i.e., 1 to 1.5 h after incubation) (Figure 1B).
Since increased plant susceptibility in dip-inoculation experiments could also be caused by a deficiency in apoplastic defense after bacterial invasion, we next tested scord mutants by infiltration inoculation with Pst DC3118 (Figure 1C). Of the eight scord mutants, scord1, -3, -6, and -8 plants showed a much higher apoplastic susceptibility to Pst DC3118, and contained 50 to 100 times more bacteria 3 days after inoculation, compared to Col-7 plants. scord2 and scord4 also showed a higher susceptibility to Pst DC3118 in infiltration experiments, harboring about 10 times more bacteria than Col-7 plants at day 3 (Figure 1C). In contrast, scord5 and scord7 did not show significantly higher susceptibility to Pst DC3118 in infiltration experiments compared to Col-7 plants (Figure 1C). Although statistically (Student's t-test) not significant, we have consistently observed in all repeats that Pst DC3118 growth is actually less in scord7 plants than in Col-7 plants. Overall, these results suggest that scord1, -3, -6, and -8 are deficient not only in stomatal defense but also in apoplastic defense. On the other hand, scord5 and scord7 are deficient in stomatal defense, and scord2 and scord4 are affected in apoplastic defense.
Next, we tested the susceptibility of the scord mutants to wild type Pst DC3000 by dipping inoculation. We found that scord1, -2, -3, -4, -6, and -8 mutants were also more susceptible to Pst DC3000 than Col-7 plants (Figure 1D). In contrast, scord5 and scord7 were not hyper-susceptible to Pst DC3000 compared to Col-7 plants. Again, although statistically not significant, Pst DC3000 growth was consistently less in scord7 plants than in Col-7 plants.
Finally, we examined the susceptibility of these eight scord mutants to the nonpathogenic Pst DC3000 hrcC mutant by dipping inoculation. None of the eight scord mutants allowed significant growth of Pst DC3000 hrcC (Figure S3). Thus, scord mutants maintained normal resistance to nonpathogenic bacteria.
We have previously shown that SA plays a critical role in bacterium-induced stomatal closure response [14], [19]. To position the scord mutants in relation to SA, we tested SA-induced stomatal closure in the scord mutants (Figure 2). In leaf peel assays, we found that scord6, -7 and -8 mutants are defective in SA-induced stomatal closure, but scord1, -2, -3, -4 and -5 had normal stomatal closure response, similar to wild type Col-7 plants.
We conducted further experiments to examine whether any of the scord mutants were affected in SA accumulation. We found that scord1, -3, -6, and -8 plants all had significantly lower basal levels of SA, containing only 10–30% of what was detected in wild-type Col-7 plants (Figure 3A). To the contrary, SA levels in scord2, -4, -5, and -7 plants were similar to those in wild-type Col-7 plants (Figure 3A).
We also examined scord mutants for SA levels 12 hours after infiltration with Pst DC3000 at 1×108 CFU/cm (Figure 3B). Similar to what was observed without bacterium challenge, the SA levels in scord1, -3, -6, and -8 plants were reduced, whereas those of scord2, -4, -5, and -7 plant were similar to those in Col-7 plants (Figure 3B).
Besides SA, we have previously shown that ABA biosynthesis and signaling are also critical for the stomatal closure response to bacteria or MAMPs [14], [19]. To position the six stomata-closure-defective scord mutations in relation to ABA, we examined the stomatal response of scord1, -3, -5, -6, -7, and -8 mutants to ABA. We found that only scord7 stomata failed to close in response to ABA (Figure 4A), whereas stomata of Col-7 and the other five scord mutants (scord1, -3, -5, -6, and -8) showed normal closure responses to ABA (Figure 4A), suggesting that the scord7 mutant might be affected in ABA response or a general step in stomatal closure that is common to multiple stomatal closure pathways. ABA response mutants often show hypersensitivity to drought stress. Indeed, when subjected to water withholding, scord7 plants wilted faster than wild-type Col-7 plants (Figure 4B). When detached leaves were tested for water loss, leaves from scord7 plants lost water at a much faster rate than those from Col-7 wild type plants (Figure 4C). The scord7 mutant, however, showed a similar ABA inhibition response as wild type Col-7 in seed germination assay (Figure 4D). We also tested scord7 plants for dark-induced stomatal response, which is thought to be independent of ABA signaling [30]. The scord7 plants were defective in dark-induced stomatal closure (Figure 4E). Thus, SCORD7 is involved either in a guard cell-specific ABA response pathway or in a later step shared by SA-, ABA- and dark signaling pathways leading to stomatal closure.
In addition to the phenotypes characterized above, we also noticed morphological phenotypes for some of the scord mutants. Among the eight mutants, the appearances of scord1, -2, -3, and -7 plants are similar to wild type Col-7 plants (Figure 5; Table 1). On the other hand, compared to Col-7 plants, scord5 and 8 plants have slightly smaller rosettes, scord5 plants have pale green leaves, scord4 and 8 plants have more serrated leaf edges, and scord6 has smoother leaf edges (Figure 5A). We also examined the stomata density and stomata morphology in these mutants, and did not notice any dramatic difference on stomata density (Figure 5B), except that scord6 stomata appear to be larger and lack the central ridges (Figure 5C). At present, it is not known whether these morphological phenotypes are caused by the same mutations that lead to defects in stomatal closure and/or bacterial susceptibility phenotypes.
We attempted plasmid rescue experiments with all scord mutants. However, we were able to correlate a T-DNA insertion with mutant phenotypes in only the scord3 and scord5 mutants. In the scord3 mutant, the T-DNA insertion is located in the first intron of At4g39030 (Figure S4), which encodes EDS5, a putative transporter protein required for SA biosynthesis [31]. Similar to the scord3 mutant, the eds5-1 mutant showed a defect in SA accumulation and stomatal response (Figure S5). Furthermore, allelism tests showed that scord3 and eds5-1 have mutations in the same gene (Figure S6). In the literature, three alleles of eds5 have been characterized [31]. We therefore have named scord3 eds5-4.
As described above, in addition to the scord3/eds5-4 mutant, the scord1 mutant also contained little SA before or after bacteria challenge (Figure 2). Like the scord3/eds5-4 mutant, the scord1mutant maintained a normal SA-induced stomatal closure response, suggesting that the scord1 mutant is deficient in the synthesis of SA, but is normal in SA signaling. We therefore examined whether the scord1 mutant also contains a mutation in the EDS5 gene. We amplified and sequenced the genomic region of EDS5 from scord1 plants. However, we did not find any mutation in EDS5 (data not shown). Therefore, the scord1 mutant likely has a mutation in a gene, other than EDS5, that is involved in SA biosynthesis.
The T-DNA insertion site in the scord5 mutant lies in the 5′UTR region of At1g64550 (108 bps upstream of the start codon ATG; Figure 6A). The existence of this T-DNA insertion in the scord5 genome was confirmed by PCR using At1g64550-specific and T-DNA-specific primers (Figure 6B). Furthermore, the transcript of At1g64550 was not detectable in scord5 plants (Figure 6C). To confirm that the phenotypes observed in scord5 plants were due to the T-DNA insertion in At1g64550, we identified an independent T-DNA insertion line, 188G03, in the GABI-Kat collection [32]. Genomic PCR confirmed the presence of a T-DNA insertion in the eighth exon of At1g64550 (Figure 6A, D). RT-PCR using primers near the insertion site showed that 188G03 is also a knockout line for At1g64550 (Figure 6E). Like the scord5 mutant, 188G03 was also impaired in stomatal closure response to Pst DC3118 (Figure 6F). Based on these molecular and phenotypic analyses, we have designated the GABI-Kat line 188G03 as scord5-2 and the original scord5 line as scord5-1.
To complement the phenotypes of scord5 mutants, a 5.4-kb genomic fragment that includes a 480-bp region upstream of the start codon and a 400-bp region downstream of the stop codon of SCORD5 (At1g64550) was amplified using Col-7 genomic DNA. This SCORD5 fragment was cloned into the plant transformation vector pCAMBIA1300 and introduced into both scord5-1 and scord5-2 mutant plants via Agrobacterium-mediated transformation. Examination of four randomly chosen, independent T1 transformants showed that stomatal closure response to Pst DC3118 was restored in both scord5-1 (gSCORD5) and scord5-2 (gSCORD5) lines (Figure 6G), providing further evidence that SCORD5 is required for bacterium-triggered stomatal closure.
Sequence analysis showed that SCORD5 encodes a gene that has been annotated as AtGCN20/AtABCF3, which belongs to the F subfamily of ATP-binding cassette (ABC) proteins. AtGCN20/AtABCF3 is a 4.8-kb gene with 17 introns, and encodes a protein of 715 amino acids. AtGCN20/AtABCF3 (SCORD5 hereinafter) is predicted to have only nucleotide-binding domains (NBDs), lacking transmembrane domains that are characteristic of the AtABCA, -B, -C, -D, and -G protein subfamilies [33]–[35].
Because T-DNA insertions in other scord mutants (i.e., other than scord 3 and scord5) either could not be amplified, or were amplified but not linked to Pst DC3118-hypersusceptibility phenotypes (data not shown), we have initiated physical mapping of two of these scord mutations: scord6 and scord7. To this end, we have narrowed the scord6 mutation to a 190-kb region on chromosome III, and the scord7 mutation to a 240-kb region on chromosome V (Table 1). Neither region contains genes that are known to be involved in MAMP, SA or ABA signaling or in stomatal closure response. Therefore, SCORD6 and SCORD7 likely encode new regulators of stomatal closure response.
In summary, our molecular and phenotype characterization of the eight scord mutants suggests that scord3, -5, -6, and -7 represent mutations in four different genes (i.e., EDS5, AtGCN20/AtABCF3, SCORD6, and SCORD7). Although we cannot ascertain the genetic relationships among the other four scord mutations (scord1, -2, -4, and -8), the distinct phenotypes (stomatal closure response, apoplast defense, SA levels, and morphology; Table 1) exhibited in the corresponding mutants suggest that at least some of them would be affected in additional genes.
Because SCORD5 has never been implicated in stomatal closure or plant defense response, we focused on this gene for further analysis. We recently showed that the signaling cascade for bacterium/MAMP-induced stomata closure involves MAMP signaling, followed by SA and ABA signaling [19], [36]. As stomata of the scord5 mutant responded normally to SA and ABA (Figures 2, 4), we investigated a possible defect in MAMP response using flg22 (a bioactive 22-aa peptide from bacterial flagellin). We found flg22 triggered stomatal closure in Col-7 plants, but not in scord5 plants (Figure 7A). Similarly, elf18 (a bioactive 18-aa peptide from E. coli EF-Tu) induced stomatal closure in Col-7 plants, but not in scord5 plants (Figure S7). These results are consistent with the observed defect in stomatal closure response to Pst DC3118 bacteria (Figure 1B), which are expected to produce flagellin and possibly other MAMPs during infection [19].
Stomatal closure is only one of several functional outputs induced by MAMPs. To determine whether other responses induced by MAMPs are also affected in the scord5 mutant, we first examined the production of reactive oxygen species and deposition of callose in scord5 leaves in response to flg22. The production of H2O2 after flg22 treatment (100 nM) was similar in scord5-1 and Col-7 plants (Figure 7B). Similarly, the amounts of callose deposition induced by infiltration of flg22 (10 nM) were comparable in scord5-1 and Col-7 leaves (Figure 7C). Next, we studied the effect of flg22 pretreatment on the induced resistance against Pst DC3000 (i.e., flg22 protection assay) in scord5-1 and Col-7 plants. One day after spraying scord5-1 and Col-7 plants with 3 µM flg22, we inoculated plants with Pst DC3000 by dipping at 1×108CFU/ml (Figure 7D). We noticed that scord5 plants were more susceptible than Col-7 to Pst DC3000 in the H2O treatment control, and allowed more Pst DC3000 multiplication, compared to Col-7 after flg22 pretreatment (Figure 7D). Nonetheless, the flg22 treatment still induced resistance in scord5 plants, albeit to a smaller degree compared to that in Col-7 plants. To find out whether the higher susceptibility of the scord5 mutant shown in the flg22 protection assay in dipping inoculation was caused by reduced apoplastic defense, we repeated the flg22 protection experiment by infiltrating Pst DC3000 (1×106 CFU/ml) directly into the apoplast. We found that the multiplication of Pst DC3000 was suppressed to similar levels in scord5-1 and Col-7 leaves (Figure 7E). These experiments suggest that flg22-induced apoplastic defense is normal in the scord5 mutant and that the partial loss of flg22-induced resistance in scord5 plants is likely caused by a defect in bacterium-triggered stomatal closure.
We conducted further experiments to increase our understanding of the involvement of the SCORD5-associated pathway in regulating bacterium-triggered stomatal closure. SCORD5 shares significant sequence similarities with yeast (46% identity and 66% similarity) and mammalian (41% identity and 61% similarity) GCN20 proteins (EDN59158 and NP_038880, respectively), which, together with GCN1, regulates stress-associated protein translation [37]–[39]. Arabidopsis has a putative GCN1 orthologue named ILITYHIA (ILA; At1g64790), which was shown recently to be involved in defense against Pst DC3000 [40]. Arabidopsis ILA shares 35% identity and 54% similarity with rat GCN1 (NP_001162135) and 34% identity and 54% similarity with yeast GCN1 (EEU04663), but shares no sequence similarity with SCORD5 (AtGCN20/AtABCF3). We examined the possibility that ILA may also be required for bacterium-triggered stomatal closure. Indeed, similar to the scord5 mutant, stomata of ila-3 mutant were completely unresponsive to Pst DC3118 (Figure 8A). This result provides further evidence that bacterium-triggered stomatal closure is likely regulated by an authentic GCN1/GCN20-associated protein translation control mechanism in plants.
In the Arabidopsis-Pst DC3000 interaction, COR has been implicated in several aspects of pathogenesis. To date, however, a random genetic screen for host mutants that rescue the virulence of COR-deficient Pst mutants has not been reported. In principle, such a genetic screen not only could identify new alleles of host genes known to be involved in COR-dependent interactions during infection, but also has the potential to uncover novel components in the host that are required for the various virulence functions of COR. Indeed, from an initial screen of ∼14,000 T-DNA insertion lines, we were able to isolate eight scord mutants. Further detailed characterization of these mutants uncovered both known (e.g., EDS5) and novel (e.g., AtGCN20/AtABCF3, and possibly SCORD6 and SCORD7) regulators of stomatal and/or apoplastic defenses, illustrating the exciting prospect of future large-scale Pst DC3118-based genetic screens in the identification of cell-type-specific defense mutants in plants.
Stomatal closure response has been shown to be an integral part of Arabidopsis resistance to COR-deficient mutants of Pst DC3000, and SA is necessary for this response [14], [19]. Indeed, of the eight scord mutants isolated in this study, six are defective in the stomatal closure response to Pst DC3118 (Figure 1B). Among them, both basal and bacterium-induced levels of SA are significantly lower in scord1, -3, -6 and -8 plants than those in wild-type plants (Figure 3), indicating that these four mutants have defects in the accumulation of SA to various degrees. The scord7 mutant is defective in stomatal closure response to SA, ABA and dark, and is hypersensitive to drought (Figure 4). These results point to a possible deficiency in a step common to multiple signaling pathways leading to stomatal closure in the scord7 mutant plant.
SA is also known to be essential for apoplast defense, which can be revealed using infiltration inoculation. In our screen, SA-compromised scord1,-3, -6 and -8 mutants allowed more bacterial growth, compared with wild-type plants, when Pst DC3118 was infiltrated into the apoplast directly (Figure 1C), supporting the importance of SA in mediating the apoplastic defense against Pst DC3118. In contrast, although the scord7 mutant is more susceptible than Col-7 plants to surface-inoculated DC3118, this mutant showed slightly more resistance to Pst DC3118 in infiltration experiments (Figure 1C). This is a particularly interesting result because ABA mutants have been observed to be more resistant to Pst DC3000 in infiltration experiments [41]. The enhanced apoplastic resistance of the scord7 mutant to Pst DC3118 therefore could be related to the inability of scord7 stomata to respond to ABA. However, considering that scord7 stomata are not responsive to SA and dark, we cannot rule out the possibility that the apoplast of the scord7 mutant has elevated resistance due to an ABA signaling-independent mechanism. Regardless, the overall higher level of susceptibility of the scord7 mutant to surface-inoculated Pst DC3118 can now be explained by a combined effect of compromised stomatal defense and enhanced apoplastic defense, with the stomatal closure defect overpowering the enhanced apoplastic defense in this mutant. We noted that the overall enhancement of Pst DC3118 growth in the scord7 mutant when inoculated by dipping was less than in other scord mutants (Figure 1A), which is consistent with the observation that the apoplast of other scord mutants was not more resistant to Pst DC3118, as revealed in the infiltration experiments (Figure 1C).
It is interesting that, although the scord5 and scord7 mutants are much more susceptible to surface-inoculated Pst DC3118, compared with Col-7 plants, they are not more susceptible to surface-inoculated Pst DC3000. Thus, the disease hyper-susceptibility of the scord5 and scord7 mutants is dependent on COR production in bacteria. This finding is consistent with the observations that stomatal defense in wild-type Col-7 plants presents a significant barrier to infection by Pst DC3118 and that the scord5 and scord7 mutants are defective in stomatal defense, thereby allowing Pst DC3118 to infect from the leaf surface. On the other hand, Pst DC3000 can infect not only scord5 and scord7 mutants, but also wild-type Col-7 plants because of its ability to produce COR, which counteracts stomatal closure in wild-type Col-7 plants. COR is not necessary for Pst DC3000 to infect scord5 and scord7 mutants, in which stomatal closure is already compromised.
In this study, we also identified two mutants, scord2 and scord4, that showed an apparently normal stomatal closure response, but compromised apoplastic defense (Figure 1). Interestingly, in contrast to scord5 and scord7 mutants, scord2 and scord4 mutants were hyper-susceptible to both Pst DC3118 and Pst DC3000, suggesting that these two mutants have a general defect in the apoplast (defense or nutrition) and that the hyper-susceptibility phenotype is not dependent on COR production in bacteria. The ability of these mutants to rescue the virulence of Pst DC3118 suggests that a hyper-susceptible apoplast can compensate, at least genetically, for the reduced ability of Pst DC3118 to overcome stomatal defense. We hypothesize that these apoplast defense mutants are likely affected in cellular processes other than the biosynthesis or signaling of SA and ABA, because defects in SA or ABA biosynthesis or signaling cause altered stomatal closure in response to bacteria [14], [19] and would have been detected in this study. Accordingly, further characterization of the scord2 and scord4 mutants has potential to yield novel genes/pathways that control apoplast defense (or nutrition) in Arabidopsis.
We were successful in cloning SCORD3 and SCORD5. Whereas SCORD3 turns out to be EDS5, which is known to be involved in plant defense, SCORD5 is a new component required for bacterium-triggered stomata closure response. The scord5 mutation affected mainly MAMP-induced stomatal closure, but not SA- or ABA-induced stomatal closure, suggesting that SCORD5 likely acts early in the stomatal closure response pathway. Although we could not detect a significant defect in apoplastic defense in the scord5 mutant, the ila-3 mutant has been shown to be more susceptible to Pst DC3000 compared with Col-0 plants by infiltration inoculation, suggesting a defect in apoplastic defense [40]. We noticed that the ila-3 plants are slightly smaller than scord5 plants (Figure 8B). Because SCORD5 belongs to a five-member gene family (Figure S8), whereas there is only a single ILA gene in Arabidopsis, we speculate that some of the SCORD5-family genes may have partially redundant functions and that one or more of SCORD5-like genes may compensate for the loss of SCORD5 in the mesophyll cells. Future research should determine whether higher-order mutants of the SCORD5 family genes would phenocopy the pleiotropic defects in both stomatal and apoplastic defenses in the ila-3 mutant.
Overall, the genetic screen reported in this paper illustrates an unbiased and useful approach in the understanding of Arabidopsis signaling pathways that govern cell-type-specific (i.e., stomatal and mesophyll) responses to Pst DC3000 infection, as well as the molecular action of COR during various stages of Pst DC3000 infection of Arabidopsis.
Unless specified, all experiments presented were repeated three times and statistical differences were detected with a two-tailed T-test (*, P<0.05; **, P<0.01; ***, P<0.001).
Arabidopsis plants were grown in soil under a 12/12 h photoperiod at 110 µmol m−2 s−1 at 22°C. Seeds of the activation-tagged T-DNA lines (CS21995, CS23153) were purchased from the Arabidopsis Biological Resource Center with Col-7 accession as the wild-type parent [42]. For ABA inhibition of germination, seeds are sterilized in 30% bleach for 15 min and washed in sterilized ddH2O for 5 times before put on MS (1% sucrose) plates. Ninety seeds were used for each treatment. Plates were left at 4°C for two days before placed in growth chambers, and the numbers of seeds germinated were recorded two days later. For stomata counting, three leaves were chosen from 5-week old plants. A middle section of each leaf was cut out and mounted in distilled water, and visualized with bright field microscopy. From each section, three areas of 0.45 µm2 were chosen for observation. The chosen areas were scanned to a depth of 40–60 µm with a Zeiss 510 Meta ConfoCor3 laser scanning confocal microscope (Carl Zeiss MicroImaging, Thornwood, NY) to generate Z-scans. Zeiss LSM software was used to assemble the z-series into stacked images representing the leaf surface for stomata counting.
Stomatal assays followed procedures described in a previous study [19]. Leaf peels from the abaxial side were collected from mature leaves of 5-week-old plants and placed in 250–300 µl of ddH2O or bacteria resuspended in ddH2O (1×108 CFU/ml), buffer (25 mM MES, 10 mM KCl, pH 6.15) or buffer containing SA or ABA (Sigma, St. Louis, MO, USA), flg22 or elf18 (EZBiolab, Westfield, IN, USA) on glass slides in square Petri dishes with lids on. To preserve the stomatal aperture status in plants used for both stomatal and bacterial pathogenesis assays, we did not further treat leaf peels in any stomatal opening buffer. The Petri dishes were left for an hour in the growth chamber in which plants were grown before being viewed under the microscope. Leaf peels on slides were observed under a light microscope and images were randomly taken to ensure that at least 30 stomata were recorded for each sample treatment. Images were opened in Adobe Photoshop for measurement of stomatal apertures. Bacteria were used at 1×108 CFU/ml (OD600 = 0.2). Stomatal apertures shown in results are means and standard errors of 30–60 measurements for each experiment.
A 10-ml low-salt LB liquid culture was started using bacteria (Pst DC3118) scraped from a plate that had been stored at −4°C for less than two weeks. After 12 h at 28°C, a larger subculture was started using 1∶100 dilution at 28°C. Bacterial cells were harvested when the OD600 of the culture reached 0.8 to 1.0. Bacteria were resuspended in ddH2O to an OD600 of 0.2 (1×108 CFU/ml). Silwet L-77 was added to a final concentration of 0.02 to 0.05%. For DC3118 dipping screening, four-week-old plants grown in meshed pots, at a density of about 20 seedlings per pot, were dipped upside down in the bacterial solution for a few seconds to coat all leaves uniformly with bacterial suspension. Plants were returned to a growth room in a tray covered with a plastic lid. At day 3 post-inoculation (3 dpi), plants showing no disease symptoms were removed from the pot. Diseased plants were allowed to set seeds. Seeds from the putative mutant plants were sown and the screening process was repeated. For other bacterial infection assays via dipping or infiltration inoculation, disease symptoms were recorded by camera and bacterial populations were monitored by serial-dilution assays according to Katagiri et al. [3]. All bacteria growth results are recorded as means of four different leaves from four different plants, with standard errors indicated.
Five-week-old Arabidopsis plants were sprayed with H2O or 3 µM flg22 (EZBiolab, Carmel, IN) 24 h before bacterial inoculation. The flg22-treated plants were inoculated with Pst DC3000 by dipping at 1×108 CFU/ml or infiltration at 1×106 CFU/ml. Two days after bacterium inoculation, leaf samples were collected for bacterial enumeration.
The plasmid rescue procedure was adapted from Weigel et al. [42]. Genomic DNA was prepared from 1 g (fresh weight) leaf tissue of 5-week-old plants using the Nucleon Phytopure Plant DNA Extraction Kit (Amersham Biosciences, Piscataway, NJ). One to five µg of purified genomic DNA was digested in a 120-µl reaction mixture overnight with restriction enzymes BamHI, SpeI, and NotI for the left border and KpnI, EcoRI, and HindIII for the right border of the T-DNA [42]. After digestion, ddH2O was added in the reaction mixture to bring the volume to 500 µl, followed with phenol-chloroform extraction and then chloroform extraction. Digested DNA was precipitated with 2 volumes of ethanol and 1/10 volume of sodium acetate (3 M, pH 5.2), and washed twice with 70% ethanol. Dried DNA was resuspended in 100 µl ddH2O, then used in a ligation reaction of 120 µl with 20–40 U of T4 DNA ligase (NEB, Beverly, MA) overnight at 16°C. DNA from the ligation mixture was precipitated in the same way as before and resuspended in 10 µl ddH2O. One to two µl of the ligation mixture was used for transformation into the SURE Electroporation-Competent Cells (Stratagene, La Jolla, CA). Plasmids from the Apr colonies were digested with the same restriction enzymes used for plasmid rescue reactions, and the inserts were subjected to sequencing using primers T3 (5′-ATTAACCCTCACTAAAGGGA-3′), T7 (5′-TAATACGACTCACTATAGGG-3′), T-1 (5′-AAGTGCAGGTCAAACCTTGAC-3′), T-3 (5′-GGTAATTACTCTTTCTTTTCCTC-3′), or sfEcoRI (5′-AGCCTTGCTTCCTATTATATCT-3′). Primers used to confirm the T-DNA insertion in scord3/eds5-4 plants also include 108 (5′-AAGGAAGCTCCATCGAACT-3′) and 109 (5′-TGTTTGGCAAGAGAAGTAGCA-3′).
Genomic DNA was prepared the same way as described for plasmid rescue. One µg of purified genomic DNA was digested in a 100-µl reaction mixture overnight with the restriction enzyme EcoRV, HindIII, DraI, or EcoRI. After digestion, ddH2O was added to the reaction mixture to bring the volume to 500 µl, and the DNA mixtures were purified using a QIAquick PCR purification kit (Qiagen, Valencia, CA) and eluted in 50 µl ddH2O. Ten µl of the purified DNA mixture was incubated overnight at 16°C in a ligation reaction of 100 µl with 10 to 30 U of T4 DNA ligase (NEB, Beverly, MA). Five µl of the ligation mixture was used in each PCR reaction, along with the following primers specific to the T-DNA [42]: iPCR1 (5′-TGGATCTCAACAGCGGTAAGA-3′), iPCR2 (5′-TTCGACGTGTCTACATTCACG-3′), iPCR3 (5′-TCGGTGTGTCGTAGATACTAG-3′), iPCR4 (5′-TGGTTGACGATGGTGCAGAC-3′), iPCR1g (5′-TGACCATCATACTCATTGCTGATCC-3′), iPCR2g (5′-CGATCCGTCGTATTTATAGGCGAAAG-3′), T-1 (5′-AAGTGCAGGTCAAACCTTGAC-3′), T-3 (5′-GGTAATTACTCTTTCTTTTCCTC-3′), and T7 (5′-TAATACGACTCACTATAGGG-3′). PCR products obtained were purified from agarose gels after electrophoresis and sequenced. The T-DNA flanking sequence was identified from a PCR fragment amplified using primers T-3 and iPCR3 and DraI-digested scord5 genomic DNA as template.
The primers used for PCR to confirm the T-DNA insertions in scord5-1 and scord5-2 plants, and for RT-PCR to confirm lack of expression of SCORD5 in the mutant plants were as follows: 101 (5′-CAAGTTATAGATAATATGAGTTTGT-3′), 102 (5′-GAATCTCGTCGCTTCTGTGTT-3′), 103 (5′-AACTTATCCGACAGTTTGCTAC-3′), 104 (5′-GTGGTCACGGACATCATCCAT-3′), 105 (5′-CTGCATATGTGACCGTGATCG-3′), 106 (5′-CAAACCAGAGACGAGTTGGAAACAG-3′), 107 (5′-CAGGTTTCTGACATTCTTTTCCTC-3′), T-1 (5′-AAGTGCAGGTCAAACCTTGAC-3′), T-2 (5′-ATATTGACCATCATACTCATTGC-3′), ACT1 (5′-GGTCGTACTACCGGTATTGTGCT-3′; 5′-TGACAATTTCACGCTCTGCTGTG-3′).
The 5.4-kb genomic fragment of SCORD5/At1g64550 was amplified using the genomic DNA of Col-7 as template. This fragment contains 480 bps upstream of the start codon ATG with an NheI restriction site attached (5′-CTGCTAGCCAATCTAAGCTCTCTTC-3′) and 400 bps downstream of the stop codon TAA with a SalI restriction site attached (5′-TGGGTCGACGGTGCTCGATTCGAA-3′). The fragment was fully sequenced before being cloned into pCambia1300 at the XbaI and SalI sites. The clone was transferred into GV3101 (pMP90) for plant transformation using the flower-dipping protocol. Independent T1 plants were identified based on hygromycin resistance and analyzed for complementation of the scord5-1 and 188G03 (scord5-2) mutations.
The site of the Tn5 insertion in Pst DC3118 was identified using iPCR performed according to Huang et al. [43]. Single colonies of Pst DC3000 and Pst DC3118 were grown overnight in 4 ml of low-salt LB culture media with appropriate antibiotics (Pst DC3000: Rif 100 µg/ml; Pst DC3118: Rif 100 µg/ml, Km 25 µg/ml). Overnight cultures were centrifuged (5 min at 2000×g) and cell pellets were resuspended in 400 µl of TES lysis solution (50 mM Tris-HCl, pH 7.5, 10 mM NaCl, 10 mM EDTA) supplemented with Sarcosyl (final concentration of 1%) and proteinase K (100 µg/ml). Lysates became clear after 45 min, whereupon 225 µl of ice cold NH4OAc (7.5 M) was added. Genomic DNA was extracted from lysates and purified using a conventional phenol chloroform extraction protocol [44], resuspended in 10 mM Tris-HCl (pH 8.0) and stored at −20°C. One µg of purified genomic DNA was digested using 20 U of restriction enzyme BamHI (New England Biolabs) for 16 h at 37°C and concentrated by ethanol precipitation before self-ligation using T4 DNA ligase (Promega) at 16°C overnight. The ligation product was used as the template for PCR with primers BR/IR0 and BL/IR0 according to manufacturer's directions (PfuUltra II Fusion HS DNA Polymerase, Agilent Technologies) and was isolated by gel electrophoresis (Figure S1). The band was purified using the QIAquick Gel Extraction Kit (Qiagen) and sequenced using the primers IR0 and BL. Amplicon sequencing indicated that the genomic region flanking the Tn5 insertion corresponds to cfa6. We confirmed the Tn5 insertion locus to be in cfa6 by PCR amplification of Pst DC3118 gDNA using Tn5- and cfa6-specific primer sets (Figure S1) and amplicon sequencing. The sequences of primers used here are: IR0 (5′-GCCGAAGAGAACACAGATTTAGC-3′), BL (5′-GGGGACCTTGCACAGATAGC-3′), BR (5′-CATTCCTGTAGCGGATGGAGATC-3′), cfa6F (5′-AGTCATGGACGGACAGGTTC-3′), and cfa6R (5′-CCAAGCTCTACGATTCCGAG-3′).
Leaves from 4-week-old plants were infiltrated with water or 10 nM flg22 using a needleless syringe. After 24 h, about eight leaves from at least five independent plants were cleared and dehydrated with 100% ethanol. Leaves were fixed in an acetic acid∶ethanol (1∶3) solution for 2 h, sequentially incubated for 15 min in 75% ethanol, 50% ethanol, and 150 mM phosphate buffer, pH 8.0, and then stained overnight at 4°C in 150 mM phosphate buffer (pH 8.0) containing 0.01% (w/v) aniline blue supplemented with carbinicillin (100 µg/ml). After staining, leaves were mounted in 50% glycerol and examined by UV epifluorescence using an Axio Image M1 microscope (Zeiss). Callose quantification was performed using Image J software.
Active oxygen species released by seedlings were assayed by H2O2-dependent luminescence of luminol [45]. Two-week-old seedlings were transferred from liquid culture into 96-well plates and incubated overnight in 200 µl H2O supplemented with carbinicillin (100 µg/ml) at room temperature. The next morning, 100 µl H2O containing 20 µM luminol and 1 µg horseradish peroxidase (Sigma) was added and luminescence was measured for 20 min after the addition of the test solutions (100 nM flg22). Luminescence was detected using a Spectra Max L plate reader (Molecular Devices, Sunnyvale, CA). Seedlings were weighed to normalize results.
Approximately 100–300 mg (fresh weight) of leaf tissues were frozen in liquid nitrogen, ground, and extracted with 1 ml methanol∶water (1∶1 v/v) containing 0.1% formic acid and 0.1 g L−1 butylated hydroxytoluene (BHT) at 4°C for 24 h. Homogenates were mixed and centrifuged at 12,000×g for 10 min at 4°C. Supernatants were filtered through 0.2 µm PTFE membrane (Millipore, Bedford, MA) and transferred to autosampler vials. Injections of plant extracts (10 µl per injection) were separated on a fused core Ascentis Express C18 column (2.1×50 mm, 2.7 µm; Supelco, Bellefonte, PA) installed in the column heater of an Acquity Ultra Performance Liquid Chromatography (UPLC) system (Waters Corporation, Milford, MA). A gradient of 0.15% aqueous formic acid (solvent A) and methanol (solvent B) was applied in a 5-min program with a mobile phase flow rate of 0.4 ml/minute. The separation consisted of a linear increase from A∶B (9∶1, v/v) to 100% B. The column, which was maintained at 50°C, was interfaced to a Quattro Premier XE tandem quadrupole mass spectrometer (Waters Corporation, Milford, MA) equipped with electrospray ionization and operated in negative ion mode. The capillary voltage, cone voltage, and extractor voltage were set at 3 kV, 30 V, and 3V, respectively. The flow rates of cone gas and desolvation gas were 100 and 800 L h−1, respectively. The source temperature was 120°C and the desolvation temperature was 350°C. Propyl 4-hydroxybenzoate was added as the internal standard for quantification of SA. Transitions from deprotonated molecules to characteristic product ions were monitored for SA (m/z 137>93) and propyl 4-hydroxybenzoate (179>93). Collision energies and source cone potentials were optimized for each transition using QuanOptimize software. Peak areas were integrated, and the analytes were quantified based on standard curves generated from peak area ratios of analytes related to the corresponding internal standard. Data acquisition and processing were performed using Masslynx 4.1 software (Waters, Milford, MA).
Sequence data for genes and proteins described in this article can be found in The Arabidopsis Information Resource (TAIR, http://www.arabidopsis.org/) under the following ID numbers: ACT1 (Arabidopsis actin 1): At2g37620; OST1: At4g33950; SCORD3/EDS5: At4g39030; SCORD5/AtGCN20/AtABCF3: At1g64550.
|
10.1371/journal.pgen.1003289 | The Flowering Repressor SVP Underlies a Novel Arabidopsis thaliana QTL Interacting with the Genetic Background | The timing of flowering initiation is a fundamental trait for the adaptation of annual plants to different environments. Large amounts of intraspecific quantitative variation have been described for it among natural accessions of many species, but the molecular and evolutionary mechanisms underlying this genetic variation are mainly being determined in the model plant Arabidopsis thaliana. To find novel A. thaliana flowering QTL, we developed introgression lines from the Japanese accession Fuk, which was selected based on the substantial transgression observed in an F2 population with the reference strain Ler. Analysis of an early flowering line carrying a single Fuk introgression identified Flowering Arabidopsis QTL1 (FAQ1). We fine-mapped FAQ1 in an 11 kb genomic region containing the MADS transcription factor gene SHORT VEGETATIVE PHASE (SVP). Complementation of the early flowering phenotype of FAQ1-Fuk with a SVP-Ler transgen demonstrated that FAQ1 is SVP. We further proved by directed mutagenesis and transgenesis that a single amino acid substitution in SVP causes the loss-of-function and early flowering of Fuk allele. Analysis of a worldwide collection of accessions detected FAQ1/SVP-Fuk allele only in Asia, with the highest frequency appearing in Japan, where we could also detect a potential ancestral genotype of FAQ1/SVP-Fuk. In addition, we evaluated allelic and epistatic interactions of SVP natural alleles by analysing more than one hundred transgenic lines carrying Ler or Fuk SVP alleles in five genetic backgrounds. Quantitative analyses of these lines showed that FAQ1/SVP effects vary from large to small depending on the genetic background. These results support that the flowering repressor SVP has been recently selected in A. thaliana as a target for early flowering, and evidence the relevance of genetic interactions for the intraspecific evolution of FAQ1/SVP and flowering time.
| In many plant species, the timing of flowering initiation shows abundant quantitative variation among natural varieties, which reflects the importance of this trait for adaptation to different environments. Currently, a major goal in plant biology is to determine the molecular and evolutionary bases of this natural genetic variation. In this study we demonstrate that the central flowering regulator SHORT VEGETATIVE PHASE (SVP), encoding a MADS transcription factor, is involved in the flowering natural variation of the model organism Arabidopsis thaliana. In particular, we prove that a structural change caused by a single amino acid substitution generates a SVP early flowering allele that is distributed only in Asia. Furthermore, genetic interactions have been shown to be a component of the natural variation for many important adaptive traits. However, very few studies, either in animals or plants, have systematically addressed the extent of genetic interactions among specific alleles responsible for the natural variation of complex traits. Our study shows that the flowering effects of SVP natural alleles depend significantly on the genetic background; and, subsequently, we demonstrate the relevance of epistasis for the evolution of this crucial transcription factor and flowering time.
| Flowering initiation is an essential developmental transition in plant life because it determines the timing of sexual reproduction. This transition is regulated by different environmental signals that synchronize reproduction with the most favourable season for seed production. Hence, the timing of flowering is a crucial adaptive trait in annual plants, since it will affect their survival and reproductive yield [1]. Supporting this relevance, considerable intraspecific quantitative variation has been classically described for flowering time among natural accessions or crop varieties for many annuals, which is presumed to reflect adaptation to local environments [2], [3]. In the past fifteen years there has been an unprecedented advance in our understanding of the molecular mechanisms of flowering regulation, mostly achieved by genetic studies of artificially induced mutants in the model plant Arabidopsis thaliana [4]. More than 100 flowering genes have been identified whose analyses are defining a complex regulatory network that involves several flowering pathways integrating different environmental signals. This network includes, among others, the photoperiod, the vernalization and the autonomous pathways, as well as various regulatory genes that play a role as pathway integrators, such as FT and SOC1 [5]–[7]. Presently, a major aim in plant biology is to decipher the molecular and evolutionary bases of the naturally-existing genetic variation, for which A. thaliana has also become a promising model species [1], [8]–[10].
A. thaliana is broadly distributed as a native species in Eurasia, whereas it has been later introduced in North America and Japan, as well as in Australia and South America (reviewed in [11]). The large amount of natural genetic variation that has been described for flowering time is likely involved in adaptation to the contrasting climates that are covered by A. thaliana geographic distribution because this variation has been associated with latitude, altitude and climatic factors [12]–[16]. A. thaliana accessions have been qualitatively classified for long time as winter- or summer-annuals depending on their extreme late or early flowering behaviours and their high or low response to vernalization, respectively [17]. Mendelian genetic analyses identified two flowering repressors, FRI and FLC, as major determinants of such qualitative flowering differences [18], [19]. In addition, numerous quantitative trait locus (QTL) analyses have been carried out with different sorts of experimental mapping populations including F2 families [20], recombinant inbred lines (RILs) [21]–[27], introgression lines (ILs) [28], [29], advanced multiparent populations [30], [31], or collections of accessions [32], [33] grown in distinct environments. Each population detected between two and 13 QTL, which together correspond to, at least, 20 different genomic regions [9], [20]. Overall, these studies identified a few large effect QTL per population and a similar or higher number of small effect loci, thus showing the contribution of both extreme kinds of loci to the quantitative flowering time variation. Furthermore, despite the limitations to find genetic interactions among QTL (epistasis), owing to the low-order (two-way) level and small population sizes that can be tested, several analyses have detected a considerable number of significant interactions [20], [24], [25], [31], which indicates that epistasis is also an important genetic component of flowering time variation [34]. Even so, until now, only the well documented genetic interactions between FRI and FLC have been confirmed at the level of specific natural flowering alleles and described in terms of genetic networks [9], [35], [36]. Understanding the functional bases of genetic interactions among the specific alleles responsible for the natural variation of complex traits goes nowadays beyond the classical distinction between Fisher's and Wright's models of evolution [37] because epistasis lies below the networks currently pursued by system biology approaches [38], [39]. Therefore, functional studies addressing epistasis among natural alleles are required to determine its extent on flowering time variation and its consequences on the estimates of flowering QTL effects.
As a first step to understand the molecular mechanisms accounting for the natural quantitative variation for flowering time, multiple laboratories are pursuing the isolation of genes underlying A. thaliana QTL and the identification of nucleotide polymorphisms affecting the function of those genes. By using combinations of different functional approaches, twelve genes have been identified as large effect flowering QTL. These include the photoreceptor genes CRY2, PhyC and PhyD; the MADS transcription factor genes FLC, FLM and MAF2; FRIGIDA (FRI) and the FRI-like genes, FRL1 and FRL2, encoding homologous proteins with unknown cellular function; the RNA processing gene HUA2; the circadian rhythm gene ELF3, and the florigen encoding gene FT (reviewed in [9], [10] and [24], [40], [41]. Detailed analyses of these genes have found indels or premature stop codons causing loss-of-function alleles, as well as amino acid substitutions and other structural modifications leading to functional changes [9], [40], [41]. In addition, several cis-regulatory polymorphisms have been demonstrated to alter gene expression levels [42], [43]. Interestingly, numerous series of independent loss-of-function alleles have been described for FRI and FLC [15], [19], [20], [42], [44]–[48], which support that late flowering is the ancestral A. thaliana state but a shift towards early flowering life cycle has recently occurred at the species level [2], [49].
In this study, we aim to determine the molecular basis of a novel A. thaliana flowering QTL named as FAQ1, which we identified in introgression lines developed by phenotypic selection from the Japanese accession Fukuyama (Fuk) and the reference strain Landsberg erecta (Ler). Complementation in transgenic lines and directed mutagenesis demonstrated that a single amino acid substitution in the MADS-box gene SHORT VEGETATIVE PHASE (SVP) causes the early flowering of FAQ1 allele present in Fuk accession. We further address the biogeography of SVP allelic variation showing that this is regionally structured because FAQ1/SVP-Fuk allele appeared confined to Asia and, most likely, it originated in Japan. In addition, we aim to quantify the extent of genetic interactions involving natural SVP alleles by developing and characterizing transgenic lines for Fuk and Ler SVP alleles in five genetic backgrounds. These analyses show that FAQ1/SVP flowering effects vary from small to large depending on the genetic background, hence revealing the significant contribution of epistasis to the evolution of the flowering time variation mediated by FAQ1/SVP.
In order to uncover natural genetic variation for flowering initiation that is not detected by direct phenotypic comparisons of wild accessions, we quantified transgressive segregation in F2 populations derived from crosses between several accessions and the reference strain Landsberg erecta (Ler). Using this approach we selected the genotype Fukuyama (Fuk) because 36% of the F2 individuals showed transgressive flowering times that duplicate the phenotypic variation observed between both parents (Figure 1A). To identify the loci responsible for this variation we developed introgression lines by phenotypic selection for flowering time during four backcross generations (see Materials and Methods). Two early flowering lines, IL-2 and IL-FAQ1, carrying single Fuk introgressions from chromosome 2 (of ∼9 and ∼2 Mb, respectively) in an otherwise Ler genetic background, were characterized for their flowering behaviour (Figure 1B). On average, the two lines flowered two days earlier and with two leaves fewer than Ler under long-day (LD) photoperiod. In contrast, under short-day (SD), both ILs flowered 21 days earlier and with 28 leaves less than the reference strain, which indicates that, similar to Fuk accession, these lines have a reduced response to photoperiod (Figure 1B). F1 hybrids derived from Ler and the ILs showed towards-early intermediate flowering phenotypes suggesting incomplete dominance (Table S1). Thus, we identified a new large effect locus contributing to the natural variation for flowering initiation and its photoperiodic response, which was named as Flowering Arabidopsis QTL1 (FAQ1).
Fine mapping using an F2 (Ler×IL-2) population of 2988 individuals located FAQ1 within a genomic interval of 11 kb where Col reference genome sequence contains only two open reading frames (Figure 1C). One of them, At2g22540, corresponded to the previously known flowering gene SHORT VEGETATIVE PHASE (SVP) encoding a MADS-box transcription factor [50]. To test if SVP might be FAQ1, we generated two SVP genomic constructs corresponding to Ler and Fuk SVP alleles, and used them to transform plants of the early flowering line IL-FAQ1 (Figure 2A and 2B). Homozygous transgenic lines carrying SVP-Fuk transgene did not differ in their flowering behaviour from IL-FAQ1 indicating that this allele, in this genetic background, has no effect on flowering initiation. By contrast, most transgenic lines for SVP-Ler flowered significantly later than control plants, under SD and/or LD photoperiods (Figure 2A and 2B). Since SVP-Ler, but not SVP-Fuk, transgenes largely complemented the early flowering and the reduced photoperiod response of IL-FAQ1, it was concluded that SVP underlies FAQ1.
Sequencing of SVP in the parental accessions identified 50 single nucleotide polymorphisms (SNPs) and small indel polymorphisms differing between Ler and Fuk (Figure 2C). Most polymorphisms were detected in non-coding genomic regions and only one non-synonymous SNP was found, which was located in the middle of the MADS domain. This mutation is predicted to change Ler Ala32 to Fuk Val32, Ala32 appearing conserved in all SVP-like proteins (Figure S1). To evaluate the functional effect of this substitution we developed two additional chimerical SVP genomic constructs corresponding to Ler and Fuk alleles where we replaced by directed mutagenesis Ala32 with Val32, and viceversa. In IL-FAQ1 genetic background, homozygous transgenic lines carrying SVP-Ler-Val32 transgene flowered similar to IL-FAQ1 and did not differ from transgenic lines for SVP-Fuk allele (P>0.05; Figure 2D and 2E). However, most transgenic lines bearing SVP-Fuk-Ala32 transgenes flowered significantly later than control plants, under LD and SD photoperiod conditions. These results demonstrated that this single amino acid substitution strongly alters SVP function, Val32 from Fuk generating a SVP loss-of-function allele that displays no effect on flowering initiation, while Ler Ala32 renders SVP functional and delays flowering initiation.
Even though most IL-FAQ1 transgenic lines carrying Ler Ala32 in SVP transgene flowered later than IL-FAQ1, quantitative analysis of these lines showed that on average they flowered earlier than Ler (Figure 2A and 2B). Therefore, FAQ1 complementation with SVP transgenes was incomplete. To test if this was due to the existence of an additional gene linked to SVP that might contribute to FAQ1, or to an interaction between the transgenic and the endogenous copies of SVP, we used the four SVP genomic constructs to transform also Ler plants (Figure 2F–2I). The four classes of Ler transgenic lines showed the same overall flowering patterns observed in IL-FAQ1 background. However, most transgenic lines carrying Fuk Val32 flowered earlier than Ler, while most lines carrying Ler Ala32 flowered significantly later than Ler under SD and/or LD photoperiods (Figure 2F and 2G). The effect of SVP alleles was estimated in each background by comparing the transgenic lines carrying Ler and Fuk transgenes (Table 1). Thus, SVP effect in Ler background was significantly larger than in IL-FAQ1 (P<0.05) and similar to FAQ1 effect estimated by comparing Ler and IL-FAQ1 control lines. These results indicated that SVP accounts for most FAQ1 effect but SVP transgenes interact with the genetic background. Since both backgrounds, Ler and IL-FAQ1, differed only in the small introgression containing SVP gene, the SVP transgene most likely interact with the endogenous allele of SVP.
To further evaluate the genetic-background-dependency of FAQ1/SVP effect, we used the two SVP genomic constructs corresponding to Ler and Fuk alleles to transform three additional accessions (Fuk, Pak-1 and Pak-3) carrying similar loss-of-function FAQ1/SVP-Fuk allele (see later). A total of 108 homozygous transgenic lines were selected in all five backgrounds and grown together under LD and SD photoperiods (Figure 3). The joint analysis of these lines showed strong additive effects of SVP transgenes and genetic backgrounds (P<0.001; Table S2). However, this quantitative analysis also detected significant SVP transgene by background interaction (P<0.01; Table S2) indicating that the allelic effect of SVP depends on the genetic background. This interaction was mainly determined by the small effect of SVP transgenes in Pak-1, since significant interactions were detected (P<0.05) in all pair comparisons of Pak-1 transgenic lines with the rest of backgrounds. As shown in Figure 3, in Pak-1, the two allelic classes of SVP transgenic lines differed weakly under both photoperiods (Table 1). In contrast, both classes of transgenic lines showed larger differences in the other backgrounds, the largest SVP allelic effect appearing in Ler (Figure 3). Furthermore, the three-way interaction among SVP transgene, genetic background and photoperiod was significant (P<0.01; Table S2) evidencing that the effect of SVP on the flowering photoperiod response also depends on the genetic background. This is illustrated with the comparable SVP effect observed in Fuk, Pak-3 and IL-FAQ1 lines when grown under SD, but not under LD photoperiod where Fuk lines displayed larger SVP allelic effect (Figure 3 and Table 1). Therefore, the differential behaviour of transgenic lines in backgrounds bearing the same endogenous FAQ1/SVP allele indicates that SVP transgenes interact with one or several genomic regions other than SVP locus, as well as with the photoperiodic environment.
Genotyping of a world-wide collection of 289 A. thaliana accessions with a CAPS marker specific for SVP causal polymorphism detected six additional accessions carrying Fuk Val32, two from Pakistan and four from Japan (Figure 4A). This showed that SVP causal polymorphism is geographically structured, Fuk loss-of-function allele appearing as rare at a global scale (<2.5% frequency) but common at a regional scale in Japan, where it displayed a frequency of ∼15%.
Sequencing analysis revealed that all seven accessions with Fuk Val32 carried the same SVP loss-of-function allele because they only differed in the length of a short AT-microsatellite located in the first intron. Further SVP sequencing in 18 accessions covering the world distribution (Figure 4B and 4C) showed an overall low nucleotide diversity in SVP coding region (π-silent = 0.0038), which increased up to average genome levels [51] only in the 5′ and 3′ flanking regions. Non-synonymous diversity was especially low because only the Ala32 to Val32 substitution was found, and no other polymorphism with obvious potential effect on SVP function was detected (Table S3). To determine the genetic relationships among accessions carrying SVP loss-of-function alleles we genotyped a sample of 54 Asian accessions for a set of 237 genome-wide SNPs (Figure 4D). The five Japanese accessions carrying Fuk Val32 were nearly identical with an average proportion of allelic differences (genetic distance) of 1.6%. However the two Pakistan genotypes carrying similar SVP allele differed substantially between them (9% genetic distance) and from Japanese accessions (average distance of 13.2%), although all these accessions were more related than other Asian genotypes.
Sequence and genotypic analyses identified YGU as a Japanese genotype that is very close to the five Japanese accessions bearing Fuk Val32, for the overall genetic background (genetic distance of 5.6%) and for SVP haplotype (Figure 4C and 4D). However, YGU carried the active Ala32 SVP allele, the only other SVP nucleotidic difference corresponding to the length of the first intron microsatellite. Furthermore, YGU flowered significantly later than Fuk and the remaining Val32 accessions (Table S1), suggesting that SVP accounts for these flowering differences. This was strongly supported by co-segregation analysis in an F2 (Fuk×YGU) population grown under LD photoperiod, where SVP causal polymorphism explained 43% of the flowering phenotypic variance (Figure 4E). Thus, in this Fuk/YGU homogeneous genetic background, SVP/FAQ1 displayed a large LD effect, in agreement with the behaviour of Fuk transgenic lines. Therefore, SVP loss-of-function allele was probably generated recently in Japan, and after outcrossing and recombination it expanded to Middle Asia.
Despite the large number of flowering time QTL identified in A. thaliana, the molecular bases of only a dozen of them have been determined until now (see Introduction). In this work, we have isolated FAQ1, a new QTL identified as a large effect locus in a population highly trangressive for flowering initiation. Most previous studies have used permanent RIL populations or F2 families to detect and map QTL [9], [10], [20]. However, we identified this locus in a population of introgression lines developed by phenotypic selection in a homogeneous reference genetic background. Although the construction of such biological materials requires considerable time, they facilitated the later characterization, the fine mapping and the molecular isolation of FAQ1, showing the power of phenotype-based ILs as an alternative mapping resource to standard experimental populations.
We have demonstrated that the well-known regulator SVP encoding a MIKC-type MADS transcription factor [50], [52] contributes to the natural variation for flowering initiation in A. thaliana. It has been previously shown that SVP is a flowering repressor that affects the photoperiod response by negatively regulating several integrator genes such as FT and SOC1 [53], [54]. SVP appears regulated by the circadian clock and by the autonomous, the thermosensory and the gibberellin pathways [53], [55], [56], which suggests that SVP is also a flowering pathway integrator. Network and protein interaction studies have further revealed that SVP is down-regulated by AP1 and interacts with AP1 and other floral MADS transcription factors like CAL and SEP3 [57]–[59] thereupon showing the close regulation between SVP and the flower identity genes. In addition, SVP binds to the promoters and regulates the expression of other transcriptional regulators including miR172 and several floral repressors of the AP2 family [60]. In this study we have proven that the natural amino acid substitution Ala32 to Val32, in the MADS domain, generates a SVP loss-of-function allele that cause early flowering, in agreement with the phenotypes described for artificial svp mutants [50], [53]. MADS domains are required for DNA binding but the Ala32, highly conserved among species, has been shown to participate also in MADS protein dimerization [61]. These functions suggest that SVP-Fuk-Val32 is likely unable to properly bind and repress SOC1 and/or FT promoters, leading to the early flowering and reduced photoperiod sensitivity observed in Fuk accession. In addition, the specificity and uniqueness of this natural structural mutation suggest that most SVP structural modifications are likely deleterious and that SVP protein is essential for A. thaliana survival in nature.
Natural regulatory and structural polymorphisms in three additional MADS-box genes, FLC, FLM and MAF2, have been shown to affect flowering in A. thaliana [41]–[43], [62], [63]. In addition, a natural amino acid substitution in the MADS-box gene AGL6 has been recently demonstrated to alter shoot branching in a flowering time dependent manner [64]. Moreover, an extensive A. thaliana genome-wide association study [32] has found SVP as associated with several flowering related traits, which suggests that additional SVP polymorphisms might affect flowering initiation. Hence, MIKC-type MADS transcription factors appear as the main class of genes accounting for the flowering natural variation in this species. Interestingly, another MADS-box gene homologous to AP1 was found to contribute to the natural variation for vernalization flowering response in cereals [65]. Several studies have shown that SVP-like genes in different families of mono- and dicotyledonous plants display partially conserved functions in the photoperiod and vernalization flowering pathways [66]–[71] despite substantial copy number variation for SVP-like genes among species. Therefore, MADS transcription factors in general, and SVP in particular, appear as important candidate genes to explain the natural variation for flowering time or related traits also in plant families that are phylogenetically distant from A. thaliana [72].
Although FAQ1/SVP was detected as a large effect flowering QTL, quantitative analysis of transgenic lines shows that FAQ1/SVP effects vary from large to rather small as consequence of its genetic interactions. On the one hand, transgenic lines differing only in a small introgression indicate that SVP effect depends on the natural alleles in a genomic region located around SVP, which strongly suggests allelic interactions. This is best illustrated with the lack of flowering effects observed for SVP-Fuk-Val32 transgenes in the SVP loss-of-function background of IL-FAQ1, whereas these transgenes accelerated flowering in the near isogenic background of Ler. Thus, the flowering repression of active SVP-Ler alleles seems to be reduced by the presence of SVP-Fuk loss-of-function alleles. This result is in agreement with the incomplete dominance observed in hybrid plants derived from IL-FAQ1 and Ler, which cannot be explained simply by a SVP dosage effect [50]. Since the function of MADS transcription factors involves homo- and hetero-dimers [57], [58] it can be speculated that in plants bearing both natural SVP alleles, protein complexes containing SVP-Val32, directly or indirectly, reduce the overall SVP transcriptional repressing capacity. On the other hand, transgenic lines in different genetic backgrounds carrying the same endogenous loss-of-function SVP allele show that SVP effects depend on the natural alleles in other genomic region(s), which implies significant SVP epistatic interactions. Interestingly, SVP interacts physically with several MADS transcription factors like FLC, AP1, SOC1 and AGL6 [53], [56], [57]. This suggests that the functional basis of the observed SVP genetic interaction is the physical interaction between SVP protein and other MADS transcription factors involved in multiple complexes. Such interactions could also account for the genetic-background-dependency observed for the incomplete dominance of SVP alleles because, in contrast to the behavior in F1(Ler×IL-FAQ1) plants, SVP-Fuk allele behaved nearly as recessive in the F2(Fuk×YGU) population (Figure 4E).
All flowering QTL isolated so far correspond to large effect alleles [9], [10], which has hampered our understanding of the molecular mechanisms involved in the natural variation for flowering initiation mediated by small effect QTL [73]. The genetic-background-dependency of FAQ1/SVP shows that QTL that are primarily detected as large effect loci may have varying effects owing to genetic interactions. Thus, epistasis appears as an important component of QTL effect estimation, which is often neglected in Fisher's views of natural quantitative variation that assume the existence of series of alleles with different additive effects [39], [74]. This result brings the possibility that some of the natural flowering alleles previously isolated might also underlie flowering QTL detected with small effect, a hypothesis whose testing requires the analysis of genetic interactions in multiple backgrounds, as shown here for FAQ1/SVP. In particular, natural variants of gene families that participate in multimer protein complexes, such as the MADS genes [57], are expected to show significant genetic interactions [39], as described for numerous artificial mutant alleles of these genes including SVP, FLM and FLC [55], [58], [75]–[78]. This view is also supported by the recent identification of a natural allele of AGL6 that affects axillary bud formation in an epistatic manner [64]. It can then be speculated that the natural SVP interacting partners are any of the MADS genes FLM, FLC, MAF2 or AGL6, as supported by their segregation in nature and their participation in SVP genetic and physical interactions, although we cannot discard other genes. Thus, our study shows the usefulness of quantitative analyses of transgenic lines in multiple genetic backgrounds as a general approach to uncover any order (di- and higher-order) genetic interactions with specific natural alleles. Nevertheless, given the significant variation found among transformants, this method demands the generation of large numbers of independent transgenic lines.
Most A. thaliana alleles that have been functionally demonstrated as contributing to the natural variation for flowering initiation are alleles found in a unique accession, which hampers inferences about their role in plant adaptation [9]. By contrast, the early flowering SVP-Fuk allele appears as a recent allele likely originated in Japan and distributed in Asia. Several arguments support that this genetic variant is involved in adaptation. First, its moderate frequency in Asia, in accessions that belong to genetically differentiated clades, indicates that this is not a deleterious allele to be purged from a unique local population. Phenotypic analysis of FAQ1 ILs did not detect any other obvious developmental alteration, further supporting flowering specificity and absence of negative pleiotropic effects of SVP-Fuk allele. Second, SVP-Val32 is the only detected amino acid substitution that has been maintained in nature at high regional frequency, whereas low silent and non-synonymous nucleotide diversities suggest that SVP is under purifying selection. Third, its early flowering phenotype is in agreement with the strong recent directional selection favouring earliness that has been described at the species level [2], [49]. The significant SVP flowering effect in Fuk/YGU genetic background, in which most likely SVP-Fuk allele was originated, supports that natural selection could act through the SVP-Fuk earliness. Thus, in addition to FRI, FLC and MAF2 genes harbouring several frequent loss-of-function mutations [13], [15], [19], [41], [46]–[48] SVP represents another flowering repressor (or vegetative growth promoter) that might be under natural selection for early flowering, in agreement with previous predictions [2]. The limited regional distribution of SVP-Fuk is probably determined by its short demographical history in a non-native region that has been recently colonized [11]. However, SVP might be involved in adaptation to particular Asian local environments. The presence of this allele in a set of genetically related accessions suggests that such potential adaptive effect of SVP-Fuk depends on the genetic background, as supported by the genetic interactions described for SVP flowering effect. Conclusive demonstration of SVP contribution to adaptation awaits the analysis of the currently unknown environmental conditions where natural SVP alleles have evolved, as recently reported for other flowering genes in more extensively sampled and documented geographic regions [15], [27].
The laboratory strain Landsberg erecta (Ler) and the wild accession Fuk, obtained from Sendai Stock centre (JW116; http://www.brc.riken.jp/lab/epd/Eng/catalog/seed.shtml) and originally collected around Fukuyama (Japan), were used as parental lines to develop a population of 31 introgression lines carrying Fuk genomic segments in Ler background. ILs were developed by phenotypic selection for early flowering time during four backcross generations, each backcross being followed by a selfing generation. Briefly, the four earliest plants of an F2 (Ler×Fuk) population of 120 plants were backcrossed to Ler to obtain four independent families. A single early plant was selected per family in each of the following selfing and backcross generations. After four backcrosses, 7–8 individual sister plants per family (a total of 31 ILs) were thoroughly genotyped with 100 AFLP, microsatellite and indel polymorphic markers previously described [26], [79], [80].
IL-2 carrying a single introgression fragment of ∼9 Mb in chromosome 2 was crossed to Ler to obtain a FAQ1 F2 mapping population. FAQ1 was fine mapped by genotyping 2988 F2 plants with 24 CAPS and indel markers developed from different sources. IL-FAQ1, carrying an introgression of ∼2 Mb between physical positions 7.6 and 9.6, was derived from the mapping population.
A world-wide collection of 189 accessions (Table S4) and a collection of 100 Iberian wild genotypes [81] were analysed for flowering behaviour, for SVP sequence, and/or for SVP causal polymorphism.
Plants were grown in pots with soil and vermiculite at 3∶1 proportion in an air-conditioned greenhouse at 21°C, supplemented with additional light to provide long-day photoperiod (16 h light∶8 h darkness). For short-day photoperiod evaluations (8 h light∶16 h darkness) plants were grown in a growth chamber illuminated with cool-white fluorescent lamps.
Flowering initiation was measured as leaf number and flowering time. Leaf number was calculated as the total number of rosette and cauline leaves in the main inflorescence. Flowering time was estimated as the number of days from the planting date until the opening of the first flower.
A SVP genomic fragment of 6.5 kb, including 3.2, 2.4 and 0.9 kb of the coding, the 5′ and the 3′ regions, respectively, were sequenced in Ler and Fuk. A 5.6 kb SVP segment was sequenced in other 15 accessions (Table S4). Nine to 12 overlapping fragments of 0.8–1.3 kb were PCR amplified (Table S5) and products were sequenced using an ABI PRISM 3700 DNA analyzer. DNA sequences were aligned using DNASTAR v.8.0 (Lasergene) and alignments were inspected and edited by hand with GENEDOC [82]. Nucleotide diversity, recombination and linkage disequilibrium were estimated with DnaSP v.5 [83]. GenBank accession numbers of DNA sequences generated in this work are JX863084–JX863100.
The two 6.5 kb SVP genomic fragments from Ler and Fuk were cloned in pCAMBIA2300 binary vector (CAMBIA, Canberra, Australia) by standard molecular biology techniques. Briefly, three successive SVP segments were PCR amplified and cloned in appropriate cloning sites, and subsequently fused in the right orientation (Table S5). Two additional SVP chimerical constructs were derived by reciprocally replacing the SNP causing Ala32 to Val32 substitution. For that, site-directed mutagenesis of this SNP was performed by PCR using the spliced overlap extension method as described by Hepworth et al. [84]. Primers containing the nucleotide to be replaced are shown in Table S5. The two PCR products of each accession were purified, mixed, and subjected to 12 PCR cycles to allow extension of heteroduplexes formed between the overlapping sequences. Extended heteroduplexes were then amplified with oligonucleotides SVP-BamHI-F and SVP-BamHI-R, digested with BamHI and XbaI, gel purified, and used to replace the fragment BamHI/XbaI in Ler and Fuk SVP constructs. All PCR amplifications were performed using high fidelity Pfu polymerase (Promega, Wisconsin, USA) and constructs were verified by sequencing.
SVP genomic constructs were transferred by electroporation to AGL0 A. tumefaciens strain [85] and plants of A. thaliana were transformed by the floral dip method [86]. T1 transformants were screened by kanamycin resistance and lines carrying single insertions were selected based on resistance segregation in T2 families. Ten to 14 independent homozygous T3 lines were selected for each construct and genetic background, their transgene and endogenous SVP alleles being verified by PCR (Table S5) previous to phenotypic analyses. Phenotypic differences among transgenic lines were tested statistically with general linear models using SPSS v 19.0.
Collections of accessions were genotyped using a CAPS marker specifically developed for SVP causal polymorphism (Table S5). Accessions from Asia were further genotyped for a genome-wide set of 320 SNPs selected from different sources, as previously described [81], [87]. A total of 237 SNPs were polymorphic and were used for genetic distance and clustering analyses, their average missing data being 4.8%. Neighbor-Joining (N-J) trees were constructed with MEGA5 [88] using 10000 bootstraps to calculate percent support for each branch node.
|
10.1371/journal.pntd.0004060 | The Biting Midge Culicoides sonorensis (Diptera: Ceratopogonidae) Is Capable of Developing Late Stage Infections of Leishmania enriettii | Despite their importance in animal and human health, the epidemiology of species of the Leishmania enriettii complex remains poorly understood, including the identity of their biological vectors. Biting midges of the genus Forcipomyia (Lasiohelea) have been implicated in the transmission of a member of the L. enriettii complex in Australia, but the far larger and more widespread genus Culicoides has not been investigated for the potential to include vectors to date.
Females from colonies of the midges Culicoides nubeculosus Meigen and C. sonorensis Wirth & Jones and the sand fly Lutzomyia longipalpis Lutz & Nevia (Diptera: Psychodidae) were experimentally infected with two different species of Leishmania, originating from Australia (Leishmania sp. AM-2004) and Brazil (Leishmania enriettii). In addition, the infectivity of L. enriettii infections generated in guinea pigs and golden hamsters for Lu. longipalpis and C. sonorensis was tested by xenodiagnosis. Development of L. enriettii in Lu. longipalpis was relatively poor compared to other Leishmania species in this permissive vector. Culicoides nubeculosus was not susceptible to infection by parasites from the L. enriettii complex. In contrast, C. sonorensis developed late stage infections with colonization of the thoracic midgut and the stomodeal valve. In hamsters, experimental infection with L. enriettii led only to mild symptoms, while in guinea pigs L. enriettii grew aggressively, producing large, ulcerated, tumour-like lesions. A high proportion of C. sonorensis (up to 80%) feeding on the ears and nose of these guinea pigs became infected.
We demonstrate that L. enriettii can develop late stage infections in the biting midge Culicoides sonorensis. This midge was found to be susceptible to L. enriettii to a similar degree as Lutzomyia longipalpis, the vector of Leishmania infantum in South America. Our results support the hypothesis that some biting midges could be natural vectors of the L. enriettii complex because of their vector competence, although not Culicoides sonorensis itself, which is not sympatric, and midges should be assessed in the field while searching for vectors of related Leishmania species including L. martiniquensis and "L. siamensis".
| This study investigates the laboratory infection of two species of Culicoides biting midges (Diptera: Ceratopogonidae) and one species of sand fly (Diptera: Psychodidae) with two species of Leishmania. These members of the L. enriettii complex were demonstrated to colonize the stomodeal valve of Culicoides sonorensis following membrane feeding on blood-parasite mixtures or direct feeding on guinea pigs that demonstrated clinical signs of infection. In contrast, three other species of Leishmania that are known to be transmitted by sand flies failed to successfully develop in C. sonorensis. A sand fly species which is highly permissive to Leishmania infection, Lu. longipalpis, a widespread vector of L. infantum in Latin America, was found to support only moderate infections of L. enriettii from Brazil and Leishmania sp. AM-2004 from Australia. In addition to establishing a suitable laboratory model for infection of Culicoides with L. enriettii, successful infection of C. sonorensis highlights that vectors other than sand flies should be considered as part of epidemiological studies on parasites belonging to the L. enriettii complex.
| The leishmaniases are widespread protozoan diseases with dermal or visceral clinical symptoms that affect humans and animals worldwide. Members of the genus Leishmania (Trypanosomatidae: Kinetoplastida) follow a digenetic life cycle, alternating between a vertebrate host and insect vector. To date, phlebotomine sand flies are considered the only proven vectors responsible for maintenance of the life cycle and transmission of these parasites. The Leishmania species infecting humans comprise about 20 species, mostly belonging to the subgenera L. (Leishmania) and L. (Viannia) [1].
Reservoir hosts may be human in some cases (anthroponotic transmission), but for the majority of Leishmania species infecting humans the reservoirs are domestic or wild animals (zoonotic transmission). Most experts studying sand fly-Leishmania interactions accept six classical criteria for vector incrimination [1,2] that would ideally be satisfied to fully prove vector status: 1, there is a strong ecological association between the vector and the reservoir host; 2, parasites are isolated and/or typed from wild caught vectors not containing recent blood meals and are shown to be identical to those in the reservoir host; 3, infections in such wild caught vectors exhibit parasites in the anterior midgut, on the stomodeal valve and the presence of metacyclic promastigotes, or such development beyond the blood meal can be demonstrated by experimental infection of the vector using laboratory colonies; 4, the vector is attracted to and bites the reservoir host; 5, the vector can be infected by biting and feeding on the reservoir host or an equivalent laboratory model (xenodiagnosis); 6, experimental transmission by bite is achieved to the reservoir or an equivalent laboratory model. However, whilst desirable, rarely are all these criteria satisfied before conclusions are drawn about the identity of Leishmania vectors. Outbreaks of known species in new foci and newly discovered species are particularly problematic as in such cases the reservoir may be uncertain or completely unknown, making the testing of many of these criteria difficult.
The epidemiology of leishmaniases caused by a group of species called the L. enriettii complex is poorly understood, but is becoming increasingly important to human health. According to phylogenetic studies the L. enriettii complex occupies a position basal to all other euleishmania species, but falls outside the established subgenera Leishmania and Viannia [3–7]. The first described species within the complex, L. enriettii, was isolated from the skin of domestic guinea pigs (Cavia porcellus) in Paraná State, Brazil [8–10] and a second species (currently unnamed, here termed Leishmania sp. AM-2004) was more recently isolated from red kangaroos in Australia [11]. These sporadic infections of guinea pigs and kangaroos were characterized by occurrence of tumour-like skin lesions on the ears, nose, feet and testicles in animals [8–10], but both species appear to be non-pathogenic to humans. However, the L. enriettii complex was also recently extended to include three species known to cause clinical disease in humans: L. martiniquensis from Martinique (Caribbean island) and Thailand [7,12]; a second species from Thailand recorded as "L. siamensis" [13]; and another new species from Ghana [14]. ("L. siamensis" has not been formally described so is used in quotation marks). In addition, DNA samples from cutaneous lesions in horses and cattle in Central Europe [15,16] and the USA [17] appear to be identical to L. martiniquensis [7]. Human infections with L. martiniquensis manifest clinically as cutaneous [12,18] or visceral disease [7,19], "L. siamensis" presented as mixed cutaneous and visceral disease [13], and in Ghana the disease has only been found in the cutaneous form [14].
Suspected vectors of the L. enriettii complex include a variety of sand fly and non-sand fly Diptera. In Brazil, Lutzomyia monticola was suggested as a possible vector for L. enriettii [9], although no definitive studies have been conducted [20]. Candidate vectors of L. martiniquensis include Lutzomyia atroclavatus and Lu. cayennensis, since these are the only known sand fly species on Martinique island [12]. In Thailand, Leishmania DNA was found in Sergentomyia species, namely Sergentomyia gemmea [21,22] and S. barraudi [22], although Sergentomyia species are not usually regarded as vectors for human-infective Leishmania. In contrast, in Australia, day-feeding biting midges of the genus Forcipomyia (Lasiohelea) were implicated as vectors of cutaneous leishmaniasis caused by Leishmania sp. AM-2004 in red kangaroos and other macropods [5,11,23]. Microscopical examination revealed that Forcipomyia produced late stage Leishmania infection of high intensities including colonization of the stomodeal valve, the presence of material resembling promastigote secretory gel (PSG) [24] and promastigotes with morphology of infectious metacyclic stages [5]. This evidence for midge-transmission of Leishmania sp. AM-2004 is compelling, and the strongest vector incrimination for any member of the L. enriettii complex, but is not conclusive as a number of the criteria set out above are yet to be satisfied or tested.
The aim of this study was to evaluate the possibility that L. enriettii is also midge-transmitted, as indicated for the related species Leishmania sp AM-2004. Direct testing of this hypothesis using wild caught midges from Brazil is currently not feasible, as there is no information on likely midge vectors or colonised insects from Brazil. Therefore, the vector competences of two species of midge available in established colonies were assessed, Culicoides (Monoculicoides) sonorensis and C. (M.) nubeculosus. Neither of these can be the true vector of L. enriettii as they are not sympatric, C. sonorensis is a north American species [25] and C. nubeculosus is European [26], but both are model systems that have been used to study a wide variety of arbovirus strains and species [25–27]. Infections of L. enriettii in these two midge species were generated by membrane feeding and compared with those produced in the neotropic sand fly Lutzomyia longipalpis, which is highly permissive for all Leishmania species tested to date [28]. To provide a parasite control, parallel infections in all three insects were also performed with Leishmania sp. AM-2004. These experiments were complemented by the use of guinea pigs (Cavia porcellus) and golden hamsters (Mesocricetus auratus) experimentally infected with L. enriettii in xenodiagnosis experiments with C. sonorensis and Lu. longipalpis, testing the ability of these insects to acquire infections by feeding on these mammalian hosts.
Animals were maintained and handled in the animal facility of Charles University in Prague in accordance with institutional guidelines and Czech legislation (Act No. 246/1992 and 359/2012 coll. on Protection of Animals against Cruelty in present statutes at large), which complies with all relevant European Union and international guidelines for experimental animals. All the experiments were approved by the Committee on the Ethics of Laboratory Experiments of the Charles University in Prague and were performed under permission no. MSMT-31114/2013-13 of the Ministry of the Environment of the Czech Republic. Investigators are certificated for experimentation with animals by the Ministry of Agriculture of the Czech Republic.
Leishmania enriettii LV90 (MCAV/BR/45/LV90) and Leishmania sp. AM-2004 (MMAC/AU/2004/AM-2004; Roo1; LV756), and three human infecting Leishmania strains, L. major FVI (MHOM/IL/81/Friedlin/FVI), L. infantum CUK3 (TOB/TR/2005/CUK3) and L. donovani GR374 (MHOM/ET/2010/GR374), were used. Parasites were maintained at 23°C in M199 medium supplemented with 10% fetal calf serum (Gibco), 1% BME vitamins (Sigma), 2% sterile urine and 250 μg/ml amikin (Amikin, Bristol-Myers Squibb), and were in culture for about 10 subpassages from an animal host before use. Before experimental feeding, parasites were washed by centrifugation and resuspended in saline solution.
Lutzomyia longipalpis (Jacobina colony) was maintained at Charles University in Prague under standard conditions [29]. Females from the colonies of Culicoides nubeculosus and C. sonorensis (both belonging to subgenus Monoculicoides) were sent to Charles University from the Pirbright Institute, UK and kept at 20°C before exposure to feeding. All insects were initially given free access to 50% sucrose supplemented with penicillin (5000 U/ml), which was replaced with sugar solution alone 3 days before experimental feeding.
All infection experiments were performed at Charles University in Prague. In each experiment, approximately 150 Lu. longipalpis or Culicoides females (5–7 days old) were fed through a chick-skin membrane on heat-inactivated rabbit blood containing 107 promastigotes/ml from one of the strains described above. Engorged females were separated, maintained at 26°C or 20°C, according to experimental design, and dissected at days 1–2, 3, 5–6 and 9–10 post-blood meal (PBM). The localization and intensity of Leishmania infection in guts were evaluated in situ under a light microscope, by scoring the proportions of flies with low (<100 parasites/gut), moderate (100–1000 parasites/gut) or heavy (>1000 parasites/gut) infections [30]. All experiments were repeated at least twice.
Smears from midguts of C. sonorensis (7 and 10 days PBM) and Lu. longipalpis (10 days PBM) infected with L. enriettii were fixed with methanol, stained with Giemsa, examined under the light microscope with an oil-immersion objective and measured using ImageJ program. Body length, flagellar length and body width of parasites were measured for determination of morphological forms according to the criteria of Walters et al. [31] and Cihakova and Volf [32]. The following morphological forms were distinguished: (i) short nectomonads: body length <14 μm and flagellar length < 2 times body length; (ii) long nectomonads: body length ≥ 14 μm and (iii) metacyclic promastigotes: body length <14 μm and flagellar length ≥ 2 times body length, as summarized in Sadlova et al. [33].
Two guinea pigs (Cavia porcellus) and two golden hamsters (Mesocricetus auratus), anaesthetized with ketamin/xylazin (150 mg/kg and 15 mg/kg, respectively), were injected with 107 late-log stage promastigotes intradermally into the ear pinnae and nose. The course of infection was recorded weekly.
Xenodiagnoses were performed on animals 3, 4, 7, 9, 12 weeks post-infection (PI) using Lu. longipalpis (5–6 days old) and C. sonorensis (5 days old). Female Lu. longipalpis or C. sonorensis were placed into plastic vials covered by fine nylon mesh and allowed to feed on the inoculated site of anaesthetized animals. Successfully blood-fed individuals were then maintained for two days at 20°C and then stored in Elution Buffer at -20°C for subsequent quantitative PCR (Q-PCR). After the last xenodiagnosis, the hosts (golden hamsters and guinea pigs) were euthanized, dissected and tissues from ears, draining lymph nodes, noses, spleens, livers and blood were stored at -20°C for subsequent Q-PCR.
Extractions of DNA from vectors and animal tissues were performed using a High Pure PCR Template Preparation Kit (Roche) according to the manufacturer´s instructions. The total DNA was used as a template for Q-PCR amplification with the primers described by Mary et al. [34] in Bio-Rad iCycler and iQ Real-Time PCR Systems using the SYBR Green detection method (iQ SYBR Green Supermix, Bio-Rad).
Infection of Lu. longipalpis was attempted with two species of Leishmania, L. enriettii (LV90 strain from Brazil) and Leishmania sp. AM-2004 (LV756 strain from Australia), by membrane feeding in flies maintained at two different temperatures (26°C and 20°C). At 26°C (Fig 1A), a high infection rate (70–80%) was observed for both parasite species on days 1–2 PBM, all parasites being located in the abdominal midgut (AMG). Then, due to defecation of the blood meal remnants, the infection rate was reduced to 40% on day 3 PBM. In late stage infections (days 5–10 PBM), L. enriettii was observed only at low numbers, all being located in the AMG with no colonisation of the stomodeal valve (SV). Leishmania sp. AM-2004 generated somewhat better infections, producing moderate or heavy infections in 12% of infected females and colonization of the SV in 15–20% of them. At 20°C parasite development was similar (Fig 1B), but no L. enriettii and very few Leishmania sp. AM-2004 infections developed to a late-stage in Lu. longipalpis.
In C. nubeculosus, L. enriettii and Leishmania sp. AM-2004 parasites were present only in the AMG before and immediately after defecation. On days 6 and 10 PBM, all 54 examined females maintained at 26°C were negative (Fig 1A), while in those maintained at 20°C very few parasites were occasionally found in the abdominal midgut (Fig 1B). Neither Leishmania species were able to establish late stage infections in C. nubeculosus or any colonization of the SV.
In C. sonorensis, L. enriettii and Leishmania sp. AM-2004 both developed early stage infections at high rates (in approximately 90% of midges), producing mostly moderate infections (Fig 1A and 1B). Immediately after defecation (2 days PBM at 26°C and 3 days PBM at 20°C), again the parasite numbers were reduced, but moderate or heavy infections were still observed in some females. However, a striking contrast to C. nubeculosus was observed in parasite development on days 5–7 PBM and onwards. The infection rates observed were comparable to Lu. longipalpis, however, in C. sonorensis, Leishmania promastigotes migrated to the thoracic midgut (TMG), forming typical rosettes, and then colonized the SV in 20–25% and 20–38% of midges, for L. enriettii and Leishmania sp. AM2004, respectively, which are significantly higher percentages than in Lu. longipalpis for both parasite species. Parasite development was similar and the rate of SV colonization was comparable at both temperatures tested (Fig 1A and 1B).
Light microscopy was used to examine L. enriettii parasites in the region of the SV (Fig 2). Large masses of parasites could be seen attached to the cuticular surface of the SV, potentially partially obstructing the opening of the SV. Morphological analysis was performed on L. enriettii parasites recovered from Lu. longipalpis and C. sonorensis at 10 days PBM. The majority of parasites were short nectomonads, 80% in Lu. longipalpis and 72% in C. sonorensis, and many of these were in rosettes. There were also long nectomonads present, 13% and 23%, respectively, and metacyclic promastigotes at 7% and 5%, respectively (Fig 3).
To evaluate the significance of the above results with L. enriettii complex parasites in C. sonorensis, by membrane feeding we tested the susceptibility of this biting midge to three Leishmania species from the subgenus L. (Leishmania) capable of infecting humans, all of which have proven sand fly vectors. None of these Leishmania species were able to develop successfully into late stage infections in C. sonorensis (Fig 4). Before defecation of blood meal remnants, high numbers of procyclic promastigotes were found in the abdominal midgut in more than 90% of females on days 1–2 PBM. However, after defecation (day 3 PBM and onward) the majority of C. sonorensis females were negative and the rest had only very few parasites in the AMG. On days 6–10, no L. major or L. infantum were present in the midges, although three out of 60 females infected by L. donovani displayed long nectomonads in the AMG, but without any parasites in the TMG or SV (Fig 4). No metacyclic promastigotes were observed in C. sonorensis infected with L. major, L. infantum or L. donovani.
The first clinical signs of infection with L. enriettii in guinea pigs were redness and swelling on the inoculated ear 3–4 weeks PI. In the following two weeks (5–6 weeks PI), the swelling developed into small cutaneous lesions (~4×3 mm), which later grew rapidly to become large and ulcerated (~14×10 mm) by weeks 9–12 PI. In addition, a secondary dermal lesion appeared in one guinea pig on the skin between the eyes and nose (4.1×4.7 mm). In the animal inoculated via a nasal route, the first clinical manifestation of infection was observed in week 5 PI (two weeks later than on the the ear), but then the signs increased in severity more rapidly and later resembled a necrotic tumour-like ulcer by the end of experiment (12 weeks PI) (Fig 5).
In total, 195 Lu. longipalpis and 125 C. sonorensis female adults were fed on the ears, and 93 Lu. longipalpis and 69 C. sonorensis fed on the noses of two guinea pigs infected with L. enriettii. Preliminary experiments confirmed that Leishmania in such insect guts transformed from amastigote to short and long nectomonads, metacyclic promastigote forms and that in C. sonorensis these proliferated vigorously and colonized the stomodeal valve. The development pattern was similar as observed during experimental feeding. Therefore, Q-PCR was an appropriate method for quantification of xenodiagnosis.
Both infected ears and noses appeared to be a good source of parasites for the insects. In C. sonorensis the infection rates were about 50% and 80% for ears and nose, while in Lu. longipalpis the infection rates were a little lower at about 30% and 50%, respectively. In both vectors, the highest infection rates were observed using animals between 4–7 weeks PI, afterwards the infectivity of guinea pigs for both vectors decreased. The xenodiagnosis results in guinea pigs are summarized in Fig 6.
After the last xenodiagnosis experiment (12 weeks PI), the guinea pigs were euthanized and Q-PCR showed high numbers of parasites present in inoculated ears and noses (ranging from 8.8×106–3.8×107 parasites in each organ). In guinea pigs L. enriettii also visceralized to the spleen, and parasites were also detected in draining lymph nodes and co-lateral ears (20–500 parasites in each organ).
In golden hamsters the first signs of disease (redness and swelling) were also observed 3 weeks PI on the inoculated ears. Then, however, the course of infections strikingly differed from those in guinea pigs. In hamsters, multiple small nodules appeared on the inoculated ears in weeks 4–7 PI (Fig 5B), but all these nodules self-healed after 7–9 weeks PI (Fig 5C). Similarly, on the nose the infection was manifested only by redness, small sores and oedema, which was reabsorbed by 7 weeks PI. At the end of the experiment neither hamster presented clinical signs of infection.
In total, 159 Lu. longipalpis and 112 C. sonorensis were fed on ears and noses of two hamsters infected with L. enriettii. Generally, the vectors were less willing to feed on hamsters than on guinea pigs and their infectivity rate was much lower. Leishmania were detected only in two groups of Lu. longipalpis females fed on hamster ears 4 and 9 weeks PI (infection rates were about 30% and 10%, respectively). No positivity was found in 80 C. sonorensis females fed on the hamster´s ears. All 85 Lu. longipalpis and 31 C. sonorensis fed on inoculated noses were negative. In hamsters euthanized 12 weeks PI the Q-PCR results demonstrated very low parasite numbers in inoculated ears (<50 parasites). Other organs tested (nose, co-lateral ear, draining lymph node, blood, spleen and liver) were negative in both hamsters.
To date phlebotomine sand flies are the only proven vectors of Leishmania species, however, based on the discovery of Leishmania sp. AM-2004 in Australian biting midges [5], we assessed the possibility that L. enriettii may also have a midge vector. Based on the examination of vector competence presented here, we conclude that it is more likely that L. enriettii is transmitted by biting midges than by sand flies. However, several important aspects of vector incrimination need to be tested in future work such as their ecological associations with reservoir hosts and transmission dynamics, which may either provide further support for midge-transmision or lead to rejection of this hypothesis. There are many neotropical species of midges and sand flies, and resolution of this will require careful fieldwork and laboratory testing of any new proposed midge or sand fly vector. In the meantime we recommend that vector studies on members of the L. enriettii complex consider both midges and sand flies as potential vectors.
Our conclusion that L. enriettii is most likely to be midge-transmitted is based on several lines of evidence presented here: L. enriettii developed in the three insects in a similar way to Leishmania sp. AM-2004 and in marked contrast to L. major, L. infantum and L. donovani; the best insect host for L. enriettii was C. sonorensis, showing a similar pattern of development to that seen in Lu. longipalpis but with a higher percentage of stomodeal valve infections, and surviving beyond the blood meal to a "late stage" infection; L. major, L. infantum and L. donovani did not survive after the bloodmeal in C. sonorensis but previous work has shown these to develop mature transmissible infections in Lu. longipalpis; and both C. sonorensis and Lu. longipalpis were infected after feeding on infected guinea pigs, but to a greater extent in C. sonorensis. These data are consistent with midge-transmission of L. enriettii, but do not prove it, and each is discussed in more detail below.
Colonization of biting midges is regarded as extremely challenging, as only a very small number of species possess life cycles traits suitable for laboratory maintanance and the vast majority will not take blood meals under laboratory conditions [25]. The Nearctic species C. sonorensis was demonstrated to be susceptible to infection and our experiments showed that L. enriettii developed late stage infections in 10–30% of C. sonorensis females. We define these as "late stage" infections, meaning that they have progressed beyond the early blood meal phase and become established in the midges. The development seen is remarkable and similar to that seen with Leishmania sp. AM-2004, but in marked contrast to L. major, L. infantum or L. donovani. Moreover, 20% of infected midges with such late stage L. enriettii infections exhibited heavy colonization of the SV. Short and long nectomonads were observed during the late-stage infection in C. sonorensis gut. The short nectomonads are responsible for forward migration and colonization of the stomodeal valve including production of promastigote secretory gel (PSG), which together with sand fly saliva are critical components for disease outcome and progress [35,36]. One area of interest for future investigation would be to see if midge saliva had disease exacerbating properties similar to those of sand fly saliva [37]. Localization of parasite masses on the SV and presence of metacyclic promastigotes is associated with Leishmania transmission in sand flies [38,39] and has been observed in Forcipomyia midges naturally infected with Leishmania sp. AM-2004 [5]. It should be noted that these experiments were performed by membrane feeding, where high doses of parasites can be ingested.
Lu. longipalpis, a widespread vector of L. infantum in Latin America, was capable of supporting L. enriettii to a similar extent as C. sonorensis, although fewer SV infections were observed. However, all other Leishmania species tested in Lu. longipalpis to date, including L. major, L. infantum and L. donovani produce mature infections with high precentages of metacyclic promastigotes and PSG [28]. The percentages and intensity of late stage infections observed here for L. enriettii are far lower than normally found for infection of Lu. longipalpis with other Leishmania species. The Palearctic species C. nubeculosus was not susceptible to Leishmania enriettii, but neither was it susceptible to Leishmania sp. AM-2004. This lack of vector competence is consistent with our previous findings that C. nubeculosus does not support development of L. infantum and L. major [40]. In fact this is the predicted outcome given that there are over 1400 known species of Culicoides known worldwide [25], so the chances of finding one that supports post-blood meal development of any Leishmania parasite must be quite low, and just further emphasises the potential significance of the results obtained with C. sonorensis.
Leishmania enriettii is known as a pathogen of guinea pigs causing tumour-like skin lesions. While some authors [41,42] reported metastatic spread of parasites to distant parts of the guinea pig body (eyelids, lips, feet and genitalia), others found parasites limited to the inoculation site [10,37]. The evolution of skin lesions caused by L. enriettii can be extremely fast within two weeks PI [42] and can be enhanced by addition of sand fly salivary gland extract to the inoculum [37]. In our study, disease manifestation differed between individuals and inoculation sites. Lesions developed 5 weeks PI on the ear (ulcerated between 7–9 weeks PI) and 7 weeks PI on the nose. However, parasites inoculated into the noses grew very quickly, producing large, ulcerated, tumour-like lesions. We did not observe a self-healing process, as previously reported [37,42,43]. According to recent studies, parasites of the L. enriettii complex do not only cause cutaneous forms of leishmaniasis, but can also produce visceral leishmaniasis [7,13,19]. These findings correlate with our results from Q-PCR, which detected L. enriettii parasites in the draining lymph nodes and spleen of infected guinea pigs.
In hamsters, L. enriettii is known to be less pathogenic than in guinea pigs and some studies suggested spontaneous self-healing [44]. Here, we demonstrated that experimental infection of hamsters led only to mild symptoms. On the ear, non-ulcerated multiple nodules appeared at four weeks PI, but had self-healed by eight weeks PI. No clear signs of disease were recorded on the nose during the entire experimental period. This is in accordance to results from xenodiagnosis showing that experimentally infected hamsters were less infectious, with a low proportion of infections found in female Lu. longipalpis fed on ears 4 and 9 weeks PI, but no infections were seen in C. sonorensis.
Xenodiagnosis is currently the gold standard method used to determine infectivity of naturally or experimentally infected hosts for insect vectors. It has been repeatedly used to prove infectivity of potential reservoirs to natural vectors of L. infantum [45–47] and L. tropica [48]. In the current study the infection rate recorded was up to 50% in Lu. longipalpis and up to 80% in C. sonorensis. This is a much higher infection rate than achieved using any rodent infected with L. major, L. tropica or L. donovani [48,49]. Similar high rates (around 60–80%), were obtained only using P. perniciosus and Lu. longipalpis fed on L. infantum-infected dogs [50,51]. It also demostrated that guinea pigs were most infective for Lu. longipalpis and C. sonorensis one month post-infection, despite more serious clinical manifestation of the disease being found later during the experiment. These results agree with previous findings using mouse models where no direct link was observed between host symptoms and infectivity to vectors [48,49].
In summary, we have demonstrated experimentally, for the first time, that two species of the L. enriettii complex, L. enriettii and Leishmania sp. AM-2004, can develop late-stage infections in the biting midge C. sonorensis. This species provides a readily manipulable experimental subject for study of the L. enriettii complex under laboratory conditions; it was found to be similarly susceptible to these parasites as a permissive sand fly species Lutzomyia longipalpis. Both promastigote and amastigote infection of C. sonorensis (performed by membrane feeding and xenodiagnoses, respectively) resulted in masses of parasites in thoracic midgut and colonization of the stomodeal valve, which was found twice as frequently in C. sonorensis as in Lu. longipalpis. These data support those of Dougall et al. [5] who reported mature infections of Leishmania sp. AM-2004 in field-collected biting midges of the genus Forcipomyia. Our results support the hypothesis that biting midges might be natural vectors of the L. enriettii complex, but more detailed studies especially focused on transmission potential and field collections need to be done. However, these results should be taken in consideration while searching for vectors of L. martiniquensis, "L. siamensis" and the recently reported species from Ghana, whose sand fly vectors are unknown.
|
10.1371/journal.pgen.1001386 | Prion Formation and Polyglutamine Aggregation Are Controlled by Two Classes of Genes | Prions are self-perpetuating aggregated proteins that are not limited to mammalian systems but also exist in lower eukaryotes including yeast. While much work has focused around chaperones involved in prion maintenance, including Hsp104, little is known about factors involved in the appearance of prions. De novo appearance of the [PSI+] prion, which is the aggregated form of the Sup35 protein, is dramatically enhanced by transient overexpression of SUP35 in the presence of the prion form of the Rnq1 protein, [PIN+]. When fused to GFP and overexpressed in [ps−] [PIN+] cells, Sup35 forms fluorescent rings, and cells with these rings bud off [PSI+] daughters. We investigated the effects of over 400 gene deletions on this de novo induction of [PSI+]. Two classes of gene deletions were identified. Class I deletions (bug1Δ, bem1Δ, arf1Δ, and hog1Δ) reduced the efficiency of [PSI+] induction, but formed rings normally. Class II deletions (las17Δ, vps5Δ, and sac6Δ) inhibited both [PSI+] induction and ring formation. Furthermore, class II deletions reduced, while class I deletions enhanced, toxicity associated with the expanded glutamine repeats of the huntingtin protein exon 1 that causes Huntington's disease. This suggests that prion formation and polyglutamine aggregation involve a multi-phase process that can be inhibited at different steps.
| Certain proteins that exist in functional unaggregated conformers can also form self-perpetuating infectious aggregates called prions. Here we investigate factors involved in the initial switch to the prion form. De novo appearance of the [PSI+] prion, which is the aggregated form of the Sup35 protein, is dramatically enhanced by overexpression of the SUP35 gene in the presence of the prion form of the Rnq1 protein, [PIN+]. When tagged with green fluorescent protein and transiently overexpressed in [psi−] [PIN+] cells, Sup35 forms fluorescent rings, and cells with these rings give rise to daughter cells that are [PSI+]. Here, we investigate factors required for this induction of [PSI+]. Analyses of over 400 gene deletions revealed two classes that reduce [PSI+] induction: one class forms fluorescent rings normally, and the other does not. Interestingly, the former class enhanced, while the latter class reduced, toxicity associated with the expanded polyglutamine repeats of the huntingtin protein exon 1 that causes Huntington's disease. These results suggest that prion formation and polyglutamine aggregation involve a multi-phase process that can be inhibited at different steps.
| Prions are associated with transmissible spongiform encephalopathies, a family of neurological diseases that include Creutzfeldt-Jakob disease in humans, Scrapie in sheep, and the well-publicized “Mad Cow” disease. Transmission of prions occurs when a normally folded protein is converted to an alternate conformation that has the ability to further convert additional molecules of the normal protein into the misfolded infectious form.
Prions also exist in Saccharomyces cerevisiae. The well characterized cytoplasmically transferred elements [URE3], [PSI+], and [PIN+] (see reviews [1]–[4]), as well as several other recently characterized elements [5]–[9], have been identified as yeast prions.
Prions occur spontaneously in laboratory strains, although at very low frequency [10]–[12]. De novo appearance of prions can be facilitated by overexpression of either the whole prion protein or distinct regions that are required for propagation, called prion domains [13]–[19]. Most known yeast prion domains are glutamine (Q) and asparagine (N) rich [3], [5]–[9], [20]. In the case of de novo appearance of the prion form of the Sup35 translational termination factor, [PSI+], overexpression of Sup35 or its prion domain (Sup35PD) dramatically increases the appearance of [PSI+]. However, this increase requires either Q/N-rich domains that are simultaneously overexpressed [21], [22] or the presence of another Q/N rich prion, like the prion form of the Rnq1 protein, [PIN+] (also called [RNQ+]; [22]–[24]).
Similar to prion strains found in mammals (reviewed in [25]), yeast prions have also been shown to exist in different conformations called “variants” [14], [26]–[30]. The introduction of in vitro generated Sup35 amyloid fibers into yeast not only infects the cells with the prion, proof of the “protein-only” hypothesis, but also demonstrates that distinct forms of the in vitro made amyloid cause the appearance of distinct variants that are heritable [31]–[32].
To further understand how prions are formed and maintained, recent studies have focused on specific host factors that affect the propagation and appearance of yeast prions. Chaperones, which are normally involved in proper protein folding, play a role in prion maintenance and appearance. Hsp104 and Sis1 break prion aggregates into smaller pieces that efficiently segregate into daughter cells, a requirement for prion transmission [33]–[41]. Deletion of the N-terminal activation domain of Hsf1, a heat shock transcription factor, prevents [PSI+] formation, while deletion of the Hsf1 C-terminal region promotes [PSI+] appearance [42]. Furthermore, disruption of the non-essential human Hsp110 ortholog, SSE1, or overexpression of HSP82 and HSC82 that encode members of the Hsp90 family of chaperones, dramatically reduces, but does not eliminate, the induction of [PSI+] caused by the overexpression of Sup35PD [43].
Factors that affect [PSI+] induction are not limited to chaperones. Deletion of actin cytoskeletal genes, such as SLA1 or SLA2, reduces [PSI+] induction [44], suggesting that the actin cytoskeleton may play a role in prion appearance. Deletion of the ubiquitin-conjugating enzyme, Ubc4, enhances the de novo appearance of [PSI+] [45], and exposure to environmental stress can also alter the frequency of [PSI+] appearance [46].
Here, we identify deletions of several genes (bug1Δ, bem1Δ, arf1Δ, hog1Δ, las17Δ, vps5Δ, and sac6Δ) that reduce the efficiency with which overexpression of Sup35PD can induce the de novo appearance of [PSI+]. Deletion of LAS17, VPS5, or SAC6, which are associated with endocytosis and the actin cytoskeleton, not only inhibit [PSI+] induction, but also suppress the toxicity and aggregation associated with the expanded glutamine repeats of the huntingtin protein exon 1 that causes Huntington's disease.
Our goal was to identify genes that influence the induction of the [PSI+] prion. Previous work approached this problem by making use of the observation that overexpression of Sup35PD-GFP in [PSI+] [PIN+] cells causes toxicity due to excessive sequestration of essential Sup35 into large aggregates [46], [47]. When Sup35PD-GFP is highly overexpressed in [psi−] [PIN+] cells, the frequent induction of [PSI+] results in an intermediate level of toxicity, because only cells that have switched to the [PSI+] state are sick [23], [46]. Therefore, genes whose deletions enhance or reduce the toxicity associated with overexpression seemed likely to increase or decrease [PSI+] induction frequency, respectively [46]. We tested 238 deletion strains that enhanced toxicity, 151 that reduced toxicity, and nine other strains studied in Tyedmers et al. [46] (Table S1; Tyedmers and Lindquist, unpublished) for their effects on [PSI+] induction.
First, to distinguish the inability to induce [PSI+] from the inability to propagate [PSI+] we tested if the 398 deletions could maintain weak and strong variants of [PSI+] in a propagating culture after cytoduction with [PSI+]. A plasmid encoding a copper inducible Sup35PD-GFP fusion was plasmiduced into the deletion strains simultaneously with the cytoduction of weak or strong [PSI+]. After over 50 generations of growth, we scored for maintenance of [PSI+] by overexpressing Sup35PD-GFP and examining cells for the presence of fluorescent aggregates. All deletion strains cytoduced with either weak or strong [PSI+] contained fluorescent aggregates, indicative of [PSI+], except for hsp104Δ, which had the characteristic diffuse fluorescence of [psi−] cells (data not shown). Additionally, we have previously shown that [PIN+] is maintained in all strains of the deletion library except rnq1Δ and hsp104Δ [48].
Next, we used a standard nonsense suppression assay for [PSI+] to screen the 398 yeast deletion library strains, in the BY4741 background, for their effects on induction of stable propagating [PSI+]. To do this, the [PIN+] prion, the Sup35PD-GFP plasmid, and a plasmid containing a ura3–14 nonsense allele to score for [PSI+] cells [49] were simultaneously cytoduced into the 398 deletion strains as described previously [48]. Following overexpression of Sup35PD-GFP to induce [PSI+], cells were plated on–Ura where suppression of the plasmid borne ura3–14 nonsense allele allowed [PSI+], but not [psi−], cells to grow.
Six novel deletion strains: bem1Δ, def1Δ, scp160Δ, rpp1aΔ, spt4Δ, and pre9Δ, as well as the expected rnq1Δ and hsp104Δ deletions, failed to grow on -Ura (Table 1). Previous work has shown that in the presence of the SUP35 R2E2 allele, which increases the appearance of [PSI+] without SUP35 overexpression, bem1Δ and pre9Δ decreased [PSI+] appearance [46]. In addition, 29 other deletions showed either low or extremely low induction of [PSI+] (Figure 1; Table 1). While strains carrying deletions of SPT4 or YML010C-B failed to express Sup35PD-GFP, all other deletion strains expressed Sup35PD-GFP at similar levels (data not shown).
Deletion strains that decrease the efficiency of [PSI+] induction (Table 1) were from both enhanced and reduced toxicity groups. This makes sense if we consider that toxic side effects of the deletions could overlay the positive effect on growth rate due to the reduced [PSI+] induction. Furthermore, in some deletions, even the overexpression of YFP alone (without the Sup35 prion domain) causes strong toxicity that must be, therefore, completely unrelated to prion induction frequency [46].
To eliminate the effects of secondary mutations known to have accumulated in library strains [50], [51], multiple independent deletions of nineteen of the best candidates that showed no or reduced induction of [PSI+] (Table 1) were re-engineered in a wildtype 74-D694 [PIN+] strain. This 74-D694 genetic background contains a [PSI+] suppressible ade1–14 allele that provides the ability to directly score for [PSI+] by examining growth on -Ade. Of the 19 re-engineered deletions, only the six deletions (bre1Δ, bug1Δ, bem1Δ, arf1Δ, pre9Δ, and hog1Δ) that reproducibly reduced the frequency of [PSI+] induction, relative to the wildtype induction frequency of approximately 7.5 X 10−3 (Figure 2), were pursued further. Since a slow growth phenotype complicates the scoring for [PSI+], deletions that significantly inhibited growth in the 74D-694 background, like lst7Δ and swa2Δ (data not shown), were eliminated from further analysis.
In addition to the above deletions, we made a deletion of LAS17, which was previously shown to inhibit the aggregation of the polyglutamine 103Q repeat [52]. We also made deletions of VPS5 and SAC6, because they, like Las17 and other proteins shown to affect the appearance of [PSI+], are associated with endocytosis and the actin cytoskeleton. Even though prion propagation is unaffected in these three deletion strains because they were all able to maintain [PSI+] over many generations after cytoduction (data not shown), vps5Δ caused a significant decrease in prion induction, and las17Δ and sac6Δ strains completely failed to induce [PSI+] (Figure 2).
To eliminate deletions that reduced [PSI+] induction by altering the levels of Sup35, Sup35PD-GFP, or chaperones, we examined these levels in all strains. None of the deletions caused significant changes in Hsp104, Ssa1, Sis1, Sse1, or Ssb1/2 (data not shown). Since Sup35PD-GFP levels were reduced in bre1Δ strains and Sup35 endogenous levels were decreased in pre9Δ strains (Figure S1), these deletions were dropped from further study.
To eliminate the possibility that the reduced [PSI+] induction was due to the loss of [PIN+] during the construction of any of the deletion strains, we showed that the maintenance of [PIN+] was unaffected by the deletions. Each independently constructed deletion was crossed to a [pin-] strain carrying a plasmid with the CUP1 controlled RNQ1-GFP fusion. When grown on medium containing copper, induction of the resulting diploid containing the RNQ-GFP construct caused the appearance of punctate dots indicative of [PIN+] (data not shown). Furthermore, we showed that [PIN+] was maintained in deletion strains by the presence of [PIN+] characteristic SDS-resistant oligomeric species after 24 hours of Sup35PD-GFP overexpression (Figure S2). Differences in the migration of Rnq1 SDS-resistant oligomers have been shown to be associated with different [PIN+] variants [53]. We observed similar migration of Rnq1 oligomeric species in the deletion strains, suggesting that the [PIN+] prion variants were not altered during construction of the strain or by Sup35PD-GFP overexpression. These results and other properties of the deletions investigated further below are summarized in Table 2.
Transient overexpression of Sup35PD-GFP in [psi−] [PIN+] cells leads to the appearance of cytoplasmic fluorescent rings and lines [54]. Daughter cells derived from micromanipulated cells that contain such rings or lines, where overexpression of Sup35PD-GFP is turned off but where endogenous SUP35 is tagged with GFP, always contained fluorescent punta indicative of [PSI+] [55]. In contrast, mother cells without ring or line aggregates always gave rise to daughter cells with diffuse fluorescence, indicative of [psi−] [44], [54]–[56]. Thus the appearance of rings/lines, while not necessarily a direct intermediate in the formation of [PSI+], is nonetheless a hallmark of the potential appearance of induced [PSI+]. No such rings have ever been observed during spontaneous appearance of [PSI+] in the absence of Sup35PD overexpression (unpublished).
To determine the deletions that inhibit the induction of [PSI+] and also prevent ring formation, we compared the number of cells that contained rings after 24 hours of expressing Sup35PD-GFP in the seven deletion strains. Ring containing cells in four of the deletions, bug1Δ, bem1Δ, arf1Δ, and hog1Δ, appeared at levels similar to wildtype (30%, Figure 3A). The rnq1Δ strains, which are [pin−], always displayed diffuse fluorescence (Figure 3A), as expected, since [PIN+] has been previously shown to be required for ring formation [54].
Interestingly, when measured after 24 hrs of Sup35PD-GFP induction, the three deletions associated with endocytosis and organization of the actin cytoskeleton, sac6Δ, las17Δ, and vps5Δ, all showed a significant reduction in cells containing rings. Since by 25 hrs of induction, wildtype, sac6Δ, and vps5Δ cultures were in stationary phase, where ring formation peaks [54], the inhibition of ring formation in sac6Δ and vps5Δ was not the result of a failure of these cultures to reach stationary phase (Figure S3). Also, since las17Δ cells reached saturation phase only after 36 hours, we also measured ring formation in las17Δ cultures following 50 hrs of induction, well after the culture reached stationary phase. These cultures still showed a 50% reduction in ring formation relative to the wildtype cultures (data not shown).
Ring formation is associated with [PSI+] appearance [54]. However, it has been shown that cells that contain rings have a higher rate of cell death than those that have diffuse fluorescence [44], [47], [54]. Therefore, we asked if the deletions might be more toxic to ring bearing cells, which would result in a decrease in [PSI+] induction. To test this, we micromanipulated individual ring containing cells from selected deletion strains and assessed whether they were viable. Similar to previous findings [47], only 36% of wildtype ring containing cells were viable, whereas 95% of wildtype cells with diffuse cytoplasmic fluorescence were viable (Figure 3B). In general, the viability of ring containing wildtype, bem1Δ, and hog1Δ cells appeared to be similar (Figure 3B) and therefore cannot explain the strong reduction of de novo induction of [PSI+]. Ring containing arf1Δ cells had reduced viability (21%), which could account for the decrease in [PSI+] induction (Figure 2). In contrast, bug1Δ ring cells showed an increase in viability (59%), but apparently these cells did not efficiently give rise to [PSI+] cells. In the absence of rings, there appeared to be no effect on the viability of any of the deletion strains (Figure 3B, green bars).
Since isolating ring cells in strains with low ring formation was difficult, we focused on one example and found that of the small population of cells in vps5Δ strains that formed rings, a majority were inviable (Figure 3B). This suggests that rings in the absence of VPS5 may be harmful to the cell and likely explains the decreased percentage of rings observed in Figure 3A.
We next asked if our deletion strains also affect the [PIN+] dependent aggregation of the polyglutamine (polyQ) expanded repeat found in the mutant huntingtin protein associated with Huntington's disease. Since las17Δ strains, carrying a galactose inducible expanded polyQ repeat (103Q) fused to GFP, were previously shown to delay 103Q-GFP aggregate formation compared to wildtype strains [52], we tested our other deletions for similar aggregation patterns. When wildtype [PIN+] cells expressed 103Q-GFP for approximately one to two hours, 82% of cells displayed strong fluorescent puncta, while [pin-] cells showed mostly cells with diffuse fluorescence (73%) and a minor population that contained a few faint fluorescent foci on a diffuse background (27%; Figure 4A and 4B). Similar analyses of [PIN+] strains with bug1Δ, bem1Δ, arf1Δ, and hog1Δ deletions resulted in a reduced number of cells that contained strong fluorescent puncta (Figure 4B, green bars) and an increased number of cells that had faint fluorescent foci with a diffuse background (Figure 4B, purple bars). As in the previous study [52], [PIN+] las17Δ strains had barely any strong fluorescent foci. Examination of [PIN+] vps5Δ and [PIN+] sac6Δ strains also showed a strong reduction in cells having bright foci and a similar distribution of cells with faint foci vs. no foci as seen in wildtype [pin-] strains.
We next examined the effect of these deletions on toxicity associated with expression of the expanded polyglutamine protein in the presence of the [PIN+] prion [57]. Our controls verified earlier findings [57] that wildtype [PIN+] cells carrying a galactose inducible 103Q-GFP plasmid show toxicity compared to cells carrying the non-toxic 25Q-GFP plasmid. As expected, polyglutamine expressing cells that are either [pin-] or rnq1Δ did not exhibit toxicity, whereas wildtype [PIN+] cells were sick (Figure 4C). Since previous studies have correlated the presence of polyQ aggregates with cell toxicity [57], the lack of strong polyQ fluorescent foci in the class of deletions that also reduced ring formation after Sup35PD-GFP overexpression (las17Δ, vps5Δ, and sac6Δ; Figure 4A and B) nicely explained the observed decrease in polyQ toxicity. Interestingly, the class of deletions that reduced [PSI+] induction, but not ring formation (bug1Δ, bem1Δ, arf1Δ, and hog1Δ), increased the frequency of cells containing faint polyQ-GFP fluorescent foci on diffuse backgrounds and clearly enhanced 103Q toxicity.
We scored 398 gene deletions, previously identified by their ability to increase or decrease toxicity caused by overexpression of Sup35PD-GFP [46], for effects on the induction of [PSI+] and established effects for four of the deletions (see Table 2 summarizing all results). Since there is not always a strong correlation between the toxicity associated with SUP35 overexpression and the induction of [PSI+], we tested all of the enhanced and reduced toxicity candidates found in the previous study [46] for [PSI+] induction. We observed that induction of [PSI+] was inhibited by deletions that either reduced (arf1Δ and bem1Δ) or enhanced (bug1Δ and hog1Δ) toxicity. Possibly the rings or other less visible aggregates have an altered property in the presence of bug1Δ or hog1Δ that inhibits detectable [PSI+] appearance and causes toxicity. It has been shown that rings cause toxicity by titrating Sup35 and/or Sup45 (the essential release factor that binds to Sup35) into aggregates and away from the ribosome [46]. The toxicity associated with bug1Δ and hog1Δ during [PSI+] induction could be an enhancement of this effect or could be by the formation of a different type of toxic [PSI+] intermediate.
We also examined three deletions (las17Δ, vps5Δ, and sac6Δ), which were chosen for their association with endocytosis and the actin cytoskeleton, and found that they reduce [PSI+] induction. While las17Δ and vps5Δ were not included in the library originally scored for SUP35PD-GFP toxicity, the sac6Δ BY4741 library strain was not associated with changes in toxicity [46]. We found that sac6Δ inhibits [PSI+] induction in both the deletion library (BY4741; data not shown) and the 74-D694 (Figure 2) backgrounds. This suggests that other non-essential deletions that inhibit [PSI+] induction were likely missed in the toxicity screen.
All seven of our deletion strains reduce [PSI+] induction caused by overexpression of Sup35PD-GFP. While a deletion of BEM1 was previously shown to reduce the spontaneous appearance of [PSI+] associated with the SUP35 R2E2 expanded repeat [46], it is unknown whether our deletions affect the spontaneous appearance of [PSI+] without overexpression or mutated alleles (e.g. R2E2) of Sup35. Since the spontaneous appearance of [PSI+] is very infrequent and Mendelian suppressors with the phenotype of [PSI+] [58] appear at a higher rate than [PSI+], scoring for mutations that lower the appearance of [PSI+] is challenging.
Time lapse examination of individually micromanipulated [psi−] [PIN+] cells, containing an endogenously tagged Sup35-GFP fusion and transiently overexpressing Sup35PD-GFP from a plasmid, previously showed that a fluorescent ring initially forms at the cell periphery and then internalizes around the vacuole. Later, such cells with rings give rise to [PSI+] daughter cells with fluorescent foci (Figure 5A) [44], [55].
In this paper, we identified two classes of gene deletions that reduce the de novo induction of stable [PSI+] but differ in effects on ring formation. Class I deletions (bug1Δ, bem1Δ, arf1Δ, and hog1Δ) form Sup35PD-GFP rings at approximately wildtype levels (Figure 3). Class II deletions (las17Δ, vps5Δ, and sac6Δ) have a significantly reduced number of cells with Sup35PD-GFP rings (Figure 3).
The existence of these two deletion classes suggests that the prion formation pathway can be inhibited at different steps. In the case of class I deletions, problems in peripheral and internal ring formation were not detected (data not shown), suggesting that these genes are important for ring containing cells to transmit heritable [PSI+] aggregates to daughter cells (Figure 5A). In the case of class II deletions, peripheral rings form infrequently, suggesting that these genes are important in the initial formation of the ring even in the presence of [PIN+] (Figure 5A). While rings contained in vps5Δ (class II) cells have reduced viability, it is unlikely that ring formation is so toxic that cells die before the ring appears because there is no corresponding increase in toxicity in non-ring containing cells (Figure 3B).
The formation of polyglutamine aggregates, upon overexpression of Q103:GFP in a [PIN+] strain, is not associated with ring formation. Instead, large bright aggregates are observed within one to two hours of induction [52]. All of our deletions affected polyglutamine aggregation, but the type of effect differed based on the class of the deletion. Class I deletions (bem1Δ, bug1Δ, arf1Δ, and hog1Δ) caused a decreased level of bright fluorescent aggregates and an increase in diffuse cells with faint foci (Figure 4A and 4B). Interestingly, these deletions enhanced the toxicity associated with 103Q-GFP (Figure 4C). In the presence of the class II deletions (las17Δ, vps5, and sac6Δ) 103Q-GFP usually remained diffuse, and very few cells had bright aggregates (Figure 4B). Additionally, all class II deletions suppressed polyglutamine toxicity (Figure 4C).
The formation of large huntingtin aggregates in mammals was initially thought to be the cause of huntingtin-mediated cell death (reviewed in [59]), but emerging evidence suggests that these large aggregates are neuroprotective ([60], reviewed in [61]) and toxicity is due to a soluble pool of oligomeric conformers [62]-[63]. Also in yeast, small multiple foci of 103Q appear to be more toxic than a large aggregate [64], [65]. Thus, we propose that class I genes are important for the formation of large protective aggregates but not small toxic oligomers (Figure 4B), and that when a class I gene product is absent, the reduction of large protective aggregates permits the increased propagation of deadly soluble oligomers.
In contrast, in cells with class II deletions, 103Q-GFP toxic oligomers or aggregates may form only rarely. While the [PSI+] and polyglutamine formation pathways are not directly comparable, we propose that class II genes affect the initial steps of polyglutamine formation, and the formation of large protective aggregates (promoted by class I genes) is a downstream event (Figure 5B).
Although endocytosis is relatively unaffected in bem1Δ, bug1Δ, arf1Δ, hog1Δ, las17Δ, vps5, and sac6Δ strains, six out of seven of our deletion mutants display fragmented vacuoles (Figure S4) [66], [67]. Possibly, vacuole fragmentation may affect the perivacuolar deposition site for aggregated proteins, called IPOD [68], which has been suggested to play a role in prion formation [55], [56]. Some of the deletion library strains we screened induced [PSI+] normally but had fragmented vacuoles (Table SI, asterisk; [66]), suggesting that intact vacuoles are not necessarily a requirement for efficient prion appearance. Vacuole fragmentation has been shown to be associated with microtubular defects [69] as well as fluctuations in the soluble actin pool [67]. We showed that treatment with the microtubule disrupting drugs Nocodazole (Figure S4B and S4C) and Thiabendazole (data not shown) at concentrations that did not inhibit Sup35PD-GFP induction do not affect [PSI+] appearance. Conversely, the actin disrupting drug Latrunculin A [70] and act1-R117A alleles [44] do inhibit [PSI+] induction. Furthermore, many of the deletions identified in this study are involved with actin (Table 2), suggesting that actin organization plays a critical role in the aggregation of prion and polyglutamine proteins.
Possibly, proper actin organization on the cell periphery is required for the initial formation of the [PSI+] ring or polyglutamine oligomers, where as actin organization elsewhere in the cell could be required for downstream events such as the formation of a propagating [PSI+] conformer or the formation of large protective polyglutamine aggregates. Actin could possibly be involved in facilitating the addition of monomer to newly formed aggregates. In assaying for [PSI+], ring cells containing functional monomer not yet integrated into an aggregate would appear to be [psi−].
Our data suggests that prion and polyglutamine formation involves a multi-step process that is dependent upon actin organization. Interestingly, six of the seven proteins identified here have mammalian homologues (Table 2), suggesting that similar mechanisms may be involved in aggregation and oligomerization of QN-rich proteins in higher eukaryotes. Further elucidation of how actin nucleation contributes to prion induction will not only shed light on how toxic oligomeric species are formed, but also could provide clues to the molecular mechanisms underlying many human aggregating neurodegenerative diseases.
In this work, 398 yeast deletion library strains (parent strain BY4741: MATa ura3Δ his3Δ1 leu2Δ met15Δ; Open Biosystems, Huntsville, AL; Table S1) previously obtained from an earlier toxicity screen ([46]; Tyedmers and Lindquist unpublished) were scored for effects on [PSI+] induction in a wildtype Sup35 background. The “kar1 plasmid donor” strain (GF667; MATα CEN1–16::pGal1-CEN1–16-URA3Kl kar1Δ15 lys2 rad5-535 leu2-3,112 can1-100 his3-11,15 trp1-1 cyhR) was used to introduce plasmids and prions into the deletion library strains via cytoduction, as described in Manogaran et al. [48]. A plasmid containing a copper inducible prion and middle domain of Sup35 (Sup35PD) fused to green fluorescent protein (Sup35PD-GFP; p1181: CEN2 HIS3 ori ARS AmpR pCup1-Sup35PD-GFP) was used to induce [PSI+] in deletion strain derivates of BY4741, while a LEU2 version of the plasmid (p1182) was used in 74-D694 (L1749; [33]; MATa ade1–14 leu2–3,115 his3Δ200 ura3–52 trp1–289 high [PIN+]). To score for [PSI+] in the BY4741 strains, a plasmid containing the [PSI+] suppressible ura3–14 allele ([49]; p1513; CEN2 LEU2 ura3–14 ori ARS AmpR) was used. A tester strain (L2174; MATα leu2 ura2 his3 [pin-]) transformed with a copper inducible RNQ1 fused to GFP (p1186; CEN LEU2 ori ARS AmpR pCUP1-RNQ1:GFP) was used to confirm the presence of [PIN+] in 74D-694 deletion strains (see below). Plasmids p1572 and p1838 [52], which contain a fusion of GFP to the galactose inducible 25Q or 103Q repeats in exon 1 of the huntingtin gene, respectively, were used to examine polyglutamine aggregation and toxicity. These fusion constructs do not contain the proline-rich region of the huntingtin exon 1 [52].
Yeast strains were cultivated using standard media and growth protocols [71] and grown at 30oC except when indicated. Complex media contained 2% dextrose (YPD), and synthetic complete media contained the required amino acids and 2% dextrose (SD) or 2% galactose (SGal).
Pre-existing prions were cured by growing strains on media containing low levels of guanidine hydrochloride (GuHCl), which cures through the inactivation of Hsp104 [72], [73]. Strains were spotted onto YPD plates containing 5mM GuHCl and repeated two to three additional times to ensure that prions were cured.
Prions and plasmids were introduced into the kar1 plasmid donor strain. To make a [PSI+] kar1 plasmid donor strain to test for [PSI+] maintenance in the deletion strains, the kar1 plasmid donor strain was crossed to either a weak [PSI+] (L1759) or strong [PSI+] (L1763) strain containing Sup35PD-GFP. kar1 plasmiductants containing [PSI+] were chosen as described in Manogaran et al. [48], and confirmed to contain [PSI+] by the formation of Sup35PD-GFP aggregates after 16 hours. The [PSI+] kar1 plasmid donor was mated to the BY4741 deletion strains, deletion strain cytoductants were confirmed by testing for auxotrophic markers, and the presence of [PSI+] Sup35PD-GFP aggregate formation was examined as described above [54].
To test for the induction of [PSI+], prions and plasmids were introduced into the kar1 plasmid donor strains by crossing to either a [PIN+] (L1749 high [PIN+]) or [pin-] strain (L2910) containing the Sup35PD-GFP and ura3–14 plasmids. [PIN+] kar1 plasmid donor strains were confirmed as above and tested for the presence of [PIN+] by the formation of Sup35PD-GFP fluorescent rings after 24 hours of induction on copper [54]. Cytoduction of plasmids and prions into the BY4741 library deletion strains has been described previously [48]. To ensure reproducibility, cytoductions were performed in duplicate.
Cytoduced BY4741 deletion strains, containing [PIN+] and both Sup35PD-GFP (HIS3) and ura3–14 (LEU2) plasmids, were spotted onto plasmid selective SD-His-Leu plates plus 50 µM copper sulfate and grown for approximately two days. Strains were resuspended in 300 µL of sterile water and either spotted onto SD-Leu (grown two days) to assess growth, or SD-Leu-Ura (grown at room temperature for five to seven days) to score for [PSI+] induction. Induction experiments on all 398 yeast deletion library strains were repeated six times to ensure reproducibility. Spots that exhibited no or reduced growth on SD-Leu-Ura, compared to controls, were chosen as candidate genes that affect [PSI+] appearance (Table 1).
Genetic recombination was used to replace candidate genes (Table 1) with HIS3 in [PIN+] 74-D694. Primers (Table S2), adjacent to sequences flanking the 5′ or 3′ ends of the candidate gene, were used to PCR amplify the HIS3 gene. PCR products were transformed and His+ transformants were confirmed for insertion of the HIS3 gene in the correct locus. Two to three independent knockout lines (Table S3) were obtained for each deletion, except for pre9Δ. To confirm the presence of [PIN+], deletion strains were mated to a [pin-] tester strain containing a RNQ1-GFP plasmid. Diploids were checked for the formation of fluorescent Rnq1-GFP aggregates after induction overnight. To test whether las17Δ, vps5Δ, or sac6Δ maintain [PSI+], the deletions were cytoduced with [PSI+] and checked for growth on –Ade.
Re-engineered deletion strains were transformed with the Sup35PD-GFP (LEU2) plasmid and grown in SD-Leu plus copper sulfate liquid media for 24 hours. After induction, approximately 10,000 cells were plated on SD–Ade, and a 50-fold dilution of cells was plated on SD+12. Colony counts were obtained from at least one transformant from each independent knock out (see Table S3) on SD+12 vs. SD-Ade. Colony counts from at least three transformants were used to determine the induction frequency.
After 24 hours of induction, cells were examined for the formation of GFP fluorescent rings [54] using a Zeiss Axioskop2 deconvolution workstation equipped with either a X40 Plan-Neofluar or X100 Plan-Apochromat objective lens (Zeiss). Approximately 300 cells were counted from at least one transformant from each independent knock out, for a total of three transformants.
Ring containing cells were simultaneously visualized and micromanipulated onto 2% sterile Noble Agar slabs. Slabs were transferred onto YPD media and grown for one to two days.
Strains containing the galactose inducible 25Q or 103Q GFP fusion plasmids were grown overnight in SD-Ura and then washed in water approximately four times to remove residual glucose. To score for aggregation, washed cells were grown in liquid Gal-Ura for one to two hours with shaking and then examined for GFP aggregates. To score for toxicity, log phase uninduced washed cells were serially diluted 20-fold and spotted onto SD-Ura or SGal-Ura.
|
10.1371/journal.pgen.1004539 | Essential Genetic Interactors of SIR2 Required for Spatial Sequestration and Asymmetrical Inheritance of Protein Aggregates | Sir2 is a central regulator of yeast aging and its deficiency increases daughter cell inheritance of stress- and aging-induced misfolded proteins deposited in aggregates and inclusion bodies. Here, by quantifying traits predicted to affect aggregate inheritance in a passive manner, we found that a passive diffusion model cannot explain Sir2-dependent failures in mother-biased segregation of either the small aggregates formed by the misfolded Huntingtin, Htt103Q, disease protein or heat-induced Hsp104-associated aggregates. Instead, we found that the genetic interaction network of SIR2 comprises specific essential genes required for mother-biased segregation including those encoding components of the actin cytoskeleton, the actin-associated myosin V motor protein Myo2, and the actin organization protein calmodulin, Cmd1. Co-staining with Hsp104-GFP demonstrated that misfolded Htt103Q is sequestered into small aggregates, akin to stress foci formed upon heat stress, that fail to coalesce into inclusion bodies. Importantly, these Htt103Q foci, as well as the ATPase-defective Hsp104Y662A-associated structures previously shown to be stable stress foci, co-localized with Cmd1 and Myo2-enriched structures and super-resolution 3-D microscopy demonstrated that they are associated with actin cables. Moreover, we found that Hsp42 is required for formation of heat-induced Hsp104Y662A foci but not Htt103Q foci suggesting that the routes employed for foci formation are not identical. In addition to genes involved in actin-dependent processes, SIR2-interactors required for asymmetrical inheritance of Htt103Q and heat-induced aggregates encode essential sec genes involved in ER-to-Golgi trafficking/ER homeostasis.
| Asymmetric cell division is key to cellular rejuvenation and budding yeast exploits this mode of cytokinesis to generate a young daughter cell from a mother cell that with each division grows progressively older. Thus, age physiognomies are reset in the progeny during division, a phenomenon that requires a mother-biased segregation of cytoplasmic ‘aging factors’, including damaged/aggregated proteins. There are two models for how aggregated proteins are segregating in a mother cell-biased fashion; one holds that asymmetric inheritance is a purely passive outcome of the aggregates' random but slow diffusion whereas the other model reasons that specific factors/organelles prevent free diffusion of aggregates into the daughter cell. In the present work, we tested whether the passive diffusion model or the factor-dependent model appear most relevant in explaining asymmetrical inheritance by quantifying traits predicted to affect inheritance by passive diffusion and identifying factors required for asymmetrical inheritance amongst essential genes interacting with SIR2; a gene shown previously to be required for mother-biased segregation. We show that passive diffusion of aggregates is not sufficient to establish mother-biased segregation and that ER to Golgi trafficking, in addition to the actin cytoskeleton, calmodulin, and the Myo2 motor protein, are key components restricting the inheritance of both heat stressed-induced aggregates and aggregates formed of the Huntington disease protein Htt103Q.
| Cell division in budding yeast, Saccharomyces cerevisiae, and specific adult stem/progenitor cells includes asymmetrical inheritance of oxidized proteins, ensuring low levels of cytosolic damage in a specific cell lineage [1]–[3]. In both yeast and adult precursor cells, the lineage inheriting less damage display a longer life expectancy [1]–[3]. Thus, these singular division events provide a tractable model for how age physiognomies are reset in the progeny, which might provide clues towards therapeutically halting, or even reversing, senescence and tissue decline
In budding yeast, the control of aggregate inheritance encompasses an Hsp104-dependent retention of damaged/aggregated proteins in the mother cell [4], [5], a spatial protein quality control (SQC) that relies also on the deposition of aggregates into specific protein inclusions called Insoluble Protein Deposit (IPOD) and JUxta Nuclear Quality control compartment (JUNQ) [6]–[8]. Besides the protein remodeling factor Hsp104, the yeast gerontogene Sir2 [9]–[11] is required for asymmetrical segregation of oxidized and aggregated proteins [1], [4], [12], [13]. The role of both Hsp104 and Sir2 in establishing damage asymmetry has been linked to actin cable-dependent processes and the polarisome [5], [14]; a complex at the tip of the daughter cell required for actin cable nucleation [15], [16]. Actin cables are suggested to play a role in aggregate retention due to their (and prions') physical association with the actin cytoskeleton preventing their free diffusion into the daughter [5], [14], [17]–[19]. Sir2 deficiency reduces actin cable abundance, cytoskeletal functions, and the velocity of retrograde actin flow from the polarisome region [4], [14], [20]. This link between Sir2 and actin cable functions are consistent with data demonstrating that Sir2 affects the rate of actin folding by modulating the activity of the chaperonin CCT [14]. Actin-cables and the small heat shock protein Hsp42 are also required for the formation of peripheral aggregates [21]. Based on such results, it has been suggested that asymmetrical segregation of damaged proteins is a factor-dependent, genetically determined process, which results in the association of aggregates with structures/organelles limiting their inheritance into the daughter cell [1], [4]–[6], [14], [19].
This view is contrasting that of Li and colleagues [21], which, based on aggregate tracking experiments and modeling, argues that asymmetric inheritance is a predictable, and purely passive, outcome of aggregates' slow, random diffusion and the geometry of yeast cells. In this view, aggregate inheritance is dictated solely by the diameter of the bud neck and for how long this neck is open (generation time) for diffusion of aggregates. However, there is a large and unexplained amount of diversity in the supposedly random movement of aggregates in the aggregate population recorded by Zhou et al., [22] such that many aggregates appears stationary in the mother cell while others move in a ballistic fashion. Thus, the usefulness of employing an average diffusion coefficient for this diverse population of aggregate movements in attempting to draw conclusions about inheritance being factor dependent or purely passive has been questioned [6]. In addition, it was shown that the large aggregates in the Zhou et al., [22] study is IPOD and JUNQ inclusions that cannot diffuse freely, or randomly, since they are tethered to the vacuole and nucleus, respectively [6].
In the present work, we tested whether the passive diffusion model or the factor-dependent tethering model appear most relevant for our understanding of asymmetrical inheritance of aggregates and the asymmetry defects observed in cells lacking Sir2. To do so, we analyzed the inheritance of two reporters; the spontaneously misfolding and aggregating Huntingtin Htt103Q protein and heat-induced, Hsp104-associated aggregates and quantified the traits of sir2 mutant cells predicted to affect the inheritance of such aggregates in a passive manner. In addition, we identified hitherto unknown factors required for asymmetrical inheritance among essential genes displaying synthetic genetic interactions with SIR2, in order to determine if inheritance defects is linked to specific biological processes/components or governed by passive traits. The data obtained suggest that slow and passive diffusion is not sufficient for establishing the mother-biased segregation displayed by wild type yeast cells. Instead, we found that the essential actin-associated myosin V motor protein Myo2 and the actin organization protein calmodulin, Cmd1, are required for asymmetrical inheritance and that both Htt103Q foci and heat-induced Hsp104-associated stress foci/peripheral aggregates co-localize with Myo2/Cmd1-enriched structures. Super-resolution 3-D structured illumination microscopy further showed that both Htt103Q and Hsp104 foci co-localize with actin cables. In addition, the data suggest that a fully functional ER-Golgi trafficking/ER homeostasis activity is required for restricting aggregate inheritance during yeast cytokinesis.
For obtaining empirical, quantitative, datasets on aggregate inheritance, we used both heat-induced aggregate formation detected by Hsp104-GFP and the aggregation-prone Huntington's disease protein Htt103Q-GFP (detailed information of this construct can be found in Wang et al. 2007 [22]), which, in contrast to heat-induced aggregates, forms small and stable aggregates rather than large IPOD/JUNQ inclusions (Figure 1A; [23]–[25]). Reduced inheritance (e.g. by aggregate retention in mother cells) and aggregate removal (e.g. by disaggregation or retrograde aggregate movement in daughter cells) [14], [19] are the two processes required for establishing asymmetric aggregate distribution. Figure 1B shows a schematic illustration of how these two processes can be distinguished experimentally. Upon HTT103Q induction (leading to Htt103Q aggregation) by the addition of galactose, cells are stained with a fluorescent conA (concanavalinA) conjugate, which binds to glycoproteins in the cell wall. During the subsequent addition of glucose, which represses further HTT103Q expression, conA is washed away. This protocol enables discrimination between daughter cells present during induction of HTT103Q expression and aggregate formation (stained with conA), and cells generated after turning off synthesis of the aggregating protein (not stained with conA) that can only display aggregates if they (or possibly small aggregation nucleation particles) have been inherited from the mother cell (Figure 1B). Analyzing the inheritance of all visible Htt103Q foci demonstrated that wild type yeast mother cells retained Htt103Q aggregates in a quantitatively similar way as heat-induced aggregates [14], [21] during cytokinesis (Figure 1C&D) and that the absence of Sir2 reduced this retention capacity about 2-fold (Figure 1C; p = 0.02). During the time frame of the experiment, we found little or no clearance of the Htt103Q protein in conA-stained daughter cells (Figure 1E). Thus, establishment of asymmetrical aggregate distribution of both small aggregation-prone disease proteins and indigenous heat-induced Hsp104-associated inclusion bodies [6], [14] are dependent on Sir2 and involves aggregate retention in mother cells.
Simulations suggest [21] that to allow for the 2-fold increased inheritance the bud neck between the mother and daughter has to be enlarged by a factor of 2.2–3.0 provided the aggregates move by random walk [21] and that the generation time and aggregate number is similar in the wild type and mutant cells. Using the septin ring component Shs1-Gfp as a reporter for the bud neck, we found no evidence that the mean and median bud neck diameter in wild type and sir2Δ mutant cells was different (Figure 2A&B). In addition, the generation time of Sir2-deficient cells was not significantly longer than that of wild type cells (Figure 2C). Moreover, the average length of a sir2Δ mutant mother cell is longer than a wild type mother cell (Figure 2D), which would mean that the average aggregate in a sir2Δ mother have to embark on a longer journey to reach the daughter, which would yield a more pronounced asymmetry in Sir2-deficient cells provided aggregate distribution was solely dependent on random walk. Finally, the distribution and average number of the Htt103Q aggregates observed was similar in wild type and Sir2-deficient cells (Figure 2E) as was the number of heat-induced, Hsp104-associated aggregates (Figure S1). Thus, changes in geometrical parameters, generation time, or aggregate abundance did not explain increased inheritance of aggregates in sir2Δ daughter cells.
The passive aggregate diffusion model predicts that cells displaying a reduced growth rate will suffer from a generally increased daughter-cell inheritance of aggregates since the aggregates are allowed a longer time to randomly find their way into, and equilibrate with, the daughter cell. Therefore, we investigated to what extent Htt103Q aggregate inheritance could be enhanced in wild type cells when the generation time was slowed-down after aggregate formation by different concentrations of the protein synthesis inhibitor cyclohexamide. It has been shown that exponential cultures treated with low concentration of cycloheximide do not display arrest in any specific cell cycle stage but instead grow at a slowed exponential fashion with a prolonged cell cycle [26]. Since septum formation occurs only after the completion of mitotic events [27] the bud neck should remain open for a prolonged time upon exposure to low concentrations of cycloheximide. The Htt103Q-GFP reporter is a useful model protein for this experiment (see Figure 2F for the experimental rationale) because Htt103Q aggregates are stable (not cleared) during long periods of time (Figure 1E) and aggregate formation does not involve changes in temperatures, which would affect diffusion rates. The segregation analysis demonstrated that prolonging the generation time more than two-fold did not result in an increased inheritance of Htt103Q aggregates (Figure 2G), suggesting that the establishment of aggregate asymmetry cannot rely on slow and random diffusion alone.
To approach the passive diffusion model and factor-dependent models further, we next identified which Sir2-dependent functions are involved in restricting aggregate transfer to daughter cells. Therefore, we supplemented the previously identified genetic interaction network of SIR2 [14] with essential alleles included in the ordered, temperature-sensitive (ts), mutant library reported by Li et al. [28]. The rational for this approach is based on data suggesting that a failure to segregate protein damage can result in a reduced fitness [14],[29],[30] and it has previously been shown [14] that machineries involved in the partitioning of protein damage could be identified among the genes interacting (as synthetic sick or lethal) with a sir2 deletion using synthetic genetic arrays (SGA) analysis [31]–[33]. The protocol for allowing a sir2Δ mutant to mate and produce spores in an SGA screen has been reported previously and includes deletion of the HMR and HML silent mating type loci in the SIR2 query strain [14]. The sir2Δ × ts-allele crosses were tested for growth at varying temperatures because different ts-mutants in the library display fitness defects under different semi-permissive conditions.
We found that 6% of the 787 alleles included in the ts-library displayed statistically significant negative genetic interaction with SIR2. As seen in Table S2 and figures 3A&B, SIR2 displayed negative genetic interactions with genes involved in actin polarity, actin folding, and actin nucleation consistent with previous results [1], [4], [14], [20]. Analysis of functional relationships and known physical interactions identified 4 additional, previously unknown, functional groups of the SIR2 interaction network: 1. ‘SPB, microtubule nucleation’, 2. ‘ER-Golgi trafficking/function’, 3. ‘chromosome/sister chromatid segregation’, and 4. ‘proteasome regulatory particle’ (Figure 3A&B).
A sir2Δ mutant contains a higher ratio of unfolded/folded actin monomers than wild type cells and the chaperonin CCT isolated from sir2Δ cells displays a reduced rate of actin folding [14]. Consistently, the cct1-2 allele, similar to the cct6-18 allele [14], was found here to cause severe synthetic sickness in combination with sir2Δ (Figure 3C). The CCT chaperonin is also providing the microtubule cytoskeletal system with folded tubulin, which could explain why tub mutants are also synthetic sick in combination with sir2Δ and why genes of the ‘SPB, microtubule nucleation’ and ‘chromosome/sister chromatid segregation’ functional groups interacts negatively with sir2Δ. The SIR2 interactors of these groups are functionally related and interconnected also by physical interactions between Cdc5, Mps3 and Smc2 (Figure 3B). Mps3 and Cdc5 are required for SPB duplication and separation, respectively, and Mps3 interacts physically with Smc2 of the Smc2/4 condensin complex. Both smc2 and smc4 mutants displayed synthetic sickness in combination with sir2Δ (Table S2; Figure 3A&B), which is interesting as the cohesin subunit Smc3 displays elevated levels of acetylation in a sir2Δ mutant following α-factor treatment [34].
Like CCT, CMD1, encoding calmodulin, is required for proper function of both the actin and microtubule cytoskeletons [35]. Consistently, cmd1-1 mutant cells were severely impaired for growth when combined with sir2Δ (Figure 3D).
Essential genetic SIR2 interactors also included a relative large number of SEC genes involved in ER/Golgi functionality and trafficking (Figure 3A&B); specifically, sec18/20/22 involved in retrograde transport between the ER and Golgi, sec7, required for intra-Golgi and ER-to-Golgi transport, sec53 required for folding and glycosylation of proteins in the ER lumen, and sec11 needed for targeting proteins to the ER. In line with Sir2 buffering against defects in ER functions, the cdc48-3 allele encoding a temperature sensitive AAA+ chaperone, which facilitates extraction of ubiquitylated misfolded proteins from the ER, also displayed negative genetic interaction with sir2Δ (Table S2; Figure 3A&B). The ‘ER-Golgi trafficking/quality control’ group of genes is more distantly connected functionally to the CCT/CMD1 groups with respect to genetic interactions [33], [36] suggesting that this group of genes display genetic interaction with SIR2 for other reasons than defects in CCT and actin/microtubule functionality.
By crossing the HSP104-GFP fusion into the essential ts-mutant library using synthetic genetic array technology, we next tested whether any of the functional groups of the essential SIR2 genetic interaction network displayed aberrant aggregate inheritance of heat-induced Hsp104-associate aggregates and then followed up by testing if asymmetrical Htt103Q inheritance required the same factors. Among all the essential alleles interacting with SIR2, about 40% caused a defect in establishing Hsp104-aggregate asymmetry. One of the most severely affected mutants, cmd1-1, encoding calmodulin, belong to the group of genes involved in actin cable organization and function (Figure 4A). In addition, defects in the organization of both tubulin (tub4-Y445D) and the SPB (spc110-220) affected asymmetry (Figure 4A), suggesting that the machineries required for nuclei segregation are also required for establishing aggregate asymmetry. This is consistent with data demonstrating that aberrant nuclei segregation can lead to daughter-cell inheritance of protein inclusions, especially JUNQ [6]. With the exception of cdc48-3, mutants of the ‘proteasome regulatory particle’ and ‘chromosome/sister chromatid segregation’ groups did not display aberrant aggregate asymmetry, whereas all alleles in the ‘ER/Golgi trafficking/function’ group did (Figure 4A).
Calmodulin regulates many processes apart from actin cable organization, including vacuole inheritance, endocytosis, microautophagy, and organization and formation of the SBP. Therefore, we next tested if any or all of these processes/components are either required for preventing the inheritance of aggregates (retention in mother cells), clearance of aggregates (in daughter cells), or both using the ConA protocol (see Figure 1). Mutations in CMD1 have been reported to cause actin cytoskeletal defects by reducing the levels of the signaling molecule phosphatidylinositol (4, 5)-bisphosphate [37]. We found that the sir2Δ interactor mss4-102, a mutant allele of the phosphatidylinositol (4, 5)-bisphosphate kinase, increased aggregate inheritance and decreased aggregate removal in daughter cells (Figure 4B). In addition, Cmd1 is required for polarized growth and inheritance of the vacuole by daughter cells through its interaction with the type V myosin motor protein Myo2 [38], [39], and cells harboring the myo2-14 or myo2-16 alleles, like cmd1-1 cells, displayed severe defects in both aggregate inheritance and removal (Figure 4B). Likewise, Spc110, which requires Cmd1 for its proper localization to the SPB [35], [40], and tubulin (tub4-Y445D) were required for both asymmetrical inheritance and removal of aggregates (Figure 4B). In contrast, deficiencies in Cmd1-dependent microautophagy, which is mediated by Vtc2 and Vtc3 [41], were not affecting aggregate asymmetry (Figure 4B).
Among the calmodulin-independent genes of the SIR2 interaction network, all involved in ER/Golgi trafficking/functionality and the UPR/ERAD, displayed deficiencies in establishing aggregate asymmetry (Figure 4A) and by testing some selected alleles in this group including sec53-6, sec20-1, sec22-1, sec18-1, kar2-ts, and cdc48-3 using the conA protocol we found that all these genes were required for preventing aggregate inheritance in daughter cells (Figure 4B).
The mutations identified causing an increased daughter-cell inheritance of protein aggregates could be doing so by affecting aggregate numbers if aggregate partitioning is predominantly due to random diffusion. Therefore, we quantified aggregates in the mutants of the functional groups found to be required for asymmetrical inheritance. This analysis demonstrated that the absence of most genes identified here as being required for aggregate asymmetry, did not significantly increase aggregate numbers (Figure 4C&D, Figure S2). However, there are some intriguing exceptions; reduced activity of Cmd1 and the ER chaperone Kar2 caused a marked increase in the average number of aggregates per cell indicating that these proteins are required for inclusion body formation (Figure 4C&D, Figure S2). Nevertheless, alterations in aggregate inheritance in the majority of the mutants identified are uncoupled from changes in aggregate numbers. Defects in aggregate partitioning could also be due to diminished levels of Hsp104 [4], [5]. However, for the mutants tested herein, the defects in inheritance was not accompanied by reduced Hsp104 levels (Figure 4E), or elevated total levels of insoluble proteins, which were separated from soluble proteins by ultracentrifugation (Figure 4F).
To test to what extent alterations in generation times might contribute to changes in aggregate inheritance, we recorded daughter cell inheritance for the ts-mutants analyzed as a function of the generation time obtained during the aggregate segregation analysis. The mutants and temperatures analyzed generated generation times within a 1.5 fold difference from the wild type cells. The data was subjected to linear regression analysis together with confidence and prediction interval determinations to quantify the contribution of generation times on inheritance. A number of important observations can be made from this analysis. First, within the confidence interval (i.e. the interval displaying little difference in generation times) vastly different degrees of inheritance were recorded (Figure 4G), demonstrating that the effects on inheritance must be governed by other means than alterations in the generation time within this group of mutants. Second, in contrast to the predictions of the passive diffusion model, the best linear fit shows a weak trend towards a decreased inheritance with increased generation times but the adjusted R-squared value and p-value of −0.04346 and 0.7752, respectively, demonstrate that this trend is not statistically significant.
To test if the segregation defect seen in sec mutants could be linked to aberrancies in actin cytoskeleton organization, we analyzed actin polarity as described in [42], and found that sec53-5, like cmd1 and myo2 mutants, displayed a markedly aberrant actin polarity (increase number of cells with more than 6 actin patches) whereas the sec18-1 mutant showed a decreased number of patches (Figure 4H&I). Thus, it is possible that some SEC and ER-associated mutants fail to segregate aggregates asymmetrically due to polarity defects.
We next tested selected alleles that markedly reduced mother cell-biased segregation of heat-induced Hsp104-associated aggregates for their effect on asymmetrical segregation of Htt103Q. We found that both Cmd1 and Myo2, as well as the SEC genes (SEC18 and SEC53) were required for asymmetrical segregation of Htt103Q (Figure 5A). As for heat induced Hsp104-associated aggregates, this defect was not due to elevated levels of unfolded and insoluble Htt103Q in these cells (Figure 5B). The requisite of the same factors for asymmetrical segregation of both heat-induced Hsp104-associated aggregates and Htt103Q is somewhat unexpected as the former is sequestered into distinct, inclusion bodies (IBs), IPOD and JUNQ, upon heat stress [7], whereas Htt103Q forms multiple small aggregates throughout the cytoplasm [23], [24], [43]. However, before the formation of IPOD/JUNQ, misfolded, Hsp104-associated, proteins assemble into small stress foci ([6]; also called Q-bodies [30] or peripheral aggregates [44]), reminiscent of the smaller Htt103Q aggregates. We therefore tested if Hsp104-GFP co-localized with Htt103Q immediately after heat stress and found this to be the case; co-localization can be observed in about 97.1% cells displaying both Htt103 and Hsp104 aggregates (Figure 5C), indicating that Htt103Q may be sequestered at terminally stable stress foci-like structures. It has been suggested that amyloids and heat-denatured proteins are sequestered to spatially different quality control sites [7]. Therefore, we tested whether Htt103Q formed amyloids using Thioflavin-T staining but found no evidence for this whereas the positive control, Rnq1-mRFP aggregates readily stained with Thioflavin-T (Figure 5D).
Next, we analyzed if Htt103Q foci co-localized with Cmd1 and/or Myo2, which could explain, in a direct physical manner, why retention in the mother cell relies on these factors. In cells where Cmd1 or Myo2 were enriched in visible structures, 94.4% and 92.2% showed co-localization between Htt103Q and such Cmd1 or Myo2 structures, respectively (Figure 5E). In addition, the ATPase-deficient Hsp104, Hsp104Y662A, which has previously been shown to be ‘locked’ in a stress foci stage [6], similarly co-localized with Cmd1 (in 74.4% of cells) and Myo2 (in 56.6% of cells) enriched structures (Figure S3). We found less co-localization between Htt103Q foci and Sec18 (about 48% of cell showing both Sec18 structures and Htt103Q foci) whereas no clear co-localization could be observed between Htt103Q foci and Sec53 (displaying a diffuse signal) (Figure 5E). In contrast, in cells with heat induced Hsp104Y662A foci a clear co-localization can be observed between Hsp104Y662A and Sec53 (82.7%; Figure S3). Some of the Htt103Q and Hsp104Y662A foci appeared to reside in the vicinity of the ER, as detected by Rtn1-GFP co-staining (Figure S4). Previous studies have shown that the small heat shock protein Hsp42 affects sequestration of misfolded proteins; specifically, in the absence of Hsp42 misfolded proteins are predominantly directed to the juxtanuclear JUNQ deposition site instead of peripheral, nucleus-distant, aggregation sites [8], [44]. Importantly, we found that the absence of Hsp42 redirected Hsp104Y662A to inclusions (cells with 1 or 2 aggregates; i.e. class 1 and 2 cells) rather than peripheral aggregates/stress foci (3 or more aggregates; class 3 cells) whereas formation of Htt103Q foci was unaltered (Figure 5F,G). Co-staining with DAPI demonstrated that the number of cells with a single juxtanuclear-localized aggregate were increased in the hsp42Δ mutant (Figure S5). These data indicate that while Hsp104Y662A and Htt103Q are directed to overlapping, Cmd1/Myo2-associated, foci, the routes/factors employed for sequestering heat-induced stress foci and Htt103Q to such sites may be different.
Cmd1 and Myo2 are intimately associated with the actin cytoskeleton and protein aggregates and prions have been found previously to reside in areas rich in actin-enriched structures using proximity ligation assays and co-localization fluorescence microscopy [14], [18], [19], [43], [45]. However, one major drawback with conventional fluorescence microscopy is that the x–y axial resolution is limited to about 250 nm and the z axial resolution to about 500 nm [46], [47]. Therefore, to more precisely analyze the spatial relationship between protein aggregates/stress foci and the actin cytoskeleton, we performed super-resolution three-dimensional structured illumination microscopy (3D-SIM) [48] to analyze possible aggregate and actin cytoskeleton interactions in vivo. With this technique an approximately 8-fold smaller volume can be resolved in comparison to conventional microscopy equating about 100 nm in x-y and 200 nm in the z axial [46], [49]. The 3D-SIM analyses revealed that both Htt103Q and Hsp104Y662A foci line up along actin cables and are in some instances wrapping around the cables (Figure 6A–D, Figure S6 and Movie S1). At this resolution, using multiple Z-stacks, it is clear that the co-localization is not due to actin oligomers residing in the aggregates themselves (Figure 6B&D). Moreover, we found that Hsp104Y662A-mCherry stress foci displayed a considerable co-localization with the actin cable-associated protein Abp140-3GFP further supporting an association between stress foci and actin cables (Figure 6E&F).
The development of an in situ protocol for detecting oxidatively damaged (carbonylated) proteins in single cells of S. cerevisiae led to the discovery that damaged proteins display a mother cell-biased segregation during cytokinesis [1] and it was later shown that such oxidatively damaged proteins coalesce into aggregates upon aging that rely on both Hsp104 [4], [5] and Sir2 [1], [14], [50] for their asymmetrical inheritance. Data accompanying the original discovery showed also that elevating damage in the mother cell by a transient exposure to oxidants rendered asymmetrical inheritance even more pronounced indicating that asymmetry was not entirely due to a passive effect of slow diffusion [1]. The data presented in this work is consistent with this notion: If aggregates would find their way into a daughter cell purely by passive and random movement, increasing the time for completing cytokinesis would enhance aggregate inheritance - we show here that this is not the case. Further, mutants with increased daughter-cell inheritance should, if aggregates diffuse randomly, either display a larger bud-neck diameter, a longer generation time, or increased number of aggregates; none of these traits could be established for sir2Δ cells or the other mutants displaying elevated inheritance. Prevention of inheritance of both the misfolded polyQ protein Htt103Q and heat-induced aggregates relies instead on specific cellular processes/components, including the actin-associated proteins Cmd1 and Myo2 and Sec proteins involved in ER to Golgi trafficking and ER homeostasis.
It is possible that a reduction in actin cable abundance affects aggregate diffusion, as an intact actin cytoskeleton in dictyostelium appears to slow down the diffusion rates of soluble GFP proteins [51]. However, since both Htt103Q and heat-induced foci co-localized with Cmd1- and Myo2-enriched structures, the retention of aggregates might be linked to a more direct physical interaction between these components and aggregates. Cmd1 and Myo2 are intimately associated with actin cables and super-resolution 3-D SIM microscopy demonstrating that stable foci of both Htt103Q and Hsp104Y662A are associated with the actin cytoskeleton (and the actin-binding protein Abp140) in the nm-scale. These data is further supporting a role of actin cable assembly [1], [4], [44], actin folding [14], and actin polarity [19] in aggregate inheritance control.
Interestingly, the Htt103Q foci co-localized with the Hsp104 stress foci formed early upon a heat shock. Thus, misfolded Htt103Q and heat-denatured proteins appears to be sequestered into the same spatial sites. In further support of this notion, the disaggregase-defective Hsp104Y662A-GFP, which upon heat stress forms stable stress foci [6], like Htt103Q, co-localized with Cmd1 and Myo2. However, we found that only Hsp104Y662A-associated misfolded proteins, and not Htt103Q, required Hsp42 for foci formation. The absence of Hsp42 has been shown previously to redirect heat-denatured misfolded proteins to the nucleus-proximal JUNQ deposition site at the expense of peripheral aggregation sites at least in the presence of a proteasome inhibitor [8], [44] but we found that Htt103Q foci formation was unaffected by Hsp42 deficiency. Inversely, we found that Cmd1- and Kar2-deficiency reduced the cells' ability to form Hsp104-associated inclusions upon a heat shock; that is, heat-induced aggregates appear ‘locked’ in the stress foci stage in these mutants.
The participation of the Myo2 motor protein and calmodulin in asymmetrical inheritance of aggregates suggests that the role of actin cables and polarity in this process may be linked also to vesicle/organelle trafficking. In support of this notion, the IPOD inclusions are associated with the vacuole [6], [7] and it is conceivable that misfolded proteins reach such deposit sites in an actin cytoskeleton- and vesicle trafficking-dependent way. Indeed, Specht et al., [44] have demonstrated that misfolded proteins fail to form peripheral aggregates when actin cables are depolarized with Latranculin and Kaganovitch et al., [7], using benomyl treatment, demonstrated the requirement also of microtubule in the formation of inclusion bodies. However, it was later shown that the effect of benomyl in inclusion body formation might be microtubule-independent [44]. The effect of abrogated ER/Golgi function on aggregate segregation could also be linked to effects on actin/calmodulin/Myo2-dependent vesicle/vacuole trafficking since the ER/Golgi is involved in lipid modifications of specific proteins, e.g palmitoylation and myristoylation, required for anchoring Myo2 to targets at vesicle membranes [52]. In this scenario, Myo2 might act as a tethering factor required for misfolded/aggregated proteins to become linked to actin cables and/or deposition sites on the surface of the vacuole since misfolded proteins have been demonstrated to associate with membrane vesicles [17], [43]. In addition, a recent report shows that misfolded Ubc9ts proteins form puncta called Q-bodies that are associated with ER [30]. However, it should be noted that Ubc9ts-Q-bodies move in an actin-cable-independent (but energy-dependent) manner suggesting that these structures are not themselves associated with actin. The apparent difference with respect to actin-association of Ubc9ts Q-bodies and Htt103Q foci is interesting and may suggest that different misfolded proteins are sequestered to different spatial locations. Another possible reason for the different results is the use of different protocols; whereas Htt103Q readily aggregate upon its production Ubc9ts aggregation is triggered by elevating the temperature, a protocol that disrupts actin cables. Also, while Ubc9ts Q-bodies move in an actin cable-independent manner [30], it is not clear if their subsequent progression to IPOD/JUNQ inclusion sites require functional actin cables or not since the dynamics, morphology and inheritance of cortical ER (which the Ubc9ts Q-bodies associate with) have been linked to actin cytoskeleton components [53]–[55]. Elucidating the exact cytological, biochemical, and genetic nature of stress foci, Q-bodies, peripheral aggregates, and IPOD/JUNQ inclusions and their relevance for different aggregate reporters appears an important task for future research.
It has recently been shown that aggregate accumulation during replicative aging of mother cells follow a delineated path; virgin and young cells display no protein aggregates, middle-aged mother cells harbor one to two protein inclusions, first JUNQ then also IPODs, while old cells display, in addition to JUNQ and IPODs, multiple peripheral aggregates resembling stress foci [56]. It will be interesting to learn to what extent these foci are connected to Cmd1/Myo2 and the actin cytoskeleton and if such associations are actually a cause of aging. We envision that the tethering of multiple aggregates will disturb actin cable-dependent trafficking processes and eventually cause a complete collapse in the physical integrity of the actin cytoskeleton. In addition, the new and previously unknown genetic interactions between SIR2 and essential genes recorded herein points to additional Sir2-related functions of potential relevance for life span control. Specifically, since Sir2 buffers against deficiencies in microtubule/spindle pole body and chromosome/sister chromatid segregation functions, it is tempting to speculate that the diminishing level/activity of Sir2 observed in aging cells [57] leads to problems also in performing proper chromosome/nuclei segregation. Further studies appear warranted to elucidate how this sirtuin is mechanistically buffering against defects in these essential functions and how they might relate to sirtuins acting as gerontogenes.
Yeast strains used in this study are listed in Supplemental Table S1. The Yeast conditional temperature-sensitive (ts) collection of essential genes for the SGA analysis was a gift from Prof. Charles Boone. Yeast cells were grown in YPD or synthetic drop-out media with antibiotics added as indicated.
The SIR2 (ts) SGA analysis was performed in duplicates as described [28]. The screen was run in the 1536-spot format using a SINGER ROTOR HDA Robot (Singer Instrument Co. Ltd.). Hits with the highest statistical probability to be true interactions were confirmed by microcultivation experiments in triplicate at 30, 34, and 38°C using the Bioscreen C system (Labsystems Oy, Helsinki, Finland). The optical density was measured every 30 minutes for 72 hours. The LSC (Logarithmic Strain Coefficient) values of growth rates were calculated and scored as described [28]. The heat map was made using TreeView [58] and Ospery 1.2.0 [59] was used for SIR2 essential gene network analysis. The physical interactions between SGA hits were obtained based on the BioGRID interaction database [60].
A Zeiss Axiovert 200 M fluorescence microscope was used to obtain images using GFP, Cy3 and DAPI channels. The ImageJ plugin “Iterative deconvolve 3-D” was used for all deconvolution images.
Cells containing the pYES2-HttQ103-GFP plasmid were grown at 30°C to exponential phase (OD600 about 0.5) in YNB-URA 2% raffinose. Htt103Q-GFP expression was induced by adding galactose to a final concentration of 2%. After 4 hours at 30°C cells were washed and resuspended in 1.5 ml Buffer P (10 mM NaH2PO4, 150 mM NaCl, 2% galactose; pH 7.2). All cells present during the expression of Htt103Q-GFP were marked by staining the cell wall components α-mannopyranosyl and α-glucopyranosyl with 0.2 mg/ml concanavalin A Alexa Fluor 647 (Invitrogen) for 30 minutes at room temperature. The cells were then washed in Buffer P, resuspended in YNB-URA with 2% glucose, which will switch off the Htt103Q-GFP expression, and grown at 30°C for one budding event. This makes it possible to distinguish between concanavalin A stained daughter cells present during Htt103Q-GFP expression and daughter cells produced subsequent to Htt103Q-GFP expression. Cells were fixed in 3.7% formaldehyde and segregation of aggregates was analyzed using fluorescence microscopy. The segregation assay of ts-alleles were performed in the same way but without concanvalin A staining. Cells were grown at 22°C to exponential phase, followed by induction and budding at different temperatures (26°C for sec18-1 and sec53-6, 28°C for cmd1-1 and 32°C for myo2-14).
The retention efficiency assay was performed as described [14] with concanavalin A staining (as described above) before the heat shock treatment. Aggregate retention and removal was distuinguished upon image analysis. Retention is determined as the percentage of aggregate-containing buds of the total number of buds generated from an aggregate-containing mother cell after the heat shock treatment (buds free of conA); Removal efficiency is determined as the perecentage of aggregate-free buds of the total number of buds already existing before heatshock; i.e. stained with conA. More than 300 budding events were quantified for each strains.
All statistical calculations were done in R-3.0.0 (www.r-project.org). Regression analysis was performed and showed that there is no statistically significant difference at p<0.05 to test that whether there is a correlation between ‘relative generation time’ and ‘deviation in asymmetry’.
The in vivo expression of mHtt103Q-GFP protein was induced by galactose addition to yeast cells in mid-exponential phase (OD600 = 0.5) grown in media with 2% raffinose. 10 µg/ml of cycloheximide (CHX) was added into the culture after 3 hours of induction to stop translation. Cells were continued to be cultured at 30°C with shaking. Aliquots were then taken and fixed at 0, 1, 2, 3, 4 and 12 hours after the addition of CHX. GFP signal intensity for each sample was quantified by flow cytometry (FACS Aria, BD equipment). 10000 events were counted for each sample. A cut-off value was picked based on an un-induced control. Stability is measured as mean signal intensity from Htt103Q-GFP aggregates as a function of time after inhibition of protein synthesis by adding CHX.
Cells containing the pYES2-HttQ103-GFP plasmid were grown at 30°C until OD600 reached approximately 0.5 in SC-URA +2% raffinose. The expression of Htt103Q-GFP was induced by adding 2% galactose for 4 hours at 30°C. Cells were then stained with ConA as described above. The cell density was adjusted to OD600 = 0.5. The cells were divided into 3 groups and recovered at 30°C for different times until the cells grow to the same optical density value as the untreated control with different concentrations of cycloheximide. Further expression of HttQ103-GFP was inhibited by adding 2% glucose to the medium during bud formation. The cycloheximide treatment was performed as follows: In the untreated group, no cycloheximide was added during the budding period and the cells were recovered for 4 hours at 30°C to let the cells generate new buds and grow to a OD600 reaching 0.8. In the two cycloheximide treatment groups, 0.05 or 0.1 mg/ml cycloheximide were added to the cell cultures and grown at 30°C until OD600 reaches the same value (0.8) as in the untreated control group. Then cells were fixed in 3.7% formaldehyde and budding events with newly generated buds were analysed for aggregate inheritance using a Zeiss fluorescence microscope.
20 mL OD600 = 0.7 Yeast cells were collected after heat shock and recovery then resuspended in 800 µl 0.1 M NaOH for 5 minutes (room temperature) and then pellet. The cells were boiled for 3 minutes in 125 µl lysis buffer (50 mM Tris, pH 7.4, 5 mM EDTA, 5 Mm NEM, 1% SDS) with protease inhibitor (Roche, 11697498001). The supernatant was mixed with equal amount laemmli buffer and heated at 95°C for 3 minutes. Denatured proteins were loaded onto NuPAGE Novex 10% Bis-Tris Gels (Invitrogen, NP0315BOX) and transferred to PVDF membranes to perform western blotting. Antibody to GFP (Roche, 11814460001) and pGK (Invitrogen, 459250, as loading control) were used in this study. After western blotting, the membranes were scanned by Odyssey Infrared Imaging System and quantified by Odyssey 2.1 software. Fold ratios were calculated based on three biological repeat experiments.
Solubility assays were carried out as described in [61]. Same volume of protein solution of each sample was loaded on precasted SDS-PAGE gels (Life Technologies). For testing heat-induced aggregates, the gel was then stained by Coomassie Brilliant Blue and scanned with a GS-800 Calibrated Densitometer (BioRad). For Htt-103Q-GFP expressing cells, the Htt-103Q-GFP protein was detected by Western blotting with mouse anti-GFP monoclonal antibody (Roche). The scanned gels and Western membranes were then quantified by ImageJ software. Relative ratio of soluble and aggregated protein for each tested strain were then calculated and plotted.
Thioflavin T staining of amyloid was performed according to a protocol from [62] with minor changes. Cells were fixed in 50 mM KPO4 (pH 6.5), 1 mM MgCl2, 4% formaldehyde for 10 minutes and then washed three times with PM buffer [0.1 M KPO4 (pH 7.5), 1 mM MgCl2] and resuspended in PMST [0.1 M KPO4 (pH 7.5), 1 mM MgCl2, 1 M Sorbitol, 0.1% Tween 20]. Cells were treated with 0.125 mg/ml Zymolase at room temperature for 15 minutes in the presence of 0.6% beta-mercaptoethanol. Spheroplasted cells were washed once and then resuspended in PMST and stained with 0.001% Thioflavin T for 20 minutes at room temperature. After five times of washing with PMST, the cells were observed under a Zeiss Observer Z1 microscope at CFP channel for the staining.
Mid-log phase cells (OD = 0.5) were incubated at 42°C for 10 minutes and then 30°C for 30 minutes. These cells were then fixed with 3.7% formaldehyde and washed twice with PBS (pH 7.4) and stored for microscopy.
At least 100 cells with both visible structures (formed by GFP-tagged proteins) and red aggregates (formed by Hsp104-Y662A-mCherry or Htt103Q-mRFP) were analyzed. Among them, cells with any overlapping green and red signals were considered as cells with co-localization, the percentage of which were then calculated. Z-stack images were used for aggregate morphology quantification and more than 100 cells showing aggregates were quantified. Cells were divided into 3 classes based on the number of aggregates in the cell (Class 1, cells with 1 aggregate; Class2, cells with 2 aggregates; Class 3, cells with 3 or more aggregates). Aggregates were counted throughout all Z-stacks.
Actin depolarization was quantified according to Ho et al. [42]. All budding events that have a small or medium bud were counted for number of actin patches in the mother cells. Mother cells bearing more than 6 actin patches were counted and the percentage of them over all mother cells were calculated and plotted.
Cells carrying Hsp104Y662A-mCherry were incubated at 42°C for 10 minutes and then 30°C for 10–20 minutes. And cells with Htt103Q-GFP was induced by adding 2% galactose for 4 hours at 30°C. Then cells were fixed with 3.7% formaldehyde and washed twice with PBS (pH 7.4). Actin cytoskeleton was stained using Alexa Fluor 568 Phalloidin (Ivitrogen, A12380) as described in the manual. Cells expressing both Abp140-3GFP and Hsp104Y662A-mCherry were heat treated as above. For super-resolution three-dimensional structured illumination microscopy, the ELYRA PS.1 LSM780 setup from Zeiss (Carl Zeiss, Jena Germany) was used. 3D-SIM images of the protein aggregates (Hsp104Y662A-GFP) and actin cytoskeleton (Alexa fluor 568 phalloidinor Abp140-3GFP) were taken with 100×/1.46 Plan-Apochromat oil-immersion objective with excitation light wavelengths of 488 nm and 561 nm. Z-stacks with an interval of 100 nm were used to scan the whole yeast in 3D-SIM. For acquisition and super-resolution processing and calculation as well as for 3D reconstruction, the Zen2011 software (Carl Zeiss, Jena Germany) was used. The ELYRA System was corrected for chromatic aberration in x-, y-, and z-directions using multicolor beads, and all obtained images were examined and aligned accordingly.
|
10.1371/journal.pgen.1006479 | Ku Binding on Telomeres Occurs at Sites Distal from the Physical Chromosome Ends | The Ku complex binds non-specifically to DNA breaks and ensures repair via NHEJ. However, Ku is also known to bind directly to telomeric DNA ends and its presence there is associated with telomere capping, but avoiding NHEJ. How the complex discriminates between a DNA break and a telomeric extremity remains unknown. Our results using a tagged Ku complex, or a chromosome end capturing method, in budding yeast show that yKu association with telomeres can occur at sites distant from the physical end, on sub-telomeric elements, as well as on interstitial telomeric repeats. Consistent with previous studies, our results also show that yKu associates with telomeres in two distinct and independent ways: either via protein-protein interactions between Yku80 and Sir4 or via direct DNA binding. Importantly, yKu associates with the new sites reported here via both modes. Therefore, in sir4Δ cells, telomere bound yKu molecules must have loaded from a DNA-end near the transition of non-telomeric to telomeric repeat sequences. Such ends may have been one sided DNA breaks that occur as a consequence of stalled replication forks on or near telomeric repeat DNA. Altogether, the results predict a new model for yKu function at telomeres that involves yKu binding at one-sided DNA breaks caused by replication stalling. On telomere proximal chromatin, this binding is not followed by initiation of non-homologous end-joining, but rather by break-induced replication or repeat elongation by telomerase. After repair, the yKu-distal portion of telomeres is bound by Rap1, which in turn reduces the potential for yKu to mediate NHEJ. These results thus propose a solution to a long-standing conundrum, namely how to accommodate the apparently conflicting functions of Ku on telomeres.
| The Ku complex binds to and mediates the rejoining of two DNA ends that were generated by a double-stranded DNA break in the genome. However, Ku is known to be present at telomeres as well. If it would induce end-to-end joining there, it would create chromosome end-fusions that inevitably will lead to gross chromosome rearrangements and genome instability, common hallmarks for cancer initiation. Our results here show that Ku actually is associated with sites on telomeric regions that are distant from the physical ends of the chromosomes. We propose that this association serves to rescue DNA replication that has difficulty passing through telomeric chromatin. If so called one-sided breaks occur near or in telomeric repeats, they will generate critically short telomeres that need to be elongated. The binding of Ku may thus either facilitate the establishment of a specialized end-copying mechanism, called break induced replication or aid in recruiting telomerase to the short ends. These findings thus propose ways to potential solutions for the major conceptual problem that arose with the finding that Ku is associated with telomeres.
| The Ku proteins, initially identified as an auto-antigen in sera from patients suffering of scleroderma-polymyositis overlap syndrome [1], are highly conserved in eukaryotes and there are also prokaryotic equivalents [2]. In eukaryotes, two subunits, Ku70 and Ku80, form a complex and its crystal structure revealed resemblances to a preformed ring [3]. This Ku-complex selectively associates with ends of double-stranded DNA molecules with high affinity but no sequence specificity [2, 4]. Ku’s primary function is to mediate Non-Homologous End Joining (NHEJ), the predominant DNA double-strand break (DSB) repair mechanism in mammals [4, 5]. However and paradoxically, in many species Ku does associate with telomeres and/or telomerase and a number of telomere-specific functions for Ku have been described [4]. How these telomere-specific functions that are thought to preclude DNA-end fusions discriminate telomeres from DSBs, where DNA-end fusions are the desired outcome, remains unknown.
The budding yeast S. cerevisiae also contains a yKu complex formed by Yku70 and Yku80 subunits [6–8]. As in mammals, yKu is essential for NHEJ, but not for Homologous Recombination (HR) [7]. yKu binds telomeres [9] and once there, supports functions such as inhibition of 5’-end resection [9, 10], telomere position effect (TPE) [9, 11, 12], and intranuclear positioning of telomeres [13]. Moreover, yKu, by its interaction with the RNA component of telomerase, is important for telomeric DNA maintenance and nuclear localization of telomerase [14, 15]. While it is clear that in principle, yKu can directly bind at an end of double stranded telomeric DNA as well as a stem-loop structure on the RNA component of telomerase, most likely those interactions occur on the same interface on yKu, and therefore are mutually exclusive [16]. Moreover, there is evidence that Yku80 interacts with Sir4 [17, 18], and at least some yKu complexes may associate with telomeres via this indirect protein-protein interaction [16, 19].
As mentioned above, the differentiation of Ku-binding at DSBs which is instrumental for NHEJ and the binding mode on telomeres, where end-fusions must be avoided, is unknown. Previous results suggested a “two faces” idea for yKu’s association with DNA-ends [20]. In this model, most of the Yku80 side faces inward from the end and is essential for yKu’s telomeric functions. Yku70, facing towards the end, would be essential for yKu’s role in NHEJ [20]. Telomeric DNA is particular and composed of tandem repeats of G-rich sequences [21]. Budding yeast telomeric repeat DNA is 300 bp +/- 75 bp long (commonly abbreviated (C1-3 A)n−(TG1-3)n) and a number of proteins are associated with these repeats: Rap1 binds directly and with high affinity to a consensus sequence in the repeats, and Rif1 and Rif2 as well as the Sir2/Sir3/Sir4 proteins associate with telomeres via Rap1 [21]. Eventually, it is the resulting nucleoproteic structure that ensures the functions ascribed to telomeres [21]. However, in addition to their localization at chromosomal termini, in many eukaryotic species telomeric repeats are also present at internal genomic sites and have been dubbed interstitial telomeric sequences (ITSs) [22, 23]. In yeast sub-telomeric regions, ITSs are relatively frequent and they are thought to set the boundaries between different telomere-associated elements [24]. These elements include heterogeneous X elements that are found on all telomeres, with sizes varying between 0.3 kb to 3.7 kb [24–26]. Y’ elements, unlike the X elements, are found on about half of the telomeres, are much more homogeneous, and occur in two size classes, ~ 5.5 kb (Y’ short) and ~ 6.7 kb (Y’ long). Y’-elements can occur in tandem with 1 to 4 copies and, if present, they are positioned immediately next to the terminal repeats [25, 26]. The ITSs between these sub-telomeric elements vary between 50 to 150 bps (Fig 1, [24]). Importantly, telomeric repeats at chromosome ends and at ITSs are well characterized natural replication barriers, causing replication forks to stall at those sites [27–30]. Furthermore, there is direct evidence that such stalled or stressed replication forks are converted to DNA double-strand breaks (DSBs) [31]. For mammals, a very close association of ITSs with chromosome breakage has been documented [32, 33], and if not repaired adequately, these breaks will compromise genome stability and cell viability [34, 35].
In order to investigate how yKu can be bound at telomeres, and at the same time be prevented from mediating NHEJ, we used in vivo Chromatin Endogenous Cleavage (ChEC; [36]) coupled to Southern blots to pinpoint yKu’s localization. The results show that the yKu complex is found associated with telomeric repeats in or near ITSs and on terminal repeats that are distal from the physical ends of chromosomes. Consistent with previous results, a fraction of this internal yKu association is dependent on Sir4, but there clearly is also Sir4 independent binding. Remarkably, telomeric yKu can be trapped on an inducibly excised circular DNA molecule with telomeric repeats, but not if there are no telomeric repeats on it. Furthermore, by using an inducibly tagged yKu, the results show that new associations of yKu with ITSs are dependent on the passage through S-phase. These observations lead us to propose that on telomeres, yKu may be bound predominantly on internal repeat sites, allowing for the presence of telomeric chromatin in the portion of the telomeric repeats that is distal to yKu. This would occlude the Yku70-NHEJ side from the physical ends and explain why yKu binding at telomeres is very important for telomere integrity, while at the same time incapable of engaging NHEJ.
The ChEC method was developed in order to map the binding sites of proteins within their endogenous chromatin landscape [36]. The method is based on cleavage of native chromatin by Miccrococal Nuclease (MN) that is fused to proteins of interest. The actual DNA cleavage is induced by external addition of calcium, the concentration of which in a yeast nucleus normally is too low to activate the MN. Determination of actual cleavage sites is done by Southern-blotting (S1A Fig). Here, we intended to pinpoint positions of the yKu-complex on genomic loci of S. cerevisiae. As a control, we first constructed MN-Rap1, which had already been shown to be amenable to this technique [37]. Yeast telomeric repeats contain the highest affinity sites for Rap1 and on average, there are 15–20 Rap1 proteins on each yeast telomere [38]. However, the protein also recognizes sites in many transcriptional promoters [37, 39]. For a first assessment of in vivo ChEC, we performed experiments with MN-Rap1 and analyzed the HIS4 locus with one Rap1 binding site (Fig 1A, S1B Fig), the RPL21a locus with two sites (S1C Fig) and telomeres with many sites (Fig 1B). For the HIS4 locus, before Ca2+ addition, the XbaI restriction fragment detected is at ~ 12.0 kb, as expected (Fig 1A, S1B Fig). Within 2 min after Ca2+ addition, a new fragment of about 2.5 kb (*) was detectable. This fragment corresponds to a cleavage at the expected Rap1 binding site and progressively becomes the major fragment. At later time points, low-intensity fragments are also generated (white arrow in Fig 1A) and those correspond to MN-hypersensitive sites without specific Rap1 binding. Such a two tiered appearance of sites (early with specific Rap1 binding and later without Rap1 binding) is consistent with a previous report on ChEC with Rap1 [37]. Quite analogous results were obtained when the RPL21a locus with two Rap1 binding sites was analyzed (S1C Fig). Finally, MN-Rap1 binding at telomeres caused a fast disappearance of the terminal restriction fragment and the appearance of two new bands at ~ 910 bp and ~ 770 bp (Fig 1B). Of note, on the Y’-elements, about 950 bp separate the XhoI restriction site from the beginning of the terminal repeats, suggesting that the detected major cleavage via induced MN-Rap1 occurred near the transition between Y’ and terminal repeats (Fig 1B). In addition and as expected, MN-Rap1 also mediated cleavage in the ITS loci. Because the Y’-specific probe used here covers sequences on both sides of the XhoI site in the Y’-element, the detected internal Y’-fragments (either a full Y’-element with the ITS in case of a tandem Y’, or the X-ITS-Y’ fragments, see drawing in Fig 1B) were shortened to yield ITS-XhoI fragments (Fig 1B, * near 4.2 and 5.4 kb). As described before [37], longer induction of MN-Rap1 cleavage also yielded some non-specific fragmentation (see empty arrow, about 2.5 kb in Fig 1B). In order to discriminate between such non-specific cleavage sites and those induced by Rap1 binding to cognate sites, we compared the MN-Rap1 cleavage pattern with that produced with a GBD-MN, where MN is fused to the Gal4 DNA binding domain (Fig 1C). GBD-MN also created the non-specific 2.5 kb fragment and a number of new fragments that could correspond to nucleosomal arrays near the probe, i.e. generating very small sized fragments at the bottom of the gel. In contrast to when MN-Rap1 was used however, with GBD-MN we did not observe cleavage at ITS sites or at the sub-telomere-telomere junctions (Fig 1C). These results indicate that at yeast telomeres, MN-Rap1 does indeed induce specific cleavages within 100–200 bp of its binding sites on terminal and ITS telomeric repeats.
The precise location of the yKu complex on telomeres still is unclear. We thus wished to determine those sites using the above described in vivo ChEC method. The Yku70 protein was tagged with MN, creating Yku70-MN, and we analyzed telomeric cleavages by southern blot analyses as above. Without Ca2+ addition, the detected terminal restriction fragment (TRF) pattern of the strain was indistinguishable from a wild-type strain and we did not observe any increase in telomeric overhang signal, indicating that the fusion of MN to Yku70 does not impinge on yKu-function (Fig 2A). Upon MN induction, a very comparable TRF pattern as the one obtained for MN-Rap1 is observed: Yku70-MN cleavage generated 910 bp and 770 bp fragments, corresponding to a cleavage at the subtelomere-telomere junction and one about 140 bp distal to that junction. Remarkably, Yku70-MN cleavage was also detected near or on the ITS sequences between the subtelomeric repeats: the same two ITS-XhoI fragments as for the MN-Rap1 cleavage are detected in the upper area of the gel (* in Fig 2B). Previous studies already suggested an association of yKu with sequences in or near subtelomeric X-elements, which may have reflected yKu association with ITSs [40]. In order to confirm the yKu association with ITSs without the complication of a nearby X-element, we performed an experimentally independent approach to assess this yKu-ITS association. We chose to use chromatin immunoprecipitation (ChIP) using Myc-tagged Yku80 followed by q-PCR using primers that are specific for ITSs that occur between two Y’-elements on chromosome 12 (TelXIIL and TelXIIR; Fig 2B and 2C). As the ChEC results above suggested, these ITS loci are indeed efficiently immunoprecipitated when the Yku80 protein is tagged, but not if an untagged construct is used (Fig 2C, left). Furthermore, as will be shown below, yKu also associates with artificial ITSs on linear plasmids and ITSs that are far from the next telomeric region. Finally, DNA samples derived from ChEC analyses with MN-Rap1, Yku70-MN or GBD-MN were also analyzed by probing with a telomere repeat specific probe (Fig 2D). Consistent with the idea that Rap1 binds throughout on telomeric tracts, after 10 min of induction, ChEC with MN-Rap1 creates very short DNA fragments of less than 250 bp (Fig 1D, lane 3). In contrast, Yku70-MN induced cutting creates telomeric repeat containing fragments that seem to plateau at around 350 bp, even after 15 min of ChEC induction (Fig 1D, lane 7). The specificity of those cuts is underscored by the fact that ChEC with GBD-MN creates an entirely different pattern (Fig 1D, lanes 8–12), creating much larger fragments of over 600 bp that could correspond to what was called the telosome previously [41]. Collectively, these data confirm that yKu is specifically associated with telomeric repeat tracts. However, as opposed to what is expected from its end-binding property, on telomeres the yKu complex appears associated with repeats near the telomere-subtelomere junction and on telomeric ITSs.
Given this presence of yKu on sites relatively distant from the actual chromosome terminus and on ITSs, we wondered whether the reason for this association was direct DNA-binding or a possible indirect association. yKu is known to interact with Sir4 via the Yku80 subunit and there is previous evidence for two pools of yKu on telomeres: one that is bound directly on DNA and one that is associated indirectly via this Sir4-Yku80 interaction [17, 19, 20]. Moreover, there are YKU80 separation-of-function (SoF) alleles, which display a drastically reduced interaction with Sir4 and are dysfunctional in telomeric gene silencing, but are proficient in NHEJ and telomeric repeat DNA maintenance [17, 20]. These alleles thus are thought to be fully proficient in DNA binding. Finally, on telomeric DNA, Rap1 association with only Sir4 is sufficient to trigger the establishment of a specialized telomeric chromatin [42]. In order to assess a possible Sir4-dependence of the yKu-telomere interactions detected in our assays, we combined a sir4Δ allele or a YKU80 SoF allele with the yku70-MN allele and performed in vivo ChEC on these strains. Qualitatively, the Yku70-MN-mediated cleavage profile on telomeres is very similar in SIR4 and in sir4Δ cells (Fig 3A). After Ca2+ addition, the same two terminal fragments of 910 bp and 770 bp are generated as in the WT cells and the subtelomeric elements are also cleaved in the ITSs that separate them. However, cleavage efficiency at the different sites was reproducibly reduced in sir4Δ cells as compared to WT, in particular at early time points of MN induction (Fig 3B and S2A Fig). For example, 2 min after Ca2+ addition, Yku70-MN cleavage efficiency in the sir4Δ cells is 2–3 fold lower for both the 910 bp and 770 bp fragments as compared to the efficiencies observed in wild type cells (WT910: 19,37%; sir4Δ910: 9,39%; WT770: 9,79%; sir4Δ770: 3,70%).
The Yku80 α-helix 5 is essential for telomeric silencing and Sir4 binding [20]. Specifically, cells harbouring the yku80-L140A allele present a silencing defect and reduced Yku80-Sir4 interaction, but telomere length and NHEJ are not affected. Thus, in order for an independent assessment of the observations made with sir4Δ cells, we tested Yku70-MN ChEC cleavage in cells with this yku80-L140A allele. We used two strains derived from the Yku70-MN strain, both harbouring a yku80Δ allele at the genomic locus. yKu function was then re-established via plasmid borne expression of wild type Yku80 from its endogenous promoter or a plasmid borne expression of yku80-L140A. As was observed in the sir4Δ strains, the cleavage profile qualitatively was not affected in cells expressing the Yku80-L140A protein (Fig 3C). Moreover, cleavage efficiencies were similarly reduced as in the sir4Δ strains (Fig 3D and S2B Fig). These findings with the ChEC technique were confirmed by ChIP with q-PCR: immunoprecipitation of ITS loci in sir4Δ cells was significantly reduced when compared to SIR4 WT cells (Fig 2C). In these ChIP experiments, we also used a strain expressing a Myc-tagged Yku80Δ36 protein, which is unable to bind any nucleic acid (either DNA or RNA)[16]. The immunoprecipitates with this protein did still contain ITS loci, albeit in reduced amounts when compared to the amounts detected with wt tagged Yku80 (Fig 2C). Finally, when we expressed the Myc-tagged Yku80Δ36 protein in sir4Δ cells, the ChIP signals were reduced to background levels. These results show that yKu associates with internal telomeric repeats in two ways: either by direct DNA binding or via an indirect Sir4-mediated association. The reduced ChEC cleavage in sir4Δ cells above therefore is due to a reduced presence of yKu on telomeric repeats, but the yKu-complexes still remaining are directly bound on the very same sites within telomeric repeat DNA.
It could be argued that the reduced cleavage reported above was due to an altered chromatin configuration at telomeres, but not due to a loss of the specific Yku80-Sir4 interaction. In order to investigate this possibility, we performed Yku70-MN-mediated ChEC in a strain harbouring a sir2Δ allele. Sir2 is a conserved NAD+ dependent histone deacetylase [43–45] that, together with the Sir3 and Sir4 proteins, is required for the specialized chromatin at telomeres [21, 42, 46]. The cleavage profile in sir2Δ cells again is similar to that observed in wild type cells (Fig 3E). However, as opposed to what was observed in the sir4Δ strains, cleavage efficiencies for the 910 bp and 770 bp fragments only decreased marginally and the decrease for the most part was not statistically significant (Fig 3F and S2C Fig). Furthermore, we also analyzed MN-Rap1 mediated cleavage in sir4Δ cells. As expected, there were no qualitative or quantitative differences in the cleavage patterns observed between SIR4 and sir4Δ cells (S3A and S3B Fig).
Altogether, these observations are in line with previous results that suggested that the Yku80-Sir4 interaction is important for yKu-mediated roles in chromatin related functions, but not for direct binding of yKu on telomeric DNA [47, 48]. Hence, the Sir4-independent Yku70-MN mediated cleavages we detect on telomeric chromatin are due to yKu being bound on DNA.
The above observations predict that at least part of yKu was in fact not bound at the very ends of chromosomes, but rather at internal sites of telomeric repeat tracts. In order to verify this prediction, we used a telomeric repeat flip-out system that should trap internally bound yKu on a circular DNA molecule, while yKu associated with the distal-most part of the telomere would remain on the chromosome, even after flip-out (see Fig 4 and [49]). We thus constructed strains in which the extremity of chromosome VIIL is modified accordingly and that also contained the yku70-MN allele. In addition, the strains contained a copy of Pgal10-FLP1 integrated in the LEU2 locus on chromosome III, which allows for a galactose-inducible Flp1-recombinase expression. The first strain, MVL022, has the URA3 gene flanked by the two Flp1-recognition target sites (FRT) (ChrVIIL-0 block, see Fig 4A) and the second strain, MVL023, has an additional TG1-3 telomeric tract of 270 bp between the first FRT site and the URA3 marker gene (ChrVIIL-1 block, Fig 4C). Flp1 recombinase induction by addition of galactose causes recombination between the repeated FRT sites and all sequences between the FRT sites will end up on an excised circular DNA molecule. In MVL022, this circular molecule will contain the URA3 marker and an FRT site, while in MVL023, it also contains the internal TG1-3 telomeric tract of the original telomere VIIL. Therefore, upon MN induction and digestion of the DNA with StuI, the latter linearized fragment with be further cut only if yKu was associated with internal telomeric repeats, but not, if it was localized exclusively in the most distal portion of the telomere. Both strains MVL022 and MVL023 were incubated in media containing 2% galactose to induce Flp1 expression or kept in 2% raffinose as non-induced controls. In addition, part of the cultures was maintained in G1 by adding α-factor for 1.5 hours prior to Flp1 induction. In vivo ChEC was performed on all strains, followed by DNA analyses on southern blots (Fig 4B and 4D). In both strains and all conditions, the URA3 probe detects a fragment at 1875 bp which corresponds to the StuI fragment from the endogenous genomic URA3 locus (-c in Fig 4B and 4D). For strain MVL022, when Flp1 is not induced, the fragment at 1560 bp corresponds to the restriction fragment between the two StuI sites on the modified ChrVIIL (marked with Θ, Fig 4B). After galactose addition, a new fragment appears at 1220 bp corresponding to the StuI linearized form of the circular molecule (marked with +, Fig 4B). Addition of Ca2+ and induction of the MN did not change this pattern, even after 20 min of induction. This suggests that in the absence of telomeric repeats, the yKu complex does not associate with sequences in between the two FRT sites on the modified ChrVIIL. For strain MVL023 that contains a block of 270 bp telomeric repeats between the FRT sites, a fragment at 1494 bp corresponding to the StuI-linearized from of the circular molecule can be detected after Flp1-induction by galactose (marked +, Fig 4D). In addition and in stark contrast to strain MVL022, Ca2+ addition to these cells generates a new fragment at ~ 750 bp (see * in Fig 4D), which matches a predicted fragment, if Yku70-MN mediated cleavage occurred in or near the inserted TG1-3 repeat tract in the circular molecule. Given that a circular DNA molecule has no physical ends for yKu to bind to, we conclude that the yKu complex was already associated with the TG1-3 tract before circular molecule excision. In addition, we also performed this experiment in sir4Δ cells in order to exclude a protein mediated association of the excised circular DNA with telomeres. Consistent with the above Sir4-independent association of yKu with telomeric repeats, the FRT mediated recombined circular fragment with telomeric repeats is cleaved after ChEC induction and this cleavage is dependent on the presence of telomeric repeats (S3C Fig).
Unexpectedly, in the above assays, yKu was found to be bound on the excised circular DNA even if cells were arrested in G1, before excision of the circular DNA (see Fig 4D, lanes marked G1). In order to distinguish whether yKu could somehow associate with internal telomeric repeats during G1 or whether this yKu detection in G1 reflected trapped yKu from the last passage through the cell cycle, we mounted a system in which only new associations of yKu with telomeric repeat DNA are detected (see S4 Fig). In essence, the system is based on an inducible tagging of the Yku80 protein via site-directed recombination of two RS sites. This recombination is mediated by the bacterial RecR protein which in our case is expressed from the conditional gal promoter (S4A Fig). Thus, in cells grown with glucose or raffinose, there is a stop codon on YKU80 ORF before the myc-peptides, the locus remains intact (S4B Fig) and no Yku80 is detected on a western blot probed with anti-myc antibodies (S4C Fig). However, 16 hrs after induction of the RecR protein by the addition of galactose to the media, most of the locus had recombined (S4B Fig) and Yku80-myc is now detectable on the western (S4C Fig). In addition to the wt YKU80 ORF on plasmid pEP22B, we also created the same taggable situation for the yku80L140A allele on plasmid pEP24C (S4D Fig). As a positive control for yKu association in the situations studied, cells also contained the plasmid YCpHOCut4, on which the HO-endonuclease is expressed from a galactose inducible promoter and an HO-cutting target sequence is integrated as well [50].The global cellular genetic make-up before the experiment is outlined in S4D Fig and the work-flow in S4E Fig. We thus assessed yKu binding after cells were arrested in G1 with α-factor, the myc-tagging of Yku80 induced by addition of galactose, followed by qChIP on ITS sequences on chromosome XII (see S4D Fig). If cells were allowed to grow after the in vivo tagging of Yku80, yKu could be found on ITSs as well as on the HO-cut plasmid, the positive control (lanes cycling + myc in Fig 4E and S4F Fig). However, if cells were retained in G1, the signal for yKu binding to ITSs remained as low as the untagged background (Fig 4E), even though the positive control clearly could be detected (S4F Fig, lane “Arrested +myc”). Arrest in G1 or release into the next cell cycle of the cell cultures was controlled by FACS analyses (Fig 4F). Similar results were also obtained with the yku80L140A allele (Fig 4E), even if the ITS binding by this protein during the next cycle was significantly lower than wt. Altogether, these results show that yKu is unable to associate with ITSs during G1 and that a passage through the next S-phase is required for this to happen.
ITSs are known to be hot spots for the initiation of genomic rearrangements [32–35]. Previous results also reported replication fork stalling leading to double-strand breaks and chromosomal rearrangements due to telomeric repeat tracts [27, 28, 30]. We therefore surmised that the above results could be the consequence of DNA breaks occurring at ITSs during replication fork passage. In order to investigate this possibility, we analyzed yKu binding onto a specific and unique ITS engineered onto linear plasmids derived from plasmids YRpRW41 and YRpRW40-2 (Fig 5A; S4A Fig; [51]). The two linear constructs differed in the location of the origin of replication (Fig 5A) and hence, the directionality the replication fork is moving through the ITS. These plasmids were transformed into a strain with Yku70-MN, the MN induced by Ca2+ addition and the integrity of the 1.4 kb StuI-XhoI restriction fragment encompassing the ITS was analyzed by southern blotting (Fig 5B). The blots revealed three new fragments of 1160 bp, 1060 bp and 915 bp that were generated in a Ca2+ dependent fashion in both strains. All three sites map very close to, or within, the ITS tract, as indicated with * on Fig 5A. We also performed ChEC analysis with GBD-MN on these same plasmids and as expected did detect some non-specific sites that are cut by GBD-MN (Fig 5D, empty arrowheads). However, these non-specific sites mapped to quite distinct locations that differed from the Yku70-MN sites on the fragment (Fig 5A). Moreover, for both plasmids the cleavage profiles obtained with Yku70-MN are virtually the same and cleavage rates are also quite comparable with only a slight but not statistically different increase for YLpRW41 (Fig 5C, S4B Fig). These results are entirely consistent with previous physical studies that have mapped orientation-independent fork stalling due to ITSs [30]. This fork stalling thus may create one-sided breaks onto which yKu can load.
All the ChEC analyses above concerned yKu associations with sites that contained telomeric repeats relatively close to an actual telomere, either terminal repeats or ITSs within about 10 kb of a telomere. If indeed telomeric repeats and associated proteins are replication fork barriers, they should cause replication blocks anywhere in the genome. In order to test this prediction, we inserted a plasmid containing either a 260 bp block of telomeric repeats (+ITS) or no such repeats (-ITS) into a non-telomeric area near the HIS3 locus on chromosome XV, about 300 kb away from the telomere on XV-L (Fig 6A and S6A Fig). After performing ChEC with Yku70-MN and probing this locus, DNA cleavage at the predicted site was detected only in the +ITS situation (Fig 6B and S6B Fig). Furthermore, the cutting at this artificial ITS was specifically mediated by yKu, since when the ChEC was performed with GBD-MN, no cleavage near this constructed ITS was observed (Fig 6C). Finally, an ITS-dependent signal after Yku70-MN ChEC was also detected at this site in cells with a sir4Δ allele (Fig 6D). These data thus confirm the association of yKu with loci in which ITS occur and also show that a nearby telomere is not required for this association.
The Rrm3 and Pif1 helicases have been proposed to facilitate replication fork passage through telomeric repeat sequences and without them, fork stalling appeared more prevalent [29, 30]. However, Yku70-MN mediated cleavage near fork stalling sites was not increased in either pif1Δ or rrm3Δ strains (Fig 7A and 7B, S7A Fig). Strains with deletions of Tof1 or Sml1, although also predicted to be more susceptible to fork disassembly, displayed only marginally increased cleavage efficiencies as compared to wt, and these differences were not statistically significant (Fig 7C and 7D; S7B Fig). These results suggest that while actual fork stalling at ITS sequences may be sensitive to repeat orientation and replisome stability, the overall frequency of converting the stall to a one-sided break is not.
Previously it was suggested that after yKu binding onto a DSB, 5’-strand resection mediated by the Mre11/Rad50/Xrs2 complex in preparation for homologous recombination may remove yKu from the DNA end [52]. If this was the case as well for the one-sided breaks that are expected to occur near replication arrests, in the absence of Mre11 we expected to observe an increase in Yku70-MN mediated cleavages near telomeric repeats. Hence, we constructed a strain harbouring an mre11Δ allele which did display short telomeres, as expected (Fig 7E, lane 0’). However, the Ca2+ dependent generation of the short telomeric fragments was not increased (Fig 7F and S7C Fig). If anything, there was a slight decrease in cleavage efficiency such that after 2 min with Ca2+, efficiencies for the two fragments were WT910: 14,7%; WT770: 7,1%; mre11Δ910: 9,3%; mre11Δ770: 4,2%. These results suggest that the Mre11/Rad50/Xrs2 complex does not play a role in yKu-release from the DNA sites near telomeric repeats analyzed here. Finally, strains harbouring a sgs1Δ allele or a combination of sgs1Δ with sir4Δ displayed slightly decreased cleavage efficiencies, albeit again not in a statistically significant manner (Fig 7G and 7H, S7D Fig).
The DNA binding complex Ku binds to dsDNA ends without any sequence specificity. This association occurs at a physical end of a DNA molecule and the DNA end will pass through a ring-like opening of Ku [3]. In the DNA-bound configuration, the majority of the Ku70 protein faces the side that is proximal to the DNA end, while the surface of the Ku80 protein faces towards the other side [3]. Previous data from budding yeast also suggested that this orientation had functional consequences: systematic screenings of mutations in both subunits showed that it is the Yku70 protein that is the major determinant for mediating NHEJ, which involves the physical DNA end side [20]. However, Ku also associates with telomeres in many organisms, including humans and yeast [2, 9]. At this location, NHEJ-induction could cause chromosome fusions with ensuing genome instability which must be avoided. Yet, how exactly NHEJ-induction by Ku is prevented at telomeres remains unknown.
In budding yeast, yKu is associated with telomeres in two ways: either the complex is bound directly to DNA, as on any DNA end and as described above, or it is associated indirectly via an interaction between Yku80 and the telomeric chromatin component Sir4 [17, 20, 47]. Previous results do show that yKu must be bound to the DNA directly in order to mediate NHEJ and the telomeric capping functions ascribed to it [16, 47]. Furthermore, yKu is associated with telomeres even in sir4Δ cells [40, 47] and the question of how NHEJ is prevented at that location remains.
Our data here show that yKu binding on telomeres can occur at sites that are distal from the physical ends of chromosomes, regardless of whether the cells contained Sir4 or not (Figs 2, 3 and 6). The sites that can be detected using the ChEC assay are near the telomeric repeat to subtelomeric DNA junctions and on ITSs (Figs 2 to 6). As expected from a general phenomenon not dependent on a specific genomic locus, yKu binding to ITSs was also detected on a chromosomal internal site (HIS3 locus), far from a telomeric region (Fig 6). The fact that these internal associations are based on direct DNA binding is underscored by ChIP assays in which the signal is only completely lost if both, the ability of binding DNA (the yku80Δ36 allele) and the interaction with Sir4 (in sir4Δ cells) are removed (Fig 2C). Consistent with this new placing of yKu on telomeres, the yKu complex can be detected on excised circular DNA that does not contain the most distal part of a modified telomere VIIL (Fig 4). This proposed localization of yKu is unexpected because it was assumed that yKu would bind to the telomeric DNA from the very ends of the chromosomes for its telomeric functions (see for example [16, 47]). We consider it highly unlikely that the detected internal binding reported here is due to end binding and then sliding of yKu on the DNA to its final position. This is particularly so for the binding detected on the ITSs, which would require yKu sliding on chromatinized DNA in vivo for at least 4 kb, or for about 300 kb in the case of the artificial ITS constructed at the HIS3 locus (Figs 1 and 6). We therefore think it more plausible that yKu associates on DNA ends that were generated near the sites detected.
The above raises the questions of how and why a DNA end is generated at ITSs and at the beginning of the terminal telomeric repeats. It is well documented that telomeric repeat sequences, including ITS tracts, can be major obstacles for the passage of a replication fork [27–30]. Furthermore, there is now direct evidence that stalled or stressed forks will generate a DNA double stranded break [31]. In line with this evidence, we propose that during S-phase, stalled replication forks near or in telomeric repeat tracts could reverse and/or be subject to strand breakages that would create what is dubbed a one-sided DSB (Fig 8). yKu could then bind those ends via its canonical binding mode on DNA. Depending on the precise end-structure generated by the break, the presence of yKu on them could prevent extensive resection but perhaps still mediate the initiation of break induced replication [53] or repeat extension by telomerase, which would secure the re-establishment of a functional telomere distal of the break. Our data also show that new yKu associations with ITSs requires that cells are growing and such new associations do not occur during G1 (Fig 4E and 4F, S4 Fig). Consistent with these results, in vitro binding studies showed that yKu cannot associate with a DNA end to which the telomeric capping protein Cdc13 was pre-associated [54]. Moreover, only when Cdc13 is actively degraded and removed from telomeres during G1 is there an end-stabilising effect exerted by yKu [55], as would be expected from the in vitro results [54]. However, yKu does not protect telomeric ends from degradation during late S-phase, when telomeric replication occurs in vivo [21, 55]. Formally, we cannot completely exclude the possibility that yKu associates with the non-terminal sites via an association that is dependent on as of yet unknown protein-protein interactions that do not involve Sir4. Arguing against this possibility is the finding that in sir4Δ cells that harbour the yku80Δ36 allele [16], a YKU80 allele that reconstitutes a yKu complex that is unable to bind to DNA, the yKu association with ITSs is completely lost (Fig 2C).
Remarkably, we can detect yKu at internal sites even if cells are in G1 of the cell cycle (Fig 4C and 4D). Given that new yKu associations with ITSs do not occur in G1 (Fig 4E and 4F), these results suggest that at least on the telomeric sites analyzed here, yKu removal is inefficient. Consistent with this idea, detection of yKu on a non-telomeric ITS appears much lower than that on telomeric ITSs (compare Figs 2B, 3A with 6B). We ignore the reason for these differences but suggest that different chromosomal locations may be differentially susceptible for yKu removal. This idea has precedence as phosphorylated H2A also appears to have a much longer persistence time in subtelomeric areas [56, 57]. In fact, the strong correlation of γ-H2A accumulation with replication barriers in telomeric areas [56] thus correlates with the accumulation of yKu on these same sites and reinforces our model (Fig 8). Recently it was proposed that yKu bound on DNA ends on a DSB would be removed by the nuclease activity of the MRX-Sae2 complex in preparation for homologous recombination (HR)[58]. Our data show that an absence of the Mre11 protein does not influence Yku70-MN mediated DNA cutting at telomeric sites (Fig 6). This finding correlates with the fact that HR is actively suppressed at telomeric loci [40, 59], a suppression that is lost in cells that lack yKu [10]. Furthermore, the nucleolytic activity of Mre11 is not required for its telomeric functions [60]. These considerations are consistent with our model that predicts that the nucleolytic activity of MRX-Sae2 for HR initiation is repressed on telomeric one sided breaks, essentially leaving yKu on the DNA. We do not know whether the MRX-complex still associates with these ends and how the potential ensuing BIR events are induced in this situation, but a recent study implied an MRX-tethering function may be important for this later step [61]. In our case, this MRX-mediated tethering as well as the possible recruitment of telomerase would be independent of the nuclease activity of Mre11. While not statistically significant, there is a trend for increased Yku70-MN cleavage in tof1Δ cells and a decrease of similar extent in sgs1Δ cells (Fig 7D, S7B Fig; Fig 7H and S7D Fig). These two genes had been found to be involved in replication barrier efficiency, albeit in different types of pathways [27]. The ChEC assay is relatively complex and not ideal for documenting smaller differences in cutting efficiencies. The significance of the above observations on TOF1 and SGS1 therefore remains unclear.
Our model hence posits an important function of yKu at replication barriers at the transition between non-telomeric and telomeric repeats DNA (Fig 8). Its binding could suppress extensive 5’-resection and mediate fork stability / fork restart by BIR, or the binding of Cdc13 and recruitment of telomerase for repeat expansion. It is important to note that yKu is not expected to mediate NHEJ in this situation: first, this would be a yKu association with a one-sided break and hence, another end for a fusion reaction is not readily available. Second, the regulatory networks directing DNA double-strand break repair choice strongly favour HR over NHEJ during late-S-phase [52]. We note that there is previous evidence for very similar non-canonical functions of Ku in fission yeast [62]. A Ku mediated stabilisation of one sided breaks occurring near telomeres also impinges on a very important conundrum in the field, namely that of accommodating Ku-binding at telomeres and at the same time complete repression of NHEJ involving chromosome ends. If yKu is trapped on DNA relatively distant from the physical end, the binding of Rap1 proteins between yKu and the actual end could prevent yKu mediated NHEJ at telomeres. This in turn would also explain why Rap1 has a strong NHEJ repressing effect [63]. However, our results do not directly address the issue of whether or not there is yKu binding at the physical ends of chromosomes, where the Cdc13/Stn1/Ten1 complex in conjunction with a specialised Rap1-based chromatin provide for the essential capping function.
Full genotypes of all strains are described in S1 Table. We constructed yeast strains EPY007 and MVY221 expressing MN-Rap1 and Yku70-MN respectively by fusing the enzymatic activity domain of micrococcal nuclease (MN) from Staphylococcus aureus to the N-terminus of the Rap1 protein or the C-terminus of Yku70. NruI linearized plasmid pRS306-MN-Rap1 was transformed into a diploid wt strain (W303). Cells that had lost the URA3 marker were then selected by restreaking one isolated colony on an FOA plate. The resulting diploid strain was sporulated and clones expressing the fusion protein were identified. Strain Yku70-MN was obtained with PCR based mutagenesis using primers flanking by the C-terminal sequences of YKU70, F2-Hdf1 and R1-Hdf1 with plasmid pFA6a-MN-TRP1 as template. The fragment was used for transformation of diploid strain MVY60 which subsequently was sporulated and clones expressing the Yku70-MN allele were identified. All strain constructs were verified by southern blotting. For expression of GBD-MN, strain W3749-1a was transformed with the replicative pRSE plasmid that contains the TRP1 marker and the Gal4 DNA binding domain fused to Micrococcal nuclease ORF, the gene being transcribed from the yeast GAL1 promoter. Strain MVL013 was derived from MVY221 in which the BAR1 gene was replaced by a natMX4 by PCR-mediated gene disruption [64, 65].
Strains MVL022 and MVL023 were derived from MVL013 in two steps. First we integrated a construct such that the 2μ-Flip protein could be induced by galactose using plasmid pFV17 [66]. This strain was then transformed with the linearized plasmids sp225 and sp229 to obtain respectively strains MVL022 and MVL023, as described previously [49]. Strains MVL047 and MVL048 were derived from MVL013 by replacing the SIR4 gene by the kanMX4 [67]. Strain MVL052 was derived from MVL013 by replacing the wt YKU80 gene by the kanMX4. Strain MVL052 was then transformed with either the pML7c-2 plasmid, which contains the wild type YKU80 locus including its endogenous promoter, or the pML7c-14 plasmid, which contains the yku80L140A allele with its native promoter. Strain MVL010 was derived from MVY221 by replacing the SIR2 gene with the natMX4 deletion cassette [64]. Strain MVL054 was derived from EPY007 by replacing the SIR4 gene with the kanMX4 deletion cassette. For analysis of linear plasmids, BamHI-linearized plasmids YLpRW40-2 and YLpRW41 [51] were transformed into MVL013 or into W3749-1a + pRSE. Clones were selected on Yc-URA-LEU plates or on Yc-URA-LEU-TRP plates. Strains MVL030, MVL031, MVL032 and MVL033 were derived from MVL023 by replacing, respectively, the PIF1, RRM3, SML1 and TOF1 genes with the kanMX4 deletion cassette. Strain MVL063 was derived from MVL013 by replacing the MRE11 gene with a HIS3 auxotrophic marker [68]. Strains EPY050 and EPY052 were derived from, respectively, MVL013 and MVL047 by replacing the SGS1 gene with a URA3 deletion cassette. Strains EPY027, EPY028 and EPY031 were derived, respectively, from MVL023, MVL022 and W303, by replacing the SIR4 gene with the kanMX4 deletion cassette. All modifications in the genome of the above mentioned strains were verified by colony PCR and Southern blotting using a probe that hybridizes to the promoter of the respective gene. Strains EPY054, EPY058, EPY061 and EPY064 were obtained from, respectively, strains W303, EPY031, MVL013 and MVL047, transformed with NheI linearized plasmid pRS303. Strains EPY056, EPY059, EPY063 and EPY066 were obtained from, respectively, strains W303, EPY031, MVL013 and MVL047, transformed with NheI linearized plasmid pEP19A. Strains EPY054, EPY058, EPY056 and EPY059 were then transformed with pRSE plasmid.
IDY80-1 strain was derived from MLY30 in which the YKU80 gene was replaced by the LEU2 gene. The SIR4 gene was replaced in this strain by natMX4 cassette, to obtain IDY82-9 strain. Strains IDY80-1 and IDY82-9 were transformed with either pJP7c (YKU80-myc) or pJP12 (yku80Δ36-myc) to perform ChIP experiments. Inducible myc tagging of Yku80 experiment was done in strain IDY80-1 transformed with YCpHOCut4 and either pEP22B or pEP24C.
All plasmids are described in S3 Table. pRS306-MN-Rap1 was derived from pRS306 [68] by insertion of three fragments; i) the gene-proximal last 490 bp of the Rap1 promoter (Rap1 promo), ii) a DNA fragment encoding the enzymatic domain of micrococcal nuclease, iii) the first 500 bp of the RAP1 coding region (Rap1-500). The Rap1 promo and Rap1-500 fragments were amplified by PCR from genomic DNA with the following primers; Rap1 Promo XhoI For/ Rap1 promo ClaI Rev and Rap1 500 bp ClaI For/ Rap1 500 bp EcorI Rev respectively (see S2 Table for details on all primers). First, both of these fragments were integrated simultaneously into the EcoRI-XhoI sites of pRS306. A second cloning step permitted to integrate the MN-encoding fragment into the ClaI restriction site. This latter fragment was amplified by PCR from pFA6a-MN-TRP1 [36]. pRSE was derived from the pRS314-Cre-EBD plasmid. First, pRS314-Cre-EBD was obtained by inserting the Gal1-Cre-EBD fragment from pSH62-EBD [69] into the SacI-EcoRI sites of pRS314. The Cre-EBD fragment was removed from pRS314-Cre-EBD by EcoRI-SalI digestion. The fragments encoding the Gal4 DNA binding domain (GBD) fragment and the MN were inserted into this plasmid by Gibson Assembly [70]. pML7c-2 and pML7c-14 plasmids were derived from pJP7c and pJP7c-L140A plasmids respectively. pJP7c is derived from the pJP7 plasmid [16] in which a point mutation had to be corrected. Essentially, these two plasmids comprise the pRS313 backbone into which either the wild type YKU80 locus or the yku80L140A allele with its native promoter were inserted. Both proteins were tagged with two Myc and ten HIS tags. YRpRW40-2 was derived from YRpRW40 by correcting the internal telomeric repeat tract to be the same as in YRpRW41. pEP19A was derived from pRS303. A 256 bp telomeric track was integrated between the XbaI and BamHI sites. pEP22B and pEP24C were derived from pJP7c and pJP7c-L140A respectively by insertion of the bacterial recR gene transcribed from the yeast GAL1 promoter. The YKU80 alleles also contained two RS sites upstream of the Myc and HIS tags (see S4 Fig). Nucleotide sequences of these plasmids are available upon request.
All culture growth was at 30°C in standard yeast cell growth conditions (YEP media with indicated carbon sources or in drop-out media). Strains with pRSE plasmid were pre-grown in Yc-TRP media with 2% raffinose to stationary phase. Cells were then diluted and grown in Yc-TRP with 2% galactose for 45 minutes up to 3 hours. Strains MVL022 and MVL023 were pre-grown in YEP media with 2% raffinose to stationary phase, the culture it was diluted in YEP media with 2% raffinose and re-grown to an OD660 of ~ 0.4. A first asynchronous/FLP non-induced aliquot was left to grow to an OD660 of ~ 0,6. A second aliquot was grown with 2% galactose to an OD660 of ~ 0,6 (asynchronous/FLP induced sample). To a third fraction, we first added α-factor (final 0.1 μM) for 90 mins. The culture was verified for G1 arrest by FACS analysis. After this treatment, one aliquot was grown with 2% galactose to an OD660 of 0,6 (synchronous/ FLP induced sample). The rest of the culture was left to grow to an OD660 of ~ 0,6 in the presence of glucose (synchronous/FLP non-induced sample). The ChEC assay was then performed on all samples.
From an overnight pre-culture, cells were diluted into 100 ml media and re-grown to an OD660 of about 0.6–0.8. Cells were harvested and washed three times in 1 ml A-PBPi buffer [36]. Cells were permeabilized in 600μl Ag-PBPi buffer for 5 min at 30°C. For MN-cleavage, CaCl2 was added to a final concentration of 2 mM and cells incubated at 30°C. A first aliquot was taken before Ca2+ addition for time point 0, and the next aliquots were removed at indicated time points after the addition of Ca2+. Aliquots were immediately mixed with an equal volume of a 2X STOP solution (400 mM NaCl; 20 mM EDTA; 4 mM EGTA; 0.2 μg/μl glycogen).
Cells were mechanically broken using glass beads and DNA extraction was realized as described previously [36]. Appropriate quantities of DNA were digested with indicated restriction enzymes, separated on 0.6% TBE agarose gels, transferred on a Hybond-XL nylon membrane (Amersham) and detected by hybridization with 32P-labelled radioactive probes. 500 ng of digested DNA was loaded on gels for hybridization with Y’-specific probe and telomeric repeats probe, and up to 2.5 μg for hybridization with other specific probes. Blots were analysed using the Typhoon FLA 9500 from GE Healthcare Life Sciences. Band intensities for cleavage efficiencies were quantified with Image quant software. For each fragment, the cleavage efficiency percentage is calculated with respect to total signal at each time. Cleavage efficiency = (fragment signal (tX)/ total signal) *100. Native in-gel analysis was performed as described [71]. As controls, DNAs derived from a wild type or a strain with a yku70Δ allele were used. After hybridization and washings, the gel was exposed to MP-high performance film (Amersham) for appropriate times. For loading controls, the DNA was then transferred to Nylon membranes which were hybridized to a probe with telomeric repeats.
Chromatin immunoprecipitation (ChIP) experiments were performed essentially as described [72] with some modifications. Briefly, cells were grown to an OD600 of 0.5–0.6. Formaldehyde solution (37%) was added to a final concentration of 1% and cells incubated for 20 min at room temperature. Cell pellets from 50 ml cultures were resuspended in 500 ml of lysis buffer containing proteases inhibitors and disrupted vigorously with glass beads three times for 30s using a FastPrep-24 (MP Biomedicals) instrument. Samples were then sonicated 10 times for 10s at 20% power using a Branson digital sonifier. Whole-cell extracts were incubated with anti-myc (9E10, Roche) antibody overnight at 4°C, and precipitated with Pro-A/G Magnetic Beads (Pierce) for 1 hour at 4°C. Quantification of the immunoprecipitated DNA was accomplished by quantitative real-time PCR, employing the SYBR Green (Life Technologies) system. Immunoprecipitated DNA was normalized to input samples to calculate the percentage of input DNA that was precipitated. Control qPCR assays were targeted to the CLN2 locus to demonstrate non-amplification of non-target loci.
For the arrested and galactose induced analyses, cells were pre-grown in raffinose and arrested with 0.1 μM α-factor (Sigma), arresting for 4 hours. Half the culture was washed and released into media containing Pronase (Roche) and galactose. To the remaining half of the culture, galactose was added. Galactose induction proceeded for 10 hours before cells were harvested for ChIP as described above.
Whole cell protein extracts were prepared as previously described [73]. Samples were analyzed on 8% SDS-PAGE followed by electroblotting onto HybondECL membrane (GE-Healthcare). Membranes blocked in 5% milk/PBS-T were incubated in 1:1000 monoclonal rabbit anti-Myc antibody (Cell signalling) diluted in 1% milk/PBS. Secondary antibodies were donkey anti-rabbit (GE Healthcare), diluted 1:5000 in 1% milk/PBS. Blots were visualized and analyzed on a LAS-4000 (GE Healthcare).
|
10.1371/journal.pgen.1005090 | LRGUK-1 Is Required for Basal Body and Manchette Function during Spermatogenesis and Male Fertility | Male infertility affects at least 5% of reproductive age males. The most common pathology is a complex presentation of decreased sperm output and abnormal sperm shape and motility referred to as oligoasthenoteratospermia (OAT). For the majority of OAT men a precise diagnosis cannot be provided. Here we demonstrate that leucine-rich repeats and guanylate kinase-domain containing isoform 1 (LRGUK-1) is required for multiple aspects of sperm assembly, including acrosome attachment, sperm head shaping and the initiation of the axoneme growth to form the core of the sperm tail. Specifically, LRGUK-1 is required for basal body attachment to the plasma membrane, the appropriate formation of the sub-distal appendages, the extension of axoneme microtubules and for microtubule movement and organisation within the manchette. Manchette dysfunction leads to abnormal sperm head shaping. Several of these functions may be achieved in association with the LRGUK-1 binding partner HOOK2. Collectively, these data establish LRGUK-1 as a major determinant of microtubule structure within the male germ line.
| Male infertility affects one in six couples in western societies and approximately half of these are the result of male factor disorders. The most common clinical presentation for male infertility is a complex mixture of abnormal sperm output, shape and motility referred to as oligoasthenoteratozoospermia (OAT). In an effort to define an origin of OAT we have analysed a mouse model of leucine-rich repeats and guanylate kinase-domain containing isoform 1 (LRGUK-1) dysfunction. Herein we show that LRGUK dynamically redistributes during the process of haploid germ cell maturation (spermiogenesis) and that LRGUK-1 function is required for multiple aspects of sperm centriole and tail development and sperm head shaping. Further, we have identified HOOK2 as a novel LRGUK-1 binding partner, thus raising the possibility that several aspects of LRGUK-1 function are achieved in partnership with HOOK2.
| Male infertility affects at least 5% men of reproductive age in the western societies [1]. Normal male fertility requires sufficient numbers of morphologically normal and motile sperm [2]. Oligoasthenoteratozoospermia (OAT), is the term used to describe semen containing low numbers of sperm with poor motility and abnormal shape, and is the most common clinical phenotype in human male infertility [3]. The underlying aetiology in the majority of men presenting with OAT is unknown, and as such, there remains an absence of specific treatments, an absence of knowledge of associated somatic pathologies, and potential consequences for offspring conceived with OAT sperm via assisted reproductive technologies (ART).
The morphological and motility aspects of OAT likely have their origins in spermiogenesis, the process wherein round haploid germ cells are transformed into highly polarized sperm with the potential for motility and fertility. This process takes approximately two weeks in the mouse and involves several thousand different gene products [4]. Three of the major aspects of spermiogenesis are acrosome development, head shaping, and growth of the sperm flagellum [5,6]. These events are critically reliant upon complex microtubule structures, including the manchette and the axoneme, and a highly orchestrated series of protein transport mechanisms [7]. The manchette is a transient microtubule structure, which encircles the spermatid nucleus during the initial steps of elongation, and has a role in sculpting the species-specific nucleus shape. Abnormalities in manchette structure result in dysmorphic sperm heads [8–10]. We note that dynamic redistribution of the actin cytoskeletal system is also required for normal manchette function [8].
The axoneme forms the core of the sperm tail. It is composed of nine microtubule doublets surrounding a central microtubule pair (9+2) [11]. In contrast to motile cilia on other cells, the sperm tail axoneme is sheathed by accessory structures required for the production of ATP and protection against shear forces [12]. These structures include the outer dense fibers, fibrous sheath and mitochondrial sheath. Sperm tail axonemal development begins in round spermatids with the maturation of the mother centriole into a basal body, followed by plasma membrane attachment and axoneme extension. Several studies suggest that the extension of the basal body into an axoneme is dependent on the intra-flagella transport (IFT) system [13]. We note, however, that while this process is well defined in primary cilia [14], its role in sperm tail biology requires more in-depth analysis. Defectives in germ cell axoneme formation leads to either an absence of a sperm tail or immotile sperm.
Interestingly, research is increasingly indicating continuity between processes governing the development of the manchette and the sperm tail [13,15]. This may explain the extremely common association between abnormal sperm head morphology and motility in infertile men and mice [16]. Indeed, pioneering research from the Kierszenbaum lab has demonstrated a protein transport highway wherein proteins processed by the Golgi apparatus are transferred to the surface of the sperm acrosome at the cranial pole of the sperm head in a microtubule-dependent manner, then onto a cytoskeletal scaffolding plate known as the axoplaxome that anchors the developing acrosome to the nucleus [17]. Subsequently, proteins then localise to the microtubules of the manchette and ultimately into the growing sperm tail. The latter part of the highway has been termed intra-manchette transport and it appears to have mechanistic similarities with, and to interface with, the classical IFT pathway required for axoneme extension in most cilia types [13,18].
In an effort to understand these processes, we have undertaken a random mouse-mutagenesis screen to identify genes with critical roles in sperm formation. One of the genes identified encodes the previously uncharacterized gene leucine-rich repeats and guanylate kinase domain contain (Lrguk). Here we have shown that LRGUK isoform 1 (LRGUK-1) has a critical role in both sperm head shaping and basal body attachment to the plasma membrane, and in the early aspects of axoneme development. LRGUK is transported along the acrosome-acroplaxome-manchette-tail axis in a potential complex with the adaptor protein HOOK2 [19,20]. LRGUK-1 dysfunction leads to abnormal manchette formation and movement, and an absence of axoneme extension from the basal body. Collectively these data define LRGUK-1 as a crucial regulator of male germ cell basal body function, microtubule dynamics and fertility.
In order to identify critical regulators of male germ cell development and fertility, we undertook a random N-ethyl-N-nitrosourea (ENU) mouse mutagenesis screen [9,21]. The ‘Kaos’ mouse line was identified based on male-specific infertility and the presentation of chaotic spermatogenesis (see below). Mapping of the mutation and sequencing of candidate genes revealed a C to T substitution in exon 14 of Lrguk (Fig. 1A). Lrguk is predicted to encode 3 splice variants, Lrguk transcript 1 (Lrguk-1), -2 (Lrguk-2) and -3 (Lrguk-3). Lrguk-1 is the longest transcript and the only isoform affected by the mutation (Fig. 1A). The C→T mutation resulted in the conversion of an arginine (R) at position 528 to a premature termination codon (R528Stop, the LrgukKaos allele) and in turn the truncation of 293 residues from the C-terminal region of LRGUK-1 (Fig. 1B). Orthologues of LRGUK can be observed in many species, and sequence alignment of LRGUK-1 from multiple species revealed that R528 residue is conserved in all (Fig. 1C).
LrgukKaos/Kaos testis showed a 75% reduction in the level of Lrguk-1 expression (8–12 weeks-of-age) compared to LrgukWT/WT siblings (Fig. 1D) and a complete absence of both the native LRGUK-1 protein of 93 kDa and the predicted truncated LRGUK-1Kaos of 60kDa (Fig. 1E). We note that the antibody used in these experiments is directed against an epitope present in both the native and truncated forms of LRGUK-1, but C-terminal to LRGUK-2 and -3 sequences. These results suggested Lrguk-1Kaos mRNA was unstable. Quantitative PCR (qPCR) showed no evidence of a compensatory up-regulation of Lrguk-2 and Lrguk-3 mRNAs in the LrgukKaos/Kaos testis (Fig. 1F-G). Collectively, these data revealed that LrgukKaos/Kaos males were sterile as a consequence of a specific absence of the LRGUK-1 protein.
All LrgukKaos/Kaos males had chaotic and disorganised spermatogenesis and were sterile (n>50, aged 7–26 weeks), whereas LrgukWT/Kaos males and LrgukKaos/Kaos females had apparently normal fertility. LrgukKaos/Kaos males had a normal body weight (S1A Fig.) and were anatomically indistinguishable from wild type siblings. LrgukKaos/Kaos male sterility was characterized by a 13% reduction in testis weight (Fig. 2A) and an 81% reduction in daily sperm production (Fig. 2B). Collectively these data suggest that the majority of missing germ cells were from the haploid phase of spermatogenesis wherein they contributed relatively little to testis weight. A similar reduction (79%) in sperm content was observed in the epididymal sperm content (Fig. 2C). The loss of germ cells from the seminiferous epithelium via sloughing was indicated by the presence of immature germ cells in LrgukKaos/Kaos epididymis (Fig. 2F). In contrast, apoptosis levels were similar between genotypes (S1B-C Fig.). Of the elongate spermatids present in the seminiferous epithelium, virtually none appeared to contain a sperm tail (Fig. 2E), and of the very few sperm that were present in the epididymis, all had grossly misshapen heads and shortened tails (Fig. 2D) that displayed no capacity for motility. A closer examination of spermatid structure also revealed the presence of fragmented acrosome in early round spermatids (Fig. 2E and below).
A more quantitative stereological analysis of the dynamics of spermatogenesis confirmed the histological observations. LrgukKaos/Kaos adult testes contained normal numbers of supporting Sertoli cells, spermatocytes and early round spermatids (S1 Table). They contained, however, a 37% reduction (p<0.05) in elongated spermatids (steps 13–16). In addition, we noted a pronounced increase in the number of elongated spermatids retained in the basal crypts of Sertoli cells in stage VII-XII tubules that was suggestive of a failure of the early phases of sperm release (spermiation)(S1D Fig.). Consistent with this interpretation, the levels of DNA double strand breaks, as determined by γH2AX staining, in the retained spermatid nuclei was notably elevated (Fig. 2G) in LrgukKaos/Kaos suggesting the initiation of degeneration [22].
Cumulatively, these data show that LrgukKaos/Kaos males were sterile as a consequence of germ cell sloughing and degeneration, abnormal sperm development and an inability of those sperm that were produced to ascend the female reproductive tract following mating.
qPCR analysis revealed that Lrguk-1 mRNA was highly enriched in the testis compared to other adult tissues (Fig. 3A). An analysis of mouse testes taken at defined periods during the establishment of the first wave of spermatogenesis also revealed that Lrguk-1 mRNA was detectable at low levels from birth, up-regulated at day 14 coincident with the appearance of pachytene spermatocytes, then maximal from day 18 coincident with the appearance of haploid germ cells (Fig. 3B). This result was suggestive of Lrguk-1 being predominantly expressed in haploid germ cells. During spermatid development, and using an antibody that should detect all of LRGUK1–3, LRGUK protein was initially localised to a supra-nuclear region of round spermatids, and was particularly evident at the leading edge of the developing acrosome and acroplaxome (Fig. 3C). As maturation proceeded and nuclear elongation initiated, LRGUK moved distally to ultimately reside on the microtubules of the manchette (Fig. 3D and S2 Fig.). LRGUK was also evident in the sperm basal body and the sperm tail (Fig. 3E). These data, and the abnormal sperm head and tail morphology in LrgukKaos/Kaos germ cells, suggested that LRGUK-1 has a role in acrosome and tail biogenesis. We note that LRGUK protein was not detectable in our hands in spermatocytes. At present it is unknown if this was due to a lack of abundance or a translational delay as is often seen in spermiogenic genes [23].
LrgukKaos/Kaos germ cells showed reduced LRGUK immunolabelling compared to wild type cells (S2 Fig.), however, residual, presumably LRGUK-2 and/or LRGUK-3 staining could be seen within mutant spermatids (Fig. 3C-D and S2 Fig.). The possibility also remains that a small amount of truncated LRGUK-1 was detected by this antibody. Within these cells the distribution of residual LRGUK was notably perturbed. Specifically, the movement of LRGUK appeared to stall at the leading edge of the acrosome/acroplaxome complex (Fig. 3C-D and S2 Fig.). LRGUK staining on the microtubules of the LrgukKaos/Kaos manchette was reduced and more variably distributed (Fig. 3D and S2 Fig.). These data are consistent with the acroplaxome operating as a loading dock for cargo proteins prior to loading onto the microtubules of the manchette [15,17] and the C-terminal 293 amino acids of LRGUK-1 having a role in this transition.
One of the features of the LrgukKaos/Kaos phenotype was the abnormal morphology of sperm heads. A detailed analysis of acrosome formation on periodic acid Schiff’s (PAS) stained sections revealed the presence of fragmented acrosomes in step 2–4 spermatids (Fig. 4A). The acrosome is formed by the sequential fusion of Golgi-derived pro-acrosomal vesicles to form a cap overlying the nucleus. The membrane overlying the acrosome is essential for the initial binding to and penetration through the oocyte complex [15,24,25]. Electron microscopy also revealed ∼20% of acrosomes in elongated spermatids were detached from sperm nuclei (Fig. 4B). A close inspection of these cells indicated that both the acrosome and the acroplaxome were detaching from the nuclear membrane suggesting that LRGUK-1 has a role in establishing the integrity of the connection between the acroplaxome and the nuclear membrane.
LrgukKaos/Kaos spermatids also contained abnormal manchettes (S2 Fig., Fig. 4D and Fig. 5). The manchette is a grass skirt-like structure that encircles the elongating spermatid nucleus. As indicated above, it has a role in both nuclear shaping and protein transport into the growing sperm tail via a process called intra-manchette transport [13,18]. In normal spermatids, the manchette contains a series of parallel microtubule bundles that extend from a perinuclear ring and lie in close proximity to, and are parallel with, the nuclear membrane into the distal cytoplasm (Fig. 4C and Fig. 5). An analysis of elongating LrgukKaos/Kaos spermatids, using α-tubulin as a microtubule marker, revealed that the manchettes formed at the correct time and that the perinuclear ring began to move caudally along the sperm head during spermiogenesis (Fig. 5). The microtubule bundles of the manchette skirt were, however, unevenly distributed and had a ‘raggedy’ appearance compared to wild type cells. Further, and based on both the light and electron microscopic images the distal movement of the manchette along the nucleus that normally occurs during elongation was also abnormal in LrgukKaos/Kaos spermatids (Fig. 5, S3 Fig.). In contrast the perinuclear ring of the manchette continued to constrict, as it would normally, thus leading to nuclear distortion (Fig. 4C, Fig. 5 and S2 Fig.).
Collectively these data illustrate a role for LRGUK-1 in acrosome attachment to the sperm head and in microtubule organisation within the manchette.
LRGUK-1 contains multiple domains with known roles in protein-protein interactions i.e. a guanylate kinase-like domain (GK), leucine rich repeats (LRRs) domains and a leucine rich repeat C-terminal domain (LRRCT) [26,27] (SMART entry IPR000483). In order to explore this potential, and to define pathways within which LRGUK-1 may be involved during haploid germ cell development, we performed a yeast two-hybrid screen to identify LRGUK-1 binding partners. One of the binding partners identified was HOOK2 (Fig. 6). The identified Hook2 clone encoded the C-terminal 348 amino acids of HOOK2. Specific transfection of the full length HOOK2 with LRGUK-1 in a separate yeast two hybid assay confirmed this interaction and the introduction of the R528Stop mutation into the Lrguk-1 sequence completely abolished binding to HOOK2 (Fig. 6A). An interaction between LRGUK-1 and endogenous HOOK2 was confirmed by co-immunoprecipitation, wherein full-length mouse Lrguk-1-GFP was transfected into HEK293T cells then the complex co-precipitated (Fig. 6B).
HOOK2 is a member of the HOOK family of proteins, which are adaptor-like proteins involved in loading cargos (including protein complexes and organelles) onto microtubules for transport [28]. Notably, HOOK2 has recently been found to function in the maintenance of centriole structure and primary cilia assembly [19]. Like sperm flagella, primary cilia contain a central axoneme, but rather than possessing a 9+2 microtubule structure, they lack the central microtubule pair (9+0) [16,29]. Specifically, within retinal epithelial cells HOOK2 bound to the essential cilia proteins PCMI and RAB8a and was required to initiate axoneme growth from the basal body [20]. These data raise the possibility that HOOK2 is involved in the delivery of LRGUK-1 to the basal body, or vice versa, and in the initiation of sperm tail growth.
As for LRGUK, HOOK2 was localised to the Golgi-derived spermatid acrosome, the acroplaxome, the manchette and the sperm basal body (Fig. 6C-E). HOOK2 localization was not appreciably altered in the presence of the LrgukKaos mutation, albeit in the presence of abnormal germ cell structure (Fig. 6D), indicating that LRGUK-1 does not define HOOK2 localization. Unfortunately genetically modified Hook2 mouse lines are not available to test the dependence of LRGUK localization on HOOK2. As such, and although the formal possibility remains that HOOK2 may be a LRGUK-1 cargo, it is more likely that HOOK2 functions in its established role to transport LRGUK-1.
The LrgukKaos/Kaos sperm phenotype has some similarities with that observed for HOOK2 depleted retinal cells, suggesting they lie in a common pathway. Greatly reduced axoneme development was evidenced in testis sections stained for the axoneme marker acetylated tubulin and the absence of 9+2 axoneme structure was seen at the electron microscopic level (Fig. 7A-E). An analysis of the early steps of centriole/basal body movement and axoneme development, however, revealed abnormalities in LrgukKaos/Kaos germ cells, strongly suggestive of a critical role for LRGUK-1 beyond that currently documented for HOOK2, specifically in the formation of the centriole appendages. In early wild type spermatids the mature centriole can be distinguished from the adjacent daughter centriole by the possession of accessory structures known as the sub-distal appendages (SAP) and distal appendages (DAPs) [30]. The mature centriole then migrates to the spermatid periphery and attaches to the plasma membrane before attaching to the nuclear pole opposite the acrosome [16]. Recent data has shown the DAPs are required for basal body-to-membrane docking [31], and SAPs are believed to be where axoneme microtubules are anchored and are thus required for axoneme extension [32–34]. Electron microscopy of spermatids from LrgukWT/WT spermatids clearly showed the docking of mature centrioles to the plasma membrane and the associated DAPs and SAPs (Fig. 7E). In contrast, the over-whelming majority of basal bodies in spermatids from LrgukKaos/Kaos mice contained SAPs that were overtly enlarged and plasma membrane attachments were very infrequent (<1% of the time) (Fig. 7E). Consistent with abnormal SAP function, microtubule extension from the LrgukKaos/Kaos basal body into an axoneme was extremely rare (Fig. 7D). Despite the absence of membrane attachment basal bodies did contain DAP-like structures. LrgukKaos/Kaos basal bodies did, however, appeared to attach to the nuclear membrane normally (Fig. 7E). In comparison to the basal body phenotype seen in LrgukKaos/Kaos spermatids, the loss of HOOK2 function during retinal cell primary cilia formation had no reported effect on SAP formation [19]. These data suggest that LRGUK-1 is required for the formation / function of the DAPs and SAPs in spermatids. Currently it is unknown if these effects are independent of HOOK2 or if they are spermatid-specific functions and thus, not seen following HOOK2 knockdown in retinal cells.
In contrast to the lack of axoneme development in LrgukKaos/Kaos spermatids, we observed considerable evidence of the assembly of outer dense fiber-like structures within the distal cytoplasm of elongated spermatids (Fig. 7B). These data suggest that while the outer dense fibers sit in close apposition to the microtubules of the axoneme in normal germs cells, their development can occur independently.
The dynamic organisation of microtubules and the ability to transport proteins over long distances are fundamental processes required for many cell types, but perhaps none more so than haploid male germ cells, where sperm head shaping and tail development occurs in the virtual absence of transcription [35]. Here we have demonstrated that LRGUK-1 is a critical component of this process with roles that impact upon multiple facets of sperm structure. LRGUK-1 is required for normal acrosome attachment, manchette function, the initiation of the axoneme extension and ultimately male fertility. In accordance with these roles, we observed LRGUK-1 within a protein transport corridor [13] involving movement from the Golgi complex in round spermatids, to the acrosome/acroplaxome, onto the manchette in elongating spermatids then ultimately into the sperm tail. Collectively our data establish LRGUK-1 as a vital component for haploid germ cell development and male fertility.
The earliest discernable abnormality seen in LrgukKaos/Kaos germ cells was in acrosome development. The acrosome is a vesicle-like structure at the caudal pole of the sperm head. The plasma membrane immediately above the acrosome is the first point of binding between the sperm and the cumulus oocyte complex prior to fertilisation, and the contents of the acrosome is required for sperm penetration through the outer vestments of the oocyte [36]. Data presented here suggests that LRGUK-1 is required for the appropriate attachment of Golgi-derived pro-acrosomal vesicles onto the nucleus and the full structural integrity of the acrosome-acroplaxome attachment to the nuclear membrane [15,17]. LRGUK-1 dysfunction resulted in the fragmentation and loss of some pro-acrosomal vesicles in early round spermatids, and the detachment of the acrosome-acroplaxome from the sperm head during late spermiogenesis. Clearly however, acrosome-like structures were formed in spermatids (albeit with apparently reduced efficiency in early round spermatids) indicating that the enhanced adhesion inferred by LRGUK-1 may only become critical when shear forces are applied, for example during stages I through to VIII wherein elongated spermatids are sequentially dragged down into Sertoli cell crypts then pushed up to the luminal aspect of the seminiferous epithelium [37]. The apparent absence of LRGUK-1 protein from the acroplaxome region in mature sperm suggests that this defect maybe mediated by an LRGUK-1 binding partner rather than LRGUK-1 itself, or that LRGUK-1 is part of a stepwise process resulting in firm acrosome-acroplaxome-nuclear attachment. Taken together our results suggest a role for LRGUK-1 in the trafficking of pro-acrosomal vesicles from the Golgi to the acroplaxome and in the attachment of acrosome-acroplaxome to the nuclear membrane.
A second major defect, and almost certainly the cause of the head abnormalities in LrgukKaos/Kaos sperm, was a defect in the structure and movement of the manchette. The manchette is a complex microtubule array that forms around the spermatid nucleus concordant with the initiation of nuclear compaction. It is composed of a perinuclear ring and a fringe of microtubule bundles that extend into the distal cytoplasm. As spermiogenesis proceeds the manchette moves distally, and in the case of species such as the mouse that contain falciform shaped sperm heads, it pivots in a manner dependent on microtubule severing [22]. In parallel, the perinuclear ring constricts, likely contributing to the tapered shape of the post-acrosomal region of the sperm head. Our data show that LRGUK-1 is required for the distal movement of the perinuclear ring. In the presence of the LrgukKaos/Kaos mutation the perinuclear ring constricted more proximally than normal leading to abnormal sperm head shape. Interestingly the remaining LRGUK present within elongating spermatids (LRGUK-2 and/or -3 or small amounts of truncated LRGUK-1) remained at the leading edge of the acroplaxome, suggesting that there are motifs C-terminal 293 amino acids of LRGUK-1 that are required for the transition to the manchette.
Of relevance, a second member of the HOOK family is required for manchette function and ultimately male fertility. The azh mouse phenotype was caused by the deletion of exons 10–11 from the Hook1 gene [38]. Homozygous azh males displayed OAT characterised by severe head abnormalities and the frequent decapitation from the sperm tail. The latter is suggestive of a weakened connection between the nuclear membrane and the basal body-derived axoneme. This phenocopying and data suggesting that HOOK family members frequently function as heterodimers [39] raise the possibility that LRGUK-1 exists in a complex with both HOOK2 and HOOK1. This possibility will be tested in future experiments.
The most striking aspect of the Kaos phenotype, and the ultimate cause of male sterility, was the almost complete block of sperm tail development. LRGUK-1 dysfunction lead to abnormal SAP formation on basal bodies and an absence of plasma membrane attachment, likely associated with DAP dysfunction. As a consequence, basal bodies failed to nucleate axoneme microtubules. Within retinal primary cilia, HOOK2 is involved in the assembly of a complex containing PCMI and RAB8a, and the initiation of axoneme microtubule extension [19]. In contrast to HOOK2 action in primary cilia, however, our data revealed that LRGUK-1 also has a critical role in the formation of basal body-plasma membrane attachment and SAP formation. Currently it is unknown if these effects are independent of HOOK2 or if they are spermatid-specific functions and thus, not seen following HOOK2 knockdown in retinal cells. Our data do not, however, support a role for LRGUK-1 in cilia broadly. Notably, the Kaos mouse line showed none of the morphometric features characteristic of primary cilia disorders such as polydactyly and craniofacial abnormalities [40], although the presence of Lrguk-1 mRNA in tissues containing motile cilia (9+2 axonemes) raises the possibility of an age-related pathology associated with motile cilia dysfunction.
The phenotypic similarity observed in LrgukKaos/Kaos and mice containing mutations in core proteins of the IFT pathway including Ift88 and Kif3a [41,42] highlights the continuity of the protein transport pathway underlying the movement of proteins involved in sperm head shaping and axoneme extension. Our data suggest that LRGUK-1 functions upstream/before the IFT pathway i.e. spermatids with defective LRGUK-1 will undergo a developmental arrest prior to the requirement for the IFT pathway. These data also highlight the exquisite value of using the testis as a model to define processes of likely fundamental importance to cell biology broadly.
Collectively, we have identified LRGUK-1 as a protein critically involved in multiple aspects of sperm assembly and function. LRGUK-1 is required for appropriate acrosome attachment to the sperm heard, sperm head shaping via the manchette and tail growth from the basal body. During the initiation of sperm tail axoneme extension from the basal body, LRGUK-1 functions in plasma membrane attachment and the formation of SAP and ultimately microtubule extension. The specific loss of LRGUK-1 function results in OAT in mice and raises the possibility that LRGUK dysfunction leads to human OAT.
Animal procedures were performed in accordance with Australian NHMRC Guidelines on Ethics in Animal Experimentation and approved by the Australian National University and Monash University Animal Experimentation Ethics Committees. Point mutant mice were generated on a C57BL/6 background and outbred to CBA and individual lines screened for sterility causing mutations as described previously [9,21]. Mating behaviour, as indicated by the presence of copulatory plugs was normal.
The sterility causing mutation in the Kaos line was initially linked to a region on chromosome 6 using a SNP-based microarray approach. The region was subsequently narrowed using additional mice and SNPs to a 12.9 Mb region (30872499–43776812 bp) that contained 74 genes. Candidate genes were selected based on expression in haploid germ cells, i.e. the site of the phenotype, in EST expression databases. 28 genes were expressed within round spermatids. The full coding region of eight of these genes was sequenced and a single C to T point mutation within the Lrguk gene was identified.
Following the identification of the causal mutation, mice were genotyped using the Amplifluor system (Chemicon) using a wild type-specific reverse primer 5’-GAAGGTCGGAGTCAACGGATTCCATAGGCACCACCAAGATATATCG-3’, a mutant allele specific reverse primer 5’-GAAGGTGACCAAGTTCATGCTCCA TAGGCACCACCAAGATATATCA-3’ and a conserved forward primer 5’- CAGCCTTGGACTATTTATAGGGAGTGTG-3’ as described previously [9].
LRGUK orthologues were identified as described previously [7].
Sterility in the Kaos mouse line was characterized using the regime outlined in Borg et al [43]. Daily sperm production (DSP) in the testis and total epididymal sperm content was determined using the Triton X100 nuclear solubilization method as described previously [44]. Sperm ultra-structure was assessed using electron microscopy as described previously [45]. Sperm motility was assessed visually using sperm back-flushed from the cauda epididymis [46]. Cauda epididymal sperm morphology was visualized using haematoxylin and eosin staining. The stages of the epithelium tubule were judged by Periodic Acid Schiff (PAS) staining [47]. Apoptosis was evaluated by TUNEL Apoptosis Detection Kit (Millipore) according to the manufacturer’s instructions; n = 5 mice per genotype and the positive cells in 100 seminiferous tubules were counted for each mouse.
Germ and Sertoli cell numbers were counted in 25μm-thick PAS stained methacrylate embedded testis sections using the optical dissector method as described previously [48]. Basal retained elongated spermatids were counted in stage VII-XII [22,48]. N = 5 mice per genotype and 240 counting frames for each mouse.
Lrguk in adult tissues and at different time points during the establishment of the first wave of spermatogenesis was defined as outlined previously [49]. Lrguk expression was detected using the Taqman assay (Mm01166701-m1). Lrguk-2 expression was detected using a custom designed assay (Forward primer 5'-CCCCAAAATCTCAAGGTATACTTATCAG-3'; reverse primer 5'–CCGCAGCTGAAGCAAAACTC-3'; probe 5'-CAAAGCACACAATGGT-3'). A custom designed assay was also used to detect both Lrguk2 and Lrguk3 transcripts (Forward primer 5'-CTGGCCTATCTGTGGATGACATC-3'; reverse primer 5'-CCGCAGCTGAAGCAAAACTC-3’; probe 5'-CAAAGCACACAATGGT-3’). All expression data was normalized to the Ppia house-keeping gene (Mm02342429-gl).
Germ cells were isolated using the Staput method described previously [50]. For immunofluorescence staining, the purified cells were fixed with 4% paraformaldehyde (PFA) and stained as described previously [9]. Primary antibodies used included: anti-LRGUK (1:200 dilution, Novus Biologicals, raised against amino acids 32–181); anti-HOOK2 (1:200 dilution, GeneTex), anti-acetylated tubulin (1:1000 dilution, Sigma), anti-α-tubulin (1:5000 dilution, Sigma). All primary antibodies were diluted in 10% non-immune horse serum (NHS) in PBS and incubated at 4°C overnight. Secondary antibodies included: Alexa Fluor 555 donkey anti-rabbit IgG and Alexa Fluor 488 donkey anti-mouse IgG, 1:500 dilution in 10% NHS in PBS for 45 minutes at room temperature. Nuclei were labelled with DAPI. The acrosome was visualized using FITC-PNA (1:2000 dilution, Sapphire Bioscience). Images were taken with a SP8 confocal microscope (Leica Mircosystems). For consistence α-tubulin was artificially coloured green in all images.
Paraffin-embedded testis sections were stained for acetylated tubulin as described previously [45].
The adult mouse testis cDNA library (pDEST22 prey vector) used for yeast two hybrid screen was as described previously [51]. The full length Lrguk-1WT and LrgukKaos cDNAs were cloned into pDEST32 vector using the Gateway cloning kit (Invitrogen). For the initial identification of LRGUK binding partners, the LRGUKWT-pDEST32 vector was used as bait in a yeast two hybrid screen with the ProQuest Two-Hybrid System (Invitrogen). Briefly, the LRGUKWT-pDEST32 vector and pDEST22 testis cDNA library constructs were co-transformed into Mav203 yeast strain. Putative interacting clones were isolated from the resulting yeast colonies and re-transformed into E. coli to obtain high purity plasmids for sequencing cDNA inserts. Sequencing of the longest HOOK2 clone revealed an open reading frame of 1078 bp that corresponded to the C-terminal amino acids 369–716 of the HOOK2 protein. To determine the effect of the LRGUKKaos mutation on the binding to HOOK2, the identified HOOK2-pDEST22 prey vector was co-transformed into the Mav203 yeast strain along with either LRGUKKaos-pDEST32, LRGUKWT-pDEST32 or pDEST32 empty vector only and plated on selection media.
Full length mouse Lrguk-1 was amplified from wild type C57BL/6J testis cDNA using Lrguk-Fw: 5’-ATATAAAGATCTGCGGCCTTCGAGCGAAAT-3’ and Lrguk-Rev: 5’-ATATAAGGGCCCCTATCGCGGCCGTGCGGGAT-3’ primers than cloned into the pEGFP-C1 expression vector (Clontech, Gen Bank Accession number U55763). The LRGUK-1/pEGFP-C1 construct was transfected into HEK293 cells using Lipofectamine 3000 Reagent (Life Technologies, Cat. No. L3000008). Co-immunoprecipitation was carried out using anti-GFP-Trap-A beads as per the manufacturer’s instructions (Chromotek, Cat. No. gta-100). Empty pEGFP-C1 vector was used as a negative control. The presence of LRGUK-GFP and endogenous HOOK2 proteins were determined using anti-GFP (Roche, Cat. No. 11814460001 and HOOK2 antibodies (GeneTex, Cat. No. GTX115898, corresponding with amino acids 197–380 of HOOK2).
For LRGUK-1 expression, protein was extracted from both wild type and mutant testes using 1% NP-40/PBS lysis buffer with 1:200 protein inhibitor cocktail (Promega). Protein (40μg) was separated on a 12% SDS-PAGE gel. Non-specific antibody binding was minimized by blocking the membrane with 5% skim milk for 1 hour room temperature and antibodies were diluted in 1% skim milk. The membrane was probed with a rabbit LRGUK-1 antibody (Sigma, raised against amino acids 336–427 of LRGUK-1 and thus will not bind to LRGUK-2 or -3). Bound antibody was detected using a goat anti-rabbit IgG HRP (Dako) secondary antibody. Antibody binding was detected using the enhanced chemiluminescence (ECL Plus) detection kit (Amersham Biosciences). Protein loading was normalized against actin.
|
10.1371/journal.ppat.1002467 | The Cysteine Rich Necrotrophic Effector SnTox1 Produced by Stagonospora nodorum Triggers Susceptibility of Wheat Lines Harboring Snn1 | The wheat pathogen Stagonospora nodorum produces multiple necrotrophic effectors (also called host-selective toxins) that promote disease by interacting with corresponding host sensitivity gene products. SnTox1 was the first necrotrophic effector identified in S. nodorum, and was shown to induce necrosis on wheat lines carrying Snn1. Here, we report the molecular cloning and validation of SnTox1 as well as the preliminary characterization of the mechanism underlying the SnTox1-Snn1 interaction which leads to susceptibility. SnTox1 was identified using bioinformatics tools and verified by heterologous expression in Pichia pastoris. SnTox1 encodes a 117 amino acid protein with the first 17 amino acids predicted as a signal peptide, and strikingly, the mature protein contains 16 cysteine residues, a common feature for some avirulence effectors. The transformation of SnTox1 into an avirulent S. nodorum isolate was sufficient to make the strain pathogenic. Additionally, the deletion of SnTox1 in virulent isolates rendered the SnTox1 mutated strains avirulent on the Snn1 differential wheat line. SnTox1 was present in 85% of a global collection of S. nodorum isolates. We identified a total of 11 protein isoforms and found evidence for strong diversifying selection operating on SnTox1. The SnTox1-Snn1 interaction results in an oxidative burst, DNA laddering, and pathogenesis related (PR) gene expression, all hallmarks of a defense response. In the absence of light, the development of SnTox1-induced necrosis and disease symptoms were completely blocked. By comparing the infection processes of a GFP-tagged avirulent isolate and the same isolate transformed with SnTox1, we conclude that SnTox1 may play a critical role during fungal penetration. This research further demonstrates that necrotrophic fungal pathogens utilize small effector proteins to exploit plant resistance pathways for their colonization, which provides important insights into the molecular basis of the wheat-S. nodorum interaction, an emerging model for necrotrophic pathosystems.
| In this manuscript we describe the cloning of SnTox1 from Stagonospora nodorum, the gene encoding the first host selective toxin (SnTox1) identified in this fungus. SnTox1 induces necrosis and promotes disease on wheat lines harboring the Snn1 gene. We verified the function of the SnTox1 gene by expressing it in a yeast culture where the resulting culture filtrate induced necrosis but only on wheat lines that carried a functional Snn1. The SnTox1 gene was also transformed into an avirulent S. nodorum isolate, resulting in an isolate that was virulent on wheat lines harboring Snn1. SnTox1 was also disrupted in virulent S. nodorum isolates resulting in the elimination of disease on Snn1 differential wheat lines. Additionally, we investigated the host response to SnTox1 and S. nodorum strains producing SnTox1 and discovered that several hallmarks of a resistance response were present during the susceptible reaction, showing that the necrotrophic pathogen S. nodorum is likely using SnTox1 to stimulate a host resistance pathway involving Snn1 to induce disease.
| Like other parasites, fungal pathogens secrete a battery of molecules known as effectors during the infection process. These effectors can alter plant biological processes resulting in successful colonization [1], [2]. Conversely, recognition of effectors by the plant innate immune system can initiate a defense response resulting in effector-triggered immunity (ETI) [3], [4]. ETI is characterized by the accumulation of reactive oxygen species (ROS), transcriptional induction of pathogenesis-related (PR) genes and production of antimicrobial chemical compounds, eventually leading to rapid and localized plant cell death, known as the hypersensitive response (HR) [5]. In ETI, the perception of the fungal effector is mediated by the corresponding plant resistance gene (R) which acts in a gene-for-gene manner [6], [7]. Currently, it is believed that this localized suicide of plant cells induced by ETI halts further growth of the biotrophic fungal pathogen, which requires living plant tissue for survival.
Necrotrophic fungal pathogens are known to produce host selective toxins (HSTs), including low molecular weight metabolites and small secreted proteins that function as essential determinants of pathogenicity or virulence [8], [9]. HSTs can therefore be viewed as effectors of necrotrophic pathogenicity and hence we prefer the term necrotrophic effector (NE) [10], [11]. These effectors play significant roles in determining the outcomes of plant-pathogen interactions by specifically interacting (directly or indirectly) with the products of corresponding host genes [12], [13]. However, in contrast to ETI in the classical gene-for-gene model, the necrosis induced by effectors from necrotrophic fungal pathogens results in disease susceptibility; thus, it can be described as effector-triggered susceptibility (ETS) [14], [15], a term which was originally used in reference to biotrophic systems [4].
The molecular basis of necrosis-induced ETS involving necrotrophic fungi is still largely unknown, but has in several cases exhibited the hallmarks of programmed cell death (PCD); DNA laddering, heterochromatin condensation, cell shrinkage, callose deposition and an oxidative burst [9], [16], [17]. ToxA, a necrotrophic effector found in both Pyrenophora tritici-repentis and Stagonospora nodorum, causes the loss of plasma membrane integrity and the accumulation of hydrogen peroxide [18], [19]. Microarray analysis revealed that several wheat genes involved in defense response and signaling pathways were strongly regulated by the ToxA-Tsn1 interaction [20], [21].
Interestingly, three plant genes involved in susceptibility to necrotrophic effectors (Pc, the sorghum sensitivity gene corresponding to PC toxin; LOV1, the Arabidopsis sensitivity gene corresponding to victorin; and Tsn1, the wheat sensitivity gene corresponding to ToxA) have been cloned and shown to be resistance-like genes containing both nucleotide binding (NB) and leucine-rich repeat (LRR) domains [15], [22], [23]. This has led to speculation that necrotrophic fungal pathogens may utilize plant resistance signaling pathways to subvert PCD and enable pathogen growth [15], [24].
Stagonospora nodorum, an ascomycete fungus (teleomorph: Phaeosphaeria nodorum), is the causal agent of wheat Stagonospora nodorum blotch (SNB), a globally distributed and economically important disease [25]. S. nodorum is a typical necrotrophic fungal pathogen [10], [26]. In recent years, it has been shown that this pathosystem is based largely on interactions involving proteinaceous necrotrophic effectors and corresponding host sensitivity genes that, when occurring together, result in ETS. To date, six interactions have been reported including SnTox1-Snn1 [27], SnToxA-Tsn1 [28], [29], SnTox2-Snn2 [12], SnTox3-Snn3-B1 [30], SnTox4-Snn4 [31], and SnTox3-Snn3-D1 [32]. In addition, several other effector-host gene interactions have been identified (Friesen and Faris, Oliver and Tan, unpublished data). Therefore, the wheat-S. nodorum system is emerging as a model to investigate the molecular mechanisms of necrotrophic pathogenesis [13]. One of our research goals has been to clone necrotrophic effector genes and decipher their molecular and biochemical functions.
Of the S. nodorum effector genes, SnToxA and SnTox3 have been cloned with the aid of the S. nodorum genome sequence information [14], [29], [33]. The SnToxA gene is essentially identical to the ToxA gene isolated from the wheat tan spot pathogen P. tritici-repentis. Mature ToxA consists of a 13.2 kDa protein containing two cysteine residues as well as an RGD-containing vitronectin-like motif that is present in a solvent-exposed loop in the active protein [34]–[38]. The RGD motif has been shown to be essential for internalization and internalization has been shown to be critical for the induction of necrosis [37], [39], [40] SnTox3 encodes an approximately 17.5 kDa mature protein with six cysteine residues and has no homology to genes in the public databases [14].
Here, we report the molecular cloning and characterization of the SnTox1 gene which encodes the SnTox1 protein, and we show that SnTox1 is specifically recognized by the corresponding wheat sensitivity/susceptibility gene Snn1. The characterization of the SnTox1-Snn1 interaction provides strong evidence that necrotrophic fungal pathogens use necrotrophic effectors to subvert the host resistance mechanism to cause disease.
Whole genome reference sequences have proven to be powerful for the identification of fungal and oomycete effector genes [1], [41]. The annotated S. nodorum genome sequence supports a minimum of 10,762 nuclear genes with 1,782 predicted to encode extracellular proteins [33]. A specific set of criteria was used to prioritize the genes and generate a list of candidates. The criteria (size less than 30 kDa, predicted to be secreted, expressed in planta, etc, see Materials and Methods) were based on the characteristics of the previously cloned SnToxA and SnTox3 genes. We focused on the top 100 genes and as expected, SnTox3 and SnToxA were identified among them. PCR analysis was conducted to confirm the absence of genes in the S. nodorum avirulent isolate Sn79-1087 (data not shown). Genes meeting these criteria were expressed in the Pichia pastoris heterologous expression system [14]. This process and the subsequent screening of a set of differential lines (see Materials and Methods) led us to identify SNOG_20078 as the SnTox1-encoding gene.
Culture filtrates of P. pastoris strain X33 transformed with the coding region of SNOG_20078 cDNA were infiltrated into the leaves of W-7984, Chinese Spring (CS), CS 1BS-18 and CS ems237. W-7984 and CS carry the dominant Snn1 allele that confers sensitivity to SnTox1 [27]. CS 1BS-18 and CS ems237 are nearly identical to CS, but harbor mutations at the Snn1 locus, resulting in insensitivity to SnTox1. Necrosis developed in the SnTox1-sensitive lines W-7984 and CS, but not in CS 1BS-18 and CS ems237 (Figure 1) suggesting that SNOG_20078 was the SnTox1-encoding gene. To map the gene conferring sensitivity, the same culture filtrates were subsequently infiltrated into the entire ITMI mapping population, which segregates for Snn1/snn1. All lines sensitive to the partially purified native SnTox1 [27] were also sensitive to the culture filtrates of the SNOG_20078 transformed yeast strain. This strongly indicated that SNOG_20078 was the SnTox1-encoding gene and therefore we designated it SnTox1.
SnTox1 is located in supercontig 10 of the assembled SN15 genome sequence ([33], Figure 2A). Within a ≈7.6 kb region, there are three genes upstream (SNOG_07154-SNOG_7156) and one downstream (SNOG_07153) of SnTox1 (Figure 2A). Similarly, there is a short truncated molly-type retrotransposable element (183 bp) sequence following SnTox1 (http://genome.jgi-psf.org/cgi-bin/browserLoad/?db=Stano1&position=scaffold_10). The sequencing of the 5′ and 3′ RACE fragments revealed three exons as well as 5′ and 3′ untranslated regions (UTRs) in the full-length transcript of SnTox1 (Figure 2B). Putative TATA and CAAT boxes were identified 114 bp and 570 bp upstream of the ATG start site, respectively (Figure S1).
The SnTox1 protein consists of 117 amino acids with the first 17 amino acids predicted as a signal peptide. Interestingly, 16 of the remaining 100 amino acids are cysteine residues (Figure 2C). No prosequence was predicted using the web-based program ProP 1.0 (http://www.cbs.dtu.dk/services/ProP/) and after the cleavage of the predicted signal sequence the mature protein was estimated to be 10.33 kDa. To demonstrate that SnTox1 was produced in yeast culture and to verify the size of SnTox1, we applied western blot analysis to the protein samples prepared from SnTox1 yeast culture filtrates. The antibody for SnTox1 was generated from rabbit immunized with a BSA-conjugated 14 amino acid long SnTox1 peptide (see Material and Methods). A single band was only observed in protein samples prepared from SnTox1 yeast culture filtrates, but not from the control culture filtrates (yeast strain transformed with an empty vector). Furthermore, the western band was visualized between the size standard of 10 and 15 kDa, but much closer to 10 kDa (Figure S2). The estimated size of SnTox1 obtained from the western blot agreed with the predicted molecular weight of 10.33 kDa for the mature protein.
A BlastP search of the NCBI non-redundant database with the SnTox1 protein sequence as a query led to the identification of three putative proteins with unknown functions, one from S. nodorum (SNOG_06487) and two from P. tritici-repentis (PTRT_04748 and PTRT_03544) with similarities of 38%, 56%, and 43%, respectively. The conserved amino acids between SnTox1 and these proteins were mostly distributed in the predicted signal sequence and the N-terminal region of the mature protein (Figure S3).
Amino acid alignment with manual adjustment indicated that SnTox1 contained local similarity with cysteine-rich Cladosporium fulvum Avr4-like fungal effectors (Figure 3A) from Cercospora beticola, Mycosphaerella fijiensis [42] and two ascomycete human pathogens, Microsporum gypseum and Geomyces pannorum (this study). These conserved motifs were identified within the chitin-binding domain (ChtBD) including the C-terminal conserved chitin-binding (CB) motif (Figure 3A). Three-dimensional (3D) structure-based sequence alignment suggested that the putative CB motif in SnTox1 was more similar to those of plant-specific ChtBDs (ChtBD1, or CBM18 superfamily, pfam00187) than to Avr4 proteins, which are related to invertebrate ChtBDs (ChBD2, or the CBM14 superfamily, pfam01607) [43] (Figure 3B). SnTox1 contained all secondary-structure-related residues including the strictly conserved β-strand-forming “CCS” motif found only in plant-specific ChtBD1 proteins [44] (Figure 3B). In contrast, all Avr4-like proteins lacked the “CCS” motif and had a loosely conserved “QWN” motif at the same positions as that found in the antimicrobial protein tachycitin, a representative ChtBD2 [44]. There were several insertions found between conserved regions in SnTox1 which also lacked the C-terminal extension after the conserved CB motif, suggesting a significant sequence divergence between SnTox1 and Avr4-like proteins.
The distribution of SnTox1 in different S. nodorum isolates and related fungal species (Table S1 and S2) was investigated using PCR assays and DNA dot blots. Among the 777 isolates that were sampled from wheat fields around the world, 85% (661) possess the SnTox1 gene (Table S1). Dot blot analysis of a subset of a global collection (Table S2) showed that SnTox1 was absent in all S. nodorum isolates collected from wild grasses which are avirulent on wheat (Figure 4A). Additionally, SnTox1 was absent in related fungal species including P. tritici-repentis, P. teres, P. bromi and M. graminicola.
To investigate sequence variation in SnTox1, the gene was PCR-amplified and sequenced from 159 global S. nodorum isolates. We found 12 different nucleotide haplotypes, 11 of which encode different protein isoforms, consistent with strong diversifying selection. The 11 protein isoforms involve amino acid changes at eight positions within SnTox1; however, all cysteine residues remain unchanged across all isoforms (Figure 2D). The nucleotide sequences of all 12 haplotypes have been submitted to GenBank and the accession number for each haplotype is provided at the end of the text. Four codons exhibit significant positive selection using PAML (Table 1). These findings provide strong evidence that positive diversifying selection, consistent with a co-evolutionary process, has been operating on SnTox1.
To investigate sequence variation of the SnTox1 genomic region in virulent and avirulent isolates, we used PCR to amplify the four genes flanking SnTox1 (SNOG_07153, SNOG_07154, SNOG_07155, and SNOG_07156, see Figure 2A for their locations). Only SNOG_07154 located directly upstream of SnTox1 could not be amplified from the avirulent isolate Sn79-1087 (data not shown), which suggested that a region containing all or part of SNOG_07154 as well as the entire SnTox1 sequence may be missing in Sn79-1087. PCR primers were designed within the two genes SNOG_07153 and SNOG_07155 and used to amplify DNA from different virulent isolates as well as Sn79-1087. The amplified fragment in SN15 was ∼4.1 kb as expected but 4.5 kb in Sn1501 and 2.3 kb in Sn79-1087 (Figure 4B). Sequencing revealed that a 3.1 kb region including SnTox1 and the last 85 bp of the 3′ end of SNOG_07154 coding region was replaced by a 1.3 kb sequence in Sn79-1087 (Figure 4C). The 1.3 kb insertion in Sn79-1087 does not share homology with any other known sequence in the NCBI database. Sequence analysis also revealed that two indels occur in the SnTox1 genomic region with one indel of 400 bp in the upstream, and the other indel of 167 bp at the end of the 3′UTR region (Figure 4C).
The avirulent isolate Sn79-1087 does not produce any known S. nodorum necrotrophic effectors, nor does it induce a susceptible response on any of the wheat lines that we have tested. Therefore, a 1.1 kb SnTox1 genomic region (Figure S1) containing the native promoter, open reading frame, and the native terminator was cloned into the pDAN vector (Figure S4A) and transformed into Sn79-1087. Southern blot analysis indicated all but one transformant possessed the 1.1 kb SnTox1 fragment (Figure S4B). Transformants A1 and A3, designated Sn79+SnTox1A1 and Sn79+SnTox1A3, were selected for further analysis. We confirmed that culture filtrates of Sn79-1087 did not cause necrosis nor did spore inoculations cause disease on CS, which contains Snn1 (Figure 5A). However, infiltration of culture filtrates from Sn79+SnTox1A1 and Sn79+SnTox1A3 produced necrosis on the leaves of CS (Figure 5A) and inoculation of CS with conidia of Sn79+SnTox1A1 and Sn79+SnTox1A3 produced lesions on the leaves of CS (Figure 5B). The two transformants did not cause disease on CS 1BS-18 or CS ems237, which lack a functional Snn1 gene (Figure 5B).
The virulent isolate Sn2000 was used in the original identification of SnTox1 and Snn1 [27]. Therefore, this isolate was used to conduct gene knock outs of SnTox1. We exploited a PCR-based split marker method to replace the majority of the SnTox1 gene with the hygromycin resistance gene (hygR). The transformants were verified using Southern blot analysis with a probe amplified from the SnTox1 region that was replaced by hygR (Figure S4C). In two transformants designated Sn2000ΔSnTox1–9 and Sn2000ΔSnTox1–15, the SnTox1 gene was successfully replaced, and one transformant designated Sn2000ΔSnTox1-ECT was identified as an ectopic insertion due to it being hygromycin resistant but still having an intact and functional SnTox1 gene (Figure S4D).
Spores of the three transformed fungal strains along with wild type Sn2000 were inoculated onto the Snn1 differential wheat line W-7984 [27]. The ectopic strain Sn2000ΔSnTox1-ECT induced similar reaction as the wild type including defined tan necrotic lesions with widespread small white flecking, whereas the two knockout strains induced almost no reaction on the leaves (Figure 6A) indicating SnTox1 is an important virulence factor for Sn2000. Sn2000ΔSnTox1–9 and the Sn2000 wild type were also inoculated onto CS. Compared to the wild type, the virulence of Sn2000ΔSnTox1–9 on CS was substantially reduced, but not completely eliminated (Figure 6B), which is due to CS having at least one additional necrotrophic effector sensitivity gene that likely interacts with another effector produced by Sn2000 (Faris and Friesen, unpublished).
The wheat ITMI population was used to originally map the QTL associated with disease susceptibility caused by Sn2000, in which two significant QTL were identified, one on chromosome 1BS corresponding to the Snn1 locus and the other on chromosome 4BL, explaining 48% and 9% of the disease, respectively [45]. We inoculated the three fungal strains: Sn2000ΔSnTox1–9, Sn2000ΔSnTox1–15 and Sn2000ΔSnTox1-ECT along with wild type Sn2000 onto the ITMI population. For Sn2000, as expected, we detected two significant QTL with one being at the Snn1 locus and the other being on chromosome 4BL accounting for 50 and 17% of the disease variation, respectively. A very similar result was obtained for Sn2000ΔSnTox1-ECT where the Snn1 QTL and the QTL on chromosome 4BL were detected explaining 50 and 15% of the variation, respectively (Figure 6C). However, in the inoculation with the two SnTox1 knock out strains, the QTL conferred by Snn1 on chromosome 1BS became undetectable showing no association with disease, but the QTL on chromosome 4B was retained and became more significant overall accounting for 40–50% of the disease variation (Figure 6C). The QTL analysis in the ITMI population clearly demonstrated that SnTox1 codes for the SnTox1 protein which plays a significant role in disease by interacting with the host gene Snn1.
SnTox1 had a very similar expression pattern as SnToxA and SnTox3 during infection in a microarray analysis that examined the expression of all fungal genes at 3, 5, 7, and 10 days post inoculation (DPI) in the wheat cultivar ‘Amery’ inoculated with SN15 (Ip-Cho and Oliver unpublished data). The analysis showed that the expression of all three genes was highest at 3 DPI (Figure S5). In this work, SnTox1 expression was examined after inoculation of CS with Sn79+SnTox1A1, in which no other toxin-sensitivity gene interactions were involved. In the current study, relative expression of SnTox1 to the fungal actin gene was examined at 10 time points ranging from 1 h to 7 d post inoculation using relative-quantitative PCR. Our analysis confirmed that SnTox1 expression was maximized at 3 DPI (Figure 7A). The expression of SnTox1 showed a slow increase between 6 and 12 HPI, increasing to about the same level as the actin gene at 24 HPI and increasing dramatically to 2.5 times higher than the actin gene expression at 48 HPI (Figure 7A). Once gene expression peaked at 3DPI, the SnTox1 transcription levels started to drop significantly from 3 to 4 DPI and returned to similar levels as the actin gene between 5 and 6 DPI. The accelerated increase of SnTox1 expression from 24 HPI to 3 DPI indicates that SnTox1 plays an important role in the early stage of infection.
The symptom development was examined macroscopically on CS inoculated with Sn79+SnTox1A1 (Figure 7B). Disease symptoms were first visible on leaves at 2 DPI as white flecks and progressed into larger necrotic and chlorotic lesions. Interestingly, tan necrotic lesions start to develop at 3 DPI within the chlorotic areas, which correlates with the maximum expression of SnTox1 (Figure 7A). By 5 DPI, necrotic lesions became evident and the chlorotic areas enlarged (Figure 7B). The overall phenotype of the lesions changed very little from 5 to 7 DPI with only a slight change in size of individual lesions (Figure 7B).
The SnTox1 protein contains 16 cysteine residues all of which are predicted to be involved in the formation of disulfide bonds with confidence levels greater than 7 (0 to 9 scale, [46]) (Figure 8A). The prediction software DiANNA [47] was used to identify the most likely connectivity of the cysteine residues as following: 1–11, 2–5, 3–6, 4–13, 7–9, 8–16, 10–12, and 14–15 (Figure 8B). The stability of SnTox1 was tested by incubation of an SnTox1-containing yeast culture filtrate with different concentrations of dithiothreitol (DTT) and different incubation periods. The complete elimination of SnTox1 activity required 4 h in 40 mM DTT (Figure 8C). Additionally, the stability of SnTox1 was tested by directly heating the SnTox1 yeast culture filtrates on a hot plate. Strikingly, the culture filtrates maintained necrotic activity even after boiling for 30 min and did not completely lose activity until after 1 h (Figure 8D). Together, these results show that SnTox1 is a highly stable protein with the ability to resist physical and chemical degradation.
The oxidative burst is one of the best-known biochemical responses of plant cells during a resistance response. The oxidative burst can be visualized by 3′–3′ diaminobenzidine (DAB) staining for H2O2 production [48]. Chinese Spring (CS, Snn1) wheat leaves were infiltrated with SnTox1 yeast culture filtrate or a control yeast culture filtrate and collected at 48 h post-infiltration. The CS ems237 line (snn1) was included for infiltration and DAB staining as a comparison. Leaves were stained with 1 mg/ml DAB solution followed by clearing of chlorophyll. Dense brown DAB staining was observed on the leaves of CS (Snn1) infiltrated with SnTox1, but DAB staining did not appear on leaves of CS infiltrated with the control culture filtrates deficient in SnTox1, nor did DAB staining appear when SnTox1 was infiltrated into leaves of CS ems237 lines (snn1) (Figure 9A), clearly showing that the production of H2O2 is induced only during the SnTox1-Snn1 interaction. A control without DAB staining was also conducted on CS leaves infiltrated with SnTox1 yeast culture. After clearing the leaf, no browning was observed indicating that, in the absence of DAB, the SnTox1 reaction itself was not able to cause brown staining on the leaf (Figure 9A).
The production of H2O2 was also detected during the fungal infection. The CS leaves inoculated with Sn79+SnTox1A1 were collected daily from 1 to 7 days post inoculation and stained with DAB followed by the same procedure for leaf clearing. The accumulation of brown staining on the leaf was readily visible from 2 DPI (Figure 9B). The generation of reactive oxygen species (ROS) associated with a hypersensitive response in planta often occurs in the chloroplast [49]. Using DAB stained CS leaves from the SnTox1 infiltration, we observed that chloroplasts had the highest intensity of brown color (Figure 9C).
Up-regulation of plant defense or signaling pathway genes including pathogenesis-related (PR) genes are hallmarks of a resistance response. Using RT-PCR, we examined the transcription level of 28 wheat genes (Table S3) in CS (Snn1) and CS ems237 (snn1) leaves that were collected at different time points from 1 h to 72 h after being infiltrated with SnTox1 culture filtrates as well as control culture filtrates. Three genes including PR-1-A1, a thaumatin-like protein gene, and a chitinase were found to be significantly up-regulated in CS leaves infiltrated with SnTox1 compared to the control leaf samples infiltrated with culture filtrates deficient in SnTox1 (Figure 10A). In the CS ems237 line which has a mutated snn1 gene, a transcript was undetectable for the PR-1-A1 gene and was at a significantly lower level for the thaumatin and chitinase genes as detected by RT-PCR (Figure 10A). Quantitative PCR (qPCR) analysis confirmed the higher expression of the three genes in SnTox1 infiltrated CS leaves compared to the control infiltrated CS leaves. Not only did all three genes show maximum expression at 36 HPI, but each had at least two-fold higher expression in SnTox1-infiltrated samples than the control (Figure 10B). qPCR also showed much lower expression of the three genes in the CS ems237 line infiltrated with either SnTox1 or the control yeast culture filtrates in comparison to CS infiltrated with control culture filtrates (Figure 10B). The reason for this is not clear, but it could be explained by the idea that Snn1 may play a role in sensing other environmental stimuli that can trigger PR gene expression.
Programmed cell death (PCD) triggered by biotrophic effectors is often evidenced by DNA laddering in plants [16], [50]. To determine if the necrosis induced by SnTox1 on Snn1 lines was a result of PCD, we isolated DNA from infiltrated CS leaf samples and checked for evidence of DNA laddering. For negative comparisons where no necrosis developed, DNA fragmentation was also examined in CS leaves infiltrated with control culture filtrates (no SnTox1) and CS ems237 (mutated snn1) infiltrated with SnTox1 or control culture filtrates. In the CS leaf samples infiltrated with SnTox1, DNA laddering was detected as early as 10 h after infiltration and was most evident at 36 h after infiltration (Figure 11); however, in the leaf samples from the other three treatments, no DNA laddering was observed at any time point (Figure 11), indicating that SnTox1-induced necrosis on lines harboring Snn1 is a result of host-controlled PCD.
Light has been found to be important in the development of necrosis induced by necrotrophic effectors from P. tritici-repentis and S. nodorum [12], [37]. Therefore, we investigated whether the development of necrosis induced by SnTox1 as well as the disease development caused by the SnTox1-Snn1 interaction was light dependent. After infiltration with SnTox1 yeast culture filtrates, CS plants were incubated in a growth chamber but covered for 2 days. The plants in the dark did not exhibit a necrotic reaction in the infiltrated area on the leaves, while the plants grown in the same growth chamber without covering showed necrosis (Figure 12) indicating the development of necrosis induced by SnTox1 is light dependent. Interestingly, necrosis did develop on the dark treated plants once they were treated with a 12 h light-dark cycle for 2 additional days.
A very similar situation was observed in the inoculation experiment. CS leaves showed no disease symptoms at 3 days post inoculation when plants were kept in the dark and similar to the infiltration experiment, the lesions developed once the dark-treated plants were moved to the light again (Figure 12).
To investigate the role of SnTox1 in disease development, we tagged both the avirulent isolate Sn79-1087 and the pathogenic strain Sn79+SnTox1A1 with GFP and compared their infection processes by fluorescence microscopy in wheat lines CS (Snn1) and the Snn1 mutant, CS ems237 (snn1). The inoculation of CS with the SnTox1-producing strain Sn79+SnTox1A1 resulted in an infection (susceptible interaction); however, the other three combinations (CS inoculated with Sn79-1087, CS ems237 inoculated with Sn79-1087, and CS ems237 inoculated with Sn79+SnTox1A1) gave no disease (resistant interaction) (Figure 13). Within 24 HPI, there was little difference observed between resistant and susceptible interactions. During this period, conidia germinated, grew short hyphae and began the penetration process. The pathogen was able to initiate penetration in both types of reactions visualized by the formation of the indistinct penetration structure called a hyphopodia [26]; Figure 13 A, B) and by autofluorescence of the damaged epidermal cell walls (Figure 13 A, B). We observed mainly direct penetration of the leaf surface over both periclinal and anticlinal epidermal cell walls.
A strong green autofluorescence was observed beneath the epidermis by 2 DPI in the susceptible interaction, suggesting that the pathogen had successfully penetrated through the epidermal cell layer and started the infection of mesophyll cells (Figure 13C). However, in the resistant interaction, the pathogen grew extensively on the leaf surface and no green autofluorescence was visible (Figure 13D). At 4 DPI, the infection area had enlarged in the susceptible interaction as shown by more mesophyll cells producing a fluorescent signal (Figure 13E). On the leaves of the resistant interactions, most of the fungal mycelium was dead, likely due to scarcity of nutrients, and only a few hyphae continued to grow over the leaf surface with repeated unsuccessful attempts to penetrate (Figure 13 F). The susceptible interaction had induced widespread lesion formation on the leaves by 7 DPI, however, no symptoms were found on the leaves of the resistant interaction (Figure 13 G, H). Examination under the fluorescent microscope of the necrotic lesion formed from the susceptible reaction revealed the extensive growth of fungal mycelium within the lesion (Figure S6).
The fungal infection process was also compared microscopically on Snn1-containing plants that were either grown under a normal light/dark cycle or in complete darkness after inoculation. The pathogen was able to germinate and generate hyphopodia within 24 HPI in both conditions (data not shown). However at 48 HPI, only the plants grown in a normal light/dark cycle showed successful penetration through the epidermal cell layer and the initiation of the infection of mesophyll cells, evidenced by the autofluorescence of the mesophyll cells (Figure 14A). In the plants that were kept in the dark, no autofluorescence was observed in the mesophyll cells and the pathogen still remained on the leaf surface without having successfully penetrated the epidermis (Figure 14B).
The necrotrophic fungal pathogen S. nodorum produces multiple necrotrophic effectors (host-selective toxins) that function as virulence factors during the infection process. The cloning of these necrotrophic effector genes is an essential step in the characterization and elucidation of the molecular and biochemical mechanism of fungal pathogenesis in the wheat-S. nodorum pathosystem. Besides the traditional biochemical and genetic tools, new genomic strategies have been recently applied for the identification and cloning of effector genes in a number of fungi and oomycetes as more genome and other sequence data becomes available. A typical procedure would include a process of data mining to identify candidate genes that meet a set of specific criteria followed by gene validation through functional analysis. High throughput functional genomics [1] as well as comparative genomics and association genetics [41] have been successfully used for the identification of pathogen effector genes in fungi and oomycetes. In the current study, we used a set of criteria to mine the S. nodorum genome sequence dataset for the identification of necrotrophic effector genes. This strategy led to the successful identification of SnTox1 from S. nodorum. Through heterologous expression, gene transformation, and gene disruption, we have provided convincing evidence that the candidate gene SNOG20078 (Gene ID: 5974395) is the SnTox1-encoding gene. This research further highlights the value of genome sequence data along with efficient bioinformatics tools in identifying effector genes. We are continuing to use this strategy to identify additional S. nodorum necrotrophic effector genes.
SnTox1 was identified using a set of criteria based on the cloned S. nodorum effector genes SnToxA and SnTox3; however, the SnTox1 gene does have some unique features. Unlike many previously identified effector genes including those from Leptosphaeria maculans [51]–[53], Magnaporthe grisea [41], [54], Fusarium oxysporum f. sp. lycopersici [55], [56], Blumeria graminis f. sp. hordei [57], and those from several Phytophthora species [58], SnTox1 lies in a gene-rich region and was flanked closely by other genes. Except for a short (≈300 bp) sequence predicted to be a truncated molly-type RE, no other obvious RE or AT-rich sequences were identified within the 300 kb genome region surrounding SnTox1 (http://genome.jgi-psf.org/Stano1/Stano1.info.html) showing that not all effector genes are associated with an abundance of repetitive or transposable elements.
The occurrence of effector genes in close proximity to one another has also been reported for several fungal and oomycete pathogens [53], [59]–[62]. This does not appear to be the case for S. nodorum. The three S. nodorum effector genes (SnToxA, SnTox1, and SnTox3) were located on different supercontigs and have been shown by pulse field gel electrophoresis and Southern analysis to reside on 2.35, 1.88 and 1.66 Mb chromosomes, respectively, in SN15 (data not shown) indicating that these genes are not clustered.
Using a worldwide collection of 777 S. nodorum isolates, SnTox1 was found to be present in ∼85% of isolates, a markedly higher frequency than found for SnToxA (∼36%) and SnTox3 (∼60%) [14], [63]. Like the other NEs, SnTox1 was shown to have a presence/absence polymorphism within individual wheat fields. This type of polymorphism has been reported in other fungal pathosystems, as reviewed in Stergiopoulos and de Wit [64]. The frequency of SnTox1 varied significantly across regional populations. We hypothesize that regional differences in the frequency of SnTox1 reflect regional differences in the frequency of Snn1. However this correlation was not apparent when tested on a small worldwide collection of wheat. We found that Snn1 is most prevalent in durum wheat lines and much more rare among hexaploid bread wheat lines throughout the world (data not shown). This could indicate that the maintenance of Snn1 in durum wheat is associated with another important trait. Widespread deployment of wheat cultivars lacking Snn1 could cause the frequency of SnTox1 to decrease if there is a fitness cost associated with producing the effector. But the large effective population sizes of S. nodorum [65] make the complete loss of the effector through genetic drift unlikely.
Observed diversity at the SnTox1 locus was found to fit a model of diversifying selection significantly better than a neutral model. Positive selection was found for 4 of the SnTox1 codons, consistent with the growing list of prokaryotic and eukaryotic effector candidates that exhibit positive selection [66]. None of the non-synonymous substitutions were found in the signal peptide, the putative chitin-binding domain, the putative Avr4-like domain or any of the cysteine codons. This suggests that the effector's functional domains were preserved, while more flexible amino acid sites were subject to diversifying selection. Possible differences in activity between different protein variants of SnTox1 are currently being tested.
Similar to SnToxA and SnTox3, SnTox1 was shown to play a significant role in disease development. Results presented here on the SnTox1-Snn1 interaction provide further evidence that the necrotrophic wheat-S. nodorum system is largely based on specific host-effector interactions that act in ETS [14], [15], which essentially has the opposite outcome of ETI that operates in many biotrophic systems [3], [4].
One of the most striking features of the SnTox1 protein as a necrotrophic effector is the high cysteine residue content. This feature is often associated with fungal avirulence gene products such as the Avr and ECP effectors from Cladosporium fulvum [64], SIX (secreted in xylem) effectors from Fusarium oxysporum f.sp. lycopersici [55], and Nip1 from Rhynchosporium secalis [67]. The predicted mature SnTox1 protein has 100 amino acids, 16 of which are cysteine residues, the richest of all effectors that have been identified. The high content of cysteine residues and high stability suggest that SnTox1 may function in the plant apoplastic space which is abundant in plant defense components. We are currently investigating the subcellular location of SnTox1. Most small cysteine-rich secreted effectors from the tomato fungal pathogen C. fulvum such as Avr2, Avr4, Avr9, and ECP2 are thought to function exclusively in the apoplast to inhibit and protect against plant hydrolytic enzymes [64]. ECP6, another C. fulvum effector containing LysM chitin binding domains was recently found functioning apoplastically as a scavenger of fungal chitin to prevent it from eliciting PAMP-triggered immunity in planta [68]. Interestingly, we observed that SnTox1 has some similarity to C. fulvum Avr4 within the chitin-binding domain and in the positions of six of the cysteine residues at the C-terminus. However, further tests are needed to determine the binding activity and functional roles of the putative CB domain in SnTox1. The presence of a potential chitin binding domain provides a point of investigation for an added function for SnTox1 in addition to its interaction with Snn1.
Successful penetration is a prerequisite for a pathogen to establish itself and fulfill its colonization in planta. For S. nodorum, previous studies have observed direct penetration through the junction of epidermal cells [69] or penetration through stomata [70] or both [26]. Based on our observation using GFP-tagging and confocal fluorescent microscopy, the fungus predominantly used direct penetration through the junction of epidermal cells, and stomatal entry was not evident. We have observed that the fungal mycelium grew over guard cells and anchored the penetration point between the junction of the guard cell and the adjacent epidermal cell instead of going through the stomata (data not shown). Although the avirulent isolate belongs to S. nodorum, the preference for direct penetration, which is different from that reported by Solomon et al. [26], may be due to its adaptation to wild grasses from which it was originally isolated.
It was our observation that the fungus could initiate direct penetration by producing a hyphopodia in both the resistant and the susceptible interaction with little difference, which agrees with previous reports [69] indicating that SnTox1 is not required for hyphopodia formation and the initial degradation of the cuticle layer and the cell wall between the junctions of the epidermal cells. Hydrolytic enzymes or other mechanisms may be employed by the fungus to breach the initial physical barrier. Several cell wall-degrading enzymes such as amylase, pectin methyl esterase, polygalacturonases, xylanases, and cellulase have been found to be produced in vitro and during the infection of wheat leaves by S. nodorum [71]. As infection progressed, the pathogen was unable to penetrate through the epidermal cell layer and therefore could not reach the mesophyll cells to establish a successful infection without the SnTox1-Snn1 interaction. This suggests that SnTox1 is significant in the initial penetration process across the epidermal cell layer. Our hypothesis is that SnTox1 interacts with Snn1 to induce cell death in epidermal cells, providing the fungus with nutrients for further invasive growth. In Cochliobolus victoria on oat and Arabidopsis systems, it was also observed that fungal penetration ceases following appressorium development and hyphae remain on the leaf surface in the absence of a compatible interaction, which requires both victorin and its corresponding sensitivity gene [22]. Our speculation was further supported by the fact that the inoculation of an Snn1 line (CS) with SnTox1 transformed avirulent isolates induced widespread necrosis - presumably programmed cell death - on leaves. Furthermore, inoculation with the SnTox1-knock out virulent strain lost the ability to cause this necrotic reaction. Additionally, qPCR revealed that SnTox1 expression was induced in planta starting as early as 12 HPI and increased at an accelerated rate from 12 to 24 HPI when the fungus was observed to penetrate. Collectively, this suggests that S. nodorum may use SnTox1 to induce cell death in the epidermal cells, providing a portal to enter the plant and subsequently feeding from dead cells to gain nutrients for further invasive growth.
It is well known that plant defenses against pests and pathogens are commonly influenced by environmental conditions, including light. Many studies have demonstrated the requirement of illumination for the interaction of plants with a diversity of bacterial and fungal pathogens as well as the isolated pathogenic elicitors [72], [73]. The effect of light on the disease development of SNB was first noticed by Baker and Smith [69] who observed that the necrotic reaction and lesion coalescence tended to be suppressed in the absence of light. The necrotrophic effector ToxA, was also shown to induce light-dependent necrosis on Tsn1 lines [37]. Among the S. nodorum necrotrophic effectors published to date, all effectors except SnTox3 have been shown to be light dependent [13], [32]. Using heterologously expressed SnTox1 and the avirulent isolate carrying the SnTox1 gene, we showed clearly that the necrosis and disease susceptibility induced by SnTox1 on Snn1 lines were completely dependent on light. The requirement of light for resistance to biotrophic disease as well as susceptibility to necrotrophic disease suggests a common host mechanism shared by reactions to the two classes of disease interactions.
The molecular mechanism underlying the light dependency of plant pathogen interactions is still poorly understood; however, research on the ToxA-Tsn1 interaction has shown that ToxA is internalized in the plant cell followed by localization to the chloroplast and induction of photosystem alterations (reviewed in [40]), providing a hint for the influence of light on this interaction. Recently, it was demonstrated that Tsn1 is regulated by light and its expression is significantly suppressed in the dark [15], providing a possible explanation for the light dependency of the ToxA-Tsn1 interaction. SnTox1 is cysteine rich and therefore possibly acts in the apoplastic space. If SnTox1 remains in the apoplastic space, different mechanisms would likely be involved even though both are dependent on light. In Arabidopsis, plants kept in the dark do not accumulate H2O2 in the chloroplasts and show significantly delayed HR cell death after a resistance signaling pathway is activated [49]. This indicates that light is required for H2O2 production in chloroplasts and that this H2O2 production is critical to programmed cell death. The DAB staining in CS (Snn1) leaves infiltrated with SnTox1 was found to be associated with the chloroplast and the CS plant infiltrated with SnTox1 showed no DAB staining if kept in the dark (data not shown), suggesting a similar mechanism underlying SnTox1-induced cell death. Very interestingly, we found that plants kept in the dark developed necrosis and disease symptoms once transitioned to a normal photoperiod. Therefore signal transduction appears to pause rather than stop in the absence of light. This may indicate that the SnTox1 signal is progressing to the chloroplast but this process is interrupted in the absence of light.
Biotrophic effectors often function as elicitors of programmed cell death (PCD) thereby activating the resistance response in host plants containing the corresponding resistance genes. The host resistance reaction begins with the direct or indirect recognition of the pathogen-produced effector by the resistance gene product, followed by a complicated signaling pathway and a series of biochemical and physiological responses in host plant cells [74]. The host response often includes an oxidative burst, cell wall restructuring, PR-gene expression and antimicrobial compound production culminating in a localized cell death at the infection site. This PCD is known as a hypersensitive response and is typically aimed at halting further colonization by the pathogen [5]. A set of biochemical tests has shown that SnTox1 is able to induce resistance-like host responses and PCD evidenced by the H2O2 production, stronger expression of PR-genes, and DNA laddering in lines carrying Snn1. It is important to note that SnTox1 physiologically evoked a widespread necrotic flecking on the Snn1 line, which is symptomatically similar to the hypersensitive response in biotrophic disease systems. However, this necrosis spreads into larger lesions resulting in susceptibility (sporulation) rather than resistance (prevention or inhibition of sporulation). Together, this indicates that SnTox1 is likely functioning biochemically and physiologically similar to a biotrophic effector (avirulence factor) in the presence of Snn1 but with a different end result.
A number of other necrotrophic effectors have also been shown to invoke a host resistance response [9], [17], [40]. It has generally been thought that necrotrophic plant pathogenic fungi possess simplistic infection mechanisms that rely on lytic and degrading enzymes [11]. In contrast, biotrophic fungal pathogen interactions have been considered more sophisticated due to the formation of special penetration and feeding structures, secretion of effectors to overcome plant PAMP triggered immunity and a constantly changing effector complement to avoid recognition by the plant innate immune system. However, three genes conferring sensitivity to necrotrophic effectors as well as susceptibility to the corresponding necrotrophic fungal pathogens have been cloned, and all possess resistance gene-like features [15], [22], [23]. Therefore, it seems that necrotrophic fungal pathogens may subvert plant resistance mechanisms for their own good. Here, we clearly showed that SnTox1 is an important virulence factor for S. nodorum in the presence of Snn1 and that the host response to SnTox1 shows several similarities to a classical resistance response induced by many biotrophic effectors, however, the outcome of the host recognition was susceptibility rather than resistance.
SnTox1 is the third effector gene that we have cloned and characterized from S. nodorum, which further strengthens the hypothesis that the wheat-S. nodorum pathosystem is based largely on host-effector interactions. The three effector genes cloned have provided molecular tools to study the mechanisms underlying disease in this system, an emerging model for necrotrophic fungal diseases.
A series of experimental and bioinformatic criteria associated with effectors were evaluated to produce a candidate gene ranking of the predicted genes in the S. nodorum genome. These criteria were based on the known and predicted properties of effectors. Genes matching different criteria were given scores from 1 to 6. The sum of scores for each gene was ranked and the top 100 genes were considered. The criteria used data from mass-spectrometry analyses of culture filtrates, a genome sequence scan of the strains Sn4 and Sn79-1087, an in planta microarray experiment and various bioinformatics analyses. The criteria were as follows: predicted to be less than 30 kDa (1 point), cysteine rich (>1 standard deviation more cys residues than expected of a protein of that size) (2 points), detected by MS in culture filtrates (6 points), located within 5 kb of repetitive sequences (2 points), absence of homologues in the NCBI nr database (2 points), presence of RXLR or RGD motifs (2 points), predicted to be secreted (3 points), presence of a modified version of the gene in Sn4 (3 points), absence of the gene in Sn79-1087 (4 points), and a gene expression profile similar to ToxA and Tox3 (3 points).
The total RNA of 7 day old mycelium of SN15 grown in Fries media [27] was prepared using the RNeasy plant mini kit (Qiagen) and treated with RNase-free DNase I (Promega). First-strand cDNA was synthesized from 2 µg of total RNA using TaqMan Reverse Transcription Reagents (Applied Biosystems). The coding region of SNOG_20078 was amplified from the above cDNA sample using primers 20078CF_EcoRI and 20078CR_ApaI containing the indicated restriction site (Table S4). The cloning of SNOG_20078 into the corresponding sequencing and expression vectors, yeast transformation, and preparation of culture filtrates from yeast cultures all followed the procedure described by Liu et al. [14]. The pGAPZ A vector containing the SNOG_20078 gene was linearized with AvrII before transformation. Culture filtrates of the yeast culture transformed with the SNOG_20078 coding region were infiltrated into wheat lines including BR34 (snn1), Grandin (snn1), BG220 (snn1), BG223 (snn1), BG261 (snn1), W-7984 (Snn1), Chinese Spring (Snn1), Opata85 (snn1), and ND495 (snn1). Because the culture filtrates caused necrosis on W-7984 and CS, which both possess Snn1 [27], it was infiltrated onto CS 1BS-18, CS ems237, and the ITMI population [27] for verification of SnTox1 based on its interaction with Snn1. CS 1BS-18 carries a deletion in the distal end of chromosome 1B that harbors the Snn1 locus [27]. CS ems237 is an SnTox1 insensitive mutant derived from CS by EMS (ethane methyl sulfonate) mutagenesis (Faris et al., unpublished data).
A 14 amino acid long peptide (sequence: CKNGKQAAHEAQKQ), designated SnTox1:50–63, was synthesized by GenScript (Piscataway, NJ). The peptide SnTox1:50–63 (4.7 mg, 0.003 mmole) was first conjugated to bovine serum albumin (BSA, 20 mg, 0.0003 mmole, Sigma-Aldrich, St Louis, MO) in the presence of 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide hydrochloride (EDC, 20 mg, Pierce Biotechnology, Rockford, IL) in 2 mL of 100 mM 2-(N-morpholino) ethanesulfonic acid buffer, pH 6 overnight at 4°C. The protein was separated from EDC through a size-based column (D-Salt Excellulose, Thermo Scientific, Rockford, IL) and concentrations were determined by the method of Bradford (Bio-Rad Laboratories, Inc. Hercules, CA) using BSA as the calibration standard. Success of the conjugation reaction was assessed on a 13% SDS-PAGE gel. One hundred milligrams of the immunogen were immunized into New Zealand White Rabbits at 3 week intervals for a total of six immunization cycles. The final sera were collected eight days after immunization and were used for western blot analysis.
To prepare the SnTox1 protein sample for western blot analysis, 5 mL of culture filtrate from an SnTox1 yeast culture and control yeast culture (yeast strain transformed with an empty vector) was precipitated by adding 20 mL of methanol and incubating in a -20 freezer overnight. After centrifuging for 10 min at 13,000 rpm on a HERMLE Z 323K centrifuge with a 220.80 V02 rotor (Labnet), the pellet was retained, air dried and re-suspended in 500 µL of a 1× sample loading buffer. Protein gels were loaded with 50 µL of the resulting sample solution. SDS-PAGE, protein transferring, and color development followed a routine protocol described in Meinhardt et al. [36]. To ensure the quality of protein sample preps, the same amount of sample solution was also run on a gel and visualized by coomassie blue staining.
The same RNA extracted from SN15 was used to amplify the 5′ and 3′ ends of the cDNA of SnTox1. The 5′ and 3′ RACE were performed using the Smart RACE cDNA amplification kit (Stratagene, LaJolla, CA) according to the instructions in the user manual with gene-specific primers 20078CF and 20078CR (Table S4). The procedure described by Liu et al. [14] was followed for the cloning and sequencing of the amplified 5′ and 3′ RACE fragments. The obtained sequences from 5′ and 3′ RACE fragments were used to assemble the full length cDNA and determine the 5′ and 3′ UTRs based on the SN15 genome sequence.
SnTox1 and Avr4 homologs were identified from the NCBI non-redundant (nr) protein database (http://www.ncbi.nlm.nih.gov/BLAST/) using BLAST searches. The chitin-binding domains of Avr4 and its homologues were identified using Reverse Position-Specific (RPS)-BLAST searches (www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi). Amino acid alignments were performed using the MegAlign programs from Lasergene 8.1 software (DNASTAR Inc. Madison, WI). Three-dimensional (3D) structure-based sequence alignment of the putative chitin-binding motifs identified in SnTox1 with those of ChtBD1 and ChBD2 proteins were performed following the previously published data on related structures [43], [44], [75].
SnTox1 presence and absence was screened in 777 S. nodorum isolates from seven geographical regions: Australia, Central Asia, East Asia, Europe, Middle East, North America, South America and South Africa (Table S1) using PCR with primer pair Tox1F_Coding and Tox1R_Coding (Table S4). A secondary PCR screen using the conserved primer pair Tox1_XF and Tox1_XR (Table S4) was conducted to confirm questionable PCR amplicons. PCR amplification was performed in 20 µl reactions containing 0.05 µM of each primer (supplied by Microsynth), 1X Dream Taq Buffer (Fermentas), 0.4 µM dNTPs (Fermentas) and 0.5 units of Dream TaqTM DNA polymerase (Fermentas). The PCR cycle parameters were: 2 min initial denaturation at 96°C followed by 35 cycles of 96°C for 30 s, 58°C for 45 s and 72°C for 1 min. A final 5 min extension was made at 72°C. To demonstrate the wide distribution of SnTox1, a subset of the global collection (79 isolates), along with 10 avirulent isolates and several related fungal species including Pti2 (P. tritici-repentis), ND89-19 (P. teres f. teres), Sm15A (P. bromi) and S. tr 9715 (M. graminicola) (Table S2) were evaluated in a dot blot analysis. For dot blot analysis, the DNA of fungal samples was isolated using a BioSprint 15 instrument (QIAGEN) with the corresponding kit (QIAGEN). The DNA samples were blotted onto a nylon membrane using a Bio-Dot microfiltration apparatus (BIO-RAD) following the instructions in the user manual. The entire SnTox1 coding region was PCR amplified from the genomic DNA of SN15 and used as a probe for Southern blot analysis. Probe preparation, DNA hybridization, membrane washing and image acquisition followed the protocol described by Faris et al. [76]. The membrane was stripped and hybridized to the S. nodorum actin gene probe to ensure the quality for all the DNA samples.
Sequences for the entire coding region were obtained using the primer pair Tox1UTR_F and Tox1UTR_R and the primer pair Tox1Fout and Tox1Rout (Table S4). In cases of poor amplification primer pair Tox1F_Coding and a new conserved reverse primer, Tox1R_Conserved (Table S4), were used to confirm observed sequence variation. Sequencing reactions were conducted in 10 µl volume using the BigDye Terminator v3.1 Sequencing Standard Kit (Applied Biosystems) with both the forward and the reverse primer. The cycling parameters were 96°C for 2 min followed by 55 or 99 cycles of 96°C for 10 s, 50°C for 5 s and 60°C for 4 min. The products were cleaned with the illustra Sephadex G-50 fine DNA Grade column (GE healthcare) according to the manufacturer's recommendations and then sequenced with a 3730x/Genetic Analyzer (Applied Biosystems). Alignment of forward and reverse sequences for each isolate was performed in SeqScape software V2.5 (Applied Biosystems, Foster City, CA). Translation and identification of protein haplotypes was also performed using this software.
Codeml implemented in the software PAML (http://abacus.gene.ucl.ac.uk/software/paml.html) was used to test for positive diversifying selection [77]. The program uses four different codon substitution models implemented in a maximum-likelihood framework to test which model, neutral or selection, best fits the data. Each model assumes a different range of values for the estimated value ω (the ratio of non-synonymous to synonymous nucleotide substitutions). Under purifying selection, non-synonymous substitutions are expected to be rare, thus ω will remain below 1. If non-synonymous mutations offer a selective advantage, they will be fixed at a higher rate than synonymous mutations and ω will be greater than one. We compared the null model M1a (neutral), which assumes two site classes, purifying (0<ω0<1) or neutral (ω1 = 1) to the alternative model M2a (selection), which adds another class of diversifying sites (ω2>1). We also compared the more complex null model M7 (neutral) that assumes a beta distribution for 0<ω<1, with the alternative model M8 (selection) which also assumes a beta distribution and adds an additional site class with ω2>1. A likelihood ratio test was used to compare the likelihood estimate scores. The model simultaneously calculates the posterior probability for each codon that belongs to a particular site class (e.i. ω>1). If the posterior probability for a codon is high and it belongs to the site class with ω>1, positive selection can be inferred for that codon, known as Bayes Empirical Bayes [78].
Based on the annotated SN15 genome sequence, four genes SNOG_07153-SNOG_07156 were predicted within a ∼7.6 kb region containing SnTox1. Primers were designed (Table S4) to amplify the gene region from start to stop codon for the four genes identified in the avirulent isolate Sn79-1087. Since all the genes except SNOG_07154 were present in Sn79-1087, one new primer designed in SNOG_07153 (20078g3R, Table S4) was used with the SNOG07155 forward primer to amplify the whole region in several virulent isolates as well as Sn79-1087. The amplified fragments with different sizes were cloned into the pCR-4 TOPO cloning vector (Invitrogen) for sequencing. The sequences from different isolates were analyzed manually to identify the variations in the SnTox1 genomic region with the aid of the genome sequence (http://genome.jgi-psf.org/Stano1/Stano1.home.html).
A ≈1.1 kb sequence of the SnTox1 genomic region including a putative promoter and terminator was amplified from the Sn2000 isolate using primers 20078gF_XbaI and 20078gR_XbaI, each containing an XbaI restriction site sequence (Table S4). The amplified fragment was cloned into the pCR-4 TOPO vector (Invitrogen) for sequencing to verify the identity and XbaI restriction sites. The SnTox1 gene fragment was then released from pCR-4 TOPO plasmid and cloned into the pDAN vector that carries the cpc-1:hygR (hygromycin-resistance gene) cassette. The resulting plasmid, designated pDAN-SnTox1 (Figure S4) containing the 1.1 kb genomic region containing SnTox1 and hygR was used to transform Sn79-1087 protoplasts. Plasmid DNA was prepared through the regular alkaline lysis method as described by Sambrook and Russell [79] followed by the purification of the plasmid DNA using precipitation with PEG 8000 [79]. The plasmid DNA was linearized with EcoR V and concentrated to 1 µg/ µl for transformation. The fungal protoplasting and PEG-mediated transformation followed the procedure described by Liu et al. [14]. The regenerated clones were screened by PCR with primers 20078gF_XbaI and 20078gR_XbaI (Table S4) and verified by Southern analysis [76]. The culture filtrate production, and infiltration and fungal inoculation with Sn79-1087 and SnTox1 transformed strains followed the protocol described previously [27], [45].
The knock out of SnTox1 was performed using a split marker strategy which employed two rounds of PCR to generate replacement fragments as described by Catlett et al. [80] (Figure S4). In the first round of PCR, the 800 bp of 5′ flanking region and 825 bp of 3′ flanking region of SnTox1 were amplified from Sn2000 using two pairs of primers 20078KOF1 with 20078KOF2 and 20078KOF3 with 20078KOF4 (Figure S4, Table S4). Simultaneously, overlapping marker fragments HY and YG of the hygromycin phosphotransferase cassette (HYG) were amplified from pDAN with two pairs of primers, M13F with HY and M13R with YG (Table S4, Figure S4). All amplified fragments were gel purified and then used in a second round of PCR. Two reactions were set up for the second round of PCR with one to fuse and amplify the SnTox1 5′ flanking region with the HY fragment and the other to fuse and amplify the SnTox1 3′ flanking region with the YG fragment by adding the corresponding first round templates and primers. At least 100 µl of PCR reaction was set up for each reaction in the second round. Standard PCR conditions and Taq polymerase (NEB BioLabs) were used for both rounds of amplification except that round 2 used a longer extension time due to the longer template. A small amount of product from the second round of PCR was evaluated on a 1.0% agarose gel to ensure a successful fusion and amplification for each fragment. The remaining product was combined and concentrated by routine ethanol precipitation [79]. The pellet was finally re-suspended in 20 µl of TE (10 mM Tris and 1 mM EDTA) for transformation of Sn2000 protoplasts.
Fungal protoplasting and transformation followed the procedure described by Liu et al. [14]. The regenerated clones were screened using the PCR primers 20078KOF and 20078KOR (Table S4) which amplifies the partial coding region of SnTox1 that was replaced by the hygromycin-resistance gene cassette. The ectopic transformant and two knock out transformants were verified by Southern blot analysis using the SnTox1 coding region as a probe. Spores of the knock out and ectopic strains as well as wild type Sn2000 were inoculated onto wheat lines W-7984, and CS for testing the effect of the SnTox1 knock out.
The International Triticeae Mapping Initiative (ITMI) mapping population was originally used to map the Snn1 gene, which confers sensitivity to SnTox1, and quantitative trait loci conferring resistance/susceptibility to Sn2000 [27], [45]. The same 106 recombinant inbred (RI) lines of this population were used to evaluate the genetically modified fungal strains including two Sn2000 SnTox1 knock out transformants (Sn2000ΔSnTox1–9 and Sn2000ΔSnTox1–15), one Sn2000 ectopic transformant (Sn2000ΔSnTox1-ECT), and the wild type Sn2000 as a control. All strains were evaluated with three biological replications by inoculating their conidia onto the ITMI population as previously described [45]. The disease rating was conducted 7 days post inoculation using a 0–5 rating scale as described by Liu et al. [45]. Composite interval mapping with the average of three disease readings was performed as previously described [30].
The web-based program DISULFIND (http://disulfind.dsi.unifi.it/) was used to predict if a particular cysteine residue was involved in the formation of a disulfide bond (DB_state) as well as the confidence level of the prediction. The state of each cysteine residue was predicted as either involved (1) or not involved (0) in a DB. The scale of confidence of disulfide bonding state prediction ranges from 0 (low) to 9 (high) [46]. The web-based program DiANNA 1.1 (http://clavius.bc.edu/~clotelab/DiANNA/) [47] was used to determine the best connectivity prediction of cysteine residues in SnTox1.
The secondary leaves of CS (≈2 week old plants) were inoculated with a fungal strain modified from the avirulent isolate Sn79-1087 by addition of the SnTox1 gene. The leaf tissues were collected from the inoculated leaves at 1 h, 3 h, 6 h, 12 h, 24 h, 2 d (day), 3 d, 4 d, 5 d, 6 d, and 7 d post inoculation. The RNA was extracted from leaf samples using the RNeasy Plant Mini Kit (QIAGEN) and treated with RNase-free DNase I (Promega). RNA sample quantification, cDNA synthesis, and gene transcript abundance analysis were performed as previously described [15]. The gene specific primers SnTox1qPCRF and SnTox1qPCRR (Table S4) designed within the two exons were used for the SnTox1 gene in qPCR. The previously reported primers ActinF and ActinR [14] were used for the S. nodorum actin gene as internal control.
Because all cysteine residues were predicted to form disulfide bonds, the protein stability of SnTox1 was investigated by DTT and heat treatment. For DTT treatment, the SnTox1 P. pastoris culture filtrate were treated with DTT (Fisher Scientific, Pittsburgh, PA) at final concentrations of 0, 20, or 40 mM and incubated at room temperature for 2 h or 4 h. For heat treatment, the P. pastoris culture filtrate was sealed in a 2 ml centrifuge tube and heated for 30 min or 1 h on a hot plate setting at 100°C. All treated culture filtrates were then infiltrated into CS leaves.
The fully expanded secondary leaves of CS and CS ems237 were infiltrated with the culture filtrates from yeast transformed with SnTox1 or culture filtrates from yeast transformed with an empty vector (as control). At 24, 48 and 72 hours post infiltration, leaf samples were collected and leaf segments with an infiltrated area were cut and stained in a freshly made 1 mg/ml 3′–3′ diaminobenzidine (DAB) (Sigma) solution. The preparation of a DAB staining solution and the staining process followed a procedure described by Thordal-Christensen et al. [48]. The stained leaf tissue was cleared for chlorophyll by placing them on a paper pre-soaked with ethanol/acetic acid solution (3∶1, V/V) in a petri dish and incubating overnight. The cleared leaves were rinsed and stored in a lactic acid/glycerol/H2O solution (1∶1∶1, v/v/v).
The fully expanded secondary leaves of CS and CS ems237 were infiltrated with SnTox1 yeast culture filtrates and control culture filtrates. Leaf samples were taken at 1, 2, 4, 8, 10, 24, 36, 48, 60, and 72 hour post infiltration. The DNA was extracted from the collected leaf samples using the CTAB method [17]. The 5 µl of DNA from each sample were separated on a 2% agarose gel. The gel was stained in ethidium bromide solution for 1 hour, destained in water for 1 h and photographed using a Gel LOGIC 100 image system (Kodak).
The fully expanded secondary leaves of CS and CS ems237 were infiltrated with SnTox1 yeast culture filtrates and control culture filtrates. Five centimeter segments of infiltrated leaf tissue was collected at 1, 2, 4, 8, 10, 24, 36, 48, 60, and 72 hour post infiltration. Three leaves from different plants were collected as three replications for each time point. Total RNA was extracted from all leaf samples using the RNeasy plant kit (QIAGEN) and treated with RNase-free DNaseI (Promega). The RNA quantification and first strand cDNA synthesis were conducted as previously described [15]. Using the cDNA samples, we examined the expression of a total of 28 wheat genes that have been reported or predicted to be involved in the defense response [Lu et al. unpublished, 21] (Table S3). The RT-PCR and agarose gel electrophorsis were performed using a standard procedure. The same cDNA samples from the three replications were used to conduct the qPCR analysis for three genes: PR-1-A1, chitinase (PR-3), and thaumatin-like protein (PR-5) following the description by Faris et al. [15].
The gGFP vector [81] was used to transform the green fluorescence protein gene into two fungal strains that were only different in the production of SnTox1. One was the avirulent Sn79-1087 that did not produce SnTox1 nor did it cause disease, and the other was an Sn79-1087 SnTox1 transformant (Sn79+SnTox1A1) that expressed SnTox1 and caused disease on Snn1 lines. Since Sn79+SnTox1A1 already carried the hygromycin resistance resulting from the SnTox1 transformation, the plasmid pII99 [82] containing geneticin resistance, was used with gGFP for co-transformation of this fungus. The plasmid DNA preparation, fungal protoplasting, and fungal transformation followed the same methods described above. For all transformations, at least 20 µg of each plasmid DNA linearized with the corresponding restriction enzyme (gGFP with BglII and pII99 with EcoRV) was used. The transformants with the strongest GFP signal were selected for both strains under the Nikon Eclipse TE-2000U microscope equipped with a GFP filter and UV light (Nikon, Japan).
The two GFP-tagged fungal strains were inoculated onto both genotypes of Snn1 (CS) and snn1 (CS ems237) as described in Liu et al. [45]. The inoculated leaves were collected at 1 h, 3 h, 6 h, 12 h, 24 h, 2 d, 4 d, and 7 d post inoculation. The leaves were cut into 5 cm long segments and directly mounted onto glass slides. The specimens were examined immediately using a Zeiss Axioplan 2 Imaging Research Microscope with ApoTome confocal component (Carl Zeiss Light Microscopy, Germany) equipped with filter blocks with spectral properties matching those of GFP.
The S. nodorum gene SNOG_20078 has been deposited in Genbank with identity numbers of 5974395 for gene ID and XP_001797505.1 for protein ID. The nucleotide sequence of 12 different haplotypes of SnTox1, designated Tox1_H1–H13, was submitted to GenBank with accession numbers from JN791682 to JN791693. The other genes and proteins referred to in this paper included Cladosporium fulvum Avr4 protein (CAA69643.1), Mycosphaerella fijiensis Avr4-like protein (Protein ID: Mycfi1:87167), Cercospora beticola Avr4-like protein (GU574324), Microsporum gypseum Avr4-like protein (GeneID:10030079) and Geomyces pannorum Avr4-like protein (DY991214).
|
10.1371/journal.ppat.1000316 | Interpain A, a Cysteine Proteinase from Prevotella intermedia, Inhibits Complement by Degrading Complement Factor C3 | Periodontitis is an inflammatory disease of the supporting structures of the teeth caused by, among other pathogens, Prevotella intermedia. Many strains of P. intermedia are resistant to killing by the human complement system, which is present at up to 70% of serum concentration in gingival crevicular fluid. Incubation of human serum with recombinant cysteine protease of P. intermedia (interpain A) resulted in a drastic decrease in bactericidal activity of the serum. Furthermore, a clinical strain 59 expressing interpain A was more serum-resistant than another clinical strain 57, which did not express interpain A, as determined by Western blotting. Moreover, in the presence of the cysteine protease inhibitor E64, the killing of strain 59 by human serum was enhanced. Importantly, we found that the majority of P. intermedia strains isolated from chronic and aggressive periodontitis carry and express the interpain A gene. The protective effect of interpain A against serum bactericidal activity was found to be attributable to its ability to inhibit all three complement pathways through the efficient degradation of the α-chain of C3—the major complement factor common to all three pathways. P. intermedia has been known to co-aggregate with P. gingivalis, which produce gingipains to efficiently degrade complement factors. Here, interpain A was found to have a synergistic effect with gingipains on complement degradation. In addition, interpain A was able to activate the C1 complex in serum, causing deposition of C1q on inert and bacterial surfaces, which may be important at initial stages of infection when local inflammatory reaction may be beneficial for a pathogen. Taken together, the newly characterized interpain A proteinase appears to be an important virulence factor of P. intermedia.
| Prevotella intermedia is one of the bacterial pathogens that has been implicated in causing periodontitis—an endemic inflammatory disease of the supporting structures of the teeth. The complement system is an important part of host innate immunity and is able to directly kill invading bacteria. To become successful pathogens, many strains of P. intermedia developed mechanisms making them very resistant to killing by complement. We found that a cysteine protease, interpain A, that is produced by many clinical strains of P. intermedia was able to destroy the bacterial killing activity of human serum. A strain of P. intermedia that produces interpain A was found to be more resistant to complement than the one lacking interpain A, and the resistance of the interpain A–producing strain could be diminished by a specific inhibitor of cysteine proteases. We attributed the protective effect of interpain A to its ability to inhibit the complement system through the efficient degradation of C3—a major complement protein that is common to all three pathways of complement activation. Understanding the mechanism governing pathogen resistance to complement may help us to design novel therapeutic strategies to prevent or treat an important bacterial disease.
| Periodontitis is an inflammatory condition with an infective etiology that leads to loss of tooth support. Prevotella intermedia is a major bacterial periodontal pathogen in humans together with Porphyromonas gingivalis and Aggregatibacter actinomycetemcomitans [1]. P. intermedia is often recovered from subgingival plaque in patients suffering from acute necrotising gingivitis, pregnancy gingivitis and chronic periodontitis [2]. Recently, P. intermedia was reported to be found in 14% of adult population in Finland and there was association between the carriage of this species and the number of teeth with deepened periodontal pockets [3]. P. intermedia was also frequently isolated from root canal infections [4]. Periodontitis is one of the most common diseases affecting humans and is primarily the result of colonization of the subgingival surfaces of teeth by bacteria. The complex interaction between these bacteria harboring many virulence factors and the host's immune response results in localized chronic inflammation and subsequent destruction of the supporting structures of the tooth. Proteinases are crucial virulence factors produced by many periodontal pathogens, which can cause the degradation of host proteins for essential nutrients but they can also protect the bacteria from the host's defenses such as the complement system [5],[6].
Complement is a major arm of the innate immune defense system and its main function is to recognize and destroy microorganisms [7]. The three pathways of human complement ensure that virtually any non-host surface is recognized as hostile. The classical pathway is usually mediated by binding of the C1 complex (composed of recognition molecule C1q and two proteinases C1s and C1r) to invading pathogens either directly or via immunoglobulins. The lectin pathway is able to recognize, via mannose-binding lectin (MBL), polysaccharide molecules normally present only on microbial surfaces. Finally, complement can also be activated through the alternative pathway, which is not so much an activation pathway but as a failure to appropriately regulate the constant low-level spontaneous activation of C3 (constantly initiated due to inherent instability of this protein). All three pathways lead to opsonisation of the pathogen with C3b (activated form of complement factor C3), which enhances phagocytosis by phagocytes. Furthermore, anaphylatoxins C5a and C3a are released as byproducts to attract phagocytes to the site of infection. Finally, the end result of the complement cascade is formation of the membrane attack complex and bacterial cell lysis. Host cells protect themselves from bystander damage following complement activation through the expression of membrane-bound or recruitment of soluble endogenous complement inhibitors.
Complement deficiencies are very rare but it has been observed that partial C4 gene deficiencies are more frequent in patients with severe chronic periodontitis [8]. A patient with aggressive periodontitis and severe edema, localized to the free gingival tissues was reported to be deficient in C1-inhibitor [9]. Furthermore, the highest salivary levels of C3 were measured in periodontally healthy subjects while low levels were often found in edentulous and chronic periodontitis patients [10].
It has been demonstrated that heat inactivation of NHS (i.e. inactivation of complement) significantly reduced opsonic activity for P. intermedia in vitro [11] suggesting that complement is important for host defense against this pathogen. Previous studies have shown that P. intermedia was opsonized by the alternative pathway in the absence of the classical pathway, probably in response to the endotoxin [12], however, kinetic studies revealed that opsonisation proceeded at significantly faster rates when the classical pathway was intact [11]. Interestingly, the alternative pathway contributed to the killing of serum sensitive strains while the classical pathway was primarily responsible for killing of strains with intermediate sensitivity [13]. Therefore, it appears that complement is able to recognize P. intermedia via several sensory molecules. However, it appears that P. intermedia is able to override to some extent the complement defenses and to establish chronic infections in the oral cavity.
Every successful human pathogen must develop means to circumvent complement. Many bacteria are able to capture human complement inhibitors such as C4b-binding protein and factor H thereby inhibiting complement and avoiding opsonisation and lysis [14]–[16]. Herpes viruses, on the other hand, produce their own homologues of complement inhibitors [17]. Furthermore, many bacteria use proteinases to incapacitate components of the complement system. For example, most strains of P. gingivalis are resistant to bacteriolytic activity of human serum [13],[18] and the gingipain proteinases have been implicated as the major factor providing protection against complement in serum [5], [19]–[22]. For a number of Prevotella subspecies and strains, including P. intermedia, the level of proteolytic activity for clinical strains was significantly higher than that recorded for commensal strains isolated from healthy mouths [23]. This, we hypothesize, may provide P. intermedia with serum protection.
We have identified three cysteine proteinases in the genome of P. intermedia that appeared to be homologues of SpeB protein of Streptococcus pyogenes [24]. Recently, the first of these genes coding for interpain A (InpA; locus PIN0048) was studied in more detail and its 3D structure was determined [25]. Based on similarity of primary and tertiary structures to known proteinases, InpA is now classified into clan CA, family C10 and registered in the peptidase database MEROPS ([26]; http://merops.sanger.ac.uk). InpA is a secreted protein composed of 868 amino acid residues including a 44-residue signal peptide, a pro-domain (Ala1-Asn111), a catalytic domain (Val112-Pro359) and a 465-residue C-terminal extension arranged in domains with putative regulatory and secretory functions. However, the specific target(s) and function of InpA have yet to be characterized. In the present study, we have examined in detail the effect of InpA on the human complement system and found that this proteinase targets mainly the C3 component, thereby inhibiting all three complement pathways simultaneously.
In order to estimate what fraction of P. intermedia strains found in periodontitis carry the inpA gene, detection by PCR was used on subgingival plaque samples obtained from 24 and 58 patients with chronic and aggressive periodontitis, respectively. We have validated specificity of the PCR assay by investigating 25 samples that were negative for P. intermedia but rich in other periodontal pathogens. No positive signal was obtained in any of the tested Prevotella-negative samples showing that the assay is specific for Prevotella inpA gene (data not shown). P. intermedia was detected in 33% and 57% of plaque samples from chronic and aggressive periodontitis, respectively (Table 1). The majority of P. intermedia-positive samples also yielded positive results regarding the inpA gene implying that InpA fulfils some important physiological function. Similarly, we found that the majority of cultivated P. intermedia strains from various sources also express InpA at the protein level as shown by Western blotting analysis of culture supernatants (Figure 1A). The upper band recognized by the specific antibody corresponds to an unprocessed form of InpA while the lower bands are products of autocatalytic processing [25]. Western blotting of lysates of bacterial cells did not yield a signal implying that InpA is mainly secreted by the bacteria and does not associate in large amounts with cell wall in the strains tested (data not shown).
Furthermore, we have detected InpA protein in gingival crevicular fluid samples collected from four chronic periodontitis patients characterized with regard to pocket depth and bleeding-on-probing. The samples were analyzed for the P. intermedia load using qPCR and subjected to Western blotting analysis. We detected InpA in various forms in samples obtained from patients with significant load of P. intermedia but not from those negative for this pathogen (Figure 1B). The 90 kDa form is the unprocessed full length protein, the 76 kDa and the 40 kDa proteins are processed on the N-terminus and the C-terminus, respectively, while the 28 kDa form is the mature, fully-processed protein. These molecular weights are calculated based on amino acid composition. The 40 kDa form runs in fact as 45 kDa protein (28 kDa form as 32 kDa protein) upon separation on 12% SDS-PAGE gel.
In order to quantitatively assess the effect of purified InpA on the bactericidal activity of human serum, we used an E. coli DH5α model system whereby cells were incubated with normal human serum (NHS) pretreated with various concentrations of InpA or its inactive mutant (InpAC154A) and surviving cells enumerated by colony counting. InpA was found to be able to destroy the bactericidal activity of human serum in a dose-dependent manner and rescued E. coli that are otherwise very sensitive to killing by NHS (Figure 1C). Moreover, P. intermedia strains have been known to vary significantly in their ability to resist killing by NHS [13], hence, various strains were investigated to see if there was a relationship between the serum resistance of a given strain and its InpA expression level. By Western blotting, P. intermedia strain 59 producing a large amount of InpA was found to have a 100% survival rate in 20% NHS while only 78% of the strain 57 with non-detectable InpA production survived (Figure 1A and 1D). Furthermore, addition of a cysteine proteinase inhibitor E64 to NHS decreased the ability of P. intermedia strain 59 to survive while it did not affect the killing of strain 57 or E. coli (Figure 1D). Taken together, the results obtained with both purified InpA and P. intermedia strains showed that InpA compromised the bactericidal activity of human serum.
In order to understand in detail how InpA destroys the bactericidal activity of NHS, i.e. complement, the enzyme was incubated at various concentrations with human serum and hemolytic assays were used to assess activity of the classical and alternative pathways of complement in the pre-treated sera. InpA was found to be an efficient inhibitor of both pathways, whereas the inactive mutant InpAC154A did not show any inhibition (Figure 2A and 2B). InpA was able to inhibit the classical pathway by 80% when present at high nanomolar concentrations (0.5 µM) while the alternative pathway was inhibited by 80% at 1.5 µM concentration. It should be noted, however, that 10% serum was used for the alternative pathway hemolytic assay versus 1.25% for the classical pathway. These concentrations were chosen on a basis of the initial titration and represent conditions in which each assay was most sensitive. The alternative pathway is known to require high concentrations of serum in order to function properly in contrast to the classical pathway that is rapidly activated even at fractions of percent of NHS. Taken together, it appears that InpA is approximately equally able to destroy activity of both the classical and alternative pathways.
Each complement pathway is composed of several factors activated in a consecutive manner. In order to assess which complement factor(s) were affected by InpA, a microtiter plate-based assay in which complement activation was initiated by various ligands depending on the pathway analyzed was used and the deposition of successive complement factors was then detected with specific antibodies. In the case of the classical pathway, complement activation was initiated by immunoglobulin deposition. We found that depositions of C1 and C4 from 2% serum were not affected by InpA (Figure 3A and 3B). However, C3 was found to be sensitive to InpA and deposition of C3b from NHS was abolished at 2 µM InpA (Figure 3C). The inactive InpAC154A mutant had no effect on activation and deposition of C3b at any concentration tested. Accordingly, deposition of C9 that appears in the cascade after C3 was inhibited at similar concentrations as C3 indicating that the inhibitory effect on deposition of C9 was due to degradation of C3 (Figure 3D).
For assessment of the lectin pathway, we used plates bound with mannan carbohydrate. In this case, InpA did not affect the binding of MBL, which is the initiator of the pathway (Figure 4A) and weakly inhibited deposition of C4b (Figure 4B). However, similar to the classical pathway, InpA strongly inhibited the deposition of C3b and C9 while the InpAC154A mutant had no effect (Figure 4C and 4D).
The alternative pathway was activated by immobilized zymosan and InpA was found to be able to inhibit deposition of C3b and C9 with a similar efficiency as previously found for the other two pathways (Figure 5A and 5B). Taken together, all three pathways were sensitive to InpA and its main target appeared to be C3, which is the key protein for all pathways of the complement system.
NHS contains several proteinase inhibitors that could potentially inhibit the activity of InpA. However, we found that the InpA activity measured with fluorogenic substrate was not affected by NHS when NHS was present at concentrations up to 30% (Figure 5C).
In order to assess the sites cleaved by InpA, in complement factors, purified C3 and structurally related C4 were incubated with InpA at various molar ratios. The proteins were then separated by SDS-PAGE and visualized using silver staining (Figure 6A and 6B). C3 is composed of covalently linked α- and β-chains while C4 contains α-, β- and γ-chains. For both proteins, InpA first attacks the α-chain while the β-chain is relatively resistant (Figure 6A–6D); which is similar to what we have previously observed for gingipains [5]. The InpAC154A mutant did not cause any degradation of C3 or C4 (Figure 6A and 6B). Interestingly, similar concentrations of InpA were required for the degradation of purified C3 and C4, whereas in the presence of NHS, InpA preferentially inactivates C3 (Figures 3 and 4).
To determine sites of proteolysis by InpA, C3 and C4 were treated with InpA and degradation products were separated by SDS-PAGE electrophoresis. The proteins were transferred to PVDF membrane, visualized with Coomassie (Figure 6E) and selected bands were subjected to N-terminal sequencing. Interestingly, cleavage of the C3 polypeptide chain at the site resulting in the N-terminal sequence SNLDEDIIA generated the exact sequence of an anaphylatoxin fragment C3a. Similarly, the cleavage of C4 producing the N-terminal sequence ALEILQEE generated the exact sequence of C4a. Sequence 2 (SPMYSII) corresponds to the N-terminus of the C3 β-chain.
When degradation of C3 and C4 was assessed at a set concentration of InpA for increasing incubation times, C4 was degraded at a faster rate than C3 (Figure 7A). To determine the kinetic parameters of degradation of C3b by InpA, surface plasmon resonance was employed. When the inactive InpAC154A proteinase was injected over immobilized C3b, no change in signal was detected (data not shown). However, upon injection of InpA, there was a rapid decrease in the signal measured in resonance units (RU) corresponding to degradation of C3b. The initial rates of proteolysis at each concentration of InpA were obtained from the initial slopes in the sensorgrams (Figure 7B). In this system, 1000 RU corresponds to a mass shift of 1 ng/mm2. The analysis demonstrated that 3 µM InpA degrades C3b at an initial rate of 7 pg/s (Figure 7B inset). The kinetic parameters of C3 and C4 degradation by InpA were also determined by fitting initial rates of degradation of α-chains of C3 and C4 into Michaelis-Menten equation. A constant amount of InpA was incubated with increasing concentrations of C3 and C4 and the initial rate of proteolysis at various substrate concentrations was estimated from the decrease of intensity of scanned bands corresponding to α-chains of C3 and C4 as resolved by SDS-PAGE. Using this approach, Km and kcat for C4 degradation was determined to be 4.3+/−0.8 µM and 0.026+/−0.005 s−1, respectively (Figure 7D). Unfortunately, a reasonably accurate measurement of the kinetic constants for C3 was not possible since there was no visible saturation of the initial rate of C3 degradation up to 2 mg/mL (10 µM) of substrate, hence, the Km could only be estimated as greater than 20 µM (Figure 7C).
We have observed previously that gingipains did not degrade C1 but instead were able to cause C1 deposition on surfaces that would not normally activate C1 [5]. In order to assess if this was also the case for InpA, human serum was incubated with InpA in the absence of any immobilized C1 activator and we observed that it did cause deposition of C1q on the empty microtiter plates blocked with BSA (Figure 8A). In the absence of InpA or in the presence of its inactive mutant, the deposition of C1q from serum was negligible as expected.
In addition, InpA was also found to be able to cause deposition of C1q on bacterial surfaces. To this end, Prevotella nigrescens was incubated with NHS containing InpA at different concentrations and the deposition of C1q was measured using flow cytometry. We found that addition of InpA to NHS caused an increase in deposition of C1q on the surface of Prevotella that mimicked results obtained using microtiter plates (Figure 8B). Taken together, our results show that InpA is able to cause deposition of active C1 complex on normally non-activating surfaces such as BSA coated plastic or bacteria. We did not observe degradation of C1q during incubation with InpA, neither when InpA was added to NHS nor when it was incubated with purified C1q (data not shown).
Since InpA and gingipains are often present simultaneously at the sites of infection colonized with P. intermedia and P. gingivalis, we assessed how they acted on complement when present together. To this end, InpA and the three gingipains (HRgpA and RgpB are arginine-specific gingipains while Kgp is lysine-specific) were pre-incubated with 4% NHS at concentrations chosen to affect the activity of the lectin pathway by only 10–30%. The deposition of C3b was assessed and we found that the proteinases acted synergistically since the deposition of C3b in combinations of InpA and the gingipains was lower than predicted if the effects of the proteinases were added separately (Figure 9). For example, InpA alone decreased the deposition of C3b by 30% at the concentration used, while Kgp yielded only 25% decrease. When used together at the same concentrations, InpA and Kgp decreased C3b deposition by 85% instead of 55% that would be expected if these proteinases had only additive effects. When all three gingipains were added together with InpA, the deposition of C3b was inhibited by 93%.
Factors governing P. intermedia infection are poorly studied when compared to other periodontal pathogens such as P. gingivalis. However, it is becoming apparent that all successful human bacterial pathogens must develop strategies to circumvent the complement system [15]. Microorganisms in gingival sulcus are immersed in serum-derived tissue exudate—gingival crevicular fluid, which is similar in composition to human serum. Since complement components are present in gingival crevicular fluid at up to 70% of serum concentration [27] and in vivo there is high level of complement activation in gingival fluid of patients with periodontitis [28],[29], successful evasion of the complement system is paramount for the survival of P. intermedia in the periodontal pocket. One such strategy of defense against complement developed by P. intermedia appears to depend on the production of InpA, which we now show, is able to degrade complement factor C3, which is the central molecule of the whole complement system. Importantly, the majority of P. intermedia strains isolated from aggressive and chronic periodontitis carry and express the inpA gene.
The proteolytic activities of oral bacteria are thought to play important roles in the etiology of periodontitis and dental abscesses. These proteinases may contribute to tissue destruction, increase availability of nutrients and impair host defense by degrading immunoglobulins and components of the complement system. Proteinases of P. intermedia display trypsin-like and dipeptidylpeptidase activities [30] and also have the properties of cysteine proteinases [31]–[33]. They have also been reported to be capable of degrading immunoglobulins, particularly IgG [34],[35], fibronectin [36] and host proteinase inhibitors [37]. The degradation of immunoglobulins was mediated mainly by cysteine proteinase(s) [35]. Now we can add C3 to this list.
Importantly, inhibition of C3 function occurred even when InpA was incubated with whole NHS showing that C3 will be specifically degraded even in the presence of all other plasma proteins (Figures 3–5). This is not the case for C4, which was degraded efficiently when purified proteins were used but its function was only weakly affected in the presence of whole serum. According to kinetic parameters determined with purified proteins, C4 should be a far better target for InpA than C3 in serum. Despite that C4 in serum seems to be resistant to proteolytic inactivation by InpA. To explain this discrepancy, we speculate that C4 may interact with other protein(s) in serum, which hinders InpA access to a cleavage site. Alternatively, the α-chain of C4 may also be susceptible to proteolysis in serum but the cleaved protein is still a functional source of C4b. The latter explanation is supported by the observation that in contrast to C3, proteolysis of C4 is more limited (Figure 6). Such phenomenon has previously been observed for α2-macroglobulin, which remained functionally active after cleavage with gingipains [38]. Importantly, it is clear that InpA will affect C3 in a way that it can no longer propagate the complement cascades; which should be of direct benefit to InpA producing P. intermedia.
Interestingly, InpA showed a preference for the α-chain of C3 and C4, similar to what we have previously observed for gingipains. At low concentrations, gingipains were able to activate complement factors C3, C4 and C5 as they preferentially target the α-chains of these proteins to cause the release of anaphylatoxins C3a and C5a as well as the activated forms C3b, C4b and C5b. Similarly, N-terminal sequencing of C3 and C4 fragments generated by InpA revealed that InpA will also release C3a and C4a. At higher concentrations, gingipains simply degrade these three complement factors, particularly C3, into smaller fragments so that they can no longer propagate the complement cascade [19],[39]. Yet again, we observe a similar phenomenon for InpA in case of C3. Also similar to gingipains, InpA was able to cause the deposition of C1 from serum onto inert surfaces without the need for a specific C1 activator; which may lead to local inflammation. However, whereas this effect could be recreated in vitro using purified C1 for gingipains [5], InpA required serum to be present for this to occur (data not shown). Thus, it appears that InpA may require a third protein to induce C1 deposition from serum. Consequently, an intricate strategy emerges: periodontal bacteria at low concentrations appear to cause non-specific activation of C1 and to generate C5a and C3a fragments—chemotactic factors for neutrophils. This may lead to a low grade inflammation that provides access to nutrients for bacterial growth and colonization. At higher concentrations of bacteria and proteinases, the complement system becomes incapacitated by multiple cleavages of critical proteins within the cascade.
P. intermedia can be highly resistant to complement and survive at very high serum concentrations but there are significant differences between various strains with regard to sensitivity to killing by complement [13]. In this study, we have shown that there is a correlation between the presence of InpA and serum resistance of P. intermedia. Using E. coli as a sensitive model to detect bactericidal activity of human serum, we have found that they were able to survive when supplemented with low micromolar concentrations of InpA in the presence of 2% NHS. In contrast, cells exposed to NHS alone or to NHS containing the inactive interpain mutant showed total loss of viability at this serum concentration. This clearly shows that purified InpA is very efficient at destroying bactericidal activity of NHS. Further, the cysteine proteinase inhibitor E64 diminished serum resistance of P. intermedia strains.
It is plausible that P. intermedia, in similarity to other bacterial pathogens, has several strategies for evasion of killing by complement. P. gingivalis employs not only proteinases for defense from complement [5] but it also produces a surface anionic polysaccharide, the presence of which strongly correlates with exceptional serum resistance of these bacteria [40]. This bacterium also attenuates the effects of complement by capturing human complement inhibitor C4b-binding protein [16]. In this study, we have found that P. intermedia was able to retain some of its ability to resist killing even when incubated with serum containing the broad-spectrum inhibitor E64. However, InpA is a secreted protein and we do not expect large amounts of it being present in our bactericidal assay that has been performed within 1.5 h of culturing. In vivo, the bacteria will have the opportunity to secrete much more interpain into its pericellular environment. Our current methodology does not allow for truly quantitative analysis of the InpA content in gingival crevicular fluid. However, we can estimate from our Western blotting analysis that 20 µL of crevicular fluid contained at least 0.1 µg of InpA. Taking into account at least 20-fold dilution of crevicular fluid upon collection, the concentration of InpA in the two positive samples analyzed must be greater than 100 µg/mL. This corresponds to approximately 4 µM of fully processed InpA, implying that the concentration of InpA is high enough for inhibition of the complement system as described to occur in vivo.
Our experiments also showed that InpA will aid survival of bystander bacterial species, thus, creating a favorable condition for the establishment of a common ecosystem that would be a beneficial habitat for all participating species. P. intermedia, together with Streptococcus gordonii may be considered to be the early colonizers of tooth surfaces, thereby promoting secondary colonization of pathogenic organisms such as P. gingivalis by providing attachment sites, growth substrates and reduced oxygen concentration locally [41],[42]. P. intermedia belongs to the “orange complex”, which encompasses bacterial species bridging between healthy state and advanced periodontitis. Thus, degradation of C3 by InpA in synergy with gingipains of P. gingivalis will complement the host immune evasion strategy of subgingival microbiota. Importantly, Prevotella species readily acquire resistance towards antibiotics [43] and deeper knowledge of how infection and serum resistance occur will be crucial for the development of alternative treatments to periodontal disease.
Purified complement proteins were purchased from Complement Technology.
InpA as well as its inactive mutant InpAC154A (the catalytic cysteine was replaced by alanine) were expressed as His-tagged recombinant proteins in Escherichia coli and purified by affinity chromatography on Fast Flow Ni-NTA Sepharose (Qiagen) followed by anion exchange chromatography (MonoQ, GE Healthcare) as described previously [25]. The amount of active enzyme in wild-type InpA preparation was determined by active site titration using inhibitor E64 (Sigma). Briefly, recombinant protein was activated at 37°C for 15 min in 0.1 M Tris-HCl, 5 mM EDTA, pH 7.5 freshly supplemented with 2 mM DTT and then preincubated with increasing concentrations of E64 for 37 min at room temperature. Residual enzyme activity was determined by measurement of fluorescence (λex = 380 nm and λem = 460 nm) of AMC released from Boc-Val-Leu-Lys-AMC (PeptaNova) added to the reaction mixture at 250 µM final concentration and using the microplate spectrofluorimeter SpectraMax Gemini EM (Molecular Devices). The concentration of active InpA was calculated from the amount of inhibitor needed for total inactivation of the proteinase. The final preparations of wild type InpA and InpAC154A were assayed for possible contamination with lipopolysaccharide using Limulus test (Hycult Biotechnology) and found to contain 7 and 1 ng/mL lipopolysaccharide, respectively. Arginine-specific (HRgpA and RgpB) and lysine-specific (Kgp) gingipains were purified from the P. gingivalis HG66 strain culture fluid as described previously [5]. Before using in any assay, InpA and InpAC154A were preactivated for 15 min by incubation in a buffer specific for the particular assay supplemented with 2 mM DTT.
InpA was activated by 15 min incubation in 0.1 M Tris·HCl, pH 7.6, 5 mM EDTA, 2 mM DTT at 37°C. InpA was mixed with increasing concentrations of NHS and incubated for 30 min at 37°C. Control samples without serum and with E64 were prepared simultaneously. After incubation, the substrate Boc-Val-Leu-Lys-AMC was added to all samples, rendering final volume 200 µL and final concentrations of 16.8 nM InpA, 0–30% NHS, 100 µM E64 and 5% DMSO. Substrate hydrolysis was monitored as AMC release. Activity was determined as the initial velocity of the reaction and expressed in relative fluorescence units (RFU)/s. Results from triplicates were plotted using GraphPad Prism software and calculated as relative activity compared to an uninhibited control.
For detection of P. intermedia in clinical samples, subgingival plaque samples were obtained from patients with severe periodontitis (aggressive periodontitis (n = 24), chronic periodontitis (n = 58)). Two paper points were inserted in each pocket for 20 s and DNA was subsequently extracted using the Genomic Mini system (A&A Biotechnology) according to the manufacturer's recommendations. PCR was carried out using primers: Pi-1: TTT GTT GGG GAG TAA AGC GGG and Pi-2: TCA ACA TCT CTG TAT CCT GCG T [44]. Presence of the inpA gene was determined using PCR with the following primers that were designed based on Oral Pathogen Sequence Database (gene pPI0032; http://www.oralgen.lanl.gov): pPI-1: GAA GGA CAA CTA CAG CGG AAA; pPI-2: TCC TTT CGT TAG TTC GCT GA. Some of the samples were cultivated on Schaedler agar and Schaedler agar supplemented with 7.5 mg/L vancomycin. Colonies typical for P. intermedia were then subcultivated yielding strains 57, 59, 120, 106, BGH10, BGH30, H13 and their identification was confirmed by PCR exactly as described previously [45]. P. intermedia OMZ 248 [46], was kindly provided by Dr. Frandsen (Department of Oral Biology, Royal Dental College, Faculty of Health Sciences, University of Aarhus, Denmark). For the experiments conducted in this study, all P. intermedia strains were grown on blood-enriched tryptic soy broth (TSB) agar plates at 37°C in an anaerobic chamber (Concept 400, Biotrace) with an atmosphere of 90% N2, 5% CO2 and 5% H2. Escherichia coli laboratory strain DH5α (Invitrogen) and Escherichia coli clinical strain were grown on standard Luria-Bertani (LB) agar plates or in LB broth. Prevotella nigrescens (ATCC 25261) was grown on BBL Columbia II agar containing 8.5% horse blood, 0.04% L-cysteine HCl, 5 mg/mL hemin and 2 mg/mL vitamin K1. Bacterial strains used in this study are listed in Table 2.
Crevicular washes were obtained using a previously described method from 4 patients with chronic periodontitis. For analysis of P. intermedia presence, DNA was extracted from 5 µL of crevicular fluid using the High Pure PCR Template Preparation Kit (Roche) according to the manufacturer's recommendations. Real-time PCR was carried out using a RotorGene 2000 (Corbett Research). Primers specific for 16S rDNA from P. intermedia were designed as described by [44]. PCR amplification was carried out as described earlier [47]. Determination of InpA in gingival crevicular fluid samples was performed by Western blotting analysis using rabbit polyclonal Ab against 40 kDa (without C-terminal profragment) form of InpA raised in rabbits by standard immunization with purified recombinant InpAC154A.
Strain E. coli DH5α was cultured in LB broth until exponential growth phase. Cells were harvested, washed once in GVB++ (5 mM veronal buffer pH 7.3, 140 mM NaCl, 0.1% gelatin, 1 mM MgCl2 and 0.15 mM CaCl2) and adjusted to an optical density at 600 nm of 0.5. NHS was prepared from blood taken from six healthy volunteers and pooled. NHS was diluted in GVB++ to a concentration of 2% and incubated with various concentrations of preactivated InpA or InpAC154A for 15 min at RT. Thereafter, 104 bacteria cells were added and incubated with serum supplemented with InpA for 20 min at 37°C in a total volume of 60 µl. After incubation, aliquots were removed, diluted serially and spread onto LB agar plates. Heat inactivated serum (56°C, 30 min) was used as a negative control. Plates were incubated for 12 h in 37°C after which colonies were counted and percentages of the surviving bacteria were calculated.
P. intermedia from four-day old agar plate culture were harvested and washed once in GVB++ and adjusted to an optical density at 600 nm of 0.6. Thereafter, 2×104 bacteria were mixed with 20% NHS diluted in GVB++ and incubated anaerobically for 1.5 h at 37°C in total volume of 110 µl. The aliquots were removed, diluted serially and spread onto TSB plates. Plates were incubated for 4 days at 37°C in an anaerobic chamber after which colonies were counted and percentages of the surviving bacteria were calculated. E. coli were treated in a similar manner except for that 40% NHS was used. All incubations were performed aerobically and the bacteria were spread on LB agar plates for counting colonies after overnight incubation.
P. intermedia strains OMZ 248, 59, 57, 120, 106, BGH 10, BGH 30, H13 and ATCC 25611 were cultured in the Schaedler liquid medium at 37°C in an anaerobic chamber for 5 days. Aliquots of cell culture media adjusted to an optical density at 600 nm of 2.0 were separated under reducing conditions by SDS-PAGE electrophoresis using 12% gel. The proteins were transferred onto PVDF membrane using semi-dry blotting system. After blocking with 50 mM Tris-HCl, 150 mM NaCl, 2 mM CaCl2, 0.1% Tween 20 and 3% fish gelatin, pH 8.0, InpA was visualized using an anti-InpA polyclonal antibody (1∶500 dilution) followed by goat anti-rabbit Abs conjugated to HRP and developed using enhanced chemiluminescence (ECL). The signals were collected using CCD camera (LAS3000, Fujifilm).
To assess activity of the classical pathway, sheep erythrocytes were washed three times with DGVB++ buffer (2.5 mM veronal buffer pH 7.3, 70 mM NaCl, 140 mM glucose, 0.1% gelatin, 1 mM MgCl2 and 0.15 mM CaCl2). The cells were incubated with a complement-fixing antibody (amboceptor; Boehringverke; diluted 1∶3000 in DGVB++ buffer) at a concentration of 109 cells/mL for 20 min at 37°C. After two washes with DGVB++, 5×108 cells/mL were incubated for 1 h at 37°C with 1.25% NHS diluted in DGVB++ buffer (total volume 200 µl). Before incubation with erythrocytes, NHS was pre-incubated with various concentrations of preactivated InpA or InpAC154A for 15 min at RT. The buffer used for activation of InpA did not interfere with the hemolytic assay or erythrocytes (data not shown). The samples were centrifuged and the amount of lysed erythrocytes was determined by spectrophotometric measurement of the amount of released hemoglobin (405 nm).
To assess activity of the alternative pathway, rabbit erythrocytes were washed three times with Mg++EGTA buffer (2.5 mM veronal buffer, containing 70 mM NaCl, 140 mM glucose, 0.1% gelatin, 7 mM MgCl2, 10 mM EGTA, pH 7.3). Erythrocytes at a concentration of 5×108 cells/mL were then incubated for 1.5 h at 37°C with 10% NHS diluted in Mg++ EGTA buffer (total volume 200 µl). NHS used was pre-treated with various concentrations of preactivated InpA or InpAC154A for 15 min at RT. The samples were centrifuged and the amount of lysed erythrocytes was determined spectrophotometrically.
Microtiter plates (Maxisorp; Nunc) were incubated overnight at 4°C with 50 µl of a solution containing 2 µg/mL human aggregated IgG (Immuno), 100 µg/mL mannan (Sigma, M-7504) or 20 µg/mL zymosan (Sigma, Z-4250) in 75 mM sodium carbonate (pH 9.6). Between each step of the procedure, the plates were washed four times with 50 mM Tris-HCl, 150 mM NaCl, and 0.1% Tween 20 (pH 7.5). The wells were blocked with 1% BSA (Sigma) in PBS for 2 h at RT. NHS was diluted in GVB++ buffer and used at a concentration of 2% for C3b, C4b, C1q (classical pathway), 4% for C3b, C4b, MBL (lectin pathway), 6% for C3 (alternative pathway) and 10% for C9 (all three pathways). These concentrations were chosen on the basis of initial titrations. NHS was mixed with various concentrations of preactivated InpA or InpAC154A and incubated in the wells of microtiter plates for 45 min at 37°C for C9 and MBL and 20 min at 37°C for C3b and C4b in case of the alternative and the lectin pathways. For the classical pathway, NHS was incubated with preactivated InpA or InpAC154A for 15 min at RT in eppendorf tubes and the enzyme was inhibited by addition of 20 µM E-64 (Calbiochem) to avoid degradation of IgM deposited on plates. Immediately after addition of inhibitor, NHS was incubated in microtiter plates for 45 min at 37°C for C9 and C1q and 20 min at 37°C for C3b and C4b. The inhibitor itself did not affect activation of complement at the concentration used (data not shown). Complement activation was assessed by detecting deposited complement factors using rabbit anti-C1q, anti-C4b, anti-C3d polyclonal antibodies (pAbs, DakoCytomation) goat anti-C9 pAb (Complement Technology) and goat anti-MBL (R&D) diluted in the blocking buffer. Bound antibodies were detected with HRP-labeled anti-rabbit or anti-goat secondary pAb (DakoCytomation). Bound HRP-labelled pAb were detected with 1,2-phenylenediamine dihydrochloride (OPD)-tablets (DakoCytomation) and the absorbance was measured at 490 nm.
To assess deposition of purified C1q on microtiter plates without any complement activator, plates were blocked with 1% BSA in PBS for 2 h at RT. NHS was diluted in GVB++ buffer to 4% and mixed with various concentration of interpain A. Plates were incubated for 45 min at 37°C with shaking and the deposited C1q was detected with specific antibodies.
P. nigrescens ATCC 25261 from two-day old agar plate cultures were harvested, washed twice in GVB++ buffer and adjusted to an optical density at 600 nm of 1.0. NHS was diluted in GVB++ to a concentration of 5%, mixed with 6×105 cells and incubated with various concentrations of preactivated InpA or InpAC154A for 30 min at 37°C. Thereafter, the cells were washed twice in the binding buffer (10 mM HEPES, 140 mM NaCl, 5 mM KCl, 1 mM MgCl2, 2 mM CaCl2, pH 7.2). C1q deposition was assessed by incubation of the cells with rabbit anti-human C1q FITC-conjugated polyclonal antibodies (DakoCytomation, diluted in the binding buffer 1∶100) for 1 h. The cells were washed twice in the binding buffer and finally resuspended in flow cytometry buffer (50 mM HEPES, 100 mM NaCl, 30 mM NaN3, 1% BSA; pH 7.4). Flow cytometry analysis was performed using FACS Calibur (Beckton Dickinson).
C4 and C3 (0.8 µM each) were incubated with InpA at concentrations ranging from 50 nM to 1250 nM. Incubations were carried out in 0.2 M Tris-HCl, pH 7.4, containing 0.1 M NaCl, 5 mM CaCl2 and 2 mM DTT for 30 min at 37°C. For the time course experiment, C4 and C3 (0.8 µM each) were incubated with 640 nM InpA for 5, 10, 20, 30, 45, 60 and 75 min. The proteins were separated by SDS-PAGE electrophoresis using standard Laemmli procedure and 12% gels. Prior to electrophoresis the samples were boiled for 5 min at 95°C in a sample loading buffer containing 25 mM DTT and 4% SDS. After separation, the gels were stained with silver salts to visualize the separated proteins and quantified by densitometry using ImageGauge (FujiFilm, Tokyo, Japan).
To determine sites of cleavage by InpA, 10 µg of C3 and C4 were incubated with 500 nM preactivated InpA for 2 h at 37°C and the proteins were separated by 12% SDS-PAGE under reducing condition. The proteins were then transferred to PVDF membranes (Pall) and stained using Coomassie Blue. Bands of interest were excised and analyzed by automated Edman degradation in an Applied Biosystems PROCISE 494 HT sequencer with on-line phenylthiohydantion HPLC analysis using a 140 C Microgradient System from Applied Biosystems, operated according to the manufacturer's recommendations.
The analysis was performed according to a previously published protocol [48]. Human C3b was diluted in 10 mM Na-acetate pH 4.0 to a concentration of 30 µg/mL and immobilized on chip CM5 to a level of 3000 RU using amino coupling kit (Biacore) and Biacore 2000. Interpain A was pre-activated by 15 min activation at 37°C in the running buffer (10 mM HEPES, 150 mM NaCl, 1 mM MgCl2, 0.15 mM CaCl2, 0.005% Tween 20, 0.2 mM DTT; pH 7.4) with 2 mM DTT and diluted in the running buffer in a concentration range 0.25–6 µM. Interpain A was then injected at the flow rate of 5 µl/min at 37°C over the immobilized C3met and its activity was quantified as decrease in RU on the sensorgram and analyzed using Biaevaluation software (Biacore).
Several concentrations of C3 (1.2–7.2 µM) and C4 (0.2–4.8 µM) diluted in DGVB++ were incubated with 110 nM or 40 nM of preactivated InpA, respectively. The incubation time was 4 h and 20 min for C3 and C4, respectively. In parallel, the same concentrations of C3 and C4 were incubated without enzyme. Proteins were separated under reducing conditions by SDS PAGE using 12% gel, stained with Coomassie and the gels were scanned followed by densitometry determination of α-chains of C3 and C4 (ImageGauge). Intensity of α-chain bands in the presence of InpA was compared to corresponding controls and expressed as the amount of substrate remaining. Initial velocity of the reaction at each concentration was calculated as amount of substrate consumed within one second and fitted by nonlinear regression into the Michaelis-Menten equation V = (kcat*[E]t*[S])/([S]+Km) using GraphPad Prism. Values Km and kcat were obtained as regression curve parameters. Similar values were obtained from two independent experiments.
The ethical board of Lund University has approved collection of sera from healthy human volunteers. The ethical committee of Jena University approved collection of periodontal plaques and crevicular fluid. Informed consent was obtained from patients and the investigation was performed according to principles of the Declaration of Helsinki.
Student's t-test was used to calculate the p values in order to estimate if the observed differences between experimental results were statistically significant.
|
10.1371/journal.ppat.1005193 | IL-4 Induced Innate CD8+ T Cells Control Persistent Viral Infection | Memory-like CD8+ T cells expressing eomesodermin are a subset of innate T cells initially identified in a number of genetically modified mice, and also exist in wild mice and human. The acquisition of memory phenotype and function by these T cells is dependent on IL–4 produced by PLZF+ innate T cells; however, their physiologic function is still not known. Here we found that these IL-4-induced innate CD8+ T cells are critical for accelerating the control of chronic virus infection. In CIITA-transgenic mice, which have a substantial population of IL-4-induced innate CD8+ T cells, this population facilitated rapid control of viremia and induction of functional anti-viral T-cell responses during infection with chronic form of lymphocytic choriomeningitis virus. Characteristically, anti-viral innate CD8+ T cells accumulated sufficiently during early phase of infection. They produced a robust amount of IFN-γ and TNF-α with enhanced expression of a degranulation marker. Furthermore, this finding was confirmed in wild-type mice. Taken together, the results from our study show that innate CD8+ T cells works as an early defense mechanism against chronic viral infection.
| Over the course of viral infection there may be a limited time period during which the host system can eliminate the virus. When viruses are not eliminated within this period of time, virus can establish persistent infection. Here, we show that IL-4-induced innate CD8+ T cells are able to effectively control chronic virus infection. Innate T cells are heterogeneous population of T cells that acquire effector/memory phenotype as a result of their maturation process in thymus, unlike conventional T cells that differentiate into memory cells after antigen encounter in periphery. Previous data suggest that innate T cells might serve as a first-line of defense against certain bacterial pathogens. IL-4-induced innate CD8+ T cells are a unique subset of innate T cells that were recently identified in both mouse and human. We found that IL-4-induced innate CD8+ T cells immediately accumulated after viral infection and produced a robust amount of effector cytokines. Thereby, IL-4-induced innate CD8+ T cells provide an effective barrier to the establishment of persistent infection via effective virus control during the early phase of viral infection. Collectively our data show that IL-4-induced innate CD8+ T cells works as an early defense mechanism against chronic viral infection.
| Conventional T cells take on naive phenotypes when they emigrate out from the thymus, whereas innate T cells from the thymus are phenotypically of the effector/memory form [1]. Compared with conventional T cells, these innate T cells, such as natural killer T (NKT) cells, mucosal-associated invariant T (MAIT) cells and H2-M3-specific T cells, are selected by interaction with hematopoietic cells rather than thymic epithelial cells, and their development is dependent on IL–15 and the SAP (SLAM-associated protein) signaling pathway [1]. Moreover, most innate T cells express T cell receptors (TCRs) specific for MHC class Ib molecules [1,2].
Memory-like CD8+ T cells expressing eomesodermin (Eomes) are another subset of innate T cells [3]. Although this type of cells is not abundant in wild type C57BL/6 mice, they initially described in Tec-kinase-deficient mice [4,5] and subsequently found in the thymus of a variety of mice in which T-cell-associated genes are deficient [6–13] or CIITA-transgenic (CIITATg) mice in which MHC class II molecules are expressed in thymocytes [14]. Recently, a substantial number of these innate CD8+ T cells was also identified in wild-type BALB/c mice [6] and in human [14]. Eomes+ CD8+ T cells from both mice and human thymus exhibit immediate effector function upon TCR stimulation [6,14]; however, this type of CD8+ T cells has unique characteristics that make them different from MHC class Ib-restricted innate T cells. Firstly, common gamma chain cytokines, particularly IL–4 in this case, drive the expression of Eomes during the intrathymic developmental process [6,14]. Promyelocytic leukemia zinc finger protein (PLZF)+ NKT cells are the major source of IL–4 in wild-type BALB/c and Klf2-deficient mice [6], whereas in CIITATg mice PLZF+ T-T CD4+ T cells are responsible for the production of IL–4 [14,15]. In humans, IL–4 would be produced by both PLZF+ T-T CD4+ T and NKT cells [14]. Secondly, MHC class Ib-restricted innate T cells have a highly restricted TCR repertoire [16], whereas IL-4-induced Eomes+ innate CD8+ T cells from CIITATg mice have a diverse TCR repertoire very much like conventional T cells [14]. This difference in TCR repertoire suggests that they are selected by diverse self-peptides presented by classical MHC class I molecules and raises the possibility that IL-4-induced innate CD8+ T cells perform some functions distinct from those of MHC class Ib-restricted innate T cells during a variety of immune responses. However, the biological relevance of IL-4-induced innate CD8+ T cells has not been elucidated.
Although CD8+ T cells are crucial for the control or elimination of various viral infections, many viruses are able to establish a chronic infection by escaping virus-specific CD8+ T cell responses. The functional inactivation of antigen-specific CD8+ T cells through the triggering of co-inhibitory receptors such as programmed death–1 (PD–1) and cytotoxic T-lymphocyte antigen–4 (CTLA–4) is currently considered to be a conserved mechanism for not only maintaining viral persistence, but also for limiting immunopathology [17,18]. Moreover, an increase in frequency of virus-specific naïve CD8+ T-cell precursors was reported to help control initial viremia but cause differing outcomes with either clearance of wild-type chronic virus or emergence of a T-cell epitope escape mutant virus [19]. However, little is known regarding the impact that the type of CD8+ T cells present during infection has on protection against viral persistence.
In the present study, we investigated the in vivo role of IL-4-induced innate CD8+ T cells in controlling initial viremia using the lymphocytic choriomeningitis virus (LCMV) clone 13 (CL–13) chronic virus infection model. One of the most notable findings from this experiment is that IL-4-induced innate CD8+ T cells produce a robust amount of cytokines such as IFN-γ and TNF-α upon LCMV infection, resulting in the efficient control of viruses from the body and providing an effective barrier to the establishment of viral persistence.
To explore the in vivo function of IL-4-induced Eomes+ CD8+ T cells, we used CIITATg mice in which thymocytes express MHC class II molecules. As reported previously [14], thymus of CIITATg mice contain high numbers of Eomes+ CD8+ T cells, whereas wild-type C57BL/6 mice have only a small number of these cells (Fig 1A). These Eomes+ CD8+ T cells exhibited a phenotype similar to that of Eomes+ memory-like CD8+ T cells identified in other types of gene-manipulated mice [3,6] in that they highly express CXCR3, CD124 (IL-4Rα), CD122 (IL-2Rβ) and CD44, and exhibit low expression of CD24 (Fig 1B). We initially infected both CIITATg and wild-type mice with a conventional dose (2 x 106 PFU/mouse) of LCMV CL–13 and found that CIITATg mice succumbed to early death, whereas wild-type mice did not (Fig 1C). Histopathological analysis of LCMV CL-13-infected CIITATg mice showed edematous lungs where most of the alveolar spaces were filled with transudate (Fig 1D), suggesting that the mice died due to immunopathologic tissue damage.
We next infected mice with a diverse range of viral doses to determine the minimal infectious dose capable of causing chronic infection in wild-type C57BL/6 mice. We found that inoculation of LCMV CL–13 into wild-type mice established chronic virus infection with sustained expression of PD–1 molecules on CD8+ T cells (S1A Fig) and virus persistency in serum (S1B Fig). Based on this, we challenged CIITATg and wild-type mice with this dose of LCMV CL–13 for subsequent studies and monitored the frequency of CD8+ T cells and viral titer in the blood for 28 DPI. Notably, the frequency and number of endogenous CD8+ T cells specific for the LCMV GP33-41 epitope (GP33) were greatly enhanced in CIITATg mice when compared with those in wild-type mice (Fig 2A and 2B). In addition, sustained expression of PD–1 on GP33-specific CD8+ T cells was not induced in CIITATg mice upon LCMV CL–13 infection, although wild-type mice exhibited strongly induced PD–1 expression patterns (Fig 2C and 2D). In the CIITATg mice, the expression of CD127, which are known to be downregulated on exhausted virus-specific CD8+ T cells [19,20] was significantly increased (Fig 2E). This data coincides with a rapid drop in virus titer in blood of CIITATg mice, but not in that of wild-type mice (Fig 2F).
We next asked whether the enhanced CD8+ T-cell response was also present in peripheral tissues, particularly during the late phase of viral infection. To this end, we dissected the CD8+ T-cell response with respect to their number and cytokine responses in the spleen and lungs at 31 DPI. In this experiment, endogenous CD8+ T cells specific for the LCMV GP276-286 epitope (GP276) as well as for LCMV GP33 were analyzed phenotypically and functionally. As expected, the CIITATg mice contained higher numbers of GP33- or GP276-specific CD8+ T cells in the spleen (Fig 3A) and these cells exhibited much lower levels of PD–1 expression on their surface (Fig 3B) compared with wild-type mice. As was the case in peripheral blood, a substantial population of CD127hi virus-specific memory CD8+ T cells was detected in the spleen and lung of CIITATg mice (Fig 3C). A particularly important point in this experiment is the fact that CD8+ T cells from CIITATg mice showed very strong cytokine responses compared with T cells from wild-type mice, in this case IFN-γ and TNF-α release upon ex vivo restimulation with GP33, GP276, or pooled peptides (Fig 3D and 3E). Like that of peripheral blood, this enhanced function of virus-specific CD8+ T cells was associated with decreased viral titers in the spleen and especially in the kidney, which is well known to be a life-long reservoir of chronic LCMV, of CIITATg mice compared with wild-type mice (Fig 3F). Taken together with data from peripheral blood, this strongly suggests that enhanced virus-specific CD8+ T-cell responses both quantitatively and qualitatively contribute to the accelerated control of viremia in CIITATg mice.
The development of IL-4-induced innate CD8+ T cells is dependent on PLZF+ T cells, such as PLZF+ innate CD4+ T cells [14] and NKT cells, as a source of IL–4 [6]. Consistent with the previous report [14], the frequency of innate CD8+ T cells co-expressing CD44 and CXCR3 in the thymus was significantly lower in CIITATgIL-4KO mice than in CIITATg mice (Fig 4A). Thus, to determine whether the enhanced anti-viral CD8+ T-cell response in CIITATg mice is dependent on IL-4-induced innate CD8+ T cells, we used CIITATgIL-4KO mice. When these mice were infected with CL–13 and virus-specific CD8+ T-cell numbers in peripheral blood were monitored until 31 DPI, the overall magnitude of these cells in CIITATgIL-4KO mice was found to be as low as that in control IL-4KO mice infected with CL–13 (Fig 4B). Functionally, virus-specific CD8+ T cells in both CIITATgIL-4KO and IL-4KO mice exhibited an exhausted phenotype in terms of sustained PD–1 expression (Fig 4C and 4D). In addition, the robust cytokine production exhibited in cells from CL-13-infected CIITATg mice upon ex vivo restimulation with epitope peptides was not present in CIITATgIL-4KO mice (Fig 4E). This defect in CD8+ T-cell function suggests a failure of virus control in CIITATgIL-4KO mice. Indeed, virus was not cleared in these mice (Fig 4F). Thus, these results imply that the enhanced CD8+ T-cell response in CIITATg mice infected with CL–13 is dependent on the existence of IL-4-induced innate CD8+ T cells.
It is possible that IL–4 deficiency probably has effects not only on the loss of the IL-4-induced innate CD8+ T cells but also on the development of antibody responses, an important component in determining the outcome of virus infection. To exclude the possibility, we examined cytotoxic T lymphocyte (CTL) and antibody response between wild-type and IL-4KO mice during the course of CL–13 infection. The number of virus-specific CD8+ T cells and their PD–1 expression were not different between wild-type and IL-4KO mice (S2A and S2B Fig). When serum level of LCMV-specific IgG was measured, the level was also similar between these mice (S2C Fig). Accordingly, absolute numbers of follicular helper T (TFH) cells and germinal center (GC) B cells was not different between IL-4KO and wild-type mice (S2D and S2E Fig), which both could not control viruses for a certain period (S2F and S2G Fig). These data suggest that IL–4 itself does not affect the control of chronic viruses.
Next, to confirm the anti-viral function of IL-4-induced innate CD8+ T cells, we generated these cells in P14 TCR transgenic mice, which express a TCR recognizing LCMV GP33 peptide presented by H-2Db molecules. To produce P14 Eomes+ CD8+ T cells, we injected a mixture of BM cells isolated from CIITATg and Thy1.1 P14 mice into irradiated CIITATgPIVKO mice (Fig 5A, T-T P14). In CIITA promoter type IV null (PIVKO) mice, MHC class II molecules are not expressed only on cortical thymic epithelial cells [21], therefore most of CD4+ T cells are selected by MHC class II+ thymocyte-thymocyte interaction pathway in CIITATgPIVKO mice [15]. As a result, CIITATgPIVKO mice generate much higher number of PLZF+ CD4+ T cells as compared to CIITATg mice, so that these mice are able to get much higher amount of IL–4 in thymic environment. As expected, this mixed chimerism allowed most of the P14 CD8+ T cells to express Eomes (Fig 5B). Consistent with a previous report [14], Eomes+ P14 cells developed with a CD127hiCD44hi memory phenotype (Fig 5C). In contrast, wild-type C57BL/6 mice that received Thy1.1 P14 BM cells contained only Eomes- naïve P14 cells (Fig 5A–5C, T-E P14). Eomes+ P14 cells also exhibited enhanced ability for effector cytokine production and cytotoxicity compared with the Eomes- counterpart: a higher fraction of Eomes+ cells produced IFN-γ and expressed a marker of degranulation, CD107a, upon in vitro stimulation with GP33 peptide (Fig 5D).
After generation of virus-specific IL-4-induced innate CD8+ T cells via mixed BM chimerism, we evaluated the anti-viral function of these innate CD8+ T cells in vivo. For this, we transferred Eomes+ or Eomes- P14 cells into congenic hosts, which were then infected with CL–13 (Fig 6A). This strategy allowed us to track the response of IL-4-induced and Eomes- conventional CD8+ T cells after CL–13 infection and to judge their individual contributions to the protection against viral persistence. To examine the proliferation capability of Eomes+ or Eomes- P14 cells after CL–13 infection, each of the cells were labelled and transferred into naïve congenic mice. After 2.5 DPI, Eomes+ P14 cells showed a more division and a higher expression of CD44 compared to Eomes- P14 cells (Fig 6B). In the spleen, a 10-fold higher number of P14 cells accumulated at an early time point (5 DPI) in the mice that received Eomes+ P14 cells than in those that received Eomes- cells, although these cells were detected with similar abundance at a later time point (18 DPI) (Fig 6C). In the mice that received Eomes+ P14 cells, PD–1 was only transiently expressed on P14 cells at 5 DPI and was then later downregulated (Fig 6C). In contrast, the high level of PD–1 expression was sustained on Eomes- P14 cells. Functionally, P14 cells from mice received Eomes+ P14 cells were superior to those from recipients of Eomes- cell, with a higher fraction of these cells capable of producing both IFN-γ and TNF-α (Fig 6D). To test whether the increased CD8+ T-cell activity corresponded to better viral control we measured viral titer in the serum. Indeed, Eomes+ P14 cells also had an effect on reducing viral load during CL–13 infection (Fig 6E). These results indicate that Eomes+ P14 cells provide superior support for anti-viral activity compared to Eomes- P14 cells.
Interestingly, most of the responding CD8+ T cells of recipients were PD-1-negative in the mice that received Eomes+ P14 cells (Fig 6C) compared to those that received Eomes- P14 cells at 18 DPI. Thus, we examined the PD–1 expression on endogenous virus-specific CD8+ T cells, which are initially Eomes-negative before infection, and their cytokine production. PD–1 expression on endogenous GP276 tetramer+ CD8+ T cells was comparable to the donor Eomes+ P14 cells (Fig 6F). In addition, endogenous virus-specific CD8+ T cells in the mice that received Eomes+ P14 cells were not exhausted (Fig 6G). These data suggest that PD–1 expression in both donor Eomes+ P14 cells and endogenous virus-specific CD8+ T cells during the course of CL–13 infection presumably depend on antigen levels rather than intrinsic property of the Eomes+ or Eomes- responding cells.
Our observation that IL-4-induced Eomes+ innate CD8+ T cells are more effective at controlling the CL–13 infection than Eomes- conventional CD8+ T cells (Fig 6) raises a question regarding the underlying mechanism. As shown in Fig 5D, Eomes+ P14 cells were able to produce more effector cytokine per cell level upon antigen stimulation than Eomes- P14 cells. In addition to an elevated effector function, Eomes+ P14 cells also showed better proliferative capability (Fig 6B), resulting in the higher frequency of these cells in the mice that received Eomes+ P14 than Eomes- P14 cells (Fig 6C). Therefore, an enhanced quantity and quality of Eomes+ P14 cells compared to Eomes- P14 cells could contribute to the accelerated control of CL–13 infection. However, it is possible that functional and numerical differences in the two populations depend on viral titers. To address this issue, we co-transferred both Eomes+ and Eomes- P14 cells into same mice (Fig 7A). At 5 DPI, we found that the frequency of Eomes+ P14 cells was significantly higher than that of Eomes- P14 cells (Fig 7B). In addition, the ability to produce effector cytokines, IFN-γ and TNF-α, was better in Eomes+ P14 cells than in Eomes- P14 cells (Fig 7C). These data indicate that functionality and proliferative capability of Eomes+ and Eomes- P14 cells upon antigen stimulation is intrinsically different independently of viral titer.
Next, we wanted to determine whether the anti-viral effect of IL-4-induced innate CD8+ T cells also occur in a wild-type host. Unlike C57BL/6 mice, BALB/c mice have an abundant population of IL-4-induced innate CD8+ T cells, whose generation is supported by IL–4 produced by intrathymic PLZF+ NKT cells [6]. Prior to investigating anti-viral CTL responses in BALB/c strain, we compared IL–4 induced innate CD8+ T-cell phenotypes on CD8+ T cells in uninfected wild-type BALB/c and CD1dKO BALB/c, in which IL-4-induced innate CD8+ T-cell generation is defective due to the absence of NKT cells. The number of virus-specific CD8+ T cells, NP118-126 (NP118) tetramer+ CD8+ T cells, was not different between wild-type and CD1dKO BALB/c mice (Fig 8A). However, the NP118 tetramer+ CD8+ T cells displayed a slightly higher Eomes expression level and the population co-expressing CD44 and CXCR3 among the NP118 tetramer+ CD8+ T cells was significantly increased in wild-type BALB/c mice than in CD1dKO BALB/c mice (Fig 8B). Similarly, the other CD8+ T cells than NP118 tetramer+ CD8+ T cells also contained the higher population of CD44hiCXCR3+ cells in wild-type BALB/c mice than in CD1dKO BALB/c mice (Fig 8C), indicating that the frequency of pre-existing innate CD8+ T cells is higher in wild-type BALB/c than in CD1dKO BALB/c prior to infection, irrespectively of antigen-specificity of CD8+ T cells.
Next, to compare anti-viral CD8+ T-cell responses in wild-type BALB/c mice with those of CD1dKO BALB/c mice, we challenged wild-type BALB/c mice and CD1dKO BALB/c mice with CL–13 (2 x 105 PFU/mouse). The viral load in peripheral blood of wild-type BALB/c mice was significantly decreased compared with that of CD1dKO hosts (Fig 8D). Moreover, wild-type BALB/c mice exhibited decreased viral titer in spleen and kidney and more abundant virus-specific CD8+ T cells in spleen and lungs than CD1dKO mice (Fig 8E and 8F). As was the case in CIITATg mice, PD–1 expression on virus-specific CD8+ T cells in wild-type BALB/c mice was not sustained at a later time point after viral infection in wild-type BALB/c mice (Fig 8G). Furthermore, we observed that CD8+ T cells from wild-type BALB/c mice produced higher amounts of effector cytokines such as IFN-γ and TNF-α than did those from CD1dKO mice (Fig 8H and 8I). These data suggest that IL-4-induced innate CD8+ T cells in a wild-type BALB/c host also contributed to reduce CL–13 viral load compared with CD1dKO mice.
Over the course of viral infection there may be a limited time period during which the host system can eliminate the virus [22]. When viruses are not eliminated within this period of time, virus-specific CD8+ T cells are exhausted via PD–1 (programmed death 1) and its ligand, PD-L1 interaction, resulting in a chronic infection [23]. In this study we demonstrated that IL-4-induced innate CD8+ T cells are able to effectively control the chronic viral infection. For this, we first compared T-cell responses to chronic viral infection induced by LCMV CL–13 in CIITATg and wild-type C57BL/6 mice. The immune system of the CIITATg mouse resembles that of humans with respect to MHC class II expression in both thymic epithelial cells and thymocytes making it a suitable model [24–27]. Thus, PLZF+ T-T CD4+ T cells are generated in response to TCR signals from the MHC class II/peptide complex expressed on thymocytes [15,27,28] and provide IL–4 for the development of IL-4-induced innate CD8+ T cells [14]. When mice were infected with CL–13, CIITATg mice were able to control viral titers below detection levels in selected tissues such as the spleen and serum within a month, whereas wild-type mice succumbed to persistent infection. Furthermore, the ability of CIITATg mice to control the virus was dependent on IL-4-induced innate CD8+ T cells as CIITATg and control mice did not show a difference in serum viral titers on the IL–4 deficient background, which causes a lack of intrathymic generation of IL-4-induced innate CD8+ T cells. Moreover, adoptive transfer of LCMV-specific IL-4-induced innate CD8+ T cells into wild-type hosts further confirmed the crucial role of these innate T cells as the primary effector mechanism for viral control. We also demonstrated that wild-type BALB/c mice, which have abundant IL-4-induced innate CD8+ T cells, exhibit notably enhanced anti-viral CTL responses compared with CD1dKO BALB/c mice, which only possess a very small fraction of these cells. As expected, expression level of Eomes and innate CD8+ T cell marker (CD44 and CXCR3) was higher in virus-specific CD8+ T cells wild-type mice, as compared to those of CD1dKO BALB/c mouse (Fig 8B), while we could not found any difference in LCMV NP118-specific CD8+ T cell numbers in these mice (Fig 8A). Considering the results from P14 cell adoptive transfer and CL–13 infection (Fig 6), these data favor the idea that IL-4-induced innate CD8+ T cells in a wild-type BALB/c host contributed to reduce CL–13 viral load compared with CD1dKO mice, although the genetic perturbation in CD1dKO mice is not specific for the innate CD8+ T cell population.
Eomes is a key transcription factor in the cytotoxic T-cell lineage [29]. During the activation and differentiation of mature CD8+ T cells Eomes induces effector function and cooperates with T-bet to sustain memory CD8+ T-cell homeostasis [30]. In particular, the central memory population is diminished in CD8+ T cells lacking Eomes [29,30]. Moreover, in the thymus Eomes also seems to confer effector function and memory phenotypes to innate CD8+ T cells as IL-4-induced innate CD8+ T cells acquire a CD44hiCD62Lhi central memory cell-like phenotype [14]. In addition to the previous reports, our comparison data of phenotype and function in between IL-4-induced innate CD8+ T cells and memory CD8+ T cells showed that the expression levels of Eomes and CXCR3 were similar but those of CD44, CD124, CD24, and NKG2D were different (S3A Fig). IFN-γ production and degranulation ability upon antigen stimulation were also different (S3B Fig). Although these IL-4-induced innate CD8+ T cells are less functional than memory CD8+ T cells, their function is evidently better than those of naïve CD8+ T cells. These data demonstrate that IL-4-induced innate CD8+ T cells are phenotypically and functionally different from conventional memory CD8+ T cells as well as naïve CD8+ T cells.
On the other hand, the expression of Eomes mRNA and protein are markedly elevated in exhausted CD8+ T cells during chronic virus infection compared to that in effector or memory CD8+ T cells [20,31]. These finding suggest that Eomes alone is not sufficient to stimulate the effector function of exhausted CD8+ T cells under the conditions of established chronic virus infection. This is despite the fact that upregulation of Eomes initially triggers the effector function of CD8+ T cells upon TCR stimulation and contributes to preserve the functionality of memory CD8+ T cells. In the present study, when CD8+ T cells already express a significantly high level of Eomes, these cells acquire obviously enhanced ability to produce effector cytokines such as IFN-γ and TNF-α upon viral antigen challenge. Taken together, these data suggest that a high level of Eomes expression allows IL-4-induced innate CD8+ T cells to exhibit their prompt and significant effector function, thereby controlling viremia during the early phase of virus infection.
Many researchers have attempted to control chronic virus infection using immunotherapeutic interventions such as blockade of the inhibitory receptors PD–1 and CTLA–4, administration of type I IFN, and regulation of microRNAs [32] [33,34]. Additionally, for chronic hepatitis B virus infection treatment, adoptive T-cell therapy using either in vitro expanded hepatitis B virus antigen-specific T cells or grafting T cells with recombinant TCR has been investigated as an approach [35]. Based on our data, virus-specific IL-4-induced innate CD8+ T cells have the potential to be used in adoptive T-cell therapy. Interestingly, when we adoptively transferred Eomes+ LCMV-specific innate CD8+ T cells into CL-13-infected mice, the CTL response was slightly increased even though the transferred cells were seemed still exhausted (S4 Fig). From this experiment, we hypothesize that combination therapy with adoptive transfer of IL-4-induced innate CD8+ T cells for prompt control of virus and PD–1 blockade for rejuvenating CTL function could be effective. Further experiments will be required to test this theory.
X-linked lymph proliferative disease (XLP) is a human immunodeficiency caused by germ-line mutations in SH2D1A gene and characterized by an inability to respond appropriately to infections such as Epstein-Barr virus [36]. The SH2D1A gene encodes the SAP molecule that is a component of the SLAM (signaling lymphocytic activation molecule) signaling pathway. Signaling through the SLAM family of receptors is crucial for the development of NKT cells, and thus the absence of the SAP causes an arrest in NKT cell development [37] [38,39]). In this context, the deficiency of NKT cells has been considered as one of the cellular bases of XLP [38]. However, considering the crucial role of SAP for T-T CD4+ T-cell development [40], the development and function of T-T CD4+ T cells would also be expected to be defective in XLP patients [27]. Both NKT and T-T CD4+ T cells are engaged in the generation of innate CD8+ T cells via IL–4 production and thus, the development of IL-4-induced innate CD8+ T cells is also dependent on the adaptor SAP [41]. In the present study, we demonstrated that IL-4-induced innate CD8+ T cells are able to rapidly proliferate, secrete cytokines, and decrease viral load after LCMV CL–13 infection. Taken together, it is necessary to consider the possibility that defective development of IL-4-induced innate CD8+ T cells causes the heighten susceptibility to Epstein-Barr virus infection in XLP patients.
Animals were maintained and procedures were performed with approval of the IACUCs of Seoul National University (permit number: SNUIBC-R100524-1) and Yonsei University (permit number: 2013–0115) in accordance to LABORATORY ANIMAL ACT of Korean Ministry of Food and Drug Safety for enhancing the ethics and reliability on animal testing through appropriate administration of laboratory animals and animal testing.
C57BL/6, IL-4KO, BALB/c and BALB/c-CD1dKO mice were purchased from the Jackson Laboratory. The CIITATg mice were previously generated at Seoul National University [26] and CIITATg mice were bred to IL-4KO and PIVKO mice in our laboratory to generate CIITATgIL-4KO and CIITATgPIVKO. LCMV epitope-specific TCR transgenic P14 Thy1.1+ Ly5.2+ mice were obtained from the Emory Vaccine Center, USA and P14 Thy1.1+ Ly5.1+ mice were obtained from POSTECH, Korea. All mice were maintained in the specific pathogen-free facility of the Yonsei Laboratory Animal Research Center at Yonsei University and the Center for Animal Resource Development at Seoul National University College of Medicine (Seoul, Korea). LCMV CL–13, a variant derived from an LCMV ARM CA1371 carrier mouse [42], was obtained from Rafi Ahmed (Emory Vaccine Center, Atlanta). Six- to ten-week old mice were infected with 1 x 105 to 2 x 106 PFU of LCMV CL–13 diluted in serum-free RPMI medium per 20 g of mouse body weight by intravenous infection or with 2 x 105 PFU of LCMV Armstrong diluted in serum-free RPMI medium per 20 g of mouse body weight by intraperitoneal infection. For serum virus titration, three to four drops of blood were individually collected by microcapillary tube at the indicated time points post infection, and the serum was directly stored at -70°C. For tissue titration, small pieces of the spleen and kidney were put in DMEM containing 1% FBS (HyClone) and stored at -70°C. The tissues were later homogenized completely using a homogenizer (Kinematica) before titration. Viral titers from sera or homogenized samples were determined by plaque assay on Vero cells as previously described [43]. Undetectable samples were given a half of each detection limit.
Peripheral blood mononuclear cells (PBMCs) were isolated from peripheral blood using density gradient centrifugation underlaid with Histopaque–1077 (Sigma-Aldrich). Lymphocytes from the thymus, spleen and lung were isolated as previously described [23]. For phenotypic analysis of lymphocytes, single-cell suspensions were stained with the following antibodies; fluorochrome-conjugated antibodies against CD4 (RM4-5), CD127 (A7R34), and PD–1 (RMP1-13) were from BioLegend, antibodies against CD8 (53–6.7), CD24 (30-F1), CD44 (IM7), CXCR3 (CXCR3-173), and CD19 (eBio1D3) were from eBioscience, and antibodies against CD44 (IM7), CD90.1 (Thy1.1; OX–1), CD122 (IL-2Rβ; TM-b1), CD124 (IL-4Rα; mIL4R-M1), CXCR5 (2G8), CD45R/B220 (RA3-6B2), CD95 (Fas; Jo2), T- and B-cell activation antigen (GL7), and NKG2D (CX5) were from BD Biosciences. H-2Db tetramers bound to GP33-41 or GP276-286 peptide and H-2Ld tetramer bound to NP118-126 peptide were generated and used as previously described [44]. To detect cytokine production by virus-specific CD8+ T cells, splenocytes from C57BL/6 mice were restimulated in vitro with 0.2 μg/ml of LCMV GP33-41, GP276-286, or peptide pool including GP33-41, GP276-286, GP70-77, GP92-101, NP166-175, NP205-212, NP235-249, and NP396-404 in the presence of Golgi plug/Golgi stop (BD Biosciences) and anti-CD107a (1D4B, BD Biosciences) Ab for 5 hours followed by intracellular cytokine staining using anti-IFN-γ (XMG1.2, BD Biosciences) and anti-TNF-α (MP6-XT22, BioLegend) antibodies. In case of in vitro restimulation of splenocytes from BALB/c mice, LCMV NP118-126, GP283-291, or NP313-322 peptide was used. For IL-4-induced innate CD8+ T-cell staining, lymphocytes were fixed and permeabilized with Fixation/Permeabilization solution (eBioscience) and stained with anti-Eomes (Dan11mag, eBioscience) Ab. The Live/Dead fixable Stain Kit (Invitrogen) was used to remove the dead cell population in most staining procedures. All stained samples were read by FACS CANTO II or LSR II (BD Biosciences), and analyzed by FlowJo software (Tree Star).
Lungs were fixed in 10% neutral buffered formalin (Sigma-Aldrich), and paraffin-embedded tissues were sectioned to a thickness of 4 μm and stained with hematoxylin and eosin. Microscopic observations were performed with an ECLIPSE 80i Bright-Field Microscope Set (Nikon) equipped with CFI 10×/22 eyepiece, Plan Fluor objectives (with 4×, 10×, 20×, and 100× objectives) and DS-Fi1 camera. We used NIS-Elements BR 3.1 software (Nikon) for image acquisition.
LCMV-specific antibodies were measured by ELISA. LCMV CL-13-infected BHK–21 cell lysate was used as capture antigen. Ninety-six-well Polysorp plates (Nunc) were coated with sonicated lysate for 3 days before performing the ELISA. Three fold serial dilutions of serum samples were incubated and detected with IgG-specific horseradish peroxidase (HRP)-conjugated goat anti-mouse immunoglobulin (Southern Biotech). TMB (Sigma-Aldrich) was used as a substrate, and the reaction was stopped by sulfuric acid and read at 450nm.
BM cells were isolated from femurs and tibias of donor mice and T cells were depleted through magnetic sorting using a mixture of CD4 and CD8 depleting microbeads (Miltenyi Biotech, Auburn, CA). Then, 3 x 106 T cell-depleted BM cells were intravenous injected into each recipient mouse preconditioned with 13,000 rad irradiation (two doses of 650 rad applied 4 h apart) from a 137Cs source 1 day prior.
For CD8+ T cell enrichment, CD8+ T cells were isolated from the spleen of wild-type BALB/c and CD1dKO BALB/c mice after negative selection using a CD8+ T-cell isolation kit (Miltenyi Biotech).
For adoptive transfer of P14 Thy1.1+ Ly5.2+ CD8+ T cells and P14 Thy1.1+ Ly5.1+ CD8+ T cells, the cells were isolated from the spleen of BMT chimera mice after negative selection using a CD8+ T-cell isolation kit and subsequent positive selection using Thy1.1+ cell microbeads (Miltenyi Biotech). After isolation, 5 x 103 purified P14 Thy1.1+ CD8+ T cells were transferred alone or with P14 Ly5.1+ CD8+ T cells into wild-type C57BL/6 mice via tail vein. Mice were infected with LCMV CL–13 at 1d after the adoptive transfer.
LCMV GP33-41-specific P14 CD8+ T cells were isolated from P14 transgenic mice using CD8+ isolation kit (Myltenyi Biotec). Purified P14 CD8+ T cells were labelled with CellTrace Violet (CTV) proliferation kit at concentration of 10uM (Invitrogen). Labelled P14 CD8+ T cells (1 x 106 cells) were adoptively transferred into naïve mice. Mice were infected with LCMV CL–13 at 1d after the adoptive transfer.
Thy1.2+ congenic C57BL/6 mice were infected with 2 x 106 PFU of LCMV CL–13. At 3 weeks post infection, P14 Thy1.1+ CD8+ T cells were isolated from the spleen of BMT chimera mice for adoptive transfer after negative selection using a CD8+ T-cell isolation kit and subsequent positive selection using Thy1.1+ cell microbeads (Miltenyi Biotech). After isolation, 2 x 106 purified P14 Thy1.1+ CD8+ T cells were transferred into CL-13-infected C57BL/6 mice via tail vein.
Survival curve was evaluated by log-rank (Mantel-Cox) test and other data were analyzed by two-tailed unpaired Student’s t-test using GraphPad Prism software. A p value less than 0.05 was considered statistically significant.
|
10.1371/journal.pcbi.1004351 | An Optimal Free Energy Dissipation Strategy of the MinCDE Oscillator in Regulating Symmetric Bacterial Cell Division | Sustained molecular oscillations are ubiquitous in biology. The obtained oscillatory patterns provide vital functions as timekeepers, pacemakers and spacemarkers. Models based on control theory have been introduced to explain how specific oscillatory behaviors stem from protein interaction feedbacks, whereas the energy dissipation through the oscillating processes and its role in the regulatory function remain unexplored. Here we developed a general framework to assess an oscillator’s regulation performance at different dissipation levels. Using the Escherichia coli MinCDE oscillator as a model system, we showed that a sufficient amount of energy dissipation is needed to switch on the oscillation, which is tightly coupled to the system’s regulatory performance. Once the dissipation level is beyond this threshold, unlike stationary regulators’ monotonic performance-to-cost relation, excess dissipation at certain steps in the oscillating process damages the oscillator’s regulatory performance. We further discovered that the chemical free energy from ATP hydrolysis has to be strategically assigned to the MinE-aided MinD release and the MinD immobilization steps for optimal performance, and a higher energy budget improves the robustness of the oscillator. These results unfold a novel mode by which living systems trade energy for regulatory function.
| This paper presents a unique dissipation mode of converting biochemical free energy in ATP to regulatory function through the MinCDE bio-oscillator that marks the mid-cell position for symmetric bacterial cell division. Through assessing the oscillator’s performance-to-cost relation, we demonstrate that some dissipation threshold needs to be satisfied to switch on the oscillation, but the oscillator’s performance can be damaged by excess free energy dissipation, which is distinct from the known monotonic tradeoff relation of stationary regulators. An optimal dissipation strategy has been unveiled: the ATP free energy must be precisely allocated to specific reaction steps for accurate mid-cell recognition, which also coincides with the dynamic requirements for robust oscillation to occur. These discoveries identify an optimizable operation scheme of free energy consumption in biological systems and provide deep insights into the evolution of dynamic regulatory networks.
| Similar to man-made systems that commonly employ sustained oscillations to measure time and length, living organisms use molecular oscillators to process spatiotemporal information for regulation. For example, the periodic pole-to-pole oscillation of Min proteins in Escherichia coli designates the mid-cell position for symmetric cell division [1, 2]; the oscillatory spindle dynamics in Caenorhabditis elegans [3, 4] and human cells [5] help position and orient the spindle at proper division site along the cell body; the genetic [6, 7] and non-genetic [8, 9] circadian rhythm networks repeatedly reset the intracellular environment every 24 hours; the RhoA and stress-fiber mediated oscillation synchronizes and coordinates the development of cells in Drosophila embryo [10, 11]; the traveling and standing waves set the differentiation markers in the zebrafish segmentation process [12, 13]. These different types of oscillators all emerge from various promotive and inhibitive interactions between the involved protein molecules, and their vital functions have been investigated extensively. However, the costs to sustain those functions have been overlooked almost completely. In particular, the free energy costs to drive the highly dissipative oscillating process, and “exchange rate” at which living organisms trade free energy for the above oscillatory regulation functions, remain largely unexplored.
We address these questions by investigating the E. coli MinCDE oscillatory network for regulation of the cells’ symmetric division. This Min oscillator comprises three proteins: the division inhibitor MinC, the ATPase MinD, and the catalytic enzyme MinE. Experiments have shown that these Min molecules interact with each other under the mediation of ATP and the phospholipid membrane [14, 15]: the ATP-bound MinD cooperatively associates with the cell membrane [16, 17], where MinE and MinC are recruited [18, 19]; MinE enhances MinD’s ATPase activity, which releases MinD from membrane after ATP is hydrolyzed [15, 20]; meanwhile MinC reduces the stability of FtsZ polymers that construct the scaffold of the division ring [21], which in turn inhibits cell division at MinC-rich locations. Based on these protein-protein interaction logics, a molecular reaction-diffusion mechanism emerges: cytoplasmic MinD molecules associate with ATP and aggregate on the cell membrane at one of the two poles; MinE molecules chase and bind to this MinD colony, catalyze the ATP hydrolysis, and eventually destroy the MinD aggregation; the released MinD molecules will then diffuse to the other pole of the cell to start the aggregation process again. [22–27]. Such reaction-diffusion process gives rise to the spatiotemporal oscillation of Min molecules between two cell poles (Fig 1A, 1B and 1C). An emerged oscillatory pattern (S1 Movie) will then result, on time average, in a “V”-shaped concentration profile of MinC along the cell’s long axis with the minimum at the mid-cell [28] (Fig 1D).
It is known that any sustained biochemical oscillation is dissipative and requires continuous free energy input [29, 30]. In line with this general concept, the energy-bearing ATP is an essential element of the Min oscillator [14, 15]. Different from the existing theoretical models that emphasize the topological and dynamic properties of the Min protein network [22–26, 31], we construct a general analysis framework to quantify the dissipative nature of the oscillator as well as the biophysical role of ATP. Our results explicitly indicate that sufficient free energy dissipation is required to switch on the oscillator, but counterintuitively, we found that free energy input does not always promote the differentiation of the mid-cell from the cell poles (defined as the “performance” [31]; Fig 1D, also see Methods). Through a global optimization analysis, we further discovered that the best performance can only be achieved when most of the ATP hydrolysis energy is dissipated in the steps of the MinE-aided MinD release and the MinD immobilization, suggesting an optimal free energy dissipation strategy for E. coli under the pressure of natural selection. These discoveries set the Min oscillator apart from stationary regulators, such as the sensory adaption systems [32, 33] and the kinetic proofreading system [34–36], whose performance is monotonically improved by higher free energy dissipation. Thus, our results suggest different free energy conversion modes for stationary and oscillatory bio-regulators.
To analyze the thermodynamic properties of the MinCDE oscillator, we constructed a detailed biochemical model (Fig 1A and 1B) based on the protein-protein interaction logics suggested in previous studies [25]. In this model, two regulatory motifs are involved for oscillation: positive auto-regulation (immobilized MinD cooperatively recruits more of itself) and negative feedback (the recruited MinE inhibits immobilized MinD via enhancing its ATPase activity). This category of interlinked positive and negative feedback loops are known to give rise to robust and tunable biological oscillations [37, 38]. Since MinC only contributes to ring-inhibition but not oscillation [2, 19], it is not explicitly included in this study of the Min oscillator.
A critical difference between our model and existing models is that we consider all reaction steps as “microscopically reversible” processes, which allows us to assess the free energy dissipation of individual steps through the forward and backward reaction fluxes [39]. These fluxes (j±i) can be expressed as:
j+1=k+1ρD:D,j−1=k−1ρD:T, j+2=k+2ρD:T,j−2=k−2ρd, j+2′=k+2′ρD:T(ρd+ρde),j−2′=k−2′ρd(ρd+ρde), j+3=k+3ρEρd,j−3=k−3ρde, j+4=k+4ρde,j−4=k−4ρD:DρE,
where ρD:D, ρD:T & ρE are the concentrations of cytoplasmic MinD:ADP, MinD:ATP complexes and MinE dimers, respectively; ρd & ρde are the concentrations of MinD:ATP and MinE:MinD:ATP complexes on the membrane, respectively. The reaction index i (= 1,2,3 & 4) represents the reactions labeled in Fig 1B accordingly. It is worth noting that j ± 2 ′ represent the cooperative recruitment of MinD:ATP to the cell membrane by existing membrane-associated MinDs (i.e. d & de).
For each reaction step, the net flux is ji = j+i−j−i (i = 1,2,3 & 4) (and j 2 ′ = j + 2 ′ − j − 2 ′ for the cooperative recruitment). In consideration of the diffusion of protein molecules in the cell volume, the dynamics of the system can therefore be described by a set of reaction-diffusion equations (see Methods for details), and if detailed balance is satisfied, the parameter
γ ≡ k + 1 k + 2 k + 3 k + 4 k - 1 k - 2 k - 3 k - 4
shall be unity; otherwise, the system is out of equilibrium and the chemical free energy from ATP hydrolysis is constantly dissipated.
The developed microscopically reversible reaction network allows us to assess the dissipation level of the oscillator. However, although the physics and methodology for evaluating free energy dissipation (or entropy production) of chemical reaction systems with stationary solutions has been nicely reviewed [29, 39], to our knowledge, dissipation of reaction-diffusion systems has yet to be broadly investigated [40]. A seemingly straightforward way to analyze such spatiotemporally varying systems is to compute and integrate the unbalanced fluxes of all reaction steps, as well as the diffusion processes (see Methods for details). Here, we obtain the total free energy dissipation rate of the reaction-diffusion system in a more elucidating way.
For any open biochemical reaction system in a steady or sustained oscillatory state, its average free energy dissipation rate is equal to the average consumption rate of the chemical free energy embedded in the environmental fuel molecules [41]. In our case of the Min oscillator, ATP is the energy source continuously supplied from the cytoplasm. The oscillator uptakes ATP molecules and excretes the hydrolysis products (ADP and inorganic phosphate, Pi) in the nucleotide exchange step (k±1) and the MinE-aided MinD release step (k±4, which is also the ATP hydrolysis step), respectively. Therefore, the chemical free energy supplied from ATP should directly be the free energy consumed by the oscillator, which is ln γ per ATP (in unit of kB T) [41]. It is worth noting that this free energy depends on the cytoplasmic concentrations of ATP, ADP and Pi, and is different from the standard free energy change of ATP hydrolysis [42]. Hence, the continuous free energy dissipation rate is:
σ ATP ( t ) = ∫ V d x j 1 ( x , t ) ln γ ≡ J 1 ( t ) ln γ , (1)
where the integral is taken over the cell volume, and the integrated flux J1(t) is the number of MinDs transiting from MinD:ADP to MinD:ATP per unit time. For a stable stationary system, J 1 = J 2 + J 2 ′ = J 3 = J 4 ≡ J is invariant over time; whereas for a sustained oscillatory system, the flux-balance relation holds only for time average over the oscillation period (P): ⟨ J 1 ⟩ ≡ P − 1 ∫ t 0 t 0 + P d t J 1 ( t ) = ⟨ J 2 + J 2 ′ ⟩ = ⟨ J 3 ⟩ = ⟨ J 4 ⟩ ≡ ⟨ J ⟩.
From an energetic point of view, σATP represents the chemical free energy “deposit” rate through uptaking ATP molecules from the cell’s cytoplasm. One should be aware that under sustained oscillating condition, the time averaged σATP over each period is equal to the time average of the instantaneous dissipation rate σtot (Fig 1C), i.e.
⟨ σ ATP ⟩ = ⟨ σ tot ⟩ = ⟨ J ⟩ ln γ ≡ ⟨ σ ⟩ . (2)
We use this average dissipation rate for sustained oscillation in our analysis. Also, to analyze the system’s behavior within the same framework, we use ⟨σ⟩ to denote the dissipation rate for non-oscillatory states whose average value is equal to the instantaneous value. It is straightforward from Eq (2) that the equilibrium situation (γ = 1) is non-dissipative.
Eq (2) provides a convenient way to compute free energy dissipation. We first investigate how the dissipation rate ⟨σ⟩ depends on the non-equilibrium parameter γ by tuning the backward nucleotide exchange step (MinD:ATP → MinD:ADP, k−1), while keeping other reaction rates fixed (see Methods).
The system exhibits two distinct regimes for the ⟨σ⟩ − γ dependence: for small γ, ⟨σ⟩ initially increases slowly with ln γ; when γ is further increased, ⟨σ⟩ increases dramatically with ln γ and then settles to a logarithmic regime where ⟨σ⟩ ∝ ln γ (Fig 2A; S1 and S2 Figs). The threshold between the two regimes is the same transition point for oscillation to occur (circle in Fig 2A), and the Min system exhibits only the regulatory function in the oscillatory regime (in the non-oscillatory stable regime, MinD is distributed uniformly on the membrane or slightly higher in the middle due to the cell-end effect as shown in S6 Fig). It is worth noting that the logarithmic dependence has been also observed in the stationary sensory adaptation systems when the net fluxes ⟨J⟩ reach their saturated levels [32, 33]. Our results suggest that this category of logarithmic relation between the non-equilibrium parameter and the associated free energy dissipation may be a universal qualitative indicator of a biochemical network providing its regulatory function, regardless of it being intrinsically stationary or oscillatory.
To investigate further how this logarithmic regime coincides with the oscillatory behavior of the system, we analyze the spatiotemporal patterns of Min protein molecules while the system becomes more dissipative. We find that the MinE dimers remain mostly in cytoplasm until the same dissipative “threshold” is reached; beyond this “threshold”, most MinEs are confined onto the membrane (Fig 2B). Due to the effective inhibitive role of MinE on MinD:ATP being associated with the membrane, oscillation can occur only at low cytoplasmic MinE molecule numbers. Therefore, when oscillation is switched on, the average flux is rate-limited by the MinE number: ⟨J⟩osc ≈ ⟨J⟩max = k+4 NE, and the free energy dissipation is then (Fig 2A, dashed line):
⟨ σ ⟩ osc ≈ k + 4 N E ln γ . (3)
Similar threshold and logarithmic dependence of free energy dissipation on γ can also be observed by constraining other reverse reaction rates from their equilibrium values (S3 and S4 Figs), and the dissipation rate can still be approximated by Eq (3) in the oscillatory/regulatory regime (see S1 Text for details).
The threshold to switch on the oscillator (Fig 2) indicates that oscillation occurs only when the dissipated free energy is large enough to sequestrate MinE from cytoplasm. Because the nucleotide exchange and MinE-aided MinD release steps are the two material exchange steps between the oscillator and the intracellular environment, we systematically investigate how the oscillator’s regulatory performance depends on these two interfaces.
Fig 3A and 3C show the performance analysis in the (k+1, k−1) and (k+4, k−4) spaces, respectively. The sharp bifurcation boundaries separate the oscillatory (colored) and stationary (white) regions: for the nucleotide exchange step, our result indicates that a fast ATP replacement rate and a large γ are both necessary for the oscillator to operate. In contrast, for the MinE-aided MinD release step, the system oscillates only when the releasing rate (k+4) remains within a moderate range while the rebinding rate (k−4) stays low. Despite the quantitative difference in the parameter scales, the regulatory performance is positive only in the oscillatory regions, and a minimum free energy dissipation needs to be satisfied for sustained Min oscillation for both material exchange interfaces.
Once the energetic threshold is met, the system reacts differently to the tuning of these two interfaces. When its dissipative behavior is tuned through the nucleotide exchange step (k±1), the system achieves its best performance in the region adjacent to the bifurcation boundary (red region in Fig 3A). A more explicit representation is given in Fig 3B: the solid and dashed lines correspond to the solid and dashed lines in Fig 3A, along which the regulatory performance and energetic costs are quantified. In both cases, the system’s performance first increases from zero to a finite peak value, and then decreases to a lower level (see panels A & B in S5 Fig for the membrane-associated MinD profiles). The negative performance before the bifurcation is due to the cell-end effect (S6 Fig).
On the other hand, when the system is tuned through the MinE-aided MinD releasing step (k±4), the high performance region is buried deeply in the oscillatory phase, and the performance exhibits simple monotonic dependence on free energy dissipation (Fig 3C and 3D and panels C & D in S5 Fig): at any given k+4, a higher free energy dissipation (i.e. smaller k−4) always leads to a better performance in mid-cell recognition. Interestingly, if we hold k−4 constant and vary k+4 (along a horizontal line in the performance map), the performance to dissipation relation would be non-monotonic with the best performance being achieved at moderate k+4 values.
These qualitatively distinct performance-to-cost relations indicate that not only different transition steps have different roles in the regulation process, but also the forward and backward reactions in the same step have different influences on the regulatory function. Moreover, the existence of an “optimal” operating region is distinct from other known stationary regulatory systems (e.g. sensory adaptation and kinetic proofreading systems) whose performance always improves monotonically with increasing free energy dissipation [32–36], suggesting that, in addition to direct free energy consumption, more complex dynamic requirements have to be satisfied for the Min oscillator to assume the role as an efficient mid-cell marker.
The results shown in Fig 3 suggest that the regulatory performance of the Min oscillator does not depend simply on how much free energy is dissipated, but indeed on how exactly the free energy is dissipated through the individual steps in the reaction-diffusion process. To investigate the E. coli cell’s dissipation strategy in mid-cell determination, we performed a systematic analysis of the Min oscillator’s performance by changing all the reverse reaction rates (k−i, i = 1 ∼ 4) while keeping γ fixed at the physiological level (i.e. ln γ is maintained at around 18 kB T to match the chemical free energy released from hydrolyzing each ATP molecule in E. coli[42]). This is equivalent to assigning different free energy consumption to different reaction steps while keeping the total free energy budget constant (given in Eq (3); see Fig 4A for illustration).
Fig 4B summarizes the results of our in silico experiments. In this figure, we use ΔGi = ln k−i/k+i to denote the standard free energy difference through the ith reaction step. It is worth special attention that the “standard” condition used here is slightly different from the conventional standard condition: since ATP, ADP and Pi are not explicitly included as reactants/products in our model but are implicitly embedded in the reaction rates k±i, we set our “standard” condition to be at unit concentration of Min molecules and at physiological levels of ATP, ADP and Pi. Since the total Δ G = ∑ i = 1 4 Δ G i = − ln γ per ATP is constant, only three out of four of the reactions are independent. In each elementary panel of Fig 4B, we present the performance contour plots in the (ΔG1, ΔG4) space at different ΔG3 values. The best performance point in each contour plot is identified and plotted in Fig 4C, where the red bars show the best performance scores at given ΔG3 levels, and the partitioned color bars indicate the calculated free energy dissipation rates for each of the individual reactions (magenta, green, yellow & blue for nucleotide exchange, MinD immobilization, MinE recruitment & MinE-aided MinD release, respectively) and the diffusion process (cyan) to achieve the best performance. The total heights of the partitioned color bars directly represent the total free energy dissipation rates, which are nearly constant (close to k+4 NE ln γ = 12,600 kB Ts−1 ∼ 700 ATPs per second).
These results provide a full spectrum picture of how energy is used to promote the Min oscillator’s performance. Firstly, oscillation occurs only within certain dissipation-strategy regions (colored in Fig 4B). A “bad” strategy can eliminate the oscillations, therefore having no regulatory function even with abundant energy source. Secondly, under the global optimal scenario (ΔG3 = −3 kB T in Fig 4C, where the performance score is globally the highest), the largest amount of free energy (4,440.9 kB Ts−1, 36.5% of ⟨σ⟩osc) is dissipated in the MinE-aided MinD release step, where ATP is hydrolyzed and Pi is released; a similar amount of free energy (4,332.5 kB Ts−1, 35.6% of ⟨σ⟩osc) is dissipated to ensure efficient MinD immobilization onto the cell membrane; a significant amount of free energy (2,681.4 kB Ts−1, 22% of ⟨σ⟩osc) is dissipated to recruit MinE to the membrane; and the least free energy (460.4 kB Ts−1, 3.8% of ⟨σ⟩osc) is used for nucleotide exchange (the rest 2.1% is dissipated in diffusion). From a structural point of view, it is known that binding of ATP helps the MinD molecule to rearrange its structure for a higher affinity to phospholipid and to MinE [43–45]. Therefore, the large amount of free energy dissipated in the membrane and the MinE involved steps might suggest that, under the pressure of natural selection, E. coli cells may have evolved to use the energy bearing ATP molecules fully to coordinate with the necessary structural changes for optimal operation. This theoretically-obtained optimal partition also clearly indicates that different reaction steps play different roles in converting free energy for the oscillatory functions.
We want to point out that the above dissipation partitions, as well as the color bars in Fig 4C, are calculated using the time averaged dissipation rates in individual reaction steps ⟨σi⟩:
⟨ σ i ⟩ = 1 P ∫ P d t σ i ( t ) = 1 P ∫ P d t ∫ x d x j i ( x , t ) ln j + i ( x , t ) j - i ( x , t ) .
Taking reaction 1 for example,
⟨ σ 1 ⟩ = - ⟨ J ⟩ Δ G 1 + 1 P ∫ P d t ∫ V d x j 1 ( x , t ) ln ρ D : D ( x , t ) ρ D : T ( x , t ) ,
which is different from ΔG1 = ln(k−1/k+1), which is the standard free energy difference between reactants and products under cytoplasmic ATP, ADP and Pi concentrations. ΔGi is useful for understanding the reaction landscape of the system as illustrated in Fig 4A, but ⟨σi⟩ represents the true dissipation at particular reaction steps (colored bars in Fig 4C). It is also worth noting that Eq (2) leads to ⟨σ⟩ = −⟨J⟩ΔG (i.e. the dissipated free energy is equal to the chemical free energy consumed through ATP hydrolysis).
Maintaining the above two dynamic conditions requires sufficient energy input to break symmetry in the proteins’ concentration distribution between the two poles. Fig 2A shows a minimum γ value obtained from tuning the nucleotide exchange step. We extended our analysis to a broader biochemical space to include all reverse reaction steps, and we further applied the global optimization method to different γ values (i.e. total energy budgets for dissipation). Our results show that, the Min oscillation can never be switched on if the total energy budget is too low, no matter how the system is optimized (Figs 5 and 6). This minimum value of ln γ is around 6 kB T. By comparing the operation phase plots at different γ values (Figs 4B and 5), we clearly show that the higher the total energy budget, the bigger the area in which oscillation could occur, implying higher robustness of the oscillator. Thus with an adequate energy budget, the organism has more freedom to arrange its internal environment while maintaining the vital oscillatory function.
Fig 6 summarizes the global optimal performance with increasing γ. By carefully tuning the Min pathway for optimization, a higher energy budget can eventually lead to a better global optimal performance until a saturation level is reached. This result indicates that the total energy budget is important for the highest achievable differentiation between the mid-cell and the pole regions, but the detailed dissipative strategy plays a more direct role in controlling the actual performance that the system can deliver.
Living organisms are all dissipative and the dissipated free energy is used to perform mechanical work, facilitate bio-synthesis/degradation processes, and process regulatory information. However, although it is intuitive to connect energy to mechanical work and to bio-mass turnover, a bigger challenge is the quantitative understanding of the free energy conversion mechanism through biological regulation processes. This difficulty is mainly because the free energy consumption in biological regulation is hardly accessible at the molecular level, and meanwhile the embedded information processing is difficult to quantify and evaluate. To date, only two energy-regulation conversion schemes have been quantitatively identified: the KP scheme and the ESA tradeoff scheme. The KP scheme was proposed by Hopfield for the Kinetic-Proofreading processes whose regulatory accuracy can be enhanced through every cycle of enzymatic checking [34–36]. In this scheme, the dissipated free energy effectively lowers the energy of an already stable state. On the other hand, it was recently discovered the ESA scheme in the sensory adaptation systems that higher free energy consumption, which is the product of the Energy dissipation rate and the inverse of the adaptation Speed, exponentially improves adaption Accuracy. In those processes, energy is used to stabilize an originally unstable state [32, 33]. Both schemes are stationary and exhibit monotonic performance-to-cost relation.
In this work, we present a third energy-regulation conversion scheme that is unique for oscillatory regulators. Our results show that, although energy is necessary to sustain biochemical oscillation, the regulatory performance does not monotonically depend on the total free energy dissipated over the full reaction cycle. Instead, the oscillator’s performance largely depends on how the energy is partitioned and dissipated through individual catalytic steps. In particular, for the MinCDE system, an optimal dissipation strategy is to allocate most of the available free energy to the MinE-aided MinD release and the MinD immobilization steps (i.e. large σ2 & σ4), whereas the nucleotide exchange step has to be kept less dissipative (i.e. small σ1). This non-monotonic optimizable feature is distinct from the KP and ESA schemes.
Our results provide energetic insights into the question of what drives the MinCDE oscillation. As predicted from the reaction-diffusion mechanism, the sustained MinCDE oscillation would require an uninterrupted pole-to-pole passage of MinD molecules together with a strong confinement of MinE molecules to the cell membrane [31]. To guarantee the uninterrupted MinD passage, a MinD concentration gradient between the two cell poles has to be established and maintained during the fast diffusion process (∼ 0.5 seconds) with the following perspectives: from a mass transfer point of view, such a gradient requires a “source” at the previously MinD-occupied pole and a “sink” at the far end of the cell; from an energetic standpoint, the “source” and “sink” are maintained by the highly “dissipative” MinE-aided MinD release and the MinD immobilization steps, respectively. In particular, a large σ4 keeps the rebinding of MinD at low levels, providing a net inward flux of MinD from the membrane; whereas a large σ2 guarantees near-perfect absorption of MinD onto the membrane at the far end.
Once a gradient is established, a remaining task is to avoid MinD binding to other parts of the cell membrane other than the far end cell pole. To this end, two additional requirements have to be satisfied: 1) the nucleotide exchange step should have a relatively high reverse rate with low dissipation level (i.e. small k+1/k−1 ratio and small σ1) so that once MinD associates with ATP (high membrane affinity), it can easily exchange back to the ADP state (low membrane affinity); 2) the nucleotide exchange rates (k±1) have to be fast so that diffusion cannot flatten out the gradient profile before MinD is ready for membrane binding. The first condition reduces the chance of MinD binding to the membrane during its first passage down the gradient; whereas the second condition reduces the chance of “backfire” in which MinD diffuses back. These two requirements are evident in Fig 3A: the performance is higher near the bifurcation boundary than deep inside the oscillatory phase region; along the γ contour line (in this case, also the k+1/k−1 contour line) passing the bifurcation boundary, the larger the rates, the better the performance will be.
Sustained MinCDE oscillation also requires MinE dimers to be confined at the old pole and meanwhile the cytoplasmic MinE has to be maintained at a low level to allow MinD accumulation at the new pole. To satisfy this condition, the cooperative recruitment step, where MinE are recruited, acquires a significant amount of free energy (σ3) to create a strong recruiting trap for MinE. Therefore, the MinCDE oscillator uses the free energy in ATP to establish alternating gradients across the long axis of the cell, so that the MinD can diffuse to and cooperatively build up a colony at the far end of the cell before the MinE can chase and destroy it. Furthermore, the limited intracellular energy budget has to be partitioned wisely to balance the requirements from individual reaction steps. And our predicted optimal free energy dissipation strategy captures these dynamic logics, suggesting that energetic and dynamic requirements are deeply connected.
It is worth noting that our study is based on a deterministic picture of the problem. In a previous study by Rex et al., the stochastic effects from finite numbers of molecules have been investigated numerically using microscopically irreversible reactions [46]. They found that, at physiological levels of molecular numbers (> 2,000 molecules), the oscillatory behavior of the stochastic MinCDE system is close to those derived from the deterministic equations, and the noise of the oscillatory period and the averaged MinD concentration profile are at a relatively low level. These results encourage us to believe that our discoveries of how free energy is traded for better identifying the mid-cell position would still be valid in the context of a stochastic cellular environment. But it would require more detailed investigation to identify whether free energy dissipation could also contribute to reducing the noise arising from the finite numbers of molecules.
All existing theoretical models, for simplification, consider biological reactions as microscopically irreversible processes [22–27]. Although these models have successfully captured many qualitative and quantitative features of the studied biological systems, we show in this paper that the predicted behaviors of the MinCDE system from irreversible models do not converge to the optimal operating mode obtained from reversible analysis. In particular, our analysis (Fig 3A) explicitly shows that the MinCDE system can benefit from its reversibility and the reversible system is capable of achieving a much higher performance compared to the irreversible counterpart (note the maximum performance near the bifurcation boundary). These results indicate that the backward reaction does not just reduce the “net” forward reaction rate, but essentially enlarges the “volume” for the system’s dynamic trajectories to occupy, which in turn leads to richer dynamic behaviors.
Furthermore, the obtained performance-to-cost relation via reversible analysis implies that “energy efficiency” as well as the “operational robustness” might be important factors for living organisms to survive the pressure of natural selection, shedding light on the evolutionary principles of regulatory networks. Therefore, we believe that it is worthwhile to reexamine other biochemical systems using the reversible modeling framework introduced here for more comprehensive thermodynamic understanding.
Given the net flux for each biochemical reaction step (ji = 1,2,3 & 4 and j 2 ′), the dynamics of the reaction-diffusion system can be described as:
∂ρD:D∂t=DD∇2ρD:D−j1,
(4) ∂ρD:T∂t=DD∇2ρD:T+j1, (5) ∂ρE∂t=DE∇2ρE, (6)
where DD & DE are the diffusion constants of MinD:ADP and MinE in cytoplasm, respectively. The system’s boundary conditions are defined by the membrane-associated reactions on the cell membrane:
∂ρd∂t=j2+j2′−j3, (7)
∂ρde∂t=j3−j4, (8)
D D n · ∇ ρ D : D = j 4 , (9)
D D n · ∇ ρ D : T = - j 2 - j 2 ′ , (10)
D E n · ∇ ρ E = - j 3 + j 4 , (11)
where n is the unit normal vector of the membrane pointing outward.
We want to point out that we adopt here the cooperative recruitment type of MinCDE oscillatory mechanism in which the diffusion on the cell membrane is considered unessential [25, 31, 47]. Similar analyses can be carried out for the other non-linear aggregation type of mechanism where 2-dimensional diffusion on the membrane plays an important role [24, 26], or for the recent comprehensive study where the two mechanisms and the direct binding of MinE to the membrane [18, 48, 49] are taken into account [27]. These would be beyond the scope of this paper, but we expect that qualitatively similar conclusions would be reached.
Using COMSOL Multiphysics 4.3, we simulate the E. coli cell as a cylinder with radius R = 0.5 μm and length L = 4 μm. The diffusion constants are set to be DD = 16 μm2 s−1 and DE = 10 μm2 s−1[50]. The total numbers of MinD particles and MinE dimers are fixed at ND = 2,000 and NE = 700, respectively [51]. Unless stated otherwise, the forward reaction rates are chosen from experimental measurements and previously theoretical studies: k+1 = 6 s−1, k+2 = 0.1 μms−1, k + 2 ′ = 0 . 01 μ m 3 s − 1, k+3 = 0.4 μm3 s−1 and k+4 = 1 s−1[17, 18, 25, 27, 31]. The backward rates, on the other hand, are tuned to alter the system from equilibrium to non-equilibrium and are specified in the corresponding sections of the paper. In addition, although many experiments have confirmed that the binding process of MinD to the membrane is cooperative, yet lipid specific [17, 18], the detailed mechanism is unclear. In this paper, we adopt a simple catalytic view that treats the spontaneous (j±2) and the cooperative attachment (j ± 2 ′) with the same free energy change. Therefore, k + 2 / k − 2 = k + 2 ′ / k − 2 ′ is satisfied in our minimal simulation setup. Using these parameters, the oscillation period is around 40s as long as k−4 stays small as shown in S7 Fig, which is in agreement with experimental observations [2, 52]. S1 Movie shows an example of the oscillatory dynamics.
Pole-to-pole oscillation is tightly coupled with the inequality of Min protein concentrations along the cell’s long axis. In our performance studies, we use an automated method to first detect the oscillatory dynamics: we collect the species’ concentrations over long time (for example, MinD concentration at one of the cell poles), and evaluate the variances of the time course data in steady state. The system is regarded to undergo oscillatory dynamics if the variance is non-zero.
Based on the time-averaged concentration profile along a cell’s long axis, we apply Halatek and Frey’s definition for the regulatory performance [31]: let h and w, respectively, be the normalized depth and width of the valley of the concentration profile of the membrane-bound MinD and MinD:MinE complex; then the performance is defined as h/w. (Fig 1D shows a particular example which has one extremum. A more detailed and general definition/illustration can be found in S8 Fig)
During oscillation, due to the canalized MinD transfer, MinD molecules periodically switch their occupancy at two cell poles. This results in a profile with lower MinD concentration at the mid-cell region. Therefore, the oscillation and the system’s positive performance are tightly coupled. A narrower and deeper valley at the mid-cell is considered superior for correct symmetric cell division, and is quantified by a higher performance score. We use such scores to demonstrate the relation between the regulatory performance and the associated energetic cost.
We use the imbalanced fluxes in the reaction-diffusion process to calculate the free energy dissipation rate.
Free energy dissipations in individual reaction steps. For a particular chemical reaction i, if the forward and backward flux densities at a spatial position x are j+i(x, t) and j−i(x, t) respectively, the dissipation rate density is [39]:
σi(x,t)=[j+i(x,t)−j−i(x,t)]lnj+i(x,t)j−i(x,t)=ji(x,t)lnj+i(x,t)j−i(x,t). (12)
For notational simplicity, the spontaneous and cooperative MinD attachments are denoted as one reaction here, i.e., σ 2 = ( j 2 + j 2 ′ ) ln ( j + 2 / j − 2 ). The total dissipation rate of all the reactions is therefore:
σ tot react ( t ) = ∫ V d x σ 1 ( x , t ) + ∫ S d x ∑ i = 2 4 σ i ( x , t ) ,
where the subscripts V and S denote the cell volume and cell surface integrals, respectively.
Free energy dissipations in diffusion. The dissipation of the diffusion process is less obvious. We model the diffusion process as a number of particles performing random walks on a uniform 3-dimensional lattice space with grid size dxdydz. Each molecule on one node (x ≡ (x, y, z)) can jump to its six neighboring nodes with rate a. Taking two neighbors along x direction for example, the forward and backward fluxes are
j + x ( x ) = j ( x → x + d x , y , z ) = a ρ ( x , y , z ) d x
j - x ( x ) = j ( x + d x → x , y , z ) = a ρ ( x + d x , y , z ) d x
This leads to the well-known equality for the net flux:
j x ( x ) = lim d x → 0 a [ ρ ( x , y , z ) - ρ ( x + d x , y , z ) ] d x = - D ∂ ρ ( x ) ∂ x
and the diffusion constant is defined as D = limdx → 0 a(dx)2. The dissipation rate between these two neighbors is:
lim d x → 0 d y d z j x ( x ) ln j + x ( x ) j - x ( x ) = d x d y d z j x 2 ( x ) D ρ ( x ) , (13)
where jx(x) is the x component of the net diffusion flux j(x) at location x. Hence the dissipation rate density of diffusion is
σ diff ( x , t ) = j x 2 + j y 2 + j z 2 D ρ ( x , t ) = | j ( x , t ) | 2 D ρ ( x , t ) . (14)
This final result in terms of the flux vector is independent of coordinates. Because we disregard the diffusion on cell membrane, the total dissipation rate of the diffusion processes of the three diffusive cytoplasmic molecules can be written as:
σ tot diff ( t ) = ∫ V d x [ σ D : T diff ( x , t ) + σ D : D diff ( x , t ) + σ E diff ( x , t ) ] , (15)
where the integral is taken over the cell volume. We find that the dissipation of the diffusion process is small compared to the reactions (see Fig 4C).
Combining reaction and diffusion, the total dissipation rate of the system can be calculated by summing these two parts:
σ tot ( t ) = σ tot react ( t ) + σ tot diff ( t ) . (16)
Please note that the instantaneous dissipation rates calculated from Eqs (1) and (16) are different for oscillatory systems. However, the total dissipated free energy over each period is the same from these two analysis methods (as shown in Eq (2) and Fig 1C).
We use COMSOL Multiphysics 4.3 to solve the partial differential equations. The cell is set up as a cylindrical volume with axial symmetry, which leaves us with a mathematical 2-dimensional model. The maximum grid size is set to be 0.1 μm. The entire system is then simulated using two physics modules provided by COMSOL: the “Transport of Diluted Species” module is used to simulate the reaction-diffusion process in cytoplasm, and the “Boundary ODEs and DAEs” module is used to account for the reactions on the cell membrane. These two modules are coupled: the membrane reactions serve as the boundary condition for the cytoplasmic reaction-diffusion process. The deterministic equations have an unstable solution with a uniform distribution of all species, so to break this unstable symmetry solution we used a step function as our initial condition for the MinD:ADP concentration in cytoplasm, while making MinD:ADP and MinE homogeneously distributed in the cytoplasm. All the simulations are run for a time long enough to cover at least 10 full periods of sustained oscillation with stable amplitude, and data are collected for these stable oscillations.
|
10.1371/journal.pgen.1005571 | The Ty1 Retrotransposon Restriction Factor p22 Targets Gag | A novel form of copy number control (CNC) helps maintain a low number of Ty1 retrovirus-like transposons in the Saccharomyces genome. Ty1 produces an alternative transcript that encodes p22, a trans-dominant negative inhibitor of Ty1 retrotransposition whose sequence is identical to the C-terminal half of Gag. The level of p22 increases with copy number and inhibits normal Ty1 virus-like particle (VLP) assembly and maturation through interactions with full length Gag. A forward genetic screen for CNC-resistant (CNCR) mutations in Ty1 identified missense mutations in GAG that restore retrotransposition in the presence of p22. Some of these mutations map within a predicted UBN2 domain found throughout the Ty1/copia family of long terminal repeat retrotransposons, and others cluster within a central region of Gag that is referred to as the CNCR domain. We generated multiple alignments of yeast Ty1-like Gag proteins and found that some Gag proteins, including those of the related Ty2 elements, contain non-Ty1 residues at multiple CNCR sites. Interestingly, the Ty2-917 element is resistant to p22 and does not undergo a Ty1-like form of CNC. Substitutions conferring CNCR map within predicted helices in Ty1 Gag that overlap with conserved sequence in Ty1/copia, suggesting that p22 disturbs a central function of the capsid during VLP assembly. When hydrophobic residues within predicted helices in Gag are mutated, Gag level remains unaffected in most cases yet VLP assembly and maturation is abnormal. Gag CNCR mutations do not alter binding to p22 as determined by co-immunoprecipitation analyses, but instead, exclude p22 from Ty1 VLPs. These findings suggest that the CNCR alleles enhance retrotransposition in the presence of p22 by allowing productive Gag-Gag interactions during VLP assembly. Our work also expands the strategies used by retroviruses for developing resistance to Gag-like restriction factors to now include retrotransposons.
| The presence of transposable elements in the eukaryotic genome threatens genomic stability and normal gene function, thus various defense mechanisms exist to silence element expression and target integration to benign locations in the genome. Even though the budding yeast Saccharomyces lacks many of the defense systems present in other eukaryotes, including RNAi, DNA methylation, and APOBEC3 proteins, they maintain low numbers of mobile elements in their genome. In the case of the Saccharomyces retrotransposon Ty1, a system called copy number control (CNC) helps determine the number of elements in the genome. Recently, we demonstrated that the mechanism of CNC relies on a trans-acting protein inhibitor of Ty1 expressed from the element itself. This protein inhibitor, called p22, impacts the replication of Ty1 as its copy number increases. To identify a molecular target of p22, mutagenized Ty1 was subjected to a forward genetic screen for CNC-resistance. Mutations in specific domains of Gag, including the UBN2 Gag motif and a novel region we have named the CNCR domain, confer CNCR by preventing the incorporation of p22 into assembling virus-like particles (VLPs), which restores maturation and completion of the Ty1 life cycle. The mechanism of Ty1 inhibition by p22 is conceptually similar to Gag-like restriction factors in mammals since they inhibit normal particle function. In particular, resistance to p22 and the enJS56A1 restriction factor of sheep involves exclusion of the restriction factor during particle assembly, although Ty1 CNCR achieves this in a way that is distinct from the Jaagsiekte retrovirus escape mutants. Our work introduces an intriguing variation on resistance mechanisms to retroviral restriction factors.
| The Ty1 and Ty2 retrotransposons of Saccharomyces belong to the Ty1/copia group of long terminal repeat (LTR) retrotransposons which replicate in a manner analogous to retroviruses [1]. Ty1 is the most abundant of five retrotransposon families (Ty1-Ty5) in the S288C reference genome of Saccharomyces cerevisiae, followed by the related Ty2 element [2, 3]. Recently, Ty2 has been shown to outnumber Ty1 in some Saccharomyces genomes [2–4], but Ty1 remains the more widely studied retrotransposon [1]. Ty1 contains two overlapping ORFs, GAG and POL, and many elements are transpositionally competent and transcriptionally active [5]. An abundant full-length Ty1 mRNA is transcribed which serves as a template for translation and reverse transcription. Two translation products are produced: Gag (p49) and Gag-Pol (p199), of which the latter comprises only 5–10% of total translation products due to its production requiring a +1 ribosomal frameshifting event. Gag, Gag-Pol and Ty1 mRNA accumulate in the cytoplasm in distinct foci called retrosomes [6–9]. Virus-like particles (VLPs) assemble from Gag and Gag-Pol proteins within retrosomes and encapsidate Ty1 mRNA, and tRNAiMet, which is used to prime reverse transcription. VLP maturation occurs via the activity of the POL-encoded enzyme, protease (PR). Pol is cleaved from p199 via a PR-dependent autocatalytic event, followed by PR cleavage of Gag at its C-terminus (from p49 to p45) and Pol at two internal sites to form mature PR, integrase (IN), and reverse transcriptase (RT). Once maturation occurs, reverse transcription of the packaged genomic Ty1 RNA forms a cDNA copy that is integrated into the host genome. Because Ty1 insertions can mutate cellular genes and mediate chromosome instability by homologous recombination with elements dispersed in the genome, it is beneficial to the host to control the process of retrotransposition [10–13].
Natural isolates of S. cerevisiae and its closest relative S. paradoxus maintain lower copy numbers of the Ty1 retrotransposon in their genomes compared to the reference laboratory strain S288C [2–4, 14], without the support of eukaryotic defense mechanisms such as RNAi or the presence of innate restriction factors like the APOBEC3 proteins [15–19]. The maintenance of Ty1 copy number is due at least in part to a mechanism called copy number control (CNC), which was first observed in an isolate of S. paradoxus that lacks complete Ty1 elements but contains numerous solo-LTRs [20]. The Ty1-less strain supports higher levels of Ty1 transposition compared to standard lab strains, as monitored using a Ty1 tagged with the his3-AI retrotransposition indicator gene [21]. Additionally, Ty1his3-AI mobility decreases dramatically when the naive genome is repopulated with Ty1 elements [20]. Introduction of a transcriptionally repressed Ty1 element on a multi-copy plasmid also inhibits Ty1his3-AI mobility. Based on these observations, CNC is conferred by a factor produced directly by the Ty1 element. The CNC phenotype, which includes decreased levels of transposition [20], the reduction of mature Ty1 RT and PR protein levels, and the absence of detectable mature IN [22], is dependent on the GAG open reading frame [20]. Overexpression of Ty1 fused to a GAL1 promoter on a multi-copy plasmid has been shown to override CNC, suggesting that CNC can be saturated [23–26].
Recently, we found that CNC functions through the protein product encoded by a subgenomic internally-initiated Ty1 sense transcript, called Ty1i (internal) RNA [26]. Transcription of Ty1i RNA initiates within GAG, about 800 nucleotides downstream of the full-length, transposition-competent Ty1 mRNA. The first AUGs are in the same reading frame as Ty1 Gag, resulting in synthesis of a 22 kD protein (p22) that shares the same sequence as the C-terminal half of Gag [26, 27]. This shared sequence includes the PR cleavage site, which is utilized within the inhibitory protein to form p18 [26]. Ectopic co-expression of p22 or p18 with Ty1 dramatically inhibits Ty1 mobility. p22/p18 co-immunoprecipitates with Ty1 Gag and co-localizes with Ty1 Gag in the cytoplasm. Ectopic expression of p22/p18 disrupts normal retrosome formation and VLP assembly, followed by a block in maturation and reverse transcription within the VLPs that are able to form. In addition, p18 interferes with the nucleic acid chaperone (NAC) function of Gag, further disrupting Ty1 replication [27, 28]. It is not clear which insult by p22/p18 is most destructive, but collectively these effects result in the strong inhibition of retrotransposition observed during CNC.
Retroelement restriction mechanisms have been aided by studying resistance mutations in retroviruses and/or sequence variation determining viral tropism. An example of particular relevance is the discovery that capsid (CA), a cleavage product of retroviral Gag, is the target of several restriction factors including Friend virus susceptibility factor–1 (Fv1), tripartite motif 5 alpha (TRIM5α), and myxovirus resistance protein 2 (Mx2), among others [29–33]. Viruses that escape restriction by these factors typically carry mutations in the CA-encoding region of the gag gene [29, 34–38]. In the case of Fv1 and TRIM5α, viral escape mutations disrupt binding between the restriction factor and CA by altering CA surface residues [37–40], while Mx2 escape mutations in CA are not fully understood but likely alter the interactions between CA and host factors [32, 33, 41]. While Fv1, TRIM5α, and Mx2 bind the incoming viral capsid during the early stages of retroviral infection, the sheep restriction factor enJS56A1 is known to interact with Jaagsiekte sheep retrovirus (JSRV) Gag at later stages when the integrated provirus is undergoing translation and particle assembly [42]. Resistance to enJS56A1 is conferred by mutations in the signal peptide of the JSRV envelope glycoprotein, which is hypothesized to ultimately alter the ratio of JSRV to enJS56A1 Gag levels to favor JSRV particle production [43]. Remarkably, Fv1 and enJS56A1 are both derived from endogenous retroelement gag genes [42, 44], similar to p22.
To further understand the mechanism of CNC, we carried out forward genetic screens for CNC-resistant (CNCR) Ty1 elements. Almost all of the CNCR elements contain missense mutations within GAG that map within predicted helices important for VLP assembly and maturation. Computational and functional analyses reveal three domains within the Ty1 Gag protein: TYA, CNCR and UBN2. All resistance mutations recovered map within the CNCR and UBN2 domains encoded by GAG. Importantly, several mutations are not present in p22 coding sequence, supporting the idea that p22 targets Gag to inhibit retrotransposition. Most CNCR mutations in GAG do not markedly alter Ty1 fitness or the interaction between Gag and p22, but prevent co-assembly of Gag and p22 into VLPs, which improves VLP maturation and progression through the retrotransposon life cycle.
We hypothesized that the generation of CNCR Ty1 mutants may identify a molecular target of p22. Since previous work implicated a physical interaction between Gag and p22 [26], isolating resistance mutations in GAG would suggest that this interaction is important for CNC. To identify Ty1 mutants that are resistant to the effects of p22 and its processed form p18, we designed a system that allowed for simultaneous expression of wild type p22/p18 and a randomly mutagenized Ty1his3-AI element fused to the regulated GAL1 promoter carried on a low copy centromere-based (CEN) plasmid (pGTy1his3-AI/CEN). The purpose of using a low copy plasmid for pGTy1his3-AI expression was to minimize CNC saturation that occurs with overexpression of Ty1 on a multi-copy plasmid. In addition, the Ty1 copy number provided by a low copy centromere-based plasmid does not result in detectable CNC [22]. Isogenic, repopulated Ty1-less S. paradoxus strains containing 1–38 copies of Ty1-H3 were analyzed, representing a wide range of Ty1 copy numbers naturally found in yeast (S1 Table) [2, 3, 14]. All strains carried a deletion of SPT3, which encodes a transcription factor required for expression of full length Ty1 mRNA from nucleotide 238 (Ty1-H3, Genbank M18706.1) and the synthesis of Ty1 Gag and Gag-Pol [45]. Ty1i RNA, which initiates internally at nucleotide ~1000, is still produced in spt3Δ mutants [26, 45, 46]. pGTy1his3-AI/CEN provided Ty1 mRNA, Gag and Gag-Pol and the strains were analyzed for CNC (Fig 1). Ty1 mRNA produced from this plasmid contains the his3-AI indicator gene, allowing transposition levels to be analyzed by growth on media lacking histidine. As expected, increasing Ty1 copy number resulted in decreased Ty1 mobility, with the strongest decrease observed in the presence of 38 genomic copies of Ty1 (Fig 1A). These strains were immunoblotted for p22/p18 levels in the presence and absence of pGTy1his3-AI expression using p18 antiserum, which detects both Gag-p49/p45 and p22/p18 [26]. Under both repressing (glucose) and inducing (galactose) growth conditions, p22 levels in cell extracts increased similarly with copy number (Fig 1B). Because p22 was not detected in the lowest Ty1 copy strain (1 Ty1) containing pGTy1his3-AI, these results confirmed that pGTy1his3-AI does not produce detectable p22. It remains possible that increasing chromosomal copies of Ty1 stimulated p22 production from pGTy1his3-AI, but this seems unlikely considering that p22 levels do not increase in the 38 Ty1 copy strain containing pGTy1his3-AI compared to an empty vector in either growth condition (Fig 1B). Therefore, genomic Ty1-H3 elements, and not the pGTy1his3-AI mutant library, were the source of p22 in the screen. When pGTy1his3-AI expression was induced by galactose, Gag-p49/p45 was detected and the maturation of p22 to p18 was observed, supporting previous findings suggestive of p22 cleavage by Ty1 PR [26]. In addition, mature RT (p63) and IN (p71) were present only in low copy number strains (Fig 1C), confirming another feature of the CNC phenotype [22]. To further establish that cleavage of p22 was Ty1 PR-dependent, wild type or PR-defective pGTy1/2μ multi-copy plasmids were introduced into the 38 Ty1 copy strain (Fig 1D). As expected, neither Gag-p49 expressed from the PR- Ty1 nor p22 expressed from genomic elements were cleaved to form mature products Gag-p45 and p18, respectively (Fig 1D).
To search for pGTy1his3-AI CNCR mutations, we utilized the 38 Ty1 copy strain described above, which produced the highest level of p22 of the strains tested (Fig 1B). pGTy1his3-AI was mutagenized by propagation in a mutator strain of E. coli (XL–1 Red, Agilent Technologies, Santa Clara, CA) and 20,000 transformants were screened for an increase in Ty1HIS3 mobility following induction on medium containing galactose (see Materials and Methods). The CNC region (nucleotides 238–1702 [48]) was sequenced from pGTy1his3-AI plasmids that conferred an increase in Ty1 mobility when compared to wild type plasmid controls. Nine unique mutations were present in GAG (S2 Table; XL–1 Red). A restriction fragment encompassing the CNC region was subcloned from the mutant plasmids into a fresh pGTy1his3-AI vector to eliminate contribution of background mutations and no loss of CNCR was observed. To avoid bias based on our mutagenesis method and to generate additional CNCR mutations, GAG and POL were mutagenized separately by error-prone PCR, followed by gap-repair with pGTy1his3-AI in the 38 Ty1 copy strain. While GAG mutagenesis by PCR revealed 8 new CNCR mutations from 500 colonies, POL mutagenesis revealed only 1 CNCR candidate from 6,000 colonies (S2 Table). While most of the mutations (8 of 9) isolated via XL–1 Red mutagenesis were single base changes, 5 of 8 mutations recovered via PCR mutagenesis had more than one base change. Interestingly, all GAG mutations recovered with either method were missense, suggesting that they function at the level of the Gag protein. The only CNCR pGTy1his3-AI plasmid isolated from POL mutagenesis contained two missense mutations within RT (D518G/V519A).
To quantify the level of resistance to p22/p18, the frequency of Ty1his3-AI mobility was determined for the mutants alongside wild type controls (Fig 2, S4 Table). In the 38 Ty1 copy strain, most mutant plasmids produced mobility frequencies between 11- and 63- fold higher than wild type (Fig 2A). Four candidates from the CNCR screen (Gag N183D, K186Q, I201T, and A273V) exhibited stronger resistance, ranging from 233- to 424-fold higher than wild type (Fig 2A). Because it was possible to obtain Ty1 mutations that globally increased transposition, rather than acting specifically in the presence of p22/p18, CNCR mobility was also measured in the 1 Ty1 copy strain (Fig 2B). Importantly, all CNCR mutants transposed at similar or decreased levels compared to wild type in the absence of p22/p18, indicating that we obtained CNC-dependent mutations. Three CNCR mutants, Gag P173L, Gag K250E, and RT D518G/V519A, showed a decrease in Ty1his3-AI mobility in the absence of p22/p18, indicating that these mutations negatively impacted Ty1 fitness (Fig 2B). Percent recovery of Ty1his3-AI mobility with CNCR plasmids was calculated by dividing Ty1his3-AI mobility in the presence or absence of p22 (Percent CNC Recovery, Fig 2C). As expected, the three CNCR mutations with decreased fitness showed higher percent recovery, with K250E and RT D518G/V519A at >100%, due to the fact that these plasmids result in higher transposition frequencies in the presence of p22 than in its absence. Since overall Ty1his3-AI mobility was extremely low, further studies were not performed with these mutants. The remaining four mutations conferring >10% CNCR include those resulting in Gag amino acid changes N183D, K186Q, I201T, and A273V. These elements exhibited 20–30% recovery, indicating that while the strongest CNCR mutations dramatically increase transposition in the presence of p22 (Fig 2A), they are only partially resistant to the effects of p22. Note that both classes of recovery should be expected since the mutant screen balanced transposition fitness and CNCR. Consequently, we focused on the N183D, K186Q, I201T, and A273V GAG mutations, since they conferred the strongest levels of resistance recovered with no effect on fitness of Ty1.
In an effort to increase CNCR, cells containing double mutant pGTy1his3-AI-K186Q/I201T or pGTy1his3-AI-A273V/I201T were tested for Ty1his3-AI mobility. Gag K186Q/I201T was defective for transposition and was not studied further. Gag A273V/I201T was able to transpose, but experienced decreased fitness. In the 38 Ty1 copy strain, the levels of Ty1his3-AI mobility with A273V/I201T was lower than either single mutant, but still 85-fold higher than wild type (Fig 2A). In the 1 Ty1 copy strain, A273V/I201T exhibited Ty1his3-AI mobility at 8% of wild type levels (Fig 2B). Interestingly, the double mutant did exhibit increased CNC recovery (45%), but at the expense of overall Ty1 mobility (Fig 2C). The loss of fitness in the double mutants reinforces the idea that mutations conferring a high level of CNCR may have been missed in the screen since they compromise Ty1 fitness. In addition, Ty1 Gag may be genetically fragile since it cannot tolerate wholesale alterations in sequence, a feature that is also observed with HIV–1 CA [49].
To determine if the CNCR elements are resistant to p22/p18 in a genomic context, wild type or CNCR pGTy1his3-AI/CEN plasmids were expressed in Ty1-less S. paradoxus, and cells with 1–2 wild type or CNCR Ty1his3-AI genomic insertions were identified. Next, an empty vector or a p22-producing plasmid, pTy1-ATGfs (S3 Table), was introduced into these strains. In the absence of p22 (Table 1, empty vector), genomic CNCR Ty1 elements N183D, K186Q, and I201T transposed at similar levels to the respective wild type control (<2-fold change), confirming that they do not globally increase Ty1his3-AI mobility in a chromosomal context. In contrast, CNCR Ty1 A273V displayed a 4.3-fold increase in Ty1 mobility compared to wild type in the presence of empty vector, indicating that A273V may not be CNC-dependent in all contexts. This may be due to the fact that A273V is the only mutation tested that maps within GAG and p22 coding sequence, thus changes in both proteins could be affecting Ty1 mobility. Dramatic differences in CNC were observed when the wild type and CNCR Ty1his3-AI elements are challenged with p22 (Table 1, pTy1-ATGfs). While p22 expression decreases wild type Ty1his3-AI mobility 56- to 120- fold, CNCR Ty1his3-AI mobility is partially resistant to the effects of p22, decreasing 2- to 13-fold.
A key feature of CNC is a decrease in Ty1 mobility of a single genomic Ty1his3-AI in the presence of elevated Ty1 copies [48]. In contrast, additional chromosomal copies of CNC-defective Ty1 elements, which are elements that do not produce p22/p18 but retain the ability to undergo retrotransposition, increase the level of Ty1his3-AI mobility [26]. To determine how chromosomal CNCR elements influence Ty1his3-AI mobility (Table 2), S. paradoxus containing a wild type Ty1his3-AI genomic insertion was repopulated with unmarked CNCR elements carrying the N183D, K186Q, I201T, and A273V mutations. It is important to note that Ty1his3-AI RNA is not preferentially packaged in cis [23] and can serve as the genomic RNA in mixed particles containing wild type and CNCR Gag, with the latter likely being produced in excess due to increased genomic copy number. Compared with the starting strain, Ty1his3-AI mobility in strains repopulated with +14 and +21 wild type Ty1 elements decreased 31- and 620-fold, respectively. Repopulation with +14–20 CNCR elements did not alter Ty1his3-AI mobility, supporting the idea that CNCR mutations relieve the inhibitory effects of p22 produced by the additional chromosomal elements. However, the fact that additional CNCR elements do not stimulate Ty1his3-AI mobility probably reflects the partial resistance phenotype imparted by the CNCR mutations.
Since little is known about the structure of Ty1 or other LTR-retrotransposon Gag proteins, we submitted Ty1 Gag protein sequence for secondary structure prediction using I-TASSER [50–52], and several other structural prediction servers (see Materials and Methods). These analyses predicted that a central portion of Ty1 Gag contains nine helical regions (Fig 3A, gray boxes), which overlap previously identified conserved Gag domains A, B and C of Ty1/copia family of retrotransposons [53]. Using profile-based methods, we identified two annotated protein families (Pfam) within Ty1 Gag called TYA (TYA transposon protein, PF01021) and UBN2 (gag-polypeptide of LTR copia-type, PF14223) (Fig 3A). The TYA domain is found strictly in yeast and corresponds to an unstructured region in the N-terminal half of Gag between residues 17–114. The UBN2 domain maps to the C-terminal half of Gag between residues 245–356, roughly overlapping Ty1/copia conserved Gag domains B and C [53], and is represented across multiple plant and fungal species. Of the 11 single GAG mutations that impart resistance to p22, 9 mapped within the helical domains, with 4 mapping within the UBN2 domain. Mutations outside of the UBN2 domain clustered between amino acids 170–220, which we refer to here as the CNCR domain (Fig 3A). The CNCR domain contains sequences belonging to Ty1/copia Gag conserved domain A [53], which is characterized by an invariant tryptophan residue corresponding to Ty1 Gag W184. Interestingly, CNCR alleles lie in close proximity to the W184 codon.
Ty1-H3 Gag (Uniprot P08405) sequence was used in a profile hidden Markov model search to identify closely related Gag proteins and an alignment was generated to highlight variations in amino acid sequence in CNCR (Fig 3B) and UBN2 (Fig 3C) domains (refer to S1 Fig for a full alignment). Redundant and partial Gag sequences were purged and the Gag sequence of the Ty2-917 element (GenBank KT203716), which was isolated as a spontaneous HIS4 mutation [58], was added to the hits. In total, we generated a multiple alignment of 15 sequences representing Ty1 and Ty1-like Gag proteins from 9 different yeast strains in the Saccharomycetaceae family [60, 61]. While substitutions of CNCR residues do not naturally occur in known Ty1 Gag sequences from S. cerevisiae, substitutions in all but one of the CNCR residues (E287) are found in the alignment of non-Ty1 Gag proteins, including those from Ty2 elements, the second most abundant Ty1/copia retrotransposon found in the S. cerevisiae S288C genome [2, 3], and Ty1-like elements present in Saccharomyces kudriavzevii and Lachancea kluyveri (Fig 3B and 3C). Most substitutions are different than those recovered in our CNCR screen, with the exception of D180N and T218A (Fig 3B). Considering all 10 CNCR residues altered in our screen, 6 of these are not conserved from Ty1 to Ty2.
To determine if Ty2 undergoes CNC, we analyzed the retrotransposition-competent Ty2-917 element [62]. Unlike pGTy1-H3, which confers CNC when GAL1-promoted Ty1 mRNA transcription is repressed, a multi-copy pGTy2-917 plasmid does not inhibit Ty1his3-AI mobility [48]. This result suggests that a p22-like protein is either not produced by Ty2-917 or does not affect Ty1 movement. A transcriptionally silent pGTy2-917 also did not affect Ty2-917his3-AI mobility, demonstrating that Ty2-917 is not under a Ty1-like form of CNC (S5 Table). In fact, a transcriptionally active Ty2-917 carried on a multi-copy plasmid stimulated Ty2-917his3-AI mobility 1.5-fold. Whether all Ty2 elements respond the same way as Ty2-917 will require further investigation. Considering the relationship between Ty1 CNCR mutations and Ty2 residues in the CNCR domain (Fig 3B), we introduced an empty vector or the p22 expression plasmid into a strain containing Ty2-917his3-AI to determine if Ty2 mobility was sensitive to inhibition by p22 (S5 Table) [26]. A decrease in Ty2-917 mobility of less than 2-fold was observed in cells expressing Ty1-p22, supporting the idea that Ty2-917 is not sensitive to Ty1 CNC.
Since CNCR residues map to putative helical domains in Gag, a series of mutations were made in hydrophobic residues within several predicted helices (Fig 3A) and their impact on Ty1 transposition and protein levels was analyzed when the mutant elements were expressed from the GAL1 promoter (Fig 4). We substituted the invariant tryptophan residue found in Ty1/copia Gag proteins to alanine (W184A, helix 1) and tested several published Gag mutations designed to interrupt hydrophobic faces of Gag helices (IM248/249NR, L252R, both in helix 4; LF339/340RD, I343K, both in helix 9) [55]. All helix substitutions abolished Ty1his3-AI mobility when expressed in both 1 and 38 Ty1 copy strains (-p22 and +p22, respectively; Fig 4A). Mature RT was not detected in whole cell extracts from the 1 or 38 Ty1 copy strains expressing mutant Ty1, indicating that Gag-Pol maturation did not occur (Fig 4B). When helix 1 (W184A) or helix 4 (IM248/249NR and L252R) was perturbed, Gag was stable and present in both immature (p49) and mature forms (p45), while p22 maturation to p18 was similar or slightly decreased. In contrast, Gag-p49 and p22 from helix 9 substitutions (LF339/340RD and I343K) did not undergo maturation.
We analyzed Gag W184A (helix 1), L252R (helix 4) and I343 K (helix 9) for VLP assembly by sedimentation of total protein extracts through 7–47% sucrose gradients (Fig 4C). This analysis was performed in the 1 Ty1 copy strain to prevent further disturbance by p22 during VLP formation. Wild type Gag migrated primarily as larger complexes, which are probably comprised of assembled VLPs (Fig 4C, fractions 5–9). W184A and L252R VLP assembly was perturbed, as Gag was present in every fraction of the sucrose gradient. There was less cleavage of Gag-p49 to p45 compared to wild type and both forms were present in each fraction. More cleavage of Gag-p49 to p45 was visible in the higher percent sucrose fractions, suggestive of Ty1 PR activity in these fractions. Gag I343K, which did not exhibit Gag cleavage (Fig 4B), remained at the top of the sucrose gradient and did not form higher order structures (Fig 4C).
CNC is associated with altered abundance and maturation of Ty1 proteins, including loss of mature RT and IN [20, 22]. Disturbing Gag helices in which CNCR mutations mapped also affected Ty1 protein maturation (Fig 4). Therefore, Ty1 protein levels produced by CNCR pGTy1his3-AI elements were examined. Cell extracts were immunoblotted for Gag-p49/p45 and p22 using the 38 Ty1 copy strain (Fig 5A). When wild type pGTy1his3-AI was expressed, there was slightly more p18 than p22. Strikingly, expressing the four CNCR pGTy1his3-AI plasmids resulted in a lower level of p18, indicating less cleavage of p22 and/or decreased stability of p18. Mature RT levels were also recovered in the CNCR strains, suggesting that VLP maturation improved (Fig 5B).
Ty1 VLPs were isolated from the 38 Ty1 copy strain expressing wild type pGTy1his3-AI or pGTy1his3-AI-I201T to determine levels of Ty1 protein and RNA within assembled particles (Fig 6). Equal amounts of VLP preparations were immunoblotted for the detection of Gag, RT and IN. Gag protein levels were similar between wild type and I201T VLPs (Fig 6A). Since wild type VLPs represent different stages of maturation, the samples contained Ty1 precursors Gag-PR-IN-RT (Gag-Pol; p199), PR-IN-RT (p154), IN-RT (p134), and PR-IN (p91). In addition, RT antibodies reacted with two bands of unknown origin around 65 and 90 kD, which we reported previously (Fig 6B, asterisks) [26]. As expected from previous work [22], mature IN (p71) was not detected from wild type VLPs isolated from the 38 Ty1 copy strain (Fig 6C). I201T VLP maturation occurred more efficiently, as indicated by increases in mature IN (Fig 6C) and a decrease in the unknown RT-reactive proteins in I201T VLPs (Fig 6B, asterisks). To determine if the increase in Ty1 mature protein products was due to less p22/p18 present in VLPs, dilutions of wild type and I201T VLPs were immunoblotted with p18 antiserum (Fig 6D). p18, rather than p22, was the main protein observed in wild type VLPs, likely due to cleavage by Ty1 PR within VLPs [26]. The level of p18 within I201T VLPs was lower than that observed in wild type VLPs, raising the possibility that less p18 within assembled CNCR VLPs results in increased Ty1 protein maturation or stability.
Ty1 RNA packaging, as demonstrated by protection from digestion when whole cell extracts are treated with the nuclease benzonase [63], is markedly decreased in the presence of p22/p18 [27]. To determine if a CNCR mutation functions by increasing the level of RNA packaged into VLPs, RNA extracted from purified WT and I201T VLPs was subjected to Northern blotting (Fig 6E). Total cellular RNA was examined to control for Ty1 mRNA expression. Wild type and I201T RNA extracts from cells or purified VLPs contained similar levels of Ty1his3-AI transcript, suggesting that CNCR does not function through the enhancement of Ty1 RNA packaging, at least in the case of I201T.
Interestingly, p22/p18 shares sequence with two regions implicated in Ty1 nucleic acid transactions: the NAC region of Gag (amino acid residues 299–401) and the N-terminus of PR (known as p4 in Gag) that participates in reverse transcription [27, 28, 64]. The first region was extensively examined using recombinant mature p18, which lacks p4 sequence [27]. Recombinant p18 possesses NAC activity and can bind Ty1 RNA, but truncated versions that lack NAC activity still inhibit Ty1 retrotransposition, suggesting that NAC activity is dispensable for p22/p18 function [27]. To test whether PR/p4 is implicated in CNC, we measured the mobility of a single genomic Ty1his3-AI element in presence of transcriptionally repressed wild type pGPOLΔ Ty1 plasmids [26], or derivatives carrying altered p4 regions. Wild type pGPOLΔ plasmids reduced Ty1his3-AI mobility by 150-fold compared to an empty GAL1 plasmid (S6 Table). The pGPOLΔd1 plasmid, which carries a deletion in PR/p4 that blocks successful reverse transcription [64, 65], and pYES2-p45 lacking p4 [27] reduced Ty1his3-AI mobility by 140- and 160-fold, respectively. These results are supported by the observations that ectopic expression of mature p18 alone, which does not contain PR/p4 sequence, inhibits pGTy1his3-AI mobility, and expression of p22/p18 or p22 mutant for the PR cleavage site exhibit similar levels of inhibition [26, 27]. Together, our results show that the PR/p4 region is not required for CNC.
To determine if the low level of p18 detected in I201T VLPs was related to altered binding of p22/p18 with wild type versus Gag I201T, pGp22-V5 and wild type pGTy1his3-AI or pGTy1his3-AI-I201T were co-expressed in a Ty1-less strain. Endogenous Gag produced by chromosomal Ty1 elements and GST-p18 have been shown to interact via a GST pull-down assay [26]. Functional p22-V5, which carries an internal V5 tag that is present in both p22 and p18, was expressed from a low copy CEN plasmid to maximize CNCR imparted by the Gag mutations. Quantitative mobility assays revealed that wild type Ty1his3-AI mobility decreased 783-fold in the presence of p22-V5, while Ty1his3-AI-I201T mobility only decreased 5-fold, confirming the CNCR phenotype (Table 3). Utilizing the V5 tag on p22/p18, co-immunoprecipitations (co-IP) were performed from total cell extracts and analyzed for the level of Gag. We detected co-IP of p22-V5/p18-V5 with wild type Gag (S2 Fig), which confirms previous pull-down analyses with p18 tagged with GST [26]. p22-V5/p18-V5 co-immunoprecipitated 1201T (S2A Fig), K186Q (S2B Fig) or wild type Gag with comparable efficiencies.
To track Gag and p22 and p18 independently during VLP assembly, the fractionation pattern of Gag was examined by sucrose gradient sedimentation as in Fig 4C using total protein extracts from cells expressing wild type pGTy1his3-AI or pGTy1his3-AI-I201T alone or co-expressed with pGp22-V5 (Fig 7). In the absence of p22-V5, wild type and I201T Gag-p49/p45 assembled into VLPs and migrated to fractions 6–9 (Fig 7A and 7B). In the presence of p22-V5, the fractionation pattern of wild type Gag was more dispersed, as reported previously [26] (Fig 7C). While Gag was present throughout the gradient, it was found at the highest concentration in fractions 4–9. In contrast, p22-V5 when co-expressed with wild type pGTy1his3-AI collects as both p22-V5 and p18-V5 and was present in the highest concentration at the top of the gradient (Fig 7D). More p18-V5 co-sedimented with wild type Gag than p22-V5, but p22-V5 and p18-V5 were detected in all fractions. The co-sedimentation of wild type Gag and p18-V5 supports the idea that cleavage by Ty1 PR occurs in complexes migrating to the lower half of the gradient. Surprisingly, p18-V5 was also present at the top of the gradient, which contains most of the soluble proteins in the extract. Considering that the introduction of the internal V5 tag does not alter the requirement of Ty1 PR for cleavage (S3 Fig), p22-V5 may be cleaved by Ty1 PR outside of fully assembled VLPs, perhaps in the Gag complexes present in retrosomes. Alternatively, p22-V5 may be cleaved within VLPs, but not all p22-V5/p18-V5 remains stably associated with the particles. Regardless, our results suggest that a fraction of p22-V5/p18-V5 co-assembles with wild type VLPs. Expression of pGTy1his3-AI-I201T and pGp22-V5 resulted in a Gag fractionation pattern similar to that observed in the absence of p22-V5 (Fig 7E). Interestingly, p22-V5/p18-V5 was detected at the top of the gradient but did not co-sediment with I201T VLPs (Fig 7F). These results suggest that the CNCR conferred by I201T results from the exclusion of p22-V5/p18-V5 from VLPs, perhaps during a step in the assembly process. However, the fact some p18-V5 is produced in these cells suggests that the restriction factor does gain access to PR.
To understand the mechanism of inhibition of Ty1 retrotransposition by the self-encoded restriction factor p22, we isolated and characterized Ty1 element CNCR mutants. All but one of the recovered resistance mutations mapped within GAG and altered Gag protein sequence. More than half of the mutations mapped outside of p22 coding sequence, including the three strongest CNCR mutations recovered (N183D, K186Q, and I201T). Importantly, most CNCR mutations do not reflect simple gain-of-function since the mutations do not increase Ty1 mobility in the absence of p22 (Fig 2B). These results, along with the observations that the mutant centromere-based pGTy1his3-AI plasmids do not produce detectable p22 levels (Fig 1B) or confer CNC [22] due to their low copy number, demonstrate that Gag is the primary molecular target of p22. Although we focused on GAG mutations as they represent the vast majority of resistance mutations recovered, one CNCR mutant contains two sequence changes (D518G/V519A) in the Ribonuclease H (RH) domain of RT (D518G/V519A) within POL that dramatically affected Ty1 fitness in the absence of p22 (0.4% recovery of wild type mobility). The RH domain of RT is responsible for degrading the RNA template during reverse transcription (reviewed in [66]), and the decrease in Ty1 mobility is probably due to the fact that D518 is a conserved residue predicted to be involved in metal chelation [67]. Mutations resulting in decreased Ty1 RT activity, including active site mutations within the polymerization domain or host mutations that inhibit RT activity by altering cytoplasmic manganese levels can be suppressed by mutations in the RH domain [67, 68]. This suppression has been attributed to allosteric communication between the RT polymerization domain and the RH domain [67]. Full-length cDNA is not detectable in cells undergoing CNC [22], because there is a low level of the initial reverse transcription product minus strand strong-stop DNA [69]. The failure of RT is likely a downstream effect of the alteration in VLP maturation and the absence of mature IN [22, 26]. Although the resistance mutation in RH may bypass the primary defect imposed by p22, it may promote a conformation of RT/RH that allows a low level of activity.
The data presented here greatly extends previous work suggesting that a Gag/p22 interaction is central to the mechanism of CNC [26, 27]. Strains undergoing CNC experience a decrease in Ty1 retrotransposition as Ty1 copy number increases [20, 22, 26]. A decrease in mobility of a genomic Ty1his3-AI element was not observed when additional genomic copies carry CNCR mutations, indicating that mutations in Gag, including those that map outside of p22 coding sequence, can relieve CNC in a genomic context. We have not identified resistance mutations with greater than 30% recovery of Ty1 mobility without additionally affecting the fitness of the element (Fig 2). Similarly, combining CNCR mutations resulted in a loss of Ty1 fitness, rather than a combinatorial increase in resistance. These results illustrate the delicate balance between resistance to p22 and overall fitness of Ty1. A similar tradeoff between resistance and fitness exists for mutations in HIV–1 CA that confer resistance to the restriction factor TRIM5α [40]. The inability to achieve complete resistance to TRIM5α is attributed to the genetic fragility of HIV–1 CA, meaning that it is highly sensitive to mutation, and the fact that TRIM5α can bind multiple surfaces on the CA lattice [40, 49]. Similarly, some CA mutations conferring resistance to the Mx2 restriction factor also have a negative impact on HIV infectivity [41]. Ty1 Gag is a multifunctional protein but unlike retroviral Gag is not cleaved into the functionally distinct proteins such as matrix, CA, and nucleocapsid (NC). Even so, Ty1 Gag is responsible for executing the same functions as retroviral CA and NC. Thus our inability to obtain fully resistant Ty1 elements strongly suggests some of the same rules concerning genetic fragility apply to Ty1 Gag, namely that its function is very sensitive to mutation. Secondly, p22 may bind multiple surfaces of the VLP during different stages of assembly, making it difficult to attain full resistance by mutating Gag at only one or two residues. A third consideration is that the surfaces or protein domains that interact with p22 may be the same or overlap with domains important for Gag function. In support of this idea, we recovered Gag mutations, P173L and K250E, that confer CNCR, yet negatively impacted Ty1 fitness in the absence of p22 (Fig 2). K250 is located within predicted helix 4, an amphipathic helix important for VLP maturation (Fig 3A) and perturbation at this site may prevent p22-mediated effects, perhaps by altering VLP assembly and maturation dynamics. However, this alteration in VLP function was not efficient in the absence of p22. Like other infectious agents, the presence of Ty elements in Saccharomyces has resulted in positive selection for certain host genes, suggesting there is an ongoing “genetic conflict” or evolutionary arms race between Ty elements and their host [70]. In our screen, we recovered mutations that mapped within Gag but not p22 coding sequence and found it difficult to recover mutations that fully restored Ty1 retrotransposition in the presence of p22. Gag and p22 share coding sequence in the natural setting, and this is likely to influence a Ty1-p22 arms race for the adaptation of Ty1 to inhibition via p22.
Our bioinformatic analysis of Ty1 Gag revealed 9 predicted helical stretches and two Pfam domains: TYA in the N-terminal half and UBN2 in the C-terminal half of the protein (Fig 3A). Four CNCR mutations mapped within UBN2; although no CNCR mutations were isolated in TYA. UBN2 is a Gag sequence motif that is found in Ty1/copia retrotransposons across plants and fungi. Because UBN2 can be recognized by profile-based methods across a wide range of organisms, this domain is likely involved in a conserved function of Gag. Because known mutations that affect VLP assembly fall within this domain [55], it is reasonable to hypothesize that the UBN2 domain is involved in VLP assembly. UBN2 also overlaps with, but does not fully encompass, the NAC region of Gag (Fig 3A) [28]. Recent work demonstrates that p18 interferes with Gag NAC function [27]. It would be interesting to investigate whether CNCR mutations modulate NAC activity of Gag, although we showed that I201T VLPs do not exhibit enhanced levels of Ty1 RNA packaging (Fig 6E). Additionally, only V336I and Q350R S395L mapped within the NAC region (S2 Table, Fig 3A), suggesting that CNCR does not primarily alter Gag NAC functions. The remaining CNCR mutations mapped to a central region of Gag, which we named the CNCR domain. Predicted helix 1 within the CNCR domain was frequently mutated in our screen and overlaps with conserved domain A present in all Ty1/copia elements that surrounds an invariant tryptophan (Gag W184 for Ty1) [53]. Ty2 Gag differs from Ty1 Gag at many positions and some CNCR mutations are present within Ty2 sequences, raising the possibility that Ty2 is naturally resistant to Ty1 CNC. Ty1 and Ty2 are closely related retrotransposons based on their near identical LTR sequences, with a single nucleotide deletion defining Ty2 LTR sequences [2, 71]. Phylogenetic analyses suggest that S. cerevisiae recently acquired Ty2 elements from S. mikatae by horizontal transfer [3, 72]. We showed that Ty2-917 is neither under CNC by Ty2-917 nor inhibited by GAL1-promoted expression of p22 from Ty1. CNCR residues are also altered in a Tsk1 element from L. kluyveri [57] and a Ty1-like element present in S. kudriavzevii and hybrids thereof [3, 56]. Further mutational analyses of specific CNCR residues within these elements will be required to address the role of naturally occurring CNCR residues in Ty1 and Ty2. In addition, the overlap between Gag and p22 coding sequence may create specificity even within Ty1 elements. It will be interesting to determine if Ty1 elements in natural Saccharomyces isolates confer CNC on Ty1-H3 or whether they have evolved specificity for elements in their native genomes.
Though the structural role of CNCR residues within Ty1 Gag’s helical domains remains to be determined, the mutational analyses presented here define the importance of key hydrophobic residues within these predicted regions. Mutation of the conserved W184 residue within helix 1 to alanine resulted in complete loss of transposition and formation of mature Pol proteins, as well as abnormal VLP assembly (Fig 4). Other Gag helix mutations we tested were previously analyzed in the context of a truncated Gag protein containing amino acids 1–381, which is deleted for 21 residues at the C-terminus of p45 and lacks a complete NAC domain [28, 55]. The truncated Gag is still able to form particles and thus was used to address the role of certain residues in particle assembly. In the context of 1–381 particles, L252R was reported to completely disrupt particle assembly, while LF339-340RD has no effect. IM248/249NR and I343K both alter the migration of 1–381 particles through a sucrose gradient and assemble into “giant” particles when visualized by negative staining and TEM. These mutations had not been characterized in the context of a full length Ty1 element nor analyzed for effects on Ty1 mobility. Therefore, our work showed that altering these hydrophobic residues within helices severely hinders Ty1 mobility (Fig 4). Our results regarding particle assembly with these Gag substitutions differ from previous work, as L252R was capable of forming higher order complexes and I343K was not (Fig 4C). These conflicting results could be due to the differences in multimerization and VLP assembly using full length Gag compared to truncated 1–381 Gag used by others [55]. Importantly, since most CNCR mutations mapped to helices required for normal Ty1 transposition, protein processing and VLP assembly, our results indicate that inhibition by p22 disturbs a central function of Gag.
Mutations exhibiting the highest level of CNCR (N183D, K186Q, I201T, and A273V) were associated with reduced levels of p18 (Fig 5), which is cleaved from p22 near its C-terminus by Ty1-PR, and there was less co-assembly between p22 and Gag I201T compared to wild type Gag (Fig 7). These results suggest that the reduction in p18 levels in the presence of the CNCR VLPs may result from diminished access to PR. More p22/p18 associated with purified wild type VLPs when compared with I201T VLPs (Fig 6D). When total cellular protein from cells co-expressing p22-V5 and wild type or Ty1his3-AI-I201T was analyzed by sucrose gradient sedimentation (Fig 7), the majority of p22-V5 and p18-V5 remained in fractions containing soluble protein rather than in fractions containing VLPs or higher order assembly intermediates. Because Ty1 PR-mediated processing is thought to occur only within assembled VLPs and cleavage of p22 and p22-V5 was Ty1 PR-specific (Fig 1D and S3 Fig), p22/p18 may be capable of moving in and out of the VLP. Perhaps this occurs by diffusion of p22/p18 through VLP pores, which are permeable to ribonuclease A (15.7 kD) but not to benzonase (30 kD) treatment in vitro [73–75]. Consequently, p22/p18 may still be within the acceptable size limit to enter and exit VLP pores. Alternatively, once p22/p18 co-assembles with Ty1 proteins in VLPs and maturation is initiated, p22/p18-containing VLPs may be subject to dissociation and degradation. Although we did observe a modest shift in Ty1 Gag fractionation towards the top of the gradient in the presence of p22/p18, Gag was not concentrated in the first two fractions with p22/p18. Lastly, our results raise the possibility that Ty1 PR may function outside of stably assembled VLPs, perhaps in assembly intermediates present in retrosomes, which are cytoplasmic foci containing Ty1 mRNA and proteins [7, 9, 76]. In support of this idea, few if any VLPs are detected in cells containing retrosomes resulting from endogenous Ty1 expression, VLP assembly increases dramatically when Ty1 is overexpressed from a strong promoter, and assembly occurs within retrosomes [7, 25, 77]. Recent work also shows that steady state Gag expressed from endogenous Ty1 elements does not co-migrate with unprocessed Gag-p49 [76], suggesting that Gag cleavage can occur in the absence of detectable VLPs. Finally, several earlier studies demonstrate the presence of mature p45 resulting from endogenous Ty1 expression [8, 24, 78–80]. Together, our results suggest that p22 cleavage may occur in the same spatiotemporal environment as pre-VLP Gag cleavage.
We observed varying degrees of p22 cleavage and/or p18 stability in the presence of altered Ty1 Gag proteins. While p22/p18 levels were comparable to WT in Gag L252R, p18 was not detected in Gag LF339-340RD and Gag I343K. CNCR mutations (N183D, K186Q, I201T, and A273V) and the helix-altering Gag W184A and IM248-249NR were associated with decreased levels of p18, but we cannot distinguish if these changes represent a decrease in p22 cleavage or a reduction in p18 stability. It is interesting to consider that some loss-of-function changes in Gag (W184A and IM248-249NR) and the gain-of-function CNCR mutations both result in decreased p18 levels. Perhaps p22 cleavage is a read-out for both Ty1 PR activity, which can be affected by several different situations, such as Gag:Gag-Pol ratio and particle assembly [47, 81–83], or access of the p22 substrate to Ty1 PR. Whereas the helix mutations alter VLP assembly, the resistance mutations likely affect access to PR, since p22 is excluded from CNCR VLPs. Thus, reduced p22 cleavage can occur in both loss-of-function and gain-of-function contexts.
Although the CNCR mutations in Gag might affect Gag/p22 binding, co-IP experiments performed using standard washing conditions did not support a simple interaction between p22 and Gag involving CNCR residues (S2 Fig). In addition, sucrose gradient fractionation indicated that most p22-V5/p18-V5 was present in the fractions containing soluble proteins and did not co-sediment with VLPs. Perhaps p22 is capable of binding several forms of Gag, whether monomeric, small assembly intermediates, or intact VLPs, and perhaps these interactions inhibit VLP assembly or maturation with different potencies. If the crucial binding substrate of p22 is multimeric and represents a minority of Gag molecules present in the cell, co-IP analysis may not show differences in binding. Interestingly, retroviral CA-binding restriction factors TRIM5α and the Gag-derived Fv1 bind to their CA target after polymerization of the lattice [84, 85]. We are considering that the interaction between p22 with polymerized/assembled Gag alone may be the defining and initial insult to Ty1 replication.
Retroviral studies involving sensitivity and escape from host restriction factors show similarities to the Ty1-p22 system. Mx2 restriction of HIV–1 is thought to involve inhibition of viral uncoating and/or nuclear entry and requires Mx2-CA binding [41, 86]. However, known Mx2 escape mutations in the CA gene do not significantly alter binding between Mx2 and CA [41], which demonstrates that viral escape mutations can promote replication in ways distinct from the disruption of restriction factor-target binding. In the case of the resistant provirus enJSRV26, increasing the levels of enJSRV26 Gag expression relative to the restriction factor enJS56A1 Gag protein is enough to allow JSRV replication in sheep [43]. Increased expression of enJSRV–26 Gag is achieved by mutation of the signal peptide in the envelope glycoprotein, which modulates proviral gene expression. Similarly, increasing the level of Ty1 expression can overcome CNC [48]. We favor the hypothesis that understanding how p22 is excluded from CNCR VLPs is central to understanding CNC. Since the steady state level of Gag was unaffected in CNCR mutants (Fig 5A), perhaps the ratio of Gag:p22 is specifically higher within retrosomes comprised of CNCR Gag.
In summary, we have shown that mutations in Gag confer resistance to the p22 restriction factor produced by Ty1 during CNC. These mutations are beneficial only in the presence of p22 and do not globally increase Ty1 mobility. CNCR mutations allow for VLP maturation, which may be the step in Ty1 replication most sensitive to CNC, by excluding p22 from assembling particles. Identification of the Gag multimerization states that bind p22 and host factors that modulate Gag assembly, in combination with studies examining VLP assembly dynamics and structure, especially regarding the newly identified Gag domains, will deepen our understanding of retroelement control.
Strains are listed in S1 Table. Strains repopulated with Ty1 elements were obtained following pGTy1 induction as described previously [20]. Ty1 insertions following repopulation experiments were estimated by Southern blotting as in [48]. Standard yeast genetic and microbiological procedures were used in this work [87].
Refer to S3 Table for plasmid descriptions and sources. Directed mutagenesis was carried out by overlap PCR using the following primer sequences: W184Ab; 5’-ATGTTTTAACAGCATTTGGAAAGTCATTAGGTGAGGTTAAC; W184Ac, 5’-GACTTTCCAAATGCTGTTAAAACATACATCAAATTTTTAC; L252Rb, 5’-ATACTTTTGGATCTAATTTTCATGATATCCGTATAATCAACG; L252Rc, 5’-TCATGAAAATTAGATCCAAAAGTATTGAAAAAATGCAATCTG; IM248/9NRb, 5’-AAAGAATTTTCCTGTTATCCGTATAATCAACGGATAGGAT; IM248/9NRc, 5’-TATACGGATAACAGGAAAATTCTTTCCAAAAGTATTGAAA; LF339/40RDb, 5’-GGATATCTAAGTCCCGTTCAGCGACTGTCATATTTAGATG; LF339/40RDc, 5’-GTCGCTGAACGGGACTTAGATATCCATGCTATTTATGAAG; I343Kb, 5’-AAATAGCATGCTTATCTAAGAACAGTTCAGCGACTGTCAT; I343Kc, 5’-CTGTTCTTAGATAAGCATGCTATTTATGAAGAACAACAGG. For pBJM24, the plasmid markers were switched from URA3 to TRP1, as described previously [26]. Galactose-inducible centromere (CEN) vectors expressing p22-V5 were created by PCR amplification of Ty1-H3 p22 coding sequence 1038–1613 with the internal V5 tag and flanking GAL1P and CYC1 TT sequences using pBDG1568 as a template [26] and primers: cla1_galp, 5’-CATGTTTCATCGATACGGATTAGAAGCCGCCGAGC; cyc1ttrevSacII, 5’-CATGTTTCCCGCGGGAGTCAGTGAGCGAGGAAGC. The insert was cloned into an empty URA3/CEN vector (pRS416) using ClaI and SacII sites. The V5 tag is located between nucleotides 1442 and 1443[26]. pTy2-917his3-AI (pBDG631) was constructed by digestion of pGTy917 with BglII and pBJC42 (his3-AI, pBDG619) with ClaI, fill-in synthesis of the linearized vector and his3-AI fragments using DNA polymerase I followed by blunt end-ligation. pGPOLΔd1 was constructed by BglII digestion and reclosure of pGTy1his3-AId1 (kindly provided by Jef Boeke [54, 64] and Joan Curcio [88]), which deletes the majority of POL. pYES2-p45 was constructed by PCR using primers specific for the coding sequence of p45 and the amplification product was cloned into the multi-copy GAL1-promoted expression vector pYES2 [27]. Recombinant plasmids were verified by restriction enzyme analysis or DNA sequencing.
Plasmid mutagenesis was performed by transforming 50 ng of pBDG606 (S3 Table) into XL–1 Red (Agilent Technologies) cells and sub-culturing transformants for 3–4 days at 37°C. Gap repair was performed with pBDG606 using mutagenized GAG template and AatII (upstream of GAL1P) and BstEII (within PR) sites, while POL mutagenesis was performed using BstEII and XbaI (within his3-AI) sites. Primers flanking these restriction sites (GAG: USAatII, 5’-ATAATACCGCGCCACATAGC; RP1, 5’-CATTGATAGTCAATAGCACTAGACC; POL: USBsteIIf, 5’-GCACGACCTTCATCTTAGGC; 3pLTRrev, 5’-ATCAATCCTTGCGTTTCAGC) were used in a standard Taq (ThermoFisher Scientific, Waltham, MA) PCR reaction with Ty1-H3 as a template to mutagenize the area of interest at a low frequency. XL–1 Red treated pBDG606 or DNA fragments for gap repair were transformed into YEM515 (see S1 Table) and plated onto SC-Ura. Transformants, were replica plated on SC-Ura + 2% galactose, incubated at 22°C for 1–2 days, and then replica plated on SC-Ura-His and incubated at 30°C for 2 days. Candidate plasmids were extracted, propagated in E. coli, transformed into YEM514 and YEM515 and retested for pTy1his3-AI mobility. Candidates with at least a 10-fold increase in retromobility in YEM515 were carried forward. After sequencing the CNC region of XL–1 Red treated plasmids, the GAL1 and GAG segments were subcloned into wild type plasmid using AatII and Eco91I sites to eliminate other mutations present outside of the region of interest. In all cases, subcloned GAG mutations conferred a similar level of CNCR as the primary isolates. For the gap repair screen, the entire region amplified by low fidelity PCR was sequenced.
The mobility frequency of Ty1his3-AI was determined as described previously [21, 48] with the following modifications. For strains transformed with only pGTy1his3-AI, single colonies were grown at 30°C overnight in 1 ml of SC-Ura + 2% raffinose and then diluted 1:25 in quadruplicate 1 ml SC-Ura + 2% galactose cultures. Galactose cultures were grown at 22°C for 2 days, and cells were then washed, diluted and spread onto SC-Ura and SC-Ura-His plates. For strains transformed with both pGTy1his3-AI and p22-containing plasmids, similar methods were used for the assay except liquid and solid media also lacked tryptophan. For qualitative mobility assays with pGTy1his3-AI, cells were patched onto SC-Ura and grown at 30°C for 2 days. Cells were replica plated onto SC-Ura +2% galactose and incubated at 22°C for 2 days, followed by replica plating onto SC-Ura-His and incubation at 30°C until His+ papillae appeared. For transposition assays involving chromosomal Ty1his3-AI, a single colony was dissolved in 10 ml water. One microliter of diluted cells was added to quadruplicate 1 ml SC-Ura or YEPD cultures and grown 2–3 days until saturation. The cells were washed, diluted and spread onto SC-Ura or YEPD and SC-Ura-His or SC-His plates, and incubated at 30°C until colonies formed.
For strains carrying pGTy1his3-AI, 1 ml of SC-Ura + 2% raffinose was inoculated with a single colony and grown overnight at 30°C, then diluted 1:10 into SC-Ura + 2% galactose and grown at 22°C for 24 h. For growth in glucose, a dilution of 1:100 was used. To detect p22/p18, 5 ml of culture was processed by trichloroacetic acid (TCA) extraction as described previously [26]. To detect all other Ty1 proteins, protein from 10 ml of culture was extracted as previously described [89] and 30 μg of protein was used for immunoblotting. Samples were separated on 10% (for RT and IN detection or to separate Gag-p49 and p45) or 15% (Gag-p49/p45 and p22/p18 detection) SDS-PAGE gels and immunoblotted as described previously [26]. Antibody dilutions were as follows: anti-p18 1:5000 [26], anti-VLP 1:10,000, anti-RT 1:5,000, anti-IN 1:2500, anti-Hts1 1:40,000, anti-TY (BB2, UAB Epitope Recognition and Immunoreagent Core, Birmingham AL) 1:50,000, anti-V5 1:20,000 (Life Technologies, Carlsbad, CA).
Ty1-H3 sequence (GenBank M18706.1) was submitted to the following online servers for secondary structure prediction: ITASSER [50–52] PredictProtein [90], Sable [91], PSIPRED [92], and SAMTO8[93]. When comparing the secondary structure predictions, the results were consistent, with the same helices predicted by all five servers. The boundaries of the helices varied slightly, but not by more than three residues. The I-TASSER results were chosen for display in Fig 3. Protein domains in Ty1 sequence were predicted using profile hidden Markov models [94] by scanning Ty1 Gag sequence against the Pfam database. Ty1 related sequences in UniProt were identified using HMMER [94] and aligned using CLUSTALW [95]. Full alignment can be found in the supplemental data (S1 Fig). Protein alignments were visualized using Jalview (http://www.jalview.org/) [96]. ClustalX coloration was used with a conservation color increment of 35. The raw alignment file is provided as S1 File.
VLPs were isolated as described previously [26], except the cells were induced in SC-Ura + 2% galactose for 24 h at 20°C. Two micrograms of final VLPs were immunoblotted to detect Gag, RT, and IN. A 1:2 dilution series was loaded to detect p18.
Equivalent total cellular RNA and VLP RNA, as estimated by OD600 or total Gag protein respectively, was extracted using the MasterPure yeast RNA purification kit (Epicentre Biotechnologies, Madison, WI) and analyzed via Northern blotting as previously described [26].
Antibodies were crosslinked to resin using a Pierce Crosslink IP Kit (ThermoFisher Scientific) and following the manufacturer’s instructions. For immunoprecipitations, a 25 ml culture was induced in SC-Ura-Trp + 2% galactose at 20°C for 24 h or until OD600 = 1.0. IP Lysis buffer was supplemented with 1 μg/ml aprotonin, pepstatin and leupeptin and 1 mM PMSF. Cells were broken in IP Lysis buffer plus protease inhibitors by vortexing with glass beads. Equal amounts of protein were applied to Protein A/G agarose crosslinked with 2 μg of V5 Antibody (Life Technologies) and allowed to bind for 2 h at 4°C. Beads were washed with IP Lysis buffer and eluted with 20 μl of elution buffer. 1/100 of the input and 1/2 of the pull-down material were loaded per lane. Beads not crosslinked to V5 antibody served as a negative control.
A 100 ml culture was induced in SC-Ura or SC-Ura-Trp + 2% galactose at 20°C for 24 h or until the culture reached an OD600 of 1. Cells were broken in 15 mM KCl, 10 mM HEPES-KOH, pH 7, 5 mM EDTA containing RNase inhibitor (100 U per ml), and protease inhibitors (16 μg/ml aprotinin, leupeptin, pepstatin A and 2 mM PMSF) in the presence of glass beads. Cell debris was removed by centrifuging the broken cells at 10,000 x g for 10 min at 4°C. Five milligrams total protein in 300–500 μl of buffer was applied to a 7–47% continuous sucrose gradient and centrifuged using an SW41 rotor at 25,000 rpm (~100,000 x g) for 3 h at 4°C. After centrifugation, 9 x 1.2 ml fractions were collected and 30 μg of the input and 15 μl of each fraction was immunoblotted to detect Ty1 proteins.
|
10.1371/journal.pcbi.1002893 | Bayesian Inference of Spatial Organizations of Chromosomes | Knowledge of spatial chromosomal organizations is critical for the study of transcriptional regulation and other nuclear processes in the cell. Recently, chromosome conformation capture (3C) based technologies, such as Hi-C and TCC, have been developed to provide a genome-wide, three-dimensional (3D) view of chromatin organization. Appropriate methods for analyzing these data and fully characterizing the 3D chromosomal structure and its structural variations are still under development. Here we describe a novel Bayesian probabilistic approach, denoted as “Bayesian 3D constructor for Hi-C data” (BACH), to infer the consensus 3D chromosomal structure. In addition, we describe a variant algorithm BACH-MIX to study the structural variations of chromatin in a cell population. Applying BACH and BACH-MIX to a high resolution Hi-C dataset generated from mouse embryonic stem cells, we found that most local genomic regions exhibit homogeneous 3D chromosomal structures. We further constructed a model for the spatial arrangement of chromatin, which reveals structural properties associated with euchromatic and heterochromatic regions in the genome. We observed strong associations between structural properties and several genomic and epigenetic features of the chromosome. Using BACH-MIX, we further found that the structural variations of chromatin are correlated with these genomic and epigenetic features. Our results demonstrate that BACH and BACH-MIX have the potential to provide new insights into the chromosomal architecture of mammalian cells.
| Understanding how chromosomes fold provides insights into the complex relationship among chromatin structure, gene activity and the functional state of the cell. Recently, chromosome conformation capture based technologies, such as Hi-C and TCC, have been developed to provide a genome-wide, high resolution and three-dimensional (3D) view of chromatin organization. However, statistical methods for analyzing these data are still under development. Here we propose two Bayesian methods, BACH to infer the consensus 3D chromosomal structure and BACH-MIX to reveal structural variations of chromatin in a cell population. Applying BACH and BACH-MIX to a high resolution Hi-C dataset, we found that most local genomic regions exhibit homogeneous 3D chromosomal structures. Furthermore, spatial properties of 3D chromosomal structures and structural variations of chromatin are associated with several genomic and epigenetic features. Noticeably, gene rich, accessible and early replicated genomic regions tend to be more elongated and exhibit higher structural variations than gene poor, inaccessible and late replicated genomic regions.
| The spatial organization of a genome plays an important role in gene regulation, DNA replication, epigenetic modification and maintenance of genome stability [1]–[5]. Understanding three-dimensional (3D) chromosomal structures and chromatin interactions is therefore essential for decoding and interpreting functions of the genome. Traditionally, the 3D organization of chromosomes has been studied by microscopic and cytogenic methods such as florescent in situ hybridization (FISH). Several FISH studies have shown that the global 3D chromosomal structures are highly dynamic [6]–[8]. However, due to the limitation of low throughput, low resolution FISH data, the 3D chromosomal structures at the fine scale are not fully understood. In particular, whether chromatin exhibits a consensus local 3D chromosomal structure is still under debate. More recently, higher throughput, higher resolution approaches based on chromosome conformation capture (3C) such as Hi-C [9] and TCC [10] allow genome-wide mapping of chromatin interactions. The chromatin interactions captured by Hi-C and TCC experiments, which are represented by the contact matrix in the original Hi-C study [9], provide an unprecedented opportunity for inferring 3D chromosomal structures at the fine resolution scale.
Much progress has been made in recent years to reconstruct 3D chromosomal structures from the Hi-C data by translating the observed chromatin contact frequency between two genomic loci to the population average spatial distance between them. Bau and colleagues [11] translated the read counts in the contact matrix to spatial constraints of 3D chromosomal structures and used the software Integrated Modeling Platform (IMP) [12] to solve a constrained optimization problem. Duan et al. [13] devised a set of constraints for all loci of the genome, and solved a similar constrained optimization problem using an open-source software IPOPT [14]. Similar optimization-based approaches have also been used in studies of the fission yeast genome [15]. Kalhor et al. [10] proposed another optimization-based approach which correlates contact frequencies with the presence or absence of chromatin contacts instead of average spatial distances. More recently, Rousseau et al. [16] developed a probabilistic model linking Hi-C data to spatial distances and designed a Markov-chain Monte Carlo-based method named MCMC5C. Different from the optimization-based approaches, MCMC5C models the uncertainties of spatial distances between two loci by assuming that the number of reads spanning those two loci follows a Gaussian distribution.
However, all the existing methods have several limitations. First, as pointed out by Yaffe and Tanay [17], the raw data obtained from Hi-C experiments exhibit multiple layers of systematic biases, such as restriction enzyme cutting frequencies, GC content and sequence uniqueness. None of the existing methods take these systematic biases into consideration. Second, optimization-based methods are prone to be trapped in local modes due to the ultra-high dimensionality and the prohibitively large search space. Third, MCMC5C suffers from the difficulty in estimating the Gaussian variance of each read count since the single Hi-C contact matrix does not provide enough information for variance estimation. Furthermore, except for MCMC5C, none of these existing methods comes with a stand-alone software [16].
More importantly, all of the existing methods focus on reconstructing consensus 3D chromosomal structures, but pay little attention to evaluating magnitudes of structural variations of chromatin at different resolution scales. To quantify structural variations of chromatin, the optimization-based methods usually require a large number of parallel runs, which is computationally intensive and not directly interpretable. Similarly, the Gaussian model in MCMC5C is derived from a consensus 3D chromosomal structure, which cannot be used to measure structural variations of chromatin either.
Since chromatin interactions captured by Hi-C experiments come from a cell population instead of a single cell, it is challenging to study structural variations of chromatin from the Hi-C data. When the cell population consists of multiple sub-populations, of which each corresponds to a distinct 3D chromosomal structure, the Hi-C data can only be interpreted as a measurement of the population average effect. The Hi-C data of mammalian genomes is further complicated by the fact that the pair of homologous chromosomes cannot be distinguished from each other without genotype information. Without fully characterizing structural variations of chromatin in a cell population, the consensus 3D chromosomal structure inferred from the Hi-C data is not directly interpretable or even misleading.
Although the global 3D chromosomal structure is indeed quite dynamic in a cell population, the local 3D chromosomal structure could be homogeneous. A recent study [18] on a high resolution Hi-C dataset has discovered that mammalian genomes are composed of thousands of mega-base-sized, evolutionarily conservative topological domains, which appear to serve as units of genomic organization and perhaps function. These findings motivate the hypothesis that each topological domain may share a consensus 3D chromosomal structure in order to keep its conservative functional forms. For local genomic regions where this hypothesis holds true, the mixture of cell populations and the ambiguity of homologous chromosomes will no longer be major barriers for 3D modeling based on Hi-C data.
In this work, we test the hypothesis of consensus 3D structure at the topological domain scale via rigorous statistical analysis of Hi-C data. To achieve this goal, we propose two integrated probabilistic approaches called BACH (which is the short name for “Bayesian 3D Constructor for Hi-C data”) and BACH-MIX. It should be noted that our approach is closely related to inferential structure determination (ISD) [19], a Bayesian approach developed to study macromolecular structure. In the BACH algorithm, we assume that the local genomic region (i.e., a topological domain) of interest exhibits a consensus 3D chromosomal structure in a cell population, and employ efficient Markov chain Monte Carlo (MCMC) computational tools to infer the underlying consensus 3D chromosomal structure. In the BACH-MIX algorithm, we assume that the genomic region of interest consists of multiple distinct 3D chromosomal structures, and explicitly model structural variations of chromatin using a mixture component model. By comparing the goodness of fit of BACH and BACH-MIX for the same genomic region via statistical model selection principles, we provide a quantitative approach to evaluate structural variations of chromatin for any given local genomic region.
Applying BACH and BACH-MIX to a high resolution Hi-C dataset, we found that BACH, instead of BACH-MIX, is preferred in about half of the topological domains. Of the topological domains in which BACH-MIX fits the data better, most contain one dominant sub-population, whose 3D chromosomal structure can be reconstructed by the BACH algorithm. These results suggest that most topological domains exhibit homogeneous 3D chromosomal structures in a cell population. We also found that geometrical properties of these topological domains, particularly the shape and the structural variations, are associated with several genomic and epigenetic features. Furthermore, we found significantly lower structural variations at domain center regions than at domain boundary regions.
The BACH algorithm takes the chromosomal contact matrix generated by Hi-C or TCC experiments and local genomic features [17], [20] (restriction enzyme cutting frequencies, GC content and sequence uniqueness) as input, and produces, via MCMC computation, the posterior distribution of 3D chromosomal structures (Methods). In the BACH algorithm, we assume that there exists a consensus 3D chromosomal structure in a cell population (this assumption will be relaxed later in the BACH-MIX algorithm). Furthermore, we assume that the number of sequencing reads spanning two genomic loci follows a Poisson distribution, where the Poisson rate is negatively associated with the corresponding spatial distance between them and is also affected by a few other factors. BACH can be used to reconstruct consensus 3D chromosomal structures from the Hi-C contact matrix, and infer the uncertainties of the spatial distance between any two genomic loci from the corresponding posterior distribution. Simulation studies have shown that the BACH algorithm works well under the posited model (Text S1).
Compared to other published methods, BACH has the following advantages: (1) It explicitly models and corrects known systematic biases associated with Hi-C data, such as restriction enzyme cutting frequencies, GC content and sequence uniqueness [17], [20]; (2) It utilizes a Poisson model that better fits the count data generated from Hi-C experiments than the Gaussian model used in MCMC5C, and performs more robustly when applied to several experimental datasets (see the following RESULTS section for validation); (3) It employs advanced MCMC techniques, such as Sequential Monte Carlo and Hybrid Monte Carlo (see Text S1 for details), that significantly improve the efficiency in exploring the vast space of possible models [21].
In the BACH algorithm, we assume that chromosomal regions of interest exhibit a consensus 3D chromosomal structure in a cell population. However, this assumption may not be true, because chromosomal regions may exist in multiple inter-convertible configurations. To test the consensus 3D chromosomal structure assumption and study structural variations of chromatin in a cell population, we propose a variant algorithm called BACH-MIX (Methods). In BACH-MIX, we assume that the genomic region of interest is composed of two adjacent sub-regions, each with a rigid consensus 3D structure, but the spatial arrangement of the two sub-structures can vary in a cell population. BACH-MIX models the uncertainty of the spatial arrangement between the two sub-structures by a mixture component model, where each component corresponds to one specific spatial arrangement. The weight of each component represents the proportion of that component in a cell population. Clearly, BACH is a special case of BACH-MIX, in which the number of the mixture component is one. We use the statistical model selection criterion, the Akaike information criterion (AIC) [22], to determine whether BACH or BACH-MIX fit the data better, so as to infer whether the structure is homogeneous (having a consensus) or variable.
BACH-MIX contains two types of parameters: the parameters to determine the local consensus 3D chromosomal structures of the two adjacent sub-regions, and the parameters to determine the spatial arrangement of the two adjacent sub-regions. In practice, the local 3D chromosomal structures of the two adjacent sub-regions can be estimated by applying BACH twice separately, each to the contact map of one sub-region. The main computation in BACH-MIX is to estimate the parameters corresponding to each spatial arrangement of the two adjacent sub-structures.
A spatial arrangement of the two adjacent sub-structures can be represented by a rotation matrix with three Euler angles [23]. We also take into account mirror symmetry structures that cannot be explained by rotations. To simplify the computation, we discretize the range of each Euler angle into four bins of equal sizes, and approximate the collection of distinct 3D chromosomal structures in a cell population by 104 spatial arrangements of two adjacent sub-regions (Text S1). The BACH-MIX algorithm takes 3D chromosomal structures BACH predicted for two adjacent sub-regions and the corresponding local genomic features [17] (restriction enzyme cutting frequencies, GC content and sequence uniqueness) as input, and produces the posterior distribution of the spatial arrangement of the two sub-regions, quantified by the proportion of each of the 104 orientations between the two. Simulation studies have shown that the BACH-MIX algorithm works well under the posited model (Text S1).
In practice, a majority of the 104 spatial arrangements of the two adjacent sub-regions are insignificant in terms of having very low proportions. To overcome over-fitting, we adopt a two-step procedure to achieve sparsity: first, we apply the full BACH-MIX model with 104 spatial arrangements to estimate the proportion for each of them; second, we remove insignificant spatial arrangements whose proportion is less than 1%, and re-estimate the proportion for the significant spatial arrangements.
We applied BACH and BACH-MIX to a dataset recently generated in our lab [18] from a mouse embryonic stem cell (mESC) line. The dataset includes 476 million reads obtained from two biological replicates processed with the use of the restriction enzyme HindIII (referred to as the HindIII sample); and 237 million reads in another biological replicate processed with the use of the restriction enzyme NcoI (referred to as the NcoI sample). To the best of our knowledge, this dataset provides the highest sequencing depth of a mammalian genome to date. Previous analysis of this dataset showed that the mouse genome is composed of 2,200 topological domains characterized by high frequencies of intra-domain interactions but infrequent inter-domain interactions [18].
We conducted a genome-wide analysis by applying BACH and BACH-MIX to this high-resolution mESC Hi-C dataset. Both BACH and BACH-MIX were applied to the 40 KB resolution Hi-C contact matrices. In the preprocessing procedure, we filtered out 300 topological domains whose length is less than 400 KB or do not contain known mouse gene (13.64% out of total 2,200 domains). We also filtered out a subset of 40 KB genomic loci within each topological domain according to restriction enzyme cutting frequencies (number of fragment end < = 5), GC content (< = 0.3) and sequence uniqueness (mappability score < = 0.8) (Figure S1), and created the 40 KB resolution Hi-C contact matrix for each topological domain. We then applied BACH to each of the remaining 1,900 topological domains to infer its 3D chromosomal structure.
To validate the spatial distances inferred by the BACH algorithm, we compared the spatial distances BACH predicted (referred to as the BACH distances) to the spatial distances measured by FISH [24] (referred to as the FISH distances). In the HindIII sample, the Pearson correlation coefficient between the BACH distances and the FISH distances is 0.88 (95% credible interval is [0.83, 0.92]). In the NcoI sample, the Pearson correlation coefficient between the BACH distances and the FISH distances is 0.83 (95% credible interval is [0.67, 0.93]). These results suggest that the spatial distances BACH predicted are consistent with the spatial distances measured by FISH (Text S1 and Figure S2). As a comparison, we applied MCMC5C and obtained the corresponding predictions of spatial distances (referred to as the MCMC5C distances). The Pearson correlation coefficients between the MCMC5C distances and the FISH distances are 0.79 and 0.11 in the HindIII sample and the NcoI sample, respectively, which are much worse than those of the BACH's results (z-test p-values <2.4e-5). In addition, we applied a modified BACH algorithm without bias correction and found it still achieved higher correlation with the FISH distances than MCMC5C (Text S1).
In the previous analysis, we obtained the 3D chromosomal structure predicted by BACH for each topological domain. Next, we divided each topological domain into two sub-regions of equal sizes, and applied BACH-MIX to infer the spatial arrangement of the two sub-regions. We evaluated the goodness of fit of the BACH model and the BACH-MIX model for each of these 1,900 topological domains in terms of AIC, which penalizes the log-likelihood of a model with the number of parameters in the model. A smaller AIC indicates a better model fitting. In the HindIII sample, BACH achieved smaller AIC than BACH-MIX in 875 out of 1,900 (46.05%) topological domains. For the rest 1,025 topological domains where BACH-MIX fits the data better than BACH, 487 topological domains have one dominant spatial arrangement of the two sub-regions with proportion greater than 80%. In 482 out of these 487 topological domains, the dominant 3D chromosomal structure can be captured by BACH. Therefore, BACH can reconstruct the consensus structure or the dominant structure in 1,357 topological domains (71.42% of 1,900 topological domains). We obtained consistent results in the NcoI sample. In the NcoI sample, BACH achieved smaller AIC than BACH-MIX in 1,156 out of 1,900 (60.84%) topological domains. For the rest 744 topological domains where BACH-MIX fits the data better than BACH, 394 topological domains have one dominant spatial arrangement of the two sub-regions with proportion greater than 80%. In 393 out of these 394 topological domains, the dominant 3D chromosomal structure can be captured by BACH. Therefore, BACH can reconstruct the consensus structure or the dominant structure in 1,549 topological domains (81.53% of 1,900 topological domains).
In the following analysis, we focus on 1,199 (the overlap between 1,357 topological domains in the HindIII sample and 1,549 topological domains in the NcoI sample, 63.11% out of 1,900) topological domains in which BACH can reconstruct the consensus 3D chromosomal structure or the 3D chromosomal structure of the dominant sub-population in both HindIII sample and NcoI sample. To summarize the structural properties of topological domains, we approximated each 3D chromosomal structure BACH predicted (40 KB resolution) by a cylinder, and computed the ratio between its height and diameter, abbreviated as HD ratio (Methods). Domains with higher HD ratios are more elongated. HD ratios of the structures inferred from the HindIII sample and the NcoI sample are highly reproducible (Pearson correlation coefficients = 0.76, p-value<2.2e-16).
To evaluate the relationship between structural properties of chromatin (measured by HD ratio) and its functional forms at the topological domain scale, we collected genomic and epigenetic features for each topological domain, including gene density (UCSC reference genome mm9), gene expression [25], five histone modification marks (H3K36me3 [26], H3K27me3 [27], H3K4me3 [25], H3K9me3 [28] and H4K20me3 [27]), RNA polymerase II [25], chromatin accessibility [29], genome-nuclear lamina interaction [30] and DNA replication time [31]. By computing the correlation between the HD ratio and each of the genomic and epigenetic features, we found that the HD ratio is highly significantly and positively correlated with gene density, gene expression, transcription elongation histone modification mark H3K36me3, repressive histone modification mark H3K27me3, promoter mark H3K4me3, RNA polymerase II, accessible chromatin and early replicated genomic regions, and negatively associated with heterochromatin marks H3K9me3, H4K20me3 and lamina associated domains (Table S1). These correlations are similarly computed based on either the HindIII sample or the NcoI sample. Two illustrative examples are shown in Figure 1 and Table S2. Consistent with other existing biological evidences, these results demonstrate that the gene rich, actively transcribed, accessible, and early replicated chromatin tends to be more elongated than the gene poor, lowly transcribed, inaccessible and late replicated chromatin, which is consistent with previous FISH experiments [32].
The original Hi-C study [9] has shown that chromatin interactions closely correlate with the genomic and epigenetic features. By applying the principle component analysis (PCA) method to the Hi-C data, Lieberman-Aiden et al. [9] demonstrated that two compartments (compartment A and compartment B) in the mammalian genome can be obtained, where compartment A is strongly associated with open, accessible, and actively transcribed chromatin [33]. Following their strategy, we also applied the PCA method to the mESC Hi-C dataset [18] and obtained the compartment label (A or B) for each topological domain. The compartments A and B represent a high level interpretation of the Hi-C data, but do not inform us on the details of chromatin folding. Recently, we and others showed that compartments A and B could be further partitioned into topological domains, which are mega-base-sized, self-interacting genomic regions [18], [34]. Using BACH, we generated 3D models of topological domains, and found that topological domains in compartment A are significantly more elongated than those in compartment B. In the HindIII sample, mean HD ratios for domains in compartment A and compartment B are 1.81 and 1.34, respectively (two sample t-test p-value<2.2e-16). Similarly, in the NcoI sample, mean HD ratios for domains in compartment A and compartment B are 1.76 and 1.26, respectively (p-value<2.2e-16). Two illustrative examples are shown in Figure 1 and Table S2. These results suggest that the HD ratio obtained in the BACH algorithm provides an intuitive visual interpretation of the Hi-C data.
We further study the structural variations of chromatin in a cell population. We first selected 562 topological domains with size larger than 1 MB, and applied BACH and BACH-MIX to the 1 MB region around the center of each selected domain center region. Additionally, we used 985 domain boundaries with size shorter than 40 KB as the control group, and applied BACH and BACH-MIX to the 1 MB region around each selected domain boundary region. We divided each 1 MB genomic region (domain center/boundary region) into two 500 KB adjacent sub-regions, predicted the 3D structure of each sub-region by BACH, and then inferred the spatial arrangements of the two sub-structures. Both BACH and BACH-MIX were applied to the 40 KB resolution Hi-C contact matrices.
Among all the possible spatial arrangements of two adjacent genomic regions, we defined the effective structures as those with their posterior mean proportions greater than 5%, and report the number of effective structures at each locus. A locus with a smaller number of effective structures exhibits lower structural variations than a locus with a larger number of effective structures. In the HindIII sample, the average number of effective structures is 2.20 for the domain center regions, and 2.82 for the domain boundary regions (Figure S3A, two sample t-test p-value<2.2e-16). Similarly, in the NcoI sample, the average number of effective structures is 2.07 for the domain center regions, and 2.54 for the domain boundary regions (Figure S3B, two sample t-test p-value = 5.2e-13). We changed the threshold for the effective structure to 10% and 1%, and observed consistent results (Figure S3 and Table S3). These results suggest that domain center regions exhibit lower structural variations than domain boundary regions.
Figure 2 shows two illustrative examples in the HindIII sample, one for the domain center region (Chromosome 2, 117,580,000∼118,580,000, Figure 2A), and one for the domain boundary region (Chromosome 1, 135,540,000∼136,540,000, Figure 2B). Under threshold 5%, BACH-MIX identified one effective structure for the domain center region with proportion 99% (Figure 2C), and three effective structures for the domain boundary region, with proportions 77% (Figure 2D), 14% (Figure 2E) and 8% (Figure 2F), respectively.
Next, we evaluated the relationship between structural variations of topological domains and its functional forms. We divided the 562 selected domain center regions into two groups, regions with high structural variations (i.e., containing multiple effective structures, threshold = 5%) and regions with low structural variations (i.e., containing one effective structure, threshold = 5%), and compared the genomic and epigenetic features between these two groups (Table S4). We observed significant enrichment of gene density, transcription elongation histone modification mark H3K36me3, repressive histone modification mark H3K27me3, promoter mark H3K4me3, RNA polymerase II, accessible chromatin and early replicated genomic regions in regions with high structural variations, and significant enrichment of heterochromatin marks H3K9me3, H4K20me3 and genome-nuclear lamina interaction in regions with low structural variations. Noticeably, we did not observe statistically significant association between gene expression levels and structural variations. These results suggest that gene rich, accessible and early replicated chromatins are more likely to exhibit multiple structural configurations than gene poor, inaccessible and late replicated chromatins.
Although it is widely accepted that the chromatin structure is highly dynamic, it is unclear whether the cell population contains one dominant chromosomal structure, or multiple distinct chromosomal structures with comparable mixture proportions. To quantify structural variations of the whole chromosome in the cell population, we designed the following two-step procedure. In the first step, we applied BACH to the whole chromosome scale Hi-C contact matrix and obtained a predicted 3D chromosomal structure (the mode of the first BACH posterior distribution, referred to as ). Then, we computed the expected Hi-C contact matrix based on this predicted structure . In the second step, we defined the residual matrix as the difference between the original Hi-C contact matrix and half of the expected Hi-C contact matrix, and applied BACH again to the residual matrix to obtain another predicted 3D chromosomal structure (the mode of the second BACH posterior distribution, referred to as ). In order to avoid the possibility of algorithmic artifacts, we ran 100 parallel chains for our two-step procedure using a large variety of initial structures and chose the structures with the highest posterior probabilities.
If there exists a dominant chromosomal structure (referred to as ) in the cell population, we will expect that and are close to each other, since is still the dominant chromosomal structure in the residual matrix. On the other hand, if there is no such dominant chromosomal structure in the cell population, we will expect that and are quite different from each other since the original contact matrix and the residual matrix should have little in common. In practice, the similarity between and can be measured by the normalized root mean square deviations, i.e., RMSD() (Methods). Simulation results (Text S1, Figure S4 and Table S5) confirmed that both and are close to (which also means that RMSD() is small) if is indeed the dominant chromosomal structure.
In practice, however, we need a reference probability distribution in order to claim that the observed RMSD() is small enough. Previous studies [35], [36] have shown that the random walk backbone model can be used to approximate the chromatin 3D structure. In this work, we use the empirical distribution of the RMSD between two 3D structures independently generated from the random walk scheme as the reference distribution to judge whether an observed RMSD() is small enough (Text S1). If the observed RMSD() falls within the lower 5% of the reference distribution, we claim that and are close enough to each other.
We applied the above two-step procedure to the real Hi-C data to generate 3D chromosomal structure for each mouse chromosome by treating each topological domain as a basic unit. Figure S5 lists the alignment of two 3D chromosomal structures BACH predicted in the two stages, and , from 20 mouse chromosomes in both HindIII sample and NcoI sample. Tail probabilities of RMSD for each chromosome are reported in Table S6. Figure S6 displays the box plots of the twenty RMSD empirical distributions, each corresponding to that between two independently generated random walks of the same length as each mouse chromosome. We found that in long chromosomes (chr 1 to chr 14 and chr X), and are similar (i.e., RMSD(,) is small, within the tail probability<0.05), suggesting the existence of a dominant 3D chromosomal structure in the cell population. It is worth noting that all these long chromosomes adopt helical structures (Figure S7A), which is unlikely to be coincidental. For short chromosomes, however, RMSD(,) is comparable to that of two independently simulated random walks (tail probability ≥0.05). We conducted similar analysis at different resolution scales by treating two domains or half of a domain as a basic unit, for both the HindIII sample and the NcoI sample. The results were almost identical to the original analysis (Text S1). These results suggest that the whole chromosome scale 3D modeling could be meaningful, especially for long chromosomes (chr 1 to chr 14 and chr X). We did not obtain consistent overall structures in the two-step procedure for short chromosomes. It is likely that such inconsistencies are caused by a lack of “leveraging” information of the Hi-C data when a chromosome is short. By further examining the differences between the two structures obtained by our two-step procedure for these short chromosomes, we observed that the large RMSD is caused by the existence of a few mirror reflections of local structures, implying that, although the local structures can be determined rather well in these chromosomes, there is not enough information to pin down the orientation of these local parts.
To further understand why shorter chromosomes appeared variable in our two-step procedure at the whole chromosome level, we also conducted a local-level structural comparison. In detail, we used a sliding window of ten domains to scan along each chromosome. For each local region of a chromosome covered by the sliding window, we evaluated the structural similarity between and locally (Figure S8), resulting in K - 9 RMSDs for each chromosome, where K is the number of domains of the corresponding chromosome. Now, for all the 20 chromosomes, we found that the local structures in and are significantly more similar than those generated from the random walk scheme. More precisely, the distribution of the K - 9 RMSDs for each chromosome is significantly and stochastically smaller than that generated from the random walk scheme (Figure S8), supporting the existence of a dominant structure in the cell population for all chromosomes, at least at a relatively local level (about 10 MB).
A competing method, MCMC5C, has been proposed to generate whole chromosome level 3D models for the human chromosomes [16]. This method, however, does not correct the systematic biases in the Hi-C data. Here we compared whole chromosome level 3D models produced by BACH and MCMC5C for the mouse chromosomes. We used BACH and MCMC5C to generate spatial models of each long chromosome (chr 1 to chr 14 and chr X) by treating each topological domain as a basic unit (Figure S7). The 3D chromosomal structures predicted by BACH from the HindIII sample and NcoI sample are significantly more consistent (measured by RMSD) than those predicted by MCMC5C (paired t-test p-value = 1.4e-7). A modified BACH algorithm without bias correction also outperformed MCMC5C (Text S1). We also conducted the same analysis using a published human Hi-C dataset [9] and found that BACH consistently outperformed MCMC5C (data not shown). The significant improvement of BACH over MCMC5C is likely due to the fact that BACH explicitly integrates the correction of known systematic biases [17], and the Poisson model used in BACH fits the count data of the Hi-C experiment better than the Gaussian model used in MCMC5C. Since other published 3D reconstruction methods do not provide stand-alone software, we were not able to conduct similar comparative studies for them.
We applied the BACH algorithm to the whole chromosome Hi-C contact matrix, and obtained the predicted 3D chromosomal structures for the 15 long chromosomes (chr 1 to chr 14 and chr X). We first investigated how compartments labeled “A” versus those labeled “B” are distributed spatially in the whole chromosome model. Among all the 1,835 topological domains in chr 1 to chr 14 and chr X, 848 belong to compartment A, 633 belong to compartment B, and the remaining 354 straddle domains contain genomic regions from both compartment A and compartment B. For each 3D chromosomal model that BACH predicted, we fitted a plane through the straddle domains using the least square method, and then counted the numbers of topological domains belonging to compartment A and compartment B, respectively, at each side of the fitted plane. The results can be represented by a two-by-two contingency table. Fisher's exact test was then used to measure the magnitude of spatial separations between two types of compartments. Among the 15 selected mouse chromosomes (chr 1 to chr 14 and chr X), we found that the compartment label (A or B) of topological domains is significantly correlated with the spatial location of these domains relative to the fitted plane (on the left side or on the right side) in 14 chromosomes in both HindIII sample and NcoI sample (Table S7). As shown in Figure 3A, topological domains with the same compartment label tend to locate on the same side of the structure, consistent with their interaction frequencies, and the observation that compartment B tends to be associated with nuclear membrane [37], [38].
We further study how genomic and epigenetic features are distributed spatially in the whole chromosome model. Similar to the previous analysis for compartment labels (A or B), we conducted the same analysis for each of the eleven genomic and epigenetic features in consideration (Table S7). We used 33rd and 67th percentiles as the thresholds and divided all the 1,835 topological domains in chr 1 to chr 14 and chr X into three groups: domains with low value, with median value, and with high value of a particular feature. For each 3D chromosomal structure BACH predicted, we fitted a plane through domains with median value of the feature using the least square method. Next, we used the Fisher's exact test p-value to measure the magnitude of association between the group label (low value group or high value group) and spatial location of topological domains relative to the fitted plane (on the left side or on the right side). Table S7 lists the number of chromosomes with significant spatial separation patterns for each genomic and epigenetic feature in both HindIII sample and NcoI sample (threshold for Fisher's exact test p-value is 0.05). We observed that the gene density, transcription elongation histone modification mark H3K36me3, repressive histone modification mark H3K27me3, promoter mark H3K4me3, RNA polymerase II, chromatin accessibility, DNA replication time, heterochromatin marks H3K9me3 and H4K20me3 and genome-nuclear lamina interaction of topological domains are significantly associated with the spatial location of topological domains relative to the fitted plane (on the left side or on the right side) among more than nine chromosomes (Table S7 and Figure 3B∼Figure 3L).
We have described BACH and BACH-MIX, two Bayesian statistical models, to study 3D chromosomal structures and structural variations of chromatins from the Hi-C data. The benefits of using a probabilistic approach are two-folds: first, rigorous statistical inference can be carried out to properly remove systematic biases and account for observational noise sources; second, sequencing depth variations can be explicitly modeled by Poisson distributions. Our results demonstrate that BACH is significantly more reproducible and achieves higher consistency with the FISH data than an existing algorithm (MCMC5C). Application of BACH to a recently published Hi-C dataset from the mouse ES cells reveals interesting structural properties of mammalian chromosomes. Specifically, we found that geometric shapes of topological domains are strongly correlated with several genomic and epigenetic features. For example, gene rich, actively transcribed, accessible and early replicated chromatins tend to be more elongated than gene poor, lowly transcribed, inaccessible and late replicated chromatins. Furthermore, by using a variant BACH-MIX algorithm, we found that structural variations of a chromatin are also correlated with several genomic and epigenetic features.
There are several issues that we have not addressed in this paper, such as biophysical properties of chromatin fiber [39], [40] and the low sequencing depth of inter-chromosomal chromatin interactions. In principle, biophysical properties can be accommodated directly in our Bayesian model as spatial constraints through an informative prior on spatial distances. With more experimental work and additional data, the BACH and BACH-MIX algorithms can be applied to study the spatial arrangement of multiple chromosomes simultaneously. With the rapid accumulation of high throughput genome-wide chromatin interaction data, the BACH and BACH-MIX algorithms could be valuable tools for understanding higher order chromatin architecture of mammalian cells.
To reconstruct the underlying consensus 3D chromosomal structure, we develop the following probabilistic model, similar to the “beads-on-a-string” model (Figure S9) that has been used extensively in chemistry. The genomic region of interest is divided into consecutive, disjoint loci of equal size , and each locus is represented by a bead in the 3D space, whose location is given by the Cartesian coordinates . The Euclidean distance between loci and represents the average spatial distance between these two loci and :Under this representation, reconstructing the 3D chromosomal structure is equivalent to placing these beads in the 3D space, i.e., specifying the Cartesian coordinates of these loci.
Let be the symmetric contact matrix generated by the Hi-C experiment, where each entry represents the number of paired-end reads spanning two loci and . The variations of can be explained by several factors. Lieberman-Aiden et al. [9] first reported the negative association between the number of paired-end reads spanning two loci () and the corresponding spatial distance (). Recently, Yaffe and Tanay [17] identified some systematic biases, including restriction enzyme cutting frequencies, GC content and sequence uniqueness of fragment ends, which substantially affect Hi-C data. Taking all these unique features into consideration, we propose the following Poisson model.
Let , and represent the number of fragment ends within locus the mean GC content of fragment ends within locus , and the mean mappability score of fragment ends within locus , respectively [17]. We assume that the off-diagonal count in the contact matrix follows a Poisson distribution with rate where:In this model, measures the magnitude of negative association between and . and are the coefficients for the enzyme effect, GC content effect and mappability effect, respectively. The link function in this Poisson model provides the relationship between the linear predictors (i.e., the spatial distance, the number of fragment ends, the mean GC content of fragment ends and the mean mappability score of fragment ends) and the mean of Poisson distribution, which can be used to translate the number of paired-end reads spanning two loci into the average spatial distance between them.
Let ( matrix) represent the Cartesian coordinates of the loci of interest, and let be the collection of all nuisance parameters. The joint likelihood is of the form:We adopt a fully Bayesian approach with non-informative priors for all model parameters, and obtain the following joint posterior distribution:Due to the high dimensionality of the parameter space, designing an efficient computational tool to draw samples from is essential for the statistical inference of our model. To achieve this goal, we propose a three-stage statistical inference procedure (Figure S10). First we assign initial values for the nuisance parameters using a Poisson regression approach [41]. We then use sequential importance sampling (SIS) [42] to generate an initial 3D chromosomal structure. At the end, we apply Gibbs sampler [43] with hybrid Monte Carlo [21], [44] and adaptive rejection sampling (ARS) [45] to further refine the 3D chromosomal structure and the nuisance parameters. More details of three-stage statistical inference procedure can be found in Text S1.
Let represent the Cartesian coordinates of the genomic region with loci, where First we shift the genomic region such that its weight center is at the original point . We then conduct the principle component analysis on the by 3 matrix , and rotate matrix to matrix , such that the x-axis is the direction of the first principle component (the one explains most variability) and the y-axis and the z-axis are the directions of the second and the third principle components, respectively. We use a cylinder to approximate the 3D chromosomal structure of the genomic region . The height of the cylinder is defined as the difference between the 90% quantile of and the 10% quantile of . The radius of the cylinder is defined as two times the median of We further define HD ratio of the genomic region as the ratio between the height of the cylinder and the diameter of the cylinder, and then normalized by the size of genomic region . By the definition, genomic regions with higher HD ratios are more elongated.
We propose the BACH-MIX algorithm to study the spatial arrangement of two adjacent genomic regions. Here we assume that each genomic region exhibits a unique consensus 3D chromosomal structure, but the spatial arrangement of two adjacent genomic regions has certain level of flexibility, and varies according to a probabilistic distribution. More precisely, let and represent the 3D chromosomal structures of two adjacent genomic region and , respectively, where The spatial arrangement of the genomic region with respect to the genomic region is determined by three Euler angles [23] and an index for mirror symmetry. Let be the collection of these four parameters, and define the rotation matrix and the mirror symmetry matrix as:The spatial arrangement of the genomic region with respect to the genomic region , denoted can be calculated by:Therefore, each corresponds to a 3D chromosomal structure of two adjacent genomic regions and , and a probabilistic distribution defines a mixture of distinct spatial arrangements between the two adjacent genomic regions and . To further simplify the statistical inference problem on , we discretize the four dimensional space of , and use a multinomial distribution to approximate .
Let be the by dimensional contact matrix for inter-region chromatin interactions, where represent the number of reads spanning the th locus in the genomic region and the th locus in the genomic region We assume that follow Poisson distribution with rate whereHere is the Poisson offset for the spatial arrangement , which is proportional to . The statistical inference problem on the multinomial distribution is equivalent to infer . is the spatial distance between the th locus in the genomic region and the th locus in the genomic region with rotation matrix and mirror symmetry matrix . , and are local genomic features which follow the previous definitions. The joint likelihood is of form:We adopt a fully Bayesian approach with non-informative priors for all model parameters, and obtain the following joint posterior distribution:We use hybrid Monte Carlo to jointly update the parameters (Figure S10). The first order partial derivatives with respect to is of the form:
Assuming and are the Cartesian coordinates of two genomic regions and , respectively, where and We first remove the scaling effect by a regression procedure. Let and be the Euclidean distance between loci and in and , respectively. We regress against and obtain the slope . Define where Assume has the singular value decomposition and then the optimal rotation matrix can minimize the sum of square error [46]. The normalized RMSD is defined as:Empirically, normalized RMSD less than 0.1 indicates high similarity, normalized RMSD between 0.1 and 0.2 indicates moderate similarity, while normalized RMSD larger than 0.2 indicates low similarity.
Under the default setting of BACH, we draw 100 3D chromosomal structures at each step of sequential importance sampling. We further enrich each 3D chromosomal structure ten times when we implement the rejection control technique. In the Gibbs sampler of BACH and BACH-MIX, we run three parallel chains with 5,000 MCMC iterations in each chain. The first 1,000 samples are dropped as the burn-in stage, and then every 50th sample in the last 4,000 samples are used for the posterior inference. We use the Gelman-Rubin statistic [43] to measure the mixing of three parallel chains. Empirically, the Gelman-Rubin statistics less than 1.1 indicates that three parallel chains converge to the same posterior distribution.
The computation time of BACH and BACH-MIX depends on the number of MCMC iterations and the number of loci in the genomic region of interest. All MCMC calculations are conducted on computing nodes in Harvard Linux cluster “Odyssey”, each with dual Xeon E5410 2.3 GHz quad core processors and 32 GB RAM. Under the default setting, BACH takes 81 seconds to predict a 3D chromosomal structure with 25 loci; BACH-MIX takes 8 minutes to predict the proportion of 104 distinct 3D chromosomal structures for two 13 loci adjacent genomic regions. The computation time increases almost quadratically with the number of loci in the genomic region of interest.
BACH and BACH-MIX can be freely downloaded at http://www.fas.harvard.edu/~junliu/BACH/.
|
10.1371/journal.pcbi.1006320 | Integration and multiplexing of positional and contextual information by the hippocampal network | The hippocampus is known to store cognitive representations, or maps, that encode both positional and contextual information, critical for episodic memories and functional behavior. How path integration and contextual cues are dynamically combined and processed by the hippocampus to maintain these representations accurate over time remains unclear. To answer this question, we propose a two-way data analysis and modeling approach to CA3 multi-electrode recordings of a moving rat submitted to rapid changes of contextual (light) cues, triggering back-and-forth instabitilies between two cognitive representations (“teleportation” experiment of Jezek et al). We develop a dual neural activity decoder, capable of independently identifying the recalled cognitive map at high temporal resolution (comparable to theta cycle) and the position of the rodent given a map. Remarkably, position can be reconstructed at any time with an accuracy comparable to fixed-context periods, even during highly unstable periods. These findings provide evidence for the capability of the hippocampal neural activity to maintain an accurate encoding of spatial and contextual variables, while one of these variables undergoes rapid changes independently of the other. To explain this result we introduce an attractor neural network model for the hippocampal activity that process inputs from external cues and the path integrator. Our model allows us to make predictions on the frequency of the cognitive map instability, its duration, and the detailed nature of the place-cell population activity, which are validated by a further analysis of the data. Our work therefore sheds light on the mechanisms by which the hippocampal network achieves and updates multi-dimensional neural representations from various input streams.
| As an animal moves in space and receives external sensory inputs, it must dynamically maintain the representations of its position and environment at all times. How the hippocampus, the brain area crucial for spatial representations, achieves this task, and manages possible conflicts between different inputs remains unclear. We propose here a comprehensive attractor neural network-based model of the hippocampus and of its multiple input streams (including self-motion). We show that this model is capable of maintaining faithful representations of positional and contextual information, and resolves conflicts by adapting internal representations to match external cues. Model predictions are confirmed by the detailed analysis of hippocampal recordings of a rat submitted to quickly varying and conflicting contextual inputs.
| Following the discovery of place cells, which specifically fire at determined positions in space [1], the hippocampus was recognized as an essential brain area for spatial representations and memories. These cognitive representations, or maps, actually code for more than position in physical space, and are also strongly informative about context [2], including physical features of the background, such as visual landmarks, light, odors, auditory stimuli, as well as more abstract conditions, such as the emotional state or the task to be performed [3–8].
A fundamental property of the hippocampus is its capacity to memorize multiple cognitive maps [1, 9–11]. This property may result from specific recurrent synaptic connectivity in the hippocampal CA3 region [12, 13], and can be theoretically understood in the framework of continuous attractor neural networks (CANN) [14, 15]. Thanks to the remapping properties of place cells, multiple maps can be memorized in the same connectivity matrix with almost no interference between them [16–19].
Cognitive maps may be retrieved when the animal explores again the corresponding environments, or be quickly and intermittently recalled depending on the most relevant behavorial information at that moment [20]. Different sources of inputs to the hippocampus concur to form, recall, and dynamically maintain cognitive maps [21]. Changes in visual cues and landmarks may substantially affect place field shape and positioning [5]. The Path Integrator (PI), capable of integrating proprioceptive, vestibular and visual flow inputs and possibly supported by the grid-cell network in the medial-enthorinal cortex (mEC) [22], allows the animal to update the neural representation during navigation [23]. The path integrator is itself sensitive to other sources of inputs, and undergoes reset in case of large disagreement with external landmarks or sensory information [24].
Insights about how these different inputs contribute to hippocampal representations were recently obtained by studying the effects of mismatches between path-integration and visual sensory information, in particular in virtual reality settings [25, 26]. In another study Jezek et al showed how abrupt changes in visual context (light conditions) during active exploration by a rodent resulted in fast flickering between context-associated maps in CA3 on the theta time scale [27] (Fig 1A). Though they are largely artificial, these conditions offer a rare window on the fast dynamics of the place-cell population, and on how this dynamics is shaped by the various inputs.
Despite these studies, how contextual and PI inputs are combined by the hippocampal network to produce cognitive maps and accurate positional encoding is not fully understood yet. In this work, we carefully reanalyze and model the experiment of Jezek et al to address this issue. We first introduce of a dual inference method capable of extracting reliably and independently the encoded map [28] and the encoded position [29, 30] from the recorded spiking activity alone (Fig 1B). Our dual decoder allows us to robustly show that the hippocampal activity always encodes the correct location in the retrieved map, even during the fast, unstable dynamics of the cognitive maps, as put forward in [27]. To explain this robust encoding, we propose a CANN model of the hippocampal circuitry, capable of storing multiple cognitive maps; the model is fed by visual-cue and path-integration inputs projecting on the place-cell populations supporting those maps [31]. The path integrator is, in turn, influenced by the hippocampal activity, closing an interaction loop between the hippocampus and the mEC [32]. Our model not only reproduces the flickering phenomenology and the stable encoding of position, but also makes several precise predictions on the dynamics of cognitive maps, the relative strength of inputs, and the intricate activation of place-cell populations supporting the two maps. These predictions are corroborated by a further detailed analysis of Jezek et al’s data. Our work therefore proposes explicit mechanisms by which the hippocampus could be capable of encoding various contextual and self-locomotion information in multi-dimensional representations, and of updating them accurately on fast time scales.
Jezek et al trained a rodent in two environments (square boxes), equal in size and shape, but differing by their light conditions [27]. A population of 34 CA3 place cells was recorded during reference sessions with fixed light conditions, and shown to define environment-specific maps, denoted by A and B. In a subsequent test session, taking place in a single box, instantaneous switches between environmental light conditions triggered the instability of the recalled cognitive map, which flickered back and forth between the two corresponding environments.
The neural activity st of the population in any theta cycle t encodes information on the context (the set of rules that connects position to activity, i.e. the place fields defining the cognitive map) as well as on the specific position within the environment (Fig 1A). We first introduce a dual decoder, able to independently infer the cognitive map and the position, at high temporal resolution. By comparing the inferred position to the true animal location, we then assess how precisely the position is represented in the population activity, irrespectively of the cognitive map in which it is neurally encoded (Fig 1B).
Due to the global remapping properties of CA3, the intensities and mutual superpositions of place fields are specific to each environment (Fig 1A). Consequently, the average firing rates and pairwise correlations of the place-cell population define a fingerprint of the corresponding cognitive map [33, 34]. We use the reference session recordings in each environment m (A or B) to compute this fingerprint statistics. We then build a model Pm(s) that approximates the probability of observing the neural activity s when the cognitive map m is recalled. This model relies on the inference of a functional network of couplings between the place cells, reproducing the fingerprint statistics of map m [34, 35] (Methods).
Given the activity st recorded in theta bin t during the test session, we then compare the two probabilistic models PA and PB to estimate which map m is more likely to have generated st. The log-ratio
Δ L ( s t ) = log [ P A ( s t ) P B ( s t ) ] (1)
indicates whether the neural activity st is more similar to the neural patterns encountered in map A than to the ones of map B (large and positive Δ L), or typical of B and not of A (large in absolute value and negative Δ L). Comparing Δ L to a statistical significance threshold allows us to infer the map mt (Methods). If the decoded map mt is discordant with the imposed light conditions the theta bin is identified as a flicker.
As a control, we check that Δ L is mostly positive in reference sessions for environment A and negative for B, see Fig 2A. Applying the decoder to the test session, we observe the presence of flickers, see Fig 2B (yellow bins); flickers were first found in [27] with correlation-based methods requiring knowledge of the true position of the animal. An analysis of the temporal correlation of these flickers reveals that they typically persist over ∼ 6 theta bins (Methods and Fig 2C); hence, cognitive maps show some inertia extending beyond the theta scale.
To assess if the fast dynamics of cognitive maps affects the quality of positional encoding we next re-use the neural activity pattern st in theta bin t, this time to infer the position of the animal. A naive Bayesian decoder [29, 30] takes as an input the above-decoded map mt (Fig 1B) and uses its place fields to estimate the position. The distance between the inferred position, r ^ t, and the true position, rt, defines the positional error ϵt. As shown in Fig 3A, the positional error ϵt (blue line) is independent of the time elapsed after the light switch, and has a value comparable to the one obtained in fixed-environment conditions (blue dashed line). This result crucially depends on the fact that position is estimated according to the decoded map mt, which varies with time t. For comparison, in Fig 3B we show the error if we decode the position according to the new, post-switch map (green line) or to the old, pre-switch one (red line) at all times. Both procedures result in similar, higher errors right after the light-switch, where flickers are frequent. The error with the post-switch map eventually decrease to fixed-environment value after few seconds, due to the rarity of flickers long after the light switch.
In summary, the output of our map decoder, mt, can be interpreted as the correct cognitive state to read the positional code, see Fig 2A. Even in the presence of fast dynamical flickers of the cognitive map, the location of the animal is robustly and coherently represented at all times. Our findings show that the hippocampus representation encodes both positional and contextual information in an independent and accurate way. Interestingly, the positional error computed with the map opposed to the decoded one (orange line in Fig 3A) shows a significant reduction in the first seconds after the light switch; this non-trivial effect will be explained in detail in the next sections.
The findings above suggest that the stream of positional information to the hippocampus is maintained despite the presence of rapid changes of cognitive representations following the abrupt modification of visual cue (V) after the light switch. A natural hypothesis is that the path-integrator (PI) sends to the hippocampus information relative to the position in the ‘old’ map [11, 31], competing with the visual cue input associated to the ‘new’ map. To formalize this assumption, we introduce a continuous attractor neural network (CANN) model that contains the minimal ingredients to understand the effect of conflicting PI and V stimuli onto the hippocampal activity. In the CANN paradigm for memory storage and retrieval of cognitive maps [15–17, 36], the animal location at a certain time is represented as a self-sustained bump of neural activity. The bump is localized in the current position within a two-dimensional manifold, where place cells are embedded according to the positions of their place field centers in the real environment. We generalize this classical model by including two informational inputs on the memory network, from allothetic (visual cues) and idiothetic (path integrator) stimuli. The proposed interaction model is composed of the following four ingredients, see Fig 4A and Methods:
From a functional point of view, the CANN model mostly behaves, for a fixed position r of the rodent, as an effective two-state model for the hippocampal activity, as sketched in Fig 4B. These two states correspond to the activity localized in map A or B; their probabilities are controlled by the intensity of, respectively, the path-integrator and visual-cue inputs. Note that the emergence of two well separated collective states from the microscopic CANN model is intrinsically due to the presence of recurrent connections, see Fig 4B; A characterization of the effective barrier between the states is reported in S1 Text (see Fig. A in S1 Text). The height of the barrier, controlled by the parameter γJ, and the amount of stochasticity in the individual neural dynamics are crucial ingredients to determine the dynamics of the model. In particular, these variables control the time-correlation of flickers (Methods); For the chosen simulation parameters, the time correlation decays over ∼ 7 theta bins (Fig 4C) in accordance with data (Fig 2C).
The typical outcome of a simulated experiment is shown in Fig 4D. During the exploration phase preceding the light switch, the visual (b) and path-integrator (c) inputs jointly contribute to the stability of the internal representation of the position. A localized bump of activity, sustained by the recurrent connections (a), can be observed in the pre-switch map (Fig 4A, left), say, m = PI = V = A. Right after the switch, the hippocampal network receives conflicting streams of information: PI = A differs from V = B. The path integrator is still activating place cells coding for the current position of the animal in the ‘old’ map, while the visual stream points to neurons coding for the same position in the ‘new’ map (Fig 4A, center). This results in a conflict between the two bump representations, which are mutually incompatible due to the orthogonality of global remapping. Flickering is produced as an alternance between these two possible states, m = PI = A and m = V = B. During this conflicting phase (Fig 4D, red region), the feedback (d) from the memory network to the path integrator tries to achieve coherence between the hippocampal and path-integrator states. When the bump is in the visually-driven, post-switch map (m = B), incoherence is strong, and the path integrator is more likely to be reset. Realigning the path integrator state with the external cue, PI = V = m = B, brings the conflict phase to an end, and the hippocampal state reaches stability (Fig 4A, right).
Despite its conceptual simplicity, the model shows a rich phenomenology and reproduces in a strikingly-accurate manner the results of the analysis of the CA3 teleportation recordings. In Fig 4D we show a representative time trace of the log-ratio Δ L (Eq (1)) in a simulated teleportation session. Alternate intervals of positive and negative Δ L signal the presence of map instability, as in [27], following the light switch (red vertical lines) and the path-integrator realignment (green vertical lines). Applying to the simulated data the same positional-error analysis as for the recorded data (Fig 3A&3B), we observe the same qualitative picture, see Fig 3C&3D. In particular, position is coherently encoded in the recalled cognitive map at all times.
Our model predicts that (1) the duration of the conflict phase, i.e. the time elapsed from a light switch to the subsequent PI realignment, is exponentially distributed (Fig 5A, right panel); (2) during the conflict phase, the flickering frequency i.e. the percentage of theta bins identified as flickers, is constant and independent of time (Fig 5B, right panel).
In order to test these two predictions on CA3 recordings, we introduce a method to disentangle the flickering dynamics of the cognitive map and the realignment of the PI, the latter bringing an end to the former. We first infer the most likely PI-realignment time for each light-switch event, given the sequence of identified flickers (Methods). The outcomes are shown as green lines in Fig 5D, and correctly separate conflicting phases (rich in flickering events) from coherent periods (during which the hippocampal representation is much more stable). The distribution of conflicting phase durations is approximately exponential in agreement with model prediction (1), with decay time τ = 53 theta bins (Fig 5A, left panel). Dividing the test session into conflicting and coherent phases, we compute the frequency of flickers in the conflicting phase only. Consistently with the model prediction (2), the frequency of flickers is independent of the delay after the switch, with about 60% of theta bins in the conflicting phase carrying flickers (Fig 5B, left panel).
Similar frequencies of flickers, close to one half, are obtained in the model when the two inputs have comparable strengths (γPI ≃ γV in Fig 4B, see also Fig. C in S1 Text). A testable consequence of this balance is that the distributions of the sojourn times (durations of the periods in which the neural activity persists in a cognitive map, see Methods) in map A and in map B are similar. This prediction is confirmed by a further analysis of the CA3 recordings: the two distributions of the sojourn times are both exponential, with roughly the same decay times (see Fig. E in S1 Text). This common time scale is related to the correlation time of the flickers (Figs 2C & 4C), see Methods.
The combination of properties (1) and (2) explains the exponential decay in the frequency of flickers with the delay after the switch reported in [27] and [38]. While the frequency of flickers is constant and large in the conflicting phase, and constant and very low in the coherent phase, the duration of the conflicting phase is exponentially distributed. Hence the frequency of flickering theta bins, irrespectively of the phase, shows the same exponential decay, see Fig 5C (right panel: simulated experiment, left panel: analysis of CA3 recordings). A detailed analysis of the data provides overwhelming statistical support to our two-fold explanation compared to a simple exponential decay of the flickering frequency (logarithmic likelihood-ratio test ∼ 150, see Methods).
Our model allows us to better understand the subtle differences between the neural encodings of position in the conflicting and the coherent phases. In the latter phase, both path-integrator and visual inputs point to the neurons with place fields overlapping the rodent position r in map. During the conflicting phase, the two inputs excite the two place-cell populations centered in r in their respective maps, respectively, m = PI and m = V. Hence, while the bump of activity is mostly localized in one of the two maps (varying over time), some dispersion may be expected due to these incoherent inputs.
Mixed activity states, in which two (distinct) populations of neurons encoding the same position in the two maps are active, can be occasionally observed in the snapshots of the simulated activity in Fig 4D, e.g. around theta bin t = 80. The overdispersion present during the conflicting phase has two consequences. First, the accuracy in position encoding is expected to be lower in the conflicting phase that in the coherent phase, see Fig 6A, left panel. Secondly, the loss in accuracy is not due to some random noise in the neural activity, but to a transient bump-like activity in the ‘wrong’ map, opposite to the decoded one. This effect is clearly seen when we choose the opposite map to infer the rodent position. While this choice leads to very poor prediction during the coherent phase, the positional error is significantly reduced during the conflicting phase (Fig 6B, left panel).
To test these two predictions in CA3 recordings, we combine our positional analysis and our PI-realignment time inference procedure. In Fig 6A (right), we compare the distributions of positional errors computed with the decoded map (according to the sign of Δ L) during conflicting and coherent phases (blue and red, respectively). Consistently with the model predictions, the positional error is significantly increased during the conflicting phase (ANOVA p < 8 × 10−8; conflicting: 14.7 ± 0.5 SEM, coherent: 12.3 ± 0.1 SEM). When computed with the opposite map, the positional error is obviously much higher than its counterpart computed with the decoded map, but a substantial decrease is found in the conflicting phase compared to the coherent phase, see Fig 6B, right panel (ANOVA p < 5 × 10−23; conflicting: 27.8 ± 0.6 SEM, coherent: 34.2 ± 0.2 SEM), in full agreement with the model prediction. This effect also explains the relatively low value of the positional error obtained with the opposite map right after the switch, i.e. deep into the conflicting phase, compared to later times, see Fig 3A&3C. While this phenomenology is clear, it could in principle be affected by the presence of visual inputs projecting onto place cells during flickering events, i.e. when the ‘opposite’ map agrees with the external cues. In order to analyze the effect of the path integrator alone, we have restricted the analysis to theta bins whose decoded maps agreed with the visual inputs, i.e. to non-flickering theta bins. Results, shown in Fig 6C&6D, are still statistically significant and in strong agreement with the model predictions (decoded map: ANOVA p < 1 × 10−12; conflicting: 17.1 ± 0.7 SEM, coherent: 12.3 ± 0.2 SEM; opposite map: ANOVA p < 1 × 10−7; conflicting: 28.9 ± 0.95 SEM, coherent: 34.2 ± 0.3 SEM). Our findings are robust against changes in the statistical threshold L0 for map decoding in the identification of conflict/coherent phases (Methods), see Fig. I in S1 Text.
The over-dispersion of the neural bump during the conflicting phase can also be observed from the reduction in (the absolute value of) the log-ratio, | Δ L |, see Eq (1). This quantity can be interpreted as a proxy for the completeness of the bump in one single map (Methods), larger | Δ L | corresponding to large bumps in either of the two maps and randomly scattered activity in the other map (Fig 4A&4D). We find that the absolute value of Δ L is significantly reduced during the conflicting phase in CA3 data, see Fig 6E (left panel, ANOVA p < 10−15; conflicting: 5.51 ± 0.16 SEM, coherent: 7.19 ± 0.08 SEM). Fig 6F shows the bimodal nature of the distributions of Δ L in the conflicting and coherent phases. While the reference and coherent-phase distributions coincide, the conflicting-phase distribution is more narrow, due to the overdispersion of the bump (reference σ2 = 80.7, coherent σ2 = 79.2, conflict σ2 = 47.3). This result provides further evidence for the predictive power of the CANN model.
Our statistical inference-based data analysis allows us to quantify how well the CA3 neural activity encode various cognitive maps, and the position therein. Correlation-based procedures, e.g. used in [27], decode the cognitive state by comparing the instantaneous population activity to the average activity recorded in reference sessions at the same position of the rodent. Our functional-network based map decoder, instead, relies on the fact that the joint pairwise spiking activity of neurons is a fingerprint of the cognitive map [28, 34]. It does not need any knowledge of the sensory correlate (here, position), and could be used to decode generic brain states in other areas.
The fast dynamics of cognitive maps studied in [27] and here results from an unrealistic sensory situation. Imposing artificial conflicts between inputs and studying their consequences is a standard approach to unveil the circuitry underlying the processing of multimodal sensory information in the hippocampus [25, 26] as well as in other brains areas, see for instance [39] for an illustration in the primary visual cortex where mismatches involve sensory and motor inputs. However, fast retrieval of functionally relevant maps, characterized by grouping and cognitive control, has also been observed in realistic settings, in which a behaving animal is required to maintain representations of two distinct spatial frames [20].
The position of the animal was accurately inferred at all times from the spiking activity using the place fields of the retrieved cognitive map (Fig 1B). As a main finding, we show that the hippocampus maintains high-quality encoding of the position even if the contextual variable undergoes fast dynamical changes. This is explained in the model by the fact that inputs point to place cells coding for the physical position in both competing maps (Fig 4A), and that the bump of activity is most often localized around these place cells in one map, and scattered all over the other map. Similar findings were reported in main text Fig 3d and Supplementary Information Fig 8 within [27], within a statistical framework assuming a priori the consistency of positional representation during flickering events, as the cognitive map was decoded by comparing the neural activity to the mean-activity vectors at the recorded real position of the rat. The emergence of unambiguous, non-mixed representations was also underlined in [27], and shown to take place in the second half of the theta cycle. However, the detailed analysis of the CA3 recordings and of the model data shows a loss of quality of the bump state (reduction in absolute value of log-ratio | Δ L |) and an increased quality of position decoding in the opposite map (Fig 6), providing evidence for the presence of partially mixed states.
Our model for the retrieval of hippocampal cognitive maps in the presence of inputs from the path integrator and visual cues is based on CANN theory [32, 40]. Two-dimensional CANN attractors, were previously applied to networks of place [15, 17] and grid [41, 42] cells. Indirect experimental evidence supporting CANN is now accumulating in various animals and brain areas. Evidence of a ring-shaped attractor region associated to head direction representation was recently reported in the drosophila central brain [43]. Attractor dynamics has also been associated to behavioral observation in a study on the monkey prefrontal cortex [44]. As for space representation, experimental support for attractor behaviour has been found in hippocampal CA1 [45] as well in grid cell [46] recordings. Further indirect evidence is provided by the pattern of connectivity in CA3, compatible with its functional role as an auto-associative attractor network [13], as suggested long ago based on anatomical and computational considerations [12, 47], and by the active nature of dendrites of mEC neurons, which enhances the robustness of attractors under environmental changes [48].
The detailed analysis of the CA3 recordings done here provides another indirect support for CANN theoretical framework, when multiple (two) cognitive maps are memorized. Memorization of the two attractors is obtained, in the model, by adding the corresponding connectivity matrices into the unique CANN connectivity matrix [16–19]. A detailed theoretical study of the mechanisms for transition from map to map was obtained in the absence of inputs, i.e. for spontaneous transitions induced by neural noise only [36]. A similar picture is found here in the presence of visual-cue inputs pointing to the ‘new’ map, while the path-integrator inputs point to the ‘old’ map in the conflicting phase. As inputs are of comparable magnitude, no single map is favored. The stochastic fluctuations resulting from the noise of the individual neurons are sufficient for the system to cross the activation barrier between the two memory states (maps) of the network, see Fig 4B and Fig. A in S1 Text. The hippocampal network jumps intermittently from one cognitive map to the other, reproducing the flickering events experimentally identified and described in [27]. Transition rates between the two maps increase with the neural noise, modeled here by the parameter β, see Eq (12) in Methods and Fig. D in S1 Text. Neural noise relative to the population activity could also be effectively increased through the introduction of periodic (theta and gamma) modulations of the activity into the model [31, 37, 49]. The presence of rhythms is known to facilitate memory formation and integration of information [50, 51]. While theta oscillations can help produce flickering events as previously reported [31, 38], our work shows that such periodic modulations are not necessary. Transitions could also be facilitated by particular ‘confounding’ landmarks or positions in space, where the maps happen to be locally similar [16, 36].
The present model reproduces accurately all the observed flickering properties, without any need for a post-learning short-term plasticity of the CA3 network hypothesized in [38]. In particular, our model predicts that the flickering frequency is independent from the time spent after the teleportation event in the conflicting phase (Fig 5B). This finding is at first sight in disagreement with the exponential decay of the flickering frequency reported in [27, 38]. However, the latter was obtained as a result of an averaging over many teleportation events. For a single event, accurate data analysis shows that our constant flickering rate hypothesis, when combined with the exponentially distributed realignment time of the path integrator (Fig 5A), is much more likely than an exponential decreasing scenario.
Our model is based on the existence of two streams of inputs conveying, respectively, external landmark and self-navigation information. Recent studies have pointed to the grid cells network in mEC as the possible region that supports path integration, as their firing patterns are maintained in the dark [52], and the relative phases of grid cells seem to be largely unaffected by global remapping between environments of similar shapes [11, 53]. CANN-based approaches have been proposed to model grid-cell networks [41, 42], differing from hippocampal CANN mostly by the short-range nature of the inhibitory couplings. In much the same way the microscopic hippocampal CANN proposed here can effectively be reduced to a 2-state model (Fig 4B), we expect CANN models for the grid-cell networks to be approximately described by a 2-state model, corresponding to the PI aligned with map A or B [11, 54, 55]. This motivates the simple model for the PI we have considered here.
In addition to sending projections towards the hippocampus, our model PI receives a feedback from the CANN, greatly increasing the probability of transition to the state agreeing with the instantaneous cognitive map [32, 56, 57]. Eventually, the state of the PI is realigned along the visual cue inputs, which stops the conflicting phase. Our model effectively implements a ratchet mechanism, locking the system into the coherent phase after a conflicting transient. Realignment of the path integrator based on visual landmarks is an important functional property, intended to limit the accumulation of errors in position estimation [58], and observed for large mismatch between external and internal inputs [24]. From a physiological point of view, projections exist from CA1 to mEC [32], and have been shown to be important for the formation of grid cells [57]. Hence, the feedback from the CANN, thought here to model CA3 activity, to the PI should be understood as effective.
As recently reported in [53], the impairement of the mEC grid firing resulted in a loss of path integrator in behaving rodents. As in our model, the recall of the pre-teleportation map, and, therefore, the whole flickering phenomenology are driven by the input stream from the path integrator to the CA3 network, we conjecture that flickering instabilities would disappear upon grid-cell impairement. Simultaneous recordings of mEC and CA3, as in [11], would be extremely useful to test our predictions and better describe the effect of the path integrator on the cognitive status of CA3.
Theta bins are identified with the Hilbert transform procedure of [27]. The activity of the N recorded neurons is binarized into each theta bin t: si,t = 1 if neuron i is active in bin t, 0 otherwise. For each cognitive map m = A, B a Ising-model probability distribution Pm(s) for the activity configurations s = (s1, s2, …, sN) is inferred,
P m ( s ) = 1 Z m exp ( ∑ i h i ( m ) s i + ∑ i < j J i j ( m ) s i s j ) , (2)
where Z m is a normalization constant. Couplings (Jm) and fields (hm) are determined such that the pairwise correlations and average activities in the neural population computed from Pm match their experimental counterparts in the reference session of environment m. These inverse Ising problems are solved using the Adaptive Cluster Expansion algorithm [59–62]. The inferred models (2) are then used to dynamically decode the map mt during the test session (s) [28], based on the log-ratio of the probabilities of the activity configuration in time bin t in the two environments (main text Eq [1]), with the result
m t = { A if Δ L ( s t ) > L 0 , B if Δ L ( s t ) < - L 0 , (3)
where the threshold L0 is chosen according to the required statistical confidence. We generally set L0 = log 10 ≃ 2.3.
After having decoded the map mt in theta bin t, we define the flicker variable ft, equal to 1 if mt does not match the light cue in theta bin t, to 0 otherwise.
The time correlation of flickering events for delay τ is defined as
C ( τ ) = 1 T t o t ∑ i = 1 S − 1 ∑ t = T i T i + 1 − τ f t f t + τ − ( 1 T t o t ∑ i = 1 S − 1 ∑ t = T i T i + 1 − τ f t ) 2 (4)
where S is the total number of switch events in the recorded data (S = 16 in [27]), Ti is the theta bin index of switch i (< S), and Ttot = TS is the total number of theta bins in the test session. The time correlation C(τ) is typically exponentially decaying, with a decay time τ0, see Fig 2C.
The correlation time τ0 is related to the sojourn time of the neural bump in the cognitive maps, defined as a sequence of contiguous theta bins decoded in the same map, see S1 Text. Theta bins whose | Δ L | are lower than the threshold L0 are considered as belonging to the same map as the last statistically significant time bin. The distribution of sojourn times in each map is shown in Fig. E in S1 Text.
The arena is discretized into 60 × 60 squared bins of 1 cm2 each, with integer coordinates (x, y) [27]. For each reference environment m ∈ (A, B) we construct the binary rate map, p i ( m ) ( x , y ), equal to the average of si,t over all theta bins t in which the rat is at position (x, y). Position is then decoded according to the naive Bayesian framework [63]: the probability of the activity configuration st = {s1, s2, …, sn} in theta bin t and at fixed position (x, y) reads
P m ( s t | x , y ) = ∏ i = 1 N [ p i ( m ) ( x , y ) · s i + ( 1 - p i ( m ) ( x , y ) ) · ( 1 - s i ) ] . (5)
Once m is known, e.g. either through the map decoder or due to constant experimental conditions, the position of the rodent can be reconstructed from the recorded neural activity through
( x ^ t , y t ^ ) = arg max ( x , y ) [ P m ( s t | x , y ) × T m ( x , y ) ] , (6)
where the maximum is computed over the 60 × 60 possible positions. Tm(x, y) is the number of theta bins spent by the rodent at position (x, y) during the reference session of map m; we use it as a prior to favor positions where the rodent is more likely to be, irrespectively of the neural activity.
The hippocampal population includes N place cells. For each cell i = 1…N the place-field centers coordinates, r i A and r i B, are drawn uniformly and independently at random in the squared environments, respectively, A and B. The linear size of each square is denoted by L.
Neural activities are represented by binary variables: si,t = 0 or 1 if neuron i is, respectively, silent or active in time bin t = 1, 2, 3, …. The duration of a time bin is the theta cycle over K; results reported here were obtained with K = 4, which corresponds to approximately 30 ms.
The total input received by neuron i at time t is
H i , t = ∑ j ≠ i J i j s j , t + h i ( V ) ( r ) + h i ( P I ) ( r ) . (7)
The three terms on the right hand side of Eq [7] represent, in order:
All neurons undergo stochastic updating of their activities from time bin t → t + 1 according to their total inputs. The activity of neuron i at time t + 1 is chosen to be
s i , t + 1 = { 0 with probability 1 1 + e β ( H i , t - θ ) 1 with probability e β ( H i , t - θ ) 1 + e β ( H i , t - θ ) . (12)
To enforce global inhibition in the population activity, the value of the threshold θ is dynamically adjusted so that an average fraction f of the neurons is active at any time. Parameter β controls the amount of noise in the neural dynamics. For β → 0 neuron activities are random and independent of their inputs, while, for β → ∞, they deterministically follow the signs of the inputs (after subtraction of the threshold θ). The average activity of cell i at time t + 1 is therefore a monotonously increasing sigmoidal function of its total input Hi,t at time t, with maximal slope equal to β/4 in Hi,t = θ.
The properties of this CANN model in the absence of any visual and PI inputs, i.e. for γV = γPI = 0, were analytically studied in [19, 36, 37], see S1 Text for further discussion. The log-ratio Δ L defined in Eq (1) for the decoding of cognitive maps has a direct counterpart in our CANN model as the difference between the contributions to the log-probability of an activity configuration s when the bump is localized in maps A and B,
Δ L ( s ) = ∑ i < j ( J i j A - J i j B ) s i s j . (13)
The path integrator is described as a two-state model, PI = A or B. Its dynamics is stochastic and Markovian: in each time bin t, the state PI can jump into state PI′ with transition probabilities R(PI → PI′), independently of the previous states. The feedback from the hippocampal network to the path integrator is expressed in the dependence of R on the hippocampal map Mt at time t. To favor transitions to the state PI′ agreeing with the current map Mt, we introduce the following witness function for the presence of the bump in map m = A, B:
W ( m ) ( s , r ) = ∑ i = 1 N s i ϕ ( r - r i m ) , (14)
where s and r are, respectively, the activity configuration and the position of the rodent at time t. Due to the normalization of ϕ in Eq [9], we expect W(m) to be close to one for the retrieved map m = Mt and to be much smaller for the opposite map.
We impose the preference for realigning the path-integrator state in accordance with the hippocampal map through the ratio between the two reciprocal transition probabilities,
R ( P I = B → P I ′ = A ) R ( P I = A → P I ′ = B ) = e γ W ( W ( A ) ( s , r ) - W ( B ) ( s , r ) ) . (15)
Here, γW is a positive parameter allowing us to tune the strength of the preference. If the hippocampal bump of activity is localized in, say, map A, the right hand side of Eq [15] will be strongly positive, and the probability of realigning the path integrator to PI′ = A will be much larger than the probability of the reciprocal transition.
A solution to the constraint expressed by Eq [15] is given by
R ( P I = B → P I ′ = A ) = R 0 × e γ W ( W ( A ) ( s , r ) - W ( B ) ( s , r ) ) / 2 , R ( P I = A → P I ′ = B ) = R 0 × e - γ W ( W ( A ) ( s , r ) - W ( B ) ( s , r ) ) / 2 , (16)
where R0 is a positive number. In the absence of bias (γW = 0), the inverse of R0 may be interpreted as the average time scale between two realignments of the path-integrator state.
The model for the path-integrator dynamics is entirely defined by the transition probabilities in Eq [17] and the probability conservation identities:
R ( P I = A → P I ′ = A ) + R ( P I = A → P I ′ = B ) = 1 , R ( P I = B → P I ′ = A ) + R ( P I = B → P I ′ = B ) = 1 . (17)
Defining τ as the PI-realignment time, t = 0 as the time bin corresponding to the light switch and T as the time bin corresponding to the next switch (end of analyzed data), we assume the probability p(t) for time bin t to be a flickering event to be
p ( t ) = { p 0 , if 1 ≤ t ≤ τ , p e , if τ + 1 ≤ t ≤ T . (18)
Here, p0 is the constant flickering probability, and pe is the baseline decoding error, see S1 Text for discussion of the values of parameters p0 and pe.
We write the log-likelihood of the parameter τ as a function of the identified flickering sequence f = {ft} as follows:
log P ( f | τ , p 0 , p e ) = log p 0 × ∑ t = 1 τ f t + log ( 1 - p 0 ) × ∑ t = 1 τ ( 1 - f t ) + + log p e × ∑ t = τ + 1 T f t + log ( 1 - p e ) × ∑ t = τ + 1 T ( 1 - f t ) (19)
We then maximize this log-likelihood over τ to infer the most likely value τ* of the realignment time. The procedure is repeated for all light switches, see Fig. G in S1 Text.
We consider two hypothesis:
Following the Bayesian information criterion [64], we parametrize each model with the same number of variables. For hypothesis Hdecay, we estimate the flickering frequency as a function of time from the average frequency computed over the full test session (Fig 5C, bottom), in bins of one second-width (8 theta cycles), up to 15 seconds after the light switch. For later delays (> 15 s) the flickering frequency is set to a baseline error probability, pe = 0.01. For hypothesis Hconstant, we infer the most likely PI realignment times for each one of the 15 light-switch events (see Section above). The associated flickering probability pt is then set to p0 = 0.55 until the inferred PI realignment time τ*, and equal to the baseline probability pe = 0.01 afterwards. We then compute the likelihoods of both hypothesis given the observed data (identified flickering theta bins f) through
ℓ ( hypothesis | data ) = ∑ t = 1 T [ log p t × f t + log ( 1 - p t ) × ( 1 - f t ) ] , (20)
where T is the total length of the test session. The above expression is then summed over all light-switch events. We define the difference of the two log-likelihoods as
Δ ℓ = ℓ ( H c o n s t a n t | data ) - ℓ ( H d e c a y | data ) . (21)
The constant flickering frequency hypothesis Hconstant is extremely more likely (Δℓ ∼ 150) than the decaying model Hdecay. The result is robust against changes in the parameters, see Fig. F in S1 Text.
|
10.1371/journal.ppat.1000078 | Trypanosomiasis-Induced B Cell Apoptosis Results in Loss of Protective Anti-Parasite Antibody Responses and Abolishment of Vaccine-Induced Memory Responses | African trypanosomes of the Trypanosoma brucei species are extra-cellular parasites that cause human African trypanosomiasis (HAT) as well as infections in game animals and livestock. Trypanosomes are known to evade the immune response of their mammalian host by continuous antigenic variation of their surface coat. Here, we aim to demonstrate that in addition, trypanosomes (i) cause the loss of various B cell populations, (ii) disable the hosts' capacity to raise a long-lasting specific protective anti-parasite antibody response, and (iii) abrogate vaccine-induced protective response to a non-related human pathogen such as Bordetella pertussis. Using a mouse model for T. brucei, various B cell populations were analyzed by FACS at different time points of infection. The results show that during early onset of a T. brucei infection, spleen remodeling results in the rapid loss of the IgM+ marginal zone (IgM+MZ) B cell population characterized as B220+IgMHighIgDInt CD21HighCD23LowCD1d+CD138−. These cells, when isolated during the first peak of infection, stained positive for Annexin V and had increased caspase-3 enzyme activity. Elevated caspase-3 mRNA levels coincided with decreased mRNA levels of the anti-apoptotic Bcl-2 protein and BAFF receptor (BAFF-R), indicating the onset of apoptosis. Moreover, affected B cells became unresponsive to stimulation by BCR cross-linking with anti-IgM Fab fragments. In vivo, infection-induced loss of IgM+ B cells coincided with the disappearance of protective variant-specific T-independent IgM responses, rendering the host rapidly susceptible to re-challenge with previously encountered parasites. Finally, using the well-established human diphtheria, tetanus, and B. pertussis (DTPa) vaccination model in mice, we show that T. brucei infections abrogate vaccine-induced protective responses to a non-related pathogen such as B. pertussis. Infections with T. brucei parasites result in the rapid loss of T–cell independent IgM+MZ B cells that are normally functioning as the primary immune barrier against blood-borne pathogens. In addition, ongoing trypanosome infections results in the rapid loss of B cell responsiveness and prevent the induction of protective memory responses. Finally, trypanosome infections disable the host's capacity to recall vaccine-induced memory responses against non-related pathogens. In particular, these last results call for detailed studies of the effect of HAT on memory recall responses in humans, prior to the planning of any mass vaccination campaign in HAT endemic areas.
| African trypanosomes are extracellular parasites that cause the deadly disease sleeping sickness in humans, and nagana in cattle. The control of infection is believed to be largely dependent on the host antibody response. We postulate here that protective anti-trypanosome responses mainly involve splenic marginal zone B cells, as they are implicated in the production of antibodies against blood-borne pathogens. In this work, we show that trypanosome infections induce the rapid loss of these marginal zone B cells, coinciding with the loss of the splenic marginal zone itself. While the infection does result in the induction of plasma cell differentiation and antibody secretion, the loss of the marginal zone B cell population results in the loss of specific protective responses. In addition, we also show that host memory responses are destroyed during infection, even affecting unrelated vaccine-induced memory responses such as those induced by the commercially available DTPa vaccine. The latter finding is crucial for the evaluation of mass vaccination approaches in African regions where trypanosome infections are prevalent.
| African trypanosomes that belong to the T. brucei species are extracellular parasites that cause Human Afican Trypanosomiasis (HAT) and Nagana, a wasting disease of cattle. As a defense barrier against the host immune response, the entire surface of the T. brucei parasite is covered with 107 densely packed molecules of a variant surface glycoprotein (VSG) that determines the antigenic phenotype of the parasite [1],[2],[3]. At any given time, a single VSG gene encodes for all the VSG molecules present on the trypanosome surface, creating a homogenous antigenic coat. There are at least 1000 different VSG genes present within the T. brucei genome [3],[4],[5]. In addition, these VSG genes undergo extensive recombination generating an extremely large and plastic antigenic repertoire. The ability to switch expression from one VSG to another is considered to be the major mechanism allowing the parasite to evade an efficient host antibody response, hence preventing parasite elimination and permitting the establishment of a chronic infection [3],[6]. In addition, the extreme degree of antigenic variation exhibited by African trypanosomes is considered to be the main reason for the failure of anti-trypanosome vaccination strategies to date [7].
Experimental T. brucei infections in mice are widely used to study host pathogen interactions [8],[9], and serve as models for anti-trypanosome vaccine development [10]. These infections are characterized by the recurring appearance of peaks of parasitemia corresponding to the newly emerging variant antigenic types (VATs) of the parasite. Parasite elimination from the blood, lymph and various host tissues can result from combined antibody-mediated killing, nitric oxide and cytokine toxicity, and parasite growth arrest in response to host and parasite-derived quorum-sensing factors [9],[11],[12],[13],[14],[15]. During the first days of infection, mice generate a rapid IgM response, followed by an immunoglobulin isotype switch and secretion of high levels of IgG2a, IgG2b and IgG3 antibodies [16]. Although infections trigger both a T-dependent and a T-independent antibody response, host resistance was shown to be mainly dependent on the latter and functions through a complement independent mechanism. Indeed, trypanosome infected athymic mice as well as complement-deficient mice mount an effective antibody response and control parasite growth with similar kinetics as observed in wild-type mice [16],[17],[18]. Interesting to note is that while early induction of anti-trypanosome antibody responses aids in the effective VAT-specific clearance of the first peak of parasitemia, mice lose the capacity to control T. brucei growth later during infection, and usually die due to multi-organ failure with high circulating parasite loads in the presence of high levels of anti-VSG antibodies [16],[18]. Interestingly, while polyclonal B cell activation was found to be a hallmark of the infection with human infective parasites [19],[20] and in trypanosome infected domestic animals [21], the mechanisms underlying this process are still not fully understood. It has been proposed that early on during infection, the densely packed VSG coat can be recognized as a highly repetitive and structured single-epitope array implicated in abnormal B cell activation as well as exhaustion [22].
To date, little is known about the fate of specific B cell populations and the tissue re-modeling during trypanosome infections, although early studies have reported changes in splenic B cell content indicating high cellular infiltration into the spleen. In general, the earliest B cells that migrate from bone marrow (BM) to the spleen are transitional type 1 (T1) B cells. These cells develop locally into transitional type 2 (T2), and next into marginal zone (MZ) B cells or mature follicular (Fo) B cells. MZ B cells play an important role in the early phases of antibody responses against mainly T-independent antigens [23],[24],[25]. As T-independent B cell responses are crucial for early control of T. brucei, the role and fate of MZ B cells was further analyzed here. Splenic MZ B cells can be characterized as B220+IgMHighIgDIntCD21HighCD23LowCD1d+CD138− [26]. Overall, B cell differentiation and survival is regulated by the B cell-activating factor (BAFF), a member of the TNF family. BAFF-R, one of the three BAFF receptors, is expressed on B cells and controls overall B cell homeostasis [27]. The binding of the BAFF to BAFF-R promotes ΝF-κB activation and increases mRNA levels of the anti-apoptotic factor Bcl-2 [28]. Despite these data, and despite the fact that T. brucei infections were reported to induce B cell unresponsiveness to mitogenic stimuli [29],[30],[31], trypanosome-induced B cell apoptosis has not received major attention so far. In general, apoptosis occurs in several pathological and non-pathological conditions and constitutes a part of a mechanism of cell replacement and tissue re-modeling, leading to maintenance of cellular homeostasis [32],[33]. The apoptotic process is characterized by the series of morphological changes such as cell shrinkage, chromatin condensation, and DNA fragmentation. Here, the family of caspases plays a central role, with the activation of caspase-3 by the release of cytochrome C regarded as a primary mechanism involved in apoptosis and the degradation of chromosomal DNA [34]. The anti-apoptotic Bcl-2 protein counteracts this process by inhibiting the release of cytochrome C, which in turn blocks activation of caspase-3 [35].
Given the crucial role of T-independent IgM responses in trypanosomiasis control, and the lack of data on the fate of B cell populations and in particular IgM+MZ B-cells during infection, this paper focuses on this rather neglected part of the host immune response.
Here we demonstrated that extensive remodeling of spleen micro-architecture takes place early after T. brucei infection. This event coincides with the drastic reduction in IgM+MZ B cells. Moreover, the loss of IgM+ B cells after in vivo infection rendered mice susceptible to a challenge with a previously encountered T. brucei variant antigenic type. Together, our data indicates that while antigenic variation might aid in perpetuating a T. brucei infection within a given host, active infection-driven elimination of IgM+ B cells renders a host susceptible to repetitive infections by the same antigenic type trypanosome. Moreover, in this paper we demonstrated that T. brucei infection can also abrogate vaccine induced protective responses that were generated against non-related pathogens such as Bordettela pertussis (B. pertussis), using a human vaccine against diphtheria, tetanus and B. pertussis (DTPa) in a mouse vaccination model.
The pleomorphic AnTat 1.1E (EATRO 1125 stock) T. brucei brucei was used in this study as previously described [12]. This infection is characterized by a multi-wave parasitemia development, in which every wave represents a parasite population of different antigenic type. Mice were infected by i.p. injection of 5000 parasites/mouse. Every 2 to 3 days, the number of parasites present in the blood was estimated using a counting chamber and a light microscope. For re-challenge experiments, cloned monomorphic T. b. brucei AnTat 1.1E or control monomorphic T. b. brucei MITat 1.4 parasites were used. These clones are characterized by the rapid killing of their host, expressing one single VAT and their lack of antigentic variation during the 4 day infection. The homogenic expression of AnTat 1.1 VSG and MITat 1.4 VSG was verified here by RT-PCR followed by VSG sequence analysis. Re-challenge experiment of AnTat 1.1E infected mice was performed on day 10 and 17 by administering 5000 cloned monomorphic parasites/mouse. T. brucei infections and re-challenge experiments were performed using the following female mice: Balb/c and C57BL/6 (Harlan), and µMT B cell deficient mice as well as nu/nu mice on the C57BL/6 background (both a kind gift of Prof. B. Ruffel, CNRS). All mice were housed in the Animal Facility under barrier conditions. The appropriate university's ethics committees approved all experimental animal procedures.
B cell proliferation assays were performed using splenic B cells from infected or non-infected mice. Cells from day 10 post infection were purified on CD19 MACS separation columns (Miltenyi Biotec). Aliquots of eluted B cells (5×104/ml) were stimulated with different concentration of LPS or anti-IgM Fab fragments. After 24 hours, cells were pulsed with 1 µCi (3H) thymidine (AEC Amersham Uppsala, Sweden) and incubated for a further 18 hours.
B cells were analyzed by flow cytomerty. Spleens were harvested from infected and non-infected mice at different time points of infection. Cell suspensions were prepared in complete RPMI 1640. Non-specific binding sites were blocked for 30 min at 4°C, using ice-cold PBS supplemented with 1% normal rat serum and 2.4G2 the anti-FcγR antibody. After washing twice, cells were stained with the following anti-mouse antibodies: anti-IgM, anti-IgD, anti-CD21, anti-CD23, anti-B220, anti-CD1d, anti-CD138, anti-GL7, anti-PNA, coupled to either FITC, or PE. All antibodies were purchased from BD Biosciences (San Jose, CA). Incubations were conducted for 30 min at 4°C. For analysis of apoptosis, cells were simultaneously stained with Annexin V (BD Bioscience) and 7AAD (BD, Biosciences) according to protocols provided by the manufacturers, in combination with the B-cell markers listed above. Stained cells were analyzed on a FACS-Calibur SE flow cytometer (BD Biosciences). Prior to analysis, PI (propidiun iodide) was added to all cell suspension in order to exclude dead cells from the data acquisition. All data was analyzed using the FlowJo (Tri Star) software package.
Gene expression levels were measured by Quantitative Real Time PCR, using the Roche/SYBR green system. B cells from infected and non-infected mice were purified using CD19 magnetic beads. Total mRNA was extracted using Trizol reagent according to the instructions supplied by the manufacturer (Invitrogen). Residual DNA was digested using Turbo DNase kit from Ambion. Reverse transcription was performed using protocol supplied by Invitrogen. Real-time PCR reactions were performed according to the Roche protocol, which included data analysis by ‘Fit Points’, normalization against 18S expression, and ‘Standard Curve Analysis’.
The following primer pairs were used for PCR amplifications:
Spleens were removed from infected (day 7 post-infection) and non-infected mice. CD21HighCD23Low MZ B cells were sorted using a FACSVantage (BD bioscience) and used for lysate preparation. Cells were re-suspended in ice-cold PBS supplemented with complete protease inhibitors (Roche). Cell suspensions were sonicated 3 times on ice by delivering the impulse for 30 seconds. Cellular lysate was spun down at 100.000×g for 30 min at 4°C and a soluble fraction was collected. Protein concentration was determined by BCA colorimetric assay (Pierce). 5 µg of soluble proteins were submitted to reducing conditions and separated on SDS PAGE followed by a transfer onto a nitrocellulose. Nitrocellulose membranes were incubated in 1%BSA/Tris pH 8.0 for 1 hour at room temperature in order to block non-specific binding sites. Membranes were washed 5 times with Tris pH 8.0/0.05% Tween 20 solution and incubated with anti-caspase-3 (BD/Pharmingen, clone C92-605, rabbit IgG) antibody diluted 1000 times in 1%BSA/Tris buffer pH 8.0, recognizing mouse pro-caspase 3 as well as the activation cleaved 12 KD and 17 kD caspase 3 products. Next, washed membranes were incubated with a secondary goat anti-rabbit IgG biotinylated antibody (30,000 diluted in Tris buffer) for 1 hour at room temperature. After a washing step, streptavidine/alkaline phosphatase (Sigma) was added to the membranes and incubated for one hour at room temperature. Proteins were visualized by using BCIP/NBT substrate solution from Sigma according to the manufacturer instructions.
Spleens were removed from control and infected mice at day 10 post infection. They were embedded in Tissue-Tek (Sakura Finetek USA, Torrance, California, United States), frozen on dry ice and stored at −80°C. Cryostat sections (6–8-µm thick) were fixed in ice-cold acetone for 10 min, rehydrated in PBS, and treated for 30 min with PBS 1% blocking reagent (PBS-BR) (Boehringer Mannheim, Mannheim, Germany). Obtained cryosections were washed in PBS and stained for 60 min with the following biotinylated-labeled antibodies: MOMA-1 (anti-metallophillic marginal macrophages (MMM)) and ERTR-9 (anti-marginal zone macrophages (MZM) (BMA Biomedicals, Augst, Switzerland) resuspended in PBS-BR buffer. Next, the sections were washed in PBS and incubated for 30 min with the secondary detection reagents streptavidin-FITC and streptavidin-Cy3 (Zymed Laboratories, Invitrogen, Carlsbad, California, United States) in PBS-BR for MZM and MMM visualization, respectively. The slides were washed in PBS and mounted in anti-fading GEL/MOUNT (Biomeda, Foster City, California, United States).
Digitized images were captured using a Zeiss AxioCam color camera and analyzed using the Photoshop software (Adobe Systems, San Jose, California, United States).
BALB/c neonatal mice were vaccinated according to the previously published protocol [36],[37]. In short, neonatal mice were vaccinated with one quarter of a human dose of the commercially available DTPa vaccine (Boostrix®) administered sub-cutaneous (s.c.). After 21 days, mice received a s.c. booster injection with the same amount of vaccine. After a further 14 days mice were infected i.p with 5000 T. brucei parasites/mouse. Ten days post-infection, mice received an intranasal dose of 5×106 B. pertussis bacteria/mouse (ATCC9797) in 10 µl PBS. Control mice received the intranasal B. pertussis challenge in the absence of a trypanosome challenge, or in the absence of vaccination and parasite infection. Lung bacterial load clearance was monitored after 3 hours, and 3, 5, 8 days post-challenge. Mice were sacrificed and whole lungs were isolated and homogenized in 5 ml PBS. Serial 10 fold dilutions were prepared and aliquots of 200 µl were plated onto the Bordet-Gongaou agar plates. The number of colony forming units (CFU's) was counted after 72 hours of incubation at 36°C.
Induction of a T-independent anti-trypanosome IgM response has been shown to be a crucial factor in T. brucei parasite elimination. As IgM+MZ B cells are the main mediators of T-independent antigen responses, the fate of this population was addressed in an experimental C57Bl/6 mouse T. brucei AnTat 1.1 infection model. Here, the trypanosome infection is characterized by multiple waves of parasitemia and a survival of about 35-40 days [16]. Hence, spleens were isolated at different time points during infection and prepared for cellular characterization by FACS. Multiple surface staining combinations can be used to characterize MZ B cells, with CD21/CD23, IgM/IgD and B220/CD1d as the most common combinations [26]. Figure 1A (upper panel) shows that all three combinations give similar MZ B cell counts when spleen populations of naïve non-infected mice were analyzed, with CD21HighCD23Low (R1) = 2.25%, IgMHighIgDInt (R2) = 2.39% and B220+CD1d+ (R3) = 2.37%. In contrast to naïve spleen populations, MZ B cells were virtually absent from spleen cell populations derived on day 10 of an AnTat 1.1 T. brucei infection. Indeed, Fig. 1A (lower panel) shows that all three surface staining combinations indicate a drastic reduction in the % of MZ B cells. In order to calculate the reduction IgM+MZ B cell numbers, the R2 gate was used (Fig. 1B). It is important to stress here that during experimental trypanosome infections, a marked splenomegaly takes place [38], resulting in a significant increase by day 10 post-infection of the total amount of cells present within the spleen (Fig. 1C). Hence, in order to incorporate the increase in cell number due to splenomegaly, the % data obtained by FACS was used to obtain the total IgM+MZ B cell count per spleen. Combined, these data clearly show that even when the increased splenic cellularity of infected mice is taken into account, a significant reduction of splenic IgM+MZ B cell numbers takes place right after the first week of infection. In order to visualize the trypanosomiasis-induced loss of the marginal zone, an additional independent strategy was applied. The MZ, delineated by the marginal sinus in the spleen, contained two distinct macrophage populations, marginal metallophilic macrophages (MMM) and marginal zone macrophages (MZM) which localize to the inner and outer rim of the MZ, respectively [39]. By day 10 post-infection, T. brucei caused a nearly complete loss of MZM and MMM, identified by the ER-TR9 and MOMA-1 markers, respectively (Fig. 2). Further analysis of the spleen remodeling, revealed that beside the loss of MZ, the entire spleen structure encompassing white and red pulp is lost permanently, confirming previous data [40].
Besides IgM+MZ B cells, splenic B cell populations comprise mainly Follicular (Fo) B cells and plasma (Pl) or germinal center B cells, depending upon the immune status of the host. Given the destruction of the spleen micro-architecture and the absence of induction of germinal center formation during infection, both Fo B and Pl B cell populations were further analyzed in detail by FACS. In naïve mice Fo B cells appear as distinct splenic population expressing CD21IntCD23High, IgMIntIgD High that distinguishes them from CD21HighCD23Low, IgMHighIgDInt expressing MZ B cells. They represent up to 25% of the total cell count in a naïve spleen (see Fig 1A). Plotting the alteration in Fo B cell numbers during a T. brucei AnTat 1.1 infection shows that also these cell counts decrease towards the end of infection, albeit not with the same magnitude as the MZ B-cells (Fig. 3A). In contrast, when plasma B cells are considered a clear and very significant increase of cell numbers during the first 10 days of infection is observed, that coincided with the disappearance of the MZ B-cells (Fig. 3B). However, these cells rapidly decrease in number again over time, although they always remain significantly increased as compared to naïve plasma spleen B cell counts. Here plasma B cells were characterized by surface expression of B220+CD138+markers. Figure 3C shows the clear increased percentage of double positive spleen cells on day 10 post-infection, and the relative reduction observed by day 17. When these cells were stained for IgM and CD138, about 50% stained double positive throughout infection, while the others stained IgM−CD138+ (data not shown).
The FACS data presented above indicate that MZ B cells are rapidly lost during the first 10 days of a T. brucei infection. Besides the possible differentiation of these cells into IgM+ plasma cells, we addressed whether apoptosis also contributes to the permanent loss of MZ B-cells. To do so, CD21HighCD23Low MZ B cells were stained for Annexin V and 7AAD at several time points of the early infection. Figure 4A indicates the induction of apoptosis in this cells population, following the first parasitemia wave and coinciding with the rapid disappearance of the population. In order to independently confirm the induction of infection-associated apoptosis in MZ B cells, CD21HighCD23Low cells were FACS-sorted and lysed for Western Blot analysis of caspase-3 activation, using a specific antibody recognizing both pro-caspase 3 (32 kDa) and the cleaved caspase 3 bands of 12 and 17 kDa that occur when the apoptosis is taking place. As shown in Fig. 4B, lysate from infection-derived MZ B-cells showed the induced presence of all three caspase 3 forms, while lysates from naïve MZ B cells, loaded on the SDS-PAGE at the same protein concentration, did not score positive for caspase-3 activation.
Additionally, we performed a number of quantitative Real-Time PCR, measuring mRNA levels for caspase-3 as well as mRNA levels coding for the anti-apoptotic protein Bcl-2 and the BAFF receptor (BAFF-R), mainly expressed on B cells and critically involved in B cell homeostasis [28]. Due to the fact that the time consuming FACS sorting technique risked affecting caspase-3 mRNA levels, RT-PCR was here performed on CD19+ MACS highly enriched splenic B cells. Figure 4C shows a significant increase in mRNA for caspase-3 (confirming the WB result presented in Fig. 4B) and down-regulation of mRNA levels for the anti-apoptotic Bcl-2 protein and BAFF-R 7 days post-infection. This pro-apoptotic gene regulation preceded the disappearance of the MZ B cells from the spleen.
The RT-PCR data presented above indicates that T. brucei infections induce apoptosis in populations of IgM+ spleen MZ B cells and MACS sorted CD19+ cells, hence, the later population was used to assess the effect of the infection on the IgM+B cell proliferative response. The sorted cells were incubated for 24 hours with different concentrations of anti-IgM Fab or LPS. Figure 5A indicates that naïve splenic B cells contain a cell fraction that is highly responsive to IgM Fab activation and proliferation. This fraction is absent in spleens of T. brucei infected C57Bl/6 mice 10 days after infection (Fig. 5B). Also when cells were stimulated in a non-specific manner with LPS, naïve CD19+ B cells showed a dose dependent proliferative response (Fig. 5C), while infection derived affected B cells showed a complete inhibition of LPS-mediated proliferation (Fig. 5D).
In order to analyze the consequence of the loss of B cell populations and in particular IgM+ B cells on the anti-trypanosome immune response, a series of re-challenge experiments were performed. Here, mice were infected with the pleomorphic AnTat 1.1 T. brucei parasite (that was used in all previous experiments presented in this paper) (Fig. 6A). The first VAT (VSG antigenic type) that emerges during this infection corresponds to the AnTat 1.1 VSG, while later parasitemia peaks correspond to subsequent VATs. In a separate experiment, mice were infected with the monomorphic AnTat 1.1 and MITat 1.4 parasites, representing clones that do not switch their VAT during their short and highly virulent infection (Fig. 6B/C). Next, infections were combined in order to analyze how mice react to a re-challenge with parasites expressing a previously encountered VAT.
The first re-challenge experiment was performed at day 10. This time point corresponds to the time point of infection in which infection-induced IgM+ plasma cells reach peak numbers (see Fig. 3), and is accompanied by peak levels of anti-trypanosome serum IgM titers that were previously shown to occur at day 10 [16]. Figure 6D shows that here a complete VAT/VSG-specific protection is obtained against the monomorphic AnTat 1.1 clone but not against the non-related MITat 1.4 parasite clone. This protection was retained in T-cell deficient nude mice (nu/nu) (Fig. 6E) but was absent in B-cell deficient µMT mice (Fig. 6F), suggesting that the protective immune response relies entirely on T-independent B-cell responses.
Next, a re-challenge experiment was done at day 17, corresponding to the time point where both IgM+MZ and IgM+ plasma B-cells were dramatically reduced. Here, both the AnTat 1.1 and MiTat 1.4 clones killed the mice rapidly (Fig. 6G). These results show that the AnTat 1.1 specific antibody mediated protection acquired by day 10 post-primary infection was already lost by day 17. A re-challenge experiment at day 24 confirmed the permanent loss of VSG-specific immunity (data not shown).
Following the finding that VAT-specific immunity is rapidly lost during trypanosome infections, we investigated whether T. brucei infection may have a general detrimental effect on the immune response generated by vaccination. As there is no efficient vaccine available against trypanosomes we tested whether T. brucei infection may abrogate the vaccine-induced protective immune response against a non-related pathogen such as B. pertussis. In order to do so, the commercially available DTPa vaccine was used. This vaccine protects children against infections such as B. pertussis, diphtheria and tetanus. The DTPa vaccination scheme has been well established and standardized in mice [36],[37]. It includes one vaccination and one booster after three weeks with the DTPa that leads to the induction of a specific protective antibody response. As indicated in Fig. 7, in contrast to non-vaccinated mice, DTPa vaccinated mice that were challenged with B. pertussis clear the bacteria from the lungs. However, this DTPa vaccine-mediated protective effect was abrogated in vaccinated mice that suffered a T. brucei infection prior to B. pertussis challenge. These results indicated that a T. brucei infection is capable of abrogating the efficacy of the vaccine-induced protective responses against non-related pathogens such as B. pertussis.
To date a large body of evidence is available addressing the molecular mechanisms underlying antigenic variation in African Trypanosomes. Antigenic variation is considered to be advantageous for the parasite as it permits continuous change of the surface antigenic coat and allows the parasites to evade antibody-mediated elimination [3],[7]. Here we demonstrate that in addition, trypanosome infections also cause the significant loss of various B cell populations and extensively remodel splenic micro-architecture. These changes (i) disable the hosts' capacity to mount a protective anti-parasite antibody responses, (ii) prevent the development of effective B-cell memory against encountered variant antigenic parasite types (VATs), and (iii) abrogate vaccine-induced protective responses to non-related pathogen such as B. pertussis in a setting where a commercially available human vaccine for diphtheria, tetanus and B. pertussis (DTPa) was used in an experimental mouse model.
Trypanosome infections are known to induce B cell unresponsiveness to mitogens and to induce polyclonal B-cell activation. In addition, highly virulent infections in which mice fail to control the first peak of infection have been shown to result in the rapid elimination of splenic FoB cells [31]. However, as natural trypanosome infections develop into a chronic phase, we analyzed here the fate of various splenic B cell populations throughout the course of an experimental T. brucei infection, using a more chronic, multi-wave pleomorphic model. In the spleen, B-cell populations can be subdivided into marginal zone (MZ) B cells, follicular (Fo) B cells, plasma (Pl) and germinal center B cells, as well as transitional T1 and T2 B cells [23],[24],[25],[26]. Our results first of all confirm previous data and show that in the early stage pleomorphic T. brucei infection extensive spleen remodeling gives rise to a rapid induction of spenomegaly due to cell proliferation and the influx of cells into the spleen [19],[38]. The detailed analysis of the splenic cell populations showed that following the first parasitemia peak mature B220+CD138+ plasma B cell numbers significantly increased, and that both IgM+ and IgM− plasma cells accumulated at this stage. Later on in infection, B220+GL7+/PNA+ ‘germinal center’ B cells accumulated in the spleen as well, despite the absence of actual germinal center formation during trypanosome infections [40]. Interestingly, while ongoing trypanosome infections continuously generate new parasite antigentic types, major p1asma B cell induction occurs only at the beginning of infection in a response to the first encountered parasite VAT. Coinciding to this early plasma B cell induction, we recorded the total destructions of the spleen marginal zone (MZ), accompanied by the disappearance of the MZ B cell population, characterized as B220+IgMHighIgDIntCD21HighCD23Low CD1d+CD138− cells [26]. As the spleen marginal zone (MZ) separates the T- and B-cell containing white pulp from the blood filled sinuses of the red pulp, the MZ is involved in the capture of blood-born pathogens, the regulation of lymphocyte trafficking, and the induction of antigen specific T-independent B cell responses, mostly resulting in IgM secretion [23],[24]. Important here is that previously published data indicated that T-independent VSG specific IgM responses are crucial in trypanosome clearance during ongoing infection [16],[17],[22]. Using a VSG-specific re-challenge model, we now show that the infection-associated destruction of the IgM+ MZ B cell compartment results in the rapid loss of IgM-mediated VSG specific protection against re-infection with a previously encountered parasite.
The infection-associated disappearance of the MZ B cells from the spleen could be explained by two independently occurring mechanisms, being cell differentiation and/or cell death. Supporting the first is the observation that the rapid disappearance of MZ B cells coincided with the temporary accumulation of IgM+ plasma cells. However, a direct link between these two events remains speculative, as no specific markers exist that allow cell fate analysis that would proof the MZ B cell origin of the occurring plasma cells. Hence, in order to address this possibility, we aimed at reconstituting JH−/− B-cell deficient mice with naïve IgM+ B cells. During T. brucei infections however, this approach did not result in efficient B cell repopulation or the accumulation of IgM+ plasma cells (unpublished results S. Black). On the other hand, our results suggest that the rapid disappearance of MZ B cells does involve the induction of parasitemia-associated B-cell apoptosis. Indeed analysis of the CD21HighCD23Low MZ B cells that remain in the spleen in the days following the clearance of the first peak of parasitemia, revealed that these cells upregulated Annexin V expression and in large stained positive for 7AAD. In addition, the induction of caspase 3 gene expression as well as the conversion of pro-caspase 3 into the cleaved 12 kD and 17 kD caspase 3 activation products in this cell population, suggests the induction of trypanosomiasis-associated apoptosis in the splenic MZ B cell population. Moreover, the affected MZ B cells showed down-regulation of mRNA for the anti-apoptotic Bcl-2 protein, involved in the inhibition of caspase 3 activation through the cytochrome C pathway [28]. Finally, also the mRNA expression for the B cell specific BAFF-R was reduced in the vanishing MZ B cell population. This receptor normally governs B cell homeostasis and provides an NF-κB activation signal that regulates the mRNA levels for the anti-apoptotic Bcl2 protein [27],[28]. Interesting to note here is that parasite-induced B-cell apoptosis has previously also been reported to occur in experimental infections with the intracellular T. cruzi parasite. In this case however, the process was mediated by Fas/Fas Ligand interactions resulting in B cell-B cell fratricide, selectively affecting IgG+ B-cells [41].
In an independent approach, T. brucei induced spleen dysfunction was assessed by immunohistochemistry. Here we show the profound destruction of splenic MZ micro-architecture is marked by the loss of MZM and MMM (macrophage) populations. Detailed analysis of the spleen remodeling, revealed that beside the loss of MZ architecture, the entire spleen structure encompassing white and red pulp is lost, showing the absence of germinal center formation. This destruction of splenic microarchitecture occurred during the first peak of infection, and was permanent. Due to the destruction of the spleen microarchitecture in combination with (i) the alteration of splenic B cell surface markers, and (ii) the drastic decrease in particular B cell subsets, immunohistochemistry could unfortunately not be used for the co-localization of apoptotic markers and specific MZ B cell markers. Important to note here is that remodeling of spleen cell architecture has also been reported in other experimental parasite infection models. Indeed, experimental Plasmodium chabaudi infections in mice also cause major disruptions in the splenic cell distribution as well as the splenic micro-architecture. However, as this infection is self-curing, in contrast to experimental trypanosome infections, these changes were found to be temporary [42].
The last B cell aspect presented in this paper, deals with the question of whether T. brucei infection can actively abrogate a vaccine-induced protective memory immune response. This point is extremely relevant to address in the context of multiple ongoing vaccination programs in the human African population such as the recent WHO Meningitis Vaccine Project (MVP), and the Pediatric Dengue Vaccine Initiative (PDVI), and in the context of future anti- HIV/AIDS and anti-malaria vaccine programs. To address this question, we applied the commercially available human DTPa vaccine against diphtheria, tetanus and B. pertussis in a T. brucei mouse model [36],[37]. While our findings showed the expected protective vaccine response in control vaccinated and B. pertussis challenged mice, the DTPa efficacy was abolished after the vaccinated host was infected with T. brucei parasites. These results indicate that the ongoing trypanosome infection either depleted the memory B cell compartment that was first generated during the DTPa vaccination, or prevented the reactivation of the memory response. Taking into account that the commercially available DTPa vaccine [37] is routinely used in Africa and other countries to protect children against diphtheria, tetanus, and B. pertussis, our results raise the worrying possibility that the protective potential of this vaccine can be abolished by a later encounter with T. brucei parasites, exposing the individual to secondary infection hazards and abolishing previous health efforts. In this context it is remarkable to note that studies performed 3 decades ago already indicated that T. brucei infections abrogate B cell memory responses to thymus-dependent and thymus-independent antigens using DNP-KLH and DNP-Ficoll as antigens for vaccination [29]. Additionally, earlier investigations in domestic animals showed that infections with African trypanosomes abrogate the efficacy of several commercial vaccines against the foot and mouth disease, swine fever, anthrax spores, Brucella abortis and louping-ill vaccine [43],[44],[45],[46],[47]. However, till now no attention has been given to the consequences of T. brucei infections on the immune response induced by commercially available human vaccines. In addition, these findings have not been seriously considered in recent years when (failed) anti-trypanosome vaccination efforts were reported. Indeed several attempts have now been made to develop anti-trypanosome vaccines using relatively non-variant trypanosome antigens, but none of these attempts have been successful so far, despite the initial induction of antigen-specific antibody responses [8],[48],[49],[50]. In view of the results presented here, it is possible that the rapid trypanosome driven B-cell dysfunction upon host contact with living parasites contributed to these vaccine failures. Our studies suggest that only vaccines that strongly suppress the level of initial parasitemia peak or induce immediate sterile immunity would be likely to have sufficient efficacy in trypanosomiasis susceptible hosts. In addition, it is crucial to realize that due to the constraints of conventional strategies, most vaccination protocols lead to the induction of T-dependent high affinity IgG responses. Taking into account that the protective response against previously encountered VAT present in ‘day 10’ re-challenged mice was T-independent and IgM-mediated, this poses an additional obstacle for vaccine development, as till now no strategies exist to generate long lasting high affinity IgM responses. |
10.1371/journal.pgen.1005182 | Ataxin-2 Regulates RGS8 Translation in a New BAC-SCA2 Transgenic Mouse Model | Spinocerebellar ataxia type 2 (SCA2) is an autosomal dominant disorder with progressive degeneration of cerebellar Purkinje cells (PCs) and other neurons caused by expansion of a glutamine (Q) tract in the ATXN2 protein. We generated BAC transgenic lines in which the full-length human ATXN2 gene was transcribed using its endogenous regulatory machinery. Mice with the ATXN2 BAC transgene with an expanded CAG repeat (BAC-Q72) developed a progressive cellular and motor phenotype, whereas BAC mice expressing wild-type human ATXN2 (BAC-Q22) were indistinguishable from control mice. Expression analysis of laser-capture microdissected (LCM) fractions and regional expression confirmed that the BAC transgene was expressed in PCs and in other neuronal groups such as granule cells (GCs) and neurons in deep cerebellar nuclei as well as in spinal cord. Transcriptome analysis by deep RNA-sequencing revealed that BAC-Q72 mice had progressive changes in steady-state levels of specific mRNAs including Rgs8, one of the earliest down-regulated transcripts in the Pcp2-ATXN2[Q127] mouse line. Consistent with LCM analysis, transcriptome changes analyzed by deep RNA-sequencing were not restricted to PCs, but were also seen in transcripts enriched in GCs such as Neurod1. BAC-Q72, but not BAC-Q22 mice had reduced Rgs8 mRNA levels and even more severely reduced steady-state protein levels. Using RNA immunoprecipitation we showed that ATXN2 interacted selectively with RGS8 mRNA. This interaction was impaired when ATXN2 harbored an expanded polyglutamine. Mutant ATXN2 also reduced RGS8 expression in an in vitro coupled translation assay when compared with equal expression of wild-type ATXN2-Q22. Reduced abundance of Rgs8 in Pcp2-ATXN2[Q127] and BAC-Q72 mice supports our observations of a hyper-excitable mGluR1-ITPR1 signaling axis in SCA2, as RGS proteins are linked to attenuating mGluR1 signaling.
| Spinocerebellar ataxia type 2 (SCA2) is an inherited neurodegenerative disorder leading to predominant loss of Purkinje cells in the cerebellum and impairment of motor coordination. The mutation is expansion of a protein domain consisting of a stretch of glutamine amino acids. We generated a mouse model of SCA2 containing the entire human normal or mutant ATXN2 gene using bacterial artificial chromosome (BAC) technology. Mice expressing a BAC with 72 glutamines (BAC-Q72) developed a progressive cerebellar degeneration and motor impairment in contrast to mice carrying the normal human gene (BAC-Q22). We found that even prior to behavioral onset of disease, the abundance of specific messenger RNAs changed using deep RNA-sequencing. One of the mRNAs with early and significant changes was Rgs8. Levels of Rgs8 protein were even further reduced than mRNA levels in BAC-Q72 cerebella suggesting to us that mutant ATXN2 might have a role in mRNA stability and translation. Using a cellular model, we showed that the ATXN2 protein interacted with RGS8 mRNA and that this interaction differed between normal and mutant ATXN2. Presence of mutant ATXN2 resulted in reduced RGS8 protein translation in a cellular model. Our studies describe a mouse model of SCA2 expressing the entire human ATXN2 gene and emphasize the role of ATXN2 in mRNA metabolism.
| Spinocerebellar ataxia type 2 (SCA2) belongs to the group of neurodegenerative diseases caused by polyglutamine (polyQ) expansion. This group includes SCA1, Machado-Joseph disease (SCA3 or MJD), SCA6, SCA7, SCA17, Huntington's disease, spinal bulbar muscular atrophy (SBMA) and dentatorubral-pallidoluysian atrophy (DRPLA). SCA2 is an autosomal dominant disorder leading to motor incoordination which is caused by progressive degeneration of cerebellar Purkinje cells, and selective loss of neurons within the brainstem and spinal cord [1]. As with most autosomal dominant ataxias, symptoms are characterized by a progressive loss of motor coordination, neuropathies, slurred speech, cognitive impairment and loss of other functional abilities arising from Purkinje cells and deep cerebellar nuclei [2,3].
In SCA2, expansion of a CAG repeat in exon 1 of the Ataxin-2 (ATXN2) gene causes expansion of a polyQ domain in the ATXN2 protein. As in the other polyQ diseases, the length of the polyQ repeat is inversely correlated with age of onset (AO) in SCA2 [1,4]. In contrast to other polyQ diseases, mutant ATXN2 does not enter the nucleus in appreciable amounts in early stages of disease. This is also confirmed by protein interaction studies that have identified ATXN2 interactors with cytoplasmic localization [5–8]. Polyglutamine disorders show their pathology through a toxic gain of function of the protein and larger polyQ expansions have been associated with greater pathology [3,9].
ATXN2 is widely expressed in the mammalian nervous system [1,10,11]. It is involved in regulation of the EGF receptor [12], and the inositol 1,4,5-triphosphate receptor (IP3R) whereby increased cytosolic Ca2+ occurs with CAG repeat expansion [13]. ATXN2 functions are also associated with the endoplasmic reticulum [14], and the Golgi complex [15]. Studies in Caenorhabditis elegans support a role for ATXN2 in translational regulation as well as embryonic development [6]. ATXN2 is also important in energy metabolism and weight regulation, as mice lacking Atxn2, developed obesity and insulin resistance [16,17]. Furthermore, ATXN2 interacts with multiple RNA binding proteins, including polyA binding protein 1 (PABP1), the RNA splicing factor A2BP1/Fox1, DDX6, TDP-43, and has been localized in polyribosomes and stress granules demonstrating its unique role in RNA metabolism [5,6,8,18].
Several SCA2 mouse models have been generated. We have reported two transgenic mouse models in which expression of full-length ATXN2 with 58 or 127 CAG repeats (ATXN2-[Q58] or ATXN2-[Q127]) is targeted to Purkinje cells (PCs) using the Purkinje cell protein-2 (Pcp2) promoter [19,20]. These lines show progressive motor phenotypes accompanied by the formation of insoluble cytoplasmic aggregates, loss of PCs, and shrinkage of the molecular layer associated with the reduction of calbindin staining in PC bodies and dendrites. Onset of the motor phenotype of Pcp2-ATXN2[Q127] mice is associated with reduced PC firing that is progressive with age [20]. Another Atxn2-CAG42 knock-in mouse model demonstrated very late-onset motor incoordination associated, but this was seen only in homozygous knock-in animals. This was associated with Pabpc1 deficiency, and upregulation of Fbxw8, but without loss of calbindin staining or downregulation of Calb1 mRNA [21].
In order to model human diseases using cis-regulatory elements, recent mouse and rat models have been created by transgenesis using human bacterial artificial chromosomes (BACs) [22–27]. In the BAC approach, an entire human gene including introns and regulatory regions is introduced into the mouse genome. BAC models often have lower genomic copy numbers than conventional cDNA transgenic models resulting in more physiological expression levels and a potentially more faithful late onset of disease.
We developed new BAC-SCA2 transgenic mouse lines expressing full-length human wild-type or mutant ATXN2 genes including upstream and downstream regulatory sequences. BAC mice with mutant ATXN2 exhibited progressive neurological symptoms and morphological changes in cerebellum. We used this mouse model to confirm changes in key PC-genes identified in a cDNA transgenic model, in particular the effects of mutant ATXN2 on Rgs8 steady state protein levels.
To understand the pathological and behavioral effects in the context of physiologic expression of human wild-type and mutant ATXN2, we engineered a 169 kb human BAC (RP11-798L5) that contained the entire 150 kb human ATXN2 locus with 16 kb of the 5’ flanking genomic sequence and 3 kb of the 3’ flanking genomic sequence (Fig 1A). The authenticities of these constructs were subsequently verified by Southern blot and restriction site analyses (S1 Fig). The CAG tract was mutation-free when sequenced from both strands. After transgenic microinjection of purified intact BAC DNAs, one line each for control (BAC-ATXN2-Q22) and one for mutant mice (BAC-ATXN2-Q72) was further analyzed. These lines will be designated as BAC-Q22 and BAC-Q72 in the remainder of the text. Quantitative PCR (qPCR) analyses of genomic DNA revealed that both BAC-Q22 and BAC-Q72 mice had tandem integrates of 10 and 4 copies of the ATXN2 transgene, respectively. In RT-PCR analyses, both BAC-Q22 and BAC-Q72 mice demonstrated the expression of intact human ATXN2 transcripts throughout the central nervous system (CNS), including cerebral hemispheres, cerebellum and spinal cord (Fig 1B). Non-CNS tissues, including heart and liver also showed ATXN2 transgene expression (Fig 1B). The authenticities of PCR products were confirmed by sequencing. We further determined relative expression of ATXN2 transcripts in the two BAC transgenic lines by quantitative RT-PCR. BAC-Q22 cerebella had higher expression of human ATXN2 than BAC-Q72 cerebella while the expression of endogenous mouse Atxn2 remained unchanged in both compared with wild-type mice (Fig 1C). To assess protein expression, we performed Western blot analysis using cerebellar extracts of 16 week-old animals and a monoclonal antibody (mAb) to human ATXN2. The results showed that BAC mice expressed full-length human wild-type or mutant ATXN2 protein. Of note, protein levels of ATXN2-Q22 were higher than those of ATXN2-Q72. Furthermore, we confirmed the ATXN2-Q72 protein expression using 1C2 mAb, an antibody against an expanded polyQ epitope in Western blot analyses (Fig 1D). These results demonstrate that human ATXN2 transgenes (ATXN2-Q22 and ATXN2-Q72) were properly expressed in BAC mice.
In addition to ATXN2, three overlapping genes (U7.1–202 snRNA, RP11-686G8.1–001 and RP11-686G8.2–001) are contained in the human BAC. Quantitative RT-PCR analyses of wild-type and BAC transgenic mouse cerebellar RNAs demonstrated that the relative expression of each overlapping gene to that of the ATXN2 transgene did not differ between BAC-Q22 and BAC-Q72 animals indicating these genes did not contribute to the phenotypes associated with CAG expansion in the ATXN2 gene (S2 Fig).
The Allen Brain Atlas shows widespread expression of human ATXN2 with very significant expression levels in the cerebellum [28]. Given the nature of ATXN2 expression in brain, we determined the expression of human ATXN2 transgene transcript in sub-regions of mouse brain including spinal cord using qRT-PCR. Expression of endogenous mAtxn2 was evident in many regions including frontal, occipital and olfactory cortex, hippocampus, thalamus, basal ganglia, cerebellum and spinal cord. Human ATXN2 transgene expression was found in all regions tested, but relatively higher expression was observed in the basal ganglia (S3 Fig).
As cerebellar degeneration is predominant in SCA2, we further examined the expression patterns of the ATXN2 transgene in discrete areas of the cerebellum using laser-capture microdissection (LCM). We captured molecular layer (ML), Purkinje cells (PCs), granule cell layer (GCL) and dentate nuclear (DN) fractions. Relative enrichment was determined by measuring expression of a cell-type specific marker genes using qRT-PCR. Evidence for expression of endogenous mAtxn2 was found in all fractions, but was highest in Purkinje cells. Expression of transgenic ATXN2 was also seen in all fractions, although small differences in expression levels existed between BAC-Q22 and BAC-Q72 (Fig 2A and 2B). LCM was remarkably successful in separating cerebellar neuronal population as shown by expression of marker genes for PCs and molecular layer (Pcp2 and Calb1), granule cells (Neurod1) and dentate neurons (Spp1) (Fig 2C and 2F). In summary, inclusion of regulatory regions in the human BAC transgene led to expression of the transgene that mirrored expression of mouse Atxn2 including low but detectable expression in GCs and DNs.
By visual inspection both BAC transgenic lines (BAC-Q22 and BAC-Q72) had a smaller body size than wild-type littermates beginning at 8 weeks of age. By 24 weeks of age, both BAC transgenic mice weighed about 30% less than their wild-type littermates (Wild-type = 33.9 ±3.8; BAC-Q22 = 24.6 ±3.6 and Wild-type = 32.1 ±2.8; BAC-Q72 = 22.9 ±3.7).
BAC-Q72 mice did not show an abnormal home cage behavior. To assess the development of motor impairment, both BAC transgenic lines and wild-type littermates were tested using the accelerating rotarod paradigm at several time points (Fig 3). BAC-Q22 mice performed as well as wild-type littermates at 8, 16 and 36 weeks of age (Fig 3) suggesting that expression of wild-type human ATXN2 was not detrimental to motor function.
BAC-Q72 mice were tested at 5, 16 and 36 weeks of age and compared with their wild-type littermates. BAC-Q72 mice showed normal performance at 5 weeks (Fig 3) and at 12 weeks (S4A Fig). Of note, testing at 12 weeks was performed on mice housed under slightly different conditions, which may explain the relatively poor performance of wild-type mice. At 16 weeks of age, performance of BAC-Q72 mice became significantly worse than wild-type mice (Fig 3; p<0.05) and mice continued to perform poorly as they aged (24 and 36 weeks old, S4A Fig and Fig 3). Taken together, these results indicate that BAC-Q72 transgenic mice develop a progressive age-dependent motor impairment.
To investigate morphological changes associated with the expression of mutant ATXN2 protein, we compared cerebellar sections from BAC transgenic lines with wild-type mice. Immunostaining with calbindin-28k antibody revealed PC morphological changes in BAC-Q72 mice at 24 weeks of age, but not in BAC-Q22 or wild-type mice (Fig 4A). To more quantitatively assess this change, we performed Western blotting and verified reduction of Calb1 and Pcp2 proteins in BAC-Q72 mouse cerebella (Fig 4B). As observed in the Pcp2-ATXN2[Q127] model, cerebellar morphology was still normal at a time when key mRNA transcripts had already declined. Thus, calbindin-stained cerebellar sections and PC counts of BAC-Q72 mice at 12 weeks showed normal cerebellar morphology and unaltered PC counts [18.8 ±1.2 in WT, n = 3 animals, and 19.4 ±1.1 in BAC-Q72 mice, n = 3 animals, p = 0.51] (S4B, S4C Fig).
We previously reported that steady-state mRNA levels of specific PC transcripts preceded behavioral onset in an SCA2 model targeting transgene expression to PCs [20]. Expression changes in these genes (Calb1, Pcp2, Grid2 and Grm1) also preceded the onset of a decrease in PC firing. Expression changes were progressive over time and paralleled deterioration of motor behavior.
To investigate whether similar changes occurred in BAC transgenic mice as we previously observed in Pcp2-ATXN2[Q127], we performed qRT-PCR to measure transcript levels of PC-specific genes at different ages. At 16 and 45 weeks, BAC-Q22 mice were indistinguishable from wild-type mice including expression of endogenous mouse Atxn2 (Fig 5A).
In BAC-Q72 mice, however, expression of Pcp2 showed significant reductions (p<0.01) as early as 5 weeks. All other genes tested remained unchanged compared to wild-type (Fig 5B). At 9 and 16 weeks of age, significant reductions in Calb1 (p<0.05) and Grid2 (p<0.01) were seen and were progressive (Fig 5B). Steady-state levels of Grm1 decreased only at 24 weeks (p<0.05). Endogenous mouse Atxn2 expression levels did not change in BAC-Q72 mice at any time point when compared with wild-type. Taken together, these data demonstrated that a subset of PC-enriched genes showed a progressive reduction in steady-state mRNA levels in BAC-Q72 mice, whereas they remained unchanged in BAC-Q22 animals.
To further characterize the BAC-Q72 line and compare it with the well-characterized Pcp2-ATXN2[Q127] line, we performed transcriptome analysis by deep RNA-sequencing of cerebellar RNA. We chose time points for both lines just prior to behavioral and morphological changes, i.e. 8 weeks for the BAC-Q72 line and 6 weeks for the Pcp2-ATXN2[Q127] line. For both sets of RNAs, quality of reads and alignments were high (see methods).
We observed significant changes of 1417 transcripts in Pcp2-ATXN2[Q127] and 491 transcripts in BAC-Q72 mice with a false discovery rate (FDR) of ≥15 and a log2 ratio of change ≥|0.30| (Fig 6A). With these filtering parameters, 255 transcripts were only seen in the BAC-Q72 line (class I), 236 transcripts were shared between the two lines (class II) and 1181 transcripts were changed only in the Pcp2-ATXN2[Q127] line (Class III). We validated changes in several of the class II transcripts by qRT-PCR using cerebellar RNA samples from BAC-Q72 mice (8 weeks old) and Pcp2-ATXN2[Q127] (6 weeks old), and compared with their respective WT littermates (Fig 6B). The concordance between RNA-seq and qRT-PCR was high (Fig 6C).
The top 50 transcripts changed in the BAC-Q72 line are shown in S1 Table and the top 50 transcripts changed in the Pcp2-ATXN2[Q127] line are presented in S2 Table. This table also shows that most of these transcripts are changed in the BAC-Q72 line as well, although with a smaller degree of change or a lower FDR. S3 Table lists the top class II genes sorted by FDR in the BAC-Q72 line. This represents a subset of the 236 overlapping genes shown in Fig 6A.
In order to gain insight into the molecular function of altered transcripts in BAC-Q72 and Pcp2-ATXN2[Q127] mice, we performed Gene Ontology (GO) analysis. This is shown in S4 Table and indicates that many of the significant GO terms are shared by the two models. Of note, GO terms relate to known functions of PC such as calcium homeostasis, glutamate-mediated signaling and voltage-gated ion channels. In summary, these data indicate a significant overlap of altered transcripts and shared functions in both SCA2 models at comparable stages just prior to onset of morphological and behavioral changes.
We were also interested in the nature and expression pattern of transcripts in class I and class III (Fig 6). We confirmed changes in several of the class I transcripts by qRT-PCR (S5 Fig). These transcripts showed a progressive reduction in BAC-Q72 mice, but remained unchanged in the Pcp2-ATXN2[Q127] line even at late time points. Of these 50, 16 genes (Grm4, Igfbp5, Fstl5, Snrk, D8Ertd82e, Dusp5, Nab2, Btg1, Adrbk2, Slc25a29, Sty12, Crhr1, Synpr, Lrrtm2, Rit2 and Cabp2) were previously identified as GC-specific using translational profiling [29].
Class III transcripts were those that showed changes only in Pcp2-ATXN2[Q127] mice, but not in BAC-Q72 at an FDR>15 and a log2 ratio of change ≥|0.3|. We verified expression changes of six class III transcripts longitudinally in Pcp2-ATXN2[Q127] mice at 4, 8, and 24 weeks of age, and BAC-Q72 mice at 5, 9, 16 and 24 weeks of age, and their respective WT littermates by qRT-PCR. Five of the six transcripts showed significant and progressive reduction with age not only in Pcp2-ATXN2[Q127] mice but also in BAC-Q72 mice (S6 Fig). This is consistent with the milder behavioral phenotype seen in BAC-Q72 mice and suggests that the overlap of the transcriptomes in the two models may potentially be even greater.
Changes in steady-state expression of a subset of genes preceded onset of physiological and behavioral changes in Pcp2-ATXN2[Q127] and BAC-Q72 mice. One of the most significantly down-regulated genes in both models prior to behavioral onset was Rgs8 (regulator of G-protein signaling 8) (S1, S2, S3 Tables). RGS proteins are regulatory and structural components of G protein-coupled receptor complexes. RGS proteins (RGS7, RGS8, RGS11, RGS17 and RGSz1) are widely expressed in cerebellum and RGS8 is specifically distributed in dendrites and cell bodies of PCs [30,31]. Several reports suggest that the RGS family proteins are also associated with motor neuron functions [32,33].
The decreased steady-state level of Rgs8 mRNA was confirmed by qRT-PCR in Pcp2-ATXN2[Q127] mice at 4, 8 and 24 weeks of age, indicating that these RNAs progressively declined with time (S7A Fig). In parallel, we also measured Rgs8 protein steady state levels in Pcp2-ATXN2[Q127] mouse cerebella at 24 weeks of age. As expected, Rgs8 protein levels were significantly reduced in Pcp2-ATXN2[Q127] mice when compared with wild-type mice (S7B Fig).
Next, we investigated the fate of Rgs8 mRNA steady-state levels in our BAC mouse models by qRT-PCR. When tested in BAC-Q72 mouse cerebella, levels of Rgs8 mRNA progressively decreased with time but remained unchanged in BAC-Q22 mice compared with wild-type mice across all ages of mice tested (Fig 7A).
To examine whether changes in steady-state mRNA levels led to decreased protein abundance, we performed Western blot analysis to measure Rgs8 protein in wild-type and BAC transgenic mouse cerebella. Western blot analyses indicated reduced steady-state levels of Rgs8 protein in BAC-Q72 mice but not in BAC-Q22 mice when compared with wild-type mice at 24 weeks of age (Fig 7B).
To assess whether these findings replicated in human cells we analyzed EBV-transformed lymphoblastoid (LB) cells derived from a control individual and two SCA2 patients with expansions of Q46 and Q52 (Fig 7C). We could not use skin fibroblasts as this cell type does not express RGS8. Two SCA2-LB cells expressing Q46 or Q52 demonstrated decreased expression of RGS8 transcript compared with control cells expressing wild-type ATXN2 with 22 repeats. Unfortunately, LB cells do not efficiently translate RGS8 message, so that Western blots did not allow detection of RGS8 protein in LB cells.
To test whether reduction of Rgs8 levels induced by mutant ATXN2 could be recapitulated in vitro, we measured steady-state levels of RGS8 mRNA and protein in hygromycin selected enriched SH-SY5Y cells expressing Flag-tagged ATXN2-Q22, -Q58 or -Q108. Western blot analyses of whole cell extracts indicated that expression of ATXN2-Q58 or Q108 resulted in decreased RGS8 levels compared to control or ATXN2-Q22 (Fig 8A). To exclude that decreased RGS8 levels were a consequence of selective cellular toxicity of ATXN2-Q58 or -Q108 expression, we measured expression of endogenous DDX6 and PABPC1, which have been shown to interact with ATXN2 [6,8] and CUG-BP1, a nuclear protein by Western blot analysis. The levels of DDX6, PABPC1 and CUG-BP1 were not altered (Fig 8A) strongly supporting that the effect of mutant ATXN2 was specific to RGS8. In parallel, qRT-PCR analyses of SH-SY5Y cell lines expressing Flag-tagged wild-type and mutant ATXN2 demonstrated a moderate reduction of RGS8 mRNA in cell expressing Flag-ATXN2-Q108 (Fig 8B).
Decrease of RGS8 levels in mutant BAC mice could be the result of transcriptional control, mRNA stability and processing or translational control. In contrast to other polyQ proteins, ATXN2 does not enter the nucleus [19] and protein interaction studies have not yielded proteins thought to be involved in transcriptional control. To examine translation of RGS8, we expressed exogenous RGS8 in hygromycin selected SH-SY5Y cells expressing Flag-tagged ATXN2-Q22, -Q58 or -Q108. MYC-tagged RGS8 cDNA including 5’ and 3’ UTRs was cloned under the transcriptional control of the CMV promoter. Forty-eight hrs post-transfection, Western blot analyses revealed that the levels of exogenous RGS8 were significantly decreased in cells expressing ATXN2-Q58 or -Q108 compared with cells expressing wild-type ATXN2-Q22 (Fig 8C). To control for equal transfection, we monitored levels of GFP, which was expressed as an independent cassette in the plasmid. Thus, presence of mutant ATXN2 reduced RGS8 protein levels in vivo and in vitro.
Reduced protein levels potentially out of proportion to reduced mRNA levels in vivo and in vitro suggested to us that ATXN2 might be directly involved in the translation or stability of specific mRNAs. In addition, ATXN2 is known to interact with RNAs through a “Like Sm (LSm) domain” [34–36]. It also interacts with cytoplasmic poly(A)-binding protein 1 (PABPC1) and assembles with polysomes [6,7]. Therefore, we first tested interaction of ATXN2 with RGS8 mRNA and then performed in vitro translation assays in the presence of wild-type and mutant ATXN2.
We performed Protein-RNA immunoprecipitation (IP) experiments in cultured SH-SY5Y cells overexpressing Flag-tagged ATXN2 containing Q22 or Q108. Whole cell extracts were incubated with Flag-mAb-beads and immunoprecipitates were washed with a buffer containing 200 mM NaCl. Bound protein-RNA complexes were eluted from the beads by Flag peptide competition. The IP products were divided equally into two aliquots and one aliquot was analyzed by Western blot. As shown in Fig 9A, the eluted proteins showed co-IP of DDX6 and PABPC1, which are known to interact with ATXN2 [6,8]. To identify RNAs that immunoprecipitated with ATXN2, the extracted RNAs from the second aliquot were subjected to RT-PCR and qPCR analyses. Our results showed that RGS8 mRNA precipitated with ATXN2-Q22 and ATXN2-Q108 (Fig 9A and 9B). Binding of RGS8 mRNA with ATXN2-Q108, however, was significantly reduced compared with ATXN2-Q22 in three independent experiments.
We next proceeded to examine in vitro RGS8 translation. For that purpose, we performed assays using Flag-tagged ATXN2 with Q22 or Q108, respectively, and determined RGS8 protein abundance by Western blot analysis. In three independent experiments, one of which is shown in Fig 9C, levels of RGS8 decreased significantly in the presence of ATXN2-Q108 when compared with the levels in the presence of ATXN2-Q22. No significant alteration in the levels of RGS8 synthesis was detected between ATXN2-Q22 and control extracts (Fig 9C and 9D). These results suggest a role for ATXN2 in translational regulation and a dysregulation of this process in the presence of mutant ATXN2.
We developed new BAC-SCA2 transgenic mouse lines and showed that a protein involved in G-coupled protein signaling was significantly down-regulated supporting our model of mGluR1-mediated enhanced Ca2+ release in SCA2. Both BAC transgenic lines express human full-length ATXN2 with Q22 or Q72 under the control of endogenous human regulatory elements. BAC-Q72, but not BAC-Q22 mice, showed motor function deficits accompanied by changes in PC morphology and steady-state mRNA levels.
Pursuing transcriptome changes in Pcp2-ATXN2[Q127] mice, we demonstrated that expression of mutant ATXN2 was associated with decreased expression of RGS8 both in vivo in BAC mice and in human cell culture models. Reduced RGS8 expression was the result of reduced interaction of RGS8 transcript with mutant ATXN2 protein and reduced in vitro translation.
Mouse models generated with tissue specific or strong promoters facilitate the evaluation of functional and anatomical consequences in many neurological disorders. The Purkinje cell protein 2 (Pcp2) and the Prion protein (PrP) promoters have been used to generate mouse models for polyQ ataxias such as SCA1, SCA2 and SCA3 [19,20,37–41]. For instance, the use of the Pcp2 promoter for expressing mutant ATXN1 or ATXN2 has been shown to recapitulate the progressive cellular and functional phenotype of human SCA1 or SCA2 [19,20,37].
Use of a BAC-transgenic approach resulted in a more widespread expression of the transgene mirroring prior observations of endogenous ATXN2 expression in mouse and human [1]. The control regions included in our BAC transgene specified expression in CNS and non-CNS tissues (Fig 1B). In the CNS, expression was seen in the cerebral hemispheres, cerebellum and spinal cord. This is consistent with expression of endogenous mouse Atxn2 [1] and in situ hybridization data as shown in the Allen Brain Atlas [28].
In the cerebellum, expression of the BAC-transgene was seen in PCs, but also in granule cells, and neurons of the dentate nucleus (Fig 2). As the transgenes were not tagged, we used LCM to establish transgene expression in these sub-regions of the cerebellum. Future physiological experiments using the cerebellar slice preparation will need to examine what role mutant ATXN2 plays in granule cells and dentate nucleus and in overall cerebellar dysfunction in comparison with the PC-targeted expression of mutant ATXN2 [20].
Motor function deficits are common to all SCA2 mouse models, although their ages of onset differ. The accelerating rotarod is used to measure motor coordination and motor learning over a number of days. Our BAC-Q72 mice developed progressive motor deficits beginning at 16 weeks of age (Fig 3B). The motor phenotype of our BAC-Q72 mice was intermediate to that of our Pcp2-ATXN2[Q58] and Pcp2-ATXN2[Q127] mice, although transgene copy numbers and precise developmental expression patterns are difficult to compare.
As with our Pcp2-ATXN2[Q22] line [19], the BAC-Q22 line did not show a motor or cellular phenotype. This study now extends these observations to mRNA measurements of key PC genes out to 45 weeks of age (Fig 5A). Lack of mRNA changes in BAC-Q22 are likely not due to differences in expression levels between lines, as transgenic ATXN2 had higher expression in the BAC-Q22 than in the BAC-Q72 line, both at the level of mRNA and protein (Fig 1C and 1D). Lack of any changes in genes that are typically altered early in Pcp2-ATXN2[Q127] and BAC-Q72 supports the notion that simple overexpression of human wild-type ATXN2 does not cause significant PC pathology.
In contrast, motor function deficits in Atxn2-CAG42 knock-in mice were not evident until the age of 18 months [21]. By comparing the motor functions in these four SCA2 transgenic mouse models, it is apparent that motor function deficits are dependent on CAG repeat length. Consistent with this interpretation, knock-in Atxn1-CAG78 SCA1 mice developed neither ataxic behavior nor a neuropathological phenotype [42], while knock-in Atxn1-CAG154 SCA1 mice did [43]. Our BAC-Q72 transgenic mouse model, although generating lower levels of mutant ATXN2 expression in the cerebellum, develop motor deficits that resemble findings in human SCA2 patients. These observations validate the notion that SCAs can be accurately modeled in mice.
Animal models for several polyQ diseases have shown alteration of body weight [21,43–45]. In this study, BAC transgenic mice demonstrated reduced body weights. The magnitude was similar to knock-in Atxn2-CAG42 mice and Atxn1-Q154/2Q mouse models [21,43]. On the other hand, mice lacking Atxn2 exhibit obesity as a consequence of insulin resistance and altered lipid metabolism pathways [16,17,46]. Increased weight loss due to reduced body fat has also been reported in other polyglutamine diseases, including Huntington disease [47,48]. Of note, reductions in body weight were similar for BAC-Q22 and BAC-Q72 mice suggesting that with regard to the body weight phenotype a simple gain of function may be operative that is mirrored by obesity in loss of function models.
RGS proteins comprise a large family of more than 20 members that negatively modulate heterotrimeric G protein signaling. They share a homologous RGS domain that functions to activate the GTPase of Gα proteins. RGS8 is widely expressed in testis, brain, and cerebellar Purkinje cells [56,57]. Mice lacking Rgs6 or Rgs9 exhibit motor function deficits and ataxia [32,33]. Rgs8 knock-out mice were viable, fertile, and showed normal development, but have not been tested in detail for motor behaviors or PC morphology [57].
Given the importance of a dysregulated mGluR1-ITPR1 axis in SCA2 pathology [13,58], reduction in RGS proteins could further increase abnormally enhanced mGluR1 signaling. We therefore examined RGS8 abundance in BAC-Q72 mice and Epstein-Barr virus immortalized human lymphoblastoid B (LB)-cells from SCA2 patients (Fig 7). The results demonstrated that Rgs8 transcripts and protein abundance were significantly decreased in BAC-Q72 mice (Fig 7A and 7B). Consistent with this, SCA2-LB cells also demonstrated decreased RGS8 transcripts (Fig 7C). Next, we developed an in vitro model using SH-SY5Y cells. Overexpression of mutant ATXN2 resulted in downregulation of RGS8 and this phenomenon was not seen for other known ATXN2 interactors (Fig 8).
As protein levels appeared somewhat depressed out of proportion to the observed reduction in steady-state mRNA levels, we hypothesized that ATXN2 might regulate translation of mRNAs directly. Consistent with this hypothesis, we showed that both wild-type and mutant ATXN2 immunoprecipitated RGS8 mRNA in human cell culture and that this interaction was weaker for mutant ATXN2 (Fig 9A and 9B). This was also reflected in in vitro translation assays as presence of an expanded polyQ tract in ATXN2 reduced translation (Fig 9C and 9D).
Our observations are consistent with studies of the Drosophila homolog of ATXN2 (Atx2). Atx2 regulates PERIOD (PER) translation by interacting with TWENTY-FOUR (TYF) that is required for circadian locomotor behavior. Depletion of Atx2 or expression of mutant Atx2 protein blocked the recruitment of PABP to the TYF-containing protein complex and decreased abundance of PER, thereby altering behavioral rhythms [59,60]. ATXN2 interactions with polyA-binding protein 1 (PABPC1), the splicing factor A2BP1/FOX1 and poly-ribosomes further support roles for ATXN2 in RNA metabolism [5–7]. Depletion of PABP from a cell free extract prevented initiation of mRNA translation [61]. Our studies now extend these observations to mammalian systems and to a gene abundantly expressed in PCs. It is quite likely that Rgs8 will be just one member of a larger set of mRNAs whose expression is regulated by ATXN2.
Aberrant RNA metabolism including processing, degradation, and translation is now recognized to play an important role in neurodegenerative diseases. Among these diseases are amyotrophic lateral sclerosis (ALS), Spinal Muscular Atrophy (SMA) and Fragile X syndrome (FXS) [62–70]. Although ATXN2 had been implicated in steps regulating mRNA translation and formation of stress granules [8,71,72], to our knowledge we describe for the first time a significant difference in these functions between wild-type and mutant ATXN2. Our observations may also have implications for ALS as long normal ATXN2 alleles are a risk factor for ALS [18,73] and some individuals with full mutant ATXN2 alleles may present as ALS [74].
In summary, BAC-SCA2 transgenic mice represent the first animal model with expression of mutant full-length human ATXN2 under the control of its endogenous human promoter including intronic regulatory sequences. These sequences resulted in widespread expression of ATXN2 mirroring expression of endogenous Atxn2. Expression of mutant ATXN2-Q72, but not wild-type ATXN2-Q22, led to a progressive motor deficit, accompanied by morphological and transcriptome changes. As previously demonstrated in C. elegans and the fly [6, 59,60,75], ATXN2 may exert translational control upon a subset of mRNAs. We showed in two independently generated models that presence of mutant ATXN2 in vivo resulted in reduced steady-state levels of RGS8 mRNA and even further reduction in RGS8 protein. ATXN2 coprecipitated with RGS8 mRNA and mutant ATXN2 reduced translation of RGS8 mRNA. RGS proteins can act via Gαq on G-protein coupled receptors. As mutant ATXN2 enhances Ca2+ release from the endoplasmic reticulum (ER) via its abnormal interaction with ITPR1, reduction of RGS8 might be predicted to further increase intracellular Ca2+ by prolonging mGluR1 stimulated Ca2+ release. Our studies now provide a framework to further examine the aberrant mGluR1-ITPR1 axis in SCA2 pathogenesis.
Human lymphoblastoid B (LB)-cells from SCA2 patients and unaffected normal controls were used. All subjects gave written consent and all work was approved by the Institutional Review Board at the University of Utah under IRB# 00035351 and IACUC- University of Utah IACUC committee, protocol number 13–0004. BAC-SCA2 mice were maintained in FVB background and bred and maintained under standard conditions consistent with National Institutes of Health guidelines and approved by the University of Utah, IACUC protocol.
A 169 kb of RP11-798L5 BAC clone (Empire Genomics., USA) containing the 150 kb human ATXN2 locus was engineered to replace the endogenous ATXN2 exon-1 CAG22 with CAG72 repeats. The BAC DNA was prepared according to published protocols [76,77] and microinjected into FVB fertilized eggs to produce transgenic mice at the University of California Irvine (UCI) Mouse Core Facility. BAC-SCA2 mice were maintained in the FVB background and bred and maintained under standard conditions consistent with National Institutes of Health guidelines and approved by the University of Utah, IACUC protocol. For genotyping of BAC-SCA2 transgenic mice, DNA was isolated from mice tails using Qiagen genomic DNA extraction kit (Qiagen Inc., USA) and genotyping PCR was performed. Three primer sets were used to identify the transgene and the primer sequences are follows: P3 forward: 5’-AATTTATGTGATGTT CACTGTTTCTTCC-3’, P3 reverse: 5’-TACGGTCCCTCCAAATAGTGTTAC-3’, P7 forward: 5’-TCTTTTTACAGTACAAGCCCACCACC-3’, P7 reverse: 5’-TTCAAAATG CACCCTTAGCACACCTG-3’, SCA2-A forward: 5’-GGGCCCCTCACCATGTCG-3’, SCA2-B reverse: 5’-CGGGCTTGCGGACATTGG-3’. For all experiments wild-type and transgenic animals were kept as littermates. From 3 to 5 litters were used per experiment dependent on actual size of litters.
Mice were deeply anesthetized with isoflurane. Mouse cerebella were removed and immediately submerged in liquid nitrogen. Tissues were kept at −80°C until the time of processing. Total RNA was extracted from mouse cerebella using the RNeasy Mini Kit (Qiagen Inc., USA) according to the manufacturer’s protocol. DNAse I treated RNAs were used to synthesize cDNAs using the ProtoScript cDNA First Strand cDNA Synthesis Kit (New England Biolabs Inc., USA). Primers for RT-PCR were designed to prevent amplification from genomic DNA (annealing sites in different exons or across intron-exon boundaries). Human ATXN2 primer sites were in exon 1 and exon 5, including Exon 1-F (5’-CTCCTCGGTGGTCGCGGCGACCTC-3’) and Exon 5-R (5’-CTCTTTTTGCATAACT GGAGTCC-3’). ATXN2 primers for amplifying CAG repeats were SCA2-A (5’-GGGCCCCTCACCATGTCG-3’) and SCA2-B (5’-CGGGCTTGCGGACATTGG-3’). Gapdh primers were GAPDH-F (5’-TGAAGGTCGGA GTCAACGGATTTGG-3’ and GAPDH-R (5’-GGAGGCCATGTGGGCCATGAG-3’). Gapdh amplification was conducted in parallel as an internal control for RNA quality and was also employed to evaluate quality the reverse transcriptase reactions. Quantitative RT-PCR was performed in Bio-Rad CFX96 (Bio-Rad Inc., USA) with the Power SYBR Green PCR Master Mix (Applied Biosystems Inc, USA). PCR reaction mixtures contained SYBR Green PCR Master Mix and 0.5 pmol primers and PCR amplification was carried out for 45 cycles: denaturation at 95°C for 10 sec, annealing at 60°C for 10 sec and extension at 72°C for 40 sec. The threshold cycle for each sample was chosen from the linear range and converted to a starting quantity by interpolation from a standard curve run on the same plate for each set of primers. All gene expression levels were normalized to the Actin or Gapdh mRNA levels. Primer pairs designed for qRT-PCR are given as forward and reverse, respectively, and listed in supplementary table (S5 Table).
Cerebella from 8 weeks old wild-type and BAC-Q72 mice (4 animals in each group), and 6 weeks old Pcp2-ATXN2[Q127] and wild-type littermates (16 animals in each group) were used for RNA sequence analyses. Total RNA was isolated using miRNeasy Mini Kit (Qiagen Inc., USA) according to the manufacturer’s protocol. RNA quality was determined using the Bioanalyzer 2100 Pico Chip (Agilent). Samples with an RNA integrity number (RIN) >8 were used for library preparation using Illumina TrueSeq Stranded Total RNA Sample Prep with Ribo-Zero rRNA Removal Kit for mouse. Single-end 50-bp reads were generated on a Hiseq 2000 sequencing machine at the University of Utah Microarray and Genomic Analysis Shared Resource using Illumina Version 4 flow cells. Reads were then aligned to the mouse reference genome (mm10) by Novoalign (http://www.novocraft.com). Quality of RNA sequencing was extremely high with an average of twenty eight million reads for BAC-Q72 and twenty two million reads for Pcp2-ATXN2[Q127]. Ninety eight percent of the reads for both sets of RNAs were aligned to the reference mouse genome. After read alignment, differentially expressed genes were identified using the DRDS application (version 1.3.0) in the USeq software package (http://useq.sourceforge.net/). Gene Ontology (GO) annotations were obtained for all differentially expressed genes (p<0.05). GO enrichment results were obtained using the software DAVID [78,79]. Overlap of BAC-Q72 and Pcp2-ATXN2[Q127] molecular function GO annotations was accomplished using only level 5 categories (p<0.05).
SH-SY5Y cells were cultured and maintained in DMEM media containing 10% fetal bovine serum. Epstein-Barr virus immortalized human lymphoblastoid B (LB)-cells from SCA2 patients and unaffected normal controls were cultured in RPMI 1640 medium supplemented with 15% fetal bovine serum, penicillin and streptomycin. All subjects gave written consent and all work was approved by the Institutional Review Board at the University of Utah.
Protein extracts were prepared by homogenization of mouse cerebella in extraction buffer (25 mM Tris-HCl pH 7.6, 300 mM NaCl, 0.5% Nonidet P-40, 2 mM EDTA, 2 mM MgCl2, 0.5 M urea and protease inhibitors; Sigma; cat# P-8340) followed by centrifugation at 4°C for 20 min at 16,100 × g. Only supernatants were used for Western blotting. Cellular extracts were prepared by the single-step lyses method [80]. The cells were harvested and suspended in SDS-PAGE sample buffer (2x Laemmli Sample Buffer; Bio-Rad; cat# 161–0737) and then boiled for 5 min. Equal amounts of the extracts were subjected to Western blot analysis to determine the steady-state levels of proteins using the antibodies listed below. Protein extracts were resolved by SDS-PAGE and transferred to Hybond P membranes (Amersham Bioscience Inc., USA). After blocking with 5% skim milk in 0.1% Tween 20/PBS, the membranes were incubated with primary antibodies in 5% skim milk in 0.1% Tween 20/PBS for 2 hrs at room temperature or overnight at 4°C. After several washes with 0.1% Tween 20/PBS, the membranes were incubated with the corresponding secondary antibodies conjugated with HRP in 5% skim milk in 0.1% Tween 20/PBS for 2 hrs at room temperature. Following three additional washes with 0.1% Tween 20/PBS, signals were detected by using the Immobilon Western Chemiluminescent HRP Substrate (Millipore Inc., USA; cat# WBKLSO100) according to the manufacturer’s protocol. The following antibodies were used throughout the study. ATXN2 mAb [(1:3000), BD Biosciences Inc.; cat# 611378], 5TF1-1C2 mAb [(1:3000), Millipore Inc.; #MAB1574], RGS8 rabbit polyclonal Ab [(1:3000), Novus Biologicals; #NBP2-20153], Calbindin-D-28K mAb [(1: 5000), Sigma Inc.; cat# C9848], PCP2 mAb [(1: 5000), Santa Cruz Inc.; cat# sc-137064], DDX6 rabbit polyclonal Abs [(1:4000), Santa Cruz Inc.; cat# sc-27127-R], PABPC1 mAb [(1:4000), Santa Cruz Inc.; cat# sc-27127-R], CUG-BP1 mAb [(1:4000), Santa Cruz Inc.; cat# sc-20003], Flag M2 mAb [(1:10,000), Sigma Inc.; cat# F3165], GFP mAb [(1:3000), Santa Cruz Inc.; cat# sc-9996] and MYC mAb conjugated with HRP [(1:5000), Invitrogen Inc.; cat# A3858]. To control for protein quality and loading, the membranes were re-probed with β-Actin mAb conjugated with HRP [(1:10,000), Sigma Inc.; cat# A3858]. The secondary antibodies were goat anti-mouse IgG-HRP [(1:5000), Sigma Inc.; cat# A2304], and donkey anti-rabbit IgG-HRP [(1:5000), Santa Cruz Inc.; cat# sc-2057].
Motor behavior of SCA2 mice was determined using the accelerating rotarod. Cohorts were age matched prior to all behavioral experiments. Male and female mice performed equally well; therefore, data were pooled and gender differences were not evaluated further. The motor performance of BAC-Q22 and BAC-Q72 mice and wild-type littermates were evaluated using the accelerating rotarod (Ugo Basile) according to our published protocol [20]. For mice clinging to the rod, the time at which a mouse had completed 5 rotations was taken as the final latency.
Mice were deeply anesthetized with isoflurane, then transcardially perfused with ice-cold phosphate buffered saline (PBS). Tissue was quickly removed and submerged into cold 4% paraformaldehyde (Electron Microscopy Sciences) and kept at 4°C overnight. The following day, PFA was replaced with 10 mM sodium citrate pH 6.0, and then incubated at 4°C overnight, after which the tissue was exposed to microwave radiation 3 times in 10 sec bursts. Following microwave radiation, tissues were cryoprotected by incubating in 20% sucrose in PBS overnight followed by 30% sucrose overnight both at 4°C. Then the samples were mounted in Tissue-Tek O.C.T. Compound (Sakura Finetek) and stored at -80°C until the time of sectioning. Tissue sections were cut into 20 μM thick slices and floated into cold PBS. Tissues were washed 3 times with PBS at RT for 15 min each time. Free-floating sections were incubated with blocking/permeabilization solution consisting of 5% skim milk, 0.3% Triton X-100 in PBS for 4 hr at RT. Sections were then incubated overnight at 4°C with primary antibodies, calbindin-28kDa mAb at 1:200 dilution. After 3 washes in PBS at 15 min each, sections were incubated with DyLight-550 (Red) (Thermo Fischer Scientific) fluorescent secondary antibodies at 1:500 dilution for 2 hr at RT. Following incubation, the sections were washed 3 times with PBS, and the sections were transferred to Superfrost Plus microscope slides (Fischer Scientific) and mounted with Prolong Gold (Invitrogen). Sections were imaged using confocal microscope (Nikon Eclipse Ti microscopy) and analyzed by Nikon EZ-C1 software. PCs were counted in parasagittal slices from 3 mice in each group.
Fresh whole cerebella from wild type or BAC-Q22 or BAC-Q72 mice was freeze-mounted in O.C.T. and sectioned onto Arcturus PEN Membrane glass slides. Sections were fixed and H&E stained using the Fast Frozen Stain Kit (EMS). Sections on slides were then dehydrated by passage through a solution series of 95% ethanol, 100% ethanol, and then xylene. Prepared slides were stored in a desiccated chamber until needed. LCM was performed using the Arcturus Veritas LCM system. RNAs were prepared from tissue on LCM caps (CapSure, Applied Biosystems) using the Arcturus PicoPure RNA Kit (Applied Biosystems Inc., USA). RNA yield was typically 5 μg/cap. cDNA was then prepared by using the ProtoScript M-MuLV First Strand cDNA Synthesis Kit (NEB Inc., USA) and used for qRT-PCR as described in Methods above.
To identify proteins and RNAs that bind to ATXN2, we carried out protein-RNA immunoprecipitation (IP) experiments from lysates of SH-SY5Y cells expressing Flag-ATXN2-Q22 and Flag-ATXN2-Q108. Whole cell extracts were prepared by the two-step lyses method [80]. First, cells were lysed with a cytoplasmic extraction buffer (25 mM Tris-HCl pH 7.6, 10 mM NaCl, 0.5% NP40, 2 mM EDTA, 2 mM MgCl2, protease and RNAse inhibitors) and cytoplasmic extracts were separated by centrifugation at 14,000 RPM for 20 min. Second, the resultant pellets were suspended in nuclear lysis buffer or high salt lyses buffer (25 mM Tris-HCl, pH 7.6, 500 mM NaCl, 0.5% Nonidet P-40, 2 mM EDTA, 2 mM MgCl2, protease and RNAse inhibitors), and the nuclear extracts were separated by centrifugation at 14,000 RPM for 20 min. The nuclear extracts were then combined with the cytoplasmic extracts and denoted as whole cell extracts. Specifically, while combining cytoplasmic and nuclear extracts, the NaCl concentration was adjusted to physiologic buffer conditions (~150 mM) to preserve in vivo interactions. Ninety percent of cell extracts were subjected to Flag monoclonal antibody (mAb) IP (Anti-Flag M2 Affinity Gel, Sigma Inc.; cat# A2220-1ML) to immunoprecipitate ATXN2 interacting protein-RNA complexes. The remaining 10% of whole cell extracts were saved as the input control for Western blotting and RT-PCR analyses. The IPs were washed with a buffer containing 200 mM NaCl and the bound protein-RNA complexes were eluted from the beads with Flag peptide competition (100 μg/ml). Eluted fractions were divided into two equal parts. One part was analyzed by SDS-PAGE followed by Western blotting to determine the efficiency and quality of immunoprecipitation. RNA was isolated from the other fraction and subjected to RT-PCR and qRT-PCR analyses to identify RNAs that bound to wild type or mutant ATXN2.
To determine the role of ATXN2 on RGS8 mRNA translation, in vitro translation assays were performed using the rabbit reticulocyte lysate-based cell free TNT Quick Coupled Transcription/Translation Kit (Promega Inc., USA) according to the manufacturer’s instructions, with minor modifications. Briefly, 1 μg of cDNA plasmids of LacZ (control) and Flag-tagged ATXN2 expressing Q22 or Q108 were added to 20 μl of the rabbit reticulocyte lysate kit component, including 20 μM amino acids in a total volume of 25 μl. The translation reaction was carried out for 2 hr at 30°C. Next 1 μg of RGS8 cDNA plasmid was added to each translation reaction with fresh rabbit reticulocyte lysate containing 20 μM amino acids in a total volume of 50 μl, and incubated further at 30°C for 4 hr. Translation assays was analyzed by SDS-PAGE followed by Western blot analyses.
For Western blot analyses, the experiments were performed at least three times, and wherever appropriate gel films were scanned and band intensities were quantified by ImageJ analyses. The p values were calculated by pairwise Student’s t-tests. Student’s t-tests were also used to compare mRNA steady state levels between BAC and wild-type mice determined by qRT-PCR. The level of significance was set at p<0.05. In the figures, a single asterisk indicates p<0.05, a double asterisk p<0.01, a triple asterisk p<0.001, and ns represents p≥0.05. For accelerating rotarod analyses, repeated measures ANOVA was used with post-hoc t-tests to compare means.
|
10.1371/journal.pgen.1000165 | Divergent Evolution of CHD3 Proteins Resulted in MOM1 Refining Epigenetic Control in Vascular Plants | Arabidopsis MOM1 is required for the heritable maintenance of transcriptional gene silencing (TGS). Unlike many other silencing factors, depletion of MOM1 evokes transcription at selected loci without major changes in DNA methylation or histone modification. These loci retain unusual, bivalent chromatin properties, intermediate to both euchromatin and heterochromatin. The structure of MOM1 previously suggested an integral nuclear membrane protein with chromatin-remodeling and actin-binding activities. Unexpected results presented here challenge these presumed MOM1 activities and demonstrate that less than 13% of MOM1 sequence is necessary and sufficient for TGS maintenance. This active sequence encompasses a novel Conserved MOM1 Motif 2 (CMM2). The high conservation suggests that CMM2 has been the subject of strong evolutionary pressure. The replacement of Arabidopsis CMM2 by a poplar motif reveals its functional conservation. Interspecies comparison suggests that MOM1 proteins emerged at the origin of vascular plants through neo-functionalization of the ubiquitous eukaryotic CHD3 chromatin remodeling factors. Interestingly, despite the divergent evolution of CHD3 and MOM1, we observed functional cooperation in epigenetic control involving unrelated protein motifs and thus probably diverse mechanisms.
| Epigenetic regulation of transcription usually involves changes in histone modifications, as well as DNA methylation changes in plants and mammals. Previously, we found an exceptional epigenetic regulator in Arabidopsis, MOM1, acting independently of these epigenetic marks. Interestingly, MOM1 controls loci associated with bivalent chromatin marks, intermediate to active euchromatin and silent heterochromatin. Such bivalent marks are often associated with newly inserted and/or potentially active transposons, silent transgenes, and certain chromosomal loci. Notably, bivalent chromatin seems to be characteristic for embryonic stem cells, where such loci change their activity and determination of epigenetic marks during cell differentiation. Here, we provide evidence that in vascular plants, the MOM1-like proteins evolved from the ubiquitous eukaryotic chromatin remodeling factor CHD3. The domains necessary for CHD3 function degenerated in MOM1, became dispensable for its gene silencing activity, and were replaced by a novel, unrelated domain providing silencing function. Therefore, MOM1-like proteins use a different silencing mechanism compared to the ancestral CHD3s. In spite of this divergent evolution, CHD3 and MOM1 seem to retain a functional cooperation in control of transcriptionally silent loci. Our results provide an unprecedented example of an evolutionary path for epigenetic components resulting in increased complexity of an epigenetic regulatory network characteristic for multicellular eukaryotes.
| TGS heritably suppresses transcription of repetitive elements, transgenes and chromosomal genes and is generally associated with repressive histone marks and hypermethylation of DNA. Mutations in Arabidopsis that affect such marks lead to the release of TGS [1]. Thus transcriptionally silent or active states of chromatin are thought to be regulated by changes in DNA and by modifications of histones. Contradicting this general view, mom1 mutations release silencing without obvious changes in DNA methylation, histone modification or degree of chromatin condensation [2],[3]. Analysis of genetic interactions between mom1 and the ddm1 mutation, which results in a severe decrease in DNA methylation and the relocation of histone modifications, suggested that MOM1and DDM1 act in independent but mutually reinforcing silencing pathways [4]. Moreover, DDM1 and MOM1 control TGS at overlapping targets that are reactivated when only a single pathway is compromised [5],[6]. Interestingly, a MOM1-specific subset of silencing targets has chromatin properties intermediate between hetero- and euchromatin. Thus, although silent these genes are poised for activation [6]. Similar bivalent chromatin properties have been found at several chromosomal loci in mammalian stem cells prior to their differentiation [7]. Unfortunately, mammalian epigenetic regulators responsible for controlling the transcriptional status of the intermediate chromatin have not been identified and MOM1 is the only example so far of a regulator determining the transcriptional status of targets associated with bivalent epigenetic marks.
MOM1 shares sequence homology with many proteins in a region containing a partial SNF2 domain [8]. SNF2 domains are found in ATP-dependent chromatin remodeling proteins involved in transcriptional control, DNA repair, and recombination. They contain seven conserved sequence motifs found in the superfamily II of DNA/RNA helicases [9]. The spatial structure of the SNF2 domain includes two lobes separated by a cleft [9]. The first lobe comprises helicase motifs I, Ia, II, III and the second includes motifs IV, V, and VI. Since the helicase motifs in the SNF2 sequence of MOM1 correspond only to the second lobe, Amedeo et al. (2000) [8] proposed that MOM1 functions as a heterodimer with an unknown Arabidopsis protein contributing the first SNF2 lobe.
The sequence close to the C-terminus of MOM1 shows similarity to an actin-binding domain (ABD) of chicken tensin [8]. Further predictions based on MOM1 protein sequence revealed a putative transmembrane domain, three putative nuclear localization signals (NLS) and several repetitive regions [8]. However, the functional relevance of all these sequence motifs was obscure.
In the present study, we demonstrate that a protein comprising 12.8% of the original MOM1 retains silencing activity through a novel motif necessary and sufficient for the MOM1 silencing function. The protein lacks all features previously considered important, except the NLS. MOM1-related proteins containing this new motif are present in the genomes of vascular plants but not in the mosses. Closer comparison of MOM1 orthologs suggests that MOM1 diverged, during the evolution of vascular plants, from the CHD3 chromatin remodeling factors common to many eukaryotes. We provide evidence that the two proteins are still able to cooperate in the control of TGS, despite the divergent evolution associated with the creation of a novel, MOM1-specific gene silencing domain and the degeneration of domains essential for CHD3 function.
Earlier homology searches with MOM1 identified no other conserved sequences than the SNF2 domain and an actin-binding region [8]. Since then, many sequences have been added to databases and the genomes of rice (Oryza sativa) and poplar (Populus trichocarpa) have been sequenced and annotated [10],[11]. In the genomes of both these species, we have detected predicted proteins with MOM1 homologies extending beyond the SNF2 domain. In the poplar database, we detected three expressed proteins sharing homology with MOM1, referred to as PtMOM1, PtMOM2 and PtMOM3 (Figure 1). In the rice database, we found two expressed MOM1 homologs, OsMOM1 and OsMOM2 (Figure 1). A gene encoding a predicted MOM1 homologue was also found in the genome of the club moss (Selaginella moellendorffii), referred as SmMOM1 (Figure 1). In addition to closely related SNF2 sequences, alignment of MOM1 and the homologs revealed three further conserved regions shared by these proteins that we named CMM1-3 (for Conserved MOM1 Motif 1-3) (Figure 1 and Figure S1). Noticeably, two poplar (PtMOM1 and 2) and rice MOM1-related proteins encode complete SNF2 domains with all seven helicase motifs. Moreover, several MOM1 homologs contain additional sequence motifs such as a Plant Homeo Domain (PHD) and chromodomains (Figure 1).
As well as in Angiosperms, further database searches revealed CMM-containing proteins that could be predicted from not fully annotated genomic databases of more distant vascular plants like pine (Pinus taeda) (Figure S1 and data not shown). Remarkably, MOM1 homologs seem to be absent from Chlamydomonas reinhardtii and the moss Physcomitrella patens. The apparent conservation of additional MOM1-specific structural features might point towards a role in MOM1-mediated gene silencing. To address this issue, we assessed the functional significance of conserved MOM1 domains in vivo.
Loss of silencing in the mom1-1 mutant (Figure 1), which is predicted to encode a MOM1 protein with a deletion spanning the sequence 1633-2001aa (MOM1Δ1633–2001) [8], implies that the missing section is essential for MOM1 function. Plants homozygous for the mom1-1 allele lose the ability to maintain TGS at previously silenced transgenic and endogenous chromosomal loci such as TSI (for Transcriptionally Silent Information) [3],[8]. In contrast, the previously uncharacterized mom1-4 allele (Figure 1), which is predicted to encode the MOM1 C-terminal truncation of 142 amino acids (MOM1Δ1860–2001), is able to maintain TSI or transgenes silencing (Figure 2A and data not shown). To exclude the possibility that the T-DNA is spliced out of transcripts of the mom1-4 locus, we performed both RT-PCR with primers corresponding to the MOM1 sequence flanking the T-DNA insert and 3′RACE. The results from both assays were consistent with the termination of MOM1 transcripts within the T-DNA insertion (data not shown). Thus, the sequence absent in mom1-4, notably including the ABD and CMM3 (Figure 1), is dispensable for MOM1 function in TGS. Contrasting phenotypes of mom1-1 and mom1-4 provide evidence that a functionally essential domain resides between amino acids 1633–1859. However, effects such as an altered protein structure or the reduced stability of the mom1-1 gene product could not be ruled out. Moreover, it is possible that other parts of MOM1 are also essential for its function, given the presence of the 1633–1859 domain. To address these questions, we performed functional assessment of a series of MOM1 gene deletions.
Although the partial SNF2 domain might be just a nonfunctional remnant of a complete domain of an MOM1 ancestor, as illustrated by the structures of MOM1-related proteins in poplar and rice, its presence may still be required for MOM1 silencing activity. To examine this, we tested a MOM1 deletion derivative lacking SNF2-related sequences as a substitute for the wild-type MOM1 protein. For this and the subsequent assays we used transgenic complementation tests of two mom1 mutant alleles: mom1-1 (MOM1Δ1633–2001) discussed above and mom1-2 (Figure 1, Figure 2A). In mom1-2, the T-DNA insertion is predicted to result in truncation of more than 85% of the protein-coding sequence (T-DNA insertion after encoding 292 aa, Figure 1). Moreover, the MOM1 transcript is undetectable in mom1-2 (Figure 2B). These features suggest that mom1-2 is a null allele. When successful, the transgenic complementation tests should restore silencing of TSI sequences in the strains with mom1 mutant alleles [3]. We introduced a modified MOM1 gene encoding MOM1Δ511–837 lacking all three helicase motifs IV, V and VI into mom1 mutants. This truncated MOM1 gene re-established silencing of TSI (Figure 2C). The apparent dispensability of the SNF2 domain unequivocally demonstrates that this domain and thus any presumed chromatin remodeling activity is not involved in MOM1-mediated silencing.
In order to assess the functional significance of other MOM1 protein sequence motifs, we performed systematic deletion/complementation analysis as for the SNF2 motif described above.
First, we constructed a series of deletions 5′ to the SNF2 region. This area encodes the first of the two repeated sequences (RS1) and an homologous region conserved between Arabidopsis, poplar and Selaginella but not present in rice MOM1 proteins (Figure 1). RS1 is composed of two repeats sharing higher similarity at the nucleotide (85.3%) than at the amino acid sequence level (78.0%) (data not shown), indicating a relatively recent duplication event. Both MOM1Δ125–345,511–837, lacking both RS1 and SNF2-related sequences, as well as MOM1Δ179–836, lacking RS1, the SNF2 domain and sequence linking those two elements, retained silencing activity (Figure 2C). Similarly, we assessed the functional importance of the N-terminal MOM1 sequence homologous to poplar MOM1- related proteins and Selaginella MOM1 (Figure 1). Since the predicted translation initiation of MOM1 is only five nucleotides after the first intron/exon junction, the most N-terminal deletion was introduced only after the first 22 amino acids of the predicted MOM1 sequence to ensure correct splicing. The resulting MOM1Δ23–122,511–837 was able to almost completely complement mom1 mutants (Figure 2C). This demonstrates that the N-terminal conserved sequence is also dispensable for MOM1-mediated gene silencing or has very minor contribution.
We next constructed a series of deletions 3′ to the SNF2 region. This area encodes the second of the two repeated sequences (RS2) and three CMM motifs conserved in poplar, rice and other vascular plants (Figure 1, Figure S1, and data not shown). MOM1Δ881–1225 lacking CMM1 retained its silencing activity, indicating that CMM1 is also dispensable for TGS (Figure 2C). The silencing activity of MOM1Δ881–1225 contradicts the previously proposed hypothesis that MOM1 acts in association with the nuclear membrane, as the deletion encompasses also the previously predicted transmembrane domain [8]. The repetitive sequence RS2 was targeted in two constructs encoding MOM1Δ1228–1557 and MOM1Δ1560–1666 carrying successive deletions of two parts of RS2 (Figure 2C). Additionally, the MOM1Δ1228–1557 deletion covers a non-conserved sequence residing 3′ to CMM1 and 5′ to RS2. Successful complementation with all these constructs showed that this entire area of the gene is functionally dispensable.
Previous predictions based on the MOM1 sequence revealed three potential NLS sequences [8]; however, re-examination with recent NLS prediction algorithms [12] confirmed only one NLS (Figure 1). This NLS is bordered on both sides by functionally dispensable sequences examined in MOM1Δ125–345,511–837, MOM1Δ179–836 and MOM1Δ881–1225 proficient in TGS (Figure 2C). In contrast, MOM1Δ125–1221 with a large deletion encompassing the NLS together with surrounding, functionally superfluous sequences failed to complement mom1-2 (Figure 2C). This suggests a requirement for NLS and is consistent with the reported nuclear localization of MOM1 [8].
Successful complementation of mom1 with a series of MOM1 truncations in the interval 22–1666 aa, with the exception of the 35 aa spanning the NLS, together with contrasting mom1-1 and mom1-4 silencing properties, suggest that the NLS and the sequence of 197 aa containing CMM2 are necessary and sufficient for the MOM1 silencing function. Therefore, we tested whether such a predicted “miniMOM1” protein (MOM1Δ23–836,872–1662,1860–2001), representing a fusion of these two sequences and comprising less than 13% of the MOM1, retains gene silencing activity (Figure 3A). The complementation tests clearly show that “miniMOM1” retains silencing activity, as reflected by a significant reduction in TSI transcription. However, TSI silencing seems to be incomplete and low levels of TSI RNAs were detected on Northern blots (Figure 3A). TSI consists of highly repeated elements residing in pericentromeric regions of all five Arabidopsis chromosomes. Therefore, it is possible that not all TSI templates are resilenced to completion by “miniMOM1” or that “miniMOM1” may require a minor contribution of conserved N-terminal sequence. Alternatively, the “miniMOM1” itself or its mRNA might be unstable and not able to reach levels allowing for complete TSI silencing. Therefore, we performed protein blots using material of randomly chosen transgenic lines. The “miniMOM1” protein was readily detected (Figure S2) and, thus, insufficient availability of “miniMOM1” cannot be considered as an explanation for incomplete TSI silencing.
To assess more precisely the silencing ability of “miniMOM1”, we introduced it into mom1 mutant strains containing the silent GUS marker locus of line L5 [13]. In these strains, the mutations mom1-1 or mom1-2 release TGS of the GUS transgene (Figure 3A). The GUS transgene was, as TSI, almost completely resilenced upon introduction of “miniMOM1” (Figure 3A). These results confirm the silencing activity of “miniMOM1” and point to CMM2 as the main and possibly the only element clearly essential for the silencing activity of MOM1, given that the NLS is provided.
Available sequences indicate that the genomes of several plant species have genes encoding MOM1 homologs. In poplar, PtMOM2 and PtMOM3 represent truncated derivatives of PtMOM1 (Figure 1). Similarly, we found a transcribed gene in Arabidopsis, hereafter referred to as MOM2 (At2g28240), predicted to encode a protein homologous to the C-terminal part of MOM1. MOM2 retained CMM2 and CMM3; however, it acquired a novel tandemly repeated sequence (RS). The absence of corresponding repeats in MOM1, along with presence of RS2 missing in MOM2, implies that these repeats were acquired independently after MOM1 and MOM2 diverged from a common ancestor. MOM2 lacks NLS and, furthermore, its CMM2 bears mutations in amino acids conserved in other MOM1 homologs (Figure S3 and data not shown). The two tested mom2 mutant alleles mom2-1 (WiscDsLox364H07) and mom2-2 (SAIL548_H02) did not affect TSI silencing (Figure S3). Additionally, mom1 mom2 double mutants had a level of TSI expression similar to that in mom1 (Figure S3), indicating that MOM2 has no silencing function redundant with MOM1. These observations are in agreement with the essential roles of the NLS and the intact CMM2 for gene silencing of MOM1.
CMM2 was detected as one of three regions of MOM1-related proteins that are conserved in addition to SNF2 motifs (Figure 1 and Figure S1). To examine whether this structural conservation also reflects conservation of a silencing function, we replaced CMM2 of Arabidopsis MOM1 by the CMM2 predicted for PtMOM1 of poplar (Figure 3B). We compared mom1-2 complementation ability of the MOM1Δ1560–1666, 1734–1815, 1860–2001 construct lacking CMM2 to the same construct containing CMM2 from poplar. Poplar CMM2 clearly restored the silencing activity of MOM1Δ1560–1666, 1734–1815, 1860–2001, suggesting that PtMOM1 is able to perform gene silencing mediated by its CMM2 (Figure 3B).
The sequence of PtMOM1 also predicts, in addition to an integral SNF2 domain with all six helicase motifs, the presence of a PHD finger and double chromodomains (Figure 1). The combination of PHD fingers, double chromodomains and an SNF2 domain is a distinctive feature of CHD3 proteins (Chromodomain-Helicase-DNA binding) [14]; noticeably plant CHD3-like proteins retained only a single PHD finger domain. The intact SNF2 domain is critical for the silencing function of CHD3 proteins. The long life-span of poplar and continuous production of “ancient” gametes is thought to reduce significantly the speed of genome evolution compared with Arabidopsis (estimated at six times) [11]. Therefore, PtMOM1 presumably reflects a more ancient sequence arrangement than those of the Arabidopsis or rice MOM1 proteins and the presence of all CHD3 domains in PtMOM1 provides strong support for an evolutionary link between MOM1 and CHD3 proteins (Figures 1, 4 and Figure S4). PtMOM1structural features were also found in SmMOM1 (Figure 1) providing additional support to this conclusion.
The Arabidopsis genome contains two genes encoding CHD3-like proteins – PICKLE (PKL) (Figure 4A) and the as yet uncharacterized At5g44800. PKL is required for postembryonic transcriptional suppression of genes involved in embryogenesis [15],[16] and seems to contribute also to the restriction of ectopic meristematic activity [17].
Since MOM1 and PKL likely diverged from a common ancestral CHD3-like gene, we were interested to examine whether their functions may still converge in the control of gene silencing. We combined the pkl and mom1 mutations and compared levels of transcriptional reactivation of TSI in the single and pkl mom1 double mutants (Figure 4B). TSI activated in mom1 remained silent in pkl, suggesting that, in contrast to MOM1, depletion of PKL was not sufficient to release TSI silencing. However, the level of TSI transcripts was increased approximately fourfold in pkl mom1 double mutants compared with the mom1 single mutant. Thus, even though PKL and MOM1 diverged in terms of their active domains, they are still able to cooperate functionally in the control of TGS.
Unexpectedly, we have found that more than 87% of MOM1 protein is dispensable for the gene silencing function, according to the functional analysis of a series of deletion derivatives of the MOM1 gene. We have also demonstrated that a “miniMOM1”, comprising 22 N-terminal amino acids, an NLS and 197 amino acids including CMM2, retains silencing activity, as reflected by drastically reduced levels of TSI expression and almost complete transcriptional suppression at a transgenic GUS locus. Therefore, minor contribution of the N-terminal part of MOM1 to its silencing activity seems to be apparent. In addition, a drastic reduction in protein size leading to alterations in physical properties (e.g. a predicted isoelectric point of 5.2 for MOM1 and 8.8 for “miniMOM1” and a change in net charge from −62 to +4.3) can also contribute to the incomplete silencing mediated by “miniMOM”. Nevertheless, the results of MOM1 deletion analysis and the successful replacement of Arabidopsis CMM2 by the CMM2 of poplar provide strong evidence that CMM2 is the most critical element of the MOM1 protein for its silencing function, not only in Arabidopsis but also in other plants. Obviously other domains, also these clearly dispensable for TSI and transgene silencing, may still be required for epigenetic regulation at other, as yet unidentified, target loci.
Although, MOM1 proteins are CHD3 derivatives, the domains shared with CHD3 chromatin remodeling factors apparently became obsolete after the acquisition of CMM2. This is also evident for PtMOM1, which has a structure largely similar to CHD3 proteins. For example, the SNF2 domain of PtMOM1, shown to be critical for the function of CHD3 proteins, acquired mutations of conserved amino acids essential for CHD3 activity [15],[18],[19] (Figure S5). Several indispensable amino acids are replaced in MOM1 homologs from different plant species and, remarkably, these replacements are identical in MOM1 proteins from different plant species. It is difficult to provide a simple explanation for this unusual sequence drift since the loss of remodeling functions of SNF2 should not be under a direct, strong selection pressure for particular types of mutations. In any case, the pattern of these mutations provides specific signatures to MOM1 SNF2 domains (Figure S5 and S6) and suggests that acquisition of CMM2 and degeneration of SNF2 occurred in species ancestral to vascular plants. The SNF2 domain of Arabidopsis MOM1 underwent the most drastic alterations due to an internal deletion. This relatively recent event seems to be accompanied by the formation of the RS1 sequence duplication. Alignment of Arabidopsis and poplar sequences flanking RS1 suggests that extensive deletion and the formation of RS1 removed not only part of the SNF2 domain but also a PHD finger and chromodomains. Clearly, this event provides the best illustration of the dispensability also of the PHD finger, and chromodomain for the MOM1 silencing function.
CHD3 proteins of human and Drosophila, known as Mi-2, act as components of a multi-subunit chromatin repression complex NuRD (Nucleosome Remodelling and Deacetylating), which combines nucleosome remodelling and histone deacetylation activities [20],[21]. The Arabidopsis genome encodes two CHD3-like proteins: PKL with a potentially functional SNF2 domain and the still uncharacterized At5g44800 with an SNF2 domain containing mutations in amino acids essential for chromatin remodeling activity (Figures S5, S6 and data not shown). PKL is involved in transcriptional repression of genes that are active only at a particular time and place during sporophyte development [15],[16],[17],[22]. However, there is little evidence at present for the involvement of chromatin remodeling and histone deacetylation in PKL-mediated gene repression.
The exact mechanism of MOM1-mediated silencing is not known, but MOM1 and PKL both seem to contribute to transcriptional suppression or restriction of levels of ectopic reactivation of TSI transcription. The multilayer nature of epigenetic regulation and the necessity for backup mechanisms have been documented recently for Arabidopsis gene silencing associated with DNA methylation changes [23]. However, in this case, interaction between the major and evolutionary highly conserved gene silencing mechanisms, such as DNA and histone methylation, was investigated and the backup deficiencies were found to have very drastic developmental consequences indicative of the destabilization of central epigenetic functions. The effects of pkl or mom1 and the combination of these mutations have much more subtle effects. This can be explained by the characteristics of MOM1 targets and their association with bivalent epigenetic marks. The number of such loci is low [6] and their reactivation is likely controlled at multiple levels, as illustrated here by the cooperative activities of MOM1 and PKL. It is remarkable that despite the clearly divergent evolution of MOM1 in terms of protein properties, it has retained its functional relationship to the CHD3 proteins. The CHD3 origin of MOM1 and the silencing in cooperation with PKL suggest that MOM1 function is also linked to histone acetylation changes. Although global changes in histone acetylation properties were not observed in mom1 mutants [2], more subtle target-specific acetylation changes cannot be ruled out.
Whatever the precise molecular mechanism(s) of heritable transcriptional repression mediated by MOM1 might be, it is remarkable that increasing complexity of epigenetic gene regulation has resulted from the emergence of supplementary and cooperating levels and/or mechanisms of epigenetic control. Genomic or cDNA sequences encoding CMM2 are present in many species of vascular plants, even as distant as club-moss Selaginella moellendorffii, pine Pinus taeda and various monocotyledonous and dicotyledonous species. Remarkably, we failed to detect CMM2 in the moss Physcomitrella patens or in green algae Chlamydomonas reinhardtii, although the sequences of both genomes are complete (http://genome.jgi-psf.org/). Therefore, it appears that the emergence of CMM2, and thus MOM1 proteins, coincided with the appearance of vascular plants. Since MOM1-related proteins are not present outside of the plant kingdom, it can be envisaged that novel, highly specialized epigenetic factors and functions can appear only in a narrow subset of organisms through diversification of the general, evolutionary conserved epigenetic regulators. So far the biological role of the CHD3/MOM1 sub-diversification remains unclear but it is intriguing that it seems to have assisted the major evolutionary step in the emergence of land plants.
Total RNA was isolated using the TRI reagent (Sigma) according to the manufacturer's instructions. Detection of TSI was as described previously [3].
The MOM1 genomic sequence was assembled in the binary vector pCAMBIA1301 after elimination of vector sequence EcoRI-BstEII encompassing a multiple cloning site and the β-glucuronidase gene. DNA fragments covering the entire MOM1 gene and ∼2 kb of sequence upstream of the transcription start were detected in the Arabidopsis thaliana Lambda Genomic Library (Stratagene) and assembled in the modified vector.
To create MOM1Δ511–837, MOM1Δ1096–1234, MOM1Δ881–1225, MOM1Δ125–345,511–837, MOM1Δ179–836, MOM1Δ125–1221, MOM1Δ1228–1557, MOM1Δ1560–1666, MOM1Δ23–122,511–837, MOM1Δ23–836,872–1662,1860–2001, the MOM1 gene sequences bordered by restriction sites BstBI-XmaI, AseI-AseI, XmaI-BbvCI, SalI-BstBI, SalI-XmaI, SalI-BbvCI, BbvCI-BlpI, BlpI-BlpI, NcoI-SalI, NcoI-BstBI, respectively, were replaced with oligonucleotide adapters or amplified fragments containing matching restriction sites. To create MOM1Δ125–345,511–837, MOM1Δ179–836, MOM1Δ23–122,511–837 and MOM1Δ23–836,872–1662,1860–2001, the corresponding deletions were introduced into MOM1Δ511–837. Constructs encoding the modified MOM1 proteins were introduced into mom1 mutant plants using the floral dip method [24].
A total protein extract in Laemmli sample buffer was fractionated by 10% SDS/PAGE and blotted onto a Hybond-P membrane (Amersham Pharmacia). Proteins were visualized using the ECL PLUS kit (Amersham Pharmacia) after membrane hybridization with anti-HA antibody (Roche).
Staining was performed on 1-week-old seedlings as described [25]. Quantitative GUS activity assay was performed on 13-day-old plantlets as described [25] with minor modifications.
AtMOM1 (At1g08060); AtMOM2 (At2g28240); PtMOM1 (eugene3.00130053); PtMOM2 (eugene3.00660276); PtMOM3 (eugene3.01310088); OsMOM1 (Os06g01320); OsMOM2 (Os02g02050); SmMOM1 (estExt_fgenesh2_pg.C_110182); PKL (At2g25170).
|
10.1371/journal.pntd.0000241 | Controlling Schistosomiasis: Significant Decrease of Anaemia Prevalence One Year after a Single Dose of Praziquantel in Nigerien Schoolchildren | In the framework of the monitoring and evaluation of the Nigerien schistosomiasis and soil-transmitted helminth control programme, a follow-up of children took place in eight sentinel sites. The objective of the study was to assess the evolution of Schistosoma haematobium infection and anaemia in schoolchildren after a single administration of praziquantel (PZQ) and albendazole.
Pre-treatment examination and follow-up at one year post-treatment of schoolchildren aged 7, 8, and 11 years, including interview, urine examination, ultrasound examination of the urinary tract, and measurement of haemoglobin. Before treatment, the overall prevalence of S. heamatobium infection was 75.4% of the 1,642 enrolled children, and 21.8% of children excreted more than 50 eggs/10 ml urine. Prevalence increased with age. The overall prevalence of anaemia (haemoglobin <11.5 g/dl) was 61.6%, decreasing significantly with increasing age. The mean haemoglobinemia was 11 g/dl. In bivariate analysis, anaemia was significantly more frequent in children infected with S. haematobium, although it was not correlated to the intensity of infection. Anaemia was also associated with micro-haematuria and to kidney distensions. In a sub-sample of 636 children tested for P. falciparum infection, anaemia was significantly more frequent in malaria-infected children. In multivariate analysis, significant predictors of anaemia were P. falciparum infection, kidney distension, and the village. One year after a single-dose praziquantel treatment (administered using the WHO PZQ dose pole) co-administered with albendazole (400 mg single dose) for de-worming, the prevalence of S. haematobium infection was 38%, while the prevalence of anaemia fell to 50.4%. The mean haemoglobinemia showed a statistically significant increase of 0.39 g/dl to reach 11.4 g/dl. Anaemia was no longer associated with S. haematobium or to P. falciparum infections, or to haematuria or ultrasound abnormalities of the urinary tract.
The high prevalence of anaemia in Nigerien children is clearly a result of many factors and not of schistosomiasis alone. Nevertheless, treatment of schistosomiasis and de-worming were followed by a partial, but significant, reduction of anaemia in schoolchildren, not explainable by any other obvious intervention.
| The World Health Organization's recommendation for the control of urinary schistosomiasis is to reduce morbidity by reducing the prevalence of heavy infections. In Niger, where urinary schistosomiasis is endemic along the Niger River valley and in proximity to ponds, a national control programme for schistosomiasis and soil-transmitted helminth was launched in 2004 with the financial support of the Gates Foundation through the Schistosomiasis Control Initiative. In the framework of the monitoring and evaluation of the control programme, a follow-up of school children took place in eight sentinel sites. The aim of this study was to assess the evolution of Schistosoma haematobium infection and associated morbidity after a single-dose administration of praziquantel and albendazole. Before treatment, the overall prevalence of S. heamatobium infection was 75.4% and anaemia (haemoglobin <11.5 g/dl) was present in 61.6% of the study sample. One year after a single-dose praziquantel treatment (administered by dose-pole) co-administered with albendazole (400 mg single dose) for de-worming, all morbidity markers of the infection decreased significantly. This study shows how a schistosomiasis control programme can benefit populations by improving their health status.
| Schistosoma haematobium is endemic in Niger, particularly throughout the Niger River valley and in villages in close proximity to permanent and temporary ponds [1]–[3]. The prevalence of infection is highly variable from one village to another. School-aged children are more often and more heavily infected than adults [4]. Overall, urinary schistosomiasis is an important public health problem in Niger. Thanks to portable ultrasound technology, urinary tract lesions due to S. haematobium infection can be easily identified in large epidemiological surveys [5]–[8]. Nevertheless, the relationship between other frequent pathologies, such as anaemia and S. haematobium infection, are less thoroughly documented [9]. Different studies have found a statistically significant association between urinary schistosomiasis and anaemia [10],[11]. However, many other factors, such as malaria, soil-transmitted helminthiasis, nutritional deficiencies and sickle-cell anaemia, are also frequently associated with anaemia [12]–[15] and controversial statements have been made about the real impact of schistosomiasis on haemoglobin status.
In the framework of the evaluation of the Nigerien national schistosomiasis and soil-transmitted control programme, launched in 2004 with the financial support of the Schistosomiasis Control Initiative (SCI), a longitudinal survey was implemented prior to the first round of mass treatment. This treatment consisted of the administration of praziquantel and albendazole to all the target population (school-aged children and high risk groups) determined by the programme. The double aim of this survey was to collect baseline data on parasitological and morbidity indicators and to monitor their evolution.
Eight villages located in schistosomiasis endemic regions were randomly selected to represent the two main transmission patterns in Niger: six villages located near permanent (Tabalak, Kokorou) or semi-permanent (Kaou, Mozague, Rouafi, and Sabon Birni) ponds and two (Saga Fondo, Sanguile) located along the Niger River. The villages represented the south-western region (Tillabéry) and the central-northern region (Tahoua) of the country, with four villages from each region. One village is located in the Sudanian climatic zone and the seven others are in the Sahelian climatic zone.
Transmission is permanent along the river and near permanent pond settings while it is intermittent near temporary pond setting. The cold season (November to March) represents the high transmission period for both transmission patterns [16].
Both male and female schoolchildren aged 7, 8 and 11 years were the study population. These particular ages were selected to be included in the monitoring and evaluation surveys because the control programme supported by the SCI targets schoolchildren and therefore these age groups would be the most representative. The chosen study population would also have the lowest likelihood of high drop-out rates over the survey years, particularly those aged 7–8 years at baseline who will stay in primary school until they the age of 11–12 years old. Furthermore, if an improvement is evident in the study population than there is a higher probability that morbidity in adults can be prevented.
The recommended sample size was 60 children by age group in S. haematobium zones. When the sample size could not be attained with enrolled schoolchildren, further school-aged children were recruited from the village.
The study was approved by the National Ethics Committee of Niger and the National Health System Local Research Ethics Committee of St. Mary's Hospital, London, prior to data collection. Once institutional consent obtained, meetings were held in all targeted villages in order to clearly present the objectives and the methods of the study to the families. They were invited to give their consent prior to inclusion of their children in the survey and they were made aware that they could refuse to permit their children from participating in the study or to remove their children without any consequences. The consent sought was verbal because of the high illiteracy rate in rural population, and the National Ethics Committee of Niger approved this method of consent.
Participating schoolchildren underwent a standardized interview focusing on clinical signs of urinary or intestinal schistosomiasis and on previous anti-schistosomal treatment.
From each child, a urine sample was collected between 10:00 AM and 2:00 PM and the presence of blood was noted, according to a standardized colorimetric scale coding from 0 to 6. Gross haematuria was defined as a visual quantification coded >3. Reagent strips (Hemastix Bayer Diagnostics Division, Tarrytown, NY, U.S.A) were used for diagnosing microscopic haematuria. Codes ranging from 0 to 5 were attributed according to the manufacturer interpretation scale and micro-haematuria was defined as urines coded ≥3.
Parasitological examination of urine and S. haematobium eggs count were performed after filtration of 10 ml of urine using Nytrel filters. The intensity of infection was categorized into two classes: light-intensity infection (egg count <50/10 ml) and heavy-intensity infection (egg count ≥50/10 ml). The geometric mean egg count of excreting individuals was used to assess the infection at community level.
S. mansoni infections were determined by preparing two 41.7 mg Kato-Katz thick smears from a single fresh stool specimen [17]. All stool specimens were also analysed for Ascaris lumbricoides, Trichuris trichiura, Enterobius vermicularis, Taenia spp and hookworm.
A Hemocue photometer (HemoCue AB, Ängelholm, Sweden) was used to determine haemoglobin concentration from a capillary blood drop collected from the finger tip of each child, using sterile single-use material. Children were considered anaemic when haemoglobin was less than 11.5 g/dl [18].
A thick smear was made for the search of Plasmodium falciparum, but this examination was performed in the four villages of the region of Tahoua only. Children with asymptomatic malaria infection during dry season are supposed to have greater risk of chronic anaemia, knowing the great role played by malaria in the physiopathology of anaemia [13],[19].
All the children underwent an ultrasound examination of the urinary tract to assess morbidity due to schistosomiasis using a portable ultrasonographer Aloka SSD 500 (Aloka, Tokyo, Japan) with a 3.5 MHz convex transductor. Bladder and urinary tract abnormalities were assessed and reported according to the WHO/Niamey protocol [20]. Bladder lesions were scored 0, 1 or 2, ureteral lesions were scored between 0, 3 or 4, and kidney distension was scored between 0, 6 or 8. By adding these elementary scores, an Echographic Severity Index (ESI) was calculated. In total, there were four classes of ESI: 1 = light impact, 2–4 = moderate impact, 5–9 = severe impact, ≥10 = very severe impact [16].
According to the national control programme procedures, praziquantel and albendazole were administered by a local team at the community level during the mass drug administration campaign, following the baseline data collection and 1 year later during the second mass drug administration campaign, following the control survey.
Praziquantel (using dose-pole corresponding to 40 mg/kg) and Albendazole (400 mg) were given to the target population regardless of infection status, during the mass drug administration campaign that took place 3–4 weeks after the surveys were conducted. Trained community drug distributors delivered the treatment in several distribution points in the villages. Schoolchildren were treated by teachers at school. Drugs were swallowed in presence of the drug distributor. Coverage rate reported by the national programme was approximately 68% for all the targeted regions (coverage rate was higher in school-based treatment than through community-based treatment).
EpiInfo and SPSS software was used for data management and analysis. The Pearson's χ2, the Fisher's exact test, the Student's t test or the Kruskal-Wallis' test were used whenever appropriate. For paired analysis, the McNemar's χ2 was used. A p value <0.05 was considered significant. Logistic regression was used for multivariate analysis to identify the risk factors of anaemia.
A total of 1642 schoolchildren were enrolled in the study. The overall sex ratio (male to female) was 1.27: 1, but the sex distribution varied greatly according to the village, ranging from 14.9% to 73.3% for girls. The overall mean age was 8.7 years, without any significant difference between schools. The number in each age group was 532, 546 and 565 for the 7, 8 and 11 years old respectively.
The most commonly reported symptoms were bloody urine (41.8%), abdominal pain (28.5%) and pain when urinating (24.5%). A previous treatment against schistosomiasis was reported by 4.5% of the schoolchildren. No previous treatment had been administered at community level in any of the eight villages.
The overall prevalence of S. haematobium infection was 75.4% [95% Confidence Interval (CI): 73.2–77.5]. The prevalence varied from 43.6% to 97.7%, according to the village. Older children were more frequently infected than were younger ones (p<0.01). The overall prevalence of heavy-intensity infections was 21.8%. Intensity increased with increasing age: 16.4%, 21.8% and 27.0% in children aged 7, 8 and 11 years, respectively (p<0.01). Among the heavily infected children, 50 (13.9%) excreted more than 500 eggs/10 ml of urine. The geometric mean egg count among egg excreting children was 15.5 eggs /10 ml urine. Geometric mean egg counts were 11.7, 16.9 and 17.6 in children aged 7, 8, and 11 years respectively (p<0.01). The prevalence of infection was significantly higher in schools located in the south-western part of the country: 85.3% in the region of Tillabéry compared to 64.8% in the region of Tahoua (p<0.01). The prevalence of S. haematobium infection was significantly associated with reported haematuria and pain during urination (p<0.01).
S. mansoni infection was observed only in 2 schools: Sabon Birni (3%) and Sanguilé (1.1%).
Hookworm infection was observed in 3 schools; Sabon Birni, where the prevalence was 18.8 and in 2 other villages were the prevalence was 0.6%. Hookworm infection was not observed in the schoolchildren of the 5 other villages.
Very low prevalence (0.3 to 0.7%) of Ascaris lumbricoides infection was observed in 5 schools, while 3% of the schoolchildren were infected in 1 school (Sanguile) and no infection was observed in 2 schools (Kaou and Tabalak).
The overall prevalence of P. falciparum infection was 8% in the 636 tested schoolchildren, ranging from 3.8% to 15.8% according to the village (p<0.03). P. falciparum infection was neither associated with age, nor S. haematobium infection.
The overall prevalence of observed gross haematuria was 6.9%, with a significant relationship with S. haematobium infection (p<0.03). Gross haematuria was observed in 21.4% of heavily infected children. The overall prevalence of micro-haematuria was 53.3%, differing significantly according to the village (p<0.03) and increasing significantly with increasing age (p<0.03). The prevalence of micro-haematuria was 9.6% in children not excreting eggs, 56.3% in children with light-intensity infections and 92.8% in children with heavy-intensity infections (p<0.03). Reported history of haematuria and pain during urination were significantly associated with the prevalence of micro- and gross haematuria (p<0.03).
Overall, the mean haemoglobinemia was 11 g/dl (range: 5.3 g–17.3 g). The prevalence of anaemia was 61.6%, without significant difference between sexes. Haemoglobinemia was significantly related to age: 66.5% in children aged 7 years, 64.6% in children aged 8 years and 54.1% in those aged 11 years (p<0.03). Prevalence of anaemia was also statistically different according to the village (p<0.03). Anaemia was significantly associated with micro-haematuria (p<0.02), but not to gross haematuria. Overall, we found a significant association of anaemia with S. haematobium infection coded as a binary variable yes/no (p = 0.045), but not with infection coded into 3 levels of intensity. After adjusting for age, the association was present only in 8 year old children, where prevalence of anaemia increased significantly with increasing intensity of infection (p<0.01). The presence of anaemia was significantly associated with P. falciparum infection (p<0.01). Table 1 shows the prevalence of anaemia according to the presence, or absence, of several potential risk factors using bivariate analysis.
The proportion of children presenting at least one ultrasound abnormality of the urinary tract was 45.8%, without significant differences between sexes or transmission patterns, and was significantly associated with age (37.8%, 46.0% and 53.0% in children aged 7, 8 and 11 years, respectively; p<0.03), to the intensity of S. haematobium infection (22.1% in children not excreting eggs, 47.5% in light-intensity infections and 68.1% in heavy-intensity infections; p<0.03) and to the presence of anaemia (p = 0.025).
Bladder wall abnormalities, evident in 41.6% of children, were the most frequently reported ultrasound abnormality. Their prevalence increased significantly with increasing age (p<0.03), with increasing intensity of infection (p<0.03), and was not associated with the presence of anaemia. Severe bladder wall abnormalities (either masses or pseudopolyps) were present in 15.5% of the children. Kidney and ureteral distensions were present in 4.2% and 4.1% of the children, respectively. Contrary to ureteral distension, kidney distension was significantly associated with the intensity of S. haematobium infection (p<0.01) and to the presence of anaemia (p<0.03).
The mean ESI, summarizing urinary tract morbidity due to S. haematobium infection, was 1.33 (range: 0–24) and the mean bladder score was 0.8 (range: 0–7). The mean ESI was not different between boys and girls. The mean ESI was significantly higher in pond-related transmission villages than in river-related transmission villages (p<0.03).
Multivariate analysis using logistic regression showed that the risk of anaemia was significantly associated with young age (OR = 1.7, 95% CI 1.3–2.2 for age = 7 and OR = 1.6, 95% CI 1.2–2.0 for age = 8), to the presence of ultrasound abnormalities of the kidneys (OR = 3.0, 95% CI 1.5–6.0) and to the village. When the same analysis was performed on the sub-sample of 4 villages where P. falciparum infection was documented, the predictors of anaemia were P. falciparum infection (OR = 2.5, 95% CI 1.3–4.7), ultrasound abnormalities of the kidneys (OR = 2.9, 95% CI 1.2–6.8) and the village (OR = 0.6, 95% CI 0.4–0.9 for Tabalak and OR = 0.4, 95% CI 0.3–0.7 for Kaou).
A total of 89% of the initial sample group were re-examined one year after baseline data collection and the first round of treatment with praziquantel and albendazole. The mean age was 9.7 years and the sex ratio (male to female) was 1.27:1. A total of 482 eight year olds, 470 nine year olds and 484 twelve year olds were re-examined.
The overall prevalence of S. haematobium infection was 38% and 4.6% of children had heavy-intensity infections; only three (4.6%) among the latter excreted more than 500 eggs/10 ml. 56.2% of the initially infected children did not excrete eggs and 91.6% of heavy-intensity infections had cleared or become light intensity infection. However, the improvement of infection markers was different according to the village and prevalence was still above 60% in 3 schools (Kaou, Kokorou and Tabalak). In children from Tabalak, only minor changes in the prevalence and intensity of infection were observed (22.7% of heavy-intensity infections at baseline and 18.2% at follow-up).
Assessment of P. falciparum infection in the 4 schools tested in 2005 showed a significant increase of the prevalence from 8.6% to 17.1% (p<0.03). The rise of prevalence of malarial infection was observed in 4 schools, although only statistically significant in Rouafi (paired analysis, p<0.03).
The overall prevalence of all the morbidity markers of S. haematobium had decreased significantly at follow-up (Table 2). Among the 1412 schoolchildren which were re-examined, the overall prevalence of anaemia decreased from 61.9% to 50.4% (paired analysis, p<0.03) and the mean haemoglobinemia showed a significant increase of 0.39 g /dl to reach 11.4 g/dl (paired analysis, p<0.01). However, the change in prevalence of anaemia was not significant in children from Kaou, Kokorou, Mozague and Tabalak. At follow-up, anaemia was not associated with S. haematobium infection. Gross and micro-haematuria decreased from 7.1% to 0.4% (paired analysis, p<0.03) and from 53.5% to 6.0% (paired analysis, p<0.03), respectively. Prevalence of gross and micro-haematuria were each significantly associated with the intensity of S. haematobium infection (p<0.03). Anaemia was neither associated with micro-haematuria nor with gross haematuria. One year after treatment, the overall prevalence of ultrasound abnormalities of the urinary tract and the prevalence of ultrasound abnormalities of the bladder decreased from 45.6% to 15.2% (paired analysis, p<0.03) and from 41.6% to 14.7% (paired analysis, p<0.03), respectively. The mean global ESI showed a significant decrease of 1.04 (paired analysis, p<0.03).
In 2006, the relationship between anaemia and P. falciparum infection was no longer observed.
Compared to those children who remained in the study cohort, the 216 children who dropped out after the initial survey differed significantly in the prevalence of S. haematobium infection (75.4% vs. 78%, respectively), but had less frequently heavy-intensity infections (22.8% vs. 16.5%, respectively). On the other hand, they did not differ in mean age (8.7 vs. 8.9 years, respectively), in the prevalence of anaemia (61.9% vs. 59.7%, respectively) nor in mean haemoglobinemia (11.04 g/dl vs. 11.03 g/dl, respectively).
According to the WHO guidelines [21], the aim of the Nigerien National Schistosomiasis Control Programme is to reduce schistosomiasis-associated morbidity by reducing the prevalence of heavy-intensity infections.
Our study, which compared baseline data collected before treatment and follow-up data collected one year post treatment, aimed to describe the evolution of S. haematobium infection and of anaemia prevalence in schoolchildren in 8 sentinel cohorts.
The predominance of boys in our samples reflects the overall low proportion of girls in full-time education in Niger. Only 11% of children of the initial sample did not take part in the follow-up survey. The characteristics of the children who dropped out of the study, for both ultrasound assessed morbidity and for anaemia, did not differ significantly from children who completed the follow-up. The children who dropped out also had slightly lower prevalence and intensity of infection at baseline.
Our data showed a high prevalence of urinary schistosomiasis at baseline, with a classical increase of both prevalence and intensity of infection with increasing age during childhood. Although studies have shown that both prevalence and intensity of infection is often higher among males in many surveys [22],[23], our data did not show such a relationship.
S. haematobium infection was well correlated to the prevalence of observed gross haematuria, confirming the value of this indicator for rapid assessment of infection at community level, as well as its intensity. Reported history of haematuria and pain during urination were both related to the infection, even if the schoolchildren underestimated it. These markers, have been used for rapid screening of urinary schistosomiasis [24].
According to the operating procedures of the national control programme, the praziquantel and albendazole treatment was administered by school teachers or health agents using the praziquantel “dose-pole” [25],[26] for praziquantel within 3–4 weeks following the baseline data collection (in November-December 2004 in Tillabéry region and in April-May 2005 in Tahoua region). Eleven to twelve months after treatment, the overall prevalence of S. haematobium infection had decreased significantly, from 75.4% to 38% and the prevalence of heavy-intensity infections had drastically reduced by 91.7%. These observations fully confirm the effectiveness of a single dose of praziquantel on infection, as demonstrated over the past two decades [27],[28].
However, in 3 schools, the prevalence of S. haematobium infection remained high, over 60%. In on village, Tabalak, the prevalence of heavy-intensity infections did not change at all. It is unlikely that this was solely due to re-infection, according to previous studies in Niger [16],[29] prevalence is unlikely to reach the initial level of intensity within a year of treatment. However, it has been demonstrated that S. haematobium reinfection patterns may vary strongly according to local epidemiological settings. Mass treatment took place after the high transmission period in the Tahoua region. In Tabalak where the transmission is year-round, the water level in local ponds usually decreases significantly during the hot and dry season. This reduction in water levels can induce a greater parasite concentration and thus a more severe infestation. Some alarming reports have been made since the 1990s about praziquantel treatment failures in S. mansoni infections [30], but despite large scale use of praziquantel in many countries there is currently no evidence to suggest that any S. haematobium has developed resistance to praziquantel as a result of its widespread use [31] and praziquantel can still be considered very effective. Furthermore, in the 4 villages where disappointing results were observed, the absence of mass treatment history should rule out a possible local problem of drug resistance induced by overuse or misuse of praziquantel. Although 75.1% of children in Tabalak reported having received tablets just after the baseline data collection, the absence of prevalence reduction is likely to be explained by the fact that praziquantel treatment was not administered as effectively as had been assumed. For children, confusion between albendazole tablets and praziquantel tablets could be quite easy and the drug delivery system in Tabalak will need to be assessed. In addition, transmission of S. haematobium in Tabalak will also be investigated.
The very low prevalence of soil-transmitted helminth infection is in accordance with previous observations made in Niger [32] and in 2 other African countries, Mali and Chad [22],[33], areas that are subjected to climatic conditions comparable to the most inhabited regions of Niger. The single school where hookworm infection prevalence was over 1% was located in the Sudanian climatic zone, in the southern-most part of the country.
The nature and the frequency of ultrasound-assessed urinary tract abnormalities, observed in 46.2% of schoolchildren, with an association with both intensity of infection and with age are fully consistent with results from other studies [5],[34].
The effect of treatment on reducing the prevalence of both bladder and ureteral lesions was 78.2% and 96.3% respectively, a clear confirmation of the relevance of the control strategy. Previous surveys in Niger reported similar favourable results, showing the rapid regression of ultrasound-assessed bladder and ureteral lesions [16],[29],[35]. Many studies in other endemic areas also demonstrated the invaluable benefit of praziquantel administration on S. haematobium associated bladder morbidity [36].
At baseline, the prevalence of ultrasound assessed-kidney lesions was low (4.1%), compared to results from other studies which reported up to 43% of hydronephrosis in schoolchildren in hyperendemic village in the Niger River valley. Other results reported between 20.9% and 23.2% upper urinary tract lesions in two villages in an irrigated area of Niger [16],[29]. Contrary to other observations [37], in our study sample kidney dilatations were significantly associated with the intensity of S. haematobium infection. Although ultrasound-assessed kidney dilatations are not specific to urinary schistosomiasis, their regression following praziquantel treatment strengthens the likelihood of their relationship with infection. According to the literature, the observed clearance rate of kidney dilatations following praziquantel treatment may vary greatly, from low [38],[39] to very satisfactory [35],[40]. This may be explained by many factors, such as the length of the post-treatment follow-up or the age of lesions.
Anaemia is considered a severe public health problem when its prevalence is over 40% within a population [41]. Anaemia is very common in developing countries and a multi-country survey in sub-Saharan Africa showed that it was generally a serious problem in schoolchildren [42]. Our results, showing a prevalence of 61.2%, confirm that anaemia in schoolchildren is a worrying public health problem in Niger.
Various causes can lead to anaemia including infections, low dietary iron intake, sickle-cell anaemia, malnutrition and inflammatory diseases. Although many cross-sectional studies and a small number of randomized controlled trials have addressed the relationship between schistosomiasis and anaemia, the conflicting results make the magnitude of the relationship unclear. Urinary schistosomiasis had been associated with anaemia in many studies. In a study in Niger, S. haematobium egg count and intensity of haematuria were negatively correlated to haemoglobin level, to haematocrit and to transferrin saturation, and positively associated with anaemia and iron deficiency [11]. Another Nigerien survey concluded that S. haematobium infections increased the risk of anaemia by 30% [10]. In Tanzania, haemoglobin concentration in infected children was 0.4 g/dl lower than in uninfected ones [43]. In Kenya, the intensity of S. haematobium infection was negatively correlated with haemoglobin level on a cross sectional basis [44]. On the other hand, a survey in Cameroon, in a context of concurrent malarial and schistosomal infections [15], did not establish a relationship between schistosomiasis haematobia and anaemia while anaemia correlated significantly with malaria infection. In Mali, in a survey focusing on interactions between helminth and malarial infections in children, results showed haemoglobin was similar in children infected with S. haematobium compared to those not infected [45]. Likewise, in a survey on anaemia in Tanzania, no relationship was observed with S. haematobium, whereas anaemia was significantly associated with hookworm infection [46].
Our results, showing a significant relationship between schistosomiasis haematobia and anaemia, were bordering on statistical significance in a bivariate analysis, and anaemia was not related to intensity of infection. In a previous analysis using an alternative haemoglobin threshold for anaemia (11 g/dl instead of 11.5, data not shown), no relationship was observed between infection and anaemia. Also, it is worthy to note that the prevalence and intensity of S. haematobium infection increased with increasing age, contrary to anaemia that was more prevalent in the youngest schoolchildren. The latter observation is consistent with the results found in Chad [22].
Our results showed that micro-haematuria increases with age as does the prevalence of schistosomiasis and that the prevalence of anaemia decreases with age. These findings raise the question about the significance of micro-haematuria in the prevalence of anaemia. By which mechanism does micro-haematuria contributes to anaemia? Why anaemia prevalence is higher in the age group where micro-haematuria is lower?
Lastly, in multivariate analysis, S. haematobium infection was not associated with anaemia. However, anaemia could be related to S. haematobium in an indirect way as it was significantly associated with micro-haematuria. The significant improvement of haemoglobin levels one year after praziquantel and albendazole treatment suggests that S. haematobium can contribute significantly to anaemia. In Burkina Faso where the SCI-supported schistosomiasis control programme provides also large-scale chemotherapy with praziquantel and albendazole, Koukounari and others found significant reduction in the prevalence of anaemia and a significant increase in haemoglobin concentration [47]. This confirms our findings on the benefit of praziquantel on anaemia.
Due to the study design, the causality cannot be formally established, but one can note that the 3 schools were anaemia did not improve corresponded to the 3 of 4 schools were S. haematobium infection showed only minor changes one year after praziquantel treatment. The overall very low prevalence of hookworm infection can rule out a significant contribution of hookworm in schoolchildren anaemia in Niger. Lastly, the observed improvement of anaemia is unlikely explained by the fact that some children dropped out of the survey before follow-up, as they did not differ significantly from the others, neither for S. haematobium infection, nor for anaemia prevalence.
The simultaneous delivery of several drug packages within the framework of the integrated neglected tropical disease control programmes (schistosomiasis, lymphatic filariasis, intestinal worms, trachoma, onchocerciasis) which are currently being implemented in several sub-Saharan African countries will probably involve an additional improvement of the prevalence of anaemia and the global health of the populations by the synergistic actions of the drugs on the worms, the ectoparasites and certain bacteria. Our study showed that anaemia can be a useful indicator for the monitoring of the impact of programs NTD as suggested by Bates et al. and Molyneux et al [48],[49].
Malarial infection is a well documented cause of anaemia in African children [13],[19]. The multivariate analysis performed on the sub-sample of 4 schools where data on P. falciparum infection were available showed that asymptomatic carriage of P. falciparum, kidney distension and the village were the only predictors of anaemia, while no relationship was found with S. haematobium infection. In Niger, 83.9% of children <5 years had an haemoglobinemia of <11 g/dl in a nationwide survey in 2006, with only minor differences between rural and urban settings [50], although both prevalence and intensity of S. haematobium infection are usually low in this age group. In total, S. haematobium alone cannot explain the high prevalence of anaemia in Nigerien schoolchildren. Actually, with 8.1% of children carrying P. falciparum at baseline, malarial infection could hardly explain the entire problem of anaemia in Nigerien children. Dietary iron availability should be investigated, as well as the prevalence of other known risk factors, such as sickle cell anaemia.
Both S. haematobium and anaemia are highly prevalent in schoolchildren in Niger. Before treatment, anaemia was significantly associated with infection in a bivariate analysis, but this association was not observed in a multivariate analysis, while malarial infection was a significant predictor of anaemia. One year after systematic praziquantel treatment, we observed a significant decrease of the prevalence and intensity of infection and of morbidity. Though the prevalence of anaemia significantly reduced, prevalence was still 50.5%. The high prevalence of anaemia in Nigerien schoolchildren is likely a result of the combination of many risk factors and should be thoroughly investigated. The role of schistosomiasis on anaemia clearly deserves more studies. It would be interesting to study the interactions of malarial infection, schistosomiasis, soil-transmitted helminthiasis as well as nutritional status in the mechanisms of anaemia. Even though schistosomiasis represents only one of multiple causes of anaemia, the effect of praziquantel treatment on morbidity clearly indicates the benefits of schistosomiasis control programmes, which must be perpetuate. |
10.1371/journal.pgen.1006216 | Levels of Ycg1 Limit Condensin Function during the Cell Cycle | During mitosis chromosomes are condensed to facilitate their segregation, through a process mediated by the condensin complex. Although several factors that promote maximal condensin activity during mitosis have been identified, the mechanisms that downregulate condensin activity during interphase are largely unknown. Here, we demonstrate that Ycg1, the Cap-G subunit of budding yeast condensin, is cell cycle-regulated with levels peaking in mitosis and decreasing as cells enter G1 phase. This cyclical expression pattern is established by a combination of cell cycle-regulated transcription and constitutive degradation. Interestingly, overexpression of YCG1 and mutations that stabilize Ycg1 each result in delayed cell-cycle entry and an overall proliferation defect. Overexpression of no other condensin subunit impacts the cell cycle, suggesting that Ycg1 is limiting for condensin complex formation. Consistent with this possibility, we find that levels of intact condensin complex are reduced in G1 phase compared to mitosis, and that increased Ycg1 expression leads to increases in both levels of condensin complex and binding to chromatin in G1. Together, these results demonstrate that Ycg1 levels limit condensin function in interphase cells, and suggest that the association of condensin with chromosomes must be reduced following mitosis to enable efficient progression through the cell cycle.
| Chromosome conformation is cell cycle-regulated so that chromosomes are highly compacted to facilitate their segregation during mitosis, and decondensed during interphase to facilitate DNA-dependent processes such as replication and transcription. Understanding how chromosomes transition between these different states is important for understanding how cells maintain a stable genome. The condensin complex is an essential five-subunit complex that controls chromosome condensation in all eukaryotes. In this study, we show that expression of the Cap-G/Ycg1 subunit of condensin in budding yeast is cell cycle-regulated, and that its reduced expression during interphase limits condensin function. When this regulation is disrupted, and Ycg1 is constitutively expressed, progression through interphase is delayed. Emerging evidence indicates that individual condensin subunits are also expressed at limiting levels in metazoan cells, which suggests that cell-cycle regulation of an individual condensin subunit is a conserved mechanism that coordinates condensin function with the cell cycle.
| The eukaryotic cell cycle is divided into two distinct parts: interphase, when cell growth and DNA replication occur, and mitosis, when chromosomes are segregated into daughter cells. One major phenotypic difference between these phases is chromosome conformation. Specifically, interphase chromosomes are decondensed and loosely packed within the nucleus, which allows for maximum accessibility of the DNA to the transcription and replication machineries, while mitotic chromosomes are tightly compacted and condensed, which facilitates their segregation during anaphase [1]. Accurate transit in and out of these conformations is paramount to proliferation, since decondensed chromosomes during mitosis impede segregation, and can generate DNA breaks that lead to genome instability [2,3], whereas condensed chromosomes during interphase hinder transcription and replication, and thus may impede cell-cycle progression.
One important factor involved in controlling interphase and mitotic chromosome conformations is the condensin complex [4]. Condensin is a conserved eukaryotic complex that is comprised of five protein subunits: two core ATPase subunits (Smc2 and Smc4), a kleisin subunit (CAP-H/Brn1), and two HEAT-repeat subunits (CAP-G/Ycg1 and CAP-D2/Ycs4), each of which is essential for complex function and cell viability [5–8]. Mammalian cells have two condensin complexes, condensin I and condensin II, which differ in their non-SMC subunits and mediate different aspects of chromosome condensation [9,10]. In contrast, yeast have only one complex, which is similar in sequence to condensin I in mammals [11].
In all organisms, condensin function is most pronounced during mitosis, when its phosphorylation-stimulated activity leads to large-scale supercoiling of DNA and chromosome compaction [12,13]. After the completion of mitosis, condensin supercoiling activity decreases, resulting in chromosome decondensation [13,14]. Although supercoiling activity is diminished after mitosis, some condensin remains associated with chromatin throughout interphase. In budding yeast, condensin associates with genes encoding tRNAs, ribosomal proteins, and small nuclear and nucleolar RNAs (SNR genes) throughout the cell cycle and aids in clustering of these loci [15–17]. Condensin also has non-mitotic roles in establishing metazoan chromosome structure [18–21]. However, the mechanisms that coordinate these different condensin functions with the appropriate cell-cycle stage are not well understood.
Previous studies investigating condensin regulation have mainly focused on how phosphorylation activates the complex during mitosis to trigger chromosome condensation. Condensin phosphorylation by Polo kinase, Aurora B, and Cdk1 has been shown to promote its localization to mitosis-specific loci, and to stimulate its supercoiling activity [13,14,22–24]. In addition, binding of budding yeast condensin to centromeres and the repetitive rDNA locus increases during mitosis via recruitment by Sgo1 and Fob1, respectively, which act as chromatin-associated receptors [25–27]. Much less is known about how chromosome condensation is reversed after mitosis is complete. However, changes in condensin phosphorylation upon mitotic exit are likely to play a role in this process. Specifically, mitotic kinases are inactivated in late mitosis [28], and inhibitory phosphorylation by CKII may limit condensin activity in interphase, as has been demonstrated for human condensin [29]. In mammals, condensin I relocalizes to the cytoplasm in interphase [30,31], thereby restricting its access to chromosomes. However, mammalian condensin II and budding yeast condensin are constitutively nuclear [5,30,31], and thus are predicted to have additional mechanisms to regulate their association with chromosomes. The precise mechanisms that downregulate the activity of these complexes after mitosis are not known.
Emerging evidence suggests that proteasomal degradation of an individual subunit may be one mechanism that limits condensin activity. In Drosophila melanogaster, the Cap-H2 subunit of condensin II is targeted for ubiquitin-mediated degradation, and blocking this degradation results in partial chromosome condensation in interphase cells [32,33]. Additional studies have reported ubiquitination of the Cap-G subunit in budding yeast [34], and that human condensin II subunits are degraded by the ubiquitin-proteasome system [35,36]. However, it is not known in any system if ubiquitin-mediated degradation leads to cyclical expression of any condensin subunit during the cell cycle, and if levels of a subunit do cycle, whether or not this regulation contributes to cell cycle-regulated changes in chromosome structure.
In this report, we show that the Cap-G subunit of budding yeast condensin (Ycg1) is expressed in a cell cycle-dependent manner due to cyclical transcription coupled with constitutive degradation. Further, we observe that cyclical expression maintains Ycg1 at limiting levels relative to the other condensin subunits. Finally, we show that increasing Ycg1 expression results in increased recruitment of condensin complex to chromosomes during G1 phase, and interferes with progression through the G1/S transition. These data suggest that downregulation of Ycg1 after mitosis contributes to a reduction in condensin activity, and that a decrease in condensin function during G1 phase is necessary to facilitate cell-cycle progression.
Although the budding yeast condensin complex associates with chromatin throughout the cell cycle [6,7,15,17], its activity increases substantially during mitosis. Previous reports have shown that this change in activity is due in part to increased phosphorylation [3,14,24], and to enhanced recruitment of the complex to a subset of sites in the genome [15,17,25–27,37]. Interestingly, several studies have also reported that transcription of the gene encoding the Cap-G subunit of condensin, YCG1, is cell cycle-regulated (Fig 1A) [38–40], with lower levels in G1 than mitosis. Additionally, Ycg1 protein levels have been reported to be lower in interphase than in mitosis [22]. This evidence suggests that regulation of Ycg1 levels may be an additional mechanism that coordinates condensin activity with the cell cycle. To investigate this possibility further, we examined expression of Ycg1 mRNA and protein following release from a G1 arrest and found that they cycled similarly: expression increased as cells progressed through interphase, peaked during mitosis, and declined upon entry into the next G1 phase, similar to the mitotic cyclin Clb2 (Fig 1A and 1B). In contrast, none of the other subunits of the condensin complex displayed this dramatic fluctuation during the cell cycle, although Brn1 expression was modestly decreased in G1-arrested cells (Fig 1C and 1D). These observations, coupled with the fact that Ycg1 is essential for condensin function [7,17,22], suggest that regulation of Ycg1 levels during the cell cycle may be a previously uncharacterized mechanism that limits condensin function during interphase.
The rapid decrease in Ycg1 levels after mitosis suggested that Ycg1 might also be regulated by proteolysis. To test this possibility and assay its stability, we monitored Ycg1 levels in asynchronous cells over time in the presence of the translation inhibitor cycloheximide, and found that Ycg1 was rapidly degraded (Fig 2A). Next, we asked whether other subunits of the complex were similarly regulated. To do this, each subunit of the complex was tagged with an identical 3HA tag, and their stabilities were compared in the same assay. This analysis revealed that Ycg1 is the least stable, and the least abundant, subunit of the condensin complex (Fig 2A).
Many cyclically expressed proteins are degraded by the ubiquitin proteasome system (UPS) [41], and Ycg1-ubiquitin conjugates were previously identified in a proteomic screen [34], which suggested that Ycg1 may undergo ubiquitin-mediated degradation. Consistent with this possibility, proteasomal inhibition impaired Ycg1 turnover in asynchronous cells, confirming that the protein is regulated by the UPS (Fig 2B). Since Ycg1 is necessary for condensin function, and condensin function is essential for the completion of mitosis [7,17,22], we speculated that Ycg1 might be stable during mitosis. To test this, we arrested cells in G1 or mitosis, and monitored Ycg1 turnover (Fig 2C). We found that although there was more protein in mitosis, consistent with its increased transcription late in the cell cycle (Fig 1A), Ycg1 was degraded in both arrests. This observation suggests that Ycg1 is degraded throughout the cell cycle, surprisingly, even during mitosis. Taken together, these data indicate that constitutive degradation, paired with cyclical transcription, leads to cell cycle-regulated expression of Ycg1.
Next, we sought to investigate the importance of cyclical Ycg1 expression for progression through the cell cycle. To do this, we engineered mutations in Ycg1 that blocked degradation. Most proteins that undergo ubiquitin-mediated degradation have short sequences termed degrons, which are essential for degradation. Many degron sequences are found in unstructured domains that are subject to other forms of regulation, such as phosphorylation [42]. Interestingly, the C-terminal domain of Ycg1 fits these criteria [14,43]. Moreover, although this domain includes several phosphorylation sites that contribute to condensin activation during mitosis, this domain is not essential for viability [14], which allowed us to replace the endogenous copy of YCG1 with alleles carrying mutations in this region. We first tested whether this domain was required for Ycg1 degradation and found that Ycg1 was completely stabilized when the C-terminal 63 amino acids were deleted (Fig 3A, Ycg1Δ973–1035). However, deletion of the C-terminal 50 amino acids had no effect on Ycg1 degradation (Ycg1Δ986–1035). These data suggested that Ycg1 turnover requires amino acids 973–985 and, consistent with this possibility, deletion of these amino acids was sufficient to stabilize the protein (Ycg1Δ973–985, Fig 3A). Additional deletions and truncations in the C-terminus were consistent with this conclusion (S1A Fig).
Since amino acids 973–985 lie within the conserved phosphoregulatory domain of Ycg1 (S1A Fig) [43], we endeavored to create a stable mutant that minimally alters the sequence of this region. To do this we mutated features within this region that might contribute to degradation, including charged residues and putative phosphorylation sites (Fig 3B). We found that positively charged residues were necessary for Ycg1 degradation, with mutation of lysine-977 or arginine-978 having the greatest effect (Fig 3C). In contrast, mutation of negatively charged residues, or all serines and threonines in the region, had little to no effect on Ycg1 stability (S1B Fig). Although our data suggest that Ycg1 is degraded throughout the cell cycle (Fig 2C), we confirmed that the increased stability of Ycg1-K977A did not result from a change in cell-cycle distribution in the mutant strain by arresting cells in G1 or mitosis and assaying Ycg1 turnover. This analysis confirmed that Ycg1-K977A is more stable than wild-type Ycg1 in both phases of the cell cycle (Fig 3D).
The prevailing model suggests that chromosome condensation needs to be reversed after mitosis to facilitate essential DNA-dependent processes during interphase, such as replication and transcription. Since Ycg1 is downregulated after mitosis, we posited that interference with this regulation might impact cell-cycle progression. To test this, we analyzed the proliferation rate of each of the strains expressing point mutations that stabilize Ycg1. Interestingly, we observed a modest increase in doubling time in mutants that partially blocked Ycg1 turnover, and a much larger increase in doubling time in mutants that fully blocked turnover (Fig 4A, S1B Fig). These data show a correlation between increased Ycg1 expression and decreased proliferation rate, suggesting that Ycg1 downregulation after mitosis may be important for cell-cycle progression.
Next, we asked whether the decreased proliferation rate that we observed in cells expressing stable Ycg1 resulted from a delay at a specific point in the cell cycle. Strains expressing Ycg1 or Ycg1-K977A were synchronized in G1 phase and released. Cell-cycle progression was then followed by flow cytometry and Ycg1 levels were monitored by Western blot. In contrast to the wild-type protein, Ycg1-K977A was expressed at a constant level throughout the cell cycle (Fig 4B, top), demonstrating that degradation is necessary for cell cycle-dependent changes in Ycg1 levels. Notably, ycg1-K977A strains exhibited delayed progression from G1 into S phase (Fig 4B, bottom), consistent with the possibility that failing to downregulate condensin might interfere with progression through interphase.
Haploid ycg1-K977A strains are viable, confirming that the allele encodes a functional protein. However, the K977A mutation falls in a domain of Ycg1 that is required for maximal condensin activity [14], raising the possibility that this mutation might both increase Ycg1 expression and reduce its function. To address this possibility, we performed additional characterization of ycg1-K977A strains. First, we confirmed that the interaction between Ycg1-K977A and the other subunits of condensin was not impaired (S2A Fig). In addition, we used an established rDNA reporter assay [44] to investigate whether ycg1-K997A cells exhibited defects in rDNA silencing, or increased recombination at the rDNA locus, both of which are phenotypes exhibited by condensin loss-of-function mutants [5,6]. We found that ycg1-K977A cells were similar to wild-type cells in this assay (S2B Fig). Moreover, the proliferation defect in ycg1-K977A strains could not be rescued by the addition of a second copy of YCG1 integrated at the URA3 locus, suggesting that the growth defect is not the result of reduced function of the mutant (S2F Fig). Although these assays suggested Ycg1-K977A is functional, we observed that multiple isolates of haploid ycg1-K977A strains exhibited non-uniform colony size (S2C Fig), exhibited increased sensitivity to the replication inhibitor hydroxyurea (HU) (S2D Fig) (a phenotype that has been reported for strains expressing hypomorphic alleles of condensin subunits in fission yeast [45]), and showed increased sensitivity to the microtubule poison benomyl (S2E Fig). Moreover, we had difficulty generating haploid strains that expressed Ycg1-K977A and had an epitope tag on any other subunit of the condensin complex. Together, these findings suggested that the K977A mutation might reduce Ycg1 function, in addition to stabilizing the protein.
To distinguish between these effects and determine whether the increased expression of Ycg1-K977A was the primary cause of the proliferation defects described above, we disrupted cell cycle-regulation of Ycg1 levels in an alternative way, using the constitutive TEF1 promoter to express Ycg1 at elevated levels throughout the cell cycle (Fig 4C and 4D). TEFp-YCG1 strains showed no alteration in rDNA stability or silencing, confirming that YCG1 overexpression does not impair condensin function (S2B Fig). Importantly, TEFp-YCG1 strains displayed an increase in doubling time, similar to ycg1-K977A strains (Fig 4C). Furthermore, both ycg1-K977A and TEFp-YCG1 strains showed a delay in G1/S progression (Fig 4B and 4D), and exhibited sensitivity to temperature stress (Fig 5A). These data argue that increasing Ycg1 abundance is sufficient to delay the cell cycle and decrease proliferation rate. Notably, overexpression of Ycg1 did not result in heterogeneous colony size or sensitivity to HU or benomyl (S2 Fig), which suggests that these phenotypes of the ycg1-K977A strain may result from its reduced function, and not increased expression of the stable protein.
The delay in cell-cycle progression described above could be the result of a delay in the G1/S transition and/or an inhibition of DNA replication in mutant strains. To determine whether the transition from G1 into S phase was delayed, we monitored budding, since bud formation is triggered by the wave of transcription that occurs at the G1/S transition, but is independent of replication initiation [46]. Interestingly, the delay in DNA synthesis in ycg1-K977A and TEFp-YCG1 strains correlated with a proportional delay in budding (Fig 5B and 5C), indicating that these strains exhibit a delay in entering S phase. The delay was most evident 22.5 minutes after release from G1, when wild-type cells were in S phase and ycg1-K977A and TEFp-YCG1 strains were largely still in G1 (Fig 5C). Consistent with a previous report [47], this delay was not observed in the condensin temperature-sensitive mutants ycg1-2 and brn1-9 [48] when they were released from G1 arrest at the restrictive temperature (S3 Fig), confirming that the G1/S delay observed upon Ycg1 overexpression is distinct from condensin loss of function.
Chromosomes decondense in telophase, so condensin activity must decrease at the end of mitosis. One possibility is that the increased Ycg1 levels in ycg1-K977A and TEFp-YCG1 strains might impair chromatin decondensation, which could induce an additional cell-cycle delay when cells exit from mitosis. We tested for this possibility by synchronizing cells in metaphase with a CDC20 shut-off allele and monitoring progression of each strain into G1 phase by flow cytometry. Although it is possible that the strains may progress through the stages of mitosis with slightly different kinetics, both strains entered G1 phase with similar timing to a wild-type strain (Fig 5D), suggesting that neither strain has a delay in exiting from mitosis. We also assayed chromosome condensation directly, by examining the structure of the rDNA locus, which undergoes compaction during mitosis that can be visualized in chromosome spreads [7,22,49]. Cells were arrested in both metaphase and G1, the rDNA was visualized by DAPI and Net1 staining of chromosome spreads, and condensation scored as previously described [22,47]. Notably, there was no significant difference in rDNA conformation between wild-type and TEFp-YCG1 strains, in either metaphase or G1-arrested cells (S4 Fig). Together, these results argue that increasing Ycg1 expression does not alter rDNA condensation, or delay exit from mitosis.
Our comparison of the expression levels of condensin subunits indicates that Ycg1 is expressed at lower levels than the other subunits (Fig 2A, S5A Fig). In addition, Ycg1 is the only condensin subunit that cycles (Fig 1C). These findings raise the possibility that Ycg1 levels might be limiting for complex formation. If this is the case, then overexpression of other subunits of the complex should not impair cell-cycle progression in the way that overexpression of Ycg1 does. To test this hypothesis, we integrated the TEF1 promoter upstream of the other four subunits of the condensin complex. Importantly, although each condensin subunit was overexpressed in these strains to similar levels (S5A Fig), increasing expression of no other condensin subunit led to an increase in doubling time (Fig 4C). Moreover, while asynchronous TEFp-YCG1 cells displayed an increased fraction of cells in G1 phase, consistent with a G1/S delay, there was no change in the fraction of G1 cells upon overexpression of any other condensin subunit (S5B Fig). These data are in agreement with the model that Ycg1 is the limiting subunit for condensin function.
Ycg1 has not been shown to function on its own, or as part of any protein complex other than condensin. Therefore, we hypothesized that increased Ycg1 expression slowed G1/S progression as a result of increased condensin complex during G1 phase. Notably, this hypothesis makes two predictions: first, that the amount of intact condensin complex varies based on cell-cycle position, and second, that modulation of Ycg1 levels is necessary to establish this variation. To test these possibilities, we assayed for changes in condensin subunit interactions in different cell-cycle phases. First, we arrested cells in G1 phase or mitosis, immunoprecipitated different subunits of the condensin complex, and determined whether more Ycg1 associated with each subunit in mitosis than in G1 phase. Importantly, more Ycg1 co-immunoprecipitated with other condensin subunits in mitosis than G1 (Fig 6, compare lanes 5 and 11 in each panel), confirming the level of intact condensin complex varies in different cell-cycle phases.
We simultaneously performed co-immunoprecipitation experiments in TEFp-YCG1 strains to determine if preventing the downregulation of Ycg1 led to an increase in the amount of intact condensin complex. Notably, more Ycg1 was associated with other subunits of the complex in the TEFp-YCG1 background compared to wild-type cells in G1 phase (Fig 6, compare lanes 5 and 6 in each panel). In mitotic cells we observed a small increase in Ycg1 interaction over the already high levels in wild-type cells when Ycg1 was overexpressed (Fig 6, compare lanes 11 and 12 in each panel). These data show that overexpression of Ycg1 increases condensin subunit interactions considerably in G1, when Ycg1 is limiting, and less so during mitosis, when Ycg1 levels peak.
A previous study demonstrated that Ycg1 is required to recruit other condensin subunits to chromatin [50]. Therefore, we investigated whether the chromatin association of the Brn1 subunit was increased in TEFp-YCG1 cells by quantifying the amount of Brn1 that associated with chromosomes in a chromosome spread assay [7,17,51]. Notably, although overexpression of Ycg1 did not lead to increased levels of Brn1 (Fig 7A), the association of Brn1 with chromatin increased in TEFp-YCG1 cells (Fig 7B and 7C). This increase in Brn1 association was observed in both asynchronous cells and cells arrested in G1 phase (Fig 7D). In contrast, there was no significant increase in bulk chromatin association of Brn1 in mitotic cells (Fig 7E). These results are consistent with the observation that increasing Ycg1 expression leads to a greater increase in the levels of intact complex in G1 than in mitosis (Fig 6), and support the possibility that an increase in the association of condensin with chromosomes in G1 phase delays cell-cycle entry.
Although condensin associates with chromosomes throughout the cell cycle, its enrichment at many of its best-characterized binding sites (including the rDNA, centromeres, and telomeres) is substantially higher in mitosis than in G1 [15,25,27,37,52]. Notably, each of these classes of binding sites requires mitosis-specific factors to stimulate this increase in condensin recruitment [24,27,53], which raises the question of whether or not Ycg1 overexpression leads to increased condensin binding to these specific loci during interphase. To address this question, we used chromatin immunoprecipitation and quantitative PCR (ChIP-qPCR) to quantify Brn1 recruitment to a representative set of these sites [16,17,27,53–55]. Interestingly, in asynchronous TEFp-YCG1 cells, Brn1 binding increased at centromeric and telomeric loci, but not the rDNA (Fig 8A). The fact that condensin recruitment to the rDNA is largely unchanged in TEFp-YCG1 strains is consistent with the fact that these cells do not show changes in rDNA condensation (S4 Fig), or in rDNA silencing or stability (S2B Fig).
We next used ChIP-qPCR to examine Brn1 recruitment to mitotic binding sites in cells that were arrested in G1 and metaphase, in order to directly compare binding at these sites to bulk chromatin binding that we had measured using chromosome spreads (Fig 7). These experiments led to two interesting observations. First, consistent with the results of the chromosome spread experiments, Brn1 binding to mitotic sites was not significantly elevated in metaphase cells upon overexpression of Ycg1 (Fig 8B). (Although binding at centromeres tended to be slightly elevated in TEFp-YCG1 cells, the data did not reach statistical significance, and a modest reduction in binding to the rDNA was observed.) The second conclusion from these data is that although Brn1 bound to the rDNA, centromeres, and a telomere in metaphase, binding at each of these sites was reduced to background levels in both wild-type and TEFp-YCG1 strains that were arrested in G1 (Fig 8B). This result indicates that although total Brn1 binding to chromosomes is elevated in TEFp-YCG1 strains in G1 (Fig 7D), the complex is not enriched at mitosis-specific target sites. In addition, the increased binding of condensin to centromeres and telomeres that is seen in asynchronous TEFp-YCG1 cells is likely to result from increased binding at a point in the cell cycle other than G1 or metaphase.
Here, we show that cyclical transcription and proteasomal degradation regulate Ycg1 levels during the cell cycle, which in turn modulates condensin complex formation. Since Ycg1 is essential for condensin function [5–8], downregulation of its expression after mitosis (Fig 1) is predicted to reduce the amount of condensin complex and thereby decrease its association with chromosomes and activity. Indeed, we demonstrate that the amount of intact condensin complex is reduced in G1, concurrent with low Ycg1 expression, and increases during mitosis, when Ycg1 expression peaks (Figs 1 and 6). Our results also argue that Ycg1 levels are limiting, since overexpression of Ycg1 was sufficient to both increase complex formation (Fig 6) and recruitment to chromatin (Fig 7), as well as slow proliferation (Fig 5). In contrast, individual overexpression of the other four condensin subunits had no effect on proliferation rate (Fig 5A). Intriguingly, we found that the reduction in proliferation rate in YCG1-overexpressing cells was caused by a delay in progression through the G1/S transition (Fig 5D and 5E). These findings suggest that downregulation of Ycg1 is important to decrease condensin activity after mitosis, thereby allowing cells to proceed through interphase.
Although several studies have reported that YCG1 mRNA is cell cycle-regulated [38–40], the question of whether or not Ycg1 protein cycles has not been addressed. Indeed, a prior study found that Ycg1 protein is expressed at lower levels in G1 than in S phase and mitosis [22], whereas others show a more constitutive expression pattern across the cell cycle [14,24]. For this reason, we analyzed the expression of Ycg1 in different strain backgrounds with different epitope tags (Fig 1B–1D). Importantly, in each case we found that Ycg1 cycled and mirrored the mRNA expression pattern (Fig 1). Furthermore, since disrupting cyclical expression of Ycg1 increased condensin complex formation and slowed proliferation (Figs 4A, 5A and 6), we conclude that cyclical expression of Ycg1 is functionally important for cell-cycle progression.
Previous studies have shown that phosphorylation of condensin subunits by mitotic kinases stimulates the supercoiling activity of the complex [13,14,22–24,37], suggesting that phosphorylation is one mechanism that helps restrict chromosome condensation to mitosis. In addition, recruitment of the complex to specific sites is known to be dependent upon mitosis-specific factors, such as Sgo1, which recruits condensin to centromeres in S phase through mitosis [26,27]. Our results reveal an additional regulatory mechanism that contributes to the reduction in condensin activity after mitosis is complete. Ycg1 levels limit the amount of condensin complex early in the cell cycle, and by extension reduce the amount of condensin that is available to act on chromatin. These findings suggest a revised model in which condensin complex formation, recruitment to a subset of binding sites, and phosphorylation are regulated to ensure that condensin activity is at its lowest level during G1 phase [13,14]. As cells progress through S phase into mitosis, Ycg1 levels rise, condensin complex formation increases, and more complex is loaded onto chromatin. Finally, during mitosis condensin is recruited to several mitosis-specific sites [25–27,37], and the complex is activated by mitotic kinases to increase its supercoiling activity [13,14]. Thus, complex formation, the availability of recruitment factors, and phosphorylation act together to establish different states of condensin activity in different cell-cycle stages.
One interesting possibility raised by our results is that constitutive expression of Ycg1 could disrupt the timing of chromosome condensation, or lead to precocious condensation of chromosomes early in the cell cycle. We tested this possibility by examining condensation of the rDNA, which undergoes the most dramatic condensin-dependent structural change during mitosis in yeast, but did not observe any change in condensation in TEFp-YCG1 cells arrested in metaphase, or any increased condensation in cells arrested in G1 (S4 Fig). Consistent with this result, condensin binding to the rDNA did not increase upon Ycg1 overexpression (Fig 8); therefore, factors that promote mitotic enrichment of condensin at the rDNA are likely necessary to drive rDNA compaction. It remains possible that the timing of condensation is altered as cells enter or exit mitosis. Alternatively, precocious condensation could occur elsewhere in the genome. However, since the activating phosphorylations on the complex are absent G1 phase [14,24], a likely possibility is that excess condensin in G1 does not drive precocious condensation but instead binds to chromatin and increases interactions between distant sites in the genome, or physically blocks the chromatin association of transcription or replication factors.
Although cells that express stable Ycg1 (ycg1-k977A) and those that overexpress wild-type Ycg1 show similar delays in cell-cycle entry, we find that they respond differently to some cell-cycle perturbations. Notably, ycg1-K977A cells are sensitive to the replication inhibitor hydroxyurea (HU), whereas TEFp-YCG1 cells are not (S2D Fig). HU sensitivity has been previously reported in fission yeast expressing a temperature-sensitive allele of the kleisin subunit of condensin, Cnd2 [45]. Thus, HU sensitivity is consistent with the possibility that the K977A mutation in Ycg1 partially impairs some aspect of condensin function, while still promoting an increase in complex levels in G1 phase that delays cell-cycle entry.
Stabilization and overexpression of Ycg1 also result in different responses to the microtubule poison benomyl, which activates the spindle assembly checkpoint (S2E Fig). This finding is intriguing because condensin has an established function at centromeres, where it promotes chromosome biorientation by biasing kinetochores for capture by microtubules from opposite poles [26,56]. Interestingly, although ycg1-K977A strains exhibit increased sensitivity to benomyl (consistent with a partial loss of function), cells overexpressing Ycg1 are more resistant to spindle disruption than wild-type cells. This raises the possibility that when Ycg1 is expressed at high levels early in the cell cycle, more condensin may be loaded at centromeres, which could enable cells to respond better to spindle disruption. Our data is consistent with this hypothesis. Although we did not observe a significant increase in condensin recruitment to centromeres in G1 or metaphase-arrested cells (Fig 8B), recruitment was increased in asynchronous TEFp-YCG1 cells compared to wild-type (Fig 8A). In the future it will be interesting to examine the dynamics of condensin recruitment to centromeres during the cell cycle, in order to determine if condensin is recruited to centromeres earlier in S-phase when Ycg1 is overexpressed, or if it persists at centromeres longer as cells progress through mitosis.
Although our data shows that condensin is regulated by limiting expression of the Cap-G subunit in budding yeast, evidence suggests that similar mechanisms control the activity of condensin in other systems. Indeed, proteolytic regulation of condensin also occurs in Drosophila melanogaster, via targeting of the kleisin subunit of condensin II, Cap-H2 [32]. In that system, blocking Cap-H2 degradation results in increased chromosome condensation in interphase cells [32,33]. However, it remains to be determined if stable Cap-H2 can delay the G1/S transition, as Ycg1 stabilization does in yeast.
Notably, although Ycg1 and Cap-H2 are similarly regulated, they are not orthologs [11]. Indeed, we posit that the existence of a rate-limiting subunit of the condensin complex may have evolved independently in fungi and animals, with different subunits being targeted for degradation. Importantly, the presence of proteolytic regulation in two evolutionarily distant eukaryotes, and the interphase phenotypes observed when proteolysis is disrupted, suggests that this regulation may be an important mechanism to limit condensin function in all eukaryotes. Human condensin II-specific subunits are also reported to undergo proteolytic regulation [35,36], and in the future it will be of interest to determine whether any of these subunits are rate limiting in mammalian cells. Limiting the levels of a condensin subunit is a mechanism that is likely to coordinate changes in chromosome structure with cell-cycle stage in all eukaryotes, and may also have broader roles in modulating condensin activity in response to specific environmental signals.
A complete list of strains used in this study can be found in S1 Table. All experiments were performed at 30°C, unless otherwise indicated. Strains were grown in rich medium with 2% dextrose, except for strains harboring MET3p-CDC20, which were grown in synthetic complete medium lacking methionine with 2% dextrose.
Epitope-tagging of genes was achieved by integrating 3HA-His3MX6, 3V5-kanMX6, or 13Myc-His3MX6 in place of the stop codon at the genomic locus of each gene, as indicated in S1 Table. To generate strains that could be synchronized in metaphase, the methionine-regulatable MET3 promoter was integrated upstream of CDC20 using plasmid pBO1105. pBO1105 is a modification of YIp22(TRP1) MET3p-CDC20 [57] in which the YIp22 vector has been replaced with pAG25 (J.J. Li, personal communication). Where indicated, the TEF1 promoter was integrated upstream of the start codons of condensin subunits, as previously described [58]. Mutations in YCG1 were introduced into the genome by deleting the non-essential 3’ end of the gene, followed by integration of PCR products that replace the 3’ sequence and include the indicated mutations. All mutations were confirmed by sequencing. For proteasome inhibition experiments, Ycg1 was tagged in strain YUS5, which carries mutations that increase its sensitivity to proteasome inhibitors [59,60]. To assay silencing and recombination at the rDNA locus, ycg1-K977A and TEFp-YCG1 were integrated into strain JS306 and strains were assayed as previously described [44]. To integrate an extra copy of YCG1 at the URA3 locus, YCG1 (with 362 base pairs of its upstream sequence) was cloned into pRS306 and the resulting vector was digested with NcoI for integration at URA3. Single copy integration was confirmed by PCR. Strains expressing temperature-sensitive condensin alleles were previously described in [48].
To assay protein degradation, cycloheximide (50 μg/mL) was added to cells and samples taken after the indicated number of minutes. At each time point equivalent optical densities of cells were collected. To assay stability upon proteasome inhibition, cells were grown in synthetic complete medium lacking proline with 0.003% SDS and 2% dextrose, then treated with DMSO or 5 μg/ml MG132 for 2 hours prior to the addition of cycloheximide. Where indicated, cells were arrested with 10 μg/ml alpha-factor for 2.5 hours, or 10 μg/ml nocodazole for 2 hours, before the addition of cycloheximide. In all experiments cell-cycle arrest was verified by flow cytometry.
Samples were prepared for Western Blotting by resuspending equivalent optical densities of cells in preheated SDS sample buffer (50 mM Tris pH 7.5, 5 mM EDTA, 5% SDS, 10% glycerol, 0.5% β-mercaptoethanol, bromophenol blue, 1 μg/ml leupeptin, 1 μg/ml bestatin, 1 mM benzamidine, 1 μg/ml pepstatin A, 17 μg/ml PMSF, 5 mM sodium fluoride, 80 mM β-glycerophosphate and 1 mM sodium orthovanadate), followed by incubation at 95°C for 5 minutes. Glass beads were then added and samples were bead beat using a Biospec Mini-Beadbeater for 3 minutes. Samples were clarified by centrifugation and analyzed by SDS-PAGE followed by Western blotting. Western blots were carried out with antibodies against GFP (clone JL-8, Clontech), Clb2 (y-180, Santa Cruz Biotechnology), Cdc28/Cdk1 (yC-20, Santa Cruz Biotechnology), HA (clone 12CA5), V5 (ThermoFisher), Myc (clone 9E10, Covance), and G6PDH (Sigma). Where indicated, quantitation was performed using a BioRad ChemiDoc Touch imaging system and the accompanying ImageLab software.
G1 cell-cycle arrest was achieved by incubating logarithmic-phase cells with 10 μg/ml alpha-factor for 2–3 hours, as indicated. Mitotic arrest was achieved by treating cells with 10 or 20 μg/ml nocodazole for 2–3 hours, or by adding 5X L-methionine (0.1 mg/L final concentration) to MET3p-CDC20 strains (growing in medium without methionine) for 3.5 hours. Where indicated MET3p-CDC20 strains were arrested in mitosis as above, then released into medium without methionine containing alpha-factor for 2.5 hours to synchronize cells in G1, followed by release into medium without methionine or alpha-factor. Details of specific arrest-release experiments are indicated in the figure legends.
Cell-cycle positions were confirmed by flow cytometry. Cells were fixed and labeled with Sytox Green (Invitrogen) as previously described [61]. Samples were analyzed using a FACScan (Becton Dickinson) and data analyzed with FlowJo (Tree Star, Inc.) software. Where indicated, fixed cells were sonicated and percentage of budded cells determined by counting at least 100 cells/sample.
rDNA silencing and stability were assayed in strains derived from JS306, as previously described [44]. In these strains, two PolII-regulated marker cassettes are integrated into different rDNA repeats: a single MET15 reporter gene (embedded in a Ty1 element) is integrated within NTS2 of one rDNA repeat, and a mURA3/HIS3 expression cassette is integrated within the 18S rRNA-coding region of a second repeat.
In this assay, the MET15 reporter is used to score an increase in recombination between rDNA repeats. The expression of MET15 results in white colonies on MLA plates (Pb+ plates), loss of the MET15 gene results in dark brown colonies or sectors (as seen in the sir2Δ strain), and if the MET15 gene is present, but is silenced, the colonies are a tan color. Strains are scored as having increased recombination between rDNA repeats if dark brown and sectored colonies are observed on MLA plates, which indicates loss of the MET15 gene. Although a tan color indicates MET15 gene is present, but silenced, the shade of tan is variable between experiments and therefore not used to infer the degree of silencing.
In the same strains the mURA3/HIS3 reporter is used to assay silencing. Strains that are capable of silencing do not express mURA3 and thus can’t grow on—Ura plates, however HIS3 is incompletely silenced so strains can grow on—His plates. For this reason, growth on—His is used as a confirmation that the strains retain the mURA3/HIS3 cassette. Strains that grow similarly on—His and—Ura plates are scored as having a loss of silencing of the rDNA locus. sir2Δ mutants were previously shown to have both decreased silencing and increased recombination [44], and serve as a positive control for both readouts.
Cell pellets from 30 optical densities of arrested cells were lysed by resuspension in HEPES lysis buffer (25mM HEPES-OH pH 7.5, 250mM NaCl, 0.2% Triton, 1mM EDTA, 10% glycerol, 1 μg/ml leupeptin, 1 μg/ml bestatin, 1 mM benzamidine, 1 μg/ml pepstatin A, 17 μg/ml PMSF, 5 mM sodium fluoride, 80 mM β-glycerophosphate and 1 mM sodium orthovanadate), followed by 3 cycles of bead-beating for one minute each (with 5 minute incubations on ice between cycles). Protein concentrations were measured by Bradford assay and equal amounts of total protein were incubated with 2μL mouse anti-V5 antibody (ThermoFisher) for 3 hours, followed by addition of 25μL protein G magnetic beads (NEB) for 1 hour. Beads were washed 3X with HEPES lysis buffer and proteins were eluted by boiling in 2X sample buffer.
Cultures were grown to logarithmic phase, then diluted to 0.1 optical densities and 100μL of each was added in triplicate to a round bottom 96-well plate. Cell proliferation was monitored by growing cultures at 30°C with shaking in a Tecan Infinite M200 Pro plate reader and measuring optical density at 600nM every 20 minutes until cultures reached approximately 0.8 OD. Doubling times were calculated by fitting data points between 0.15 OD and 0.6 OD to an exponential growth equation using GraphPad Prism software.
Chromosome spreads to analyze condensin association with chromatin and rDNA morphology were performed as previously described [7,17,51]. 3HA-tagged Ycg1 was detected with mouse anti-HA antibody (clone 12CA5), 3V5-tagged Brn1 and 3V5-tagged Net1 were detected with mouse anti-V5 antibody (ThermoFisher), all in combination with Alexa Fluor 488-conjugated goat anti-mouse IgG (ThermoFisher) and DAPI. A wild-type strain lacking both epitope tags was used as a negative control in all experiments. To quantify Ycg1 and Brn1 chromatin binding, Alexa Fluor 488 fluorescence intensities within an area encompassing the merged Alexa Fluor and DAPI images were measured after background subtraction in ImageJ software. At least 190 cells were quantified for each sample, in each experiment. To score condensation of the rDNA, the rDNA structure (evident both by Net1 staining and the conformation of the DAPI-stained nucleolar DNA) in at least 200 cells were classified as either puffs (decondensed) or loop/lines (condensed), as previously described [7,22,49]. For all chromosome spreads performed on synchronized cultures, cells were first arrested with 10μg/ml alpha-factor or 20μg/ml nocodazole for 3 hours.
Chromosome immunoprecipitation (ChIP) was performed as previously described [37] with the following modifications. For asynchronous and nocodazole-arrested cultures, 40 optical densities (ODs) of each culture were lysed in a Mini-Beadbeater (Biospec) and lysates were sonicated using a Diagenode Biorupter. For alpha-factor arrested cultures, 70 OD were used. Brn1-3V5 was immunoprecipitated with mouse anti-V5 (ThermoFisher) coupled to Protein G magnetic beads (New England Biolabs). Eluted DNA was quantified by qPCR on an Eppendorf Realplex system. Primers used for qPCR are listed in S2 Table.
|
10.1371/journal.ppat.1004979 | EBV BART MicroRNAs Target Multiple Pro-apoptotic Cellular Genes to Promote Epithelial Cell Survival | Epstein-Barr virus (EBV) is a ubiquitous human γ-herpesvirus that can give rise to cancers of both B-cell and epithelial cell origin. In EBV-induced cancers of epithelial origin, including nasopharyngeal carcinomas (NPCs) and gastric carcinomas, the latent EBV genome expresses very high levels of a cluster of 22 viral pre-miRNAs, called the miR-BARTs, and these have previously been shown to confer a degree of resistance to pro-apoptotic drugs. Here, we present an analysis of the ability of individual miR-BART pre-miRNAs to confer an anti-apoptotic phenotype and report that five of the 22 miR-BARTs demonstrate this ability. We next used photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP) to globally identify the mRNA targets bound by these miR-BARTs in latently infected epithelial cells. This led to the identification of ten mRNAs encoding pro-apoptotic mRNA targets, all of which could be confirmed as valid targets for the five anti-apoptotic miR-BARTs by indicator assays and by demonstrating that ectopic expression of physiological levels of the relevant miR-BART in the epithelial cell line AGS resulted in a significant repression of the target mRNA as well as the encoded protein product. Using RNA interference, we further demonstrated that knockdown of at least seven of these cellular miR-BART target transcripts phenocopies the anti-apoptotic activity seen upon expression of the relevant EBV miR-BART miRNA. Together, these observations validate previously published reports arguing that the miR-BARTs can exert an anti-apoptotic effect in EBV-infected epithelial cells and provide a mechanistic explanation for this activity. Moreover, these results identify and validate a substantial number of novel mRNA targets for the anti-apoptotic miR-BARTs.
| One important innate immune response to viral infection is apoptosis, also called programmed cell death, whereby the infected cells commit suicide rather than serve as factories for virus production. As a result, many viruses have developed strategies to inhibit apoptosis. Here, we demonstrate that five of the Epstein-Barr virus (EBV) miR-BART microRNAs that are expressed in EBV-transformed epithelial cell tumors display anti-apoptotic activity. We have identified ten cellular mRNAs that are bound and downregulated by one of these five anti-apoptotic microRNAs and show that this downregulation can explain the observed reduction in apoptosis in miR-BART-expressing cells. Together, these data demonstrate that the EBV miR-BARTs can help sustain latently EBV-infected cells in the face of pro-apoptotic innate immune signals and this may explain the resistance to DNA damaging agents, including chemotherapeutics and radiation, seen in a subset of EBV-induced epithelial tumors.
| MicroRNAs (miRNAs) are 22 ± 2 nucleotide (nt) non-coding RNAs that are expressed by all multicellular eukaryotes as well as by several viruses [1–3]. MiRNAs are generally initially transcribed by RNA polymerase II in the form of a long primary miRNA (pri-miRNA) precursor that is sequentially processed by the RNase III enzymes Drosha, in the nucleus, to generate the pre-miRNA intermediate and Dicer, in the cytoplasm, to yield the mature miRNA [1, 4]. Upon loading into the RNA-induced silencing complex (RISC), the miRNA serves as a guide RNA to direct RISC to partially complementary target sites [5]. Particularly important in this regard is the miRNA seed sequence, extending from position 2 to 8 on the miRNA, which is exposed during mRNA binding by RISC and plays a key role in target mRNA recognition [5, 6]. Because seed sequence complementarity to an mRNA target is generally not only necessary but frequently also sufficient for effective RISC recruitment, it is predicted that each miRNA functionally interacts with >100 mRNA targets. RISC binding in turn results in the translational inhibition and partial destabilization of the target mRNA [5]. The accurate identification of these mRNA targets, and more importantly, the discovery of mRNA targets that are phenotypically relevant, remains the most difficult challenge in understanding miRNA function. This is particularly difficult in the case of virally encoded miRNAs as these are subject to rapid evolution and, unlike cellular miRNA target sites, which have co-evolved with host cell miRNAs, cellular mRNA targets for viral miRNAs are generally not evolutionarily conserved. Efforts to identify important mRNA targets for viral miRNAs have therefore generally followed one of two approaches, which have been respectively referred to as the “bottom up” and “top down” approach [2]. In the “top down” approaches, the investigator first identifies a phenotype exerted by a miRNA then seeks to determine which mRNA target(s) is responsible for this phenotype. Conversely, in the “bottom up” approach, the investigator first uses computational methods or experimental techniques, such as microarray analysis or a cross-linking/immunoprecipitation approach, to globally identify mRNA targets for a given viral miRNA then seeks to confirm that the phenotypic effect predicted upon downregulation of a given mRNA target is actually observed. These approaches are not, of course, mutually exclusive as tools for the global identification of mRNA targets for a given viral miRNA can provide critical information for efforts to identify the mRNA target(s) that explain a miRNA phenotype.
Epstein-Barr virus (EBV) encodes two miRNA clusters that are differentially expressed during latent EBV infection [7–10]. In latency III, as seen for example in lymphoblastoid cell lines (LCLs) of primary B-cell origin, EBV expresses a high level of the three viral pre-miRNAs encoded in the miR-BHRF1 cluster and moderate levels of the 22 pre-miRNAs encoded in the miR-BART cluster [7, 10, 11]. Consistent with this expression pattern, mutational inactivation of the miR-BHRF1 cluster severely impairs B-cell transformation by EBV, with the resultant LCLs showing a slow growth phenotype, while loss of all 22 miR-BARTs has at most a modest effect on B-cell transformation [12–15]. Conversely, in EBV-transformed epithelial cells that are in latency II, including nasopharyngeal carcinoma (NPC) cells and EBV-induced gastric carcinomas, the miR-BHRF1 cluster is not expressed while the miR-BARTs are transcribed at substantial levels [7, 9, 10, 16, 17]. Whether the miR-BART miRNAs are required for the transformation of primary human epithelial cells by EBV remains unclear, due to the lack of good in vitro systems to study this process. However, analysis using the gastric carcinoma cell line AGS strongly suggests that this is likely to be the case. AGS cells are normally EBV-negative but can be readily infected by EBV to establish a latent infection marked by high level expression of the miR-BARTs, as well as the viral EBNA1 protein and the EBER non-coding RNAs, but only very low levels of the other viral latent proteins, including LMP1 and EBNA2 [18, 19]. Strikingly, EBV+ AGS cells show enhanced anchorage independent cell growth and the ectopic expression of the miR-BART miRNAs in AGS cells also inhibits apoptosis [18, 20, 21]. This latter result is consistent with a number of reports that have provided evidence for the downregulation of pro-apoptotic cellular genes by individual miR-BART miRNAs [14, 20–24]. However, at present a full understanding of how the EBV miR-BART miRNAs inhibit apoptosis to promote the viability of EBV-infected epithelial cells remains elusive. Here, we report a systematic effort to identify pro-apoptotic mRNA targets for the EBV miR-BART miRNAs. We demonstrate that at least five of the 22 miR-BART pre-miRNAs have anti-apoptotic activity and we identify seven pro-apoptotic cellular mRNA targets, six of them novel, that contribute significantly to the observed anti-apoptotic phenotype. Together, these data represent a substantial increase in our understanding of the role of the miR-BART miRNAs in promoting EBV infection and latency.
Although no system for the study of transformation of primary human epithelial cells by EBV is currently available, the human EBV-negative epithelial cell line AGS, derived from a gastric carcinoma, has emerged as a useful model system [18]. In particular, infection of AGS with EBV, which results in latently EBV infected AGS cells that express high levels of the miR-BART miRNAs, has been associated with enhanced anchorage independent growth in vitro, enhanced tumor formation in vivo in mice and a reduction in apoptosis [18, 21, 25]. As apoptosis is well established as a key innate immune response to viral infection [26, 27], and given several reports suggesting that individual miR-BARTs can target specific pro-apoptotic cellular genes to promote cell survival [14, 20–24], we decided to systematically analyze the anti-apoptotic potential of the 22 miR-BART pre-miRNAs in the human epithelial cell line AGS.
The phenotypic effect of a given miRNA is in part determined by the expression level of the miRNA relative to its potential pool of mRNA targets so that ectopic overexpression of a given miRNA can give rise to phenotypic effects that are not seen at physiological levels of expression [28]. Therefore, to identify EBV miR-BART miRNAs with anti-apoptotic potential, we decided to express each of the 22 miR-BARTs at approximately physiological levels using lentiviral miRNA expression vectors, constructed as previously reported by inserting the entire pri-miR stem-loop, together with ~100 bp of 5’ and 3’ flanking sequence, into the 3’ UTR of the turbo red fluorescent protein (turboRFP) gene present in pTRIPZ [29, 30]. After selection for the included puromycin marker, the cells were sorted for high turboRFP expression (upper 30%) and expanded. To analyze the expression level of each miR-BART miRNA in each AGS transductant, we harvested total RNA from each culture and then used qRT-PCR to compare the level of expression to that seen in the EBV latency II NPC cell line C666. As may be observed in Fig 1, we achieved stable expression of levels of several of the miR-BARTs in AGS cells that were comparable to the endogenous levels seen in C666. Eight of the miR-BARTs (miR-BART 2, 7, 8, 9, 10, 11, 16 and 18) were expressed at levels slightly above that seen in C666 while a further nine (miR-BART1, 3, 5, 6, 14, 17, 19, 21 and 22) were expressed at levels close to, but slightly below, the levels seen in C666. Finally, the remaining five miR-BARTs were either expressed at levels >10-fold lower than seen in C666 (miR-BART4, 12, 13 and 15) or were not detected (miR-BART20). Therefore, these latter transductants can be viewed essentially as negative controls.
Previously, Marquitz et al. [20] reported that ectopic expression of clusters of EBV BART miRNAs in AGS cells (either miR-BART1, 3, 4, 5, 6, 15, 16 and 17 or miR-BART7, 8, 9, 10, 11, 12, 13, 14, 18, 19 and 20) confers resistance to apoptosis induced by treatment with etoposide and we therefore initially examined whether expression of any of these individual BART miRNAs would exert a similar phenotypic effect. As shown in Fig 2A and 2B, we observed a significant reduction in the level of apoptotic cells in the AGS cultures expressing pre-miR-BART3, 6, 8, 16 and 22 when compared to the other 18 cultures. This result, which was initially obtained by quantitation of the sub-G1 population of AGS cells by FACS, could also be largely confirmed by Western blot analysis for cleaved and uncleaved PARP expression, with the AGS cultures expressing miR-BART6, 8, 16 or 22 showing significantly reduced levels of cleaved PARP after etoposide treatment (we did not see a statistically significant reduction in the case of miR-BART3) (S1A and S1B Fig).
The inhibition of etoposide-induced apoptosis observed in Fig 2 and S1 Fig in the cultures expressing miR-BART6, 8, 16, 22, and possibly also miR-BART3, presumably reflects the downregulation of one or more mRNAs with pro-apoptotic potential by each of these viral miRNAs. A number of possible mRNA targets for individual miR-BARTs have been reported, some of which have clear pro-apoptotic potential [14, 20–24]. We therefore wondered if the anti-apoptotic activity of the five miR-BART miRNAs defined in Fig 2 could be explained by these previously published target mRNAs. To address this question, we generated indicator constructs in which the 3’ UTR of a cellular gene of interest (either the complete 3’UTR or a segment of 540 bp or more, see S1 Table) was inserted 3’ to the firefly luciferase (FLuc) indicator gene [29]. Then, 293T cells were co-transfected with the relevant indicator construct, a miR-BART miRNA or control expression vector and an internal control plasmid expressing Renilla luciferase (RLuc). At ~72 h post-transfection, the cells were lysed and the relative expression of FLuc and RLuc, in the presence and absence of the miR-BART miRNA, quantified. In general, our experience has been that this assay format produces a readily detectable, >20% reduction in FLuc expression that is both reproducible and statistically significant. All miR-BART miRNA expression vectors tested were fully biologically active, as previously demonstrated [30] by their ability to downregulate a similar FLuc-based indicator construct containing a perfectly complementary target site inserted into the 3’UTR.
As shown in Fig 3, the 3’UTRs of the cellular mRNAs encoding DICE1, a proposed target for miR-BART3 [23], Dicer, a proposed target for miR-BART6 [31], and TOMM22, a proposed target for miR-BART16 [32], all produced a readily detectable inhibitory effect on FLuc expression when present in cis in cells expressing the cognate miR-BART miRNA. In contrast, we did not see a significant reduction mediated by the 3’UTR of STAT1, a proposed target for miR-BART8 [33], or by the 3’UTR of caspase 3 (CASP3), a proposed target for miR-BART16 [14]. We note that neither of these 3’UTRs contains an intact seed target for miR-BART8 or miR-BART16, respectively, so the observed lack of inhibition was not unexpected. While the 3’ UTR of CASP3 has been suggested to function as a target for several miRNAs encoded by the human γ-herpesvirus KSHV [34], we did not observe a significant inhibition of an FLuc indicator bearing the 3’UTR of CASP3 in the presence of miR-BART3, 6, 8, 16 or 22, although the modest, ~20% repression seen with miR-BART3 did approach significance (S2B Fig). Similarly, the 3’UTR of the pro-apoptotic cellular gene BIM, another proposed miR-BART target [20], also did not function as an effective target for miR-BART3, 6, 8, 16 or 22 in this indicator assay (S2A Fig). It remains possible that the simultaneous expression of several miR-BARTs might induce a more marked inhibitory effect. Indeed, Marquitz et al. [20] reported that an analogous indicator containing the BIM 3’UTR was not affected by co-expression of any individual miR-BART miRNA but was inhibited by the simultaneous expression of multiple BART miRNAs. However, as the anti-apoptotic phenotypes shown in Fig 2 result from the expression of individual miR-BART miRNAs, it is apparent that mRNA targets relevant to these phenotypes must be significantly responsive to these individual miRNAs. Based on these results, it therefore appears that while the anti-apoptotic effect observed for miR-BART3 (Fig 2) might be explained by downregulation of the pro-apoptotic gene product encoded by cellular DICE1 [23], the analogous effects exerted by miR-BART6, miR-BART8, miR-BART16 or miR-BART22 are not readily accounted for by previously reported mRNA targets for those miRNAs. Specifically, while we could confirm downregulation of Dicer mRNA function by miR-BART6 and TOMM22 mRNA function by miR-BART16 (Fig 3), neither of these two proteins is known to be pro-apoptotic [31, 35]. Conversely, we did not observe significant downregulation mediated by the 3’UTRs of the pro-apoptotic genes STAT1, CASP3 or BIM in the presence of any of these individual miRNAs (Fig 3 and S2 Fig). We therefore next sought to globally identify the mRNA targets for the EBV miR-BART miRNAs in the naturally EBV-infected epithelial cell line C666 using the previously described photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP) technique [11, 36, 37].
We initially performed deep sequencing of the small RNA population (~18 to ~24 nt in size) in C666 cells, as previously described [11, 37]. This resulted in a total of 26.6x106 reads, of which 25.8x106 (~97%) mapped to either the human or EBV genome (See S2 Table for the complete data set). Of these, 23.9x106 (~93%) represent known mature miRNAs or miRNA passenger strands, with 6.7x106 (~28%) mapping to the EBV miR-BART locus and the remaining 17.2 x106 reads (~72%) representing known human miRNAs (Fig 4A). Among the miR-BARTs, we recovered reads from all 22 known miR-BART miRNA and miRNA passenger strands, but only those miRNA passenger strands representing ≥10% of the total reads derived from a given pre-miR-BART intermediate are shown in Fig 4B. The most highly expressed miR-BARTs detected in C666 cells were miR-BART2-5p, miR-BART9-3p and miR-BART19-3p.
Next we performed PAR-CLIP to globally identify the mRNA targets for the miR-BARTs in C666 cells using an antibody that immunoprecipitates all four human Argonaut (Ago) proteins. The PAR-CLIP library gave 16.7x106 reads, of which 6.6x106 could be mapped to a unique sequence present in either the human or EBV genome (see S3 Table for the complete data set). Computational definition of binding site clusters and assignment to expressed miRNAs [38] revealed that the majority of both the cellular miRNA and miR-BART clusters mapped to 3’UTRs, although significant numbers of clusters also were observed in mRNA coding sequences (CDS) or in intronic regions (Fig 4C). Of the total number of 3’UTR miRNA binding clusters that were detected, 792 were computationally assigned to one of the EBV miR-BART miRNAs or to a miR-BART passenger strand, based on seed homology, including a moderate number of potential mRNA targets for the anti-apoptotic miRNAs miR-BART3, 6, 8, 16 and 22. Inspection of these mRNA targets revealed several with known pro-apoptotic activity including FEM1B and CASZ1a (miR-BART3), OCT1 (miR-BART6), ARID2 (miR-BART8), CREBBP and SH2B3 (miR-BART16) and finally PPP3R1, PAK2 and TP53INP1 (miR-BART22) (see S4 Table for a summary the known functions of these gene products and relevant citations). As noted above, all of these mRNAs contained 3’UTR targets, identified by PAR-CLIP, that bear full seed homology to the indicated miR-BART miRNA (Fig 5A). We therefore used PCR to clone the 3’UTRs of each of these human mRNAs (see S1 Table for sequence coordinates) and inserted these 3’ to the FLuc gene, as described in Fig 3. As may be observed (Fig 5B), every 3’ UTR tested conferred substantial inhibition of FLuc activity in cells co-expressing the cognate miR-BART miRNA that was statistically significant (p<0.05). All these 3’UTRs contained a single predicted miR-BART target site that was captured by PAR-CLIP except for CASZ1a, which also contained a captured site with seed homology to miR-BART18. However, this potential target was not responsive to co-expressed miR-BART18 in co-transfected cells (Fig 5B).
In order to test whether the suppression of Fluc activity by a given miR-BART is indeed due to the seed homology present in the clusters captured by PAR-CLIP, we introduced transversion mutations at the 3’UTR nucleotides pairing to miRNA seed positions 2, 4 and 6 in eight of the nine captured clusters in these 3’UTRs. In the ninth 3’UTR, derived from SH2B3, the predicted target site for miR-BART16 was removed by deletion. The 3’UTR of DICE1 was not mutated, as the single miR-BART3 binding site present in this 3’UTR has been previously validated (23). By assaying 3’UTR-containing FLuc reporters containing these mutations, in parallel with the wild-type 3’UTR-based FLuc reporters used in Fig 5B, we observed that loss of seed homology in the captured PAR-CLIP clusters resulted in the complete loss of miR-BART mediated inhibition of FLuc activity, consistent with our hypothesis that the clusters captured by PAR-CLIP are bound by the predicted miR-BARTs (S3A Fig). However, in the case of CASZ1a, mutation of the single predicted miR-BART3-5p target site only led to a partial recovery of FLuc activity, indicating an additional miR-BART3 target site(s) is present in the CASZ1a 3’UTR. Indeed, we computationally identified a potential site with seed homology to miR-BART3-3p (S3B Fig) that was not detected by PAR-CLIP but that could account for this residual inhibition. Although miR-BART3-3p is in principle a passenger strand, with miR-BART3-5p representing the dominant arm, both strands are actually recovered at similar levels and have a comparable number of binding clusters (Fig 4B). In conclusion, this mutational analysis confirmed that all 9 candidate 3’UTR target sites are indeed binding sites for the predicted miR-BART.
If the pro-apoptotic genes listed in Fig 5 and S4 Table are indeed authentic targets of the miR-BART miRNA listed in the same figure, then expression of physiological levels of that miR-BART in AGS cells should result in a reduction in the expression of that gene [1]. We initially performed qRT-PCR analysis of control AGS cells and of the AGS cells described in Fig 1 that express close to physiological levels of one of the five anti-apoptotic miR-BARTs (Fig 2) looking at the nine cellular mRNAs listed in Fig 5 as well as the mRNA encoding DICE1, a previously reported [23] potentially pro-apoptotic target for miR-BART3 confirmed by indicator assay in Fig 3. As noted in Fig 6A, this analysis is more complex in the case of CASZ1, which is expressed in two spliced variants, encoding CASZ1a and a shorter protein called CASZ1b [39], as only the 3’UTR found in the longer mRNA splicing isoform encoding CASZ1a is predicted to be a target for miR-BART3.
As shown in Fig 6B, we observed a significant (p<0.05) reduction in the level of expression of all the predicted mRNA targets with the exception of the TP53INP1/miR-BART22 combination, where the observed reduction in mRNA expression fell slightly short of significance. Importantly, while expression of the CASZ1a mRNA was significantly reduced in the presence of pre-miR-BART3, expression of the CASZ1b mRNA was not, as predicted (Fig 6A).
While miRNAs can clearly reduce the steady-state expression level of target mRNAs, evidence suggests that a major, and possibly the primary, effect of RISC binding to the 3’UTR of a target mRNA is to reduce the translation of that mRNA [40, 41]. To address whether the anti-apoptotic miR-BARTs indeed reduce the expression of the proteins encoded by the 10 pro-apoptotic genes listed in Figs 5 and 6, we performed Western analyses of all ten proteins in the transduced AGS cells expressing the individual anti-apoptotic miR-BART miRNAs. Representative Western blots are shown in Fig 7A and a compilation of data derived from four independent experiments for each of the 10 proteins is shown in Fig 7B. As may be observed, these data uncover statistically significant (p<0.05) decreases in expression for all 10 cellular proteins under analysis, with the only non-repressed protein being CASZ1b which, as shown in Fig 6A, is encoded by a spliced mRNA isoform that is not expected to bind miR-BART3. In contrast, expression of the CASZ1a protein was, as expected, repressed upon miR-BART3 expression (Fig 7).
The work described so far has identified several putatively pro-apoptotic cellular mRNA targets for the five anti-apoptotic EBV miR-BART miRNAs miR-BART3, 6, 8, 16 and 22 and shown that these are, in fact, downregulated at both the mRNA and protein level in AGS cells expressing physiological levels of the miR-BART miRNA in question. However, these data do not address whether this downregulation is, in fact, causatively related to the observed reduction in apoptosis. To test this hypothesis, we constructed two lentiviral vectors expressing artificial miRNAs (amiRNAs) specific for each of the 10 candidate mRNA targets, a total of 20 vectors [42]. These were used to transduce AGS cells that were then selected for blasticidin resistance and tested for knockdown of the encoded protein target by Western blot. As shown in S4 Fig, 16 distinct amiRNAs demonstrated some degree of knockdown ranging from >10-fold to as little as ~30%, for the targets CASZ1, OCT1, SH2B3, ARID2, PAK2, TP53INP1 and CREBBP as well as DICE1. Unfortunately, we did not observe significant knockdown of PPP3R1 or FEM1B with either amiRNA tested and these two potential targets were therefore not further addressed. However, we were able to test the other 16 amiRNA- expressing AGS cell lines for their ability to resist the induction of apoptosis seen upon incubation in 5 μM etoposide. As shown in Fig 8, we observed a statistically significant (p<0.05) reduction in apoptosis for both amiRNAs specific for CASZ1, DICE1, SH2B3, PAK2 and TP53INP1. We also observed a significant reduction in apoptosis in one of the two cell lines expressing an amiRNA specific for OCT1 and CREBBP, with the other cell line showing a trend towards lower apoptosis that did not achieve significance due to a high standard deviation between assay replicates. Finally, neither amiRNA specific for ARID2 resulted in reduced apoptosis, suggesting that this protein is perhaps not, in fact, functionally pro-apoptotic in AGS cells. In conclusion, our data demonstrate that amiRNAs specific for seven distinct cellular genes identified as targets for anti-apoptotic EBV miR-BART miRNAs are able to phenocopy the anti-apoptotic activity of these viral miRNAs.
The primary goal of this study was to determine if any of the miR-BART miRNAs expressed at high levels in EBV transformed epithelial cells have an anti-apoptotic phenotype and, if so, to identify and validate the cellular mRNA targets that mediate this phenotype. This work was initially prompted by published reports arguing that the expression of clusters of miR-BART miRNAs in the gastric carcinoma cell line AGS inhibits the apoptosis caused by exposure to the topoisomerase II inhibitor etoposide [20] and reports, based largely on computational approaches, that identified several individual pro-apoptotic cellular genes as potential targets for specific miR-BART miRNAs [14, 20–24].
Because the phenotypes exerted by miRNAs can be influenced by their expression level [28], we initially decided to construct stable cell lines, derived from human AGS cells, that individually expressed a close to physiological level of each of the miR-BART miRNAs using lentiviral vector transduction. As shown in Fig 1, we were indeed able to achieve a level of expression in AGS cells that was closely comparable to that seen in the naturally EBV transformed NPC cell line C666 for 17 of the 22 miR-BARTs. Analysis of these cell lines then showed that five of the EBV miRNAs, that is miR-BART3, 6, 8, 16 and 22, were able to reduce the level of apoptosis seen after etoposide treatment (Fig 2 and S1 Fig). We next globally identified the mRNA targets bound by RISC-loaded miR-BART miRNAs by PAR-CLIP analysis of the EBV-transformed epithelial cell line C666, using a pan-Ago antibody. This resulted in the identification of several cellular mRNA targets bound by the five anti-apoptotic miR-BARTs (Figs 4D and 5), nine of which were predicted to encode proteins with pro-apoptotic activity (S4 Table). We were able to further validate these cellular mRNAs as authentic targets for the five anti-apoptotic EBV miR-BART miRNAs by several criteria:
Insertion of the 3’ UTR, including the PAR-CLIP identified miR-BART seed target, 3’ to the FLuc indicator gene conferred specific downregulation of FLuc when the cognate miR-BART miRNA was expressed in trans (Fig 5B). Moreover, this downregulation was dependent on the integrity of the seed target (S3 Fig).
Expression of any one of these miR-BART miRNAs at physiological levels in AGS cells (Fig 1) resulted in a specific and significant downregulation of the level of expression of the endogenous mRNA and protein encoded by the predicted target gene (Figs 6 and 7).
While these four lines of evidence provide strong support for the hypothesis that the 10 genes listed in Figs 6 and 7 (nine of which are novel while one, DICE1, has been previously described [23]) are indeed authentic targets for one of the five anti-apoptotic miR-BARTs, they do not address whether these mRNA targets are directly relevant to the observed anti-apoptotic phenotype (Fig 2). To address this question, we constructed two artificial miRNA (amiRNA) lentiviral expression vectors specific for each of the potential miR-BART mRNA targets tested in Figs 6 and 7. These lentiviral vectors, which are closely similar to the miR-BART lentivectors used in Figs 1, 2, 6 and 7, were then used to generate stably transduced AGS cell lines expressing these amiRNAs. As shown in S4 Fig, we obtained two amiRNAs that each effectively and stably downregulated the expression of eight of these potentially pro-apoptotic genes in AGS cells (we did not obtain amiRNAs able to stably downregulate FEM1B or PPP3R1, either because our amiRNA designs were ineffective or because these proteins are required in AGS cells). Analysis of the resultant 16 stable knockdown AGS cell lines obtained showed a significant reduction in apoptosis levels after etoposide treatment in both cell lines expressing an amiRNA specific for CASZ1, DICE1, SH2B3, PAK2 or TP531NP1 and in one of the two cell lines expressing an amiRNA specific for OCT1 or CREBBP1. Neither amiRNA specific for ARID2 showed an anti-apoptotic phenotype, though both effectively inhibited ARID2 protein expression (S4D Fig). We therefore conclude that we have identified at least seven authentic pro-apoptotic cellular mRNA targets that are significantly downregulated upon expression of one of the anti-apoptotic miR-BART miRNAs at physiological levels in human epithelial cells. These findings can at least partly explain the previously reported anti-apoptotic activity of the miR-BART miRNA cluster in AGS cells [20].
Of the seven anti-apoptotic mRNA targets validated in this manuscript, i.e., CASZ1, DICE1, OCT1, CREBBP, SH2B3, PAK2 and TP53INP1, only one, DICE1 has been previously reported as an mRNA target for miR-BART3 [23]. This was surprising, as a number of other pro-apoptotic cellular mRNAs have also been reported to be targets for miR-BARTs [14, 20–22, 24, 33]. However, as shown in Fig 3 and S2 Fig, we were not able to validate STAT1, CASP3 or BIM as targets for any of the five pro-apoptotic EBV miRNAs miR-BART3, 6, 8, 16 and 22. It remains possible, as previously proposed [20], that the simultaneous expression of multiple miR-BARTs, as seen in EBV-transformed epithelial tumors, would result in significantly reduced expression of STAT1, CASP3 and/or BIM. However, we note that the 3’UTRs of STAT1 and CASP3, which have been reported to be targets for miR-BART8 and miR-BART16 respectively [14, 33], do not contain full seed targets for either of these two EBV miRNAs and neither 3’UTR was in fact identified as a target for miR-BART8 or miR-BART16 binding in the PAR-CLIP analysis reported in Fig 4D and S3 Table.
In addition to the previously reported mRNA targets for the anti-apoptotic miR-BARTs analyzed in Fig 3, several other potentially pro-apoptotic cellular mRNAs have also been previously reported as targets for other miR-BARTs that did not exert a detectable anti-apoptotic phenotype when expressed individually in AGS cells (Fig 2). These include BID, a proposed target for miR-BART4 [21]; PUMA, a proposed target for miR-BART5 [22]; PTEN, a proposed target for miR-BART7 [24]; E-Cadherin (E-CAD), a proposed target for miR-BART9 [43]; and finally, EBF1, a proposed target for miR-BART11 [44]. All of these miRNAs, except miR-BART4, were expressed at physiological levels in the AGS transductants (Fig 1), so the lack of a detectable anti-apoptotic phenotype was unexpected.
Analysis of our PAR-CLIP data, obtained in C666 cells, as well as previous PAR-CLIP experiments, using Ago-specific antibodies, performed using LCLs or PEL cells latently infected with wildtype EBV [11, 37], and expressing readily detectable levels of the miR-BARTs, failed to identify miR-BART binding sites at the proposed locations in the 3’UTRs of any of these mRNAs (S3 Table). Moreover, FLuc-based indicator constructs containing 3’UTRs derived from these five mRNA species either failed to show any evidence of downregulation in the presence of the cognate miR-BART expression plasmid (BID/miR-BART4; PTEN/miR-BART7; E-CAD/miR-BART9) or showed a minimal level of inhibition (PUMA/miR-BART5 and EBF1/BART11) (S5 Fig). We note that the EBF1 3’UTR does not, in fact, contain a seed target for miR-BART11 and is therefore not predicted to be highly responsive to this miRNA. Others have also failed to confirm the identification of PUMA as an authentic target for miR-BART5 using RISC immunoprecipitation or indicator assays [14, 20], so the PUMA 3’UTR, despite the presence of a highly complementary potential 3’UTR target, may not in fact be an authentic target for downregulation by miR-BART5.
In conclusion, we have identified a series of at least seven mRNA targets for EBV miR-BART miRNAs that encode pro-apoptotic proteins. The BART miRNA-induced reduction in the expression of these proteins can at least partly explain the previously reported anti-apoptotic activity of the EBV miR-BART locus in EBV latency II epithelial cells [20]. Clearly, this activity could be highly advantageous to EBV in ensuring the survival of these latently infected cells despite the known ability of EBV to activate innate immune pathways that have the potential to induce programmed cell death pathways [45] and may also contribute to the development of resistance seen in a significant percentage of EBV+ NPC tumors in patients undergoing chemotherapy or radiation therapy [46, 47].
C666 cells (a gift from Dr. Nancy Raab-Traub) [7], AGS cells (a gift from Dr. Lindsey Hutt-Fletcher) [16] and 293T cells (Duke Cancer Institute Cell Culture Facility) were cultured using RPMI 1640, Ham’s F-12 and Dulbecco's Modified Eagle Medium (Gibco), respectively, supplemented with 10% fetal bovine serum and 10 μg/ml gentamicin. All cell cultures were maintained at 37C with 5% CO2.
Lentiviral miR-BART miRNA expression vectors used for FLuc-based 3’UTR reporter assays were generated in the pLenti-CMV-Blasticidin (pLCB) backbone, and individual ~300 bp EBV miRNA expression regions, as previously described [30], were inserted into the 3’UTR of the Blasticidin gene using unique XhoI and XbaI sites. Functional expression of individual miR-BART miRNAs was confirmed using miRNA indicator assays [30] and stem-loop-qRT-PCR (Fig 1).
Lentiviral miR-BART miRNA expression vectors used for transduction of AGS cells were generated in the pTRIPZ backbone (Open Bioystems) (doxycycline inducible turboRFP, puromycin selectable), with EBV miRNA expression regions inserted into the 3’ UTR of the turboRFP gene using XhoI and EcoRI sites.
FLuc-based 3’UTR reporter plasmids were generated using the pLenti-SV40-GL3 (pLSG) backbone [37] by inserting 3’UTRs of candidate cellular target mRNAs (see S1 Table for full description of the inserted sequences) into the 3’UTR of FLuc between unique XhoI and XbaI sites. PCR primers used to clone the 3’UTRs are listed in S1 Table. To generate mutant 3’UTR reporter plasmids, internal primers bearing transversion mutations of the nucleotides pairing to seed positions 2, 4 and 6 of the miRNA were utilized, together with the primers listed in S1 Table, to clone mutant forms of the 3’UTR regions from the wild-type 3’UTR reporter plasmids by overlap extension PCR. To clone a truncated SH2B3 3’UTR, an internal forward primer and the reverse primer listed in S1 Table were utilized. The PCR primers used to clone the mutant 3’UTRs are listed in S7 Table.
Lentiviral amiRNA expression vectors were generated in the pLenti-CMV-Blasticidin-Hairpin (pLCBH) vector. pLCBH was derived from pLCB by inserting a miR-30-based amiRNA cassette [42] into the 3’UTR of the blasticidin gene between the unique XhoI and EcoRI sites. Oligonucleotides used to clone specific amiRNAs are listed in S6 Table.
qRT-PCR for determination of relative mRNA expression and stem-loop qRT-PCR for relative miRNA expression were performed based on vendor protocols. Briefly, total RNA was first harvested using TRIzol (Ambion). For qRT-PCR analysis, RNAs were reverse transcribed using a high capacity reverse transcription kit (Applied Biosystems) and assayed with Power SYBR Green PCR Master Mix (Applied Biosystems). Relative gene expression was first normalized to GAPDH and was then compared to the negative control. Primers used to detect distinct isoforms of CASZ1 mRNA were as previously described [39]. All the qPCR primers used are listed in S5 Table.
For stem-loop qRT-PCR, total RNA preparations were reverse transcribed using a Taqman miRNA reverse transcription kit (Applied Biosystems), and assayed with Taqman Universal PCR Master Mix, no UNG (Applied Biosystems). The relative miRNA expression level of individual miR-BART miRNAs expressed in transduced AGS cells was first normalized to endogenous U6, and then to the miR-BART miRNA level detected in C666 cells. All the EBV miR-BART reverse transcription primers and stem-loop qPCR probes were purchased from Life Technologies.
105 AGS cells were plated into each well of a 6-well plate, and after 24 h, a final concentration of 5 μM/ml etoposide (SigmaAldrich) (25 μM/ml for PARP cleavage) was added to the medium. After 24 h, both floating and adherent cells were harvested, pooled together and fixed with 80% ethanol for ~4 h. Cells were then stained with PI solution with RNase A (BD Pharmingen) and analyzed by flow cytometry. Data were further analyzed by FlowJo (Treestar).
Cells were lysed in NP40 lysis buffer supplemented with Complete Mini EDTA-free proteinase inhibitors (Roche). Cell lysates were separated by SDS-PAGE and subsequently transferred to nitrocellulose membranes. Western blots were probed using primary antibodies including anti-FEM1B (sc-67568, Santa Cruz), anti-CASZ1 (sc-135453, Santa-Cruz), anti-DICE1 (sc-376524, Santa Cruz), anti-OCT1 (sc-232, Santa Cruz), anti-ARID2 (sc-166117, Santa Cruz), anti-CREBBP (sc-369, Santa Cruz), anti-SH2B3 (sc-393709, Santa Cruz), anti-PAK2 (sc-1872, Santa Cruz), anti-PPP3R1 (sc-6119, Santa Cruz), anti-TP53INP1 (sc-689919, Santa Cruz), anti-beta-Actin (sc-47778, Santa Cruz), and anti-PARP (9542p, Cell Signaling). The secondary antibodies used included anti-Goat IgG (sc-2020, Santa Cruz), anti-Mouse IgG (A9044, Sigma) and anti-Rabbit IgG (A6145, Sigma). All images were obtained using G:BOX (Syngene) and GeneSys (Syngene) acquisition software, and were subsequently analyzed by Genetools software (Syngene).
10 ng of a pLSG-based 3’UTR reporter, 10 ng pLenti-SV40-Rluc, along with either 500 ng of a miR-BART expression vector or a matched negative control, were co-transfected into 293T cells in 24-well plates using polyethylenimine (PEI). Cells were lysed ~72 h post-transfection with passive lysis buffer (Promega) and FLuc and RLuc expression analyzed using a dual luciferase assay kit (Promega). All 3’UTR reporter assays were performed on three separate occasions using technical triplicates.
The small RNA deep sequencing library for C666 cells was generated as previously described [11]. Briefly, C666 total RNA was first harvested using TRIzol (Ambion), and the small RNA fraction (~18 to ~24 nt) was subsequently isolated using 15% TBE-Urea polyacrylamide gels (Bio-Rad). The harvested RNAs were then ligated to 3’ and 5’ Illumina adapters, reverse transcribed using SSIII (Invitrogen) and subjected to Illumina deep sequencing.
The PAR-CLIP library for C666 was generated as previously described [11, 37]. Briefly, C666 cells were first expanded to 30 150-mm dishes at ~80% confluency, and were then cultured in the presence of 100 μM 4-thiouridine (4SU) for ~20 h. The cells were then UV radiated at 365 nm for 1 minute, harvested and lysed on ice in NP40 lysis buffer. Cross-linked Ago:RNA complexes were then immunoprecipitated using a pan-Ago antibody (ab57113; Abcam) and protein G Dynabeads (Invitrogen). Ago-bound RNAs were digested with RNaseT1, radio-labeled, gel purified, proteinase K treated, phenol-chloroform extracted, ethanol precipitated and ligated to 3’ and 5’ Illumina adapters. After reverse transcription and limited PCR amplification, the recovered cDNAs were deep-sequenced.
The C666-derived small RNA deep sequencing library and PAR-CLIP library were processed as previously described [11, 37]. Briefly, sequencing reads were pre-processed using the FAST-X toolkit (http://hannonlab.cshl.edu/fastx_toolkit/), and were aligned to the human genome (hg19) and EBV1 wild type genome using Bowtie with up to two (three for PAR-CLIP) mismatches allowed.
The PAR-CLIP library was further processed using the PARalyzer program, as previously described [11, 37, 38]. Briefly, reads were first filtered allowing for up to three mismatches but with only one or zero non-T-to-C mutations. Subsequently, reads that aligned to a unique genomic location, that contained at least one T-to-C mutation and that overlapped by at least one nucleotide were grouped together as clusters. Clusters with a read depth of at least five reads were presented as miRNA:mRNA interaction sites in the PAR-CLIP dataset. Each cluster in the PAR-CLIP dataset was further examined for canonical miRNA seed match sites, using the miRNA expression data generated from the small RNA deep sequencing library derived in parallel, and miRNAs with seed matches equal to or greater than 7mer1A (perfect base pairing to seed nt 2–7 with an A across from nt 1 of the miRNA, see [5]) to the cluster were identified as candidate miRNAs putatively responsible for the cluster. The raw sequencing data from the C666 small RNA deep sequencing and PAR-CLIP analysis have been submitted to the NCBI Sequence Read Archive (SRA), and both dataset can be accessed with the accession number GSE67990. The sub-accession numbers of the individual C666 small RNA deep sequencing and PAR-CLIP libraries are GSM1660655 and GSM1660656, respectively.
|
10.1371/journal.ppat.1005175 | Ganglioside and Non-ganglioside Mediated Host Responses to the Mouse Polyomavirus | Gangliosides serve as receptors for internalization and infection by members of the polyomavirus family. Specificity is determined by recognition of carbohydrate moieties on the ganglioside by the major viral capsid protein VP1. For the mouse polyomavirus (MuPyV), gangliosides with terminal sialic acids in specific linkages are essential. Although many biochemical and cell culture experiments have implicated gangliosides as MuPyV receptions, the role of gangliosides in the MuPyV-infected mouse has not been investigated. Here we report results of studies using ganglioside-deficient mice and derived cell lines. Knockout mice lacking complex gangliosides were completely resistant to the cytolytic and pathogenic effects of the virus. Embryo fibroblasts from these mice were likewise resistant to infection, and supplementation with specific gangliosides restored infectibility. Although lacking receptors for viral infection, cells from ganglioside-deficient mice retained the ability to respond to the virus. Ganglioside-deficient fibroblasts responded rapidly to virus exposure with a transient induction of c-fos as an early manifestation of a mitogenic response. Additionally, splenocytes from ganglioside-deficient mice responded to MuPyV by secretion of IL-12, previously recognized as a key mediator of the innate immune response. Thus, while gangliosides are essential for infection in the animal, gangliosides are not required for mitogenic responses and innate immune responses to the virus.
| Biological and structural studies have combined to give a detailed understanding of how the mouse polyomavirus binds to sialyloligosaccharides, how polymorphisms in the sialic acid binding pocket of the major virus capsid protein constitute important determinants of pathogenicity, and how gangliosides function as receptors for cell entry and infection by the virus. We used mice with knockouts in defined ganglioside biosynthetic pathways to determine whether gangliosides alone suffice to mediate lethal infection in the intact host and whether non-gangliosides are also recognized by the virus and utilized for important physiological responses. We confirmed the requirement of specific gangliosides for infection and determined that not all gangliosides that bind in vitro serve as receptors in vivo. Results also revealed two physiologically important responses that do not require MuPyV-ganglioside interactions: i) rapid induction of c-fos in fibroblasts as an early step in cell cycle progression on which the virus depends for its own replication, and ii). activation of cytokine secretion by antigen presenting cells as a critical innate immune response to the virus. We infer that these responses are mediated by non-ganglioside receptors bearing sialic acid. These results serve to illustrate the multiplicity of MuPyV receptors and the complexity of virus-cell surface interactions.
| The Polyomaviridae comprise an expanding family of viruses of human, non-human primate and rodent origin as well as several avian species [1]. These small non-enveloped icosahedral DNA viruses are similar in their structural and genetic organization. Studies in cell culture with several members of the group have demonstrated that gangliosides serve as necessary receptors for infection. Initial studies showed mouse polyomavirus (MuPyV) binding to specific gangliosides in the plasma membrane leads to internalization and transport via endolysosomes to the endoplasmic reticulum [2]. There the virus is thought to undergo partial disassembly followed by translocation to the cytosol and nuclear entry. Steps of virus disassembly leading to export from the endoplasmic reticulum are partially understood [3–8].
Gangliosides are sialic acid containing glycosphingolipids that are ubiquitously expressed. Gangliosides are anchored in the outer leaflet of the plasma membrane by a ceramide tail with their sialylated oligosaccharide portion (glycan) facing extracellularly. The level of gangliosides required for infection by MuPyV are controlled in part through regulation of a sialidase activity by tyrosine kinases of the Abl family [9]. Binding specificity among the polyomaviruses is based on recognition of the glycan by the major viral capsid protein VP1. High-resolution structural and biochemical studies have revealed details of how recognition of sialic acids in various linkages occur with different polyomaviruses [10–17]. MuPyV binds to oligosaccharides carrying terminal sialic acids in specific linkages found in several gangliosides. Studies with different strains of MuPyV have shown how differences in glycan recognition underlie biological properties. MuPyV has also been shown to bind to the α4β1 integrin. Mutagenesis of the integrin binding site on VP1 decreases infectivity by 50%, suggesting that α4β1 may serve as a ‘co-receptor’ mediating a post-attachment step of infection [18, 19].
The outcome of infection by MuPyV depends on the genetic background of both virus and host. Inbred strains of mice have been used to identify host determinants that underlie susceptibility or resistance to the virus [20]. Strains of MuPyV differing widely in pathogenicity owe their differences to polymorphisms in VP1 that allow the virus to discriminate among different oligosaccharides or that affect avidity of binding to sialic acid [10, 11, 21–25]. High-resolution structural studies of complexes between recombinant VP1s of several MuPyV strains and various glycans have extended and refined our understanding of receptor interactions [26]. Here we utilize mice with knockouts in ganglioside biosynthetic pathways to investigate the importance of specific gangliosides for infection and to determine whether gangliosides are essential for other host-responses such as mitogenic gene induction and innate immunity.
Previous studies have used ganglioside-deficient cell lines (i.e., rat glioma C6 cells, R- mouse cells) to evaluate the importance of ganglioside receptors for MuPyV infection [2, 9]. These cell lines are often from a heterologous-host for MuPyV, and are not genetically defined. Thus, we generated ganglioside-deficient mice with known ganglioside composition to clearly identify the role of specific gangliosides in MuPyV infection. The B4 KO mouse is blocked in a β1–4 GalNAc transferase (GM2/GD2 synthase) and is expected to lack the previously identified MuPyV ganglioside receptors, GD1a and GT1b, while maintaining expression of GM3 and GD3 (Fig 1A). We validated the ganglioside composition in B4 KO mice by analyzing total acidic lipid fractions from kidneys, a major site of replication and tissue destruction by MuPyV. High performance thin layer chromatography of kidney lipid fractions from uninfected wild type and B4 KO mice confirmed that only GD3 and its precursor GM3 are made in B4 KO mice (Fig 1B, lane 3). B4 heterozygous (+/-) mice showed decreased levels of gangliosides compared to wild-type mice (Fig 1B, lane 4). Immunofluorescence staining also showed that B4 KO mice lack a-series gangliosides such as GD1a (Fig 1C).
GD3 has been shown to bind pentamers of MuPyV VP1 in an in vitro screen, but its function as a receptor has not been evaluated. To determine the possible role of GD3 in MuPyV infection, St8 mice lacking -2,8 sialyltransferase (GD3 synthase) were generated. St8 mice cannot synthesize b-series gangliosides, including GD3 and its derivatives (Fig 1A), but retain a-series gangliosides, such as GD1a as verified by immunofluorescence staining with a GD1a antibody (Fig 1C). Thus, the B4St8 double KO mouse is expected to synthesize only GM3, which was previously shown to be unable to bind or mediate infection by MuPyV [2]. Protein glycosylation pathways are expected to be unaltered in these ganglioside-deficient mice.
The LID strain of MuPyV induces a rapidly lethal infection of newborn mice of most backgrounds [23, 27]. Mice typically succumb within a few weeks due to a widely disseminated infection with extensive destruction of the kidney and other vital tissues. LID owes its virulence to an amino acid substitution in the major capsid protein VP1 that reduces hydrophobic interactions with the sialic acid ring. This lower avidity binding of virus to cells facilitates release from cell debris and promotes virus spread. To establish the importance of gangliosides in mediating this infection, newborn mice from a cross of heterozygous B4 KO mice were inoculated with LID. Mice were followed daily and death was used as an endpoint (Fig 2A). Genotyping was carried out retrospectively, i.e., at time of death or at the end of the experiment.
Wild-type mice (B4 +/+) all succumbed within 14 days, as expected. Homozygous knockout mice (B4 -/-) all survived and showed no overt signs of illness at 35 days post infection when the experiment was terminated. Heterozygous mice (B4 +/-) also succumbed, though with a delay compared to wild-type mice. These mice express slightly decreased levels of gangliosides compared to wild-type mice (Fig 1B). A single copy of the GM2/GD2 synthase gene targeted in the B4 KO mouse thus sufficed to confer susceptibility. The extended survival of B4 heterozygotes is consistent with a gene dosage effect, whereby overall levels of enzyme activity (i.e., ganglioside synthesis) correlate inversely with mean survival time. Results with homozygous St8 KO mice were also consistent with this view. These mice (St8 -/-) all succumbed, but like B4 +/-, survived longer than wild-type mice (Fig 2A). The St8 mice retain GD1a and other a-series gangliosides (Fig 1A and 1C) indicating that these receptors are sufficient for lethal LID infection in the absence of GT1b and other b-series gangliosides. Thus, MuPyV infection is delayed by either decreased ganglioside diversity or decreased abundance of complex gangliosides. These results establish the importance of complex gangliosides lacking in the B4 KO mouse for mediating infection, and confirm for the first time that specific gangliosides are required for virus infection in vivo. These results establish the importance of complex gangliosides lacking in the B4 KO mouse for mediating infection by the LID strain of MuPyV. They confirm for the first time that specific gangliosides are required for virus infection in vivo.
Mouse embryo fibroblast cultures (MEFs) were established from wild-type and ganglioside-deficient mice. These cells were used to access the degree of resistance and the roles of specific gangliosides in mediating infection. MEFs were first infected with the standard laboratory small plaque strain RA at a multiplicity of infection (MOI) of 1–2 plaque forming units (PFU) per cell. Cells were fixed at 24 hrs post-infection and stained with anti-T (tumor) antigen antibody to determine the number of infected cells expressing the nuclear large T protein (Fig 2B). B4 KO MEFs were resistant while St8 KO MEFs were susceptible to infection. The B4St8 double KO cells, like those from B4, were resistant. Two independently derived lines of MEFs of each genotype were tested and showed the same results. B4 KO and B4St8 KO MEFs produced normal yields of virus following transfection with MuPyV DNA or supplementation with gangliosides, confirming that their resistance is due to a block in cell entry prior to uncoating of the virus and not to a block in the virus replication cycle per se. The resistance of B4 and B4St8 KO MEFs extended equally to small plaque (RA) and large plaque (PTA) strains (Table 1). When the MOI was increased from 1–2 PFU/cell to 5–10 PFU/cell, a few percent of the B4 KO MEFs were infected. In contrast, B4St8 KO MEFs remained completely resistant at the higher MOI (see further below).
B4 KO mice express only GD3 and GM3 and are not susceptible to lethal LID infection (Fig 2A). GM3 was previously shown to be unable to bind virus based on a flotation assay using ganglioside-supplemented liposomes [2]. However, GD3 has been identified in a glycan array screen as a strong binder of recombinant VP1 pentamers. A co-crystal structure of recombinant VP1 pentamers with the GD3 glycan has been determined [26]. GD3 also emerges as a possible receptor based on the observation that at high MOI a small percent of B4 KO, but not B4St8 double KO MEFs, become infected (Table 1). B4 KO MEFs synthesize GD3 while B4St8 MEFs do not (Fig 1A). Supplementation experiments were carried out to directly test whether GD3 is a functional receptor. B4St8 KO MEFs were incubated for 2 hrs in media with increasing concentrations of GD3, then washed and infected with RA at a MOI of 5 to10 PFU/cell. Resistance was overcome, but only a few percent of cells were capable of being infected even at the highest concentrations of GD3 tested (Table 2). Thus, under conditions of supplementation with high concentrations of GD3 and a high virus input, some virus particles apparently engage enough of the ganglioside to allow cell entry and infection. Although these conditions may exist in vivo, they do not lead to lethal infection in B4 KO mice (Fig 2A). GD3 accumulates to high levels in kidneys of B4 KO mice (see Fig 1B, lane 3). Despite the high endogenous levels of GD3, B4 KO mice are resistant to infection (Fig 2A). We conclude that GD3 does not serve as an efficient functional receptor in vivo despite its ability to bind the viral capsid subunits in vitro.
GD1a and GT1b have previously been shown to confer susceptibility to MuPyV infection when added to ganglioside-deficient rat and mouse cell lines [2, 28, 29]. We sought to extend these results using our B4St8 KO cells, which are genetically defined and have known ganglioside composition. GD1a was used as a positive control, and GM1, the SV40 receptor, as a negative control to confirm previous results. We then tested the ability of additional gangliosides, GT1a and GD1b, to confer susceptibility using the RA strain of virus. GT1a had been suggested as a possible receptor based on co-crystallization with MuPyV VP1 [26]. Cells were pre-incubated with 0.5 to 2.0 μM gangliosides in serum-free medium for 16 hrs, then infected and scored for T-antigen expression 24 hrs post-infection. GT1a conferred infectibility slightly more efficiently than GD1a, a result consistent with in vitro affinity studies [26], and GD1b conferred low levels of infectibility, much less efficiently than GT1a or GD1a (Fig 3A).
The VP1 binding pocket of MuPyV is thought to accommodate both glycolipid (ganglioside) and glycoprotein binding [10,11]. Thus, we investigated the dynamics and levels of cell surface binding of virus in the presence or absence of ganglioside receptors. Wild-type, B4St8 KO, and GD1a-supplemented B4St8 KO MEFs were infected (RA MuPyV, 10 PFU/cell) at 4°C followed by fixation and staining for cell surface VP1 using a VP1 antibody. Virus binding to wild-type MEFs at 4°C was time dependent with two cell populations at 1 hr post virus addition: a cell population with low virus accumulation (<102 VP1 staining) and a cell population with high virus accumulation (>102 VP1 staining) (Fig 3B). B4St8 KO MEFs were also bound by virus in a time dependent manner; however, these cells displayed only low virus accumulation after 1 hr at 4°C (<102 VP1 staining) (Fig 3B). Supplementation of B4St8 KO MEFs with 5μM GD1a prior to virus addition restored virus accumulation on the cell surface after 1 hr at 4°C (>102 VP1 staining) (Fig 3B). These data indicate that although virus binds cells in the absence of ganglioside receptors, the dynamics of binding are changed. Gangliosides result in high levels of virus accumulation on the cell surface, whereas alternative interactions result in lower levels of overall virus binding.
Using the B4St8 ganglioside KO MEFs we determined whether gangliosides are required for virus entry. Wild-type MEFs and B4St8 KO MEFs were infected with MuPyV (RA, 50 PFU/cell) and then fixed at the indicated times post-infection (30 min, 3 hrs). At 30 mins post-infection in wild-type MEFs line scan analysis showed that MEFs exhibit similar staining for cell surface (shown in red) and total VP1 (shown in green), indicating minimal virus internalization at this early time (Fig 4B). Similar results were seen in B4St8 KO MEFs at 30 min post infection (Fig 5B). At 3 hrs post-infection in wild-type MEFs line scan analysis showed that MEFs had abundant intracellular VP1 staining (green only), indicating a large fraction of internalized virus (Fig 4C). B4St8 ganglioside KO-MEFs also displayed high levels of internalized virus as shown by line scan analysis (green only) (Fig 5C). These data demonstrate that gangliosides are not required for virus entry into MEFs, and virus can enter cells through non-ganglioside mediated pathways. Given the complete resistance of these B4St8 KO cells to MuPyV infection (Table 1), it can be assumed that virus uptake via these alternative routes proceeds along a non-infectious or ‘dead end’ pathway. While the presence of glycoproteins such as α4β1 integrin have been observed to enhance infection [18, 19], gangliosides are required for virus uptake along infectious pathways.
An earlier study demonstrated the ability of MuPyV to induce expression of c-fos and other ‘early response’ genes in established mouse fibroblasts [30]. This response, measured by mRNA synthesis, is biphasic. A rapid but transient response occurs within the first hour followed by a second sustained wave of expression beginning around 12 hrs post-infection. The second wave requires early viral gene expression, while the initial transient phase can be induced by empty capsids or recombinant VP1. To determine if the induction of an immediate early response depends on recognition of gangliosides or other virus receptors, wild-type and B4St8 KO MEFs were exposed to purified virus in serum-free medium (Fig 6). Extracts were analyzed for c-fos protein by western blot. The double KO MEFs responded indistinguishably from wild-type MEFs with a clear induction at 1 hr post-infection. At 6 hrs post-infection, the response was diminished but showed the slower migrating form(s) of the c-fos protein indicative of phosphorylation which is known to accompany activation of mitogen receptors (Fig 6) [31]. We conclude that the rapid induction of c-fos in ganglioside-deficient cells results from virus binding to non-ganglioside receptor(s).
Antigen-presenting cells from mice that are naturally resistant to tumor induction by MuPyV respond to the virus by secretion of IL-12, a type 1 cytokine [32, 33]. This IL-12 response does not require infection and can be elicited by exposure to virus-like particles assembled from recombinant VP1. Pretreatment of splenocytes with neuraminidase prevents the IL-12 response, consistent with roles of gangliosides or glycoprotein, or both, as receptors mediating the innate immune response [33].
TLR4, a known sialoglycoprotein, has previously been implicated in mediating cytokine responses to MuPyV. Quantitave Trait Locus analysis of an F2 cross between susceptible and resistant mice pinpointed a region of chromosome 4 encompassing TLR4 as the determinant of the cytokine response. Transfection of macrophages from a TLR2/TLR4 double KO mouse with TLR4 cDNA from a resistant strain conferred an IL-12 response. While pointing clearly to a role of TLR4, these results do not rule out the possibility that gangliosides may play an essential or supporting role in mediating the IL-12 response. We therefore investigated the immune response in the absence of ganglioside receptors.
B4 KO and St8 KO mice were derived on a background of the resistant C57BL/6 strain that responds to MuPyV with IL-12. To determine if this innate immune response is retained in ganglioside-deficient mice, splenocytes were harvested from naive wild-type, B4 KO and St8 KO mice. Cells were exposed to virus for 1 hr and incubated further for 40 hrs. Culture supernatants were then assayed for IL-12 by ELISA. Splenocytes from all three strains responded to the virus by secretion of varying levels of IL-12 indicating that gangliosides are not required for the IL-12 response (Table 3). Interestingly, splenocytes from B4 KO mice showed the strongest response, roughly 2-fold greater than the wild-type, indicating that gangliosides may dampen the IL-12 response possibly through competition with TLR4 for virus binding. Pre-treatment of cells with neuraminidase from Vibrio cholera which cleaves sialic acids from both glycolipids and glycoproteins completely inhibited the response, as expected. Importantly, pretreatment with PNGase F, an endoglycosidase that cleaves oligosaccharide chains from N-linked glycoproteins, also inhibited the response. These results demonstrate that gangliosides are not required for the innate immune response to MuPyV, and ganglioside-deficient cells have a heightened IL-12 response upon virus challenge compared to wild-type and St8 (GD1a and GT1a containing) cells.
Five million spleen cells were pretreated with either Neuraminidase (Neurase) (400 units/mL) or PNGase (10,000 units/mL). Cells were exposed to MuPyV and cultured for 40 hr. Supernatants were tested for IL-12 by ELISA (pg/ml). Cells stimulated with medium alone gave <25 pg/ml. Shown are the mean ± SD of four 4–5 week old mice for each determination.
Ganglioside supplementation experiments in ganglioside-deficient rat and mouse cell lines have identified specific gangliosides (GD1a and GT1b) as MuPyV receptors; however, these findings have not previously been validated in vivo. Additionally, the role of gangliosides in other host-responses to MuPyV infection such as mitogenic gene induction and innate immunity has not been investigated. We generated mouse strains that are deficient (St8 KO) and null (B4 KO) for complex gangliosides to further characterize the specificity of ganglioside-mediated host responses to MuPyV in vivo. The St8 KO mice are deficient in GD3 synthase and do not synthesize b-series gangliosides (GD1b and GT1b) but retain synthesis of the a-series gangliosides (GD1a and GT1a) (Fig 1A and 1C). St8 KO mice succumb to LID infection, indicating that a-series gangliosides are sufficient to mediate a lethal virus infection. (Fig 2A). The B4 KO mouse lacks GM2/GD2 synthase, which is required for both a-series and b-series ganglioside synthesis (Fig 1A and 1C). Our finding that newborn B4 KO mice are completely resistant to infection by the normally lethal LID strain of MuPyV provides clear evidence that gangliosides are required for infection and spread of the virus in the animal. The LID strain was chosen to evaluate the role of gangliosides in the animal because of its rapid effects ending in death as a discrete endpoint. As the KO strains derive from tumor-resistant C57BL/6 mice, the PTA and RA strains cannot be evaluated for their ganglioside dependence in a tumorigenic setting using these mice. However, based on the fact that these strains of MuPyV, like LID, are unable to infect B4St8 KO MEFs, it is expected that they would be unable to infect ganglioside-deficient mice. Results using LID have shown for the first time that the GM2/GD2 synthase pathway is necessary and sufficient for MuPyV infection in vivo. This finding in the natural host is an important in vivo validation of earlier biochemical and cell culture results [2, 28, 29]. Additionally, these mice provide an excellent model to test the role of gangliosides in other host responses to MuPyV infection.
Early biochemical and in vitro experiments were extremely valuable in first identifying gangliosides as possible receptors for MuPyV based on ganglioside binding to VP1-pentamers [2, 28, 29]. However, results presented here have shown that distinctions must be made between gangliosides that bind in vitro and those that operate as functional receptors in vivo. GD3 was identified in a screen of a glycan array as an effective binder of VP1-pentamers. The GD3 glycan also binds recombinant VP1 pentamers in crystallographic studies [26]. This ganglioside however does not confer susceptibility to infection under normal conditions (B4 KO, Fig 2A). The observation that B4 KO mice have high levels of GD3 in the kidney (Fig 1B), yet are resistant to infection, is convincing evidence that this ganglioside is unable to mediate infection in vivo despite binding to VP1 in vitro. Discrepancies between results of biochemical binding and in vivo infection are important to recognize for a full evaluation of receptors from a functional standpoint.
Previous experiments have shown that GD1a or GT1b are sufficient for infection of ganglioside-deficient rat glioma cells [2, 28]. We identified additional gangliosides, the a-series ganglioside GT1a and the b-series ganglioside GD1b, as receptors for RA MuPyV. When fibroblasts from B4St8 ganglioside-deficient mice were pre-incubated with gangliosides GT1a or GD1b their infectibility was restored (Fig 5). We also confirmed that GD1a as a receptor for MuPyV in the ganglioside-deficient B4St8 MEFs. Because B4St8 MEFs lack all complex gangliosides, these data show that supplementation with single gangliosides is sufficient for MuPyV infection.
Structural studies of the MuPyV capsid protein VP1 have revealed that VP1 binds sialic acid within a pre-formed sialic acid binding pocket on the virus surface [10,11]. These results suggest that MuPyV could potentially bind sialyated oligosaccharides on either glycoproteins or glycolipids (i.e., gangliosides) through interactions with sialic acid. Thus, we evaluated MuPyV cell surface binding in the presence or absence of ganglioside receptors. We found that while virus binds to the cell surface of B4St8 KO MEFs, it accumulates to lower levels than in wild-type MEFs or GD1a-supplemented B4St8 MEFs. These results suggest that virus binding to gangliosides is required for high levels of virus accumulation on the cell surface, although the presence of glycoprotein or other interactions on ganglioside-deficient cells allows for some virus binding to the cell surface. Cell surface binding of virus is not necessarily indicative of virus entry. Therefore we determined if virus enters cells in the absence of ganglioside receptors. We observed that MuPyV is internalized in the absence of gangliosides. Importantly, uptake of virus under these conditions does not lead to infection. Previous studies have shown that proteinase treatment of cells prior to MuPyV addition leads to slight increases in infection, suggesting that MuPyV-glycoprotein interactions inhibition MuPyV infection [34]. This inhibition may be due to altered trafficking of MuPyV, whereby ganglioside receptors mediate transport of MuPyV to the ER along an infectious pathway while glycoproteins act as “decoy receptors” and lead to MuPyV degradation [34]. Our results are consistent with the idea that MuPyV-glycoprotein interactions lead to non-infectious pathways of entry. It is possible that glycoprotein interactions mediate other host responses to virus infection, such as mitogenic signaling, or the immune response. Alterations of these responses are not readily apparent in infection-based cell culture experiments.
In order to understand how host responses change when ganglioside interactions are lost we evaluated two known cellular responses to MuPyV infection in the B4St8 KO cells: a) activation of mitogenic gene induction, and b) activation of innate immune responses. Fibroblasts from B4St8 KO MEFs responded to MuPyV binding with a rapid transient induction of c-fos that is indistinguishable from the response of wild-type MEFs. Thus gangliosides are not required for mitogenic gene induction by MuPyV and this response may be triggered by virus binding to any of a number of sialoglycoproteins (or other molecules) that normally serve as mitogen receptors in the plasma membrane. These include receptors for platelet-derived growth factor, epidermal growth factor, insulin-like growth factor, and others, whose activation leads to cell cycle progression and entry of cells into S phase essential for facilitating the initial replication of viral DNA [30].
It has previously been shown that antigen-presenting cells from wild-type mice that mount effective adaptive anti-tumor responses respond to virus challenge at the innate level by secretion of the type 1 cytokine IL-12. TLR-4 is required for the IL-12 response in these mice and mice lacking TLR-4 are suspectible to MuPyV tumor induction, likely due to a loss of IL-12 secretion [34]. MuPyV is thought to bind to TLR-4 through sialic acid containing oligosacharride regions on the extracellular domains of TLR-4. To confirm that sialic acid interactions are required for IL-12 induction we treated splenocytes with neuraminidase prior to challenge with virus. As expected, pretreatment with neuraminidase blocks the cytokine response to MuPyV in both wild-type, St8 KO, and B4 KO cells; however, neuraminidase treatment also abrogrates MuPyV-ganglioside interactions and thus is not informative about the role of gangliosides in IL-12 induction. We sought to abolish MuPyV-TLR4 interactions while retaining ganglioside interactions, by treating with PNGase, which removes N-linked carbohydrate chains from glycoproteins, prior to virus challenge. We found inhibition of the IL-12 response by pretreatment with PNGase providing further support for TLR4 as a required receptor in the innate immune response to MuPyV and confirming that MuPyV-ganglioside interactions are not sufficient for IL-12 induction. Lastly, we wanted to determine whether gangliosides contribute positively or negatively to the innate immune response induced by virus binding. Splenocytes from B4 KO mice display an increased IL-12 response compared to wild-type or St8 KO splenocytes. These data indicate that gangliosides may dampen the cytokine response, possibly by competing for virus binding with the TLR4 glycoprotein receptor. This observation suggests that glycolipids and glycoprotein receptors act in an opposing manner in multiple ways, even at the level of the innate immune response.
Gangliosides contribute to a diverse array of physiological responses involved in viral infection. Results of experiments in ganglioside-deficient mice show that while gangliosides are essential as receptors for MuPyV infection, they are not essential for cell surface binding, cell entry, or for activating the early mitogenic and innate immune responses of the host. Additionally, the antiviral immune response was heightened in ganglioside-deficient splenocytes, indicating that gangliosides somehow serve to dampen the antiviral cytokine response. These data establish that multiple types of receptors bearing sialic acid are utilized by the virus to mediate different aspects of virus-host interaction. These results could have implications for tissue tropism and immune response generated in vivo by other Polyomaviruses.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocols were approved by the Harvard Medical Area Standing Committee on Animals (approval numbers 298 and 781). All mouse strains were maintained in specific pathogen-free conditions in the animal facilities of the Harvard Medical School. All efforts were made to minimize suffering and provide humane treatment to the animals included in the study.
Two knockout strains of mice were obtained through the Mutant Mouse Regional Resource Center, Missouri/Harlan Consortium, by cryo-resuscitation and embryo transfer. Strain B6;129S-B4galnt1tm1Rlp/Mmmh [Stock #: MMRRC:000036-MU] is a knockout in the B4galnact1 (beta-1,4-N-acetyl-galactosaminyl transferase 1) gene. It is referred to here as the B4 KO. B4 -/- KO mice are male sterile; the colony was maintained by crossing heterozygous B4 +/- males with B4 -/- or B4 +/- females. Strain B6;129S-St8sia1tm1Rlp/Mmmh [Stock #: MMRRC:000037-MU] is a knockout in the St8sia1 (ST8 alpha-N-acetyl-neuraminide alpha-2,8-sialyltransferase 1) gene. It is referred to here as the St8 KO. A double knockout mouse, B4St8 KO, was generated by crossing B4 KO and St8 KO mice. Genotyping was carried out on tail DNA by PCR using the suppliers protocols.
Crude gangliosides were extracted from kidneys of wild-type and B4 KO mice by homogenization in 20 mL solvent B with a Polytron homogenizer. The solution was centrifuged, the supernatant transferred and the homogenization repeated two times with 20 mL solvent A and once more with 20 mL solvent B. The four extracts were pooled and dried on a rotary evaporator. Gangliosides were purified from the crude fraction by DEAE-Sephadex Ion Exchange chromatography. The crude fraction was resuspended in a minimal amount of solvent C and applied to a column of DEAE-Sephadex A-25 (acetate form). Neutral species were eluted with 5 volumes of solvent C. Acidic species were eluted with 5 volumes of 0.5 M sodium acetate in methanol. The acidic fraction was dried, dialyzed exhaustively using a Pierce Slide-a-lyzer cartridge (MWCO = 3500) in deionized water and redried. Neutral and acidic fractions were dried and taken up in Solvent A for HPTLC analysis.
Analytical high-performance thin-layer chromatography (HPTLC) was performed on silica gel 60 plates (E. Merck, Darmstadt, Germany) using solvent D as the mobile phase. Lipid samples were dissolved in solvent B and applied by streaking from 5 ml Micro-caps (Drummond, Broomall, PA). Detection was with Bial’s orcinol reagent [0.55% (w/v) orcinol and 5.5% (v/v) H2SO4 in ethanol-water (9:1, v/v); the plate was sprayed and heated briefly to ~200–250°C.
The PTA and RA strains have been described (CD—AJP, 1987; RF, GMDubensky 1991; Freund, AC 1991; CD,RF 1987). PTA is a standard large plaque ‘high tumor’ strain that binds straight chain oligosaccharides with terminal α 2,3-linked sialic acid. RA is a standard small plaque ‘low tumor’ strain that binds branched as well as straight chain sialic acids [23, 24, 35]. LID is a virulent strain derived from PTA [23, 27]. All strains were propagated in primary baby mouse kidney cells.
Newborn WT, B4 (-/-), B4 (+/-) and St8 (-/-) mice were inoculated intraperitoneally within 24 hrs of birth with 1–2 x 106 plaque forming units of the LID strain of virus. Mice were followed on a daily basis and results recorded based on death as an endpoint. Mice were genotyped retrospectively, i.e., after death or at the termination of the experiment at 35 days.
Fibroblast cultures (MEFs) were prepared from embryos of 18 to 19 days gestation and genotyped to generate B4 KO, St8 KO and B4St8 KO mouse embryo fibroblast lines (MEFs). MEFs were maintained by serial passage in Dulbecco’s Modified Eagle’s medium with 10% fetal bovine serum and used for viral infections at passages between 2 and 5. Cells on coverslips were infected by MuPyV strains RA or PTA at various multiplicities of infection and fixed at 24 hrs post-infection with 4% neutral buffered paraformaldehyde (Electron Microscopy Sciences, Ft Washington, PA). Cells were permeabilized with 0.3% Triton X-100 in PBS and stained with rat polyclonal anti-T antibody (Goldman & Benjamin 1975) and rhodamine-conjugated donkey anti-rat IgG.
WT and B4St8 MEFs were seeded onto glass coverslips in Dulbecco's Modified Eagle's Medium supplemented with 10% fetal bovine serum (FBS). Cells were incubated overnight in serum free media prior to infection. Cells were then infected with RA MuPyV (MOI 50). At indicated times post infection (30 min, 3 hrs) cells were washed in phosphate buffered saline (PBS) and fixed with 4% paraformaldehyde (PFA) at room temperature (RT) for 10 mins. Cells were blocked in 10% FBS in PBS overnight at 4°C followed by staining for cell surface VP1 (I58 antibody/Alexa Flour secondary 546). Cells were then re-fixed with 4% PFA at RT for 10 mins followed by permeabilization with 0.5% Triton X-100 for 15 mins at RT. Cells were blocked in 10% FBS in PBS overnight at 4°C followed by staining for total VP1 (I58 antibody/Alexa Flour secondary 488). Confocal images were taken as a 5 step (.125 μm step size) z-stack and slices were taken through the center of the cells. Each z-stack was aligned and compressed into a max intensity Z projection image for quantification of cell surface and total VP1 staining. Using the Nikon software, line scans were taken sampling the cell surface and cytoplasm of each cell to measure both cell surface and internalized virus as indicated by VP1 staining.
Cells were seeded onto glass coverslips in Dulbecco's Modified Eagle's Medium supplemented with 10% fetal bovine serum (FBS). Cells were incubated overnight in serum free media prior to infection. For ganglioside-supplemented cells, serum free media containing the indicated concentration of gangliosides was used. Cells were then infected with RA (MOI 5 to 10). At indicated times post infection cells were washed in phosphate buffered saline and fixed with 4% paraformaldehyde at room temperature (RT). Cells were blocked in 10% FBS and then stained for GD1a using the MAB5606 (Millipore). Samples were then incubated with Alexa Fluor labeled secondary antibodies. Cells were imaged on a Nikon A1R confocal microscope.
Confocal images were taken as a 9 to 13 step (.25 μm) z-stack. Each z-stack was aligned and compressed into a max intensity Z projection image for quantification of T-antigen staining. To quantify infection, T-ag staining was measured per each DAPI labeled nuclei. The DAPI channel on each image was thresholded and nuclei were counted using ImageJ (Analyze Particles). These particles were marked as “Regions of Interest” (ROI) and then the average pixel intensity of T-ag staining was measured for each nuclei (ROI). These were then binned into T-ag positive or T-ag negative nuclei to determine % infected.
Cells were dissociated from the plate with Versene solution (EDTA) for 10 mins at room temperature (RT). Resuspended cells were then washed in cold PBS followed by incubation with MuPyV (RA, MOI 10) on ice. Samples were removed at indicated time points and temperatures, washed with cold PBS, and fixed with 0.5% paraformaldehyde (RT for 5 mins), followed by staining for VP1 (I58). Cells were not permeabilized. Cell surface virus levels were measured for >10,000 cells per sample by flow cytometry using a CyAN ADP Analyzer.
Wild-type and B4St8 MEFs were plated at 3 x 105 cells per 35 mm plastic tissue culture dish in DMEM with 10% fetal bovine serum. Twenty-four hrs after plating, the plates were washed and the medium replaced by DMEM with 0.1% platelet-poor plasma for an additional 12 hrs prior to infection. Infection was carried out with purified virus in serum-free PBS buffer at a MOI of 20 to 40 PFU/cell. Virus was purified by cesium chloride density gradient centrifugation as described [36, 37]. Cells were incubated for 1, 2 and 6 hrs post-infection in serum-free DMEM. Total cell proteins were extracted, separated by SDS-PAGE and blotted for c-fos protein with antibody from Santa Cruz Biochemicals.
Assays were carried out as described [38]. Briefly, five million splenocytes were harvested from naïve 4–5 week old wild-type B6, St8 KO and B4 KO mice. 5 x 106 cells were exposed to MuPyV RA strain at an MOI of 2 to 5. Culture supernatants were harvested after 40 hrs and levels of IL-12 determined by ELISA. Determinations were carried out on four mice of each strain. Measurements were also made on cells pretreated with neuraminidase from V. cholera (400 Units/ml) or by endoglycosidase PNGase F (10,000 Units/ml) (Calbiochem) prior to infection.
|
10.1371/journal.ppat.1006745 | Association of Marek’s Disease induced immunosuppression with activation of a novel regulatory T cells in chickens | Marek’s Disease Virus (MDV) is an alphaherpesvirus that infects chickens, transforms CD4+ T cells and causes deadly lymphomas. In addition, MDV induces immunosuppression early during infection by inducing cell death of the infected lymphocytes, and potentially due to activation of regulatory T (Treg)-cells. Furthermore, immunosuppression also occurs during the transformation phase of the disease; however, it is still unknown how the disease can suppress immune response prior or after lymphoma formation. Here, we demonstrated that chicken TGF-beta+ Treg cells are found in different lymphoid tissues, with the highest levels found in the gut-associated lymphoid tissue (cecal tonsil: CT), fostering an immune-privileged microenvironment exerted by TGF-beta. Surprisingly, significantly higher frequencies of TGF-beta+ Treg cells are found in the spleens of MDV-susceptible chicken lines compared to the resistant line, suggesting an association between TGF-beta+ Treg cells and host susceptibility to lymphoma formation. Experimental infection with a virulent MDV elevated the levels of TGF-beta+ Treg cells in the lungs as early as 4 days post infection, and during the transformation phase of the disease in the spleens. In contrast to TGF-beta+ Treg cells, the levels of CD4+CD25+ T cells remained unchanged during the infection and transformation phase of the disease. Furthermore, our results demonstrate that the induction of TGF-beta+ Treg cells is associated with pathogenesis of the disease, as the vaccine strain of MDV did not induce TGF-beta+ Treg cells. Similar to human haematopoietic malignant cells, MDV-induced lymphoma cells expressed high levels of TGF-beta but very low levels of TGF-beta receptor I and II genes. The results confirm that COX-2/ PGE2 pathway is involved in immunosuppression induced by MDV-lymphoma cells. Taken together, our results revealed a novel TGF-beta+ Treg subset in chickens that is activated during MDV infection and tumour formation.
| Treg cells are crucial for the maintenance of tolerance and control of immune responses, especially during viral infection and tumour formation. Marek’s Disease Virus (MDV) infection causes immunosuppression and induces transformation of CD4+ T cells in chicken. Here we demonstrate that a population of chicken CD4+ T cells express inhibitory molecules including TGF-beta and have immune-regulatory properties. TGF-beta+ Treg cells are detected in different chicken lymphoid tissues, the highest being detected in cecal tonsils. Chicken lines susceptible to MDV-induced lymphoma formation have higher frequencies of TGF-beta+ Treg cells compared to the MDV resistant chicken line. Infection of chicken with a virulent MDV increased the numbers of TGF-beta+ Treg cells, which was not changed after infection with the MDV vaccine strain. MDV-transformed CD4+ T cells produce high levels of TGF-beta, while they express very low levels of TGF-beta receptors compared to non-transformed CD4+ T cells. In addition, MDV-induced lymphoma cells express soluble suppressive factors that can inhibit T cell function; however these soluble factors cannot suppress proliferation of the lymphoma cells. For the first time, we identified chicken TGF-beta+ Treg cells and demonstrate that these cells are involved in pathogenesis and immunosuppression of MDV infection.
| Regulatory T cells (Tregs) are critical for maintenance of immune-homeostasis and immunological tolerance by enforcing negative regulation on T helper (Th) cells. Transcription factor Foxp3 (Foxp3) is a lineage specific factor for human and murine CD4+CD25+ Treg cells and is crucial for Treg development and function. TGF-beta can bind to the surface of human Foxp3+ Treg cells via GARP (LRRC32) a membrane anchoring molecule, and these cells can be classified as activated Treg cells with a highly potent immune-regulatory properties [1–3].
In chickens, CD4+CD25+ T cells have been classified as Treg cells which are present in most tissues including thymus [4] thus, they are thought to be equivalent to mammalian natural regulatory T cells (nTreg cells). Interestingly, expression of Foxp3 is restricted to jawed vertebrate and no Foxp3-like genes has been identified in the chicken genome [5]. Therefore, CD25 is currently the only marker for identification of Treg but CD25 is also a marker for activated T cells [2]. Apart from immune regulatory activity, Tregs cells are implicated in progression of the tumour and pathogenesis of viral as well as bacterial diseases in humans and mice. Depletion of tumour-induced Treg cells can reduce tumour progression via the activation of T cell responses [6–9]. It has been postulated that Treg cells are involved in the development of malignant lymphomas, however, their role is more complex in lymphoma than that in other cancers such as carcinomas [10]. Transformed CD4+ T cells (lymphoma) may express inhibitory markers, and suppress anti-tumour immunity. Therefore, Treg cells isolated from malignant lymphoma patients can be categorized as primary Treg cells or malignant Treg cells [10].
Marek’s Disease (MD) is caused by a highly contagious alphaherpesvirus, Marek’s Disease Virus (MDV), leading to development of malignant CD4+ T cell lymphoma in domestic chickens. The pathogenesis of the MD can be classified into three distinct phases: i) the early cytolytic phase with the infection of B and T cells which is associated with transient immunosuppression, ii) the latency phase which is defined by absence of viral protein expression and viral replication and iii) the transformation phase that leads to a deadly lymphoma. In MD-susceptible chicken lines, a second cytolytic phase may occur, resulting in atrophy of lymphoid tissues and a severe immunosuppression [11]. During the early phase of infection, both T and B cells are the predominant target cells for MDV infection in the thymus, spleens and the bursa of Fabricius. MDV replicates in the infected B and T cells, resulting in a depletion of lymphocytes and transient immunosuppression in the host [12]. The activation of T cells via the antigen presenting B cells is important for the establishment of primary infection, as the resting T cells are resistant to the MDV infection [13].
Here we identified and characterized TGF-beta+ Treg cells in chickens and demonstrated that MD-susceptible chicken lines have higher levels of TGF-beta+ Treg cells. Infection with virulent virus, but not vaccine strains, increased the number of TGF-beta+ Treg cells. Interestingly, MDV-induced lymphoma cells adopt a Treg phenotype and release soluble molecules that can inhibit T cell function. Taken together, our results suggest that activated Treg cells may be involved in MDV-induced immunosuppression in chickens.
CD25 molecule is not only a marker for identification of human and murine Treg cells but also for activated T cells. We confirmed this observation in chickens by analysing the surface expression levels of CD25 on CD4+ T cells isolated from spleens (3 weeks old Rhode Island Red (RIR) chickens) upon activation with Con-A or PHA. As expected, addition of Con-A or PHA to splenocytes increased the expression of CD25 molecules on the activated CD4+ T cells 3 days after in vitro activation (Fig 1A). Another marker for identification of a subpopulation of human Treg cells is membrane bound TGF-beta. To examine whether chicken Treg cells also express membrane bound TGF-beta, mononuclear cells from different tissues (blood, spleen, lungs, cecal tonsil, and thymus) were isolated from RIR chickens (3 weeks old) and stained with anti-CD4 and anti-TGF-beta mAbs. Flow cytometry analysis of these cells revealed that a subpopulation of chicken CD4+ T cells express membrane bound TGF-beta (Fig 1B). The next step was to determine whether membrane bound TGF-beta are preferentially expressed on CD4+CD25+ T cells or CD4+CD25- T cells. CD4+ T cells were subdivided into three subpopulations; CD25-, CD25low and CD25high CD4+ T cells, and the expression of TGF-beta was analysed in these three subpopulations. The data revealed that TGF-beta is expressed on both CD4+CD25- and CD4+CD25+ T cells, however a higher percentages of CD4+CD25high T cells expressed TGF-beta compared to CD4+CD25- T cells (Fig 1C). Only 15–20% of TGF-beta+CD4+ T cells expressed CD25 molecules (Fig 1D), suggesting that TGF-beta+CD4+ T cells and CD4+CD25+ T cells are two distinct cell types with some overlap. Flow cytometry analysis of surface expression of TGF-beta on CD4+ cells show that TGF-beta+CD4+ T cells are present in various lymphoid tissues including spleen, cecal tonsils, lungs, and blood of chickens (Fig 1B, C and 1E). Intriguingly, unlike CD4+CD25+ T cells, TGF-beta+CD4+ T cells were not found in thymus of chickens (Fig 1E), indicating that TGF-beta+CD4+ T cells are induced in the periphery. In addition, confocal microscopy confirmed that a population of primary CD4+ T cells express TGF-beta as both an intracellular and membrane bound form (Fig 1F).
One of the immune-regulatory functions of human and murine Tregs is their ability to inhibit T cell proliferation in vitro. Hence, we utilized a typical inhibitory T cell proliferation assay to examine the inhibitory properties of chicken TGF-beta+CD4+ T cells. The cells were isolated from spleens of 3-weeks old RIR chickens and the CFSE-labelled responder cells were co-cultured with TGF-beta+CD4+ T cells or TGF-beta-CD4+ T cells and were stimulated with Concanavalin (Con)-A, and the proliferation of T cells were analysed 3 days after co-culture using flow cytometry. The result demonstrated that TGF-beta+CD4+ T cells reduced T cell proliferation in vitro (Fig 1G), suggesting that these cells exhibit immune-regulatory activity.
A defining characteristic of human CD4+ Treg cells is their defect in phosphorylation of AKT at S473 [14]. To determine the AKT phosphorylation levels in chicken TGF-beta+ Treg cells, mononuclear cells from 3-weeks old RIR chickens were isolated and were stimulated with PMA for 15 minutes. AKT phosphorylation at S473 were analysed within TGF-beta+ Treg cells and TGF-beta-CD4+ T cells using Phosflow. Fig 1H demonstrates that chicken TGF-beta+ Treg cells (R1 gate) have lower AKT activation at S473 compared to TGF-beta- CD4+ T cells (R2 gate), confirming that chicken TGF-beta+ Tregs have also a defect in phosphorylation of AKT.
As described above, we demonstrated that TGF-beta+ Treg cells are present in various tissues of chickens. However, the highest frequencies of these cells were found in cecal tonsils (CT) as detected using flow cytometry (Fig 1E). A representative of Forward and Sideward scatter of cells isolated from different tissues including spleens, blood, lungs, thymus and CT are shown (S1A Fig). Exclusion of dead cells (7AAD positive cells) (Fig 1E) and the light microscopy of the cells isolated from CT demonstrated that these cells are alive and intact (S1B Fig). Various subsets of Treg cells that promote immunological non-responsiveness have been suggested to be involved in sustaining immune privilege in the gastrointestinal tract under normal homeostatic conditions. Here we demonstrated that the frequencies of both TGF-beta+CD4+ cells and CD4+CD25+ T cells in the CT of RIR chickens (3 weeks old) were significantly higher than that observed in the spleens (p = 0.002, n = 8) (Fig 2A, 2B and 2C). We also examined the differential expression levels of inhibitory molecules including CTLA-4, IL-10, PDL1 and PD1 in CD4+ T cells from CT and spleens of RIR chickens using semi-quantitative RT-PCR. Chicken CD4+ T cells isolated from CT expressed significantly higher levels of inhibitory genes including CTLA-4, PD1 and IL-10 compared to the cells isolated from the spleens (Fig 2D). Another characteristic of Treg cells is their ability to inhibit the production of proinflammatory Th1 cytokines such as IFN-gamma. To analyse the differential IFN-gamma expression in CT and spleens, mononuclear cells isolated from spleens and CT of RIR chickens were stimulated with PMA and Ionomycin for 4hrs and the expression of IFN-gamma gene was determined using semi-quantitative RT-PCR. The results confirmed that the cells isolated from CT are unable to up-regulate the expression of IFN-gamma gene (Fig 2E). Similarly, we observed a significantly lower frequencies of IFN-gamma producing T cells in mononuclear cells isolated from CT, stimulated with PMA and Ionomycin for 18 hrs, compared to that from the spleens using chicken IFN-gamma ELISPOT assay (Fig 2F). Generally, Treg cells exhibit reduced proliferative response in the absence of rIL-2 in vitro. The unresponsiveness of mononuclear cells isolated from CT was demonstrated in a CFSE-based proliferation assay. In contrast to mononuclear cells from spleens, the cells isolated from CT did not proliferate in response to Con-A stimulation (Fig 2G and 2H). Fig 2I shows proliferation index of mononuclear cells from CT cells compared to splenocytes. The results show that CT cells are unresponsive and do not proliferate after the stimulation. One of the mechanisms involved in Treg-induced suppression is their ability to inhibit IL-2 mRNA expression [15]. To examine whether T cells from CT can respond to the stimulation and up-regulate IL-2 genes, CD4+ T cells from CT and spleens of RIR chickens were stimulated with PMA and Ionomycin for 4 hrs and IL-2 mRNA expression was analysed using RT-PCR. IL-2 expression was not upregulated in CT, while IL-2 gene expression was significantly increased in the cells isolated from spleens (Fig 2J). To examine the role of TGF-beta in providing an inhibitory microenvironment in CT, the mononuclear cells from CT were treated with either anti-TGF-beta or control MAbs and the cells were stimulated with Con A, and the proliferation was analysed using a CFSE-based proliferation assay. The treatment of mononuclear cells isolated from CT with anti-TGF-beta blocking antibody, but not an isotype control MAb, increased their ability to proliferate in vitro (Fig 2K). Con-A was used in all T cell proliferation assays performed in this study. Unlike Con-A, PMA (ranging from 12.5–100 ng/ml) did not stimulate chicken T cell proliferation (S2 Fig). Taken together, the above results suggest that high concentration of TGF-beta+ Treg cells in CT is associated with immune-privileged microenvironment in the largest gut associated lymphoid tissues in chickens, and TGF-beta is involved in the induction of unresponsiveness to the stimulation in CT.
To understand how TGF-beta+ Treg cell are generated in vivo, we developed a protocol for generation of chicken Treg cells in vitro. CD4+ T cells isolated from RIR chickens were cultured for 1–3 days in the presence of plate-bound anti-chicken CD3 antibody. TGF-beta expression was up-regulated on CD4+ T cells 3 days after culture with anti-CD3 antibody (Fig 3A). Addition of rIL-2 increases the expression of TGF-beta on CD4+ T cells (Fig 3B). These in vitro generated TGF-beta+ Treg cells inhibited T cell proliferation upon Con-A-stimulation (Fig 3C), indicating that these cells have immune-regulatory activity. Activation of CD4+ T cells or splenocytes with PMA, PHA or Con-A did not induce TGF-beta expression on chicken CD4+ T cells (ranging from 4.5 to 7% of CD4+ T cells; p = 0.6).
The genetic background plays an important role in the susceptibility of chicken lines to the formation of MDV-induced lymphoma. While, the resistant lines display fewer tumours than susceptible lines, the resistant lines will develop tumours, just at a much-reduced rate (e.g. line P and line 7 birds being susceptible and line N birds being resistant to lymphoma formation). MDV infection induces immunosuppression, however the exact mechanisms involved in MDV-induced immunosuppression is still unknown [11]. Mononuclear cells from the spleens of aged matched naïve chickens from four different chicken lines (line P, line 7, line N and RIR) were isolated and counted prior to staining for flow cytometry analysis. The results showed that there was no significant difference in the total numbers of splenocytes (ranging from 52 to 68 x 106 cells) and total CD4+ T cells among these chicken lines (p = 0.6). The cells were stained with anti-CD4, anti-CD25 and anti-TGF-beta antibodies and the percentages of TGF-beta+CD4+ T cells, TGF-beta+CD4+CD25+ and CD4+CD25+ T cells were analysed using flow cytometry. The MD-susceptible line P and 7 had significantly higher percentages of TGF-beta+CD4+ T cells (Fig 4B) and TGF-beta+CD4+CD25+ T cells (Fig 4C) than the MD-resistant line N, demonstrating that there is an association between high frequencies of TGF-beta+ Treg cells in the spleens and susceptibility to MDV-induced lymphoma formation (p = 0.0008). Intriguingly, there were no significant differences in the frequencies of CD4+CD25+ T cells (Fig 4D) among these chicken lines.
Virulent MDV strains cause immunosuppression and efficiently transform CD4+ T cells resulting in deadly lymphoma [11]. To assess the impact of MDV infection on TGF-beta+ Treg cells in vivo, one day old line P chickens were infected via intra-tracheal route with the virulent MDV strain RB1B. Spleen samples were taken from infected and mock-infected birds during (1) early phase of infection at 4 days post infection (dpi), (2) latent phase at 11 dpi, and (3) the transformation phase of MD at 21 and 28 dpi (Fig 5A). Mononuclear cells were stained with anti-CD4, anti-CD25 and anti-TGF-beta mAbs, and 7AAD was used for the exclusion of dead cells. The percentages and absolute numbers of TGF-beta+CD4+ T cells were determined by flow cytometry. A significant increase (p<0.05, n = 5) in the percentages of TGF-beta+CD4+ T cells was observed in the spleens of MDV infected birds at 21 dpi (Fig 5D and 5E), while there was no difference in the percentages of total CD4+CD25+ T cells (Fig 5B) or CD4+CD25high T cells (Fig 5C) between the infected and mock infected birds at any time points post infection. The absolute numbers of TGF-beta+ Treg cells were calculated and the data confirm that MDV-infection leads to an increase in both percentages and absolute numbers of TGF-beta+ Treg cells at 21 dpi (Fig 5F). However, no difference was found in the percentages or absolute numbers of TGF-beta+ Treg cells between infected and non-infected birds in the spleens at 4, 11, or 28 dpi (Fig 5). Analysis of mean fluorescent intensities (MFI) of TGF-beta expression on CD4+ T cells demonstrated that TGF-beta expression levels were significantly increased on CD4+CD25+ T cells (p = 0.008, n = 5), but not on CD4+CD25- T cells, at 21 dpi (Fig 5G and 5H).
As MDV initially infects lung epithelial cells, we hypothesized that TGF-beta+ Treg cells may be induced in the lungs prior to an increase in their frequencies in spleens. FACS analyses revealed that a higher percentage of TGF-beta+ Treg cells are detected in the lung tissues isolated from MDV-infected birds compared to mock-infected chickens at 4 dpi (p = 0.004) (Fig 5I), while no increase in the percentages of these cells were observed in the spleens at this time. There was also no difference in the percentages of CD4+CD25+ (ranging from 10.5 to 11.5% of CD4+ T cells; p = 0.4) or CD4+CD25high (ranging from 4.5 to 5.3% of CD4+ T cells; p = 0.4) T cells in infected and non-infected lung tissues. The results indicate that the induction of TGF-beta+ Treg cells occur at an earlier stage of infection in the lungs.
In this animal experiment, we set to address three questions regarding to the effects of MDV infection on chicken TGF-beta+ Treg cells; Does the expansion occur (i) in other chicken lines (ii) upon infection via other routes (iii) with oncogenic and vaccine strains of the virus. To address these questions, RIR birds, an outbred chicken line which is moderately susceptible to MDV lymphoma formation, were selected for this experiment. The birds were infected via the intra-abdominal route with the oncogenic virus (RB-1B), the vaccine strains (CVI988-Rispens) or non-infected CEF cells (mock infected group). This route of infection is routinely used to study MDV pathogenesis. Similar to line P birds, there were no changes in the frequencies of total CD4+CD25+ T cells (Fig 6A), or CD4+CD25high T cells (Fig 6B) on 21 dpi. In contrast, the percentages (Fig 6C) of TGF-beta+CD4+ T cells were increased (p = 0.003) in the MDV-infected birds. The attenuated MDV vaccine CVI988-Rispens, (98% sequence identity with RB1B) which can infect and replicate in chickens, but cannot induce lymphoma formation, did not induce TGF-beta+ Treg (Fig 6C). We also assessed the absolute numbers of TGF-beta Treg cells in the vaccinated group and compared it with that in the control birds. In average, the numbers of splenocytes isolated from vaccinated group were not significantly increased (60–87 x 106 cells per spleen) compared to the CEF control group (61–83 x 106 cells per spleen) at 21 dpi. Similarly the absolute numbers of TGF-beta Treg cells were not altered after vaccination (7.3± 1.2 x 105 cells in vaccinated compared to 7.62 ± 0.85 105 cells in the control group per spleen).These results reveal that the induction of TGF-beta+ Treg cells is associated with pathogenesis of the disease and it can also occur in outbred chickens.
Next we examined if the oncogenic RB1B strain also induces TGF-beta Treg cells in a MDV resistant chicken. One day old line N chickens were infected with 1,000 pfu RB1B via intra-abdominal route and the frequencies of TGF-beta Treg cells were compared with the control group (non-infected) 3-weeks post infection using flow cytometry as described above. As expected, no obvious gross pathology was observed in the infected MDV resistant chickens at 3-weeks post infection. The results demonstrated that the infection with the oncogenic virus does not increase the frequencies of TGF-beta Treg cells in the MDV resistant chicken line (TGF-beta Treg cells as % of CD4 T cells were 1.95± 0.54 in the infected chickens compared to %1.86± 0.47 in the control group).
TGF-beta pathway has been linked to tumour progression because it plays an essential role in modulation of cell proliferation and migration [16]. At different stages of tumour development, TGF-beta can either inhibit or promote tumour growth. There is evidence suggesting that overexpression of TGF-beta is associated with poor prognosis in human cancers [17, 18]. Expression of TGF-beta was analysed in spleen tissues obtained from MDV or mock infected line P chickens 3 months post infection, when gross pathological lesions were evident. Tissues were stained with anti-CD4 and anti-TGF-beta antibodies and analysed by confocal microscopy. In mock infected animals, only a small number of CD4+ T cells in the splenic T cell zone co-expressed TGF-beta (Fig 7Ai and 7Aii). In contrast, the majority of CD4+ T cells in the infected spleens expressed TGF-beta as demonstrated by co-localization of CD4 and TGF-beta molecules (Fig 7Bi and 7Bii). Similar to the primary lymphoma cells, MDV-transformed CD4+ T cell clone (clone 265L; generated from line P) expressed both membrane bound (Fig 7C) and intracellular TGF-beta (Fig 7D), as demonstrated using flow cytometry and confocal microscopy, respectively. Approximately 20% of MDV-induced lymphoma cells expressed membrane bound TGF-beta (Fig 7C), while the majority of the lymphoma cells expressed intracellular TGF-beta (Fig 7D). The expression of TGF-beta was not exclusive to 265L clones, as 15–22% of MSB1 clone also expressed mTGF-beta. Interestingly, MDV-induced lymphoma cells (265L) did not express CD25 molecules (Fig 7C). As TGF-beta can act as both paracrine and autocrine factor to inhibit T cell proliferation, we determined the gene expression levels of TGF-beta receptor I and II in 265 L cells (MDV-induced lymphoma cells from line P) using quantitative RT-PCR. MDV-induce lymphoma cells expressed significantly lower levels of TGF-beta receptors compared to the primary naïve CD4+ T cells isolated from line P birds (Fig 7E), suggesting that these cells may be resistant to inhibitory effects of TGF-beta on proliferation. Human lymphoma cells may produce many inhibitory factors such as PGE2, VEGF, IL-10 and TGF-beta that can suppress T cell function, and several of these factors may be involved in the induction of Treg cells. In transwell experiments (splenocytes isolated from RIR birds were cultured in the lower chamber and 265L cells in the upper chamber), 265L cells inhibited Con-A induced T cell proliferation in vitro in a contact independent manner (Fig 7F). As expected, addition of cell culture supernatant from 265 L lymphoma cells (as low as 5% supernatant/ cell culture media ratio) also inhibited T cell proliferation 3 days after stimulation (Fig 7G), while the addition of supernatant to the cell culture had no effect on proliferation of lymphoma cells even at high ratios (60% supernatant/ cell culture media ratio). Fig 7I depicts the proliferation rate of lymphoma cells, in the presence (coloured histograms) or absence (grey histograms) of the supernatant, 24–96 hrs after culture using a CFSE-based proliferation assay. The results suggest that MDV-lymphoma cells express inhibitory factors that can suppress T cell proliferation in a contact independent manner, while it has no effect on the proliferation of the lymphoma cells. One of the inhibitory molecules produced by MDV-lymphoma cells is TGF-beta, which has been shown to inhibit T cell function in vitro [19]. To determine whether TGF-beta produced by the tumour cells can also exert inhibitory effects on T cells, splenocytes isolated from RIR birds were stimulated with Con-A in the presence of the tumour supernatant (5% of total volume) and anti-TGF-beta blocking antibody or a control antibody. The results demonstrate that anti-TGF-beta antibody could not restore T cell function, indicating that other factors secreted by those tumour cells are likely to be more important. To examine the role of COX-2/PGE2 pathway in the inhibitory effects of MDV-induced lymphoma cells, splenocytes isolated from RIR birds were stimulated with Con-A in the presence of the tumour supernatant (5% of total volume) with or without cyclooxygenase-2 (COX-2) inhibitor (SC-236) and anti-TGF-beta blocking antibody. The results demonstrate that COX-2 inhibitor can partially restore T cell proliferation and treatment with both COX-2 inhibitor and anti-TGF-beta blocking MAb further increased Con A-induced T cell proliferation (Fig 7I).Taken together, our data demonstrate that PGE2 and TGF-beta are involved in the inhibition of T cell function in a contact independent manner.
Early studies identified a population of human and murine CD4+ T cells that co-express high levels of CD25, IL-2 receptor α-chain, and demonstrated that these cells are essential for maintenance of peripheral self-tolerance [20]. Because activation of T cells can lead to upregulation of CD25 molecules, subsequent studies proposed Foxp3, a member of the fork-head/winged-helix family of transcriptional factor, as a unique marker for identification of human and murine naturally occurring Treg cells. However, activated human T cells may also transiently expressed Foxp3 [21, 22]. The presence of Foxp subfamily members including Foxp3 is limited to jawed vertebrates and Foxp3-like genes are not found in chicken [5]. Here we show that a subpopulation of chicken CD4+ T cells express membrane bound TGF-beta, an inhibitory cytokine that possess immune-regulatory properties. The results demonstrate that TGF-beta+CD4+ T cells are distinct from CD4+CD25+ T cells as the majority of TGF-beta+CD4+ T cells do not express CD25 molecules.
Intracellular signalling within human and murine Treg cells differs from that of effector T cells. In the latter, activation of AKT results in cytoskeletal rearrangements, cytokine production, cell-cycle progression and engagement of T-cell effector functions. In contrast, Treg cells that are functionally hypo-responsive have a defect in AKT pathway [14], and AKT activation can impair de novo induction of Treg cells by TGF-beta [23, 24]. Our data show that chicken TGF-beta+ Treg cells have a defect in AKT phosphorylation, confirming that these cells are hypo-responsive. In this study, we also demonstrated that chicken TGF-beta+ Treg cells are present in the spleen, blood and lung of naïve birds, but, unexpectedly, these cells were not found in the thymus. This suggests that chicken TGF-beta+ Treg cells are induced in the periphery and that thymic Treg cells do not express membrane bound TGF-beta. However, our result does not exclude the possibility that other cells in the thymus, e.g. apoptotic thymocytes, may express and release TGF-beta which is known to be required for T cell development in the thymus [25, 26]. Induction of TGF-beta+ Treg cells in the periphery could be induced by stimulants such as TLR ligands and metabolites produced by gut microbial fermentation [27–29]. Similar to human Treg cells, chicken TGF-beta+ Treg could be induced via activation of non-Treg cells with plate-bound anti-CD3 antibody in vitro. However, it is unclear how exactly TGF-beta+ Treg cells can be induced in vivo. It is possible that metabolites such as short chain fatty acids produced by gut microbial fermentation induce TGF-beta+ Treg cells [29]. This notion is supported by our results demonstrating that high frequencies of TGF-beta+ Treg cells are found in the largest avian gut associated lymphoid tissue, cecal tonsil (CT). Mononuclear cells from CT were hypo-responsive and expressed inhibitory molecules, suggesting that gut associated lymphoid tissues have an immuno-privileged microenvironment in the chickens, and these cells can modulate the inflammatory response. Further work is required to determine the exact mechanism involved in the induction of large numbers of TGF-beta+ Treg cell in gut associated lymphoid tissues in chickens and study the role of gut TGF-beta+ Treg cells in health and diseases of intestinal tract in chickens. The main function of TGF-beta+ CD4+ T cells in gut associated lymphoid tissues may be to act as T follicular helper cells and to provide help for B cells to produce IgA, as well as suppress pro-inflammatory responses.
Viruses can trigger and expand both nTreg and induced regulatory T cells (iTregs) via poorly understood mechanisms. It is believed that non-TCR mediated mechanisms, e.g. innate ligands, may be involved in the induction of Treg cells. For example, it has been suggested that hepatitis C and influenza viruses can increase circulating bioactive TGF-beta levels by activating latent TGF-beta [30, 31], an inhibitory cytokine which can induce Treg cells. On the other hand, some viruses activate Treg cell expansion via ligation of TLR, induction of IL-2 and generation of inhibitory antigen presenting cells [32]. In this study, we demonstrated that RB1B (very virulent MDV) can induce the expansion of Treg cells in the susceptible chicken lines. Further studies are required to examine the effect of mild MDV or very virulent plus MDV on Treg cell activation/expansion. Based on our data, it can be speculated that the formation of tumour lesions in the host is associated with activation/expansion of Treg cells. High mortality rates observed in the chickens infected with very virulent plus MDV may occur prior to the formation of tumour. This may affect the expansion of Treg cells, however further experiments are required to determine the effects of these viruses on Treg cell expansion/activation. Here, we demonstrated that chicken lines that are genetically susceptible to Marek’s Disease (MD) (line P and line 7) have intrinsically higher concentrations of TGF-beta+ Treg cells. At the time of these experiments, we had no access to tissues from line 6 (another resistant line with similar MHC molecules as line 7) chickens. Therefore, we were unable to examine the frequencies of intrinsic TGF-beta Treg cells in these birds. However, these experiments are planned and the transfer of TGF-beta Treg cells from line 7 (susceptible:B2 haplotype) to line 6 (resistant:B2 haplotype) chickens will also provide further information on the role of TGF-beta Treg cells in susceptibility to the disease. Our current results demonstrate an association between susceptibility to virus-induced lymphoma and concentrations of Treg cells. This hypothesis is in agreement with reports indicating that Treg cells can confer susceptibility to pathogens and the removal of Treg cells protects the animals from the disease [33–36]. In our study, we did not observe any differences in percentages of CD4+CD25+ or CD4+CD25high T cells in the chicken lines, indicating that only TGF-beta+ Treg cells are associated with susceptibility to MDV. But how can TGF-beta+ Treg cells contribute to increased susceptibility to MDV? It is possible that TGF-beta+ Treg cells dampen down the inflammatory response induced by MDV infection, leading to an increase in virus replication and expression of viral genes involved in the transformation. This notion is supported by the results demonstrating that higher MDV replication occurs in the susceptible chicken lines compared to the resistant lines [11]. Since MDV is a cell-associated virus, it is believed that cell-mediated immunity plays an important role in control of the disease. A suppressive effect of TGF-beta+ Treg cells on cell-mediated immunity will result in enhanced susceptibility to the disease. The expression of TGF-beta by MDV-transformed T cells indicates that these cells may exert their immuno-regulatory properties by production of this inhibitory cytokine. TGF-beta inhibits activation and proliferation of naïve T cells, and induces apoptosis in activated T cells [37]. On the other hand, T cells exposed to TGF-beta and IL-6, which is produced during inflammatory response, can develop into Th17 cells, known to be involved in immunopathology [38]. It has been shown that IL-6 gene expression is upregulated in chicken lines that are susceptible to Marek’s Disease. [39]. Therefore, it is possible that induction of TGF-beta and IL-6 in the MDV-infected susceptible birds can lead to the development of Th17 cells. In addition to the immunological effects of TGF-beta, this cytokine can induce apoptosis of various cells including virus-infected cells or transformed cells. How do MDV-induced lymphoma cells escape anti-proliferative effects of autocrine and paracrine TGF-beta? Some cancer cells escape the tumour suppressor effects of TGF-beta by down-regulation/mutation of the type 1 and 2 receptors, Smad2 and Smad4 [40]. It has been shown that a viral analog of cellular miR-155 in MDV, which is critical for oncogenicity of MDV, down-regulate TGF-beta signalling [41]. In agreement with this report, our results show that MDV-transformed TGF-beta+ lymphoma cells down-regulate TGF-beta receptors, altering tumour suppressive activity of TGF-beta to tumour-promoting activity. Lymphoma cells may release many inhibitory factors (TGF-beta, VEGF, PGE2, etc.) which can cause immunosuppression. Here we demonstrated that COX-2 inhibitor can partially restore the inhibitory effects of soluble factors released by the tumour cells confirming that MDV-induced tumour cells modulate the function of immune cells via COX-2/PGE2 pathway. In addition, we observed that anti-TGF-beta have synergistic effect with COX-2 inhibitors, while has no or very negligible effect when it is used on its own. The results indicate that PGE2 and TGF-beta released by lymphoma cells may be involved in the immunosuppression.
One of the prominent features of Marek’s Disease (MD) is a generalized immuno-suppression in chickens. The exact mechanism responsible for MDV-induced immuno-suppression has not yet been defined [11]. As summarized in the proposed model shown in Fig 8, we demonstrate that the induction and expansion of TGF-beta+ Treg cells are involved in MD-induced immunosuppression. It should be noted that Treg cells and TGF-beta producing tumour cells are probably induced by different mechanisms and are not necessarily the same in either phenotype or behaviour. The effects of MDV infection on induction/expansion of TGF-beta+ Treg cells suggest that these cells may be involved in pathogenesis of the disease, and expansion of these cells can contribute to MDV-induced immunosuppression. This notion is supported by observation that the MDV vaccine strain (CVI988-Rispens) which shares a 98% homology with virulent MDV strains does not trigger the induction of TGF-beta+ Treg cells. Furthermore, we demonstrate that MDV-induced tumour cells produce PGE2 and TGF-beta and perhaps other soluble factors that can inhibit T cell function, and may contribute to MDV-induced immunosuppression after transformation phase of the disease. Taken together, our results indicate that Treg cells may be involved in MDV-induced immunosuppression at early stage of infection and soluble factors produced by MDV-induced lymphoma cells can cause immunosuppression.
All animal experiments were conducted based on the guidelines and care approved (project license PPL 30/3169) by the UK government Home Office and the personnel involved in the procedures had obtained personal license from the UK Home Office.
SPF inbred chicken lines including line P (homozygous for the B19 haplotype), line N (homozygous for the B21 haplotype), line 7 (homozygous for the B2 haplotype), and Rhode Island Red (RIR; outbred chicken line) were raised and kept at isolation unit at The Pirbright Institute. Tissues (spleen, lungs, cecal tonsil, and thymus) were harvested after cervical dislocation of the birds. In total, 40 one-day-old line P birds were inoculated with dusts via intra-tracheal route using a Penn-Century device. Twenty a-day old line P birds per group were inoculated with 5 mg of RB1B-containing chicken dust via intra-tracheal route or mock-infected controls. The dust was collected from our previous experiment in which one-day-old RIR birds were infected with CEF cells-RB1B (1,000 PFU) or mock-infected CEF cells via intra-abdominal route. MDV-genome copies were confirmed in the dust using qPCR as previously described [42]. Five MDV-infected birds and 5 mock-infected birds were taken at 4, 11, 21 and 28 dpi. Replication of MDV (RB1B) was confirmed by analyzing genome copy numbers were above 1.00E+3 in spleens of birds using qPCR. MDV-infected birds showed splenomegaly in post-mortem examination on 21 and 28 dpi. Day-old Rhode Island Red chickens were inoculated with 1,000 PFU CEF cells-RB1B (oncogenic virus; n = 8), 1,000 PFU CEF cells-CVI988/Rispens (vaccine strain; n = 8), or mock infected-CEF cells (controls only; n = 8) via the intra-abdominal route, and splenocytes were harvested on 21 dpi for further analysis. Day-old line N chicks (n = 3) were injected with 1,000 PFU RB1B via intra-abdominal route, and the frequencies of TGF-beta+ Treg cells in the spleens were analysed 3-weeks post infection.
A day-old Rhode Island Red (RIR) chickens were infected via the intra-abdominal route with 1,000 plaque forming units (PFU) of RB1B. Splenocytes were harvested at 14 dpi and were co-cultured with primary CEF cells for 7 days. Once the cytopathic effects were observed, the cell-associated MDV-infected CEF cells were passaged two times in fresh CEF to prepare virus stocks. The virus stocks were titrated and stored in liquid nitrogen. Commercial CVI988/Rispens vaccine virus (Nobilis Rismavac) was obtained from Intervet.
Dust samples were prepared as 5 mg aliquots and total DNA was isolated from dust samples using a DNeasy kit (Qiagen, Manchester, UK). For absolute quantification of MDV genome copies, we performed real-time quantitative duplex PCR (q-PCR) for the detections of the MDV meq gene in DNS isolated from dust and splenoyctes. The chicken ovotransferrin (ovo) was used as a reference gene as previously described [42]. Ten microgram bovine serum albumin was added to reactions using dust samples to block the inhibitory effect of melanin pigment. MDV genomes (for 104 cells or 1 μg dust) were quantified using standard curves which were calibrated against plasmid contracts of known target gene copy number.
Total RNA was isolated from CD4+ T cells isolated from splenocytes or cecal tonsils after stimulation with PMA (50 ng/ml) and ionomycin (100 ng/ml) at 41°C, 5% CO2 for 4 hrs. The RNA was quantified and treated with DNA-free DNA removal kit (Thermo Fisher Scientific, Paisley, UK) for removal of genomic DNA contamination and reverse transcribed into cDNA with maxima H Minus First Strand cDNA Synthesis Kit according to the manufacturer's instruction (Thermo Fisher Scientific, Paisley, UK). Primer sequences for all target cytokines and housekeeping genes are enlisted in S1 Table. The SYBR green based real-time PCR was performed with a LightCycler 480 SYBR Green I Master kit (Roche Applied Science, USA). The Quantitative PCR was performed in QuantStudio5 (Thermo Fisher Scientific, Paisley, UK) with the following conditions: initial denaturation was performed at 95°C for 1 min, followed by 40 cycles of denaturation at 95°C for 15s and annealing / extension at 60°C for 30s, with end point melt-curve analysis. The relative fold change of target genes in cecal tonsils was calculated by 2−ΔΔCT method. The Ct value for each sample was normalized against GAPDH housekeeping gene for respective sample. The amplified TGF-beta receptor I and II were cloned in pGEMT-easy vector (Promega, USA) and used for generation of the standard curve equation to calculate the absolute copy number of the genes.
Mononuclear cells isolated from spleens, lungs; cecal tonsils, blood, and thymus were stained with fluorochrome-conjugated monoclonal antibodies, and the expression of surface molecules were analysed using flow cytometry. Prior to staining, the cells were incubated with chicken serum to block FC receptors, and then the cells were incubated with anti-CD4-PE (Cambridge Bioscience, Cambridge, UK), anti-TGF-beta1,2,3 (R&D systems, Abingdon, UK), anti-CD25 FITC mAbs (Bio-Rad, Watford, UK) at 4°C for 15 min. The cells were washed twice and 7-AAD staining was used to exclude dead cells. The minimal number of 100,000 cells was acquired for FACS analysis. Cells were acquired on a MACSQuant analyser and the data were processed by FlowJo V10 software.
For Phosflow analysis of the cells, mononuclear cells were plated at 5.0 x105 cells per well and stimulated (15 min; 41°C, 5% CO2) with PMA (50 ng/ml) or cell culture media only. The staining was conducted according to the manufacturer’s instruction as previously described [43]. After fixation of the cells for 30 minutes using the fixation/ permeabilization kit (Thermo Fisher Scientific, Paisley, UK), the cells were stained with anti-TGF-beta (R&D Systems, Abingdon, UK) and anti-CD4 mAbs (Cambridge Bioscience, Cambridge, UK), and were permeabilized using Perm buffer for 10 min at 4°C, followed by staining with anti-AKT (S473) PE-conjugated mAbs (BD Bioscience, Oxford, UK) at 4°C for 30 minutes. The cells were washed twice in Perm buffer and re-suspended in FACS buffer and acquired using flow cytometry.
The inhibitory effects of TGF-beta+ Treg cells on T cell proliferation were analysed in an in vitro inhibition assay. TGF-beta+ Treg and TGF-beta- CD4+ T cells were purified from splenocytes using anti-PE and anti-APC micro-bead kits according to the manufacturer’s instructions (Miltenyi Biotec, Woking, UK). TGF-beta+ Treg cells or control cells were co-cultured with the responder cells (CFSE-labelled splenocytes [depleted from TGF-beta+ Treg cells]; 3 X 104) at an effector/responder cell ratio of 1:1 in three replicates. The cells were stimulated with 2.5μg/ml Con-A to induce T cell proliferation. In each experiment, the cells from 7 birds were pooled to isolate enough cells for the inhibition assays. T cell proliferation was analysed using flow cytometry at 72 hrs after the co-culture, and proliferation index was calculated using FlowJo v10.07.
For assessing the inhibitory effects of in vitro generated TGF-beta+ Treg cells, the responder cells (3 x 104 cells/ well) were co-cultured with TGF-beta+ Treg cells or TGF-beta- CD4+ T cells at an effector/responder ratio of 1:1 in three replicates as described above. The numbers of live cells (7AAD negative) in the culture were counted using flow cytometry.
To study the role of PGE2 and TGF-beta in the inhibitory effects, the cells were treated with 10 μM SC-236 (COX-2 inhibitor) (Tocris, Abingdon, UK) and /or 1 μg/ml anti-TGF-beta blocking antibody (clone 1D11) or a control MAb (mouse IgG1) (R&D Systems, Abingdon, UK) prior to Con-A stimulation (2.5μg/ml).
MDV-derived lymphoma cell lines were adhered to coverslips using Cell-Tak adhesive (BD Bioscience, Oxford, UK). Tissues were immediately frozen in isopentane mixed with dry ice, and the frozen samples were kept at -80°C. Tissue sections were cut at 8μm and mounted on gelatin-coated histological slides. Sections were fixed 4% paraformaldehyde at Room temperature for 1 hr and permeabilized in 0.1% Triton X100 in PBS for 10 minutes. After rehydration and blocked (0.5% BSA in PBS), the tissue sections or cells were incubated with mouse anti-chicken CD4 (IgG2b) (Cambridge Bioscience, Cambridge, UK), and mouse anti-TGF-beta (IgG1) (R& D Systems, Abingdon, UK) overnight at 2–8°C. An isotype matched control (R&D Systems, Abingdon, UK) was used to identify non-specific staining. After three washes in PBS at room temperature, the tissues/ cells were incubated with fluorochrome-labeled anti-mouse IgG1 or IgG2b antibodies (Thermo Fisher Scientific, Paisley, UK), and DAPI solution was added for 10 minutes at room temperature. The sections were mounted onto glass slides using Vectashield mounting medium, and coverslips were sealed with nail varnish before visualisation by confocal microscopy (Leica Microsystems).
Chicken ELISPOT assay was performed as described previously [43]. Briefly, nitrocellulose-backed plates were coated with anti-chicken IFN-gamma capture antibody overnight at 4°C. The wells were washed and blocked using RPMI1640 containing 2% FCS (1hr at 41°C). Mononuclear cells were seeded (3 x106 cells/ml) and stimulated with Phorbol Myristate Acetate (PMA; 50 ng/ml) and ionomycin (100 ng/ml) in six replicates and incubated at 41°C for 18hrs. The wells were washed with PBS and distilled water, and then incubated with anti-chicken IFN-gamma biotin (1 μg / ml) (Thermo Fisher Scientific, Paisley, UK) for 1 h. After several washes, the plate was incubated with 100 μl of streptavidin-HRP (1:1000) at 41°C for 1 h. The assay was developed with AEC substrate as per manufacturer’s instruction (BD Bioscience, Oxford, UK). The reaction was terminated by washing plates under running water. The spots were enumerated in the dried plate with ELISPOT reader (Advanced imaging devices, GmbH, Germany). The counted spots were finally expressed as spot forming units (SFU) per million mononuclear cells.
MDV-transformed CD4+ T-cell lines were cultured (41°C, 5% CO2 incubator) in RPMI 1640 growth medium (Thermo Fisher Scientific, Paisley, UK); supplemented 100 μg/mL of penicillin/streptomycin, 10% fetal calf serum, 10% tryptose phosphate broth, 1 mM sodium pyruvate; (Sigma-Aldrich, Dorset, UK), and 50 μM 2-mercaptoethanol (Thermo Fisher Scientific, Paisley, UK).
Statistical calculation included Wilcoxon and Mann Whitney non-parametric to determine significance. Results were considered statistically significant at P < 0.05 (*).Statistical analysis of the data was performed using Graph Pad Prism 7. The data are presented as mean +-SD.
|
10.1371/journal.pgen.1008047 | The dynamin-like protein Fzl promotes thylakoid fusion and resistance to light stress in Chlamydomonas reinhardtii | Large GTPases of the Dynamin Related Proteins (DRP) family shape lipid bilayers through membrane fission or fusion processes. Despite the highly organized photosynthetic membranes of thylakoids, a single DRP is known to be targeted inside the chloroplast. Fzl from the land plant Arabidopsis thaliana is inserted in the inner envelope and thylakoid membranes to regulate their morphology. Fzl may promote the fusion of thylakoids but this remains to be proven. Moreover, the physiological requirement for fusing thylakoids is currently unknown. Here, we find that the unicellular microalga Chlamydomonas reinhardtii encodes an Fzl ortholog (CrFzl) that is localized in the chloroplast where it is soluble. To explore its function, the CRISPR/Cas9 technology was employed to generate multiple CrFzl knock out strains. Phenotypic analyzes revealed a specific requirement of CrFzl for survival upon light stress. Consistent with this, strong irradiance lead to increased photoinhibition of photosynthesis in mutant cells. Fluorescence and electron microscopy analysis demonstrated that upon exposure to high light, CrFzl mutants show defects in chloroplast morphology but also large cytosolic vacuoles in close contact with the plastid. We further observe that strong irradiance induces an increased recruitment of the DRP to thylakoid membranes. Most importantly, we show that CrFzl is required for the fusion of thylakoids during mating. Together, our results suggest that thylakoids fusion may be necessary for resistance to light stress.
| All eukaryotic cells are composed of compartments with defined functions. Among those, mitochondria generate the main source of energy in human and animal cells. Their capacity to generate and diffuse energy in the cell is regulated by fusion and fragmentation processes.
Together with mitochondria that produce energy from oxygen, plant cells include an additional compartment called the chloroplast that produces energy from light. The machinery that converts light into energy is more precisely located inside the chloroplast within stacks of membranes called the thylakoids. Here, we elucidate the function of CrFzl, a previously uncharacterized protein encoded by the genome of the unicellular alga Chlamydomonas reinhardtii. Algal cells that do not contain CrFzl are impaired in their capacity to grow when they receive too much light and our results indicate that CrFzl promotes the fusion of thylakoids during mating. These results suggest that membrane fusion is an essential tool for energy production in stress conditions by living organisms.
| Dynamin-Related-Proteins (DRPs) are large GTPases that bind lipid bilayers and auto-oligomerize into high order macro-molecular structures [1]. These features provide DRPs with a dedicated ability to remodel intracellular membranes through fusion, fission and tubulation processes. The founding member of the DRP family, Dynamin 1 (Dyn1), is critical for clathrin-coated vesicle endocytosis in mammals [2]. DRPs are on the other hand essential for mitochondrial dynamics in fungi, metazoans and plants where mitochondria organize into tubular cellular networks through balanced events of division and fusion of their tubules [3,4]. In line with their strong involvement in mitochondrial dynamics and with the endosymbiotic origin of mitochondria, DRPs have evolved from prokaryotic ancestors and have been characterized in bacterial models [5,6,7]. They are highly conserved in the cyanobacterial subgroup [8] and consistent with the cyanobacterial origin of the chloroplast [9], DRPs also participate in the membrane dynamics of this fundamental photosynthetic organelle [10].
From a molecular standpoint, GTP-binding promotes Dyn1 recruitment and self-assembly into spirals around the neck of budding vesicles. GTP hydrolysis then induces constriction of the spiral resulting in scission of the lipid tube, which releases the vesicle from the plasma membrane [11]. DRP1, in mammals, and Dnm1, in yeast, act similarly to Dyn1 to promote the separation of mitochondrial tubules [1]. These “fission DRPs” are cytosolic GTPases that are recruited to their cognate membranes by specific protein and lipid adaptors. In contrast, “fusion DRPs” are generally trans-membrane GTPases [1]. Consistent with this, fusion of mitochondria is mediated by two distinct sets of DRPs that are integral to outer and inner membranes. The mitofusins (MFNs in mammals, Fzo1 in yeast) oligomerize in cis (i.e. on the same membrane) and then in trans (i.e. from opposing membranes) to tether mitochondrial outer membranes and trigger their homotypic fusion. Tethering and fusion of inner membranes is subsequently mediated by OPA1 in mammals and Mgm1 in yeast. GTP binding and hydrolysis are crucial for both membrane fusion processes, which are themselves essential for oxidative phosphorylation.
The cyanobacterial BDLP (Bacterial Dynamin Like Protein) from Nostoc punctiforme shares significant homology with mitofusins, suggesting its involvement in membrane fusion processes [12]. Notably, NpBDLP is soluble and self-associates as a dimer in its nucleotide-free form [12]. GTP binding promotes an extensive conformational switch of the protein from a compact structure to an open configuration [13]. In this GTP-bound state, NpBDLP inserts into lipid bilayers through a hydrophobic paddle and is able to generate tubulation of liposomes through its auto-oligomerization. In vivo, NpBDLP was mostly observed at the cellular internal periphery and was proposed to regulate thylakoid organization because of the cyanobacterial origin of plant chloroplasts [12].
Plant DRPs have been mainly studied in the Arabidopsis thaliana model, which revealed their involvement in endocytosis [14,15], in division of mitochondria and peroxisomes [16], and in cytokinesis [17]. Most importantly, plants host the essential photosynthetic electron transport chain within the highly complex and dynamic thylakoid membrane network [18]. Thylakoids reside in the chloroplast which is itself delimited by outer and inner envelope membranes. DRP5B, formerly known as ARC5, is responsible for the chloroplast division through a mechanism similar to mitochondrial fission [19]. The only other DRP known to be implicated in maintenance of the chloroplast morphology is called Fzl due to its similarity with the yeast mitofusin Fzo1 [20].
Fzl is the sole mitofusin-like protein in Arabidopsis thaliana. It does not regulate mitochondrial dynamics but was found inserted in the inner envelope of the chloroplast and within thylakoid membranes [20]. Its homology with NpBDLP, suggested a conserved function from cyanobacteria to plants [12]. The absence of Fzl induces a slow growth phenotype associated with the development of pale green leaves in the Columbia (Col-0) ecotype [20,21]. In the Landsberg erecta (Ler) background, the slow growth was conserved but the older leaves developed chlorotic areas instead of homogenous bleaching [22,23]. fzl cells display enlarged chloroplasts with morphologically altered thylakoids. In addition, an accumulation of stromal vesicles was detected in Col-0 [20] whereas autophagosomes were visualized in the cytoplasm of fzl-Ler mutant cells [23]. This upregulation of autophagy was associated with increased hydrogen peroxide deposit in the cell-wall and with activation of cell death signaling pathways [22,23]. The analysis of photosynthetic electron transport chain components from thylakoids of the fzl Col-0 background revealed reduced levels of chlorophyll and decreased accumulation of cytochrome b6f complex in older pale green leaves exclusively [21].
Significant but heterogeneous phenotypes have thus been linked to the absence of Fzl in Col-0 and Ler ecotypes of Arabidopsis thaliana. In this context, reciprocal causality between the structural alterations of thylakoids, the photosynthetic defect observed in older leaves or the activation of autophagy and cell death pathways can hardly get established. Importantly, a possible involvement of Fzl in fusion of thylakoids has been proposed based on its similarity with mitofusins [20]. Consistent with this, cryo-electron tomography on germinating Arabidopsis thaliana cotyledons recently reported a delayed interconnection between developing thylakoid stacks in the absence of the DRP. Moreover, immunogold labeling localized Fzl at the points of interconnection between thylakoids [24]. Nonetheless, the established involvement of Fzl in fusion of thylakoids remains to be demonstrated and besides its potential role in architecture maintenance, the physiological function of thylakoids fusion is unknown.
With the objective of clarifying the function of Fzl in chloroplast biology and given its conservation in the green lineage [20], we focused on the counterpart of Arabidopsis thaliana Fzl from the unicellular microalga Chlamydomonas reinhardtii, which contains a single chloroplast [25]. Chlamydomonas Fzl (CrFzl) was found targeted to the chloroplast, in which it is soluble. Insertional knock out mutagenesis by CRISPR Cas9 allowed the generation of strains lacking CrFzl which revealed a requirement of the DRP during high light stress. High light treatment in the mutant induces morphological changes in the chloroplast accompanied by a significant photoinhibition of the photosynthetic electron transport chain, an accumulation of cytosolic vesicles in contact with damaged plastids and an activation of autophagy. Our observations indicate that the GTPase and hydrophobic domains of CrFzl are essential for its recruitment to intra-plastidial membranes under high light and demonstrate that CrFzl is essential for effective fusion of thylakoids during mating. Together, these results indicate that CrFzl promotes the merging of thylakoids and is required for resistance to light stress.
Using the Phytozome plant genome database, we found that the Cre14.g616600 locus of Chlamydomonas reinhardtii encodes a putative CrFzl ortholog of AtFzl. Bioinformatics analysis of the putative CrFzl protein identified a predicted GTPase domain (Fig 1a, top). The GTPase domains of CrFzl, AtFzl and NpBDLP counterparts, presented the strongest similarities with those of mitofusins (Fig 1a, bottom). Online prediction tools also identified a chloroplast transit peptide (TP), a thiamine monophosphate synthase (TMP) domain, a hydrophobic domain and one C-terminal coiled-coil fragment (Fig 1a, top). The precise properties of the hydrophobic region were uncertain as bioinformatics tools predicted two, one or no transmembrane domain at all in this region (see Materials and methods).
The genomic CrFzl locus was subcloned under the promoter of the PhotoSystem I (PSI) subunit PsaD and in frame with three C-terminal HA epitopes to allow constitutive expression of CrFzl-3HA in a wild-type strain. A specific signal migrating around 110 kDa was detected close to the expected molecular weight of CrFzl-3HA (CrFzl = 108 kDa + 3HA = 4.5 kDa), which validates the gene model and predicted size of CrFzl (Fig 1b). Subcellular fractionation designed to distinguish cytosolic, chloroplast and thylakoid fractions [26], revealed that CrFzl-3HA is mainly associated with the chloroplast fraction defined by the presence of the large Rubisco subunit (RbcL) and the thylakoid β-CF1 subunit of ATPase (S1 Fig). Moreover, CrFzl fused at its C-terminus with the Yellow Fluorescent Protein variant mVenus (CrFzl-mVenus) co-localized with the chlorophyll autofluorescence of the chloroplast, whereas free mVenus localized in the cytosol of Chlamydomonas reinhardtii cells (Fig 1c and 1d).
While these results confirm that constitutively expressed CrFzl is localized in the chloroplast, we generated polyclonal antibodies to evaluate the expression and localization of the endogenous CrFzl protein. As expected, these antibodies (anti-CrFzl) recognized both the endogenous and the 3HA-tagged CrFzl proteins (Fig 1e). In subcellular fractionation assays, endogenous CrFzl was largely associated with the chloroplast fraction positive for RbcL and β-CF1 (Fig 1f, Cp). CrFzl was also detected in the thylakoid fraction enriched for β-CF1 but this fraction was often contaminated with significant amounts of soluble RbcL (Fig 1f, Thy). Treating this thylakoid fraction with increasing amounts of salts (50, 150 and 500 mM NaCl) did not displace β-CF1, which is tightly bound to membranes, but induced progressive dissociation of both RbcL and CrFzl (Fig 1g). This suggests that CrFzl is mainly soluble in the chloroplast, which led to monitor the association of CrFzl-3HA with membranes by dissociating soluble and membrane fractions from whole cells. Interestingly, the DRP accumulated in the same fraction as the soluble Nucleic Acid Binding protein 1 (NAB1) while a smaller portion remained associated with the membrane fraction that contains β-CF1 (Fig 1h).
These results thus indicate that CrFzl is targeted to the chloroplast where the protein is mainly soluble.
Functional analysis of CrFzl required strains that lack expression of the protein. The Chlamydomonas Mutant Library [27] contains one strain, LMJ.RY0402.175738, in which a paromomycin resistance cassette is inserted in the open reading frame of the Cre14.g616600 locus. However, preliminary analyzes revealed that this strain presents phenotypes that do not segregate with the insertion cassette, suggesting additional mutations in its genome. We thus took the challenge of generating CrFzl knock out strains using the still emerging CRISPR/Cas9 technology in Chlamydomonas reinhardtii [28,29,30,31]. We designed a small guide RNA (sgRNA) that targets a 20 bp sequence from the first exon of the Cre14.g616600 locus, leading to a cleavage by the Cas9 nuclease after the 66th nucleotide (S2a Fig). The wild-type cw15.J3 strain was co-transformed with the sgRNA/Cas9 RNP complex and a cassette containing the AphVII hygromycin resistance gene. Roughly 800 hygromycin resistant colonies were obtained and 322 of those were screened using a PCR strategy with 3 distinct primers (S2b Fig) expected to generate 3 distinct products (Fig 2a): a 760 bp PCR product when the CrFzl locus is uninterrupted and 1261 bp (sense) or 715 bp (antisense) products depending on the orientation of the cassette upon insertion at the expected genomic site. Out of 322 candidate strains subjected to PCR screening, 153 (47.5%) displayed a product different from the 760 bp wild-type PCR fragment (Fig 2b; the screen for the first 15 clones is shown). Among these, 41 presented the sense product (Fig 2b, S, clones #10 and #15), 62 the antisense fragment (Fig 2b, AS; clones #4, #8, #12 and #14) and 48 clones had more complex rearrangements resulting in different lengths and number of the PCR products (Fig 2b, clone #1, #2 and #11). We sequenced the flanking regions of the predicted cut site at the CrFzl locus of 4 distinct clones (#1, #8, #10 and #12) and confirmed insertion of the resistance cassette at that site (S2c Fig).
Expression of the endogenous protein in five candidate CrFzl mutants (clones #1, #4, #8, #10 and #12) was then assessed by western blot of total cellular extracts with the anti-CrFzl (Fig 2c, dashed boxes). The wild-type strain from the Chlamydomonas Mutant Library, CC-4533, and the mutant strain LMJ.RY0402.175738 (ΔCrFzl CLiP) for which the cassette insertion is predicted in the 8th intron were used as additional controls. We confirmed the successful CRISPR/Cas9-based inactivation of CrFzl since its expression was undetectable in all mutants. The CRISPR/Cas9 technology thus allowed the generation of several strains defective for the expression of CrFzl in a timely fashion (6 days for selection of hygromycin resistant strains, 4 days for amplification of isolated mutants, 1 day for DNA extraction and PCR and 3 days for growth, protein extraction and western blotting of selected mutants).
The effect of abrogating CrFzl expression was first assessed by growth of wild-type and CrFzl mutant strains in mixotrophy on TAP medium (growth relies on supplied acetate and photosynthesis) or in photoautotrophy on minimal medium (TP, growth solely relies on photosynthesis). On TAP medium, three distinct CrFzl mutants (clones #1, #8 and #10) displayed a slight growth defect as compared to the wild-type (Fig 2d, TAP, 50 μmol photons.m-2.s-1). On minimal medium, the wild-type strain grew better with higher light intensity (compare 50, 100 and 200 μmol photons.m-2.s-1), whereas growth of the mutants was nearly abrogated at 100 and 200 μmol photons.m-2.s-1 (Fig 2d). In mixotrophic conditions, the growth of mutant strains was also gradually impaired with higher irradiance (Fig 2d, TAP, 100 μmol photons.m-2.s-1, 200 μmol photons.m-2.s-1). Importantly, reintroducing CrFzl expression in the mutants fully rescued the growth phenotype, both on TP and TAP media under all light regimes (Fig 2e). These results thus demonstrate that CrFzl is required to maintain cell division and growth under high light.
Consistent with this observation, the growth of mutant strains (clones #1 and #8) was undistinguishable from that of the wild-type in heterotrophic conditions, i.e. liquid TAP medium and darkness (Fig 2f). In contrast, moderate light (50 μmol photons.m-2.s-1) induced a slight delay in growth induction of the mutants in both TAP and TP liquid media (Fig 2g and 2h). Most strikingly, growth of the mutants in high light regime (400 μmol photons.m-2.s-1) was strongly delayed in TP and even completely inhibited in TAP liquid cultures (Fig 2i and 2j).
CRISPR/Cas9 mutagenesis thus allowed the successful generation of multiple ΔCrFzl mutants and their preliminary analysis revealed the requirement of CrFzl for survival in conditions of high illumination.
Growth on agar and in liquid media indicated a high sensitivity of ΔCrFzl mutant strains to light stress. Consistent with a specific effect on the chloroplast, we observed that both wild-type and mutant strains display undistinguishable tubular mitochondrial networks when stained with the mitochondrial dye rhodamine 123 (Fig 3a and 3b). This confirms that, in contrast to its yeast homolog Fzo1, Chlamydomonas Fzl is not involved in the regulation of mitochondrial dynamics [20]. Moreover, the wild-type strain and a CrFzl mutant were grown under normal light (50 μmol photons.m-2.sec-1) in TAP medium with or without NaCl to assess the effect of high salt concentration, another abiotic stress [32], upon loss of CrFzl expression. As expected, incubation under normal light in TAP medium without NaCl caused a very weak delay in the induction of growth for the mutant strains (Fig 3c). However, increasing salt concentration to 100 mM NaCl [33] did not further affect growth of the ΔCrFzl strain (Fig 3d), confirming that the growth phenotypes we observed are not affected by high salt and may thus be specifically associated with light stress.
During exposure to high light, photosynthetic electron transport is saturated and chlorophylls excited by light will dissipate their energy by forming reactive oxygen species (ROS), inducing damage to the photosynthetic machinery. Excess damage promotes degradation of the D1 protein, which is the core subunit of Photosystem II (PSII), resulting in partial inactivation of the global PSII pool and photoinhibition [34]. This manifests as a decrease of the maximum PSII quantum yield (Fv/Fm, see Material and methods). This indicator of PSII efficiency can be monitored in vivo by measurements of chlorophyll fluorescence during photoinhibition (Fig 3e). The wild-type strain and a ΔCrFzl mutant, grown in TAP or TP medium, were thus exposed to photoinhibitory light irradiance (850 μmol photons.m-2.s-1) during one hour followed by recovery in very low light for three hours. In both growth conditions, a significant decrease of the PSII quantum yield was observed in the ΔCrFzl mutant upon photoinhibition whereas both strains recovered from light induced damage at a similar rate (Fig 3f and 3g), These results therefore indicate that CrFzl protects photosynthesis against photoinhibition. This protection may allow limiting the amount of ROS induced by photoinhibition, whereas excess ROS may contribute to the growth inhibition in high light of ΔCrFzl mutants (Fig 2j). In agreement with these possibilities, the growth of the wild-type strain was progressively inhibited with increasing concentrations of NAC (Fig 3h; WT) whereas the ΔCrFzl mutant was progressively rescued by the antioxidant (Fig 3h; ΔCrFzl). CrFzl is thus required for growth during high light stress to possibly protect photosynthesis against ROS generated during photoinhibition.
Our data indicate that CrFzl is required for growth during light stress and that strong irradiance induces reduced resistance to photoinhibition in the mutant. Given that CrFzl belongs to the DRP membrane remodeling factors family, we hypothesized that these defects might result from perturbations in the plastid membrane system. We thus examined the morphology of the chloroplast taking advantage of the fluorescence emitted by the chlorophyll pigment that is mainly inserted in the thylakoid membrane.
Under low irradiance, the plastids from wild-type and CrFzl mutant strains are detected as 2 lobes appressed against the plasma membrane and joining around the pyrenoid where chlorophyll fluorescence cannot be seen as it mostly consists in non-chlorophyll-associated proteins [35,36] (Fig 4a and S3a Fig). After 6 hours of light stress, this typical cup shape of the chloroplast was unchanged in the wild-type strain but ΔCrFzl cells accumulated dark structures within the fluorescent lobes of the plastids (Fig 4a, arrows).
While analyzing the cells with distinct wavelengths and filter sets, we also detected fluorescent dot-like structures (Fig 4b, S3b Fig, arrowheads). These dots were spotted with settings usually used for detection of DAPI, CFP, GFP or YFP but not for detection of chlorophyll (Chloro setting), indicating that their spectral properties spanned wavelength of 365–505 nm for excitation and 397–600 nm for emission. In normal light conditions, the fluorescent dots were hardly detected in wild-type cells (< 3%) but seen in more than 20% of mutant cells (Fig 4c). High light stress increased their occurrence by 3-fold in both wild-type (27%) and mutant cells (63%). The origin of their fluorescence is unknown but these dots appeared nearly systematically located in the cytoplasm and in close juxtaposition with the chloroplast (Fig 4b).
Conventional transmission electron microscopy (TEM) analysis revealed that absence of CrFzl in fact induces the accumulation of starch granules in lobe regions of the chloroplast (Fig 4d and 4e and S3c Fig; asterisks). Notably, this increase in starch granules correlated with an altered shape of the pyrenoid that, in normal light conditions, was clearly identifiable in only ~7% of mutant cells against 20% in the wild-type (S3d Fig). These starch granules devoid of chlorophyll likely correspond to the dark structures that accumulate in the fluorescent lobes of plastids from ΔCrFzl cells.
Another important feature of the ΔCrFzl strain analyzed by TEM was the systematic observation of very large cytoplasmic vesicular structures in both normal and high light conditions (Fig 4d and 4e, S3 Fig; arrows and v). In the ΔCrFzl strain, these vesicles were much larger than vesicles seen in the wild-type (Fig 4f) and were filled with electron dense material (Fig 4d). Notably, the vesicles and the chloroplast were in such contact that the plastid envelope was often undistinguishable from the membrane of the vesicle (Fig 4e; arrows). This led us evaluating a possible involvement of autophagy by monitoring the expression of ATG8 which is known to be stimulated upon induction of autophagy by Rapamycin ([37] and Fig 4g). Interestingly, the levels of this autophagy marker were increased upon strong illumination of WT cells and this increase was even stronger in ΔCrFzl mutants (Fig 4g).
Upon treatment with high light, cells lacking CrFzl thus simultaneously combine two features that may relate to the degradation of the chloroplast: an activation of autophagy (Fig 4g) and large vesicles that adjoin the chloroplast and may correspond to vacuoles (Fig 4d and 4e). In this context, carotenoids that absorb light between 400 and 520nm [38] and that accumulate in thylakoid membranes may represent candidates for the fluorescence of the spots that were found in the vicinity of the plastid (Fig 4b and 4c).
These results indicate that lack of CrFzl perturbs the organization of the chloroplast. This may be linked to the decreased resistance to photoinhibition during light stress, which would lead to degradation of the plastid and cell death upon extended exposure. This prompted us to begin investigating the mean by which CrFzl regulates the morphology of the chloroplast.
The high light phenotypes induced by the absence of CrFzl likely result from defects that are secondary to the loss of function of the DRP. We thus reasoned that following the behavior of the CrFzl protein during high light stress could begin hinting at its primary function. For this purpose, we generated a ΔCrFzl strain expressing CrFzl-mVenus as the unique source of CrFzl in the cell and followed the distribution of the protein in normal and high light conditions by fluorescence microscopy (Fig 5a, S4a Fig). In the wild-type strain expressing the untagged CrFzl, a diffuse background signal was detected with the YFP filter set in both normal and high light conditions but this signal did not demarcate any specific subcellular structure (Fig 5a; WT NL, WT HL). In CrFzl-mVenus cells, a specific YFP signal was detected. In normal light (NL), this YFP signal delimited tubular structures and colocalized with the fluorescence of Chlorophyll, which is inserted in thylakoids membranes. Most importantly, this tubular labelling seemed stimulated under high light (HL). These results suggest that CrFzl may be recruited to thylakoid membranes and that this recruitment would be stimulated upon high light stress.
To confirm this possibility, we reasoned that the dynamic recruitment of CrFzl onto intra-plastidial tubules may require integrity of the distinct sub-domains of the DRP. We thus began by evaluating the capacity of CrFzl mutants (Fig 5b) to rescue the strong phenotypes generated by the absence of CrFzl. Wild-type CrFzl-3HA or CrFzl-3HA mutant versions with a substitution of a methionine for the conserved lysine 446 within the G1 section of the GTPase domain (CrFzl-G1*-3HA), a deletion of the predicted hydrophobic domain (CrFzl-ΔHD-3HA) or a deletion of the C-terminal coiled-coil domain (CrFzl-ΔCC-3HA) were expressed in ΔCrFzl cells (Fig 5c and 5d). The resulting strains were then monitored for their growth on complete media under low light (TAP) or minimal media under high light (TP) as compared to the WT, ΔCrFzl and CrFzl-3HA control strains (Fig 5e). While the expression of wild-type CrFzl-3HA or CrFzl-ΔCC-3HA totally corrected the growth defect of ΔCrFzl on TP media under high light, the GTPase and the ΔHD mutants lost this rescuing capacity. Notably, the same conclusions were obtained by assessing the occurrence of fluorescent dots to monitor CrFzl functionality (S4b Fig). The integrity of the GTPase domain and HD region of CrFzl are thus essential for the function of the DRP whereas the predicted C-terminal CC domain is dispensable.
These findings led us isolating the thylakoid fractions from CrFzl-3HA, CrFzl-G1*-3HA and CrFzl-ΔHD-3HA cells that were grown under normal light (NL) or under high light (HL) in TAP medium. We then quantified the amount of CrFzl remaining on each thylakoid fractions before and after washing with 500 mM NaCl (Fig 5f). Importantly, this approach revealed that the binding of WT CrFzl-3HA to thylakoids is stronger under high light (50% remaining) than under low light (35% remaining), thus confirming that the recruitment of CrFzl to thylakoids is stimulated upon strong illumination (Fig 5g; CrFzl-3HA). More surprisingly, we observed that while the amount of CrFzl-ΔHD-3HA associated with intra-plastidial membranes was not significantly modified by light (Fig 5g; CrFzl-ΔHD-3HA), that of CrFzl-G1*-3HA increased by more than 35% upon strong illumination (Fig 5g; CrFzl-G1*-3HA).
These observations are consistent with the recruitment of CrFzl on thylakoids upon high light stress. In addition, the hydrophobic and GTPase domains would regulate this recruitment to preserve the integrity of thylakoids, possibly by triggering their fusion.
To directly assess the capacity of CrFzl to mediate fusion between thylakoids, we turned to the only thylakoid fusion assay established to date [39]. This assay requires two distinct strains with opposing mating types (mt) in which the photosynthetic activity is abolished: the mt+ Photosystem I mutant F15 (hereafter referred as ΔPSI) and the mt- Photosystem II mutant Fud34 (ΔPSII). ΔPSI is a mutant of the nuclear gene TAB1, whose absence prevents the translation of the PSI subunit PsaB from the plastid genome (Fig 6a and 6b, top schemes) [40,41]. ΔPSII is a mutant of the plastid DNA in the 5’UTR of the gene encoding the PSII subunit PsbC, which prevents its translation (Fig 6a and 6c, top schemes) [42]. When ΔPSI is crossed with ΔPSII, the chloroplasts of each mutant fuse together and functional complementation in the zygote allows recovery of photosynthetic activity by de novo synthesis of PSI and PSII subunits. In the presence of chloramphenicol (CAP), de novo synthesis of PSI and PSII subunits is prevented and the recovery of photosynthetic activity can be achieved if and only if the chloroplasts from parental strains mix their respective contents after fusion and their thylakoids then fuse together and mix their pre-existing PSI and PSII complexes (Fig 6d, top scheme).
The photosynthetic activity in parental strains and zygotes can be assessed as previously described by measurement of chlorophyll fluorescence induction kinetics [39]. In a wild-type strain, chlorophyll fluorescence progressively rose upon continuous illumination due to PSII photoreduction, and then decreased upon reoxidation because electrons were transferred toward PSI (Fig 6a, bottom graph). In the PSIΔ strain, chlorophyll fluorescence was induced as in the wild-type but did not decrease because no PSI can receive the electrons from PSII (Fig 6b, bottom graph). In the PSIIΔ strain, chlorophyll fluorescence was low but readily saturated because emitted photons cannot get derived to PSII (Fig 6c, bottom graph). After mating of ΔPSI and ΔPSII strains in the presence of CAP, the progressive increase in fluorescence and its subsequent decrease were detected as in wild-type cells because thylakoid fusion occurred (Fig 6d, bottom graph). We also mixed ΔPSI and ΔPSII strains and took fluorescence measurements before mating to mimic a lack of thylakoid fusion (Fig 6e, top scheme). In this context, the progressive increase occurred but no subsequent decrease in fluorescence could be detected (Fig 6e, bottom graph).
After confirming that chlorophyll fluorescence induction kinetics is a reliable readout to assess photosynthesis activity and thylakoid fusion, we repeated these experiments by crossing ΔPSI and ΔPSII strains with ΔPSII ΔCrFzl and ΔPSI ΔCrFzl strains, respectively (Fig 6f, first and second columns). The crosses were performed in the presence of CAP to assess thylakoid fusion but also in its absence to monitor mating efficiency by functional complementation. Photosynthesis was rescued in each situation indicating efficient mating (-CAP) and functional homotypic fusion of chloroplast and thylakoid membranes (+CAP). In striking contrast, we observed that upon crossing of ΔPSI ΔCrFzl with ΔPSII ΔCrFzl, photosynthesis was rescued in the absence but not in the presence of CAP (Fig 6f, third column and S5 Fig). This indicates that CrFzl is not required for mating or for fusion of chloroplasts but is required for the fusion of thylakoids.
Electron microscopy analysis previously revealed that the thylakoids of ΔPSI and ΔPSII gametes have very discernable characteristics in their stacking properties [39]. In agreement with these observations, we easily noticed that the thylakoids of ΔPSI cells (14 cells; S1 Table) are highly stacked and electron-dense (Fig 7a, S6a and S7a Figs) whereas those from ΔPSII cells (6 cells; S1 Table) are much less stacked with more discernable lumens (Fig 7b, S6b and S7b Figs). Interestingly, the ΔPSI ΔCrFzl and ΔPSII ΔCrFzI gametes kept these differences but also accumulated additional specific features. Together with their highly stacked thylakoids, all the ΔPSI ΔCrFzl gametes that could be analyzed (34 cells; S1 Table) presented a significant increase in starch granules (Fig 7c, S6c and S7c Figs), consistent with the phenotype of vegetative ΔCrFzl mutants (Fig 4a). In contrast, starch granules did not accumulate in the ΔPSII ΔCrFzl gametes (12 cells; S1 Table). However, all cells had thylakoids that were less stacked but also curved (Fig 7d). Remarkably, this curvature often (9 cells out of the 12; S1 Table) resulted in thylakoids organized in concentric stacks similar to fingerprints or onion rings (S6d and S7d Figs). Intriguingly, this phenotype is reminiscent of the fuzzy onion mitochondrial morphology previously observed in early post-meiotic spermatids from mitofusin mutant flies [43].
All the zygotes resulting from ΔPSI and ΔPSII (Fig 7e, S6e and S7e Figs; 23 zygotes observed, S1 Table) or ΔPSI ΔCrFzl and ΔPSII (Fig 7f, S6f and S7f Figs; 4 zygotes observed, S1 Table) crossings in the presence of CAP displayed both highly (black arrowheads) and less stacked thylakoids (orange arrowheads). Nonetheless, the two populations were organized in obvious continuity yielding long and properly assembled thylakoid networks, suggesting efficient thylakoid fusion as expected from fluorescence induction kinetics measurements (Fig 6f). On the other hand, in the absence of CrFzl (10 zygotes observed, S1 Table), the network of intra-plastidial membranes was disorganized (Fig 7g, S6g and S7g Figs). In all zygotes, highly stacked thylakoids (black arrowheads) were mixed with less stacked thylakoids (orange arrowheads) but any continuity between both populations was not discernable. This together with fluorescence measurements (Fig 6f) supports that CrFzl is required for the fusion of thylakoids.
Previous studies performed in the plant model Arabidopsis thaliana demonstrated that the absence of Fzl is responsible for a slow growth phenotype associated with pale green or partly chlorotic leaves [20,22], enlarged chloroplasts containing morphologically altered thylakoids [20] as well as activation of autophagy and cell death pathways [22,23]. In addition, photosynthetic impairment was observed in pale green leaves [21]. However, these phenotypes were obtained from distinct mutants and ecotypes which hampered the establishment of any mutual causality between the heterogeneous defects caused by the absence of Fzl expression. Moreover, the primary role of Fzl in maintenance of chloroplast and thylakoid morphology was clearly demonstrated but its precise function in the remodeling of plastidial membranes remained to be addressed. We thus aimed at clarifying the function of Fzl by shifting its analysis in the unicellular microalgal model Chlamydomonas reinhardtii.
Our data confirmed the expression and plastidial localization of CrFzl (Fig 1). Using the CRISPR Cas9 technology in Chlamydomonas reinhardtii, we generated ΔCrFzl strains with a surprisingly high efficiency (Fig 2) when compared to previous results [31], which might be explained by the absence of cell wall in our recipient strain, our much shorter donor DNA or the nature of our target gene. Phenotypic analysis revealed a strong sensitivity of the mutants to high light stress with decreased resistance to photoinhibition of photosynthesis, which was rescued by antioxidants (Fig 3). An intriguing observation was the increased sensitivity of the CrFzl mutant in mixotrophic (TAP) as compared to photoautotrophic (TP) conditions upon high light stress (Fig 2i and 2j). Interestingly, LHCSR3, one of the major energy dissipating protein in stressful conditions, is far less expressed in mixotrophic than in photoautotrophic conditions upon high light [44]. Whether ΔCrFzl cells have a distinct ability to dissipate excess energy according to their growth in TAP or in TP could thus be tested in the future.
In response to high light treatment, the morphology of ΔCrFzl chloroplasts was also altered with increased starch granules inserted between thylakoids, a marked accumulation of cytosolic vesicles in close physical contact with plastids and an activation of autophagy (Fig 4). This overall sensitivity to light stress allows reconciling the distinct phenotypes previously observed in Arabidopsis thaliana. The absence of CrFzl which causes light stress impairment of photosynthesis is likely associated with elevated reactive oxygen species that induce photo-oxidative damage to thylakoid membranes [34]. This extreme damage would then promote the selective degradation of the chloroplast content by cytoplasmic vacuoles and ultimately cell death, upon prolonged exposure to high light. Most importantly, the defective resistance to photo-inhibition that induces this cascade of events may result from abrogation of CrFzl primary function in mediating fusion of thylakoids.
We should at this point not exclude that CrFzl could modulate membrane organization and stabilization but our work indicates that Fzl may trigger the fusion of thylakoid membranes. In conditions where thylakoid fusion is required, zygotes from a cross between ΔPSI and ΔPSII strains lacking CrFzl failed to recover photosynthetic activity (Fig 6) and presented disorganized thylakoid networks (Fig 7). Conversely, when CrFzl was expressed in only one of the two strains to be mated, ΔPSI and ΔPSII thylakoids remained competent for mixing their membrane content as attested by their recovered photosynthetic activity (Fig 6) and the proper morphology of their thylakoid networks (Fig 7). The most straightforward interpretation for this thylakoid fusion occurring after mating of CrFzl positive and negative strains relies on our data suggesting that CrFzl is soluble in the stroma of the chloroplast (Fig 1). In this context, the CrFzl protein expressed in one of the two chloroplasts could diffuse and bind thylakoids of the ΔCrFzl strain. The DRP present on both sets of thylakoids would then promote their tethering and subsequent fusion. Yet, fusion DRPs such as mitofusins or OPA1 are integral to mitochondrial outer and inner membranes, which contrasts with CrFzl being essentially soluble in the stroma of the chloroplast. In fact, such feature is actually shared by NpBDLP, for which insertion into membranes is triggered by a GTPase domain-dependent conformational switch [12,13]. NpBDLP includes a membrane paddle instead of a regular transmembrane domain. In the apo or GDP bound form of BDLP, this paddle is not competent for membrane binding but becomes accessible upon the binding of GTP that induces the conformational switch of the DRP. Consistent with its essential GTPase and HD domains, CrFzl may function in a similar fashion and our data further suggest that the targeting to thylakoid membranes is stimulated upon light stress (Fig 5). We found that the GTPase and HD region both regulate this light-dependent recruitment of CrFzl to thylakoids (Fig 5g). Deletion of the HD inhibited the association of CrFzl to membranes under high illumination. While this likely reflects the requirement of the HD for CrFzl recruitment to membranes, we cannot exclude that this effect was caused by a disturbed structure of the mutant. On the other hand, the G1 mutation surprisingly stimulated the binding of CrFzl to thylakoids upon high light. Whether this G1 mutation (K448M) abolishes the binding of GTP or abrogates GTP hydrolysis is currently unclear. However, similar to BDLP [12,13], binding of GTP in the absence of hydrolysis may maintain CrFzl in a conformation competent for stable membrane association. Moreover, since the G1 mutant only responded to high but not low illumination (Fig 5g), an additional light stress-dependent signal may participate in the recruitment of CrFzl to thylakoids. In this regard, CrFzl has recently been identified as a thioredoxin target [45], suggesting that CrFzl-mediated membrane fusion induced by high light treatment may be redox regulated by the opposing actions of ROS and thioredoxin, as previously established for the regulation of autophagy [46,47].
Intriguingly, obvious phenotypes in thylakoids from Crfzl mutants were not observed in vegetative conditions. In this regard, Fzl-mediated fusion may occur in between thylakoid stacks [24]. This may explain that defects in thylakoid organization caused by the absence of CrFzl could only be observed in cells with weak thylakoid stacking (compare thylakoids from Figs 4d and 7c Vs those from Fig 7d and S6d Fig). The mating assay we employed indicates that zygotes lacking CrFzl do not fuse their thylakoids. In this context, fusion of thylakoids is not essential for growth in “normal” conditions but becomes required during light stress. This echoes the stress-related function attributed to the NpBDLP paralogue of Bacillus subtilis, DynA, which has been implicated in the repair of membrane pores formed by antibiotics [48]. In the case of CrFzl, promoting fusion between two distinct thylakoid saccules during light stress could participate in driving a continuous redistribution of functional photosystems between thylakoids, thus maintaining homogenous photosynthetic electron transfer. In this scenario, CrFzl would avoid the accumulation of thylakoids with very low photosynthetic efficiency that may otherwise produce significant amounts of ROS. The absence of thylakoid fusion may thus lead to such damages that the cells would induce the selective degradation of chloroplast material to cope with further insults resulting from photosynthetic dysfunction. The cytoplasmic vesicles accolated to chloroplasts (Fig 4, S3 Fig) and the activation of autophagy (Fig 4g) that we observed in this study give credence to this possibility. However, the mechanism of formation of these vesicles and their relationship to autophagy remains obscure, which warrants future investigations.
While CrFzl triggers thylakoid fusion during light stress, outer and inner envelops of chloroplasts must also fuse during mating. EZY8 (Early Zygote 8; Cre06.g250650) represents a good candidate for this function as microarray and RNA-Seq studies detected increased expression of this DRP during early stages of zygote development [49,50]. In parallel, Fzl proteins are the only mitofusin-like proteins in plants and this study further confirms that they do not participate in mitochondrial dynamics. Yet, several studies have demonstrated that mitochondrial fusion takes place in plant cells and microalgae [51,52,53]. Future discoveries will likely resolve this fascinating issue and extend the contribution of DRPs to the control of chloroplast membrane dynamics.
The strain used for initial expression of CrFzl and fluorescence subcellular localization was CC-4533 [27]. The strain used for subcellular fractionation and mutagenesis and as a wild-type control in all following experiments was cw15.J3 (mt-). As a control for CrFzl knock-out in the screening of CRISPR Cas9 generated strains, the mutant LMJ.RY0402.175738, containing a paromomycin cassette insertion in the CrFzl locus, was obtained from the CLiP library [27]. Cells were maintained on TAP agar 1,5% containing the appropriate antibiotic if necessary, and inoculated in TAP for preculture prior to any experiment. Unless otherwise stated, cultures were grown in continuous light at 50 μmol photons.m−2·s−1, 26°C on a rotary shaker at 120 rpm. For minimal medium cultures, the TP medium is the same composition as TAP but with pH adjusted with HCl instead of acetate.
Crosses required to obtain ΔPSI ΔCrFzl mt+, ΔPSII mt- and ΔPSII ΔCrFzl mt- strains were performed according to [25] and are described in the “Thylakoid fusion assay” section of materials and methods.
Chloroplast transit peptide length was predicted using Wold PSort [54]. CrFzl sequence was blasted in SMART [55] for subdomain prediction and Phobius [56], TMHMM2.0 [57], CCTOP [58], PredictProtein [59] and TCDB [60], for transmembrane subdomain prediction. Three out of 5 online programs predicted two transmembrane domains (CCTOP: 921–940 and 948–965; TCDB: 922–938 and 948–968; PredictProtein: 924–943 and 947–959) while Phobius only predicted one (948–967), and TMHMM2.0 none. Protein sequences for different DRPs have been collected from Phytozome (for Arabidopsis and Chlamydomonas proteins), SGD (for Yeast) and UniProt (identifiers: Ns_BDLP, B2IZD3; Hs_Mfn1, Q8IWA4; Hs_dynamin, Q05193; Hs_dynamin-like, O00429; Hs_MxA, P20591; Hs_OPA1, O60313; Hs_GBP1, P32455). GTPases domains were delimitated using SMART prediction and all remaining sequences were analyzed on ClustalW online Software.
Cloning of Crfzl locus into pSL26 was done using NEBuilder HiFi DNA Assembly Cloning Kit with 4 fragments. Backbone vector was produced by restriction of pSL26 with NdeI/NruI. 5’ and 3’ “adaptor” fragments were synthesized by PCR using the primer pairs CrFzl_ATG_For/CrFzl_ATG_Rev and CrFzl_STOP-HA_For/CrFzl_STOP_HA_Rev, respectively, and the base pairs from +711 to +6199 of the CrFzl locus were excised from BAC vector PTQ13345 using a SphI/BciVI restriction reaction. The same strategy was adopted for cloning into pLM005-CrVenus vector, using different 5’ forward and 3’ reverse primers containing sequence homologous to the backbone vector (See S2 Table).
For construction of the mutated CrFzl versions, the plasmid for expression of the full length CrFzl-3HA was digested with AgeI (insertion of the GTPase domain mutation), or MluI/NruI (deletion of the TM and CC domains). The CrFzl region containing the G1-box of the GTPase domain was amplified with the nucleotide substitution A to T at position +2709 (AAG→ATG) using the primers pairs PreG1_For/Pre-G1_Rev and Post-G1_For/Post-G1_Rev. 3’ part of the gene without the TM regions was amplified with Pre-TM_For/Pre-TM_Rev and Post-TM_For/Post-TM_Rev. The ΔCC fragment was produced using the pair Pre-TM_For/ΔCC_Rev.
Cells in mid-log phase were centrifuged and resuspended in TAP medium containing 40 mM sucrose at a density of ~2.108 cells.mL-1. 250 μL aliquots were transferred to 4 mm electroporation cuvettes and incubated for 20 mn on ice. 500 ng of linearized plasmid were added and the suspension was homogenized by flicking the cuvette. A time constant electric pulse of 800V was then applied (11 ms, 800 V, Xcell Pulser Electroporation System, Biorad). Cells were left for 5 mn at room temperature, diluted in TAP 40 mM sucrose and incubated for a 18 hours recovery in low light with agitation. After recovery, the cells were pelleted, resuspended in fresh TAP sucrose and plated on TAP medium containing paromomycin (10 μg.mL-1). Resistant clones were allowed to grow for 6 days before picking, ordering and screening.
One OD750nm (1 mL) of cells was lyzed with 100 μL of 1 M NaOH for 10 mn on ice then proteins were precipitated by addition of 100 μL of 50% TCA on ice for 30 mn. Proteins were then pelleted, resuspended in Sample Buffer (1.6 mM EDTA, 1.6% SDS, 40 mM DTT, 8% Glycerol, 0.016% Bromophenol Blue, 333 mM TrisBase) and solubilized at 70°C for 10 mn prior to loading. After SDS-PAGE, proteins were transferred onto a nitrocellulose membrane and detected with specific commercial (anti-RbcL, anti-NAB1 and anti-βCF1 from Agrisera, anti-HA [clone 12ca5] and anti-Tub from Sigma-Aldrich) and custom made (anti-CrFzl from Covalab) primary antibodies. Unless otherwise stated, 10 μL of protein extract were loaded and Tubulin antibodies served as loading control.
Chloroplast isolation was performed as described in [26] with some modifications. Briefly, mid-log phase cells (OD750nm ~ 0,8) were concentrated to 0.7 mg chlorophyll per mL in isolation buffer (HEPES 50 mM, Sorbitol 300 mM, EDTA 2 mM, MgCl2 1 mM, pH 7,5). Cells were then broken by passing through a 27 gauge needle with a flow of ~0.1 mL.s-1. Unbroken cells and chloroplasts were pelleted by centrifugation at 800 g, 4°C for 5 mn. Supernatant was collected as the cytosolic fraction. The pellet was gently resuspended in isolation buffer using a fine paintbrush. The chloroplast suspension was then layered on top of a 3 cautions Percoll gradient (20%/45%/65%) and centrifuged at 5525 g, 4°C for 30 mn with slow acceleration and no brake. While unbroken cells cross the three cautions and pellet at the bottom of the gradient, the thylakoid enriched fraction was recovered at the 20/45% Percoll interface and intact chloroplasts were harvested from the 45/65% interface. Both fractions were washed from Percoll in 50 mL isolation buffer and centrifuged at 700 g, 4°C for 3 mn. Pelleted chloroplasts were collected and frozen while the thylakoid fraction was split in four sub-fractions. Those sub-fractions were further washed and pelleted twice in 5 mL HEPES 50 mM, Sorbitol 300 mM containing 0, 50, 150 or 500 mM NaCl, respectively; and centrifuged at 2759 g, 4°C for 10 mn. Thylakoid pellets were finally frozen. Proteins were then extracted as described above (Western blot) and loading was adapted to obtain equivalent amounts of thylakoid protein marker (β-CF1) in the total, chloroplast and thylakoid fractions.
For membrane/soluble fractions separation, cells were broken in HEPES 50mM by ten cycles of 10 s shaking/15 s on ice with glass beads (diameter 425–600 μm) in a cold room. Unbroken cells were pelleted by centrifugation at 800 g, 4°C for 5 mn. The supernatant was then further centrifuged at 21000 g, 4°C for 30 mn. Resulting supernatant containing soluble proteins and pellet containing membrane proteins were harvested and proteins extracted as described above.
After treatment, cells were fixed by adding formaldehyde at a final concentration of 3,7% and incubated at room temperature with agitation for 30 mn. Cells were mounted between glass slide and coverslips after 2 subsequent washes with TAP medium. Imaging was performed with a Zeiss Axio Observer.Z1 microscope (Carl Zeiss S.A.S.) with a X100 oil immersion objective equipped with the filter sets 10 Alexa Fluor (Excitation BP 450/490, Beam Splitter FT 510, Emission BP 515–565), DAPI (Excitation BP 359–371, Beam Splitter FT 395, Emission BP 397-∞), 47 HE CFP (Excitation BP 424/448, Beam Splitter FT 455, Emission BP 460–500), 46 HE YFP (Excitation BP 488–512, Beam Splitter FT 515, Emission BP 520–550) and Chloro (Excitation BP 450/490, Beam Splitter FT 505, Emission BP 600-∞). Cell contours were visualized with Nomarski optics. Images were acquired with an ORCA-R2 charge-coupled device camera (Hamamatsu) and analyzed with ImageJ.
Custom made antibodies were ordered from Covalab. Anti-CrFzl antibodies were produced by immunization of a rabbit with two synthesized peptides (peptide 1 [289–304]: C-FDLAENATAEDYAQA-coNH2 and peptide 2 (727–741): C-GRQLGRFRAEMEKDA-coNH2).
Protocol for targeted disruption of Cre14.g616600 locus was adapted from Shin et al., 2016 [31]. Cells were transformed as described above. Prior to the electric pulse, sgRNA (33,33 μg.mL-1) and purified Cas9 protein (25 μg.mL-1, Labomics), preincubated at 37°C for 30 min, and a DNA cassette (1 μg.mL-1) containing the aph7” gene conferring resistance to hygromycin, were added to the 250 μL aliquots. After recovery, the cells were pelleted, resuspended in fresh TAP sucrose and plated on TAP medium containing hygromycin (20 μg.mL-1).
The hygromycin cassette was amplified from the pSLHyg plasmid using the primers HygFor and HygRev, and the Phusion High-Fidelity polymerase (NEB) following the manufacturer’s instructions with an annealing temperature of 70°C and 40 s of extension for 35 cycles.
Hygromycin resistant clones were screened for targeted cassette insertion using a 3 primers PCR strategy. After quick DNA extraction by boiling in 50 μL of 10 mM EDTA, 1 μL of supernatant was used for PCR reaction with primers CrFzl_ATG_For, CrFzl_ATG_Rev and Hyg_500_Rev with hybridization temperature set at 60°C and polymerization allowed for 90 s. PCR conditions are set such as three possible products are expected depending on the insertion of the cassette in the GOI and its orientation: 760 bp from the pair CrFzl_ATG_For/CrFzl_ATG_Rev, 1246 bp from CrFzl_ATG_For/Hyg_500_Rev and 715 bp from CrFzl_ATG_Rev/Hyg_500_Rev.
For spot tests, Chlamydomonas cells in log-phase where harvested by centrifugation at 1500g for 5 mn. They were then washed one time with minimal medium before being resuspended at 2 OD750nm/mL. Drops (3 μL) were deposited on minimal TP and enriched TAP media and grown at 50, 100 and 200 μmol photons.m−2·s−1 for 5 days.
For liquid cultures, mid-log phase cells were inoculated in fresh TP or TAP medium at a starting OD750nm of 0.1. Growth in darkness and at 50 and 400 μmol photons.m−2·s−1 was monitored by measuring optical density at 750 nm.
Chlamydomonas mid-log cells were harvested and Rhodamine 123 added at a final concentration of 25 μM. Cells were incubated for 25 mn at RT, washed twice and concentrated in TAP medium before observations.
For photoinhibition measurements, mid-log phase cells from wild-type and mutant background were diluted in TAP or TP medium at OD750nm = 0.15, left for 2 hours to adapt to the fresh medium and exposed to high light (850 μmol photons.m−2·s−1) for 1 hour. They were then left for recovery in dim light (< 5 μmol photons.m−2·s−1) for 3 h. PSII efficiency (Fv/Fm) was measured every 30 mn for the first two hours and every hour thereafter.
For detection of fluorescent dots and fluorescence microscopy study of CrFzl membrane association, cells inoculated at an OD750nm = 0.15 were incubated for 6 hours at 400 μmol photons.m−2·s−1. Treated cells were immediately fixed following light stress as described previously.
For the biochemical study of CrFzl membrane association, cells were grown to OD750nm = 0.4 and treated for 6 hours at 800 μmol photons.m−2·s−1 before proceeding to cellular subfractionation.
Fluorescence measurements were performed using a DeepGreen Fluorometer (JBeamBio). PSII efficiency, Fv/Fm (Fv = Fm − F0), was determined using shortly (1 mn) dark-adapted cells. F0 is the initial chlorophyll fluorescence measured after dark adaptation and Fm is the maximum fluorescence measured after a brief and saturating flash of green light [61,62].
Cells of wild-type and mutant background were inoculated at OD750nm = 1 in TAP medium containing 0, 1, 2.5 or 5 mM N-acetyl cysteine (NAC; buffered with 100 mM HEPES, pH 7.5); and treated for 24 hours with high light stress at 800 μmol photons.m−2·s−1. As a control, the NAC-free cultures were duplicated and grown at normal light conditions (50 μmol photons.m−2·s−1).
Cells incubated in darkness, normal or high light for 6 hours, as well as gametic cells and zygotes from the crosses were concentrated, deposited on polylysine-coated cover slips, fixed with a mixture of 2% (wt/vol) paraformaldehyde, 1% (wt/vol) glutaraldehyde in 0.2 M phosphate buffer (PB), pH 7.4, post-fixed with 1% (wt/vol) OsO4 supplemented with 1.5% (wt/vol) potassium ferrocyanide, dehydrated in ethanol, and embedded in epon 812 (TAAB Laboratories Equipment). Ultrathin sections were prepared with a Reichert UltracutS ultramicrotome (Leica), counter-stained with uranyl acetate, and viewed at 80kV with a Transmission Electron Microscope (TEM; Tecnai Spirit G2; Thermo Fisher Scientific, Eindhoven, The Netherlands) equipped with a QUEMESA CCD 4K camera (EMSIS GmbH, Münster, Germany) using iTEM software (EMSIS).
As a control of ATG8 induction, the wild-type cw15 strain (OD750nm = 0.25) was treated with 0.5 and 1 μM rapamycin for 16 hours. In parallel, wild-type and ΔCrFzl strains were treated for 6 hours at 400 μmol photons.m−2·s−1 at OD750nm = 0.15. Proteins were then extracted as described in [63]. 20 μg of proteins were loaded for each sample.
Strains required for this assay were obtained by cross the ΔPSI mutant F15 (mt+) and the ΔPSII mutant Fud34 (mt+) with the ΔCrFzl strain (mt-). Zygotes were germinated in TAP sorbitol containing 2% sorbitol in dim light for 24 hours. Aliquot were then spread on agar plates containing Hygromycin B (20 μg.mL-1) to select for strain defective for CrFzl. Single colonies were then picked and arrayed on new agar plates before measurements of chlorophyll induction kinetics. Clones displaying a fluorescence kinetic typical of ΔPSI or ΔPSII mutants were selected for mating type identification by PCR [64].
For the thylakoid fusion assay, induction of gametogenesis was done by an overnight incubation in nitrogen deprived medium under dim light. For each cross, parental gametes were mixed in equivalent amount of cells to reach a total of ~1,5.107 cells.mL-1. Chloramphenicol was added simultaneously at a concentration of 100 μg.mL-1. For antibiotic-free control crosses, ethanol, in which CAP is solubilized, was added to a final concentration of 0,3%. Crosses were left to process at 50 μmol photons.m−2·s−1 without agitation for 48 hours which represents the light phase of zygote development. The zygote pellicle deposited at the bottom of Petri dishes was washed from gametic cells and the zygotes resuspended in water containing 2% sorbitol with or without CAP prior to fluorescence measurements, which were performed using a DeepGreen Fluorometer after dark adaptation of the cells (1 mn).
|
10.1371/journal.pntd.0002129 | Investigating the Role for IL-21 in Rabies Virus Vaccine-induced Immunity | Over two-thirds of the world's population lives in regions where rabies is endemic, resulting in over 15 million people receiving multi-dose post-exposure prophylaxis (PEP) and over 55,000 deaths per year globally. A major goal in rabies virus (RABV) research is to develop a single-dose PEP that would simplify vaccination protocols, reduce costs associated with RABV prevention, and save lives. Protection against RABV infections requires virus neutralizing antibodies; however, factors influencing the development of protective RABV-specific B cell responses remain to be elucidated. Here we used a mouse model of IL-21 receptor-deficiency (IL-21R−/−) to characterize the role for IL-21 in RABV vaccine-induced immunity. IL-21R−/− mice immunized with a low dose of a live recombinant RABV-based vaccine (rRABV) produced only low levels of primary or secondary anti-RABV antibody response while wild-type mice developed potent anti-RABV antibodies. Furthermore, IL-21R−/− mice immunized with low-dose rRABV were only minimally protected against pathogenic RABV challenge, while all wild-type mice survived challenge, indicating that IL-21R signaling is required for antibody production in response to low-dose RABV-based vaccination. IL-21R−/− mice immunized with a higher dose of vaccine produced suboptimal anti-RABV primary antibody responses, but showed potent secondary antibodies and protection similar to wild-type mice upon challenge with pathogenic RABV, indicating that IL-21 is dispensable for secondary antibody responses to live RABV-based vaccines when a primary response develops. Furthermore, we show that IL-21 is dispensable for the generation of Tfh cells and memory B cells in the draining lymph nodes of immunized mice but is required for the detection of optimal GC B cells or plasma cells in the lymph node or bone marrow, respectively, in a vaccine dose-dependent manner. Collectively, our preliminary data show that IL-21 is critical for the development of optimal vaccine-induced primary but not secondary antibody responses against RABV infections.
| Over two-thirds of the world's population lives in regions where rabies is endemic, resulting in over 15 million people receiving post-exposure treatment. A person, disproportionately a child, dies of rabies every 20 minutes and the cost of rabies prevention exceeds $1 billion US dollars per year. The development of a single-dose human rabies vaccine would greatly reduce the burden of rabies globally by lowering the cost associated with rabies vaccination and saving lives. Understanding how B cells develop to produce protective virus neutralizing antibodies would greatly help to achieve the goal of developing a single-dose vaccine. In this report, we show that IL-21 is critical for the induction of primary vaccine-induced anti-RABV G antibody titers and that the effects of IL-21 are highly dependent on the dose of vaccine administered. In our model of rabies immunogenicity and protection, the lack of IL-21 receptor influenced the detection of B cells in germinal centers in lymph nodes or of plasma cells in bone marrow after immunization with low or high doses of vaccine, respectively. Overall, these preliminary results indicate that IL-21 has the potential to influence B cell development and functions in the context of rabies vaccine-induced immunity and protection.
| RABV is a single-stranded negative sense RNA virus of the genus lyssavirus in the Rhabdoviridae family that kills approximately 55,000 people annually. Up to 60% of rabies cases are in children, making rabies the seventh most important infectious disease in terms of years lost [1]. In Africa, a person dies of rabies every 20 minutes [2]. In China, rabies became the leading cause of infectious disease mortality in 2006, which increased by more than 27% from 2005 [3]. In the United States, cases of rabies in wildlife are detected in virtually all states and Puerto Rico (Hawaii is considered rabies-free). Except for cattle and foxes, the incidence of rabies in domesticated or wildlife remained unchanged or significantly increased in the US in 2011 compared to the five-year average for each species [4], exemplifying the difficulty in containing zoonotic viral infections even in industrialized nations. The cost associated with rabies in the US, Africa and Asia is almost $1 billion annually [5], [6] contributing to the financial burden of global health care costs. Furthermore, rabies is a NIAID Category C Priority Pathogen, indicating rabies is an emerging infectious disease with the potential for mass dissemination and harm to people [7]. Together, rabies is considered a neglected global zoonotic infectious disease that disproportionately affects children and, therefore, understanding how B cells develop in response to experimental RABV-based vaccination may help to support efforts to develop a single-dose human rabies vaccine for use in both developing and industrialized countries.
A wide array of RABV variants exist, ranging from highly pathogenic strains to attenuated RABV vaccine strains such as the molecular clone SAD B19 [8]. Live attenuated RABV vaccine strains are highly immunogenic and potentially could serve as a single-dose human RABV vaccine to replace currently used multi-dose inactivated RABV-based vaccine regimens. Due to residual pathogenicity of these live virus strains, however, several “second-generation” RABV-based vaccines are under investigation in which entire genes are deleted from the RABV genome [9]–[12], or multiple pathogenic markers are genetically modified [13]. Data from these studies indicate that very safe and effective live RABV-based vaccine vectors can be generated. Despite extensive efforts to attenuate live RABV-based vaccine vectors for safety, little information is available on factors that influence the generation of effective antibodies in response to live RABV-based vaccines. Virus neutralizing antibodies (IgG but not IgM) directed against the RABV glycoprotein (G) are protective against pathogenic RABV strains [14], [15]. In the case of a replication-deficient RABV-based vaccine in which the matrix gene is deleted, VNAs are generated by T cell-independent and –dependent (extrafollicular and germinal center) mechanisms [16], suggesting multiple pathways of B cell activation and differentiation could be exploited to rationally design a single-dose RABV vaccine for use in both pre- and post-exposure settings. With respect to typical vaccine-induced antibody responses, APC-primed T cells most likely display an intermediate Tfh phenotype (i.e., “pre-Tfh cell”) characterized phenotypically as CD4+CXCR5hiPD1lo, which migrate the T and B cell border of secondary lymphoid organs and interact with their cognate antigen-primed B cells [17]. This T∶B cell interaction typically results in the Tfh cells producing optimal amounts of IL-21, and in the B cells differentiating into early short-lived extrafollicular antibody secreting cells or migrating into the follicles and forming GCs. With additional signals provided by Tfh cells in GCs, B cells mature and differentiate into long-lived plasma cells (PCs) secreting high affinity antibodies or into memory B cells. Due to the importance for PCs secreting high affinity antibodies and memory B cells in vaccine-induced immunity, the development of Tfh cells and CG B cells is critical for vaccine-induced protection against future exposures. In the context of RABV-specific vaccination in post-exposure settings, the rapid induction of extrafollicular B cell responses may also be critical to prevent infection of the CNS, especially in cases where treatment is delayed after exposure to a potentially infected animal. As such, understanding factors that generate short- and long-term anti-viral B cell responses will help design more efficacious RABV vaccines for use in humans.
Cytokines present at the time of antigen exposure influence T and B cell activation and GC formation and, therefore, also affect the outcome of vaccination. IL-21 [18], [19] is a type 1 cytokine that is a member of the common γ-chain receptor family, which also includes IL-2, IL-4, IL-7, IL-9, and IL-15. It is produced primarily by activated Tfh and Th17 cells and has pleiotropic effects throughout innate and adaptive immunity [reviewed in [20]]. The role for IL-21 in regulating Tfh and B cell functions was originally identified using model antigens [21]–[23]. In addition, the role for IL-21 in immunity and protection against helminth [24], viral [25], [26] and bacterial [22] infections has been studied. IL-21 is a key mediator for the control of persistent viral infections in mouse models of LCMV [27]–[29] and hepatitis B virus [30], or in humans infected with HIV [31]–[33] or HIV in combination with the 2009 H1N1 influenza virus vaccine [34]. However, the complexity and diversity of persistent infections makes it is difficult to pinpoint the effects of IL-21 on anti-viral B cells versus CD8+ T cells, both of which can contribute the control of many chronic infections [35]. Conversely, B cells play a critical role for clearance of most acute viral infections and for the efficacy of vaccines against most vaccine-preventable diseases. Clearance of RABV infections relies strictly on B cell-mediated effector functions, but not CD8+ T cells, for protection, making RABV infection an excellent mouse model to pinpoint the role for IL-21 in vaccine-induced immunity against RABV infections and potentially for other pathogens that rely solely on B cells for protection. In this report, our preliminary data indicate that IL-21 is critical for the development of effective vaccine-induced primary antibody responses against RABV infections by influencing GC B cells or PC generation in a vaccine dose-dependent manner, while also showing IL-21 is dispensable for RABV-specific secondary antibody responses when a primary antibody response develops.
All animal work was reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Jefferson Medical College, Thomas Jefferson University. Work was completed in accordance with international standards [Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC)] and in compliance with Public Health Service Policy on Humane Care and Use of Laboratory Animals, The Guide for the Care and Use of Laboratory Animals of the National Institutes of Health (NIH).
The construction of the live RABV-based vaccine (rRABV) used in this study was described elsewhere and was previously named SPBN [9], [36], [37]. This vaccine is a molecular clone derived from the attenuated SAD-B19 vaccine strain of RABV. Virus stocks were propagated on baby hamster kidney cells and then concentrated and purified over a 20% sucrose cushion. The challenge virus used was the pathogenic Challenge Virus Strain-N2c (CVS-N2c), which is a mouse-adapted sublclone of CVS-21 RABV [38]. CVS-N2c was initially propagated in neonatal mouse brains and then passaged once in-vitro on a neuroblastoma cell line (NA cells). The titer of CVS-N2c required to kill unvaccinated mice was determined experimentally by inoculating serial ten-fold dilutions into naïve immune-competent mice [39] i.m. and then observing mice daily for clinical neurological symptoms of rabies. The titer required to kill unvaccinated mice within 8 days post-infection, which is typical for CVS-N2c [9], [10], was determined to be 105 focus forming units (ffu)/mouse. Cryopreserved embryos of mice deficient in the IL-21 receptor (B6;129-IL21r tm1Wjl/Mmucd); #015505-UCD) [39] were obtained from the Mutant Mouse Regional Resource Centers (NIH) and implanted and bred in-house at Thomas Jefferson University in a pathogen-free animal facility. Control C57BL/6 mice were obtained from the Frederick National Laboratory for Cancer Research (NCI).
The following antibodies where purchased from BD Biosciences, unless otherwise noted, and used for flow cytometry staining: APC-Cy7-B220 (clone RA3-6B2), PerCP-CY5.5-CXCR5 (clone 2G8), APC-CD138 (clone 281-2), PE-Cy7-CD95/Fas (clone Jo2), FITC-GL7, eFluor 450-CD4 (clone RM4–5, eBioscience), PE-PD1 (clone J43, eBioscience), Alexa Fluor 700-CD38 (clone 90, eBioscience), rat anti-mouse CD16/32 (FcBlock; Pharmingen).
Groups of 6- to 10-week-old female IL-21R−/− or wild-type C57BL/6 mice were inoculated intramuscularly (i.m.) with 103 or 105 focus forming units (ffu)/mouse with rRABV or an equivalent volume of PBS as controls. Five weeks post-immunization, mice were challenged i.m. with 105 ffu/mouse with CVS-N2c and then observed for three weeks for clinical signs of rabies. Mice were euthanized at the onset of neurological symptoms. At various times post-immunization and challenge, blood was collected retro-orbitally. Three-fold serial dilutions of sera were tested by ELISA to determine RABV G-specific IgG antibodies as described [10] and reported as the reciprocal serial dilution. Data represented two independent experiments (N = 9–11 mice per group). To measure virus neutralizing antibody titers, the Rapid Fluorescent Foci Inhibition Test (RFFIT) was completed on pooled sera from two independent experiments as described previously [16], [40], [41].
Groups of 6- to 10-week-old female IL-21R−/− or wild-type mice were inoculated i.m. with 103 or 105 ffu/mouse with rRABV or PBS/naive as controls. Draining lymph nodes and bone marrow cells were collected 7 and/or 14 days post-immunization. Single cell suspensions (106 cells/sample) were incubated with rat anti-mouse CD16/32 (1 ug/106 cells) in fluorescence-activated cell sorter (FACS) buffer (PBS supplemented with 2% fetal bovine serum) for 1 h on ice. Cells were washed twice with FACS buffer and incubated with fluorescently conjugated antibodies (0.2 ug/106 cells) for 30 min at RT in the dark. Cells were subsequently washed 2 times with FACS buffer and fixed in 2% paraformaldehyde in PBS for 30 minutes. Flow cytometry was completed using FACScan (BD LSRII) and analyzed by FlowJo software. Data represents samples completed in duplicate (N = 3–5 mice) [16].
Kaplan-Meier survival curves were analyzed by the log rank test; *p = <0.05 indicates significant survivorship between two immunization groups [9], [10]. Statistical difference between two groups of data was compared using an unpaired, two-tailed t test and data is presented at the mean ± SEM; *p<0.05, **p = 0.01–0.001, ***p≤0.001 [9], [10].
Cytokines present at the time of immunization have the ability to affect the outcome of vaccine-induced B cell responses. Due to the importance for IL-21 in promoting effective T-dependent B cell responses [21], [22], we examined the requirement for IL-21R signaling in the generation of antibodies in a mouse model of RABV immunogenicity and protection using a mouse model of IL-21R-deficiency. These mice, designated here as IL-21R−/− mice, lack the IL-21R extracellular and transmembrane domains but show normal lymphoid development [39]. T and B cells exhibit similar proliferative responses to CD3-speficic antibodies or LPS, respectively, when compared to wild-type controls [39]. As a group, IL-21R−/− mice immunized with a low dose of a live recombinant RABV-based vaccine (rRABV) (103 ffu/mouse) showed only low levels of anti-RABV G antibodies that were not significantly different from PBS-immunized mice at all time points tested post-immunization (Figure 1, left panels, a–e), although at least three IL-21R−/− mice developed anti-RABV antibodies by day 28 post-immunization (Figure 1B). The seroconversion of these three IL-21−/− mice may explain the limited protection observed in the pathogenic challenge experiments described later in this report. Wild-type mice immunized with the same dose of rRABV showed significant levels of anti-RABV G antibodies as early as 7 days post-inoculation compared to rRABV-immunized IL-21R−/− mice, and these antibody responses continued to increase through day 21 post-immunization. VNA titers detected 28 days post-immunization (Figure 1C) are consistent with the antibody titers detected by ELISA (Figure 1B) indicating anti-RV antibody titers detected by ELISA are representative of the ability for vaccine-induced antibodies to neutralize rabies virus. IL-21R−/− mice immunized with a higher dose of rRABV (105 ffu/mouse) showed significantly reduced anti-RABV antibody responses when compared to wild-type mice also immunized with 105 ffu/mouse of rRABV (Figure 1, right panels, f–j). Together, this data indicates that IL-21 is critical for the induction of optimal primary anti-RABV antibody responses, especially when low doses of vaccine are used.
We next wanted to evaluate the effect of IL-21 on vaccine-induced antibody recall responses and protection against pathogenic RABV challenge. Five weeks post-immunization with rRABV, mice from Figure 1 were challenged with 105 ffu/mouse of a highly pathogenic mouse-adapted RABV strain (Challenge Virus Strain-N2c; CVS-N2c) [38], which typically kills naïve mice within 8 days post-infection [9], [10]. Consistent with the low antibody titers detected during the primary antibody response, significantly less anti-RABV antibodies were detected three or five days post-challenge in IL21R−/− mice immunized with 103 ffu/mouse of rRABV compared to the antibody recall response detected in wild-type mice (Figure 2, panels a and b, and Figure 2B) and antibody recall titers were not significantly different from PBS-immunized/CVS-N2c-challenged mice (Figure 2, e). Only 40% of IL-21R−/− mice immunized with 103 ffu/mouse of rRABV were protected against pathogenic RABV challenge, while all wild-type mice that were similarly immunized were protected against challenge (Figure 3, left panel). As expected, those mice with higher antibody titers showed protection compared to mice with lower antibody titers (Figure 2B). Conversely, an antibody recall response was induced in IL-21R−/− mice immunized with 105 ffu/mouse of rRABV within three days post-challenge with CVS-N2c at levels equivalent to rRABV-immunized wild-type mice (Figure 2, panels c and d) and all immunized IL-21R−/− mice were protected against pathogenic RABV challenge (Figure 3, right panel). Taken together, IL-21 is required for optimal primary (Figure 1) but not secondary (Figure 2) antibody responses to RABV vaccination. Furthermore, IL-21 is required for protection against pathogenic RABV in a vaccine dose-dependent manner (Figure 3).
Next we wanted to determine whether the impaired primary anti-RABV antibody response in IL-21R−/− mice was due to an overall defect in the generation of Tfh and/or GC B cells. Lymph nodes from IL-21R−/− or wild-type mice were collected 7 or 14 days post-immunization with 103 or 105 ffu/mouse of rRABV or PBS alone to determine the influence of IL-21 on Tfh and B cell populations. Representative gating strategies [17], [42] to identify CD4+ T cells from the total live lymph node cultures (Figure 4A) or Tfh (CD4+CXCR5hiPD1hi) cells from the CD4+ T cell populations (Figure 4B) are shown. A significant increase in the number of CD4+ T cells displaying a Tfh phenotype was detected in IL-21R−/− mice 14 days post-immunization with 103 or 105 ffu/mouse rRABV compared to similarly immunized wild-type mice (Figure 4C and Figure 4D, respectively). However, the formation of optimal GC B cells appears to be dependent on the dose of vaccine administered (Figure 5). Representative gating strategies [22], [43], [44] to identify B220+ B cells from the total live lymph node cultures (Figure 5A) or GC B cells (B220+GL7hiCD95/Fashi) from the B220+ B cell population (Figure 5B) are shown. IL-21R−/− mice immunized with a low dose of vaccine failed to induce optimal GC B cell formation compared to wild-type mice 14 days post-immunization, as shown by a significant decrease in the number of GC B cells in IL-21R−/− mice compared to wild-type mice (Figure 5C). However, a significant increase in the number of GC B cells was detected in IL-21R−/− mice immunized with 105 ffu/mouse of rRABV compared to wild-type mice 14 days post-immunization (Figure 5D). The data indicates that the suboptimal primary antibody responses detected in IL-21R−/− mice immunized with 103 ffu/mouse of rRABV appears to be due to the lack of GC B cell formation, while the suboptimal primary antibody response detected in IL-21R−/− mice immunized with 105 ffu/mouse most likely does not result from a defect in Tfh or GC B cell development.
Our analysis above shows that IL-21R signaling is dispensable for the formation of the Tfh and GC B cell populations in response to higher doses of rRABV, indicating that other B cell types are more likely responsible for the suboptimal primary antibody titers detected in IL-21R−/− mice immunized with 105 ffu/mouse. Since IL-21 can also influence the balance between the generation of memory B cells and PCs [24], [45]–[47], we investigated the role for IL-21R signaling to regulate memory B cell and PCs populations in response to RABV vaccination. Figure 6A and Figure 6B show representative gating strategies [48]–[50] to identify the memory B220+ B cells (CD38+CD138−) from the lymph node and PC (B220loCD138+) populations from the bone marrow from mice immunized with 103 or 105 ffu/mouse of rRABV or PBS alone. The presence or absence of IL-21R does not appear to influence the development of memory B cells in mice immunized with 103 ffu/mouse of rRABV (Figure 6C). However, the percentage of memory B cells was significantly increased in the lymph node cell cultures from IL-21R−/− mice immunized with 105 ffu/mouse of rRABV as early as 7 days post-immunization compared to immunized wild-type mice (Figure 6C). By day 14 post-immunization, similar memory B cell populations were measured (data not shown), indicating that IL-21 is not required for the formation of memory B cells in response to live RABV-based vaccination. This is consistent with the findings in Figure 1 and Figure 2 indicating that IL-21 is dispensable for secondary antibody responses against RABV infection when a primary antibody response develops. However, the percentage of PCs was reduced in the bone marrow of IL-21R−/− mice immunized with 105 ffu/mouse compared to rRABV- or PBS-immunized wild-type mice 14 days post-immunization (Figure 6D), consistent with the suboptimal primary antibody titers detected in IL-21R−/− mice.
Current rabies PEP regimens are based on multiple doses of inactivated RABV-based vaccines administered intramuscularly or intradermally. In cases of severe exposure, rabies immune globulin (RIG) is administered [51]–[53]. The development of a single-dose vaccine would greatly benefit human rabies prevention by reducing the cost of vaccination and saving lives. Understanding immune parameters that influence the magnitude and/or quality of anti-RABV antibody responses may lead to more effective single-dose vaccines [54]. Correlates of protection against rabies infections are defined as virus neutralizing antibodies directed against the single viral transmembrane glycoprotein (G) [15], [52], [54], [55]. CD8+ T cells do not appear to be important for the clearance of RABV infections [56]. Protection against RABV infection typically requires CD4+ T cell help [56]–[60], although we recently showed that this requirement is not absolute and that protection against pathogenic RABV challenge can be afforded in mice devoid of all T cells (TCRβδ−/− mice) vaccinated with a matrix gene-deleted RABV-based vaccine (rRABV-ΔM) [16]. Furthermore, we show that mice immunized with rRABV-ΔM also induce antibodies by T cell-dependent extrafollicular B cell responses before GC-derived B cells are detected. Together, our previous work identified multiple pathways of B cell development that can be exploited to make more efficacious RABV-based vaccines for use in humans. Nonetheless, very little information is available on how effective B cells develop in response to live RABV-based vaccination. IL-21 is a pleotropic cytokine that is produced by NKT cells and CD4+ T cells, most notably Th17 and Tfh cells. IL-21 binds to the IL-21R on a wide variety of cells involved in innate immunity, including DCs, NK cells, NKT cells, and macrophages, as well as on cells involved in adaptive immunity, such as B cells and CD4+ or CD8+ T cells [reviewed in [20]]. Due to its multiple roles in innate and adaptive immunity, IL-21 has the potential to influence the quality and magnitude of vaccine-induced immunity to acute viral infections. Here we used a mouse model of IL-21R-deficiency to evaluate the role for IL-21R signaling in vaccine-induced protection against RABV; i.e., an acute viral infection that relies on B cells for protection that has implications for global public health initiatives.
In this report, we showed that IL-21R signaling is critical for the generation of optimal primary anti-RABV antibody responses to vaccination. Primary anti-RABV antibody titers were significantly reduced in immunized IL-21R−/− mice compared to wild-type mice at almost all time points tested post-immunization, suggesting IL-21R signaling plays important roles throughout RABV-specific primary B cell responses. Nonetheless, IL-21R signaling appears to influence immunity in a vaccine dose-dependent manner, which is consistent with findings by others suggesting the influence of IL-21 is dependent on the model studied [22], [61]. Indeed, significantly less IL-21R−/− mice immunized with low-dose vaccination were protected against pathogenic challenge compared to wild-type mice while all IL-21R−/− and wild-type mice immunized with high-doses of vaccine survived challenge similarly. Despite the differences in protection elicited in IL-21R−/− mice immunized with different doses of vaccine, it appears that IL-21 is critical for the generation of optimal RABV-specific primary B cell responses.
One potential explanation for the suboptimal primary antibody responses observed in immunized IL-21R−/− mice compared to immunized wild-type mice might be that GC B cells failed to form in IL-21R−/− mice, therefore, the GC B cell compartment was analyzed in mice immunized with different doses of vaccine. GC-derived B cells were reduced in IL-21R−/− mice immunized with a low dose of vaccine compared to similarly immunized wild-type mice, indicating that IL-21 is required for optimal GC B cell formation in response to low-dose RABV-based vaccination. On the other hand, GC-derived B cells expanded in IL-21R−/− mice immunized with a high dose of vaccine compared to similarly immunized wild-type mice, indicating that IL-21 is dispensable for GC B cell formation after high-dose vaccination with rRABV-based vaccines. Furthermore, the data indicates that factors other than IL-21 were responsible for GC B cell formation in IL21R−/− mice immunized with higher doses of vaccine. Multiple signals lead to B cell activation and functions. These signals can come from BCR or Toll-like receptor (TLR) ligation, TNF superfamily receptor engagement (eg., via BAFF and APRIL) or cytokine signaling. Furthermore, B cell activation is contextual, meaning B cells are differentially activated in the presence of different signals at the time of antigen exposure. Due to the repetitive display of rabies antigen on the surface of infectious particles, the potential exists that cross-linking BCRs and/or TLRs on the surface of B cells overcame the requirement for IL-21R signaling in B cell activation when high doses of vaccine are administered. The influences of these and other B cell signaling events in the context of RABV-based vaccine-induced B cell activation were not directly measured in these studies and remain to be elucidated. Nonetheless, based on the results reported here, it appears that IL-21R signaling is important for optimal primary vaccine-induced antibody responses to RABV vaccination especially when low doses of vaccine are administered.
As noted above, a rapid antibody response is critical for rabies PEP to neutralize virus before it reaches the CNS. We have recently shown that RABV-based vaccines are able to induce early and rapid T cell-dependent extrafollicular antibody responses before GC B cells are formed [16]. These early pre-GC B cell responses contributed to the protection against pathogenic RV challenge early post-immunization, which is an important attribute for PEP [16]. In the studies described in this report, we detected a significant reduction in antibody titers in IL-21R−/− mice as early as 5 days post-immunization with a high dose of rRABV compared to immunized wild-type mice, suggesting that IL-21 may be influencing the outcome of extrafollicular antibody responses in the context of RABV vaccination, although this was not directly studied in this report. Nonetheless, the frequency of Tfh and GC B cells was similar in IL-21R−/− mice compared to wild-type mice 7 days post-immunization and, therefore, it would appear that the early suboptimal antibody responses in IL-21R−/− mice may be due to impaired extrafollicular PCs directly and not through impaired Tfh or GC B cell formation. This is consistent with the findings that IL-21 can promote Blimp-1 expression and PC development [47], IL-21- or IL-21R-deficiency decreases extrafollicular PCs in a model of NP-KLH immunity [62], and that IL-21 acts on early stages of B cell differentiation before GC or PC B cells are formed [42]. Finally, IL-21 has been reported to be important for Tfh cell maintenance but not formation. Together, existing data suggests that IL-21R−/− signaling influences early events in pre-GC B cell development in the context of RABV vaccination [22].
IL-21 can also influence the balance of B cell differentiation into memory B cells or PCs [45]–[47]. The specific role for IL-21 in memory B cell responses is not completely clear and appears to rely on the type of antigen used and the model studied [61], [62]. In the context of RABV vaccination, IL-21R signaling was not required for the generation of B cells displaying a memory B cell phenotype in IL-21R−/− mice immunized with either vaccine dose. This is consistent with our finding showing that IL-21R signaling is not required for optimal secondary anti-RABV G antibody titers after challenge with pathogenic RABV. However, we detected a decrease in the number of PCs in IL-21R−/− mice immunized with either dose of vaccine compared to wild-type mice. We cannot determine whether the slightly suboptimal PC subset detected in IL-21R−/− mice immunized with low doses of vaccine was indirectly a result of impaired GC B cell development or directly as a result of impaired PC formation itself. However, in mice immunized with a high dose of vaccine where we observed an expansion of GC-derived B cells in IL-21R−/− mice, we also observed a decrease in PCs in the bone marrow compared to wild-type mice, indicating that IL-21 acts directly on the formation of PCs. Together, the data shows that IL-21 influences the balance between memory and PC B cell formation in the context of RABV vaccination.
Despite the impaired primary antibody response and PC B cell formation in immunized IL-21R−/− mice, we detected an expansion of CD4+ T cells displaying a pre-Tfh (data not shown), Tfh cell and GC B cell phenotype in IL-21R−/− mice compared to wild-type mice at 14 days post-immunization. The expansion of GC B cells and Tfh cells in the absence of IL-21R signaling was also shown by King I.L. et al in a model of Heligmosomoides polygyrus immunity [24], suggesting that IL-21R signaling may not be necessary for the generation of these cell types in response to a wide range of pathogens or vaccination. Furthermore, the increase in GC B cells and Tfh cells in H. polygyrus-infected or RABV-vaccinated IL-21R−/− mice suggests that IL-21R signaling may play an inhibitory role in the development of T and B cells in the context of some pathogens, which is consistent with the ability for IL-21 to activate or inhibit immune function depending on the antigen and available co-stimulatory signals [20]. The expansion of Tfh and GC B cells in IL-21R−/− mice compared to wild type mice is also consistent with the finding that IL-21 has the ability to mediate apoptosis in primary resting and activated murine B or to promote apoptosis or growth arrest for non-specifically activated B cells [63]. Alternatively, the elevated number of GC Tfh cells could be a result of the lack of PC that developed in the IL21R−/− mice. Pelletier et al described a negative regulatory feedback-loop in which antigen-specific PCs negatively regulate antigen-specific Tfh cell development and function [64]. In this report, they also observed a significant expansion of Tfh cells and GC B cells in the absence of PC development. Together, the role for IL-21 in the homeostatic balance of T and B cell development in the context of infectious diseases appears to be important and remains to be fully elucidated.
Additional studies are needed to identify the exact cell type(s) responsible for the affects described in this report. While we speculate that B cell-intrinsic IL-21R signaling is responsible for the induction of optimal anti-RABV antibody responses, we cannot rule out the influence of other cell types that also express IL-21R. IL-21 has the ability to influence the function of macrophages, NK cells and NKT cells by affecting survival/apoptosis, antigen processing, and cytokine secretion [reviewed in [20]]. The function of these cells of the innate immune system may indirectly be affecting the outcome of B or T cell functions in the context of RV vaccination. Nonetheless, IL-21 has the potential to influence a wide range of B cell functions and pathways. Our preliminary data indicates that IL-21 is critical for the formation of optimal vaccine-induced primary antibody responses and demonstrates an important role for IL-21 in the generation of vaccine-induced immunity against RABV infection and perhaps other acute infections that rely on B cell-mediated effector functions for protection.
|
10.1371/journal.ppat.1005986 | Mycobacterium abscessus-Induced Granuloma Formation Is Strictly Dependent on TNF Signaling and Neutrophil Trafficking | Mycobacterium abscessus is considered the most common respiratory pathogen among the rapidly growing non-tuberculous mycobacteria. Infections with M. abscessus are increasingly found in patients with chronic lung diseases, especially cystic fibrosis, and are often refractory to antibiotic therapy. M. abscessus has two morphotypes with distinct effects on host cells and biological responses. The smooth (S) variant is recognized as the initial airway colonizer while the rough (R) is known to be a potent inflammatory inducer associated with invasive disease, but the underlying immunopathological mechanisms of the infection remain unsolved. We conducted a comparative stepwise dissection of the inflammatory response in S and R pathogenesis by monitoring infected transparent zebrafish embryos. Loss of TNFR1 function resulted in increased mortality with both variants, and was associated with unrestricted intramacrophage bacterial growth and decreased bactericidal activity. The use of transgenic zebrafish lines harboring fluorescent macrophages and neutrophils revealed that neutrophils, like macrophages, interact with M. abscessus at the initial infection sites. Impaired TNF signaling disrupted the IL8-dependent neutrophil mobilization, and the defect in neutrophil trafficking led to the formation of aberrant granulomas, extensive mycobacterial cording, unrestricted bacterial growth and subsequent larval death. Our findings emphasize the central role of neutrophils for the establishment and maintenance of the protective M. abscessus granulomas. These results also suggest that the TNF/IL8 inflammatory axis is necessary for protective immunity against M. abscessus and may be of clinical relevance to explain why immunosuppressive TNF therapy leads to the exacerbation of M. abscessus infections.
| The incidence of non-tuberculous mycobacterial infections has recently increased and has even surpassed tuberculosis as a public health concern in many developed countries. These infections require long treatment regimens that are often unsuccessful. Among these, Mycobacterium abscessus has emerged as perhaps the most difficult-to-manage pathogen, especially in cystic fibrosis patients. Unfortunately, very little is known regarding the contributions of the pro-inflammatory and innate immune responses during M. abscessus infection. Here, we exploited the transparency of zebrafish embryos to study, at high resolution, the interactions of M. abscessus with macrophages and neutrophils, and found that both cell types are required to control the infection. We also describe the dramatic consequences of impaired TNF/IL8 immunity on the outcome of the infection. Most importantly, by tracking the dynamics of neutrophil mobilization, we demonstrated the crucial role of these cells in the formation and integrity of protective granulomas. Together, our data provide a significant advance in deciphering the immunopathology of M. abscessus infection, which is particularly relevant for understanding the exquisite vulnerability of cystic fibrosis patients to this bacterium.
| The rapidly-growing non-tuberculous mycobacteria (NTM), Mycobacterium abscessus (Mabs), is an emerging pathogen that causes a wide clinical spectrum of syndromes, including skin and soft tissues infections and pseudotuberculous pulmonary infections, especially in patients with underlying lung disorders [1–3]. Mabs is considered to be the most pathogenic NTM affecting cystic fibrosis (CF) patients, and is often associated with a dramatic decline in lung function and even death in these patients [4]. This organism is notorious for being intrinsically resistant to most antibiotics and disinfectants [5,6,7] and unsuccessful eradication of Mabs is a contraindication to lung transplantation in many CF centers, leaving patients without any therapeutic option. Despite being a rapid grower, Mabs can persist for years or decades within the lungs of infected patients [8,9] where it forms organized granulomas [10]. These hallmarks of pathogenic mycobacteria are composed of infected macrophages surrounded by additional macrophages, neutrophils and lymphocytes, and the centers of these tightly aggregated structures can develop caseous necrosis [10]. The pathways leading to Mabs granuloma formation and maintenance, however, have been poorly characterized.
Mabs exhibit rough (R) and smooth (S) morphotypes that, as a consequence of alterations within the GPL biosynthetic/transport gene cluster, differ in the amounts of surface-associated glycopeptidolipids (GPL) [11]. Pulmonary Mabs infections in CF patients that are characterized by chronic airway colonization and poor outcomes have been linked to a genetic conversion allowing the S variant to become R [8,12,13]. This is also supported by results in mice and in cultured macrophages, emphasizing the hyper-virulent phenotype of the R compared to the S form [14–16]. MmpL4b is involved in translocation of GPL to the bacterial surface and its absence correlates with the lack of GPL and the highly virulent phenotype of the R variant [11,17,18]. A plausible explanation for the enhanced virulence of the Mabs R morphotype is that loss of GPL unmasks cell wall inflammatory-provoking lipoproteins and/or phosphatidyl-myo-inositol mannosides known to be TLR2 agonists [19,20]. Despite the demonstration that Mabs R induces a stronger TLR2-mediated TNF response than Mabs S [15,20], there is little information regarding the events leading to the inflammatory response in Mabs infection and how the inflammation impacts on the outcome of the disease.
By exploiting the optical transparency of the zebrafish embryo model, we confirmed the hyper-virulence of Mabs R and revealed the ability of both Mabs morphotypes to induce granulomas [17]. The increased virulence of the R variant correlated with a massive production of extracellular serpentine cords of bacteria which, due to their size, prevent phagocytosis, thus suggesting cording as a mechanism of immune evasion. Cords also initiated abscess formation, particularly in the central nervous system (CNS) of the infected animal, with subsequent tissue damage presumably caused by the induction of a potent inflammatory response. Lung injury in CF patients is caused by an intense and persistent pulmonary infection associated with a massive influx of neutrophils into the airways [21]. Thus, scrutinizing the inflammatory and neutrophilic response to Mabs infection could provide important advances in our understanding of the Mabs immunopathology and the unexplained innate susceptibility of CF patients to these infections.
Herein, through the use of loss- and gain-of-function approaches coupled with fluorescent reporter zebrafish lines and high resolution imaging, we have dissected the TNF/IL8-mediated signaling pathway that contributes to immuno-protection against Mabs infection. Our findings unraveled the crucial and dual role of TNF in activating the macrophage bactericidal activity, in restricting intracellular bacterial growth and, importantly, in neutrophil recruitment for the generation and maintenance of protective granulomas.
To identify key effectors of the Mabs-induced inflammation, the pro-inflammatory profile in R- and S-systemic infected zebrafish embryos was determined and compared at various time points post-infection. Quantitative RT-PCR revealed an induction of tnf-α (Fig 1A), il1-β and ifn-γ2 (S1 Fig). Expression of tnf-α is induced by both variants at 2 days post-infection (dpi), corresponding to the early appearance of granulomas, and prior to abscess formation [17]. This response was further increased at 5 dpi, with higher levels for R, consistent with the increased TNF release in murine macrophages [15]. This indicates that Mabs, notably the R variant, induces a robust TNF response in zebrafish.
Tg(tnfα:eGFP-F) zebrafish larvae, which express farnesylated eGFP under the control of the tnf-α promoter [22], were infected with Mabs to investigate the nature and the spatiotemporal distribution of the TNF-producing cells. While the control PBS injection in the otic cavity failed to induce eGFP expression, all animals injected with Mabs exhibited green fluorescent cells that were recruited and clustered around the injection site as early as 2 hours post-infection (hpi) (Fig 1B and 1C), with equal numbers of eGFP-positive cells recruited in response to S and R variants (Fig 1D). In intravenously (iv) infected embryos, eGFP expression was detected from the earliest hours post-infection (S2A Fig), increased over time and peaked at 5 dpi, in an expanding bacterial-dependent manner, usually in close vicinity of the infection foci, particularly after R infection (Fig 1E).
Both Mabs R and S were found close to or within eGFP-positive cells near to the injection site at 1 dpi (S2B Fig). At later time points, in both S- and R-infected embryos, a strong eGFP signal was detected in Mabs-containing granulomas (Fig 1F and 1G) but not around or within R-abscesses (Fig 1G), the latter consisting essentially of highly replicating extracellular mycobacteria associated with tissue damage [17]. eGFP-positive cells were also surrounding Mabs serpentine cords (Fig 1H). Because TNF-α can be produced by numerous cells [23], we attempted to identify the TNF-producing cells in response to Mabs, with the double transgenic Tg(tnfα:eGFP-F/mpeg1:mCherry-F) or Tg(tnfα:eGFP-F/LysC:DsRed) embryos. It was possible to visualize green tnf-α expression concomitantly with red macrophages or red neutrophils, respectively. For both R and S infections, cells producing TNF-α co-localized with Mabs-containing macrophages, either isolated or within granulomas (S2B Fig and Fig 1I). In sharp contrast, the eGFP signal failed to co-localize with neutrophils (Fig 1J and S2C Fig). The total lack of TNF-positive cells in macrophage-depleted Tg(tnfα:eGFP-F) larvae (S2D Fig), generated after lipo-clodronate injection [17], confirmed macrophages as the primary source of TNF in response to Mabs infection.
Taken collectively, these results indicate that TNF is principally produced by macrophages following infection with both Mabs variants, from very early phagocytosis after infection to later time points when the characteristic granulomas have appeared.
TNF-α is a multifunctional cytokine playing a pivotal role in the regulation of inflammation and infection via the stimulation and engagement of the specific cell surface receptor 1 (TNFR1, ZDB-GENE-040426-2252). To address the role of TNF signaling in Mabs infections, loss-of-function experiments for TNFR1 were carried out using a specific morpholino, leading to complete abrogation of the native tnfr1 mRNA (S3A Fig) and thereby subsequent TNF production (S3B–S3D Fig). Although showing morphological defects (S3A Fig), uninfected tnfr1 (tnfr) morphants exhibited survival rates similar to those of wild type (WT) embryos (Fig 2A). Importantly, TNFR1 impairment led to an increase in the severity of the infection and hyper-susceptibility to R and S variants (Fig 2A). This correlated with higher bacterial burdens as demonstrated by whole embryo imaging (Fig 2B) and fluorescent pixel counts (FPC) (Fig 2C). Imaging of the R- and S-infected tnfr morphants showed that the increased bacteremia coincided with the presence of highly-replicating extracellular bacteria, resulting in the rapid development of abscesses (Fig 2B and 2D). Whereas abscesses remain the exclusive attribute of R infections in WT embryos, 60% of S-infected tnfr morphants developed abscesses (Fig 2D). Similarly, after R-infection, rapid and massive cord formation occurred in nearly all tnfr morphants within 1 dpi (Fig 2E and 2F). At 1 dpi, WT embryos had fewer cords (<5) while 70% of the morphants had high cord numbers (>5) (Fig 2G). Whereas cords developed essentially within the CNS in WT fish, all tnfr morphants exhibited widespread cording in the vasculature, in addition to the CNS (Fig 2F and 2H). Thus, hyper-cording of Mabs R occurs very rapidly in the absence of TNF-mediated immunity, leading to early larval death, as has been described for Mycobacterium marinum in embryos lacking TNF signaling [24].
These results demonstrate the crucial and protective role of the TNFR1-dependent pathway in response to Mabs S and R by restricting extracellular multiplication and pathogenesis.
To define how TNF orchestrates the events leading to the Mabs pathophysiology, we addressed whether the increased mortality and unrestricted mycobacterial growth in tnfr morphants are linked to a defect in macrophage recruitment, phagocytosis and/or bactericidal activity. The ability of macrophages to traffic across the epithelial and endothelial barriers was evaluated following S and R injection into the muscle (S4A Fig) and the otic cavity (S4B and S4C Fig) of Tg(mpeg1:mCherry-F) larvae. Irrespective of the infection site, the number of early recruited macrophages is comparable in both tnfr morphants and WT embryos at 2 hpi, implying that TNF signaling is not required for the early trafficking of macrophages, consistent with previous studies with M. marinum [24]. Although bacteria were rapidly engulfed by macrophages after infection of tnfr morphants (S4A Fig), the number of infected macrophages harboring either R or S bacteria was lower in morphants than in WT embryos at 4 hpi (Fig 3A). Knocking-down TNFR1 expression altered TNF expression (S3B–S3D Fig) and a defective TNF-positive feedback loop seems to be required for the efficient recruitment of macrophages at later stages (S4B Fig) and subsequent phagocytosis. The ability of tnfr morphants to develop abscesses in the CNS at later time points after intravenous infection suggests that macrophages remain efficient in transporting and disseminating the bacteria from the bloodstream to deeper tissues (Fig 2).
To assess the contribution of TNF signaling in modulating the mycobactericidal activity of macrophages, the number of intracellular bacteria in individual infected macrophages was determined. The proportion of slightly infected (<5 bacteria), moderately infected (5–10 bacteria) or heavily infected (>10 bacteria) macrophages was enumerated at 1 dpi (Fig 3B). Compared to the WT embryos, the tnfr morphants displayed a greater percentage of macrophages in the high burden category. This was true for both R and S variants and suggests that TNF restricts intracellular growth by stimulating the macrophage bactericidal activity. Further confirmation was obtained following staining of the infected embryos with a probe that detects reactive oxygen species (ROS) (CellROX) (Fig 3C). At 1 dpi, ROS-labeled infected macrophages were found in WT embryos (Fig 3D), in agreement with previous reports showing that macrophages can produce ROS to control Mabs infections [25,26]. No differences in the proportion of ROS-positive macrophages containing either Mabs S or R were noticed (Fig 3D). However, fewer ROS-positive infected macrophages were found in morphants as compared to WT embryos (Fig 3D), and ROS-positive macrophages in granulomas were only seen in WT animals (Fig 3E). Aggregating cells positive for CellROX staining were occasionally observed, but only in WT larvae (S5A and S5B Fig). However, the relatively low numbers of ROS-labeled infected macrophages, reflecting a reduced bactericidal activity in tnfr morphants, is unlikely to explain the extreme susceptibility of the tnfr morphants to Mabs and the very high extracellular bacterial loads. Because macrophage death has been previously shown to release extracellular Mabs, we examined the extent of macrophage death in infected larvae [17]. Imaging of the acridine orange (AO)-infected larvae (S6A Fig) and quantitative determination of AO-positive macrophages (S6B Fig) showed the significantly higher numbers of dead infected-macrophages in the tnfr morphants at 2 dpi as compared to the WT embryos. As expected, the basal levels of dead macrophages were very low in PBS-injected WT embryos or uninfected tnfr morphants (S6A Fig).
Overall, these results indicate that TNF signaling is pivotal in establishing the initial innate immune response by: i) triggering the early bactericidal activity in macrophages and granulomas to restrict intracellular growth; ii) reducing uncontrolled extracellular bacterial growth by preventing macrophage death; and iii) promoting the formation of inflammatory cell aggregates which, in turn, may contribute to amplifying the local inflammation and recruitment of other immune cells [27].
Neutrophils are the first line of defense against pathogenic microorganisms and are rapidly recruited to infection sites where they engulf microorganisms and excrete their granule contents [28]. Mabs-containing neutrophils were previously identified in granulomas [17] but how these cells are recruited and contribute the immunity against Mabs remains unknown. Time-lapse microscopy of neutrophil mobilization in the caudal vein (Fig 4A) or in the muscle (Fig 4B and S1 Movie) of Tg(mpx:eGFP) embryos revealed a massive influx of neutrophils at the infection site starting at 10–20 min post-infection (mpi). Isolated Mabs were rapidly engulfed by neutrophils (Fig 4C and 4D), however the number of neutrophils harboring either R or S bacteria remained lower than the number of infected macrophages at 4 hpi (Fig 4E). Time-lapse microscopy showed a massive mobilization of neutrophils within the deeper CNS infection foci (S7A Fig), especially around abscesses (Fig 4F). Interestingly, the R-abscesses continued to expand, concomitant with a time-dependent disappearance of the neutrophils, reminiscent of the neutropenia (S7A Fig), and high bacteremia occurs prior to larval death as reported in embryos infected with Shigella [29].
IL8 is a central chemokine in neutrophil mobilization from hematopoietic tissues to infection sites. qRT-PCR revealed up-regulation of il8 early after infection, coinciding with granuloma formation and peaking at 5dpi, with higher levels for R infection than for S infection (Fig 5A). Injection of a morpholino targeting il8 expression [30] in WT embryos strongly inhibited early neutrophil mobilization into infected muscle, hindbrain or otic cavity (Fig 5B) while leaving the baseline number of neutrophils unchanged, as reported earlier [30]. At later stages of infection, no neutrophils were associated to the CNS abscesses of il8 morphants, a phenomenon unrelated to neutropenia (Fig 5C). In sharp contrast, while infection of il8 morphants resulted in strongly impaired neutrophil recruitment in the muscle, the hindbrain and the otic cavity (Fig 5B), mobilization (S8A Fig) and phagocytosis (S8B Fig) of macrophages were unaffected by the il8 morpholino injection. Importantly, IL8 ablation correlated with reduced larval survival (Fig 5D) and with increased S and R loads (Fig 5E).
The lack of neutrophil recruitment seen with IL8 ablation paralleled the pronounced increase in the bacterial loads, as evidenced by the numerous large abscesses (Fig 5E), suggesting that the absence of neutrophils at the site of infection may be deleterious for the host. To test this hypothesis, specific neutrophil depletion was carried out though injection of csf3r morpholino [31], which at the concentration used did not affect the macrophage recruitment (S9A Fig) or macrophage phagocytic activity (S9B Fig). As seen in the il8 morphants, the csf3r morphants showed hyper-susceptibility to both R and S infections with 100% larval mortality at 6 dpi (Fig 5F) and massive extracellular bacteremia (Fig 5G). Overall, these findings indicate that IL8 mediates neutrophil mobilization to the infection sites and plays a critical role in host defense against Mabs.
TNF orchestrates the early regulation of chemokine induction, including IL8, essential for neutrophil activation and recruitment to inflamed tissues [32]. To address the role of TNF in neutrophil mobilization, local infections were done in the otic (Fig 6A and 6B) and hindbrain (Fig 6C and 6D) cavities of Tg(mpx:eGFP) tnfr morphants. While neutrophils were rapidly recruited to the infected ear or hindbrain in WT animals, their recruitment was severely reduced in the tnfr morphant for both S and R variants (Fig 6A–6D), supporting a major role of TNF in early neutrophil mobilization. Whilst confocal microscopy of the cord/abscess-containing environments demonstrated massive mobilization of neutrophils around cords (Fig 6E and S2 Movie) and abscesses (Fig 6F and S4 Movie) in WT embryos, this was not true in tnfr morphants (Fig 6E and S3 Movie, Fig 6F and S5 Movie). The reduced number of neutrophils in tnfr morphants (S10A and S10B Fig), was consistent with a study reporting the influence of TNF on hematopoietic stem cell formation [33]. To inquire whether the decreased neutrophil recruitment in tnfr morphants is linked to a possible alteration in a basal neutrophil number, we performed a neutrophil mobilization assay using fMLP, a synthetic neutrophil chemoattractant [34]. After injection of fMLP into the otic cavity of tnfr morphants, neutrophils were recruited to the injection site to the same extent as in the WT embryos (S10C Fig). Comparable results were also obtained following injection of recombinant IL8 into the tnfr morphants (S10D Fig). Thus, the decreased neutrophil recruitment in tnfr morphants (Fig 6) is due neither to the general reduction of the neutrophilic population nor to nonspecific effects of the tnfr morpholino used.
Together, these results further position TNF as a critical mediator in initiating the early and late phases of neutrophil recruitment and substantiate the crucial role of neutrophils in controlling Mabs infections.
Despite the co-existence of macrophages and neutrophils in Mabs-induced granulomas [17], the importance of neutrophils in the maintenance and/or integrity of this organized cellular structure remains elusive. Monitoring the kinetics of granuloma development revealed that granulomas induced by both R- and S-variants appeared at 2 dpi and expanded in most WT embryos at 5 dpi (Fig 7A and 7B and [17]). In sharp contrast, the granuloma-like structures in the tnfr morphants appeared as poorly delimited loose cellular aggregates (Fig 7A–7E and S11 Fig), similar to those previously documented in TNF defective M. marinum [24], and highlighting the absolute requirement of a functional TNF pathway for granuloma formation. Confocal microscopy revealed a more open and disjointed structure with the presence of high numbers of extracellular bacterial aggregates (Fig 7C), prompting us to examine whether the increased proportion of dead macrophages in infected tnfr morphants (S6A and S6B Fig) may contribute to the morphologically altered granulomas and the reduced granuloma numbers. As shown in S6C Fig, despite the higher proportion of dead macrophages in the tnfr morphants as compared to the WT embryos, there was no correlation with the number of “defective” granulomas found in the tnfr morphants. While granulomas in WT embryos contained numerous neutrophils, they were nearly absent in the corresponding structures in Tg(mpx:eGFP) tnfr morphants, which were characterized by substantial numbers of extracellular bacteria/cords (Fig 7D, S6 and S7 Movies). Time-lapse monitoring (Fig 7E) and determination of the number of infected neutrophils recruited to the granulomas (Fig 7F) established a linear relationship between the number of recruited neutrophils and the granuloma volume in WT embryos. In contrast, tnfr morphants exhibited an important lack of neutrophils (Fig 7E) and no linear correlation with the size of the granuloma-like structures could be established (Fig 7F), suggesting a direct participation of neutrophils in elaborating and shaping Mabs granulomas.
That TNF modulates the early mobilization of neutrophils into granulomas led us to determine whether TNF depletion influences IL8 expression. Quantitative RT-PCR revealed that, while Mabs stimulated il8 expression levels in WT larvae, its expression was severely decreased in the TNFR1-depleted larvae (Fig 8A) and was even further reduced in macrophage-depleted larvae following injection of lipoclodronate (Fig 8A), indicating that macrophages are part of the pathway that triggers IL8 release in response to Mabs infections. This suggests that in WT embryos, TNF production by macrophages (Fig 1I) governs IL8-driven chemotaxis, and thus the absence of TNF signaling profoundly restricts neutrophil mobility. Having confirmed that macrophages are required for IL8 production, we next examined whether neutrophil mobilization to the infection site is dependent on macrophages. Neutrophil recruitment was dramatically reduced in the macrophage-depleted larvae (Fig 8B), demonstrating that macrophages are key players in neutrophil mobilization in response to Mabs infection. In addition, while neutrophil recruitment is strongly impaired in the absence of either macrophage or TNFR signaling, the injection of exogenous IL8 fully rescued neutrophil mobility (Fig 8B and 8C) and restored survival of tnfr morphants infected with R or S in the otic cavity (Fig 8D). This correlated with decreased bacterial loads compared to untreated tnfr morphants, as evidence by the determination of the FPC (Fig 8E) and fluorescence microscopy (Fig 8F). These findings highlight the immune-protective role of IL8-dependent neutrophil mobilization during Mabs infections. Furthermore, granuloma formation was severely impaired in il8 morphants and, consistent with these findings, granuloma formation was abrogated in the neutrophil-depleted csf3r morphants (Fig 8G and 8H). Similarly to tnfr morphants (Fig 7F), il8 morphants exhibited neutrophil-poor granulomas (Fig 8I), further supporting the direct participation of neutrophils in elaborating and shaping Mabs granulomas.
Overall, these results emphasize the absolute requirement of an IL8/TNF-dependent neutrophil mobilization for granuloma formation and control of Mabs infections.
Herein, we report the first stepwise dissection study of the immune control of Mabs using a non-invasive visualization approach with special emphasis on the inflammatory response. The spatiotemporal immunopathological events (Fig 9) can be summarized as follows: i) rapid engulfment of Mabs by macrophages; ii) TNF release by activated macrophages, leading to ROS production and intracellular killing of Mabs, and IL8-driven chemotaxis that guides neutrophils to the infection site; iii) proficient granulomatogenesis and development of chronic infections. In contrast, defective TNF signaling correlates with i) impaired macrophage activation with increasing intramacrophage bacterial loads and disruption of IL8 production, resulting in impaired neutrophil recruitment; ii) absence of bona fide granulomas. Additionally, the increased macrophage death releases free bacilli that multiply extracellularly in an uncontrolled manner, resulting in mycobacterial cords that prevent phagocytosis by macrophages and neutrophils [17], acute infections and larval killing.
Mabs infections are characterized by growth in highly inflamed tissues, suggesting a role for neutrophils in the host response. Supporting this hypothesis, patients with CF, a disease that is dominated by persistent neutrophil-mediated inflammation, are particularly susceptible to Mabs in addition to other extracellular pathogens [4,35]. However, despite the prevalence of neutrophils in Mabs infections, information regarding the neutrophil response is scarce. Previous work suggested that human neutrophils mediate killing of Mabs, but phagocytosis was reduced when compared to Staphylococcus aureus, another important CF pathogen [36]. In experimental models of Mabs infection, the presence of increased numbers of neutrophils is associated with a worse response to Mabs [37]. We assessed here the in vivo capacity of neutrophils to migrate in response to Mabs, to engulf the bacilli and to participate directly to granuloma formation in infected zebrafish. Our results are distinct from those with Mycobacterium tuberculosis or M. marinum where neutrophils appear to be less mobilized [34,38]. Ex vivo infections of human lung tissues indicated that neutrophils had a greater tendency to phagocytize Mabs than M. tuberculosis and that Mabs has a higher capacity to induce the migration of neutrophils than other mycobacterial species [38]. Additionally, while neutrophils are unable to kill virulent strains of M. tuberculosis or M. marinum [39,40], they appear to be important for controlling virulent and avirulent Mabs because neutrophil-depleted embryos are extremely susceptible to Mabs infection. The mobilization of neutrophils toward the large cords and abscesses is also in agreement with recent studies reporting that neutrophils, through a microbial size-sensing mechanism, tailor their antimicrobial responses to pathogens based on microbial size [41].
Recent studies have indicated that M. tuberculosis-infected neutrophils can be considered as biomarkers for poorly controlled mycobacterial replication and are associated with severity in human tuberculosis [42]. Our data show that infected neutrophil-depleted embryos exhibit increased CNS pathology, high mycobacterial loads, decreased survival rates and reduced granuloma numbers. They also support the crucial role of IL8 in the control of Mabs at the infection site due to its function as the main mediator of neutrophil mobilization to the infection site, and through this recruitment of neutrophils, the formation of granulomas. Indeed, the absence of granulomas caused by il8 or tnfr knock-down was associated with compromised survival and reduced bacterial clearance of Mabs, phenotypes that were restored upon injection of recombinant IL8. Despite the important role of macrophages in IL8 signaling, production of IL8 was not completely abrogated in the macrophage-depleted embryos, suggesting that other cell types may also contribute to IL8 production.
Our results thus shed new light on the role of neutrophils in early granuloma formation, integrity and maintenance, and reveals striking differences with the dynamics of granuloma formation in the M. marinum zebrafish model, where granulomas contribute to early bacterial growth and expansion of the infection [43]. Although one cannot exclude the possibility that Mabs exploits the granuloma to manipulate host immune responses for its own benefit as suggested for M. marinum, our conclusions support the primary role of granulomas in preventing a widespread expansion of Mabs in the extracellular milieu. Granuloma-defective embryos (tnfr, il8 or csf3r morphants) were all hyper-susceptible to S and R infections with pronounced larval mortality and unrestricted extracellular bacterial growth. Both R and S strains simulated granuloma formation at comparable levels [17], indicating that granuloma formation is not correlated with virulence of Mabs. In contrast, granuloma formation with M. marinum is linked to virulence, as demonstrated using the attenuated RD1-deficient mutant [44]. Apparently, the distinctive kinetics and functions of granulomas are species-specific and conclusions drawn from one mycobacterial species cannot be extrapolated to others.
Furthermore, while M. marinum-infected embryos deficient in TNF signaling showed increased granuloma formation in early stages of infection [24], after Mabs infection of tnfr morphants the granuloma formation was almost completely abrogated. Instead of compact organized granuloma, Mabs infection of tnfr morphants elicited only the formation of disorganized cellular structures that consisted of aggregated macrophages. A similar result was seen in TNF-α KO mice where impaired Mabs control was associated with profound alteration of granuloma formation [16]. Why would TNF exert distinct responses when stimulated with different mycobacterial species? It is well known that the mycobacterial cell envelope possesses a large panel of surface-exposed glycolipids, some of which are particularly granulomatogenic. Although the overall architecture of the cell wall is conserved among mycobacteria, each species can be typified by a specific lipid/glycolipid signature, and subtle variations in the lipid composition, structure and size can affect granulomatogenesis. Other effectors, such as the ESX-1 are also required for efficient granuloma formation with M. tuberculosis and M. marinum [44,45]. While the loss of RD1 in a M. marinum is associated with altered early aggregation of infected macrophages and a delay in granuloma formation [44], Mabs, which naturally lacks RD1, induces “normal” granuloma formation in embryos [17], adult zebrafish [46] and in mice [16]. M. marinum mutants with reduced granuloma formation were also found to be defective in EspL, located in the ESX-1 cluster [47]. Mabs presents two ESX-like loci [48] but whether these clusters participate in Mabs-induced granulomas and pathogenesis awaits further investigation. The identification of the Mabs-specific effectors participating in granuloma formation would contribute to our knowledge of Mabs pathogenicity and could foster the development of needed therapeutic interventions, as Mabs is refractory to most antimicrobials.
In conclusion, we report here new and unexpected insights into several aspects of Mabs immunopathogenesis by demonstrating the crucial role of TNF signaling in a continuum of effects starting from limiting intracellular bacterial multiplication to the induction of granulomas that together exert a protective effect by limiting Mabs dissemination. Through its orchestration of the inflammatory cytokine/chemokine network, dominated by IL8 production, TNF modulates the engagement of neutrophils to the infection site and their subsequent recruitment to granulomas, which are essential for control of the early and later stages of Mabs infection, respectively. This explains why immunosuppressive TNF therapy increases the risk of Mabs infections [49]. Airways of CF patients are characterized by a severe inflammation [50] but whether this directly impacts on subsequent Mabs colonization is difficult to predict. The airways of these patients are chronically colonized with complex, polymicrobial infections [51] and these polymicrobial communities promote intricate inter-microbial and host-pathogen interactions which alter the lung environment, impact the response to treatment, and drive the course of the disease. In addition, the relationship between abnormal CFTR expression and the predisposition of CF patients to chronic Mabs infections remains elusive, but the results presented here suggest that it would be interesting to determine whether the CFTR defect has a detrimental effect on neutrophil-mediated immunity.
Zebrafish experiments were conducted in accordance with the guidelines from the European Union for handling of laboratory animals (http://ec.europa.eu/environment/chemicals/lab_animals/home_en.htm) and approved by the Direction Sanitaire et Vétérinaire de l'Hérault et Comité d'Ethique pour l'Expérimentation Animale de la région Languedoc Roussillon under the reference CEEA-LR-1145. For zebrafish anaesthesia procedures, embryos are immersed in a 270 mg/L Tricaine solution in osmotic water. When required, larvae were cryo-anesthetized by incubation on ice for 10 min and then euthanized using an overdose of Tricaine (500 mg/L). Experimental procedures were performed using the golden zebrafish mutant [52], along with transgenic lines: Tg(mpx:eGFP)i114 [53] or Tg(LysC:DsRed)nz5 [54], harboring either green- or red-fluorescent neutrophils, respectively; Tg(mpeg1:mCherry-F)ump2 [17], harboring red-fluorescent macrophages; and Tg(tnfα:eGFP-F)ump5 [22], which allows visualization in green of the transcriptomic expression of tnf-α under appropriate stimulatory conditions.
R and S variants of M. abscessus sensu stricto strain CIP104536T (ATCC19977T) carrying pTEC15 (Addgene, plasmid 30174), pTEC27 (Addgene, plasmid 30182) or pTEC19 (Addgene, plasmid 30178) that express green fluorescent protein (Wasabi), red fluorescent protein (tdTomato) or bright far-red fluorescent protein (E2-Crimson), respectively, were prepared and microinjected in zebrafish embryos, according to procedures described earlier [17,55]. Briefly, systemic infections were carried by the injection of 100–200 CFU into the caudal vein of 30 hours post-fertilization (hpf) embryos. For phagocyte recruitment assays, 100 CFU were injected locally into the otic vesicle, the muscle or the hindbrain compartments of 3 days post-fertilization (dpf) larvae or into the caudal vein of 2 dpf embryos.
Neutrophil recruitment was elicited in zebrafish embryos through injection of the f-Met-Leu-Phe (fMLP, Sigma-Aldrich) or recombinant human IL-8 (rhIL-8, R&D Systems, Inc.), chemoattractants previously described [34]. IL-8 (15–25 pg) or 3 nl of 300 nM fMLP were injected into the otic cavity of 3 dpf larvae. The quantity of recruited neutrophils was determined at the injection site at 3 hpi using fluorescence microscopy.
Morpholinos purchased from Gene Tools were injected into 1–4 cell stage embryos. tnfrsf1a splice-blocking morpholinos targeting TNF receptor 1 (5’-GGAAGCATGAGGAACTTACAGTTCT-3’) were used at a concentration of 0.5 mM. The efficiency of gene knockdown was confirmed by RT-PCR with the following primers for both sides of the morpholino target sequence: tnfrsf1a, CCCGCATGCTCCACGTCTCC and TTATAGCGGCCGCCCGACTCTCAAGCTTCA. The zcxcl8 splice blocking morpholino for IL8 knock-down (5’-TATTTATGCTTACTTGACAATGATC-3’) was prepared and injected as described earlier [30]. To generate neutrophil depleted-embryos, csf3r translation morpholino (5’-GAAGCACAAGCGAGACGGATGCCAT-3’) targeting the csf3r gene was used [31]. For the selective depletion of macrophages into embryos, lipo-clodronate (Lipo-C) [56] was injected into the caudal vein of 24 hpf embryos as previously reported [17,55].
Dead cells in living zebrafish embryos were detected using Acridine Orange (AO), as described [17]. For detecting ROS, living embryos were soaked in 5 μM CellROX Green Reagent (Invitrogen) in Hanks’s Balanced Salt Solution (HBSS) for 30 min at 28.5°C, followed by two washes in HBSS, then transferred into a dish for fluorescent microscopic observation and analyses.
To determine cytokine/chemokine expression levels, total RNA from a pool of 10–15 larvae per biological experiment was isolated using the Nucleospin RNAII kit (Macherey-Nagel). cDNA synthesis was performed with M-MLV reverse transcriptase (Invitrogen) and then quantitative RT-PCR was performed using a homemade SYBR Green mix on a LightCycler 480 instrument (Roche) as described [57] with the following pairs of primers (sense and antisense): ef1α, TTCTGTTACCTGGCAAAGGG and TTCAGTTTGTCCAACACCCA; tnfα, TTCACGCTCCATAAGACCCA and CCGTAGGATTCAGAAAAGCG; il8, CCTGGCATTTCTGACCATCAT and GATCTCCTGTCCAGTTGTCAT; ifnγ2, TGCACACCCCATCTTCCTGCGAA and GTGTTGCTTCTCTATAGACACGCTT; il1β, TGGACTTCGCAGCACAAAATG and GTTCACTTCACGCTCTTGGATG. Each experiment was run in triplicate. qRT-PCR datas were calculated using the ΔCt or ΔΔCt method and normalized to the housekeeping gene ef1α.
To quantify bacterial loads, granulomas, cords and neutrophil recruitment, infected larvae were tricaine-anesthetized, positioned on 35-mm dishes, immobilized in 1% low-melting-point agarose and covered with water containing tricaine. Bright-field and fluorescence pictures of live infected embryos were taken with a Zeiss microscope equipped with a Zeiss Plan Neo Fluor Z 1x/0.25 FWD objective and a Axiocam503 monochrome (Zeiss) camera, with acquisition and processing using ZEN 2 (blue edition). Evaluation of intracellular mycobacterial growth, enumeration of macrophages recruitment, phagocytosis, mortality and granuloma organization, infected embryo were prepared for fixed microscopy. Animals were tricaine-anesthetized and fixed overnight at 4°C in 4% (vol/vol) paraformaldehyde in PBS, then washed twice in PBS, and transferred gradually from PBS to 50% (vol/vol) glycerol for microscopic observation. Confocal microscopy was performed using a Leica SPE upright microscope with 40x ACS APO 1.15 oil objective. Images were captured by LAS-AF software (Leica Microsystems).
Overlays of fluorescent and bright-field images and 2D reconstructions of images stacks were produced, assembled and adjusted using LAS-AF software or GIMP 2.6 freeware. Three-dimensional volume reconstitutions and movies were performed using Imaris 7.0 software (Bitplan AG).
Statistical analyses were performed using Prism 4.0 (Graphpad, Inc) or R 3.0.3 and detailed in each figure legend. *p< 0.05; **p< 0.01; ***p<0.001; ****p< 0.0001; ns, not significant (p≥ 0.05).
|
10.1371/journal.pgen.1000552 | Mre11-Rad50 Promotes Rapid Repair of DNA Damage in the Polyploid Archaeon Haloferax volcanii by Restraining Homologous Recombination | Polyploidy is frequent in nature and is a hallmark of cancer cells, but little is known about the strategy of DNA repair in polyploid organisms. We have studied DNA repair in the polyploid archaeon Haloferax volcanii, which contains up to 20 genome copies. We have focused on the role of Mre11 and Rad50 proteins, which are found in all domains of life and which form a complex that binds to and coordinates the repair of DNA double-strand breaks (DSBs). Surprisingly, mre11 rad50 mutants are more resistant to DNA damage than the wild-type. However, wild-type cells recover faster from DNA damage, and pulsed-field gel electrophoresis shows that DNA double-strand breaks are repaired more slowly in mre11 rad50 mutants. Using a plasmid repair assay, we show that wild-type and mre11 rad50 cells use different strategies of DSB repair. In the wild-type, Mre11-Rad50 appears to prevent the repair of DSBs by homologous recombination (HR), allowing microhomology-mediated end-joining to act as the primary repair pathway. However, genetic analysis of recombination-defective radA mutants suggests that DNA repair in wild-type cells ultimately requires HR, therefore Mre11-Rad50 merely delays this mode of repair. In polyploid organisms, DSB repair by HR is potentially hazardous, since each DNA end will have multiple partners. We show that in the polyploid archaeon H. volcanii the repair of DSBs by HR is restrained by Mre11-Rad50. The unrestrained use of HR in mre11 rad50 mutants enhances cell survival but leads to slow recovery from DNA damage, presumably due to difficulties in the resolution of DNA repair intermediates. Our results suggest that recombination might be similarly repressed in other polyploid organisms and at repetitive sequences in haploid and diploid species.
| Most organisms contain only one or two copies of their genome, but in some species multiple copies are found. The presence of multiple genome copies (polyploidy) has profound implications for DNA repair and is frequently seen in cancer cells. We have studied DNA repair in the archaeon Haloferax volcanii, which contains up to 20 genome copies. Archaea are a third form of life distinct from bacteria and eukaryotes. We have focused on the DNA repair proteins Mre11 and Rad50, which are found in virtually all organisms and which in humans act to prevent cancer. Surprisingly, we have found that H. volcanii cells deficient in Mre11-Rad50 are more resistant to DNA damage than wild-type cells. The DNA damage resistance of mre11 rad50 mutant cells appears to be due to the exclusive use of homologous recombination, a DNA repair mechanism that is accurate but has the potential to generate genome rearrangements that require time to resolve. Correspondingly, we have found repair of DNA damage in mre11 rad50 mutants takes longer than in wild-type cells. Our results suggest that polyploid organisms employ a program of DNA repair that minimizes their reliance on homologous recombination.
| Bacterial and eukaryotic cells are normally assumed to be haploid and diploid, respectively, but polyploidy is surprisingly widespread. Polyploid cells can arise naturally during development of otherwise haploid or diploid organisms (e.g. hepatocytes), or as a consequence of cellular stress and disease (e.g. cancer, reviewed in [1]). Organisms that are constitutively polyploid are common amongst eukaryotes, and include plants, fish and amphibians. Polyploid bacteria include the radiotolerant species Deinococcus radiodurans, which harbors ∼8 copies of its genome [2], and Epulopiscium spp., which contain tens of thousands of genome copies [3]. Amongst archaea, Methanocaldococcus jannaschii, Halobacterium salinarum and Haloferax volcanii have been shown to be naturally polyploid [4],[5].
The presence of multiple genome copies affects many aspects of cell metabolism, in particular pathways of DNA repair. Since homologous recombination (HR) requires an identical genome copy, its usage for DNA repair is influenced by cell ploidy. When only one genome copy is present in the G1 phase of the eukaryotic cell cycle, DNA double-strand breaks (DSBs) are repaired by non-homologous end-joining (NHEJ), while HR is the predominant form of DSB repair in the G2 phase (reviewed in [6]). A further doubling of the ploidy of eukaryotic cells can result in increased reliance on HR, since genes involved in HR become essential for viability in tetraploid yeast [7]. In the presence of 8 genome copies in D. radiodurans, RecA-dependent HR is also required for DSB repair. However, HR is the second part of a two-stage DSB repair process, and is preceded by RecA-independent extended synthesis-dependent strand annealing [8].
It is a common assumption that additional genome copies might help protect polyploid cells from DNA damage. This is not the case, since tetraploid Saccharomyces cerevisiae cells are no more resistant to DNA damage than diploids [9],[10]. Furthermore, D. radiodurans cells have the same survival rate after ionizing radiation, whether they contain 4 or 10 genome copies [11].
We have undertaken a study of DNA repair in the halophilic archaeon H. volcanii, which is naturally polyploid and contains 10–20 copies of the genome, depending on growth phase [4]. Archaea are of great interest in their own right, and share many core components of their DNA processing machinery with eukaryotes (reviewed in [12]). We have focused on the role of the Mre11-Rad50 complex, which is present in all domains of life and is involved in several pathways of DSB repair including HR and NHEJ (reviewed in [13]). Mre11 is a nuclease, while Rad50 consists of two globular DNA-binding domains (reviewed in [14]). Together, Mre11 and Rad50 form a complex that binds to and tethers DNA ends, in order to erect a scaffold for the subsequent processing and repair of DSBs [15]–[17]. For instance, Mre11-Rad50 has recently been shown to initiate 5′-strand resection at DSBs [18],[19].
Mre11-Rad50 is critical for DSB repair, and S. cerevisiae mutants in mre11 or rad50 are acutely sensitive to agents that induce DSBs [20],[21]. Mre11 is one of the first proteins to localize to the sites of DSBs [22], where it activates the ATM/Tel1 kinase that is central to the DNA damage-induced checkpoint [23]. It is noteworthy that Mre11 foci at the sites of DSBs dissociate before the appearance of “classical” HR proteins such as Rad51 and Rad52 [22]. The temporal separation between binding by Mre11-Rad50 and the subsequent repair of DSBs presumably allows for the appropriate pathway (HR or NHEJ) to be chosen. Mre11-Rad50 is also essential for the repair of meiotic DSBs in both S. cerevisiae and Schizosaccharomyces pombe, but only in S. cerevisiae does the formation of meiotic DSBs depend on Mre11-Rad50 [24],[25]. Similar differences are found with respect to NHEJ, where Mre11-Rad50 is required for NHEJ in S. cerevisiae but not in S. pombe [26]–[28].
The bacterial homolog of Mre11-Rad50 is SbcCD. Sensitivity to DNA damage is also seen in sbcCD mutants of some bacterial species such as D. radiodurans and Bacillus subtilis [29],[30], and D. radiodurans sbcCD cells exhibit delayed repair of DSBs after ionizing radiation [29]. Similarly retarded kinetics of DSB repair is seen in mre11 mutants of the archaeon Halobacterium sp. NRC-1, although in this species deletion of mre11 or rad50 does not result in sensitivity to DNA damage [31]. We have deleted mre11 and rad50 genes in the polyploid archaeon H. volcanii and have found that mutants are more resistant to DNA damage than the wild-type. Our results indicate that the use of HR is restrained by Mre11-Rad50, and that the unrestrained use of HR in mre11 rad50 mutants enhances cell survival but leads to slower recovery from DNA damage.
The mre11 and rad50 genes were identified in an operon in the H. volcanii genome (Figure 1A). All motifs diagnostic for Mre11 and Rad50 [32] are conserved in the H. volcanii proteins (Figure S1). Genes for NurA and HerA, which cluster with mre11 and rad50 in thermophilic archaea [33],[34], are not apparent in the H. volcanii sequence. Xrs2 and Nbs1, which form part of the Mre11-Rad50 complex in yeast and higher eukaryotes respectively [13],[14], are not found in archaea. Sequence analysis of the mre11-rad50 region failed to identify additional genes in the operon.
Deletion mutants of rad50, mre11, and mre11 rad50 were constructed using a gene knockout system for H. volcanii (Figure 1B) [35]. The generation time of the mutants during exponential growth in rich medium (Hv-YPC broth) was similar to the wild-type (WT) (∼2 hours). However, in a pairwise growth competition assay, the mre11 rad50 mutant was out-competed by the WT (Figure 1C). The growth advantage of the WT is ∼1% per generation.
We examined the sensitivity of H. volcanii rad50, mre11 and mre11 rad50 mutants to DNA damage, specifically ultraviolet (UV) and γ radiation, the radiomimetic chemical phleomycin, and the alkylating agent methyl methanesulphonate (MMS). In all cases, the mutants are significantly more resistant to DNA damage than the WT strain (Figure 2). The UV sensitivity of mre11 rad50 mutants was restored to WT levels by expression of the mre11-rad50 operon from a replicative plasmid (pTA795), confirming that hyper-resistance to DNA damage is due to mre11 rad50 deletion (data not shown).
After UV irradiation, mre11 rad50 colonies were smaller than WT colonies (Figure 3A). The small mre11 rad50 colonies yielded normal-sized colonies on restreaking (data not shown), therefore the small-colony phenotype is probably due to a temporary delay in growth of mre11 rad50 cells after UV irradiation. To investigate this further, we carried out pairwise growth competition assays after UV irradiation (Figure 3B). After 180 J/m2 UV, only a small fraction of WT cells survive, but these survivors exhibit a rapid recovery that results in restoration of the WT cell fraction to pre-UV levels after 24 hours. Irradiation with 60 J/m2 UV results in <50% cell death and there is no difference in survival between WT and mutant cells (Figure 2), but pairwise growth competition shows that the WT has a significantly faster recovery from DNA damage than the mre11 rad50 mutant (Figure 3B).
Repair of DNA damage was monitored directly by pulsed-field gel electrophoresis. Within 1 hour after irradiation with 180 J/m2 UV, the genome is fragmented by DSBs (Figure 3C). Formation of these DSBs requires processing of UV-induced DNA damage, since the total disappearance of bands corresponding to an intact chromosome is not seen until 10–30 minutes after irradiation (data not shown). Bands corresponding to an intact chromosome are visible again by 16 hours in both WT and mre11 rad50 cells, but in contrast to the WT, the majority of DNA in the mre11 rad50 mutant is found in a characteristic smear of broken fragments. By 24 hours, the WT has reconstituted the genome, whereas fragmented DNA still persists in the mre11 rad50 mutant. Therefore, repair of DNA damage is more rapid in the WT than the mutant.
We examined DSB repair using the recombination assay shown in Figure 4A. Cells are transformed with a replicative plasmid carrying the beta-galactosidase gene bgaHa (Text S1, Figure S3). This allele can recombine with a mutant bgaHa-Kp allele on the chromosome. If the plasmid is cut with KpnI prior to transformation, the DSB can be repaired either by end-joining or by HR. Repair by accurate end-joining (religation) results in colonies that stain blue with Xgal, whereas inaccurate end-joining or HR results in colonies that remain red (bgaHa−). Inaccurate end-joining and HR can be distinguished by a restriction digest of plasmid DNA, since the StuI site in the bgaHa-Kp allele is diagnostic of HR (Figure 4A and 4C).
If the plasmid is left uncut, gene conversion of the plasmid-borne bgaHa allele does not differ significantly between the WT and the mre11 rad50 mutants (Figure 4B, left graph). Therefore, inactivation of Mre11-Rad50 does not affect HR of circular DNA. If the plasmid is cut with KpnI, the efficiency of DSB repair is similar in the WT and mutants, however the mode of DSB repair differs markedly between these strains (Figure 4B, right graph). In the WT, the vast majority of DSBs are repaired by accurate end-joining, with very little contribution of HR or inaccurate end-joining (Figure 4B and 4C). By contrast, in mre11 rad50 mutants most repair is by HR, while accurate end-joining is reduced by ∼50%. It is notable that in the WT, the frequency of HR between the plasmid and chromosome is reduced almost 300-fold when the plasmid is cut with KpnI, whereas in the mre11 rad50 mutants HR is not affected by the presence of a DSB (Figure 4B, compare left and right graphs). Therefore, Mre11-Rad50 restrains the use of HR at DSBs.
Since Mre11 is a nuclease and Mre11-Rad50 has been shown to initiate 5′-strand resection at DSBs [18],[19],[36], it is possible that H. volcanii Mre11-Rad50 might prevent HR by nucleolytic degradation of DSBs. To test this, we modified the DSB assay by inserting a trpA marker at the end of the plasmid-borne bgaHa gene (Figure S2). Degradation of KpnI-cut DNA is measured by loss of the trpA marker, but HR with the chromosomal bgaHa-Kp allele is still possible using downstream sequences. There was no significant difference between WT and mutant strains in the frequency of marker loss, suggesting that Mre11 is not responsible for DSB degradation in H. volcanii.
We also measured the fraction of single-stranded DNA in WT and mre11 rad50 cells, by slot-blotting of native (undenatured) and denatured genomic DNA, and hybridization with a genomic DNA probe (Figure S2C). After irradiation with 180 J/m2 UV, the fraction of single-stranded DNA increases ∼2.5-fold (Figure S2D), but there was no significant difference between WT and mre11 rad50 strains. Thus, Mre11-Rad50 is not responsible for the formation of single-stranded DNA after UV damage.
Mre11 has been shown to facilitate end-joining at microhomologies in vitro [37]. In WT H. volcanii, repair of the cut plasmid is primarily by accurate re-ligation of cohesive KpnI ends. The efficiency of this end-joining is reduced by ∼50% in mre11 rad50 mutants, suggesting that it is partially dependent on Mre11-Rad50. End-joining also depends on micro-homology, since its efficiency is decreased by >90% if the DSB is blunt-ended (data not shown). Furthermore, plasmid repair events by inaccurate end-joining are characterized by deletions of the bgaHa gene, which are flanked by microhomologies of 3–5 bp (Figure S2E). This repair pathway is Ku-independent, since H. volcanii (and almost all other archaea) does not encode a Ku homolog [38], and resembles the microhomology-mediated end-joining seen in other organisms [39].
RadA is the H. volcanii ortholog of RecA/Rad51 recombinase and is more similar to eukaryotic Rad51 than bacterial RecA [40]. Mutants of radA are slow-growing, sensitive to DNA damage, and completely deficient in recombination [41]. However, microhomology-mediated end-joining of a cut plasmid (as described in Figure 4) is observed in a radA mutant (H607, data not shown). Deletion of radA in an mre11 rad50 background leads to a growth defect that is worse than that seen with radA alone (Figure 5A). This might suggest that Mre11-Rad50 and RadA act in different pathways of DNA repair. However, we found that radA mre11 rad50 mutants are no more sensitive to UV radiation than radA strains (Figure 5B). This indicates that in both WT and mre11 rad50 cells, the repair of UV-induced DNA damage requires RadA, and therefore most likely HR.
We have deleted the mre11 and rad50 genes of the polyploid archaeon H. volcanii and have found that: (1) mre11 rad50 mutants are hyper-resistant to DNA damage, but recover from DNA damage and repair DSBs more slowly than the WT; (2) mre11 rad50 mutants exhibit more homologous recombination at DSBs than the WT; (3) RadA recombinase is ultimately required for DNA repair.
In S. cerevisiae, null mutations of mre11 or rad50 confer sensitivity to ionizing radiation and MMS, although not to UV [20],[21]. In bacteria, Deinococcus radiodurans and Bacillus subtilis sbcCD mutants are sensitive to DNA damage [29],[30]. However, it is not always the case that lack of Mre11-Rad50 or SbcCD results in decreased resistance to DNA damage. E. coli sbcC mutants and mre11 rad50 mutants of the archaeon Halobacterium sp. NRC-1 do not show increased sensitivity to DNA damage [31],[42]. Intriguingly, we have found that H. volcanii mre11 and rad50 mutants are more resistant to DNA damage than the WT (Figure 2).
Mutations in DNA repair genes that increase resistance to DNA damage have been reported elsewhere. Defects in DNA ligase IV or the end-binding Ku70-Ku80 complex increase the resistance of chicken DT40 and yeast cells to high doses of γ radiation or phleomycin [43],[44]. The increased DNA damage resistance of NHEJ-defective cells is thought to be due to a failure to suppress HR [45]. Our work shows that this is also the case for mre11 rad50 mutants of H. volcanii, since the hyper-resistance to DNA damage correlates with increased repair by HR.
Although mre11 rad50 mutants of H. volcanii are more resistant to DNA damage than the WT, they take longer to recover. This is evident in the size of colonies formed by surviving cells (Figure 3A), and in pairwise competition assays after UV irradiation (Figure 3B, left graph). WT survivors are less numerous but recover more rapidly than mre11 rad50 mutants after irradiation with high doses of UV. This suggests that WT and mre11 rad50 cells use two different DNA repair strategies. The speed of the Mre11-Rad50-dependent repair strategy gives WT cells a long-term advantage, in spite of being incapable of repairing high levels of DNA damage. At lower doses of DNA damage (60 J/m2 UV), the advantage of the WT repair strategy is also apparent (Figure 3B, right graph). One reason might be slower repair of DSBs in mre11 rad50 mutants, as indicated by pulsed-field gels of genomic DNA after UV irradiation (Figure 3C). A similar delay in the repair of DSBs has been observed in sbcCD mutants of D. radiodurans and mre11 mutants of the archaeon Halobacterium sp. NRC-1 [29],[31].
Our results indicate that the DNA repair strategy used in mre11 rad50 cells involves unrestrained HR. In WT cells, cut plasmid molecules are repaired by microhomology-mediated end-joining, whereas in the absence of Mre11-Rad50, HR is increased ∼100-fold and is the predominant mode of repair (Figure 4). This is in contrast to S. cerevisiae, where deletion of mre11 reduces HR between plasmid and chromosome ∼20-fold [46]. Furthermore, in S. cerevisiae the presence of a DSB stimulates HR between plasmid and chromosome [47], whereas in WT H. volcanii the presence of a DSB reduces HR between plasmid and chromosome (Figure 4B, compare left and right graphs). This suggests that in H. volcanii, the preferred substrate for HR is not a DSB.
There are two (non-exclusive) hypotheses to account for our results: (1) Mre11-Rad50 binds to DSBs and directly prevents HR (e.g. by blocking assembly of RadA filaments); (2) Mre11-Rad50 stimulates an alternative pathway of DSB repair (e.g. by microhomology-mediated end-joining), thereby removing the substrate for HR. If the sole function of Mre11-Rad50 is to promote an alternative to HR, then mutations that eliminate HR should be synergistic with deletion of mre11 rad50. Mutation of radA renders cells sensitive to DNA damage and deficient in HR [41], and we show here that radA mre11 rad50 mutants are no more sensitive to UV radiation than radA cells (Figure 5B). RadA might have other roles in addition to HR, such as activation of an SOS response to DNA damage, as seen in bacteria [48]. However, to date there is no evidence for an SOS response in archaea [49],[50]. For these reasons we favor the first hypothesis, where Mre11-Rad50 directly restrains HR and allows another pathway to act as the primary mode of DSB repair. The phenotype of radA mre11 rad50 mutants suggests that HR is ultimately required for repair of DSBs, therefore the restraint on HR imposed by Mre11-Rad50 can only be temporary.
Why does H. volcanii Mre11-Rad50 appear to act differently when compared to other organisms? We suggest that HR is restrained in H. volcanii because this species is highly polyploid [4]. Coordinating HR is likely to be problematic when each DSB has 20 partners to choose from. This problem is exacerbated in organisms with a circular genome, since DSB repair by HR runs the risk of generating chromosome concatemers, which require resolution before cell division. However, our results with radA mutants suggest that DNA repair ultimately requires HR. We propose that Mre11-Rad50 temporarily restrains HR, and promotes (directly or indirectly) a repair mechanism that reduces the number of DNA ends, which ultimately have to be repaired by HR. A two-step process of DNA repair has been proposed for the polyploid bacterium D. radiodurans, where DNA fragments are first reassembled by extended synthesis-dependent strand annealing (ESDSA) before chromosome reconstitution by HR [8]. However, the initial step used in H. volcanii is unlikely to be ESDSA, since we find no evidence for increased DNA synthesis after UV irradiation (data not shown). As we suggest above, H. volcanii might use microhomology-mediated end-joining, but other mechanisms for the initial DNA repair are also possible. In any case, this initial mechanism appears to be incapable of repairing large numbers of DSBs, as evident by the hyper-resistance to DNA damage seen in mre11 rad50 mutants.
Mre11-Rad50 might also restrain HR in order to promote a physiological change, such as induction of a DNA damage response that allows time for processing of DSBs, and/or arrest at a cell cycle checkpoint. In eukaryotes, hyper-recombination is a common phenotype of checkpoint defective cells [51], and the small size of H. volcanii mre11 rad50 colonies after UV irradiation (Figure 3A) is strikingly reminiscent of yeast cells that have adapted to the DNA damage checkpoint and re-entered the cell cycle [52].
Why would evolution restrain HR, if the unrestrained use of HR (in H. volcanii mre11 rad50 mutants) results in higher cell survival? Yeast and chicken DT40 cells defective in NHEJ show increased DNA damage resistance, which is suggested to be due to a failure to suppress HR [43],[44]. Perhaps the cost of HR results from the time taken to repair DNA damage, and this cost is particularly acute for polyploid organisms. In the polyploid species D. radiodurans and Halobacterium sp. NRC-1, mutations in sbcCD and mre11, respectively, result in slower repair of DSBs [29],[31], similar to what we have observed for H. volcanii. We suggest that HR is restrained in these and other polyploid organisms, and at repetitive sequences in haploid/diploid species, for example the rDNA locus [53]. When many copies of a gene or genome are present, restraining HR might be necessary to prevent each DNA end from engaging with multiple partners.
Unless stated otherwise, chemicals were from Sigma and restriction enzymes from New England Biolabs. Standard molecular techniques were used [54].
H. volcanii strains are shown in Table 1, plasmids in Table S1, and oligonucleotides in Table S2. H. volcanii strains were grown at 45°C on complete (Hv-YPC) or casamino acids (Hv-Ca) agar, or in Hv-YPC or Hv-Ca broth, as described previously [55],[56]. To estimate generation times, a culture of Hv-YPC broth (+thymidine) was inoculated with ∼103 cells/ml, viable cells/ml was determined by plating aliquots at regular intervals. Isolation of genomic and plasmid DNA, and transformation of H. volcanii were carried out as described previously [55],[57].
The rad50 gene was identified in the genome sequence. A 469 bp fragment of rad50 was amplified by PCR and used to probe a Southern blot of genomic DNA digested with MluI; a 5.3 kb fragment hybridized with the probe. A genomic library of MluI 5.3 kb fragments was constructed and screened by colony hybridization with the rad50 probe. One clone (pTA42) was found to contain the mre11-rad50 operon (Figure 1A). The sequence of the mre11-rad50 operon has been deposited in the EMBL database (accession number AJ635369).
To generate the Δrad50 construct, a BspEI–BsaBI rad50 fragment of pTA42 was deleted (Figure 1A). The Δmre11 construct was generated by PCR, to ensure precise gene deletion that would not exert a polar effect on rad50 (see Table S2). To generate the Δmre11Δrad50 construct, the BspEI–BsaBI rad50 fragment was deleted from the Δmre11 construct. A pyrE2 marker from pGB70 [35] was inserted into plasmids carrying Δrad50, Δmre11 and Δmre11 Δrad50 constructs, generating pTA137, pTA138 and pTA171 respectively, which were used to construct deletion strains by a gene knockout system [35]. Deletion of radA in mre11 rad50 strains required complementation by a plasmid-borne radA gene (see Text S1 and Figure S4).
To compare strains during normal growth, a 10 ml culture of Hv-YPC broth was inoculated with ∼103 WT and ∼104 mre11 rad50 cells (1∶10 ratio). The mixed culture was grown for 2 days (∼108 cell/ml), diluted 1000-fold and ∼104 cells was used to inoculate 10 ml Hv-YPC; this was repeated a further 3 times. After each inoculation, the ratio of WT and mutant cells was determined by plating, transferring 100 colonies to nylon membranes and probing with the 469 bp rad50 PCR product. To compare the ratio of WT and mutant cells after UV irradiation, bgaHa+ derivatives of H115 and H204 that stain blue with Xgal (5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside) were generated (H642 and H645, respectively, Text S1 and Figure S3). Cultures were grown to ∼108 cell/ml, centrifuged and resuspended in an equal volume of 18% SW (salt water) [55]. A 1∶1 mixture of WT and mre11 rad50 cells (differentially marked with bgaHa+ or bgaHa-Kp) was exposed to 180 J/m2 UV (254 nm, 1 J/m2/s) or left unirradiated. The mixed culture was centrifuged, resuspended in an equal volume of Hv-YPC broth and grown in the dark at 45°C. Aliquots of 10 µl were taken at the intervals shown and plated on Hv-YPC. After 5 days incubation, plates were sprayed with Xgal solution (BlueTech, Mirador) and cells scored the next day. Pairwise growth competition after 60 J/m2 UV was performed in a similar manner, except that after irradiation the culture was diluted 106-fold in 1 ml Hv-YPC broth and grown in the dark for 3 days.
For radiation assays, cultures were grown to ∼108 cell/ml, diluted in 18% SW and 20 µl aliquots spotted on Hv-YPC plates. Once spots had dried, cells were exposed to UV or γ radiation (137Cs, 395 Gy/s). After UV exposure, cells were shielded from visible light. For chemical mutagen assays, cultures were divided into 1 ml aliquots, and phleomycin (Apollo Scientific) or methyl methanesulphonate was added. Cultures were returned to 45°C for 1 hour, diluted in 18% SW and plated on Hv-YPC. Survivors were counted after 5–14 days incubation.
Cultures were grown to ∼108 cell/ml, centrifuged, and resuspended in an equal volume of 18% SW. Cells were exposed to 180 J/m2 UV with shaking, centrifuged, resuspended in an equal volume of Hv-YPC broth and grown in the dark at 45°C. Samples were taken at the time indicated. Cells were centrifuged, resuspended in 20 µl 18% SW and embedded in 100 µl plugs at a final concentration of 0.8% SeaPlaque GTG agarose (Cambrex) prepared in 18% SW. Agarose plugs were incubated for ≥16 hours at 45°C in lysis solution (20 mM Tris-HCl pH 8.8, 500 mM EDTA pH 8, 1% N-lauroylsarcosine, 1 mg/ml proteinase K), then transferred to fresh lysis solution containing 10 µg/ml RNaseA for 4 hours at 45°C. Plugs were incubated in wash solution (25 mM Tris-HCl pH 7.5, 100 mM EDTA pH 8) for 30 minutes at 37°C, then transferred to fresh wash solution containing 1 mM phenylmethane sulfonyl fluoride for 1 hour at 37°C. Plugs were washed two more times for 30 minutes each, the final time in 1/10× wash solution. The plugs were transferred into 300 µL of restriction enzyme buffer and incubated for 30 min at 37°C, the buffer was replaced, and 40 U of PmeI was added. Chromosomal DNA was digested overnight and the fragments separated on a 1.2% agarose gel in 0.5× TBE using a CHEF Mapper PFGE system (Bio-Rad) running with a gradient voltage of 6 V/cm, an included angle of 120°, and initial and final switch times of 0.64 sec and 1 min 13.22 sec, respectively, with a run time of 20 h 46 min at 14°C.
Plasmids pTA274 and pTA329 were cut to completion with KpnI or AarI, and purified by excision from agarose gels and extraction using QIAquick columns (Qiagen). Equal aliquots were treated either with shrimp alkaline phosphatase (Promega) or mock-dephosphorylated (no phosphatase). DNA concentration was determined by A260, and 0.1–1 µg was used to transform WT, rad50, mre11 and mre11 rad50 strains. Uncut plasmid was used to measure transformation efficiency. Cells were plated on Hv-Ca (+tryptophan for H292–295 transformed with pTA329). After 5 days incubation, plates were sprayed with Xgal solution and transformants scored the next day. To measure loss of the trpA marker in pTA329, red transformants of H292–295 were patched onto Hv-Ca agar without tryptophan, and scored after 3 days incubation.
|
10.1371/journal.ppat.1004450 | The Host Protein Calprotectin Modulates the Helicobacter pylori cag Type IV Secretion System via Zinc Sequestration | Transition metals are necessary for all forms of life including microorganisms, evidenced by the fact that 30% of all proteins are predicted to interact with a metal cofactor. Through a process termed nutritional immunity, the host actively sequesters essential nutrient metals away from invading pathogenic bacteria. Neutrophils participate in this process by producing several metal chelating proteins, including lactoferrin and calprotectin (CP). As neutrophils are an important component of the inflammatory response directed against the bacterium Helicobacter pylori, a major risk factor for gastric cancer, it was hypothesized that CP plays a role in the host response to H. pylori. Utilizing a murine model of H. pylori infection and gastric epithelial cell co-cultures, the role CP plays in modifying H. pylori -host interactions and the function of the cag Type IV Secretion System (cag T4SS) was investigated. This study indicates elevated gastric levels of CP are associated with the infiltration of neutrophils to the H. pylori-infected tissue. When infected with an H. pylori strain harboring a functional cag T4SS, calprotectin-deficient mice exhibited decreased bacterial burdens and a trend toward increased cag T4SS -dependent inflammation compared to wild-type mice. In vitro data demonstrate that culturing H. pylori with sub-inhibitory doses of CP reduces the activity of the cag T4SS and the biogenesis of cag T4SS-associated pili in a zinc-dependent fashion. Taken together, these data indicate that zinc homeostasis plays a role in regulating the proinflammatory activity of the cag T4SS.
| Helicobacter pylori is a bacterium that colonizes the stomach and causes gastric diseases. Some strains of H. pylori possess a secretion system that has the capacity to inject a cancer-causing protein into host cells. The activity of this secretion system contributes to the development of inflammation and is linked to the development of gastric cancer. Here, we show that the host protein calprotectin, which has the ability to bind and sequester nutrient metals from invading pathogens, can directly repress H. pylori secretory activity and the production of secretion-associated pili in a zinc-dependent manner. H. pylori-infected animals lacking calprotectin trend toward having more gastric inflammation and a significantly lower bacterial burden than infected animals that express calprotectin; these differences are not observed when animals are infected with a strain of H. pylori that lacks an active secretion system. Thus, a better understanding of how nutritional immunity modulates this secretion system could help us develop novel antimicrobial therapeutic strategies targeting secretory processes in H. pylori.
| Helicobacter pylori is a Gram-negative bacterial pathogen that colonizes half of the world's population and contributes to a variety of disease outcomes, including peptic or duodenal ulcer disease, gastric adenocarcinoma, and mucosal-associated lymphoid tissue (MALT) lymphoma [1]. The most common manifestation of H. pylori-related disease is chronic gastric inflammation (non-atrophic chronic gastritis), which can potentially advance to multifocal atrophic gastritis, a precancerous lesion [2]. H. pylori-induced gastritis is characterized by neutrophil and mononuclear leukocyte infiltration to the lamina propria, involving cells in both the innate and adaptive arms of the immune response.
This gastric mucosal inflammatory response to H. pylori is enhanced if persons are infected with strains that possess a cag- type IV secretion system (cag T4SS). The cag T4SS is a macromolecular assembly that is responsible for translocating the oncogenic effector molecule, CagA and peptidoglycan, into host cells [3], [4]. These translocated effectors elicit a variety of host cell responses, including activation of nuclear factor κB (NFκB) and secretion of the proinflammatory cytokines, IL-1β and IL-8 [5]–[8]. The latter is associated with recruitment of innate immune cells, including neutrophils [9]. Neutrophil recruitment to the gastric mucosa is also enhanced by Th17 and Th1 responses.
Neutrophils are capable of controlling microbial infections via phagocytosis and subsequent production of reactive oxygen and reactive nitrogen intermediates, neutrophil extracellular trap (NET) formation, and production of antimicrobial factors [10], [11]. One such example is calprotectin (CP) [12]. CP, a heterodimer of S100A8 and S100A9 subunits (also known as Mrp8/14, calgranulin A/B, cystic fibrosis antigen). CP comprises about 50% of the neutrophil's cytoplasmic protein content and is a critical component of the host nutrient withholding process termed nutritional immunity [13], [14]. Humans and other mammals restrict access to essential metals through this nutritional immunity mechanism as a means to prevent infection with pathogenic organisms.
CP binds manganese and zinc with high affinity, effectively starving bacteria of these essential nutrients. There are two transition metal binding sites in CP; site 1 (S1; 6 His site) binds manganese and zinc, and site 2 (S2; 3 His Asp site) binds zinc only [15], [16]. Mutagenesis of CP's two metal binding sites has produced a site 1 mutant (ΔS1) that binds zinc only, a site 2 mutant (ΔS2) capable of binding both manganese and zinc, and a double site mutant (DS CP) incapable of binding manganese and zinc. Previous reports have indicated that CP exhibits antimicrobial activity against numerous microorganisms, including Salmonella enterica serovar Typhimurium, Staphylococcus aureus, Escherichia coli, Borrelia burgdorferi, Listeria monocytogenes, Candida albicans, Acinetobacter baumannii, Staphylococcus epidermidis, Staphylococcus lugdunensis, Enterococcus faecalis, Pseudomonas aeruginosa, and Shigella flexneri [12], [14], [15], [17]–[22]. CP has been demonstrated to enhance pathogenic Salmonella persistence in the inflamed gut [18] as well as increase neutrophil killing of S. aureus [14]. Concentrations of calprotectin in the tissue have been reported to be as high as 20 mg/ml in response to bacterial infections [23]. Expression of CP subunits S100A8 and S100A9 in inflamed gastric tissues of H. pylori-infected persons has been reported in the literature [24]. However, the interaction between CP and H. pylori has not been previously investigated.
H. pylori-associated inflammation is dynamic and recently published results indicate that H. pylori can modulate its cag T4SS activity in response to inflammation [25]. Here, we report a study to determine the role of CP in control of H. pylori colonization and pathogenesis. We find that gastric CP levels are elevated in H. pylori-infected humans and rodents, and that most of the CP localizes to neutrophils in H. pylori-infected tissues. We also demonstrate that H. pylori-infected CP-deficient mice (A9−/− mice) have decreased bacterial burden and a trend towards increased gastric inflammation compared to infected WT mice, a phenotype which is not observed in mice infected with an H. pylori isogenic cagE mutant, which lacks cag T4SS activity. Finally, we show that CP represses cag T4SS activity and cag T4SS-associated pilus production via zinc sequestration.
In order to determine if CP is elevated in the context of H. pylori infection, real-time RT-PCR analysis of s100a8 and s100a9 transcripts in RNA isolated from either mouse or human gastric tissue was performed. For the first mouse study, RNA was isolated from gastric tissue of mice that had been infected with H. pylori PMSS1 or SS1 for 1, 2, and 3 months and s100a8 and s100a9 expression was compared to that of uninfected mice. CP subunit s100a8 was significantly increased in gastric tissue in response to H. pylori infection (Figure 1A). Transcript levels of CP subunit s100a9 did not significantly increase, but it has been demonstrated that the S100A9 protein is stabilized by increased S100A8 expression [18]. Corresponding inflammation scores are presented for these infections (Figure 1B). In a second study, human gastric biopsy samples were divided into H. pylori-negative samples and H. pylori-positive samples. Gene expression of both s100a8 and s100a9 subunits were elevated in H. pylori-infected biopsy samples compared to H. pylori-negative samples (Figure 1C).
CP was elevated in response to H. pylori infection in both human and murine stomach tissues, a phenomenon we hypothesize is driven by the increased presence of neutrophils. An immunohistochemical staining approach was used to evaluate the localization of CP within H. pylori PMSS1 or SS1-infected murine stomach tissues. Microscopy analyses revealed that the majority of CP is localized in association with neutrophils within the gastric tissue (Figure 1D).
Since immunohistochemistry staining revealed that CP was mainly localized in proximity to host neutrophils (Figure 1D) and since CP comprises 40–60% of the total protein in the neutrophil cytoplasm [26], [27], we hypothesized that recruitment of neutrophils to the H. pylori-infected stomachs correlates with increases in CP expression. To test this hypothesis, IL-17 receptor A-deficient mice (IL-17RA-/-) were used. Previously published data indicate that these mice have a defect in IL-17 signaling, a prerequisite for the maintenance of neutrophil recruitment to the stomach during chronic H. pylori infection [28]. At 3 months post-infection, H. pylori infected IL-17RA-/- mice exhibit significantly decreased PMN infiltration compared to H. pylori infected WT mice [28]. Thus, the 3 month time point was chosen for these analyses. Real-time RT-PCR analysis of CP subunit expression revealed that PMSS1-infected IL-17RA-/- mice have diminished s100a8 and s100a9 expression compared to PMSS1-infected wild-type (WT) animals (Figure 1E), demonstrating that increased abundance of CP was correlated to the presence of a neutrophilic infiltrate.
CP has been demonstrated to inhibit bacterial growth via sequestration of nutrient manganese and zinc [16]–[18]. We hypothesized that CP is elevated in response to H. pylori infection as part of a host strategy to inhibit bacterial proliferation within the gastric niche. To test this proposal, in vitro growth assays were performed in modified bacteriological medium. Analysis of bacterial growth curves (OD600) and colony forming units (CFU/mL) revealed that wild-type CP (CP) at 300 µg/mL significantly inhibited H. pylori growth (Figure 2 and Figure S1). The addition of exogenous manganese and zinc (50 µM of zinc chloride and 50 µM manganese chloride) restored growth to control levels. In addition to investigating the ability of CP to inhibit growth of H. pylori, the effects of three previously generated mutants of CP's metal binding sites (DS CP, ΔS1, and ΔS2) were investigated [15]. The DS CP harboring inactivation of both S1 (manganese and zinc binding) and S2 (zinc binding alone) sites was unable to inhibit bacterial growth (Table S1). The ΔS1 mutant at 1200 µg/mL was able to inhibit bacterial growth, as was the ΔS2 mutant (Table S1). These results indicate that CP inhibited H. pylori growth in vitro at concentrations above 300 µg/mL, and that the antibacterial activity is dependent on CP's ability to sequester metal.
Because CP can inhibit the growth of H. pylori in vitro, we hypothesized that this host protein would also contribute to control of the bacterial burden in vivo. To test this hypothesis, both WT and CP-deficient (A9-/-) mice were orogastrically infected with H. pylori strain SS1 or PMSS1; the former strain lacks a functional cag T4SS, while the latter expresses a functional cag T4SS. In analysis of animals at 6 weeks post-infection, A9-/- mice infected with the SS1 strain had significantly higher levels of H. pylori compared to WT mice infected with the SS1 strain (Figure 3A, p = 0.014), A9-/- mice infected with PMSS1 had significantly fewer CFU per gram of tissue compared to PMSS1-infected WT mice (Figure 3C, p = 0.0325). Bacterial burden and inflammation have been shown to have a reciprocal relationship in H. pylori models of murine infection [29]–[31]. This suggests that CP-deficient mice may have an increase in the inflammatory response to H. pylori. To test this hypothesis, inflammation in WT and A9-/- mice was evaluated via histological analysis and scoring. The A9-/- mice infected with SS1 did not have significant differences in inflammation compared to WT mice (Figure 3B), but A9-/- mice infected with PMSS1 had significantly higher inflammation scores than WT-infected animals (p = 0.04; Figure 3D). These results suggested that the absence of CP results in increased gastric inflammation during H. pylori infection with a strain expressing a functional cag T4SS (the SS1 strain lacks a functional cag T4SS), which may explain the decreased bacterial burden. We also investigated the effect of CP on chronic infection in the mice at later time points, up to 3 months post infection. At this timepoint, there were no significant differences in the colonization in the H. pylori-infected A9-/- mice compared to WT mice, but there was a trend toward increased inflammation in H. pylori-infected A9-/- mice compared to H. pylori-infected WT mice (SS1 infection p = 0.06; PMSS1 infection p = 0.13). We observed that the stomachs of A9-/- mice became infected with fungus by 3 months, which complicates the interpretation of results at this timepoint.
We hypothesized that CP modulates the cag T4SS since the activity of this virulence factor is associated with H. pylori-induced inflammation. To test this hypothesis, WT and A9-/- mice were infected with either PMSS1 or an isogenic H. pylori PMSS1 cagE mutant. Since CagE is the ATPase that powers cag T4SS assembly, H. pylori strains deficient for CagE do not have a functional cag T4SS and do not form cag T4SS pili [32]. Again, when infected with the PMSS1 strain, the bacterial burden was significantly lower (Figure 3E) and there was a trend toward higher inflammation in the A9-/- infected mice compared to the WT mice (Figure S2A). When infected with the PMSS1 cagE mutant, there was no significant difference in bacterial burden (Figure 3E) or inflammation (Figure S2) in A9-/- mice compared to WT mice at 6 weeks post-infection. To further quantify the gastritis in these mice, flow cytometry was performed at 6 weeks post-infection. There was a trend toward greater numbers of gastric PMNs and monocytes in PMSS1-infected A9-/- mice compared to PMSS1-infected WT mice (p = 0.057 and p = 0.097, respectively; Figure S2B, C). Assessment of CP levels (transcript levels of s100a8 and s100a9) by realtime rtPCR showed no difference in expression when comparing H. pylori infected WT mice infected with PMSS1 or the PMSS1 cagE mutant (Table S2). Thus differences in bacterial burden and inflammation between PMSS1-infected WT and the A9-/- mice are likely cag T4SS-dependent (Figure 3E and Figure S2). These data suggest that CP may repress the activity of the cag T4SS in vivo.
A functional cag T4SS translocates the effector molecule, CagA, into host cells, where it is then phosphorylated [25]. Moreover, a functional cag T4SS is necessary for activation of NFκB (nuclear factor kappa-light-chain-enhancer of activated B cells) in human AGS gastric epithelial cells, which is a result of both CagA translocation and peptidoglycan recognition by NOD1 [33]. As a result of these cellular signaling events, IL-8 is produced and secreted by the AGS gastric epithelial cells. Therefore, the functional activity of the cag T4SS was measured with three assays; CagA translocation and phosphorylation (Figure 4), NFκB activation (Figure 5A), and IL-8 secretion by gastric epithelial cells (Figure 5B).
As an initial assessment of cag T4SS function, an assay for CagA translocation into AGS cells based on detecting phosphorylated CagA was performed. H. pylori were cultured in the presence or absence of CP for 4–6 hours prior to a 4 hour co-culture with AGS cells. Following washes to remove unbound bacteria, lysates of H. pylori-bound AGS cells were generated and separated by SDS-PAGE. The levels of phospho-CagA and total CagA in the extracts were determined by immunoblot analyses [25]. As shown in Figure 4A and C, reduced levels of phosphorylated CagA in H. pylori-AGS co-cultures were observed when H. pylori was pre-treated with sub-inhibitory (growth) doses of CP.
As mentioned earlier, activity of the cag T4SS leads to NFĸB activation and IL-8 secretion by gastric epithelial cells. To assess the ability of CP to inhibit cellular activation, H. pylori were cultured in the presence or absence of CP for 4–6 hours prior to a 1 hour co-culture with an AGS-NFĸB luciferase reporter cell line. As show in Figure 5, CP inhibits the ability of the cag T4SS to activate NFĸB (Figure 5A). To determine if CP also inhibits IL-8 secretion in this co-culture system, H. pylori were cultured in the presence or absence of CP for 4–6 hours prior to a 4 hour co-culture with AGS. IL-8 secretion was significantly lower in supernatants from AGS co-cultures with H. pylori pre-treated with CP compared to AGS co-cultures with untreated H. pylori (Figure 5B). To investigate whether these results were an effect of reduced adherence, a bacterial adherence assay was performed. There were no significant changes in H. pylori adherence as a consequence of treatment with either CP or TPEN alone or in the presence of an exogenous source of nutrient zinc (Figure S3).
To investigate which metal binding sites are critical for this phenotype and to determine if this phenotype is due to manganese sequestration or zinc sequestration, the same experiments were performed with CP ΔS1, ΔS2, and DS mutants. Mutagenesis of either the S1 site (manganese and zinc binding) or the S2 site (zinc binding alone) decreased the efficiency of CP's ability to repress the CagA translocation (Figure 4) and cag T4SS activity (Figure 5A and B). While 200 µg/ml of CP was sufficient to repress CagA translocation, NFkB activation and IL-8 production, due to a reduced ability to bind metal, higher levels (600 µg/ml) of ΔS1 and ΔS2 CP were needed to observe the same effects. The DS mutant, which does not bind any metals, was unable to ablate the cag T4SS-dependent NFĸB activation and had no effect on IL-8 induction (Figure 5A and B). Since the S1 mutant, which can only bind zinc, is still functional, nutrient zinc sequestration is responsible for the cag T4SS repression phenotype. The observation that CP is acting on the cag T4SS through a zinc sequestration-dependent pathway was consistent with experiments showing that the inhibitory effects of CP on both CagA phosphorylation and the activity of the cag T4SS were reversible with the addition of an exogenous zinc source (Figure 4 and 5).
To confirm that the effect of CP on the repression of cag T4SS was due to zinc chelation, a synthetic metal chelator which preferentially binds zinc, N,N,N′,N′-tetrakis (2-pyridylmethyl) ethylenediamine (TPEN), was tested for its ability to repress the cag T4SS. Bacteria pre-exposed to TPEN at concentrations below that necessary to inhibit growth (5 µM, Figures 2 and S1B) translocated less CagA than did untreated bacteria, as determined by immunoblot analysis for phospho-CagA (Figure 4A). Bacteria pre-exposed to TPEN caused less NFkB activation within the host cell (Figure 5A) and showed a diminished capacity to induce IL-8 secretion (Figure 5B). The addition of exogenous zinc to the H. pylori cultures reversed the phenotype, and the activity of the cag T4SS was restored (Figure 4 and 5). These data suggest that the activity of the cag T4SS is reduced by CP through a zinc-sequestration-dependent process.
Our co-culture experiments demonstrated that metal sequestration by CP leads to abrogated phosphorylation of CagA and inhibition of downstream cellular activation in gastric epithelial cells. We next tested the hypothesis that the observed inhibition of the cag T4SS-dependent phenotype was attributable to inhibition of cag T4SS pilus production. To test this, field emission gun-scanning electron microscopy (FEG-SEM) analysis of the bacterial-gastric epithelial cell (H. pylori-AGS cell) co-cultures was performed to visualize the cag T4SS pili, as previously described [25], [32], [34]. Briefly, bacteria were cultured for 4–6 hours prior to co-culture with AGS cells in the presence or absence of WT or mutant forms of CP or the synthetic zinc chelator, TPEN, alone or in the presence of an exogenous source of nutrient zinc. Bacteria were co-cultured with gastric cells in the absence of additives for 4 hours before samples were processed and analyzed by high resolution FEG-SEM. Pre-treatment of H. pylori with CP or TPEN reduced the number of cag T4SS pili visible at the host pathogen interface (Figure 6). The addition of exogenous zinc restored WT pili formation, suggesting that the zinc-chelation is responsible for this reduced pilus formation. Similarly, the ΔS1 and ΔS2 mutant CP proteins were unable to repress pilus formation at the same concentration as WT (200 µg/mL). However, when increasing the concentration of the ΔS1 and ΔS2 mutants to high concentrations (600 µg/ml), cag T4SS pilus formation was repressed. The higher concentration of both ΔS1 and ΔS2 mutant CP proteins was consistent with earlier results, where 600 µg/ml of these mutant proteins was necessary to observe decreased NFkB and IL-8 expression. Conversely, the DS mutant CP did not repress cag T4SS pilus formation, even at very high concentrations (1200 µg/ml). These data indicate that CP inhibits the production of cag T4SS-associated pili through zinc sequestration.
Results from these studies indicate that zinc homeostasis plays an important role in regulating the cag T4SS in H. pylori. Previous reports have shown that when H. pylori interacts with gastric epithelial cells, the cag T4SS translocates CagA into host cells leading to activation of c-Src and changes in the cytoskeleton [34]–[36]. A functional cag T4SS also increases the activation of NOD and NFkB [33], which ultimately results in the production of IL-8 and the recruitment of neutrophils [34]–[36]. Based on our results, we propose a model which is presented in Figure 7. In response to H. pylori infection, neutrophils are recruited to the site of infection. CP is deposited, and nutrient manganese and zinc are sequestered. Sequestration of zinc by CP represses cag T4SS pilus formation and CagA translocation, and results in diminished NFκB activation and IL-8 secretion. With reduced IL-8, less inflammation develops and the end result of CP-dependent zinc-sequestration is increased bacterial persistence.
H. pylori infection elicits a robust neutrophil response, resulting in increased expression of CP in the infected tissue, a result that agrees with previously published observations [24], [37]. In other models of infection, CP levels increase in response to bacterial infection [14], [18], [38], and levels vary from tissue to tissue [39]. It is likely that CP levels are dynamic within host gastric tissue during H. pylori infection because inflammation varies in a time- and location-dependent manner and is shaped by host genetics [2], [40], [41]. We hypothesized that CP is an important mediator of host-H. pylori interactions. Our results suggest that CP's activity against H. pylori may be dose-dependent. CP inhibits growth of H. pylori by sequestration of nutrient metals, an observation that agrees with the results seen with numerous other pathogens and supports the proposal that CP is part of the host response designed to restrict H. pylori burden in vivo. Yet when CP is absent, the bacterial burden is reduced by the increased inflammatory response. These data suggest that whether CP is present or absent, there is cross regulation between the bacteria and the host immune response, leading to a level of inflammation which controls bacterial burden but does not necessarily induce enough inflammation to completely clear the infection, and therefore, the bacteria persist.
In addition to indicating that CP restricts the growth of H. pylori, our work reveals that CP represses the activity of the cag T4SS, an important virulence factor that has been associated with carcinogenesis [42]. By utilizing CP mutants harboring inactivation of the metal binding sites within CP, our work demonstrated that both the site 1 and the site 2 mutant proteins have the capacity to inhibit the activity and biogenesis of pili associated with the cag T4SS. Since the CP mutant that can only bind zinc is still capable of inhibiting cag T4SS function and pilus biogenesis, this result proves that nutrient zinc sequestration is responsible for the cag T4SS repression phenotype, a result that is supported by restoration of cag T4SS activity and pilus biogenesis when an exogenous source of nutrient zinc is provided. At present, the components of the cag T4SS are not well defined and controversial [43], therefore it is not possible to determine whether pilus production is blocked at the level of pilus assembly, at the level of translation of pilus components or through some other mechanism. Interestingly, the cag T4SS has been reported to be induced by chelation of nutrient iron, a result that is reciprocal to the regulation imposed by sequestration of nutrient zinc, but still supports the contribution of micronutrients to the regulation of this important bacterial organelle [44]. There are other examples of reciprocal regulation of virulence factors by nutrient iron and zinc among bacterial pathogens. For example, in the gastrointestinal pathogen, enteropathogenic E. coli (EPEC), the expression and secretion of EPEC-associated secreted proteins (Esp) have been shown to be repressed by nutrient zinc and induced by nutrient iron [45].
Furthermore, CP-mediated metal sequestration has been associated with changes in bacterial virulence in other pathogens. For example, in the presence of CP, S. aureus superoxide defense is repressed, leading to diminished resistance to innate immune cells [16]. These previously published results reveal that CP can, in addition to repressing bacterial growth, also repress bacterial virulence to promote host defense strategies. In our murine model of infection, CP represses a major inflammation-promoting virulence factor, the cag T4SS. This activity may contribute to bacterial evasion of the immune system.
CP has been shown to be induced in the context of bacterial infection, and the utility of CP-deficient mice (A9 -/- mice) in these studies has proven to be important [14], [17], [18]. In a model of S. aureus infection, CP-deficient mice had an approximately 1-log increase in bacterial burden in their livers compared to WT mice [16]. Similarly, in a model of A. baumannii infection, CP-deficient mice exhibited a significant increase in bacterial burden relative to WT mice at 36 hours post-infection, but not at 72 hour post-infection, indicating that CP is important for controlling early infection, but other components of the host system can clear bacteria independently of CP [17]. Conversely, in a model of S. Typhimurium, Salmonella overcomes zinc sequestration by calprotectin to colonize the gut through production of the CznABC transporter. CznABC was also required to promote the growth of S. Typhimurium over that of competing commensal bacteria in the inflamed gut [18]. H. pylori bacterial burden is decreased in CP-deficient mice compared to WT mice, indicating that the activity of CP is not restricted to antimicrobial activity against H. pylori, but can also regulate bacterial virulence which may promote chronic colonization within the host gastric niche. In other words, CP repression of the cag T4SS could promote chronic colonization by reducing inflammation and increasing bacterial burdens.
In addition to repressing bacterial virulence, CP has other effects on bacterial pathogens. S. aureus upregulate two metal uptake systems, MntABC and MntH, to promote manganese acquisition in the presence of CP [39]. The metal-acquisition properties are important for staphylococcal resistance to CP as well, and mutants in mntABC or mntH have decreased growth compared to WT strains in increasing concentrations of CP [39]. Similarly, both A. baumannii and S. Typhimurium upregulate zinc acquisition machinery to resist the nutrient sequestration imposed by CP in inflamed tissues. Mutants in zinc acquisition in both of these pathogens have diminished capacity to compete with WT bacteria in rodent models of infection [17], [18]. Taken together, these data support a model in which zinc acquisition provides a selective advantage for invading pathogens in the presence of inflammation-associated antimicrobial peptides such as CP. It also supports the diverse roles that both manganese and zinc availability play in microbial responses within the vertebrate host.
The trace element zinc is essential for cell function, and zinc co-factors are prevalent in the bacterial proteome [46]. Less than 0.1% of zinc is found in serum plasma [47], indicating the vast majority of zinc is stored in the intracellular space, and thereby, unavailable to extracellular pathogens. Additionally, the immune system produces zinc-chelating molecules such as S100A8 and S100A9 that have the capacity to bind and sequester nutrient zinc from invaders [48]. Thus, zinc homeostasis is emerging as an important feature of the host-pathogen interaction. Zinc exposure has been associated with decreased adherence, biofilm formation and virulence factor expression in enteroaggregative Escherichia coli (EAEC) [49]. Additionally, macrophages have been shown to utilize intraphagosomal zinc accumulation as a strategy to poison bacterial pathogens such as Mycobacterium tuberculosis [50]. To combat this, M. tuberculosis elaborates heavy metal efflux P-type ATPases, metallothioneins, and a zinc exporter [50]. This strategy is not uncommon, as pathogens such as Streptococcus pyogenes and Pseudomonas aeruginosa, have been shown to express zinc efflux systems that are crucial for colonization of an animal host [51], [52]. Similarly, H. pylori encodes a novel metal efflux pump, CznABC, which is required for resistance to cadmium, nickel and zinc intoxication, as well as colonization of a vertebrate host [53]. These results implicate zinc as an important micronutrient signal at the host-pathogen interface that bacteria sense and respond to accordingly. H. pylori has likely evolved to sense the presence of zinc as well as the deprivation of zinc by CP to induce the appropriate cellular responses. Thus, CP is an important inflammation-associated environmental signal to the bacterial cell.
CP also has the ability to act as a damage-associated molecular pattern molecule (DAMP) within the host [54]. It has been demonstrated that CP activates toll-like receptor 4 (TLR4) and signaling through nuclear factor ĸB, which ultimately leads to increased inflammatory responses [54], [55]. Additionally, CP is hypothesized to interact with the receptor for advanced glycation end products (RAGE) to promote chronic inflammation [56]. In the context of our murine model of infection, CP itself could potentially contribute to inflammatory processes. However, in the presence of an H. pylori infection, the presence of CP is associated with diminished inflammation, which can be attributed to the zinc-dependent regulation of the proinflammatory cag T4SS.
In conclusion, we propose a model (Figure 7) in which nutrient zinc acts as a signal to induce the cag T4SS and promote H. pylori-dependent inflammation. After a neutrophil response develops through cag T4SS-driven IL-8 production or through induction of the Th17 response, deposition of CP sequesters zinc and manganese. The result of this sequestration is that the cag T4SS is switched off. This tightly controlled regulation may contribute to bacterial immune evasion and promote chronicity.
All animal experiments were done in concordance with the Animal Welfare Act, U.S. federal law, and NIH guidelines. All experiments were carried out under an ACORP protocol approved by Vanderbilt University Institutional Animal Care and Use Committee (IACUC; V/10/410 and V/13/240) and the Department of Veteran's Affairs, a body that has been accredited by the Association of Assessment and Accreditation of Laboratory Animal Care (AAALAC). The human study protocol was approved by the Vanderbilt University and the Nashville Department of Veterans Affairs Institutional Review Board (#5190). Human subjects gave informed written consent.
H. pylori strains SS1 (a mouse-adapted clinical isolate), PMSS1 (the clinical isolate from which SS1 was derived), and the PMSS1 cagE isogenic mutant (PMSS1 cagE::aphA, a gift from M. Amieva) were used for these studies. PMSS1 was selected because it has a functional cag PAI and has the ability to colonize mice, and SS1 was selected due to its ability to colonize mice, although it lacks a functional cag PAI [57]. For infection assays, H. pylori strains were cultured on tryptic soy agar plates supplemented with 5% sheep blood or in Brucella broth supplemented with 10% fetal bovine serum at 37°C in room air supplemented with 5% CO2. For bacterial growth assays, H. pylori were grown in 60% Brucella broth plus 40% calprotectin (CP) buffer [16] supplemented with 10% fetal bovine serum (FBS) alone or supplemented with 50 µM zinc chloride and 50 µM manganese chloride, with increasing concentrations of CP at 37°C in room air supplemented with 5% CO2. Bacterial growth was quantified at 4, 12 and 24 hours by spectrophotometric reading of OD600, and at both 4 and 24 hours bacteria were subjected to serial dilution and plating onto tryptic soy agar plates supplemented with 5% sheep blood for enumeration of viable bacterial cells (CFU/mL).
For H. pylori-AGS cell co-culture assays, bacteria were grown in 60% Brucella broth plus 40% CP buffer supplemented with 10% FBS alone or supplemented with 100 µM zinc chloride, in the presence of 200 µg/mL WT CP, or 200–1200 µg/ml of ΔS1, ΔS2, double site mutant (DS CP) CP mutants. H. pylori were also cultured in the presence of the synthetic zinc chelator, N,N,N′,N′-tetrakis (2-pyridylmethyl) ethylenediamine (TPEN) (Sigma Aldrich) at a concentration of 5 µM. Bacteria for AGS co-cultures were enumerated by using the OD600 and bacterial coefficient established for this H. pylori strain using our spectrophotometer (this coefficient was determined by plating to enumerated viable bacteria). Purification of WT and mutant CP proteins was performed as previously described [16]. Adherence assay methods are in the Materials and Methods S1 file.
AGS human gastric cells (ATCC) were cultured to 70% confluency in RPMI medium supplemented with 10% FBS, 2 mM L-glutamine, and 10 mM HEPES buffer at 37°C in room air plus 5% CO2.
H. pylori strains grown in various conditions of zinc availability were co-cultured with AGS cells at a multiplicity of infection of 100∶1 (as determined by spectrophotometric reading at OD600) for 4 hours. Cellular supernatants were collected and IL-8 was measured using an anti-human IL-8 ELISA (R&D), as previously described [32]. Bacteria grown in the presence of 5 µM TPEN or CP protein (WT, ΔS1 CP mutant, ΔS2 CP mutant, DS CP mutant) at either 200 µg/mL or 600 µg/mL were compared to those grown in the medium alone or in the presence of chelator plus 100 µM zinc chloride.
Human gastric AGS cells were stably transfected with an NF-kB-luciferase reporter, as previously described [25]. Briefly, eukaryotic cells were grown to 70% confluency, bacteria were added at a multiplicity of infection of 20∶1 (as determined by spectrophotometric reading at OD600), and co-cultured for 4 hours. Supernatants were collected, cells were lysed and luciferase activity was measured with the Promega E4030 luciferase assay system (Promega, Madison, WI). As a positive control, H. pylori PMSS1 grown in medium alone was used and normalized for 100% luciferase activity.
Phospho-CagA was detected by immunoblot, as previously described [25]. For detection of phospho-CagA, AGS cells were co-cultured with bacteria, as described for the IL-8 ELISA assays. AGS cells were washed twice with RPMI medium containing 1 mM sodium orthovanadate and pelleted by centrifugation (15,000 RPM for 3 minutes). Pellets were lysed in NP-40 lysis buffer containing Completer Mini EDTA-free protease inhibitor (Roche, Indianapolis, IN) and 2 mM sodium orthovanadate. Proteins were separated by soluble fractionation using 7.5% SDS-PAGE and immunoblotting with an anti-phosphotyrosine antibody (anti-PY99, Santa Cruz) or anti-H. pylori antibody, a polyclonal rabbit anti-H. pylori antibody described in [58]). Immunoreactive bands were visualized by ECL following incubation of the blot with HRP-conjugated secondary antibody. Detection of CagA and H. pylori soluble proteins was performed by stripping the blot with Restore Buffer (Pierce), and reprobing with either anti-CagA or anti-H. pylori primary antibodies.
Permission to use male and female IL-17RA-/- mice for the establishment of a breeding colony was obtained from Amgen (Seattle, WA). S100A9-/- mice were a gift from Wolfgang Nacken (Institute of Experimental Dermatology, University of Münster, 48149 Münster, Germany) to E.P.S. The generation of these mice was previously described [59]. CP-deficient mice lack the S100A9 component of the heterodimer and exhibit destabilization of S100A8 protein as well [60], resulting in deficiencies in metal sequestration [14]. Amgen's IL-17RA-/- mouse breeding colony is maintained at Taconic Farms. Helicobacter-free male IL-17RA-/-, S100A9-/- and WT mice (all C57BL/6 background; 8–10 weeks old, were used in all experiments). Mice were orogastrically infected with 5×108 CFU H. pylori in 0.5 mL of Brucella broth twice over a 2 day period.
At different time points post-infection mice were sacrificed by carbon dioxide inhalation and the glandular stomach was removed for analyses. The stomach was removed from each mouse by excising between the esophagus and the duodenum. The forestomach (nonglandular portion) was removed from the glandular stomach and discarded. The stomach was rinsed gently with PBS to remove food and cut into three longitudinal strips spanning both the antrum and corpus, which were used for quantitative bacterial culture, RNA extraction/real-time RT-PCR analyses, and histological examination. For culturing of H. pylori from the stomach, gastric tissue was placed into Brucella broth-10% FBS for immediate processing. For RNA extraction, the stomach was placed in RNAlater solution (Ambion) before being processed. A longitudinal strip from the greater curvature of the stomach was excised and placed in 10% normal buffered formalin for 24 hours, embedded in paraffin and processed routinely for hematoxylin and eosin (H&E) staining. Indices of inflammation and injury were scored by a single pathologist (M.B.P.) who was blinded to the identity of the mice. Acute and chronic inflammation in the gastric antrum and corpus were graded on a 0–3 scale. Acute inflammation was graded based on density of neutrophils and chronic inflammation was graded based on the density of lamina propria mononuclear cell infiltration independent of lymphoid follicles. Total inflammation was calculated as a sum of acute and chronic inflammation in the corpus and the antrum allowing for quantification of total inflammation on a scale of 0–12 [61]. Immunohistochemistry was performed using commercially available polyclonal rabbit anti-S100A9 antibody (Cat # NB110-89726, Novus Biologicals, Littleton, CO).
For quantitative culture, the gastric tissue was homogenized in Brucella broth using a tissue tearor (BioSpec Products, Inc. Bartlesville, OK). Serial dilutions of the homogenate were plated on trypticase soy agar plates containing 5% sheep blood, 10 µg/ml nalidixic acid, 100 µg/ml vancomycin, 2 µg/ml amphotericin, and 200 µg/ml bacitracin. After 5 to 7 days of culture under microaerobic conditions generated by CampyPak Plus Gas Pak system, H. pylori colonies were counted. Flow cytometry methods are in the Materials and Methods S1 file.
Because our study comparing PMSS1 and PMSS1 cagE mutant was designed to elucidate the regulation of the cag T4SS in response to a neutrophil-associated antimicrobial protein, the six week post-infection time point was an appropriate timepoint when neutrophils would be present [29] and the cag T4SS would still be functional [25], [57], [62].
RNA was isolated from the stomach using the TRIZOL isolation protocol (Invitrogen, Carlsbad, CA) with slight modifications, as previously described [61]. RNA was reverse transcribed using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA). For real time RT-PCR, we used the relative gene expression method [63]. Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) served as the normalizer, and tissue from a 5 uninfected WT mouse stomachs (or 5 uninfected human biospies) served as the pooled calibrator sample. All real time RT-PCR was performed using an Applied Biosystems StepOne Plus real time PCR instrument. Levels of cytokine expression are indicated as “relative units”, based on comparison of tissue from H. pylori-infected mice with tissue from uninfected mice (calibrator tissue) [63]. Primer and probe sets were purchased as Taqman Gene Expression Assays from Applied Biosystems (as pre-designed assays the annealing temperatures and amplicon length are available on their website). Primer and probe sets for eukaryotic genes were purchased as Taqman Gene Expression Assays from Applied Biosystems [S100A8 (Mm01220132_g1), S100A9 (Mm00656925_m1), GAPDH (Mm99999915_g1), human S100A8 (Hs00374264_g1), human S100A9 (Hs00610058_m1), and human GAPDH (Hs99999905_m1)].
H. pylori cag T4SS pili were imaged by field emission gun-scanning electron microscopy (FEG-SEM) analysis using methods previously described [32]. Briefly, H. pylori cells grown under various conditions of zinc availability were co-cultured at a multiplicity of infection of 50∶1 with AGS human gastric epithelial cells on poly-L-lysine-treated coverslips (BD Biosciences) for 4 h at 37°C in the presence of 5% CO2. Cells were fixed with 2.0% paraformaldehyde, 2.5% glutaraldehyde in 0.05 M sodium cacodylic acid buffer for 1 h at room temperature. Samples were washed three times with cacodylic acid buffer before secondary fixation with 1% osmium tetroxide. Cells were subjected to sequential dehydration with increasing concentrations of ethanol before being dried at the critical point, mounted onto SEM stubs, painted with colloidal silver at the sample edge, and sputtered with 20 nm of gold-palladium coating. Samples were visualized with an FEI Quanta 250 FEG-SEM at high vacuum and micrographs were analyzed with Image J software.
Statistical analysis of bacterial burden was performed after log transformation using unpaired two-tailed Student's t-test. Statistical analyses of IL-8 secretion, pilus quantification, expression data and luciferase activity were performed using paired two-tailed Student's t-test. Statistical analyses of histological inflammation scores were performed using Mann-Whitney U analysis. All data were derived from at least three separate biological replicates unless specified otherwise. Statistical analyses were performed using GraphPad Prism Software.
|
10.1371/journal.pntd.0004329 | Sustaining Control of Schistosomiasis Mansoni in Western Côte d’Ivoire: Results from a SCORE Study, One Year after Initial Praziquantel Administration | The Schistosomiasis Consortium for Operational Research and Evaluation (SCORE) has launched several large-scale trials to determine the best strategies for gaining and sustaining control of schistosomiasis and transitioning toward elimination. In Côte d’Ivoire, a 5-year cluster-randomized trial is being implemented in 75 schools to sustain the control of schistosomiasis mansoni. We report Schistosoma mansoni infection levels in children one year after the initial school-based treatment (SBT) with praziquantel and compare with baseline results to determine the effect of the intervention.
The baseline cross-sectional survey was conducted in late 2011/early 2012 and the first follow-up in May 2013. Three consecutive stool samples were collected from 9- to 12-year-old children in 75 schools at baseline and 50 schools at follow-up. Stool samples were subjected to duplicate Kato-Katz thick smears. Directly observed treatment (DOT) coverage of the SBT was assessed and the prevalence and intensity of S. mansoni infection compared between baseline and follow-up.
The S. mansoni prevalence in the 75 schools surveyed at baseline was 22.1% (95% confidence interval (CI): 19.5–24.4%). The DOT coverage was 84.2%. In the 50 schools surveyed at baseline and one year after treatment, the overall prevalence of S. mansoni infection decreased significantly from 19.7% (95% CI: 18.5–20.8%) to 12.8% (95% CI: 11.9–13.8%), while the arithmetic mean S. mansoni eggs per gram of stool (EPG) among infected children slightly increased from 92.2 EPG (95% CI: 79.2–105.3 EPG) to 109.3 EPG (95% CI: 82.7–135.9 EPG). In two of the 50 schools, the prevalence increased significantly, despite a DOT coverage of >75%.
One year after the initial SBT, the S. mansoni prevalence had decreased. Despite this positive trend, an increase was observed in some schools. Moreover, the infection intensity among S. mansoni-infected children was slightly higher at the 1-year follow-up compared to the baseline situation. Our results emphasize the heterogeneity of transmission dynamics and provide a benchmark for the future yearly follow-up surveys of this multi-year SCORE intervention study.
| Schistosomiasis is a parasitic worm disease that is widespread in sub-Saharan Africa. To better understand how to gain and sustain the control of schistosomiasis and how to eliminate this disease in different epidemiologic settings, the Schistosomiasis Consortium for Operational Research and Evaluation (SCORE) has launched several multi-year studies that are being implemented in East and West Africa. This article highlights how the Schistosoma mansoni infection levels changed one year after an initial treatment with the anti-worm drug praziquantel given to children aged 5–15 years in western Côte d’Ivoire. Infection and treatment data at school level were available from more than 4,600 children in 50 schools. One year after the treatment that had been received by more than 80% of the children, the overall S. mansoni prevalence decreased from 19.7% to 12.8%, while the intensity of infection among S. mansoni-positive children slightly increased. In several schools, the S. mansoni intensity and, particularly the prevalence, increased unexpectedly. Our findings show that the dynamics of schistosomiasis transmission varies from one village to another. It will be interesting to monitor changes over longer time periods as this SCORE study unfolds.
| Schistosomiasis is a neglected tropical disease that exerts a considerable public health problem in 78 tropical and subtropical countries [1]. In 2013, it was estimated that schistosomiasis affected more than 250 million people worldwide with 90% of the reported cases concentrated in sub-Saharan Africa [2]. Since the mid-1980s, the World Health Organization (WHO) emphasizes morbidity control using the drug praziquantel as the main pillar of the global strategy to fight schistosomiasis [3]. Praziquantel is the drug of choice because it is efficacious against the adult stages of all Schistosoma species parasitizing humans, is inexpensive (the average cost to treat a school-aged child was US$ 0.2 per treatment in 2013), and has a good safety profile [4–8]. For morbidity control, praziquantel is being administered to at-risk populations without prior diagnosis, a strategy commonly known as ‘preventive chemotherapy’ [9].
The recommended frequency of drug administration is based on the level of endemicity in a given study area. According to WHO, in areas with high schistosomiasis endemicity (prevalence ≥50%), all school-aged children and adult people at risk of infection should be treated annually [10]. In areas with moderate endemicity (prevalence 10–50%), all school-aged children should be treated once every two years. In low endemic areas (prevalence <10%), school-aged children should be treated twice during their time in school; first at school entry and then again in their last year of schooling [11,12]. However, these prevalence thresholds are arbitrary. Hence, the Schistosomiasis Consortium for Operational Research and Evaluation (SCORE) launched a series of studies to strengthen the evidence-base how best to gain and sustain the control of schistosomiasis, including cost considerations [13]. Two 5-year cluster-randomized trials are being implemented in Côte d’Ivoire and Kenya [14,15]. These trials are school-based with three treatment arms (25 schools per arm) and aim to assess whether annual school-based treatment (SBT) with praziquantel for four years (arm A), annual SBT in years 1 and 2, followed by “drug holidays” in years 3 and 4 (arm B), or SBT in years 1 and 3, spaced by“drug holidays” in years 2 and 4 (arm C) will substantially reduce the prevalence and intensity of Schistosoma infection and keep infection at low levels.
Here, we present the effect of the first SBT with praziquantel on Schistosoma mansoni infection among school-aged children in western Côte d’Ivoire, as revealed by a detailed follow-up survey conducted in May 2013, compared to baseline data collected from December 2011 to February 2012. Specifically, we determined changes in the prevalence and intensity of S. mansoni infections among children in the 50 schools that belong to treatment arms A and B, and discuss consequences for the ongoing cluster-randomized trial and, more generally, for schistosomiasis control interventions in Côte d’Ivoire and elsewhere.
The study protocol was approved by the institutional research commissions of the Swiss Tropical and Public Health Institute (Basel, Switzerland) and the ‘Centre Suisse de Recherches Scientifiques en Côte d’Ivoire’ (CSRS; Abidjan, Côte d’Ivoire). Ethical approval was obtained from the ethics committees in Basel (reference no. EKBB 279/10) and the Ministry of Public Health in Côte d’Ivoire (reference no. 1994 MSHP/CNER).
At the onset of the study, regional directors of the education and health sectors, education inspectors, village authorities, local community members, and teachers were sensitized in detail about the objectives of the research project. Parents and guardians of study participants provided written informed consent for children to participate. After the baseline parasitologic survey, in the frame of the first SBT conducted in June 2012, school-aged children living in the catchment area of participating schools were offered treatment with praziquantel at a single oral dose of 40 mg/kg of body weight [16].
The baseline survey was carried out from December 2011 to February 2012, the SBT in June 2012, and the first follow-up survey was conducted in May 2013 in eligible schools located in four regions of western Côte d’Ivoire: Cavally, Guemon, Haut-Sassandra, and Tonkpi. Details of the study area and population surveyed have been described elsewhere [15,17]. The Cavally and Sassandra rivers and their tributaries represent the major hydrographic features of the study area [18,19]. Buyo, a hydroelectric dam built across the Sassandra River in 1981, formed a lake with an estimated surface area of 600 km² [20]. In western Côte d’Ivoire, the sources of water are traditional wells, rain water, rivers, water supply dams, ponds, creeks, fountains, natural spring water, and tap water [21]. The main reasons for human water contact that might lead to schistosomiasis transmission are washing dishes, washing children, fetching water, fishing, swimming, farming, and playing [22]. Despite the existence of latrines in numerous households, open defecation is commonly practiced [22–24].
The aim of the SCORE sustaining schistosomiasis control study implemented in western Côte d’Ivoire is to determine the best strategy of preventive chemotherapy with praziquantel to sustain schistosomiasis mansoni control in moderate endemicity settings [15,17]. For this purpose, the S. mansoni prevalence in n schools in three treatments arms is compared over a study period of four years. The prevalence of S. mansoni is determined by testing m children in those schools where there is subsequent treatment. The effect of the different treatment intervals on the S. mansoni prevalence will be estimated using the following logistic regression model: log (pijt / (1 − pijt)) = μ +αi + βt + ɣik, where pijt denotes the prevalence of S. mansoni in school j receiving treatment i in year t, μ is an intercept term, αi is the effect of treatment i, βt is the effect of time t, and γik is the time by treatment interaction. Generalized estimating equations have been used to fit these longitudinal data [25]. To take into account variation in the S. mansoni prevalence among schools, an overdispersion parameter φ was introduced into the model. When φ = 1, all schools under the same treatment have identical prevalences, whereas φ increases with increasing variation of prevalence levels between villages.
The calculations revealed that studying 20 schools per arm and evaluating 100 individuals per school would result in minimum effect sizes of 5–12% with or without overdispersion. In order to increase the chance of detecting differences between the intervention arms, the number of intervention units was increased to 25 per arm. Consequently, a total of 75 schools with a S. mansoni prevalence of 10–24% according to results from an eligibility survey were randomized to one of the three treatment arms [15]. Treatment arm A receives SBT with praziquantel once every year for four years, arm B receives SBT in years 1 and 2, followed by “drug holidays” in years 3 and 4, and arm C receives SBT in years 1 and 3, alternated by “drug holidays” in years 2 and 4 [15]. Before administration of the first round of treatment, a detailed baseline survey was conducted.
Following the SCORE harmonization protocol, all 75 schools were included in the baseline parasitologic survey implemented in Côte d’Ivoire from December 2011 to February 2012. Children were treated with praziquantel in June 2012. Only the 50 schools belonging to treatment arms A and B were subjected to the first follow-up survey carried out in May 2013, while the 25 schools belonging to treatment arm C were not subjected to a follow-up survey, as they were on “drug holidays” in year 2.
Baseline and follow-up surveys pursued cross-sectional designs. Study procedures have been detailed elsewhere [15,17]. In brief, in each of the selected schools, approximately 100 children were invited to participate in the study. Inclusion criteria were as follows: (i) age of children ranging between 9 and 12 years; (ii) presence of an informed consent sheet signed by parents/guardians; and (iii) children themselves assented orally. Over three consecutive days, children were invited to submit a portion of their own morning stool in a 125-ml plastic container. Every day, filled stool containers were collected by trained field enumerators and sent to the hospital laboratories in the towns of Biankouma, Danané, Douékoué, Guiglo, Kouibly, and Man for processing.
Stool specimens were subjected to the Kato-Katz method [26]. In brief, duplicate Kato-Katz thick smears were prepared from a single stool sample, using 41.7 mg plastic templates. The thick smears were allowed to clear for at least 60 min and examined by experienced laboratory technicians under a light microscope at low magnification. Eggs from S. mansoni, and additionally from soil-transmitted helminth species, were counted and recorded for each species separately. For quality control, 10% of the slides were randomly selected and re-read by a senior microscopist. In case of discrepancies, the results were discussed with the concerned microscopists and the slides re-read until agreement was reached [27].
In June 2012, children aged 5–15 years enrolled in the 75 study schools and non-enrolled school-aged children living in the school catchment areas were offered free-of-charge treatment with praziquantel (40 mg/kg) using a dose pole according to WHO guidelines [16]. Praziquantel was administered by trained teachers to children, following a directly observed treatment (DOT) approach. Children remained under medical observation and adverse events were recorded within 4 hours post-treatment. Treatment was led by the ‘Programme National de Lutte contre la Schistosomiase, les Géohelminthiases et la Filariose Lymphatique’ (PNL-SGF), and supported by staff from the ‘Programme National de Santé Scolaire et Universitaire’ (PNSSU), the CSRS, and the ‘Université Félix Houphouët-Boigny’. Praziquantel tablets were supplied by the Schistosomiasis Control Initiative (SCI; London, United Kingdom).
The overall number of school-aged children residing in each village was obtained by adding up the number of school-aged but non-school attending children as recorded by the community health workers and the number of children registered in school, as detailed by school teachers. Trained teachers administered praziquantel to children (those attending school, and the non-enrolled children) and recorded the number of treated children.
Baseline survey data were entered into Microsoft Excel (2010 Microsoft Corporation), while data from the first follow-up survey were directly entered into smartphones and then uploaded to a database maintained on a central server (Open Data Kit) in Atlanta, United States of America. Statistical analyses were performed with STATA version IC13.1 (Stata Corporation; College Station, United States of America). The final analysis included children aged 9–12 years who had at least four Kato-Katz thick smear readings at the parasitologic surveys done both at baseline and follow-up. To obtain individuals’ eggs per gram of feces (EPG), we divided the total S. mansoni egg counts from the multiple Kato-Katz slides per child by the total number of Kato-Katz thick smears and multiplied by a factor of 24. S. mansoni-positive individuals were stratified into three infection intensity categories: (i) light (1–99 EPG), (ii) moderate (100–399 EPG), and (iii) heavy (≥400 EPG) [16]. Moreover, we calculated S. mansoni prevalence and arithmetic mean (AM) EPG for positive individuals per school and treatment arm. With regard to soil-transmitted helminth infections that were also identified with the Kato-Katz technique, a child was considered positive if at least one egg of Ascaris lumbricoides, hookworm, or Trichuris trichiura was detected in one of the slides.
We employed a χ² test to assess a potential association between S. mansoni prevalence and age or sex. Reduction in the prevalence and intensity of S. mansoni infection per school was calculated using the following formulae [28]: prevalence reduction = [(prevalence at baseline—prevalence at first follow-up) / (prevalence at baseline)] X 100. Reduction in the intensity of infection = [(AM EPG at baseline—AM EPG at first follow-up) / (AM EPG at baseline)] X 100.
The treatment coverage rate was assessed by using the following formula: coverage of the mass drug administration (MDA) = [(number of school-aged children with DOT recorded by teachers) / (overall number of school-aged children registered in school and recorded by health workers)] X 100.
Geographic coordinates of each school were recorded using a hand-held global positioning system (GPS) receiver (Garmin Etrex 30; Olathe, United States of America). Arc Map 10.2.1 (Environmental Systems Research Institute Inc.; Redlands, United States of America) was used to generate maps of the changes of S. mansoni prevalence and intensity of infection (AM EPG) from baseline to follow-up.
The baseline survey was conducted in the 75 schools meeting eligibility criteria from December 2011 to February 2012, and 7,478 children were invited to participate (Fig 1). Among them, 168 pupils were excluded from further analyses, because their age was outside the 9–12 years range. Additionally, 299 children were excluded because they did not provide sufficient stool to prepare at least quadruplicate Kato-Katz thick smears. The final study population for analysis of the baseline survey consisted of 7,011 children. There were more boys (n = 4,173) than girls (n = 2,838). The mean age was 10.5 years. The number of children in treatment arms A, B, and C was 2,410 (34.4%), 2,348 (33.5%), and 2,253 (32.1%), respectively.
In May 2013, 4,966 children from the 50 schools belonging to intervention arms A and B were invited to participate in the first follow-up survey. According to the SCORE harmonization protocol, children attending schools belonging to study arm C were not surveyed. Among the pupils attending schools included in arms A and B, who were invited to participate, 49 children had an age outside the 9–12 years range, and 250 children did not provide enough stool for at least quadruplicate Kato-Katz thick smears. Hence, results of 4,667 children were included for further statistical analyses. There were more boys (n = 2,640) than girls (n = 2,027). The children’s mean age was 10.3 years. There were 2,379 children in treatment arm A and 2,288 in treatment arm B.
At baseline, before the implementation of the first SBT with praziquantel, the examination of at least four Kato-Katz thick smears per child revealed an overall S. mansoni prevalence of 22.1% among the 75 schools surveyed. The prevalence at the unit of the school ranged from 1.0% to 54.0%. S. mansoni infection was significantly associated with age (χ² = 25.2, p <0.001), higher prevalence was predominantly observed among older children. The prevalence of S. mansoni was significantly higher among boys than girls (24.3% versus 18.7%; χ² = 29.9, p <0.001). The overall S. mansoni prevalence in treatment arms A, B, and C was 18.8% (95% CI: 17.2–20.3%), 20.5% (95% CI: 18.9–22.2%), and 27.2% (95% CI: 25.3–29.0%), respectively. With regard to the AM infection intensity, the respective values were 93.5 EPG (95% CI: 62.6–124.4 EPG), 96.2 EPG (95% CI: 74.5–117.9 EPG), and 88.1 EPG (95% CI: 71.5–104.7 EPG) (Table 1).
As summarized in Table 1, at the first follow-up survey, the overall S. mansoni prevalence in arms A and B showed a statistically significant decline from 19.7% (95% CI: 18.5–20.8%) at baseline to 12.8% (95% CI: 11.9–13.8%) at the 1-year follow-up. In arm A, a decrease from 18.8% (95% CI: 17.2–20.3%) to 11.2% (95% CI: 9.9–12.4%) was observed, corresponding to a reduction of 40.4%, while in arm B the prevalence declined from 20.5% (95% CI: 18.9–22.2%) to 14.5% (95% CI: 13.1–16.0%), a reduction of 29.3%.
Fig 2 indicates the dynamics of the S. mansoni prevalence from baseline to first follow-up survey on a school-by-school basis, stratified by treatment arm. Among the 25 schools belonging to treatment arm A, the S. mansoni prevalence dropped in 23 schools (S1 Table). The most significant decreases occurred in Dio, Pona 2, Siambly, and Gregbeu, where at the 1-year follow-up, no eggs of S. mansoni were found in the stool of the children examined. However, in Biélé, the S. mansoni prevalence increased significantly from 36.0% (95% CI: 26.4–45.6%) to 79.0% (95% CI: 70.9–87.1%), while a non-significant increase from 12.0% (95% CI: 5.5–18.5%) to 20.7% (95% CI: 12.0–29.4%) was observed in Séohoun-Guiglo.
In treatment arm B, the prevalence of S. mansoni decreased in 20 out of the 25 schools included (S1 Table). In two schools, the prevalence dropped prominently to zero from 24.0% in Semien and from 25.6% in Diehiba. A significant increase in the S. mansoni prevalence was observed in Ziondrou from 31.6% (95% CI: 22.0–41.1%) to 62.0% (95% CI: 52.3–71.7%). An increase in prevalence was also observed in Dah, Douandrou 1, Koulouan, and Guessabo 2, but without statistical significance.
Taken together, as shown in Fig 3A, among the 50 schools surveyed at the first follow-up, a reduction of the S. mansoni prevalence of 25% and above was observed in 39 schools (78.0%). In six schools, the changes ranged from -25% to +25%. An increase of 25% and above was recorded in five schools (10.0%). The increase in prevalence was observed mainly in the central part of Guemon region, eastern Tonkpi region, and western part of Haut-Sassandra region.
The overall S. mansoni AM EPG in arms A and B increased from 94.9 EPG (95% CI: 76.2–113.6 EPG) at baseline to 109.3 EPG (95% CI: 82.7–135.9 EPG) at the 1-year follow-up survey. However, this increase was not statistically significant. As shown in Table 2, in arm A, an increase from 93.5 EPG (95% CI: 62.6–124.4 EPG) to 123.7 EPG (95% CI: 70.7–176.7 EPG) was observed, corresponding to an increase of 32.3%, while in arm B the AM EPG at baseline (96.2 EPG, 95% CI: 74.5–117.9 EPG) and the 1-year follow-up (97.8 EPG, 95% CI: 75.5–120.0 EPG) remained basically the same. The proportion of children with heavy infections (≥400 EPG) increased from 4.9% to 6.3%.
Fig 4 displays the changes of the S. mansoni AM EPG in all the schools of treatment arms A and B from baseline to the first follow-up. In arm A, the S. mansoni AM EPG decreased in 16 (64.0%) out of the 25 surveyed schools (S1 Table). However, a statistically significant decrease in AM EPG from 33.0 EPG (95% CI: 13.9–52.0 EPG) to 5.5 EPG (95% CI: 3.8–7.2 EPG) was observed in only one school; Tobly Bangolo. Increases in S. mansoni AM EPG were observed in nine schools. However, the increase lacked statistical significance in all schools. In treatment arm A, the proportion of children with moderate (100–399 EPG) and heavy infections (≥400 EPG) increased from 9.5% to 13.5% and from 4.6% to 7.2%, respectively.
In arm B, a decrease of the S. mansoni infection intensity was observed in 13 (52.0%) out of the 25 schools (S1 Table). With the exception of one school, this decrease was not statistically significant. The AM EPG decreased significantly in Mangouin school from 178.0 EPG (95% CI: 77.7–278.3 EPG) to 30.4 EPG (95% CI: 4.2–56.6 EPG). In the remaining 12 schools, the AM EPG increased, but these increases lacked statistical significance. The proportion of children with moderate and heavy infection intensities increased from 12.9% to 18.7%, while the proportion of heavy infections decreased slightly from 6.0% to 5.7%.
Fig 3B shows the spatial distribution of S. mansoni AM EPG reduction after the intervention in the study area. The AM EPG decreased by at least 25% in 25 schools (50.0%). In eight schools (16.0%), the change varied from -25% to +25%. The AM EPG increased by 25% and more in 17 schools (34.0%). An increase of S. mansoni infection intensity by 25% and more was only focally observed; in Tonkpi region and central Guemon region.
During the SBT carried out in June 2012, the estimated number of the school-aged population in the study area was 31,832 children. Among them, 26,804 swallowed praziquantel tablets at the SBT, resulting in an overall DOT coverage of 84.2%. Stratified by treatment arm, we found a DOT coverage of 79.2% (range: 31.9–97.9%) for arm A, 84.8% (range: 61.5–98.5%) for arm B, and 88.4% (range: 75.1–98.9%) for arm C.
The individual DOT coverage rates achieved in the 75 villages are shown in S2 Table. A coverage of 75% and above was achieved in 57 schools (76.0%), while a coverage of less than 75% was reported in the remaining 18 schools. Yaoudé (in arm A) reported a coverage below 50%. The DOT coverage was not significantly correlated with changing levels of S. mansoni prevalence (Spearman ρ = -0.11; p = 0.43), while it was significantly correlated with AM EPG (Spearman ρ = 0.32; p = 0.02) (Fig 5).
Preventive chemotherapy with praziquantel is the backbone of the global strategy against schistosomiasis and other helminthiases [12,29]. Our findings show that one year after an initial treatment with praziquantel in 50 schools of western Côte d’Ivoire that met inclusion criteria of a SCORE harmonization protocol (prevalence ranging between 10% and 24%) [15], the overall S. mansoni prevalence was reduced from 19.7% to 12.8%, while there was no significant change in the overall AM EPG. The overall DOT coverage in the study area was 84.2%; hence, above the 75% coverage recommended by WHO [16]. At school level, the picture on the impact of the SBT was less clear cut. Decreases in prevalence and infection intensity were observed in some schools and increases in others. Among the six schools that showed higher prevalences of S. mansoni at the 1-year follow-up compared with baseline, in only one school, the treatment coverage was <75%. The changes in the AM EPG level were significantly correlated with the coverage rate.
The overall reduction of the S. mansoni prevalence in the first year of this SCORE project (35.0%) is in line with studies assessing the S. mansoni prevalence 12 months post-MDA in central Sudan and Uganda, where reductions of S. mansoni prevalence of 36.7% and 39.5% were observed, respectively [30,31]. The treatment coverage in these two studies was reported to be 100% and 79.2%, respectively [31,32]. In the Sudan study, treatment of children with praziquantel was conducted by trained nurses and medical officers, while in Uganda, the treatment was carried out by trained teachers and community drug distributors [31,32]. A survey conducted 6 months after praziquantel treatment in Sierra Leone where the overall treatment coverage was 94.0% found a reduction of the S. mansoni prevalence of 44.6% [33]. Another study carried out in Sierra Leone reported an even higher reduction in the S. mansoni prevalence of 67.2%, as determined three years after three rounds of praziquantel administration [34]. In contrast, studies conducted in Zambia and Kenya showed that 2 years after the withdrawal of praziquantel treatment led to an increase of S. mansoni prevalence [35,36]. It is important to note that these studies showed that the impact of MDA on the S. mansoni prevalence varied depending on the infection status in a given area, and the frequency and number of treatment rounds. Repeated treatments over short time periods can lead to a high reduction in S. mansoni prevalence compared to longer treatment intervals. Similar baseline S. mansoni prevalences were observed in two preceding studies in Sierra Leone and Uganda (49% and 42%, respectively), but the decrease in S. mansoni prevalence was lower in Uganda, where the intensity of infection, and thus the level of transmission, was higher. A plausible explanation of this observation arises from rapid re-infection, which is related to the force of infection, and which is likely higher where S. mansoni transmission is intense. Indeed, the level of schistosomiasis transmission, which is governed by various factors, such as local environmental determinants, climate, water contact patterns, intermediate host snail distribution, and ecology, may affect the impact of MDA [37–40]. When interpreting these results, one has to bear in mind, however, that the prevalence of S. mansoni was determined by an insensitive diagnostic approach; single stool samples subjected by single (Uganda) or duplicate Kato-Katz thick smears (Sierra Leone). Hence, the diagnostic approach was less rigorous than in the current study in Côte d’Ivoire, where only those children who had at least quadruplicate Kato-Katz thick smears examined before and after treatment were included in the final analysis.
In our study, in the schools Biélé and Ziondrou, the S. mansoni prevalence had significantly increased one year after SBT with levels in excess of 60%. Since the DOT coverage in both schools was high (75.2% in Biélé and 91.9% in Ziondrou), we assume that there are major transmission hotspots in the area, where children become rapidly re-infected. Re-emergence of S. mansoni and S. haematobium after treatment in high-endemicity areas has previously been reported from other studies in Côte d’Ivoire and Niger [41,42]. One explanation might be migration of people, including those infected with S. mansoni or S. haematobium, into treated villages. A considerable population movement has, for example, been observed in Côte d’Ivoire due to socio-political unrest in 2011 [43], hence at the start of our study. A lack of access to safe water, sanitation, and hygiene (WASH) might also be the reason for rapid reinfection. Noteworthy, when interviewing the local village leaders, they reported that people in the area frequently use well water for washing and bathing, while ponds and rivers serve as the main natural water contact sites. While some houses have latrines, many people still practice open defecation. Another explanation of the increase in S. mansoni prevalence might be the target population of the treatment strategy. The present study focused on school-aged children. Preschool-aged children and adults also harbor Schistosoma worms, and hence, they act as reservoir of transmission source of re-infections [36]. Yet, there are other local conditions that might foster S. mansoni transmission in Biélé and Ziondrou that warrant further investigation. For example, one might want to assess the frequency and duration of water contact in children and associated re-infection patterns, and the transmission force caused by intermediate host snails populating waterbodies located in close proximity to the surveyed schools. It will be important to assess in future surveys whether individuals had indeed received praziquantel in the past treatment round, or whether they were immigrating from other areas after the last survey, or had traveled to highly endemic areas over the past year. Ideally, the reinfection pattern would be determined by following a cohort of children, including immunological markers of the individuals that might favor or delay reinfection, and molecular markers of the infecting parasites.
An increase of S. mansoni infection within the frame of ongoing treatment programs has also been observed elsewhere. In Senegal, for example, an elevated S. mansoni prevalence was found 10 months after praziquantel administration [44]. More recently, in Ségou district in Mali, the national control program had revealed an increase of the S. mansoni prevalence after four rounds of MDA in 7- to 14-year-old children [45]. It has been assumed that these increases of S. mansoni infections after praziquantel treatment might be explained by partial resistance to praziquantel, the acquisition of new infection, and high force of transmission [46–48].
Taken together, our data show that SBT resulted in marked decreases of S. mansoni prevalence, but the intensity of infection among infected children did not change significantly. Hence, with a single treatment round, the force of transmission in terms of egg excretion in the school-aged population has not been changed in most of our study schools. Monitoring the impact of multiple treatment rounds and “drug holidays” over the next years will provide stronger evidence of what multiple SBT rounds can achieve [13,15].
Clearly, sustainable control and eventual elimination of schistosomiasis requires multiple intervention packages, such as preventive chemotherapy (perhaps extended to all age groups), intensified case management, control of intermediate host snails, provision of WASH, and setting-specific information, education, and communication (IEC) [49–51]. In Côte d’Ivoire, the control of schistosomiasis at a national scale is still at an early stage. Indeed, the PNL-SGF was only launched shortly before this SCORE project. For the success and sustainability of schistosomiasis control in Côte d’Ivoire–and elsewhere in sub-Saharan Africa–it will be important that, in addition to preventive chemotherapy, other control measures are considered and implemented [6,7,49,52].
The present study showed that one year after SBT with praziquantel, the overall prevalence of S. mansoni infection had decreased significantly. However, in certain hotspot schools, the S. mansoni prevalence had increased unexpectedly. The infection intensity among S. mansoni-infected children was similar at the 1-year follow-up. These results demonstrated that the dynamic of schistosomiasis in the study areas is heterogeneous and that a single round of treatment is insufficient to have a lasting effect. It will be important to monitor the dynamic of schistosomiasis over the course of this SCORE study, in order to deepen our understanding of the dynamics of schistosomiasis transmission in a moderately endemic setting.
|
10.1371/journal.ppat.1002077 | Spatial Dynamics of Human-Origin H1 Influenza A Virus in North American Swine | The emergence and rapid global spread of the swine-origin H1N1/09 pandemic influenza A virus in humans underscores the importance of swine populations as reservoirs for genetically diverse influenza viruses with the potential to infect humans. However, despite their significance for animal and human health, relatively little is known about the phylogeography of swine influenza viruses in the United States. This study utilizes an expansive data set of hemagglutinin (HA1) sequences (n = 1516) from swine influenza viruses collected in North America during the period 2003–2010. With these data we investigate the spatial dissemination of a novel influenza virus of the H1 subtype that was introduced into the North American swine population via two separate human-to-swine transmission events around 2003. Bayesian phylogeographic analysis reveals that the spatial dissemination of this influenza virus in the US swine population follows long-distance swine movements from the Southern US to the Midwest, a corn-rich commercial center that imports millions of swine annually. Hence, multiple genetically diverse influenza viruses are introduced and co-circulate in the Midwest, providing the opportunity for genomic reassortment. Overall, the Midwest serves primarily as an ecological sink for swine influenza in the US, with sources of virus genetic diversity instead located in the Southeast (mainly North Carolina) and South-central (mainly Oklahoma) regions. Understanding the importance of long-distance pig transportation in the evolution and spatial dissemination of the influenza virus in swine may inform future strategies for the surveillance and control of influenza, and perhaps other swine pathogens.
| Since 1998, genetically and antigenically diverse influenza A viruses have circulated in North American swine due to continuous cross-species transmission and reassortment with avian and human influenza viruses, presenting a pandemic threat to humans. Millions of swine are transported year-round from the southern United States into the corn-rich Midwest, but the importance of these movements in the spatial dissemination and evolution of the influenza virus in swine is unknown. Using a large data set of influenza virus sequences collected in North American swine during 2003–2010, we investigated the spatial dynamics of two influenza viruses of the H1 subtype that were introduced into swine from humans around 2003. Employing recently developed Bayesian phylogeography methods, we find that the spread of this influenza virus follows the large-scale transport of swine from the South to the Midwest. Based on this pattern of viral migration, we suggest that the genetic diversity of swine influenza viruses in the Midwest is continually augmented by the importation of viruses from source populations located in the South. Understanding the importance of long-distance pig movements in the evolution and spatial dissemination of influenza virus in swine may inform future strategies for the surveillance and control of influenza, and perhaps other swine pathogens.
| Swine influenza A viruses cause severe respiratory disease in pigs, similar to that which presents in humans, and constitute an important economic concern for the US swine industry and threat to public health. Influenza was first clinically recognized in pigs in the Midwestern US in conjunction with the severe 1918 ‘Spanish flu’ H1N1 pandemic in humans [1], although whether the pandemic originated in humans or pigs remains unresolved [2]. Periodic transmission of influenza viruses between humans and swine occurs in both directions, including such notable cases as the 1976 outbreak of swine A/H1N1 influenza virus in humans in Fort Dix, New Jersey [3] and the 2009 swine-origin A/H1N1 pandemic virus in humans [4], [5]. The 1918-origin ‘classical’ H1N1 swine influenza virus circulated in US swine for 80 years with relatively few antigenic changes [6], but in the last decade the antigenic diversity of swine influenza viruses in the US has multiplied, stimulating research, development, and uptake of influenza vaccines in the US swine industry.
Currently, influenza A viruses of the H1N1, H1N2, and H3N2 subtypes all co-circulate in US swine. In 1998–1999, a triple reassortant H3N2 influenza virus emerged in US swine that possessed HA (H3), NA (N2), and PB1 segments of human H3N2 virus origin, PB2 and PA segments of avian virus origin, and NP, M1/2, and NS1/2 segments of classical swine virus origin [7] (Fig. 1). Over the next decade these H3N2 triple reassortant swine viruses further reassorted with human H3N2 viruses [8], [9], as well as with the co-circulating H1N1 classical swine viruses [10], [11]. Mainly these reassortment events involved the HA and NA segments, preserving what has been termed the ‘triple reassortant internal genes’ (TRIG) constellation (avian-origin PB2 and PA, human H3N2-origin PB1, and classical swine-origin NP, M1/2, and NS1/2).
In 2003 influenza A virus of entirely human H1N2 origin was identified in Canadian swine [12], and in 2005 H1N1 viruses with human-origin H1 and N1 segments were identified in the United States, representing two separate introductions of human H1 virus into swine that were referred to as ‘δ-1’ (H1N2) and ‘δ -2’ (H1N1) lineages based on the order of identification [13]. These human-H1 origin swine viruses also acquired novel genome segments via reassortment with other swine and human influenza viruses [12], [13].
Globally, the swine influenza virus population is spatially separated into the North American and Eurasian lineages, although both lineages co-circulate in Asia, which imports swine from North America and Europe. In the US the traditional center of swine production is located in the ‘Corn Belt’ of the Midwest, including Iowa, Illinois, Indiana, and Minnesota [14]. Beginning in the 1970's, swine production expanded into large new facilities located in the Southeastern US, mainly North Carolina, and more recently into Oklahoma in the South-central US [15]. Due to the lower cost of transporting swine versus the required amount of feed, the majority of swine born in the South-central and Southeastern regions are transported by road to the Midwestern Corn Belt to be fattened and slaughtered, resulting in continuous large-scale movements of swine (‘swine-flows’) into the Midwest [14]. However, the role of local, regional, and global swine-flows in the ecology and evolution of swine influenza viruses remains unclear.
The aim of our study was to investigate the role of inter-regional swine-flows in the spatial dissemination of newly introduced swine viruses in the US, using the human-origin A/H1 influenza virus as a case study. We utilize HA1 sequence data from a large data set of swine influenza virus isolates (n = 1,516 sequences) collected from 23 US states during 2003–2010 and apply recently developed methods of Bayesian phylogeography. The strength of the Bayesian approach is that the diffusion process among discrete location states is integrated with time-scaled phylogenies that incorporate phylogenetic uncertainty. This approach provides a formal framework to test hypotheses about viral diffusion processes driven by known population distributions and movements.
Of the 1,516 HA1 (H1) influenza virus sequences collected from swine in the United States and Canada from 2003–2010 that were included in this study, 41 were related to the human pandemic H1N1/09 virus, all of which were collected in 2009–2010 and appear to result from multiple human-to-swine transmission events. These pandemic viruses have been described previously and thus are not the focus of the present study [16]. Of the remaining 1,475 swine viruses, 327 were phylogenetically related to seasonal human H1 viruses (Fig. S1), which constitute two phylogenetically distinct clusters, representing two contemporaneous, but independent introductions of different human influenza viruses into swine (Fig. 2), consistent with previous findings [13]. Both of these clusters are phylogenetically most closely related to human H1 influenza viruses collected in early 2003. One cluster (n = 138 sequences) is related to widespread human seasonal A/H1N1 virus, while the other cluster (n = 187 sequences) is related to a less common human reassortant A/H1N2 virus that circulated globally in humans from 2001–2003. The A/H1N2 reassortant virus contains an HA derived from human seasonal H1N1 viruses and 7 segments of human H3N2 influenza virus origin [17].
We estimated the Time to the Most Recent Common Ancestor (TMRCA) for the nodes adjoining the branch that represents the human-to-swine transmission events of the H1N1 and H1N2 viruses. Accordingly, the cross-species transmission of H1N1 from humans into swine is estimated to have occurred during the period October 2002–March 2003, which coincides with the timing of the A/H1N1-dominant 2002–2003 winter influenza epidemic in humans in North America [18] (Fig. 2, Table S1). Similarly, the timeframe for the cross-species transmission of the H1N2 virus into swine is estimated to be August 2002–February 2003, which overlaps with the time period when A/H1N2 viruses circulated in humans in North America (Table S1).
To explore the whole-genome evolution of these human-origin swine influenza viruses, maximum likelihood trees were inferred for the subset (n = 31) of the human-origin swine influenza virus HA1 sequences for which the NA and internal gene sequences were publicly available at GenBank [19]. Major reassortment events are summarized in Table 1 and Fig. 1, including the H1N1 and 2003–2004 H1N2 reassortment events (#1 and #2/3 respectively, Table 1) that have been described previously [12], [13]. The PB2 phylogeny is depicted in Fig. 3, the NA (N2) phylogeny is depicted in Fig. 4, and the phylogenies of other 5 segments and N1 are available in the Supporting Information (Figs. S2, S3, S4, S5, S6, and S7). Notably, all H1N1 and H1N2 isolates collected after 2004 have acquired the triple reassortant internal genes (TRIG) cassette, which were originally derived in 1998 from avian influenza viruses (PB2 and PA), human influenza viruses (PB1), and classical swine influenza viruses (NP, M, and NS). The topology of these trees suggests that the human H1N2-origin lineage may have acquired components of the TRIG cassette approximately 3–4 times over the course of 2007–2008 via multiple reassortment events (Fig. 3, Fig. S2, S3, S4, S5, S6, and S). The largest clade (n = 21) of 2008 human H1N2-origin swine isolates (#7, Table 1) contains the TRIG, but also has acquired via reassortment a human H3N2-origin NA (N2) segment that had circulated in swine at least since 2003, when human H3N2 viruses appear to have reassorted with a lineage of swine A/H3N2 triple reassortant swine viruses that is referred to ‘clade IV’ in the nomenclature for the HA segment [9] (Fig. 4).
To investigate the spatial dissemination of these novel viruses within the US swine population, we inferred separate Bayesian phylogenies for the H1N1 and H1N2 data sets, considering the three discrete US regions that are well sampled in our data: the Midwest (IL, IN, IA, KS, MI, MN, MO, NE, OH, SD, WI), South-central (OK, TX), and Southeast (NC, SC), which are delineated broadly according to the US farm production regions defined by the USDA [20]. Distinct spatial patterns are clearly evident for both the H1N1 and H1N2 lineages that are depicted in the phylogeny presented in Fig. 2, as all of the H1N1 viruses are from the Southeast (83/138 isolates), mainly representing North Carolina, or the Midwest (55/138 isolates), whereas the H1N2 isolates are predominantly collected in the Midwest (97/169 isolates) and South-central (70/169 isolates) regions (Fig. 2). Both phylogenetic trees exhibit strong spatial structuring, and we observe a statistically significant correlation between phylogeny and location state for the Midwest (p<0.01), South-central (p<0.01), and Southeast (p<0.05) regions on both the H1N1 and H1N2 trees using the parsimony score (PS) and association index (AI) statistics [21].
The maximum clade credibility (MCC) trees annotated with most probable nodal locations indicate multiple introductions of both H1N1 and H1N2 viruses into the Midwest, with the H1N1 virus disseminating Southeast-to-Midwest, and the H1N2 virus disseminating South-central-to-Midwest. In contrast, there is little evidence of viral migration in the opposite directions, or between the South-central and Southeast regions (Fig. 2). ‘Markov jump’ counts [22] of the expected number of location state transitions along the phylogenetic branches provide a quantitative measure of gene flow between regions, representing successful viral introductions from one region to another (Fig. S8). Across the posterior distribution of trees inferred for both subtypes, the vast majority of inter-regional introductions occur in the directions of Southeast-to-Midwest (mean, 13.1) and South-central-to-Midwest (mean, 9.4), with less frequent viral migration also detected from Midwest-to-Southeast (mean, 3.3) (Table 2). Based on the number of swine transported from one region to another over the years of high sampling (2005–2008) (Table S2), we estimate that an introduction of a human-origin H1 swine influenza virus occurs roughly per million swine transported from one region to another (Table 2), although this provides only a lower boundary as the introductions are estimated based on our limited sampling, and we can only detect introductions with substantial onward transmission.
To quantitatively estimate the importance of known geographical swine population distributions and movements in the spatial dynamics of the virus, we encoded four potential predictors of viral dissemination between pairwise regions as phylogeographic models [23] and fitted these models individually to the sequence data: (i) the number of swine transported annually from one region to another (with directionality), (ii) the swine population size in the region of origin, (iii) the swine population size in the region of destination, and (iv) the product of the swine population sizes in the region of origin and the region of destination (Tables S2 and S3). Given that the South-central, Southeast, and Midwest regions are approximately equidistant from each other by road and geodesic distance, we did not consider geographical distances to be a potential predictor of viral movements in our inter-regional analysis. Bayes factor comparisons [24] via marginal likelihood estimates of the model fit for each potential predictor indicates that the spatial dynamics of the human-origin H1 virus in swine are best described by the number of swine transported annually from one region to another (Table 3). Fixing the rates relative to the swine population size of the region of destination also improved the marginal likelihood, reflecting the directionality of swine-flows from regions of relatively lower swine population size in the South-central and Southeast regions to the largest swine population found in the Midwest. The poorest marginal likelihood was obtained when rates were fixed relative to the swine population in the region of origin, indicating low rates of viral dissemination out of the large swine populations in the Midwest.
Finally, to ensure that the observed geographical patterns were not an artifact of sampling (Fig. S9), we repeated the phylogeographic analysis using a balanced data set that was randomly subsampled from the original data to obtain equal numbers of sequences from each region (n = 70). Using this balanced data set we find very similar patterns as those derived from the full data set, with substantial viral movement from South-central to Midwest and Southeast to Midwest and strongest support for the ‘swine-flows’ model (Tables S4 and S5). The numbers of viral introductions are somewhat lower than in the original analysis (Table S4) and there is weaker support for the ‘swine-flows’ model (Table S5), but this is expected given the smaller number of sequences used in the sensitivity analysis.
To capture the early spatial patterns of a newly emergent virus in swine populations prior to extensive geographical mixing, this study focused on an H1 influenza virus that was introduced twice from humans into swine around 2003. The fact that this human H1 virus was introduced into swine on two separate occasions (H1N1 and H1N2) allows, uniquely, a side-by-side comparison of the spatial dynamics of two similar emergent viruses. In our statistical analysis, we also take advantage of the independent nature of these two introductions through a model that simultaneously draws information from the H1N1 and H1N2 evolutionary histories to inform the rates of movement in an asymmetric diffusion model. The latter allows us to fully characterize the bidirectional movement between the three major sampling regions despite the fact that the independent lineages provide very different numbers of samples from these regions.
We find that the key source population of the human-origin H1N1 virus is likely to be swine in the Southeastern US, particularly North Carolina, whereas the source population of the H1N2 virus appears to be swine in the South-central US, including Oklahoma. Subsequently, both the H1N1 and H1N2 virus rapidly disseminated to the Midwestern US, apparently following the main swine transportation routes (‘swine-ways’) to the Midwest, the traditional center of American pig farming, to be fattened on the feed corn produced in the region prior to slaughter. Although the Midwest swine population is >4-fold larger than the Southeast swine population and >12-fold greater than the South-central population, the Midwest effectively serves as an ecological sink for the virus due to its commercial function as a final marketing destination and net importer of pigs. These results appear to be robust to sampling bias, as we found similar patterns of viral migration using a subsampled data set comprising 70 isolates that were randomly sampled from each of the three US regions (Tables S4 and S5).
It is certainly possible for novel lineages of influenza virus to begin their spread in the Midwest, and we have not considered farm density, climatic conditions, husbandry practices, biosecurity, vaccination status, or any other factors that would favor viral emergence in the South-central or Southeast versus the Midwest. The role of newer high-density swine production facilities in Oklahoma and North Carolina in viral evolution, in tandem with other immunological or environmental factors, clearly requires study at a finer spatial scale. Rather, our findings suggest that any viral lineage that originates in the Midwest would be less likely to spread to other US regions due to lower rates of regional exportation of Midwestern swine, whereas viruses that originate in the South-central or Southeast are likely to rapidly disseminate to the Midwest.
Although the Midwest does not appear to be a source population for swine influenza viruses, the region is likely to provide a reservoir for multiple genetically distinct variants to co-circulate and exchange segments via reassortment due to the continual importation of swine influenza viruses from other regions. Even a limited sampling (31 whole-genome sequences) revealed extensive reassortment between the human-origin swine viruses and other swine and human influenza viruses over a 7-year period. Both the human H1N1- and H1N2-origin swine viral genomes exhibit a pattern of HA and NA segments that are closely related to human viruses, but internal segments related to triple reassortant swine viruses (TRIG), suggesting that such genomic arrangements may be selectively favored (although this clearly requires further study).
Overall, our study captures the effects of at least a decade of large-scale structural changes in the US commercial swine industry on the evolution and spread of one of the most economically important pathogens in US swine. Further understanding of the role of long-distance pig transport in the ecology and evolution of swine influenza viruses may inform targeted surveillance and mitigation strategies in the future, including intensified surveillance in the less sampled Southern regions. While increased genetic and antigenic diversity observed in swine influenza viruses in recent years has stimulated ongoing research into the development of new influenza vaccines for swine, including live-virus and DNA-based approaches [25], identifying key geographical sources of the virus and reservoirs of genetic diversity may direct vaccination strategies in pigs of different age groups and specified localities. Although the patterns of viral dissemination we identify using the human-origin H1 influenza virus as a case study are striking, these findings invite further study into the phylogeography of swine influenza viruses at more precise spatial scales, including within our broadly defined Midwest region, as well as globally.
For this study we newly generated a total of 1,412 HA1 sequences (889 nt) from H1 influenza A viruses collected from swine in the United States and Canada that exhibited respiratory disease during the period 2003–2008 [26] (Table S6). Two of the isolates were swine viruses that were isolated from turkeys: A/turkey/North Carolina/00533/2005 and A/turkey/North Carolina/00536/2005, but these were triple reassortant viruses and not included in the phylogeographic analysis. HA1 gene sequences were obtained either from virus isolates or directly from the originally submitted nasal swab or lung tissue material. To isolate viruses, the swab or tissue supernatant (in 400-µl amounts) was inoculated on monolayers of MDCK cells grown in 25-cm2 flasks with 5 ml of MEM+ media [27]. All cultures were incubated at 37°C under a 5% CO2 atmosphere. All flasks were examined daily for 7 days under an inverted light microscope to observe virus-induced cytopathic effects (CPE). Viral RNA was extracted from 50 µl of swab supernatant using a magnetic bead procedure (Ambion MagMAX AM1835 and AM1836, Applied Biosystems, Foster City, CA). Segment specific PCR fragments were obtained with One-Step RT-PCR (Qiagen, CA) using influenza A specific primers for HA as described previously [28].
These data were supplemented with 104 additional HA1 sequences from H1 North American swine influenza viruses sampled during 2003–2010 that were downloaded from the National Center for Biotechnology Information (NCBI) Influenza Virus Resource (http://www.ncbi.nlm.nih.gov/genomes/FLU/FLU.html) available at GenBank [19]. This overall total of 1,516 sequences were collected from 23 US states and Canada: Arkansas (AR), Colorado (CO), Georgia (GA), Illinois (IL), Indiana (IN), Iowa (IA), Kansas (KS), Kentucky (KY), Michigan (MI), Minnesota (MN), Missouri (MO), Nebraska (NE), North Carolina (NC), Ohio (OH), Oklahoma (OK), Oregon (OR), Pennsylvania (PA), South Carolina (SC), South Dakota (SD), Tennessee (TN), Texas (TX), Virginia (VA), and Wisconsin (WI). The majority of isolates were collected from the Midwest (n = 921), followed by Southeast (n = 426) and South-central (n = 139) regions (Table S6, Fig. S9). We excluded the possibility that the spatial patterns detected were simply an artifact of uneven sampling during early emergence of the human-like H1 influenza virus in swine (2003–2005) by observing no statistical difference between the number of isolates collected in each region during 2003–2005 compared to 2006 when the virus was widespread in the US (p-value = 0.9055, Pearson's Chi-square test).
Nucleotide alignments were manually constructed for the HA1 region (889 nt) using the Se-Al program [29]. To infer the evolutionary relationships for the complete data set of 1,516 HA1 sequences, we employed maximum likelihood (ML) methods available through the PhyML program, incorporating a GTR model of nucleotide substitution with gamma-distributed rate variation among sites, and a heuristic SPR branch-swapping search [30]. This phylogenetic analysis identified a cluster of 327 sequences that were separated by a very high number of expected substitutions from the remaining 1,193 swine sequences. To explore the evolutionary origins of these highly divergent sequences in greater detail, a second tree was inferred for the 325 divergent swine sequences (two were excluded due to poor sequence quality) and 92 randomly selected human H1 (HA1) sequences: 3 H1N1 sequences selected from each of the following years: 2000, 2001, 2004, 2005, 2006, 2007, 2008, and 2009; 3 H1N2 sequences selected from 2001; plus an additional 33 H1N1 and 32 H1N2 sequences for the years 2002–2003 during which human-to-swine transmission occurred (the XML file is available in Supplemental Information, Text S1). For this data set, posterior distributions were estimated under a phylogenetic model using a Bayesian Markov chain Monte Carlo (MCMC) method implemented in the BEAST package (v1.6), incorporating the date of sampling [31]. Given the time span of our data set, sequences for which only the year of sampling was known were included and assigned a mid-year sampling date of June 1st. Only 30 of 325 isolates did not have an exact date of collection, mainly because collection dates were not available on GenBank [19]; the majority of isolates without exact dates were collected in 2008 in Oklahoma (Table S6). We employed a strict molecular clock, a flexible Bayesian skyline plot (BSP) prior (10 piece-wise constant groups), HKY85 +Γ4 model of nucleotide substitution, and the SRD06 codon position model with two partitions for codon positions (1st+2nd positions, 3rd position), with substitution model, rate heterogeneity model, and base frequencies unlinked across all codon positions. The MCMC chain was run for 100 million iterations, with sub-sampling every 50,000 iterations. All parameters reached convergence, as assessed visually using Tracer (v.1.5). The initial 10% of the chain was removed as burn-in, and maximum clade credibility (MCC) trees were summarized using TreeAnnotator (v.1.5.4).
A phylogenetic analysis also was conducted upon the 31 human-origin swine influenza viruses (3 H1N1, 28 H1N2) for which whole-genome sequences were available at the NCBI Influenza Virus Resource [19] at GenBank (http://www.ncbi.nlm.nih.gov/genomes/FLU/FLU.html) (Table S6). As the evolutionary relationships of the H1 already had been extensively analyzed (Fig. S1), we downloaded only the remaining 7 genome sequences from GenBank. Due to the divergence of the NA (N1) and NA (N2) sequences, two separate alignments were constructed. In each alignment, 15 representative human influenza viruses collected during 2001–2003 were included, representing the H3N2 (n = 3), H1N2 (n = 5), and H1N1 (n = 7) subtypes. Given the complexity of phylogenetic relationships on the NA (N2) tree arising from frequent reassortment, 99 additional human H3N2 NA sequences were included. Twenty-three swine triple reassortant H3N2 viruses collected during 1998–2009 were included as background. Varying numbers of swine H1N1 influenza virus sequences were available on GenBank for each segment as background: PB2 (n = 38), PB1 (n = 47), PA (n = 36), NP (n = 31), N1 (n = 35), N2 (n = 60), M1/2 (n = 47), NS1/2 (n = 67). Sequence alignments were manually constructed for the major coding regions of PB2 (2,277 nt), PB1 (2,271 nt), PA (2,148 nt), NP (1,494 nt), NA (1,407 nt), M1/2 (979 nt), and NS1/2 (835 nt). Regions of overlapping reading frame were deleted in the case of M1/2 and NS1/2. Here, phylogenetic trees were inferred using the maximum likelihood (ML) method under a GTR+I+Γ4 model available in PAUP* [32] for each of these 8 alignments. In all cases TBR branch-swapping was employed to determine the globally optimal tree. To assess the robustness of each node, a bootstrap re-sampling process (1,000 replications) using the neighbor-joining (NJ) method was used, incorporating the ML substitution model. Clades of related isolates were identified by high bootstrap values (>70%) and exceptionally long branch length estimates.
Due to high sampling heterogeneity among US states, we categorized each isolate into three US regions: Midwestern (IL, IN, IA, KS, MI, MN, MO, NE, OH, SD, WI), South-central (OK, TX), and Southeastern (NC, SC). These regions generally correspond to the US farm production regions defined by the US Department of Agriculture (USDA) [20], with the Midwest region including the Corn Belt (IL, IN, IA, MO, OH), Lake States (MI, MN, and WI), and Northern Plains (KS, NE, ND, SD); the Southeast region including Appalachia (KY, TN, NC, VA, WV) and the Southeast (AL, FL, GA, SC); and the South-central region corresponding to the Southern Plains region (OK, TX). Sequences from the other geographic regions that were sampled at relatively low levels were excluded, as were highly phylogenetically divergent sequences that might represent possible sequencing error. This resulted in a final data set of 127 H1N1 and 169 H1N2 isolates that could be used in our detailed spatial analysis. Although we considered separate evolutionary histories for our 127 H1N1 and 169 H1N2 human-like swine HA1 sequences, we jointly inferred the asymmetric rates of movement under a single model of discrete diffusion among the three regions to perform spatial model testing (see below). Moreover, estimating the rates of a single diffusion matrix applied to independent phylogenies may also improve statistical efficiency [23]. Posterior distributions under the Bayesian phylogeographic model [23] were estimated using a MCMC method implemented in BEAST using BEAGLE [33] to improve computational performance. The model incorporated the date of sampling and used a strict molecular clock, BSP prior, and the SRD06 model of nucleotide substitution described. The MCMC chain was run for 100 million iterations, with sub-sampling every 10,000 iterations. All parameters reached convergence, as assessed visually using Tracer (v.1.5). The initial 10% of the chain was removed as burn-in, and MCC trees were summarized using TreeAnnotator (v.1.5.4). The expected number of location state transitions conditional on the observed data was obtained using Markov jump counts [22], [34] again implemented in BEAGLE [33], and summarized per branch and for the complete evolutionary history. Ad hoc measures of the extent of geographic structure in the MCC trees were determined for the H1N1 and H1N2 data sets using the parsimony score (PS) and association index (AI) tests as available in the Bayesian Tip-association Significance testing (BaTS) program [21].
To test the importance of swine population sizes and movements in the US in the spatial patterns that were observed, we parameterized the discrete phylogeographic diffusion model in terms of four sources of state-level information on swine populations, aggregated to the regional level and normalized (mean of 1) (Tables S2 and S3). First, we used the number of swine transported annually between states in a pairwise manner for the year 2001, available through the United States Department of Agriculture (USDA) Economic Research Service (http://www.ers.usda.gov/Data/InterstateLivestockMovements/view.asp) (XML file available in the Supplemental Information, Text S2). Second, we obtained data from the USDA 2007 Census of Agriculture [35] to integrate as instantaneous diffusion rates (i) the swine population size of the region of origin (XML file, Text S3), (ii) the swine population size of the region of destination (XML file, Text S4), and (iii) the product of the swine population sizes from the region of origin and the region of destination (XML file, Text S5). Each of these predictors was incorporated into an asymmetric transition matrix that allows for separate directional rates between each pair of locations. A Bayes factor comparison [36] via the relative marginal model likelihoods was used to select the most appropriate model for the data, compared to equal migration rates (XML file, Text S6). Finally, the phylogeographic analysis was repeated using a balanced data set that was randomly subsampled from the original data to obtain equal numbers of sequences from each region (n = 70) (XML file, Text S7) and using independent rate matrices (XML file, Text S8).
All sequences were submitted to GenBank and given accession numbers CY040460 – CY082963 (Table S6).
|
10.1371/journal.pbio.1000446 | Foxp1 and Lhx1 Coordinate Motor Neuron Migration with Axon Trajectory Choice by Gating Reelin Signalling | Topographic neuronal maps arise as a consequence of axon trajectory choice correlated with the localisation of neuronal soma, but the identity of the pathways coordinating these processes is unknown. We addressed this question in the context of the myotopic map formed by limb muscles innervated by spinal lateral motor column (LMC) motor axons where the Eph receptor signals specifying growth cone trajectory are restricted by Foxp1 and Lhx1 transcription factors. We show that the localisation of LMC neuron cell bodies can be dissociated from axon trajectory choice by either the loss or gain of function of the Reelin signalling pathway. The response of LMC motor neurons to Reelin is gated by Foxp1- and Lhx1-mediated regulation of expression of the critical Reelin signalling intermediate Dab1. Together, these observations point to identical transcription factors that control motor axon guidance and soma migration and reveal the molecular hierarchy of myotopic organisation.
| Many areas of our nervous system are organized in a topographic manner, such that the location of a neuron relative to its neighbors is often spatially correlated with its axonal trajectory and therefore target identity. In this study, we focus on the spinal myotopic map, which is characterized by the stereotyped organization of motor neuron cell bodies that is correlated with the trajectory of their axons to limb muscles. An open question for how this map forms is the identity of the molecules that coordinate the expression of effectors of neuronal migration and axonal guidance. Here, we first show that Dab1, a key protein that relays signals directing neuronal migration, is expressed at different concentrations in specific populations of limb-innervating motor neurons and determines the position of their cell bodies in the spinal cord. We then demonstrate that Foxp1 and Lhx1, the same transcription factors that regulate the expression of receptors for motor axon guidance signals, also modulate Dab1 expression. The significance of our findings is that we identify a molecular hierarchy linking effectors of both neuronal migration and axonal projections, and therefore coordinating neuronal soma position with choice of axon trajectory. In general, our findings provide a framework in which to address the general question of how the nervous system is organized.
| Neural circuits are frequently organised in a topographic manner such that the position of a neuronal cell body is correlated with the location of the post-synaptic target and therefore its axon trajectory. Since the inference of such organisational principles [1], the molecular identity of many neuronal migration and axon guidance cues has been uncovered [2],[3]. Recent studies have also begun to identify the transcription factors that control neuronal identity and deploy the repertoire of neuronal migration and axon guidance receptors and signals employed in neural circuit assembly [4],[5],[6]. These observations raise the possibility that correlated neuronal soma localisation and axon trajectory of topographically ordered neural circuits arise as a consequence of specific transcription factors directing both axon guidance and cell body migration effector expression.
Vertebrate spinal motor neurons are organised myotopically in longitudinal columns such that the location of their soma in the ventral spinal cord corresponds to the position of their muscle targets in the periphery [7]. In mouse and chick, motor neurons innervating axial and body wall muscles are located in medially positioned columns, whereas motor neurons innervating limb muscles are located in the lateral motor column (LMC) present only at spinal cord levels in register with limbs. LMC neurons are further subdivided according to their axon trajectory within the limb: lateral LMC (LMCl) neurons innervate dorsal limb muscles, whereas medial LMC (LMCm) neurons innervate ventral limb muscles [8],[9],[10]. Motor pools are also organised myotopically such that, in general, the anterio-posterior location of a pool within the LMC correlates with the proximo-distal location of its limb muscle target [7],[9],[11],[12]. A motor axon guidance decision point is at the base of the limb where LMC axons interact with mesenchymal cells resulting in the selection of a dorsal or a ventral limb nerve trajectory [10],[13]. Concomitant with this process, LMC somata migrate from the progenitor-rich ventricular zone to the ventral horn of the spinal cord [14],[15], with the later-born LMCl neurons migrating past the earlier-born LMCm neurons in a manner reminiscent of the inside-out lamination of the developing cerebral cortex [16],[17],[18]. Recent studies also describe a topographic relationship between motor neuron soma and dendrite localisation in Drosophila and the patterns of motor neuron recruitment during swimming in fish [19],[20].
The molecular signals controlling the trajectory of LMC axons are characterised, but those controlling LMC soma position in the spinal cord are poorly understood. The LIM homeodomain proteins Isl1 and Lhx1, expressed by LMCm and LMCl neurons respectively, act in conjunction with the pan-LMC forkhead domain transcription factor Foxp1 to specify the dorsoventral axon trajectory in the limb by regulating the expression of axonal Eph tyrosine kinase receptors that enable LMC growth cones to respond to ephrin ligands in the limb mesenchyme. Genetic evidence argues that ephrin-A ligands in the ventral limb repulse EphA-expressing LMCl axons into the dorsal limb nerve, while ephrin-B ligands in the dorsal limb repulse EphB-expressing LMCm axons into the ventral limb nerve [21],[22],[23],[24],[25],[26]. The clustering of some motor pools relies on EphA4, type II cadherins, and the ETS transcription factor Pea3 [27],[28],[29], while migration of LMCl and LMCm neurons into their appropriate columnar location can be biased by Lhx1 and Isl1 and requires Foxp1 [21],[22],[23]. These observations raise the possibility that Foxp1, Lhx1, and Isl1 control the migration of LMC cell bodies within the ventral horn by restricting the expression of specific effectors of neuronal migration.
The extracellular matrix protein Reelin is a crucial neuronal migration signal that acts through the lipoprotein receptors VLDLR or ApoER2 to induce the phosphorylation of the intracellular adaptor protein Dab1 leading to remodelling of the actin cytoskeleton [30]. Loss of Reelin or its signalling effectors disrupts the layering of the neuronal somata within the cerebral cortex [31],[32],[33] but the role of Reelin in neuronal migration remains controversial. Reelin has been proposed to act as a neuronal migration stop signal [34]; however, since Reelin expression in the ventricular zone can partially rescue the pre-plate splitting defects in Reelin-deficient mice, Reelin has also been proposed to act as a permissive signal enabling neurons to interpret distinct migration cues [35]. Similar to cortical neurons, spinal neuron progenitor clones migrate away from the ventricular zone in radial spoke-like trajectories [14] and the migration of preganglionic (PG) motor neurons and the layering of the dorsal horn laminae is controlled by Reelin [36],[37]. These studies raise the possibility that Reelin may also regulate the localisation of LMC neurons and is thus a general migration cue specifying the position of many different classes of spinal neurons including LMC motor neurons.
Using gain and loss of function experiments in chick and mouse, we provide evidence that Reelin directs LMC neuron migration but not the selection of limb axon trajectory. We also show that Foxp1 and Lhx1, the transcription factors specifying LMC axon trajectory choice, gate Reelin signalling through the restriction of Dab1, a key signalling intermediate. Thus, the same transcription factors are directing neuronal soma migration and axon trajectory selection revealing the molecular hierarchy controlling the establishment of a somatotopic map.
To explore the possibility that Reelin signalling might control LMC soma migration, we monitored the expression of Reelin, its receptors, and their adaptor protein Dab1 in mouse embryos between embryonic day of development (e) 11.5 and e12.5 and in chick embryos between Hamburger and Hamilton (HH) stages (St) 23 and 30 [38] in limb-level spinal cord. These stages correspond to the times at which LMCl neurons are migrating out of the ventricular zone and reach their final position lateral to LMCm neurons [17],[22]. We used the transcription factor Foxp1 as a pan-LMC marker and subdivided the LMC based on the presence of Isl1 and Lhx1 transcription factors [21],[23],[25].
Reelin has previously been detected in the thoracic spinal cord adjacent to PG neurons [36]. At limb levels Reelin is expressed from e10.5 (Figure S1) and in e11.5 mouse embryos we observed Reelin expression in cells medio-dorsal to LMC neurons, and by e12.5 this domain expanded ventrally, resulting in a Reelin-rich band intercalated between the ventricular zone and the LMC (Figure 1A–H). We also observed a similar Reelin mRNA and protein distribution in chick embryos (Figure S1).
We next monitored the expression of Reelin receptors VLDLR and ApoER2 and their intracellular adaptor protein Dab1 in mouse and chick spinal cords. In e11.5 mouse embryos at both limb levels, VLDLR protein and mRNA were apparently expressed in all LMC neurons (Figure 1I–L; unpublished data). However, VLDLR protein levels appeared higher in LMCl neurons relative to LMCm neurons (Figure 1K). By e12.5 VLDLR mRNA and protein levels appeared uniform throughout the LMC (Figure 1M–P; unpublished data). In chick embryos, VLDLR mRNA was present in apparently all lumbar LMC neurons at both HH St 24 and HH St 30 (Figure S1). At the stages examined, ApoER2 mRNA was expressed in the ventricular zone adjacent to the floor plate of both mouse and chick embryos; however, its expression in LMC neurons was only apparent in mouse embryos (Figure 1Q–T; Figure S1; unpublished data).
In mouse, Dab1 mRNA and protein were present throughout the LMC from e10.5, at both limb levels; however, at later ages examined, an LMC subpopulation expressed Dab1 mRNA and protein at noticeably higher levels (Figure 1U–AF; Figure S1, Figure S4; unpublished data). At e11.5, this expression domain (Dab1high) was confined to the medio-ventral aspect of the LMC corresponding to Foxp1+Isl1− LMCl neurons while the low-level Dab1 expression domain (Dab1low) was confined to the dorsally positioned Isl1+Foxp1+ LMCm neurons (Figure 1U–X). By e12.5, Dab1high and Dab1low LMC neurons were found in, respectively, lateral and medial aspect of the LMC, and corresponded to LMCl and LMCm neurons (Figure 1Y–AB). Similar Dab1 mRNA distribution was observed in chick embryos (Figure S1). Together, our expression data raise the possibility that Reelin signalling directs LMC soma migration and the disparate Dab1 expression levels in LMCl and LMCm neurons suggest that these neuronal populations may differ in their responsiveness to Reelin.
To determine whether Reelin signalling influences LMC neuron migration, we examined the spinal cord of Dab1 and Reelin (Reln) mutant mice (Figure 2) [31],[32]. Since Reelin signalling is required for the appropriate positioning of PG neurons which share a part of their migration trajectory with LMC neurons [36],[39], we focused our analysis on caudal lumbar-sacral (LS) levels, which contain no PG neurons, as assessed by phospho-Smad1 expression [23]. During LMC migration, the total number of LMC neurons, LMCl and LMCm subtype specification, and radial glia development was unaffected by Dab1 and Reln loss of function (Figure S2, Figure S3; unpublished data). Additionally, most likely because of its impaired degradation [40], Dab1 protein levels in LMC neurons were increased in Reln mutants, suggesting that all LMC neurons are responsive to Reelin (Figure S4).
We next analysed the localisation of lumbar LMC neurons in Dab1 and Reln mutants at e12.5, the time at which, in control embryos, the majority of wild type LMCl neurons have terminated their migration and are positioned lateral to LMCm neurons (Figure 2A–D). In Dab1 mutants, LMCl neurons settled ventral to LMCm neurons, which were abnormally shifted to a lateral position in the ventral horn, and many LMCl and LMCm neurons were intermingled (Figure 2E–H). This neuronal displacement was more evident when we superimposed the position of LMCl and LMCm neurons in images of adjacent wild type (wt) and Dab1 mutant spinal cords sections (Figure 2D, H). To assess the expressivity of this phenotype and to account for LMC neuron displacement along mediolateral (ML) and dorsoventral (DV) axes simultaneously, we performed a two-dimensional position analysis of LMC neuron position using the bivariate statistical Hotelling's T2 test. We measured the mean ML and DV coordinates of wild type and Dab1 mutant LMC neurons within the ventral spinal cord. To compensate for sectioning artefacts, we normalised the ML coordinates to the distance from the ventricular zone to the lateral edge of the Foxp1+ expression domain and the DV coordinates to the dorsoventral extent of the Foxp1+ expression domain, two standard measurements that are not different between Dab1 mutants and wild type littermates (see Experimental procedures for details; unpublished data). Thus, with the lateral-most edge of the LMC defined as ML: 100%, and with the dorsal-most domain of the LMC defined as DV: 100%, in wild type embryos, the mean position of LMCm neurons was not changed significantly by Dab1 mutation; however, these neurons were spread over a larger mediolateral zone compared to wild type littermates (Figure 2I; Table S2). In contrast, by visual inspection of at least six spinal cord sections per embryo, we noted that in six out of six embryos analysed, LMCl neurons were positioned aberrantly. Quantification revealed that LMCl neuron position was significantly shifted in a medio-ventral direction in Dab1 mutants relative to wild type littermates ((ML: 73%; DV: 33%) versus (ML: 79%; DV: 39%); p<0.0035, Hotelling's T2 test; Table S2), which could be observed at least until e15.5 (Figure 2S–U, W–Y; unpublished data). A similar LMC migration phenotype was also observed in the cervical spinal cord as well (unpublished data), and in chick LMC neurons expressing a Dab1 protein in which the five tyrosines essential for Reelin signalling have been mutated (Dab15YF; Figure S5, Table S3; [41]). We also noted that in four out of four embryos, the position within the ventral spinal cord of a Pea3-expressing motor neuron pool was shifted medio-ventrally at e15.5 (Figure 2V, Z). Together, these results demonstrate that in the limb-level spinal cord, Dab1 is essential for the normal migration of LMC neurons and motor pool position.
We next examined the position of lumbar LMC neurons in Reln mutant embryos at e12.5: Reln mutation did not alter the mean position of LMCm neurons (Figure 2J–Q; Table S2), although as in Dab1−/− embryos, these neurons were spread over a larger area of the LMC when compared to controls (Figure 2R). In contrast, in three out of four embryos, we observed that LMCl neurons were positioned abnormally, with quantification revealing that the mean LMCl neuron position in Reln mutants was significantly shifted in the medio-ventral direction relative to wild type, with many LMCl neurons found intermingled with LMCm neurons ((ML: 75%; DV: 35%) versus (ML: 80%; DV: 41%); p<0.0473, Hotelling's T2 test; Figure 2J–R; Table S2). Migration defects observed in Reln mutants mirrored those observed in Dab1 mutants, thus implicating Reelin signalling in the specification of LMC soma position in the ventral spinal cord.
Based on the differential expression and the requirement for its function in LMCm and LMCl neurons, we reasoned that the levels of Dab1 expression, rather than simply its presence or absence, might influence the migration of LMC neurons. We therefore asked whether increasing Dab1 expression would shift the position of LMC soma laterally. To do this, we used in ovo electroporation to introduce a Dab1::GFP fusion protein or GFP expression plasmids into the lumbar spinal cord of HH St 17/19 embryos and monitored the position of GFP+ LMC neurons at HH St 29 [22]. Dab1::GFP was expressed with equal efficiency in LMCl and LMCm neurons and did not change their identity nor affect their axon trajectory in the limb (Figure S6; unpublished data). The mean position of LMCl neurons with elevated Dab1 levels was the same as that of LMCl neurons expressing GFP (Figure 3A–G, I; Table S3). However, in four out of five embryos, we observed that LMCm neurons with elevated Dab1 expression were observed in a more ventro-lateral position (Figure 3E–I; (ML: 70%; DV: 49%)) compared to LMCm neurons expressing GFP (Figure 3A–D, I; (ML: 67%; DV: 59%), p = 0.0165, Hotelling's T2 test; Table S3), demonstrating that increasing Dab1 expression levels in LMC neurons is sufficient to shift their position laterally.
The myotopic relationship between LMC soma position and axon trajectory within the limb raises the possibility that changes in LMC soma position in Dab1 or Reln mutants could result in the selection of inappropriate limb trajectory by LMC axons. To examine the LMCl axon limb trajectory in Dab1 mutants, we used the Lhx1tlz marker line [42] and quantified the proportion of LacZ+ LMCl axons projecting into e11.5 forelimb dorsal and ventral limb nerves in Dab1−/−; Lhx1tlz/+, and Lhx1tlz/+ littermate embryos [24]. In Lhx1tlz/+ embryos we observed ∼99% of LacZ+ axons within the dorsal limb nerves and ∼1% of LacZ+ axons within the ventral limb nerves (Figure 4A, B, E). The proportions of LacZ+ in dorsal and ventral limb nerves of littermate Dab1−/−; Lhx1tlz/+ embryos were not significantly different (Figure 4C–E; 98% and 2%, respectively, p>0.5, Student's t test). Additionally, in whole mount e12.5 Dab1−/−; Lhx1tlz/+ embryos, we did not detect any aberrantly projecting LMCl axons at either limb level (unpublished data).
To trace LMCm axons we used the hcrest/Isl1-PLAP reporter line in which the Isl1 enhancer-promoter drives the expression of placental alkaline phosphatase (PLAP) in LMCm neurons at forelimb levels [43]. PLAP enzymatic reaction was used to detect LMCm axons in Dab1−/−; hcrest/Isl1-PLAP+ and control hcrest/Isl1-PLAP+ e11.5 forelimbs, followed by axonal signal quantification. In hcrest/Isl1-PLAP+ embryos, ∼99% of PLAP+ axons were found in the ventral limb nerve, while ∼1% of PLAP+ axons were found in the dorsal limb nerve (Figure 4F, G, J), proportions not significantly different from Dab1−/−; hcrest/Isl1-PLAP+ embryos (Figure 4H–J; 99% and 1%, respectively; p = 0.335, Student's t test). LMCm limb trajectory in Reln mutants was also apparently normal (unpublished data), indicating that neither Dab1 nor Reelin are required for the selection of limb trajectory by LMC axons and demonstrating that the LMC soma position can be dissociated from axon trajectory selection.
Since our results indicated that the Dab1 protein level determines the position of LMC neuron somata but not their axon trajectory, we next evaluated whether the deployment of effector pathways governing these processes might be coordinated by a common set of transcriptional inputs. To determine whether Foxp1, a transcription factor specifying LMC cell fate, participates in the control of Dab1 expression in LMC neurons, we analyzed the embryonic spinal cords in which Foxp1 is expressed in all motor neurons (Hb9::Foxp1 transgenic) as well as in those lacking Foxp1 function [21],[23]. We first focused our analysis on upper cervical levels, where Foxp1 and Dab1 expression levels are normally low or undetectable (Figure 5A–C; Figure S7; unpublished data). In e12.5 Hb9::Foxp1+ spinal cords, compared to control embryos, we observed a significant increase in Dab1 mRNA levels (30 arbitrary (arb.) units versus 16 in controls; p = 0.002, Student's t test; Figure 5A, C, D, F, M) as well as protein expression levels associated with ectopic Foxp1+ neurons, without any obvious changes in Reelin expression (Figure 5A, B, D, E, M; Figure S7; 30 arb. units versus 16 in controls; p<0.001, Student's t test). To determine whether Foxp1 is required for Dab1 expression, we examined the lower cervical spinal cord of Foxp1 mutant mice at e12.5. When compared to controls, Foxp1 mutant spinal cords exhibited a significant decrease in Dab1 mRNA levels (15 arb. units versus 33 in control littermates; p<0.001, Student's t test; Figure 5G, I, J, L, M) as well as Dab1 protein levels (Figure 5G, H, J, K, M; Figure S7; 12 arb. units versus 37 in control littermates; p<0.001, Student's t test), demonstrating that Foxp1 is both sufficient and required for Dab1 expression in migrating LMC neurons.
Although Foxp1 controls Dab1 expression, because of its uniform expression throughout the LMC, it appeared to us an unlikely determinant of the differential level of Dab1 expression in LMCl and LMCm neurons. LIM homeodomain proteins Isl1 and Lhx1 are determinants of, respectively, LMCm and LMCl neuronal fate, can influence their migration, and can control their axon trajectory by modulating Eph receptor expression (Figure S8 and Text S1; [22],[24],[25],[42]). We thus hypothesized that while Foxp1 activates Dab1 expression in all LMC neurons, Isl1 and Lhx1 have opposing effects on Dab1: (1) Isl1 lowers Dab1 expression in LMCm neurons while (2) Lhx1 elevates Dab1 expression in LMCl neurons. We tested the first of these hypotheses by electroporating Isl1 and LacZ expression plasmids, or a control LacZ expression plasmid alone into HH St 17/19 chick lumbar spinal cords and measuring changes in Dab1 mRNA levels relative to the unelectroporated control side at HH St 29 [22]. Expression of LacZ did not affect Isl1 or Dab1 mRNA expression while overexpression of Isl1 significantly reduced Dab1 mRNA expression levels in LMC neurons (Figure S9; e/u values: 1.4 for LacZ versus 0.7 for Isl1, p<0.001, Student's t test) indicating that Isl1 can suppress Dab1 mRNA expression. To test whether Isl1 is required to control Dab1 expression, we examined the effects of siRNAs directed against Isl1 in LMC neurons but observed no significant difference in Dab1 expression when compared to controls (Figure S9 and Text S1). Together, these data suggest that Isl1 is sufficient but might be dispensable for the modulation of Dab1 expression in LMC neurons.
We next tested whether Lhx1 is required to specify the position of LMCl neurons by examining embryos with a conditional loss of Lhx1 function in LMC neurons, obtained by crossing Lhx1flox homozygotes with Isl1Cre/+; Lhx1tlz/+ mice, in which Isl1Cre drives Cre recombinase expression in all LMC neurons. We focused our analysis on e12.5 lumbosacral levels in two groups of embryos obtained from these crosses: Lhx1tlz/flox; Isl1Cre/+, designated as Lhx1COND, and control Lhx1tlz/+, designated as Lhx1+/−. Lhx1 loss of function did not affect the total number of LMC or LMCm neurons but resulted in ∼60% of LMCl neurons (Foxp1+Isl1−) losing their Lhx1 expression (Isl1−Lhx1/5+Foxp1+: 37.3% versus 95.2% in controls; p<0.001, Student's t test, Figure 6I, unpublished data). We determined the soma position of three LMC neuronal populations: LMCm, LMCl, and LMCl neurons lacking Lhx1 expression, which were defined as Isl1−Foxp1+Lhx1/5− (LMCl*). As in control embryos, in which the majority of LMCl neurons settled in the most lateral part of the LMC, in Lhx1COND embryos, a significant proportion of LMCl* neurons settled laterally and the mean position of LMCm, LMCl, or LMCl* neurons was not changed when compared to controls (Figure 6A–J; Table S4). However, in Lhx1COND embryos, many LMCl* neurons were found in medial locations, intermingled with LMCm neurons (Figure 6A–H), and these neuronal displacements were more evident when we superimposed the positions of LMCl*, LMCl, and LMCm neurons in images of adjacent control and Lhx1COND spinal cords sections (Figure S10). To further characterise the medially displaced population of LMCl* neurons, we counted the number of LMC neurons in four equal quadrants of the LMC (Figure 6J, K, unpublished data). In both Lhx1 mutant and control embryos the majority of LMCm neurons were in the medial half of the LMC (unpublished data). In control embryos, 60% of LMCl neurons were in the lateral half of the LMC, compared to 42% of LMCl* neurons in Lhx1 mutants, representing a significant change (p = 0.003, Student's t test, Figure 6K), indicating that Lhx1 is required for LMCl position specification.
To determine whether Lhx1 directs LMCl migration by controlling Dab1 expression, we compared Dab1 protein levels in the lumbar spinal cord of e12.5 Lhx1 mutants in which at least 50% of LMCl neurons lost their Lhx1 expression and littermate controls [22]. Our analysis revealed that in Lhx1 mutants, Dab1 protein expression in LMC neurons was decreased by ∼20% when compared to control embryos (Figure 7A–H, O; p = 0.038, Student's t test). We also quantified Dab1 mRNA and protein levels in the LMCm, defined as containing >90% of Isl1+Foxp1+ neurons and LMCl defined as Isl1−Foxp1+. Within the LMCm, Dab1 mRNA and protein levels were not significantly different from controls, while in LMCl of Lhx1 mutants, relative to controls, Dab1 mRNA was decreased significantly by approximately 40% (p = 0.01, Student's t test) and Dab1 protein was decreased significantly by ∼14% (p = 0.017, Student's t test, Figure 7O), indicating that Lhx1 is required for the differential expression of Dab1 in LMC neurons. Together, our results reveal that Foxp1 and Lhx1 coordinate LMC myotopy through their modulation of expression of neuronal migration and axon guidance effectors.
Our observation that Reelin is an essential signal specifying the location of LMC neurons in the ventral spinal cord allowed us to address how neuronal migration and axon guidance are coordinated to achieve topographical organisation. Our experiments demonstrate that the transcription factors specifying the axon trajectory of LMC neurons occupy a privileged position in the molecular hierarchy controlling myotopy as they also control LMC soma migration by gating Reelin signalling. Here we discuss Reelin as a motor neuron migration signal, coordination of axon trajectory selection and soma placement, and the possible functional consequences of myotopic organisation of motor neurons.
Following their birth near the ventricular zone, spinal neurons first migrate radially by perikaryal translocation, then tangentially, either in dorsal or ventral direction [14]. Reelin has been proposed as a radial migration signal; however, our observations argue that the initial, apparently radial trajectory of LMC motor neurons is Reelin signalling independent as is the case of PG and hindbrain motor neurons [36],[39]. Thus, in general, the radial migration trajectory of motor neurons might not require Reelin signalling, but once it is terminated, Reelin becomes an important guidance signal, suggesting that unlike cortical neurons that rely on Reelin for their localisation in the radial plane, motor neurons at different rostrocaudal levels of the spinal cord depend on Reelin for the tangential aspect of their migration.
How does Reelin act in motor neuron migration? The initial model where Reelin is a migration stop signal has been challenged by observations that Reelin overexpression in the cortical ventricular zone can rescue, at least in part, pre-plate splitting defects associated with Reelin loss of function [34],[35]. Likewise, overexpression of Reelin in the ventricular zone of the spinal cord rescues Reln mutant PG neuron migration defects but does not cause an overt phenotype in a wild type background [44]. In the context of LMC neurons, the Reelin expression domain is intercalated between the emerging postmitotic neurons and their final lateral position, thus precluding a function as a migration stop signal, unless at the time of their early migration LMC motor neurons are insensitive to Reelin. Our functional Reelin fragment overexpression in the ventral spinal cord resulted in LMCl motor neurons moving beyond their normal lateral position (E.P., T.-J.K., and A.K., unpublished observations); thus, in the context of motor neurons, Reelin is unlikely to function as a migration stop signal, rather, it likely promotes migration or enables LMC neurons to respond to a cue that provides spatial information.
What is the relationship of the Reelin-mediated LMC position specification to that mediated by cadherins, Eph receptors, and the transcription factor Pea3 [27],[28],[29]? Because of their restricted expression patterns and functional analysis phenotypes, these are thought to operate at the level of motor pools, in contrast to Reelin signalling which appears to specify the position of the entire LMCl division. Cadherins have been shown to be involved in the clustering of specific motor pools via their combinatorial expression imparting different adhesion properties on specific motor pools. Similarly, although the early migration of LMC motor neurons in EphA4 mutants appears to be normal, eventually the position of the tibialis motor pool is shifted. Because of these observations, it is likely that Cadherins, EphA4, and Pea3 act at a step following Reelin-mediated migration of LMCl neurons. Unfortunately, since ETS genes, arguably the earliest molecular markers of motor pools, begin to be expressed at the time when LMCl somata attain their lateral position [45], it is technically difficult to ascertain experimentally whether motor pool clustering precedes or coincides with LMCl lateral migration. The differences between the LMC position phenotypes in Dab1 and Lhx1COND mutants might shed some light on this hierarchy. In Dab1 mutants, although shifted medio-ventrally, LMCl neurons remain clustered, in contrast to Lhx1 mutant LMCl motor neurons that can be found intermingled with LMCm neurons. These observations suggest that while the Dab1 mutation probably only leads to the absence of sensitivity to Reelin, the loss of the transcription factor Lhx1 might have consequences beyond the loss of Dab1, resulting, for example, in a change in expression of cell surface adhesion molecules allowing LMCl and LMCm neurons to intermingle.
Our findings demonstrate that migration of LMC neurons within the ventral spinal cord requires Reelin signalling through the intracellular adaptor protein Dab1. This requirement is principally evident in LMCl neurons and corresponds to the high level of Dab1 protein and mRNA expressed in this population when compared to LMCm neurons. Other studies have also implicated Dab1 protein levels controlled by Cullin5 and Notch signalling as a determinant of neuronal migration [46],[47], raising the question of how might differential Dab1 expression specify LMC soma position in the ventral spinal cord. Upon activation of the Reelin pathway, Dab1 is phosphorylated and rapidly degraded [30],[34]. Therefore, in the presence of Reelin, the low Dab1 protein levels in LMCm neurons might be depleted faster than the higher Dab1 protein levels in LMCl neurons, resulting in the termination of Reelin signalling and thus a migration stop occurring sooner in LMCm neurons than in LMCl neurons. This mode of Dab1 function assumes that Reelin promotes migration of LMC neurons, or is a factor enabling their reception of a migration cue and is consistent with our observation that both LMCl and LMCm neurons can respond to Reelin. Thus similar to the Toll-like receptor (TLR) [48] and chemokine [49] signalling pathways regulated by the level of expression of a signalling intermediate, Reelin signal is differentially gated in two neuronal populations through opposing levels of Dab1 expression. In such a model, we would favour the idea that Dab1 concentration, in the presence of Reelin, is an instructive determinant of LMC neuron position, although the formal demonstration of this through, for example, the change of LMCm Dab1 levels to match exactly those in LMCl neurons is technically challenging. Following its phosphorylation, Dab1 is targeted for polyubiquitination and degradation by Cullin5 [47], raising the possibility that in LMC neurons, Dab1 protein stability might contribute to the differences in Dab1 protein in LMC neurons. However, since in LMC neurons Cullin5 is apparently expressed at equal levels by LMCl and LMCm neurons (E.P. and A.K., unpublished observations), and because of the selective enrichment of Dab1 mRNA in LMCl neurons, compared to LMCm neurons, we favour the hypothesis that differential transcriptional regulation of the Dab1 gene or its mRNA stability is an important factor contributing to Dab1 protein levels in LMC neurons.
Our results demonstrate that Dab1 expression levels in LMC neurons are set by Foxp1 and Lhx1, two transcription factors that are essential for the specification of LMC soma position [21],[22],[23]. Our data suggest the following model of Dab1 expression control in LMC neurons: a basal level of Dab1 expression in LMC neurons is induced or maintained by Foxp1, while Lhx1, a transcription factor selectively expressed in LMCl neurons, could act to elevate Dab1 expression in LMCl neurons. Additionally, based on its ability to suppress Lhx1 [22] and Dab1 mRNA expression in LMC neurons, Isl1 might function to diminish Dab1 expression in LMCm neurons. Thus, although we cannot exclude the influence of other transcription factors or distinguish whether the control of Dab1 expression by Foxp1 and Lhx1 occurs at the level of the Dab1 promoter, through intermediary transcription factors or regulation of Dab1 mRNA stability, we propose that the concerted action of Foxp1 and Lhx1 leads to differential Dab1 expression levels in LMC neurons.
Could transcription factor control of Dab1 expression be a general mechanism gating Reelin signalling in the CNS? In the cortex, examples of control of migration effectors by transcription factors include the coupling of neurogenesis to migration by bHLH control of doublecortin and p35, Tbx20 control of the planar cell-polarity pathway, and Nkx2.1 control of Neuropilin2 expression [6], but to our knowledge, a general link between a specific transcription factor and Dab1 expression has so far only been established for CREB/CREM [50]. Intriguingly, in the spinal cord, like LMC neurons, PG neurons migrate in response to Reelin and also require Foxp1 for their specification [21],[23],[36], yet although their initial lateral migration path is shared, they eventually occupy two distinct locations in the spinal cord, raising the question of the identity of the divergent migration cues that act on these two motor neuron populations.
The myotopic organisation of spinal motor neurons is the consequence of the selection of a specific axon trajectory in the limb mesenchyme and of a particular soma location within the spinal cord. The two processes can be uncoupled by loss of Reelin, Eph signalling, or mutation of Lmx1b, a LIM homeodomain transcription factor that controls ephrin ligand expression in the limb [24],[26],[42], raising the question of the molecular hierarchy controlling myotopy. Foxp1 and Lhx1 determine the selection of a dorsal or ventral LMC axon trajectory through restriction of Eph receptor expression [21],[22],[23], and our data suggest that they gate LMC neuron sensitivity to Reelin signals, thereby specifying the position of LMC soma in the ventral spinal cord. These observations imply that the selection of an LMC axon trajectory in the limb and soma position within the ventral horn are normally controlled coordinately by Foxp1 and LIM homeodomain transcription factors. Based on these observations, we propose a simple hierarchy for motor axon trajectory and soma position selection coordination (Figure 8). Foxp1 together with Lhx1 and Isl1 transcription factors are required for the expression of Eph receptors in LMC axons, and thus their repulsion from ephrin ligands in the limb mesenchyme, leading to their selection of a dorsal or a ventral limb trajectory. Foxp1, Lhx1, and possibly Isl1 also establish disparate Dab1 protein levels in LMC neurons, thus enabling their cell bodies to segregate into distinct mediolateral positions. A number of transcription factors regulating reception of specific axon guidance receptors has already been described [4],[5], implying that some of them may also direct neuronal migration, thus coordinating topographic organisation of neuronal circuits. Moreover, topographical organisation also extends to dendrite arborisation and synaptic activity [19],[51], and since Foxp1 regulates the position of motor neuron dendrites [21], it remains plausible that the transcription factors controlling migration and axon projections may be used to control other facets of topographic organisation.
Why should neuronal migrations and axon trajectories be controlled coordinately? LMC neurons within a specific motor pool, i.e. those innervating a particular muscle, are electrically coupled through gap junctions, possibly to consolidate their electrical activity patterns during the time of spinal motor circuit assembly [52]. Aberrant soma position could result in the inability of LMC neurons to form electrically coupled motor pools even though neuromuscular junctions with appropriate muscle targets in the limb might be maintained. Thus, a motor neuron might receive appropriate signals from its muscle target but is unable to synchronise its electrophysiological maturation, such as calcium transient waves [53], with other motor neurons in its pool because of their dispersed position. The emergence of functional motor circuitry also depends on the formation of specific sensory-motor contacts achieved by sensory axons synapsing on the dendrites of homonymous motor neurons within the ventral spinal cord [54]. Motor neurons in distinct pools have stereotypic dendritic arbor shapes which in principle could be dictated by the position of the motor neuron soma [28], although it remains to be determined whether motor neuron soma displacement, without any effects on molecular markers of cell fate, results in dendritic arborisation defects and whether such defects alter the sensory-motor connectivity. Reelin signalling has also been implicated in cortical dendrite formation, raising the possibility that Reln mutation might lead to LMC dendritic arbour defects independently of its effect on soma localisation. Moreover, in Reln mutant mice, although retrograde and electrophysiological analysis reveals relatively normal cortico-thalamic connectivity, retinal circuit connectivity is perturbed possibly due to defects in neuronal layer formation [55],[56]. Because of the involvement of Reelin in synapse function [57], it is difficult to dissociate the functional consequences of altered topography in Reelin signalling loss of function from altered synaptic function. However, examples of severe functional deficits caused by neural circuit topography disruption apparently independent of Reelin signalling [58] highlight the importance of topographic organisation of the nervous system.
All mice were maintained and genotyped by PCR as previously described [21],[31],[43],[59],[60],[61],[62]; Reln allele was Relnrl/J (Jackson Laboratory, USA). Fertilised chick eggs (Couvoir Simentin, Canada) were staged according to Hamburger and Hamilton [38].
Chicken Dab1L isoform (NM_204238) [63] was cloned by RT-PCR (Invitrogen, USA) and fused in frame to GFP at the C-terminus in pN2-eGFP (Invitrogen, USA).
Chick spinal cord electroporation was performed using a Ovodyne TSS20 square pulse generator (Intracell, UK) as described [24],[64].
Immunofluorescence stainings were carried out on 12 µm cryosections as described [22],[24]. For antisera used and dilutions, see Table S1.
In situ mRNA detection was performed as previously described [65],[66]. Probe sequence details are available upon request.
Images were acquired using a Zeiss LSM confocal microscope or a Leica DM6000 microscope with Improvision Volocity software. Quantification of protein and mRNA expression, GFP- and β-gal-labelled axon projections was as described [24],[65]. To quantify axon projections in hCrest/Isl1-PLAP embryos, 12 µm cryosections were immunostained (see Table S1), post-fixed, washed, and incubated at 65°C. Phosphatase activity was revealed simultaneously in sections containing mutant and control tissue. The signal was quantified in sections sampled at 30–50 µm rostrocaudal intervals at the cervical level with at least six sections analysed per embryo.
All quantifications were done between lumbosacral (LS)4 and LS6 levels as assessed by vertebra counts and absence of pSmad1+ PG neurons [23]. Neurons were imaged in 12 µm cryosections sampled at 100 µm intervals using a Zeiss LSM confocal or Leica DM6000 fluorescent light microscope; ML and DV values were calculated using ImageJ software measurements of distance (D) and angle (α) of motor neuron soma from the ventral edge of the ventricular zone (see Text S1 for details) and then plotted using Matlab software running the “dscatter” function, which creates a scatter plot with contour lines linking data points with similar frequency and colour intensities that increase with data point frequency.
In all cases, to compare the vectors of means between experimental and control groups, we used a two-sample Hotelling's T2, which is a two-dimensional generalization of the Student's t test, combined with a randomization test under the assumption of unequal variances, which does not rely on the stringent assumptions of the parametric Hotelling's T2, to circumvent the difficulty of having moderately sized samples. The analysis was implemented using the NCSS software package (Hitze J. (2007); Kaysville, Utah, www.ncss.com).
|
10.1371/journal.pgen.1002378 | Effect of Host Species on the Distribution of Mutational Fitness Effects for an RNA Virus | Knowledge about the distribution of mutational fitness effects (DMFE) is essential for many evolutionary models. In recent years, the properties of the DMFE have been carefully described for some microorganisms. In most cases, however, this information has been obtained only for a single environment, and very few studies have explored the effect that environmental variation may have on the DMFE. Environmental effects are particularly relevant for the evolution of multi-host parasites and thus for the emergence of new pathogens. Here we characterize the DMFE for a collection of twenty single-nucleotide substitution mutants of Tobacco etch potyvirus (TEV) across a set of eight host environments. Five of these host species were naturally infected by TEV, all belonging to family Solanaceae, whereas the other three were partially susceptible hosts belonging to three other plant families. First, we found a significant virus genotype-by-host species interaction, which was sustained by differences in genetic variance for fitness and the pleiotropic effect of mutations among hosts. Second, we found that the DMFEs were markedly different between Solanaceae and non-Solanaceae hosts. Exposure of TEV genotypes to non-Solanaceae hosts led to a large reduction of mean viral fitness, while the variance remained constant and skewness increased towards the right tail. Within the Solanaceae hosts, the distribution contained an excess of deleterious mutations, whereas for the non-Solanaceae the fraction of beneficial mutations was significantly larger. All together, this result suggests that TEV may easily broaden its host range and improve fitness in new hosts, and that knowledge about the DMFE in the natural host does not allow for making predictions about its properties in an alternative host.
| Mutations are the raw material on which natural selection operates to optimize the fitness of populations. The occurrence of selection and its strength depend on the effect that mutations may have on the survival and reproduction of individuals: mutations can be lethal, deleterious, neutral, or beneficial. Thus, determining how many mutations belong to each of these categories is of importance for predicting the evolutionary fate of a population. For emerging infectious diseases, this distribution determines the likelihood that a pathogen crosses the species barrier and successfully infects a new host. We characterized such distributions across a panel of alternative hosts for a plant virus and found that fitness effects of individual mutations varied across hosts in an unpredictable way and that many mutations considered deleterious in the natural host may turn out to be beneficial in other hosts.
| The emergence of new epidemic viruses is a critical issue for public health and economic welfare [1]–[7]. Virus emergence is a complex, multilevel problem that results from a combination of ecological and genetic factors [5]–[8]. The increasing threats imposed by emerging and re-emerging viruses make it even more urgent to predict whether and when virus populations replicating in their reservoir hosts will acquire the ability to successfully infect individuals of a new host species, adapt to it and, eventually, turn into an epidemic. To make such predictions we must first identify the factors determining why some viruses, like Hepatitis C virus, Human immunodeficiency virus type 1 (HIV-1), Influenza A virus or Cucumber mosaic virus have been so successful in causing pandemics whereas other viruses such as SARS coronavirus, Ebola virus, Hantan virus, or Cocoa swollen shoot disease virus produced outbreaks limited in time and space. A pre-requisite for viral emergence is the existence of standing genetic variation within the reservoir host that enables successful virus replication within naïve hosts after spillover by chance [2], [3], [8]. As a first approximation, and neglecting the effect of genetic drift, the frequency of these host-range mutants in the reservoir population will directly depend on the equilibrium between (i) the rate at which they are produced and (ii) the fitness effects they may have in the reservoir host.
If host-range mutations are deleterious in the reservoir host, their frequency will be low and thus the likelihood of emergence will be low as well, whereas if they are neutral or perhaps even beneficial, their frequency will increase, which will in turn increase the chances of emergence. RNA viruses are characterized not only by extremely high mutation rates [9], but also by short generation times and large population sizes [3], [8]. For these reasons RNA viruses have a high evolutionary potential and are over-represented among emerging viruses. Regarding fitness effects, extensive data have shown that host-range mutants have high fitness in the new host but pay fitness penalties in the reservoir host [10]–[13]. This fitness trade-offs should also preclude the evolution of generalist, multi-host viruses [11], [13]–[15]. Antagonistic pleiotropy is often called to explain the existence of such fitness trade-offs [11], [13]. However, an alternative, although not mutually exclusive, mechanism promoting host specialization is the accumulation of neutral mutations in the genes that are not necessary in a given host but are essential in alternative hosts, making these mutations deleterious in the alternative host environment [14], [15].
Therefore, to predict the probability of a virus to infect new hosts, it is necessary to characterize the distribution of mutational fitness effects (DMFE) on its primary hosts as well as on potential new hosts. DMFE across hosts show the fraction of all possible mutations that may be beneficial in new hosts and reveal their fitness effects in the primary host. DMFE have been characterized in recent years for a handful of single-stranded DNA [16], [17] and RNA viruses [16], [18]–[20] in their primary hosts. All these studies but one [18] took a similar experimental approach to the characterization of DMFEs. In all cases, site-directed mutagenesis was performed on infectious clones, generating collections of random single-nucleotide substitution mutants. The fitness of these mutants was then measured by means of competition experiments against the parental non-mutated virus. In [18] (and in some experiments described in [16]), an undetermined number of mutations were fixed by genetic drift in the absence of purifying selection (Muller's ratchet). Three commonalities can be found in these studies [21], which we will briefly summarize. First, all viruses examined show very low tolerance to mutation, as demonstrated by a large fraction of lethal mutations (between 20% and 40%). Second, for non-lethal mutations, the mean fitness loss associated to a single nucleotide substitution is about 10%. Third, DMFEs characterized are left-skewed (i.e., containing more negative values than the Gaussian distribution) and leptokurtic (i.e., comprising less central values than the Gaussian and having longer tails). Accordingly, the probability density functions that better fitted the data were from the heavy-tailed family (Log-normal or Weibull) or highly skewed ones (Gamma or Beta). Still, probably due to the overwhelming amount of work associated with these studies, the effect of host heterogeneity on the properties of DMFE have not been experimentally addressed; with the exception of the work done by Van Opijnen et al. [22] with HIV-1. However, this study was limited to a few single nucleotide-substitution mutations that were not randomly scattered along the viral genome but concentrated in a regulatory non-coding region.
The situation that we have just described in the context of emerging viruses is a particular case of a more general biological problem: the extent to which a phenotype (here viral fitness) is determined by the interaction between the genotype and the environment (here the host species), or the G×E interaction [23]. Understanding how genotype and environment interact to determine the phenotype and fitness has been a central aim of ecology, genetics, and evolution. Therefore, it should also be central for the epidemiology and evolution of infectious diseases; even more so in light of the reasons given above. The fate of genetic variation in populations depends on the form of the G×E interactions [24], [25] and, for instance, a change in the rank order of genotypic fitness in different environments may support a balanced polymorphism [25]. Despite this centrality, not much is known about the extent and underlying form of G×E interactions. Previous attempts to answer these questions suffer from one or another weakness (e.g., non-random samples of mutations taken from standing variation formerly filtered by selection, unknown number of mutations, traits of unclear relationship with fitness, etc.) [26]. To overcome these problems, Remold and Lenski [26] proposed using a collection of mutant genotypes that differ from the wildtype in a single and well defined mutation. Mutational fitness effects should further be evaluated in environments not previously experienced by the organism. By applying this simple experimental design to the bacterium Escherichia coli, these authors found that G×E interactions were quite common even for genotypes that differed by only one mutation and across environments that differed in a single component.
In this study, we sought to study how different host species affect the parameters describing the DMFE for a plant RNA virus, Tobacco etch potyvirus (TEV). Furthermore, we were interested in testing whether single point mutations are sufficient to give rise to G×E interactions in simple and compacted RNA genomes. To do so, we randomly selected 20 single-nucleotide substitution mutants from the collection previously described in Carrasco et al. [20]. Then, we quantified the absolute fitness (i.e., Malthusian growth rate) of all these mutants in eight different host species and characterized the parameters describing the DMFE and how they varied across hosts. Furthermore, we evaluated the amount of observed variability that was explained by genetic differences among viral genotypes, by differences among host species and, more interestingly, by the non-linear interaction between these two factors (e.g., the genotype-by-environment variance). In nature, TEV infects five of these hosts (Nicotiana tabacum, Nicotiana benthamiana, Solanum lycopersicum, Capsicum annuum, and Datura stramonium), all belonging to the same plant family, the Solanacea. The other three species are not TEV natural hosts, although they are experimentally susceptible to systemic infection. They belong to two plant families, the Asteraceae (Helianthus annuus) and the Amaranthaceae (Gomphrena globosa and Spinacea oleracea). Both the Solanaceae and the Asteraceae are within the Asterids, while the Amaranthaceae are not [27].
For this study, we have used a collection of 21 TEV genotypes (20 mutants plus the wildtype) drawn from a larger collection of mutants obtained by Carrasco et al. [20]. Each mutant contained a single nucleotide change whose position and substitution were chosen at random. In 14 cases, the mutation resulted in an amino acid substitution (Table 1). Our set of mutants consisted in changes that were randomly dispersed throughout the TEV genome (Table 1). Selected mutants were all viable in the natural host N. tabacum. The absolute fitness effects of these genotypes were evaluated in eight susceptible host species. The observed DMFEs for the 21 genotypes in all eight hosts are shown in Figure 1. A quick look at these histograms suggests that in the natural host N. tabacum and in its close relative N. benthamiana (both species belong to the same genus of the Nicotianoideae subfamily) most mutants have absolute fitness indistinguishable from or below the value of the wildtype (indicated by the vertical dashed line; enumerated in Table 2). Indeed, the average absolute fitness values for all mutant genotypes on these two hosts were significantly smaller than the values estimated for the wildtype (Table 2; one-sample t-tests, P≤0.019 in both cases). Also supporting this excess of deleterious effects, the distributions had significant negative skewness values (Table 2; t-test comparing to the Gaussian null expectation, P<0.001 in both cases). The average absolute fitness effect of all genotypes together was undistinguishable in these two hosts (Mann-Whitney test, P = 0.232). Both distributions are also significantly leptokurtic (Table 2; t-test comparing to the Gaussian null expectation, P<0.001 in both cases), indicating that many mutations have mild fitness effects and, therefore, the DMFEs are more peaked than expected for a Gaussian distribution. When the absolute fitness of the different TEV mutants was evaluated in hosts whose genetic relatedness to N. tabacum decreased, while still belonging to the Solanaceae (Solanoideae subfamily: D. stramonium, C. annuum and S. lycopersicum), the average value of the distributions did not shift significantly compared to Nicotianoideae (Mann-Whitney test, P = 0.348). In addition, it remained skewed towards the left tail, that is, the values were smaller than the median of the distribution (Table 2; t-test, P≤0.026). In D. stramonium and S. lycopersicum, a few mutations were lethal (see below the arguments supporting the lethality of these mutants), thus making the distributions even more negatively skewed. The change in shape of DMFE noticeably affected the kurtosis parameter. In the three Solanoideae hosts DMFEs have no significant kurtosis (Table 2; t-tests, P≥0.195 in all cases), and thus they are effectively mesokurtic (e.g., Gaussian-like). In general, DMFE dramatically change in several aspects within non-Solanaceae hosts. First, the distribution mean shifts towards lower values; a comparison of absolute fitness values between Solanaceae and non-Solanaceae hosts indicates that the difference is highly significant (Mann-Whitney test, P<0.001). Second, the distributions become positively skewed, although the asymmetry was significant only for S. oleracea (Table 2; t-test, P = 0.008). Positive skewness means that the tail of the distribution containing fitness effects higher than the mean is significantly heavier than the negative tail. This finding is particularly interesting when observed that the fitness of the wildtype is always in the negative tail of the distribution.
To further expand the analyses of the data shown in Figure 1, we compared the absolute fitness of each mutant to that of the wildtype TEV on each host using the bootstrap method described in [18]. Based on the bootstrap results, mutations were classified into lethal, deleterious (i.e., significantly smaller absolute fitness than wildtype), neutral, and beneficial (i.e., significantly larger absolute fitness than wildtype) on each alternative host (Table 2). The analysis of this contingency table shows that there is a significant heterogeneity in the distributions of discrete mutational classes among hosts (χ2 = 163.262, 21 d.f., P<0.001). However, this heterogeneity is entirely driven by the differences among TEV absolute fitness in Solanaceae hosts (χ2 = 96.161, 12 d.f., P<0.001), but not among non-Solanaceae hosts (χ2 = 0.891, 6 d.f., P = 0.989). Indeed, if a new contingency table is constructed by grouping hosts into Solanaceae and non-Solanaceae, a significant heterogeneity is observed among the two host classes (χ2 = 37.884, 3 d.f., P<0.001). These results are explained by the shift from more neutral mutations in the two Nicotianeae towards more beneficial and lethal in the three Solanoideae, while the three non-Solanaceae species had similar counts of neutral and beneficial mutations. Interestingly, neutral and non-neutral cases were evenly distributed among synonymous and nonsynonymous mutations for all hosts (Fisher's exact test, P≥0.131 in all hosts). In recent years, increasing evidence supports the notion that, for compacted RNA genomes, synonymous mutations are not necessarily neutral mutations [20], [28]. This observation is most likely due to the overlapping nature of many viral genes, the existence of secondary RNA structures essential for regulating gene expression, the adaptation to the host's codon usage bias, and the pressure for evading RNAi-based host defenses.
The above classification of viable mutants into deleterious, neutral or beneficial depends on whether their fitness values deviates significantly from that of the wildtype TEV in the bootstrap test. However, given the statistical uncertainties inherent to our measurements, it is difficult to distinguish between small-effect mutations and lack of fitness effects. For the Solanaceae, relative fitness values<−0.03 were generally significantly deleterious, whereas mutations were assigned to the beneficial class if they had relative fitness >0.05 as in S. lycopersicum, although the threshold for C. annuum rose up to >0.2. For the non-Solanaceae, in general, mutations were considered as beneficial if they had relative fitness values >0.05. However, since the concept of neutrality depends on the effective population size [29], modeling the continuous DMFE rather than their discretization, at length, is to be more informative. In the next section we will address this problem.
Failed inoculation experiments and lethal mutations produce the same apparent result: a lack of viral accumulation in the inoculated plants. To rule out the possibility that the putative lethal mutations observed in D. stramonium and S. lycopersicum are just a succession of failed inoculation experiments, we applied the following statistical argument. First, we evaluated our rate of failure to produce an infection when starting the experiment with viruses that are viable in each host species. In the case of D. stramonium, two mutants were assigned to the class of lethals. Out of 171 D. stramonium plants inoculated with viable viruses, 72 plants were infected and thus the failure rate was 1–72/171 = 0.579 per inoculation event. After nine trials (corresponding to the number of replicates per mutant and per host species), the probability of failing all cases should be 0.5799 = 0.007. Therefore, in a sample of 21 genotypes, we expect less than one case (21×0.007 = 0.153) to be erroneously assigned to the category of lethal mutations. Similarly, in the case of S. lycopersicum, where eight mutants were putatively lethal, 72 out of 117 plants inoculated with viable viruses were infected, which represents a failure rate of 0.385 per inoculation experiment. From this, we expect (21×0.3859 = 0.004) much less than one case to be classified as lethal but resulting from multiple inoculation failures. Therefore, on these grounds, we are confident that the mutations classified as lethal on these two hosts were really so.
Next, we sought to determine which of several competing statistical models better describes the observed DMFEs. Following previous analyses of the DMFE for RNA viruses [16], [18], [19], [20], we evaluated the goodness-of-fit of distributions sharing the property of asymmetry and with heavy tails to the empirical DMFEs observed in each host. Lethal mutations were excluded from the analyses. The probability density functions (pdf) tested were: Exponential, Gaussian, Gamma, Beta, Log-normal, Laplace, Pareto, and Weibull. Nonlinear regression techniques were used to fit models to the data. Table 3 shows the best-fitting model for each host and the relevant parameters describing each distribution, as well as the statistics measuring the goodness of fit (Akaike's weight and R2). The Weibull pdf was the model that better described the DMFEs measured in N. tabacum, N. benthamiana, D. stramonium, S. lycopersicum, and G. globosa. A Weibull pdf is described by two parameters, the scale λ and the shape κ, related to the expected value of the distribution as , where Γ(·) is the gamma function evaluated at the given argument. However, the Akaike's weight for this pdf is <0.95 in all cases, suggesting that alternative models, or combinations of models, can still contribute to better describe the observed distributions. In the cases of C. annuum and S. oleracea the pdf that better explained the observed DMFEs were Laplace and Pareto, respectively. These two distributions are from the power-law family. In the case of the Laplace pdf, the expected fitness value is equal to the location parameter E(m) = μ, whereas in the case of the Pareto, the expected value is , where α is the shape parameter and c the threshold value. For the two non-Asterids hosts (e.g., G. globosa and S. oleracea) the expected fitness values were negative, whereas in all other cases the expected fitness values were positive and in the range 0.02–0.311.
The Akaike's weight informs about which one among a set of competing models is best supported by the data, after ranking them according to their AIC values. However, given the uncertainties associated to the small sample size here used (21 TEV genotypes), one may be interested in evaluating how much better performs the best fitting model relative to any other model. To make this analysis, we used an evidence ratio (ER) computed as the likelihood of the best model divided by the likelihood of the alternative model of interest [30]. The last column in Table 3 shows the ER values computed for models ranked in second place. The Weibull pdf is the best descriptor in five out of eight host species. Hence, one may ask how good a descriptor it is for the three remaining hosts. In the case of C. annuum, the Weibull was ranked as the second best fitting, performing only ∼5.5 times worse than the Laplace pdf. For H. annuus, the Weibull pdf ranked in third position, with an ER = 38609.153, thus providing a much worse fit than the Laplace pdf. Finally, in the case of S. oleracea the Weibull pdf ranked in seventh position, with an ER = 190935.254, indicative of a very poor fit compared to the best fitting Pareto pdf.
Next, we sought to evaluate whether the location and shape characteristics of the DMFE were affected by the genetic relationship between the hosts. Figure 2a shows that a statistically significant negative correlation (Spearman's rS = −0.798, 6 d.f., P = 0.018) exists between the expected centrality parameter of the DMFE, E(m) (taken from Table 3), and the ranked phylogenetic distance of each host to the natural one; N. tabacum. This negative correlation indicates that the average absolute fitness decreases as the host becomes more and more distant from the one to which the virus was originally adapted. By contrast, a significant positive correlation has been observed between the skewness of the DMFE and host's phylogenetic distance from the natural one (Figure 2b; Spearman's rS = 0.877, 6 d.f., P = 0.004). This result is congruent with the above observation that the skewness of the DMFE shifts from negative to positive as hosts become more phylogenetically distant from the natural one. The phylogenetic distance did not significantly affect the variance and kurtosis of the distributions (in both cases Spearman's rS≤0.569, 6 d.f., P≥0.153).
Model I in Table 4 shows the GLM analysis of the absolute fitness data using host species and TEV genotype as random factors. First, there is a highly significant difference among TEV genotypes in their absolute fitness. This is in agreement with previous analyses of the larger collection of genotypes from which these 20 were drawn [20]. However, only ∼4% of total observed variability is explained by genetic differences among TEV genotypes. There is also a highly significant effect of the host species on viral fitness, which explains ca. 26% of the observed variability in absolute fitness. Finally, and more interestingly from the perspective of predicting emerging viral infections by using information about fitness effects in natural hosts, the G×E interaction term is also highly significant, and explains ca. 67% of the observed variability in absolute fitness. This significant interaction means that we cannot accurately predict a particular genotype's absolute fitness in a given host from the main effects, thus adding an unpredictability component to viral emergence. Finally, it is worth noting that only 2.76% of the observed variance remained unexplained by the model and was used as error variance in the computation of the different variance components.
To account for the fact that hosts are not independent but phylogenetically related, we fitted a more complicated model to the data (Model II in Table 4). This alternative model treated the host species as a binary factor; belonging to one of two classes (Solanaceae vs. non-Solanaceae). Then, host species were nested within these two classes and the G×E component was evaluated by looking the significance of the interaction between hosts within classes and TEV genotype. This model has an appreciably lower AIC value than the Model I and thus should be taken as a better one, although the conclusions do not qualitatively depart from those reached from the simpler model (Model I): the genetic component only explains a minor fraction of observed fitness variance whereas most of it is explained by the G×E interaction term.
A significant G×E interaction can be produced by two non-mutually exclusive mechanisms [26]. First, pleiotropic effects may change the rank order of mutations across environments (e.g., a mutation beneficial in one environment may not be so in an alternative one). Second, while still retaining the rank order of fitness effects, G×E can also be generated by altering the genetic component of phenotypic variance () across hosts. To evaluate the contribution of these two mechanisms to the observed G×E, we run two different analyses.
As a first statistical test, we computed Spearman's rank correlation coefficients between absolute fitness values in the primary host N. tabacum and the values estimated on each alternative host (Figure 3). Lethal mutations were assigned to the lowest rank. A negative correlation would indicate negative or antagonistic pleiotropy (e.g., mutations change the strength and sign of their effects on different hosts), whereas a positive correlation would indicate positive pleiotropy. Interestingly, the correlations were positive for all the Solanaceae hosts (although only reached significance in two cases, N. benthamiana and D. stramonium). By contrast, for the three non-Solanaceae hosts the correlation coefficients had negative non-significant values. We used the frequency of discrete mutational signs on each host class to construct a contingency table, and applied a Fisher's exact test to confirm that the difference in correlation signs among host classes was significant (P = 0.029) despite the small sample size. Furthermore, the shift from negatively skewed DMFE (excess of deleterious effects) in the Solanaceae to positively skewed distributions (excess of beneficial effects) in the non-Solanaceae described above is also consistent with antagonistic pleiotropy. Therefore, from these analyses we concluded that antagonistic pleiotropy contributed to generate G×E when the new host species are phylogenetically distant from the natural host (i.e., outside the plant family), but not when host species belong to the same family. Nevertheless, this conclusion needs to be qualified because the most extreme cases of antagonistic pleiotropy are mutations that were viable in N. tabacum but lethal in D. stramonium and S. lycopersicum, all being from the same family.
A non-significant correlation test, however, cannot be taken as an evidence of a lack of pleiotropic effects across hosts. For instance, one can imagine a situation in which, in a given host, some mutations may have negative pleiotropic effects, some others positive ones and some even being independent on the host. In such situation, the correlation would turn out to be non-significant while still some mutations may be pleiotropic. To overcome this drawback, we performed a second statistical test based on the frequency of mutations that changed the sign of its fitness effects (compared to that of the wildtype TEV) across hosts. For each mutation on each host, we recorded whether fitness was lower (negative sign) or higher (positive sign) than the wildtype TEV. Then we counted the number of cases for which the sign changed between the primary host, N. tabacum, and each alternative one. If a mutation has the same sign both in the primary and in the alternative hosts, it is considered not to be pleiotropic. By contrast, if sign changes, then it is considered as pleiotropic. Under the null hypothesis of no excess of pleiotropic effects, mutations would distribute evenly across both categories. Departures from this null hypothesis were evaluated using Binomial tests. Only in N. benthamiana (x = 2) and D. stramonium (x = 4) the number of observed mutations with putative pleiotropic effects was not significantly larger than expected under the null expectation (probability of having x or more cases of pleiotropic mutations than expected by sheer chance: P<0.001 and P = 0.006, respectively). By contrast, the number of mutations whose fitness effects switched signs were significantly larger than expected by chance in all other hosts: x = 18 in C. annuum (P>0.999), x = 19 in S. lycopersicum (P>0.999), x = 14 in H. annuus (P = 0.942), 15 in G. globosa (P = 0.979), and 17 in S. oleracea (P>0.999). Therefore, this second test of antagonistic pleiotropy confirmed the conclusions drawn from the Spearman's correlation test. Moreover, it showed that antagonistic pleiotropy also made an important contribution to the fitness variability observed in the two hosts (C. annuum and S. lycopersicum) in which no overall tendency was observed in Figure 2.
Next, to evaluate the importance of changes in genetic variance, , for absolute fitness as a source of G×E we computed it for each of the eight host species. Table 5 shows the estimates of , of error variance () as well as the broad sense heritability (H2) that indicates the percentage of total phenotypic variance explained by genetic differences among TEV genotypes. For the five Solanaceae hosts, ranged from 0.051 to 0.115, with an average value of 0.083, and explaining >95% of the observed phenotypic variance. By contrast, within the non-Solanacea hosts was significantly smaller (Mann-Whitney test, P = 0.036), with an average value of ∼0.002. Besides, for these hosts only ca. 25% of phenotypic variance for absolute fitness was explained by genetic differences among TEV mutants. Henceforth, from these analyses we conclude that changes in genetic variance for absolute fitness contributed to the observed G×E only when comparing phylogenetically distant hosts.
All together, these results suggest that G×E arises from the combined effect of antagonistic pleiotropy and reductions in genetic variance associated to the shift from hosts that belong to the same family as the natural host to hosts that do not belong to this family.
New emerging epidemic viruses represent one of the most serious threats to human, animal and crops health [1]–[8]. The problem of viral emergence is complex and depends on the interaction between host's genetics, vectors' abundance, ecology, and virus evolvability. Predicting the potential of a virus to spillover from its natural host reservoir to few individuals of a new host species and successfully establish a productive infection that will trigger a new epidemic seems an insurmountable problem. However, from the perspective of evolutionary genetics, the problem can be simplified by considering that the fate of the viral population entering into the new host depends, in a first instance, on whether it contains genetic variants with a positive fitness value. In other words, a pre-requisite for predicting the ability of a virus to expand its host range is to have information about the distribution of fitness effects associated to mutations (DMFE) across all possible hosts. In this study, we have characterized DMFE across a set of hosts for the plant virus TEV. The host species selected widely ranged in their degree of genetic relatedness with the natural host, N. tabacum: from very close relatives (members of the same genus) to members of other genera within the same family, and finally, to species belonging to different families within the same class or even to different classes. We found that the central parameter of the DMFE shifted towards smaller values as the phylogenetic distance of each host from tobacco increased (Figure 2a). The distributions did not just displace; they also changed in shape, moving most of the probability mass from the negative to the positive tails. This means that, on average, the absolute fitness of TEV decreased as hosts became more different from the natural one. However, if the fitness of individual mutant genotypes is expressed relative to wildtype virus, the change in shape means that the number of (conditional) beneficial mutations increases as hosts become more phylogenetically distant from tobacco. This suggests that the number of mutations that may potentially expand TEV host range is large. A similar abundance of host-range mutants was also observed for phage φ6 [12]. In this case, the mutations were concentrated in the P3 gene that encodes for the protein responsible for attaching the virion to the bacterial pili. However, in our case, host-range mutations do not concentrate in any particular gene but were scattered along the genome. Notably, Gaussian fitness landscape models [31] predict an increase in the proportion of beneficial mutations under stressful conditions (here represented by those hosts in which absolute fitness was dramatically reduced).
The shape of DMFE is a critical component of many mathematical models of evolutionary dynamics, including the molecular clock, the rate of genomic contamination by Muller's ratchet, the maintenance of genetic variation at the molecular level, and the evolution of sex and recombination [32]. In more practical terms, characterizing the shape of DMFE is essential for understanding the nature of quantitative genetic variation, here including complex human diseases as well as pathogens virulence [32]. Therefore, it is not surprising that much effort has been recently invested in characterizing the DMFE for many organisms (reviewed in [32]), including several RNA and DNA viruses. Despite differences in the genetic material of these viruses, their sizes and gene contents, the methodology applied has been similar in all cases, namely, generating collections of single-nucleotide substitutions mutants and then characterizing the fitness of each of these mutants relative to the non-mutated parental. In RNA viruses such as bacteriophage Qβ [16], Vesicular stomatitis virus (VSV) [19] and TEV [20], over one third of mutations generated unviable viruses, whereas viable mutations reduced fitness, on average, by ∼10% [21]. Regarding the theoretical pdf that better explained these datasets, VSV fitness data conformed to a complex distribution combining a Log-normal and an Uniform pdfs, the original TEV larger dataset was best fitted by a Beta pdf (notice that in [20] fitness was measured as a relative value, which may justify the difference to the Weibull pdf conclusion reached here), and the Qβ DMFE was well described by a Gamma pdf. In the case of DNA phages φX174 [16] and f1 [17] the fraction of unviable mutations was lower (one fifth) but the average effect of viable mutations was almost identical to the one reported for RNA viruses [21]. φX174 best fitting was to the Exponential pdf whereas for f1 the Log-Normal and the Weibull fitted equally well. Taken together, all these results suggested the existence of certain common rules: a large fraction of mutations are lethal or have a large negative fitness effects (displaying the fragility of viral genomes). In addition, DMFE for viruses are highly asymmetric and can be reasonably well described by theoretical pdfs with heavy tails. In a recent study [33], the reason for this generality was grounded into the thermodynamic properties of protein folding, suggesting that the effect of mutations on protein folding and stability was a good explanation for the observed DMFEs. Despite being important for understanding the evolution of a virus in its natural host, these results were, even so, insufficient to understand the likelihood of a virus expanding its host range. Here, we have contributed to cover this lack of knowledge by describing the effect of changing hosts on the properties of DMFE. One of the most striking conclusions from our study is that the fraction of lethal, deleterious, neutral and beneficial mutations, and hence the shape and location of the distributions, radically depends on the host in which the fitness effects of mutations is evaluated, and that this dependence is, itself, conditioned by the phylogenetic distance among hosts. Furthermore for host species belonging to the same family as the primary host, the Weibull pdf fitted best (or second to best for C. annuum) model to describe DMFE, although for hosts outside the family this model is the best only in one out of three cases (Table 3).
Martin and Lenormand [31] proposed three possible outcomes for the DMFEs measured in permissive vs. stressful environments: (i) conditional expression means that some mutations have a detectable fitness effect in some environments but are neutral in others, (ii) conditional average means that the average mutational effect differs between the two types of environments and (iii) conditional variance, meaning that variance in mutational effects changes between the two types of environments. In a survey of DMFE across benign and stressful environments for organisms as diverse as the fungi Saccharomyces cerevisiae and Cryptococcus neoformans, the nematode Caenorhabditis elegans, and the fruitfly Drosophila melanogaster, Martin and Lenormad [31] found that stressful conditions tend to inflate the variance of the DMFE while leaving the central value of the distributions almost unaffected. These results contrast with those reported here: for TEV, DMFE evaluated in stressful hosts (the non-Solanaceae) had lower average (Figure 2a) and more positive skewness (Figure 2b) than in permissive hosts (the Solanaceae), while no significant effects on variance were observed. Furthermore, we found that some mutations that were neutral in the natural host had reduced absolute fitness in alternative ones. Therefore, our data contain all three possible outcomes proposed by Martin and Lenormand [31], thus suggesting that their expectations were somewhat simplistic.
A compelling idea of the phylogenetic constraints for a virus jumping the host species barrier resides in the argument that the more closely related the primary host and the new host are, the greater are the chances for a successful spillover [34]. There are good mechanistic reasons that argue for it; if the ability to recognize and infect a host cell is important for cross-species transmission, then phylogenetically related species are more likely to share related cell receptors and defense pathways. However, others support the opposed view based on the observation that spillovers have occurred between hosts that can be either closely or distantly related, and no rule appears to predict the susceptibility of a new host [35]. Whether or not phylogenetic relatedness between reservoir and new hosts may be a factor for host switching, the rate and intensity of contact may be even more critical. Viral host switches between closely related species (e.g., species within the same genera) may also be limited by cross-immunity to related pathogens [2]; paraphrasing Holmes and Drummond [35] “although a species might be exposed to a novel pathogen, they might, through a combination of shared common ancestry and good fortune, already posses a sufficient immune response to prevent the infection from being established”. Our results shed some light into this debate: certainly the absolute fitness of a virus may be reduced when colonizing a new host, especially those distantly related ones, but the fraction of mutations that may be beneficial in this new host also increases with phylogenetic distance between the new host and the reservoir.
The existence of G×E interactions in determining fitness has been well established for many organisms, however, many of these studies used genotypes that differed in a large and unknown number of mutations [23], [36]–[39], making unclear whether G×E depended on single plasticity genes or on the quantitative contribution of multiple genes. Furthermore, in many examples, these studies used genotypes sampled from natural populations and thus have been filtered out by natural selection. Interestingly, our data demonstrate that single random nucleotide substitutions are sufficient to produce a significant G×E interaction. Mutations involved in significant G×E were scattered along the genome and they were randomly chosen irrespective of their fitness effects, provided they were viable in the primary host N. tabacum. Thus, we can conclude that phenotypic plasticity of TEV is not associated to the expression of any particular gene but results from the contribution of different genes. The concordance of these results with those previously reported by Remold and Lenski [26] for the bacterium E. coli and using knockout mutations suggests that the contribution of individual mutations to G×E is a general norm. In the context of emerging viral infections, the existence of a significant G×E interaction means that by knowing the absolute viral fitness in the natural host informs us little about what it may be in an alternative one, thus minimizing our ability to predict which genetic variants may be relevant for expanding TEV host-range.
Two non-mutually exclusive explanations can be brought forward to explain the existence of G×E: a change in the rank order of mutational effects across hosts (i.e., pleiotropy) and a change in the magnitude of the genetic variance but without changing the rank order. The evolutionary implications for these two mechanisms are different. Changes in genetic variance imply that the relative influence of selection and drift on the fate of mutations depends on the host. Exposure to hosts where the genetic variance in absolute fitness effects is low minimizes the efficiency by which selection operates either removing deleterious alleles or fixing beneficial ones and thus enhances the role of drift. By contrast, changes in rank order imply that selection favor different mutations in different hosts thus driving to a balanced polymorphism and specialization. We have assessed the extent to which these two possibilities may contribute to the observed G×E and found that both indeed coexist. Antagonistic pleiotropy does not contribute significantly to G×E when the novel host is closely related to the natural one, however, it becomes an important factor when hosts are distantly related (Figure 3). Similarly, genetic variance for absolute fitness was similar within Solanaceae hosts, but approximately one order of magnitude smaller for hosts outside the Solanaceae. Therefore, we conclude that the observed G×E interaction can be explained both by antagonistic pleiotropy and by changes in the genetic component of variance. Previous studies with E. coli showed that G×E was mainly explained by changes in genetic variance but not by changes in the rank order of fitness effects across environments [26]. However, other authors found that the contribution of new mutations to G×E for fitness traits in D. melanogaster was mostly via antagonistic pleiotropy [40].
The significant positive pleiotropy observed between absolute fitness in the natural host N. tabacum and in two closely related alternative ones (N. benthamiana and D. stramonium) suggests that mutations ameliorate aspects of the virus interaction with host factors that may be common to all three hosts but not to the other hosts. By contrast, the antagonistic pleiotropy observed between absolute fitness in N. tabacum and in the non-Solanaceae hosts suggests that TEV may be interacting with different host factors and that the improved interaction with tobacco may led to less efficient interaction with an orthologous factor, if available, in the alternative hosts. In this regard, many examples exist in the plant virology literature showing that host-range mutations have negative pleiotropic effects in the natural host (reviewed in [8], [41]). An illustrative example is the interaction between the VPg protein of other potyviruses and the host translation initiation factor eIF4E [42], [43]. Translation of the viral genomic RNA into the polyprotein depends upon the correct attachment between VPg and eIF4E. Mutations in eIF4E have been identified as the cause of the Potato virus Y (PVY) resistant phenotype of pepper cultivars. However, PVY overcomes the resistance by fixing amino acid changes in the central domain of VPg that reconstitutes the correct binding. These mutants pay a fitness cost in the non-resistant pepper.
Here we have shown for the first time how DMFE for an RNA virus vary across hosts. Our results suggest that the location of the DMFE moves towards smaller values as the phylogenetic distance to the natural host increases. In parallel, the distribution switches from negative to positive skewness, indicating that the probability of potential beneficial mutations increases along with host genetic distance. Similarly, we have found that the virus genotype and the host species interact in a non-linear manner to determine viral fitness. Both pleiotropic effects and reductions in genetic variance contribute to generate this genotype-by-host interaction. The implications of these observations for our understanding of emerging viral infections are multiple, but basically all hint on the unpredictability at the level of individual mutations: in the light of information collected on the primary host one can not anticipate which particular viral genotypes will be more likely to emerge. However, antagonistic pleiotropy still leaves some room for predictability at the level of classes of mutations: beneficial mutations, as a class, in the natural host may become deleterious in an alternative one, or vice versa.
For this study, a subset of 20 mutants non-lethal in N. tabacum (Table 1) was randomly chosen from a larger collection used in a previous study [20]. A plasmid containing the TEV genome, pMTEV [44], generously gifted by Dr. J.A. Daròs, was used to generate both the wildtype virus and the mutant genotypes. Single-nucleotide substitution mutants were generated by site-directed mutagenesis using QuikChange II XL Site-Directed Mutagenesis Kit (Stratagene) as described in [20] and following the manufacturer's instructions. The kit incorporates PfuUltra high fidelity DNA polymerase that minimizes the introduction of undesired mutations. The uniqueness of each mutation was confirmed by sequencing an 800 bp fragment encompassing the mutagenized nucleotide.
Infectious RNA of each genotype was obtained by in vitro transcription after BglII linearization of the corresponding plasmid as described in [45]. The infectivity of each RNA genotype was tested by inoculating five N. tabacum plants. All TEV genotypes were confirmed to be infectious on N. tabacum.
Eight host species previously described as susceptible to TEV systemic infection (VIDE database; pvo.bio-mirror.cn/refs.htm) were chosen for these experiments. Five hosts belong to the Solanaceae family: N. tabacum, N. benthamiana, D. stramonium, C. annuum, and S. lycopersicum. The first two belong to the same genus of the Nicotianoideae subfamily whereas the other three belong to the Solanoideae subfamily [27]. One host, H. annuus, pertains to the Asteraceae family. Both Solanaceae and Asteraceae are classified as Asterids [27]. The remaining two hosts, G. globosa and S. oleracea belong to the family Amaranthaceae. The three plant families are Eudicots [27].
All hosts were at similar growth stages when inoculated in order to minimize infectivity error due to possible variation in defense response to infection with developmental stage. All inoculations were done in a single experimental block. Nine plants per host per TEV genotype (9×8×21 = 1512) were inoculated by rubbing the first true leaf with 5 µL containing 5 µg RNA in vitro transcript of the virus and 10% carborundum (100 mg/mL). Solanaceae hosts show clear symptoms when infected and thus visual inspection was enough for determining infection. Nonetheless, some randomly chosen asymptomatic Solanaceae plants were subjected to RT-PCR for detection of infection as described in [46]. None was positive in this test. In the case of the non-Solanaceae hosts, symptoms were not recognizable and thus, infection was confirmed by RT-PCR.
Ten days post-inoculation (dpi), the whole infected plant, except the inoculated leaf, was collected. The whole tissue was frozen in liquid nitrogen and ground with mortar and pestle.
An aliquot of approximately 100 mg of grounded tissue was taken and mixed with 200 µL of extraction buffer (0.2 M Tris, 0.2 M NaCl, 50 mM EDTA, 2% SDS; pH 8). An equal volume of phenol∶chloroform∶isoamylic alcohol (25∶25∶1) was added, thoroughly vortexed and centrifuged at 14000 g for 5 min at 25°C. Ca. 160 µL of the upper aqueous phase were mixed with 80 µL of a solution containing 7.5 M LiCl and 50 mM EDTA and incubated overnight on ice at 4°C for RNA precipitation. The precipitated RNAs were centrifuged at 14000 g for 15 min at 4°C, washed once with 70% ice-cold ethanol, dried in a SpeedVac (Thermo) and resuspended in 30 µL of DEPC-treated ultrapure water. RNA concentration was measured spectrophotometrically and the samples were diluted to a final concentration of 50 ng/µL.
Within-plant virus accumulation was measured by absolute RT-qPCR using external standard [47]. Standard curves were constructed using five serial dilutions of TEV RNA produced by in vitro transcription and diluted in RNA obtained from the corresponding healthy host plant species. Samples were grouped by hosts and quantity of viral RNA was calculated using the corresponding standard curve.
RT-qPCR reactions were performed in 20 µL volume using One Step SYBR PrimeScript RT-PCR Kit II (TaKaRa) following the instructions provided by the manufacturer. The primers forward TEV-CP 5′-TTGGTCTTGATGGCAACGTG and reverse TEV-CP 5′-TGTGCCGTTCAGTGTCTTCCT amplify a 71 nt fragment within the TEV CP cistron. CP was chosen because it locates in the 3′ end of TEV genome and hence would only quantify complete genomes but not partial incomplete amplicons. Each RNA sample was quantified three times in independent experiments. Amplifications were done using the ABI PRISM Sequence Analyzer 7000 (Applied Biosystems). The thermal profile was as follows: RT phase consisted of 5 min. at 42°C followed by 10 s at 95°C; and PCR phase of 40 cycles of 5 s at 95°C and 31 s at 60°C. Quantification results were examined using SDS7000 software v. 1.2.3 (Applied Biosystems).
Absolute fitness was estimated as Malthusian growth rate per day, according to expression , where Q is the number of pg of TEV RNA per 100 ng of total plant RNA quantified at t = 10 dpi.
Unless otherwise indicated, all statistical tests were performed using SPSS version 19. Generalized linear models (GLM) were used to explore the effect of the different factors on TEV fitness. We assumed that m was distributed either as a Gaussian pdf or as a more stretched Gamma pdf. In both cases an identity link function was used. No qualitative differences were observed between the results obtained with these alternative distributions. Results reported will be those obtained using the Gaussian model.
|
10.1371/journal.pntd.0003704 | Fitness of Leishmania donovani Parasites Resistant to Drug Combinations | Drug resistance represents one of the main problems for the use of chemotherapy to treat leishmaniasis. Additionally, it could provide some advantages to Leishmania parasites, such as a higher capacity to survive in stress conditions. In this work, in mixed populations of Leishmania donovani parasites, we have analyzed whether experimentally resistant lines to one or two combined anti-leishmanial drugs better support the stress conditions than a susceptible line expressing luciferase (Luc line). In the absence of stress, none of the Leishmania lines showed growth advantage relative to the other when mixed at a 1:1 parasite ratio. However, when promastigotes from resistant lines and the Luc line were mixed and exposed to different stresses, we observed that the resistant lines are more tolerant of different stress conditions: nutrient starvation and heat shock-pH stress. Further to this, we observed that intracellular amastigotes from resistant lines present a higher capacity to survive inside the macrophages than those of the control line. These results suggest that resistant parasites acquire an overall fitness increase and that resistance to drug combinations presents significant differences in their fitness capacity versus single-drug resistant parasites, particularly in intracellular amastigotes. These results contribute to the assessment of the possible impact of drug resistance on leishmaniasis control programs.
| Chemotherapy is currently the only treatment option for leishmaniasis, a neglected tropical disease produced by the protozoan parasite Leishmania. However, first-line drugs have different types of limitations including toxicity, price, efficacy and mainly emerging resistance. The WHO has recently recommended a combined therapy in order to extend the life expectancy of these compounds. The emergence and spread of Leishmania antimonial-resistant parasites have led to a high rate of antimonial failure in India and have raised questions about the selection and propagation risk of drug resistant parasites. The spread of drug-resistant parasites in the field probably depends on their transmission potential, which is influenced by, among other factors, the relative fitness of drug-resistant versus drug-susceptible parasites. In light of this, we have designed experimental studies to determine whether Leishmania donovani parasites resistant to single and combinations of anti-leishmanial drugs present any advantages in their ability to bear the different stress conditions versus a susceptible L. donovani line. Our results suggest that resistant parasites acquire an overall fitness increase and that resistance to drug combinations presents significant differences in their fitness capacity, particularly in intracellular amastigotes.
| Leishmaniasis, a neglected tropical parasitic disease that is prevalent in 98 countries spread across three continents, is caused by protozoan parasites belonging to the genus Leishmania [1]; visceral leishmaniasis (VL), caused by species of the Leishmania donovani complex, is a lethal disease if left untreated. The recommended first-line therapies for VL include: i) pentavalent antimonials (meglumine antimoniate and sodium stibogluconate), except in some regions in the Indian subcontinent where there are significant areas with drug resistance [2]; ii) the polyene antibiotic amphotericin B or the liposomal amphotericin B formulation AmBisome; iii) the aminoglycoside paromomycin; and iv) the oral drug miltefosine. Although WHO [1, 3] recommended the use of either a single dose of AmBisome or combinations of anti-leishmanial drugs in order to reduce the duration and toxicity of treatment, to prolong the therapeutic life-span of existing drugs and to delay the emergence of resistance, recent experimental findings have demonstrated the ability of Leishmania to develop experimental resistance to different drug combinations [4]. The emergence and spread of Leishmania antimonial-resistant parasites have led to a high rate of antimonial failure in India [5] and have raised questions about the selection and propagation risk of drug resistant parasites [6, 7]. The spread of drug-resistant parasites in the field probably depends on their transmission potential, which is influenced by, among other factors, the relative fitness of drug-resistant versus drug-susceptible parasites.
Previous results have demonstrated that the acquisition of drug resistance could have an impact on parasite fitness, which could in turn influence other important biological properties involved in the regulation of proliferation and differentiation of parasites [6, 8]. As defined previously [6, 9], the fitness of Leishmania parasites can be measured by: i) capacity to survive, grow and generate infective metacyclic forms in the vector, ii) capacity to survive and grow in the mammalian host, and iii) capacity of transmission between the host and the vector.
Throughout its natural life cycle, Leishmania encounters adverse conditions that include: i) nutrient starvation and acidification of the medium, conditions that induce metacyclogenesis [10], ii) heat shock, when the parasite moves from growth at 28°C in the sandfly to 37°C inside the mammalian host macrophage, and iii) reactive oxygen species (ROS) and reactive nitrogen species (RNS), when it is phagocytized by macrophages of the host [11, 12]. Leishmania has evolved a broad spectrum of mechanisms to protect itself against these host defenses, including: i) enzymes that detoxify ROS and RNS [13], ii) use of thiols as antioxidant defenses [14], and iii) inhibition of the host's oxidative defense mechanisms [15].
However, it is not strictly true that all of the above increased the fitness capacity of all L. donovani resistant strains, as some strains showed little or no difference in their in vivo survival capacity compared to antimony-sensitive strains [16]. As suggested, other factors such as the genetic background of parasites could be important in the in vivo fitness capacity of L. donovani isolates that are clinically resistant to antimonials [16].
In this work, we evaluate whether L. donovani lines that are experimentally resistant to single anti-leishmanial drugs [amphotericin B (AmB), miltefosine (MIL), paromomycin (PMM) and trivalent antimony (SbIII)] and to drug combinations (AmB-MIL, AmB-PMM, AmB-SbIII, MIL-PMM and SbIII-PMM), present any advantages in their ability to bear the different stress conditions with respect to susceptible cells expressing the luciferase gene (Luc gene). For this purpose, we have studied the susceptibility of mixed promastigote populations under different stress conditions and their ability to infect and survive in mouse peritoneal macrophages. The results of this study using parasites that are experimentally resistant to single and multi-drug combinations are discussed in relation with their potential impact on future leishmaniasis control programs.
L. donovani promastigotes (MHOM/ET/67/HU3) and the previously described derivative resistant lines A, M, P, S, AM, AP, AS, MP, and SP (resistant to AmB, MIL, PMM, SbIII, AmB+MIL, AmB+PMM, AmB+SbIII, MIL+PMM and SbIII+PMM, respectively) [4] were grown at 28°C in an RPMI 1640-modified medium (Invitrogen) supplemented with 20% heat-inactivated fetal bovine serum (hiFBS, Invitrogen). L. donovani with the luciferase gene integrated into the parasite genome (Luc line) was grown in the same conditions.
Phothinus pyralis luciferase gene (luc) was amplified from vector pX63NEO-3Luc [17] by PCR using the primers LucNcoIF 5’-GACGCCCATGGATGGAAGACGCCAAAAACAT-3’ and LucNotIR 5’-GACGTAGCGGCCGCTTACAATTTGGACTTTCCGC-3’ including (in bold) NcoI and NotI restriction sites, respectively. The luc gene was then cloned into the NcoI-NotI sites of vector pLEXSY-hyg2 (Jena bioscience, Jena, Germany) which harbor a marker gene for selection with hygromycin-B (hyg gene). The vector generated was denominated pLEXSYHyg-Luc. In this construct, sequences of the 18S rRNA gene flanked the luc and hyg genes. Following linearization with SwaI, stationary promastigotes were transfected with 3 μg of linearized pLEXSYHyg-Luc plasmid to integrate the luc and hyg genes into the 18S rRNA (ssu) locus by homologous recombination, using a previously described protocol [18]. Twenty-four hours after transfection, the culture medium was supplemented with 25 μg/mL of hygromycin-B. Hygromycin-resistant parasites were usually selected after 7 days. After establishing the transgenic parasites, they were plated onto 1.5% agar plates containing culture medium plus 100 μg/mL hygromycin-B. After 10 days incubating at 28°C, clones were selected on agar plates and further propagated in liquid RPMI-modified medium supplemented with 100 μg/mL hygromycin-B. Integration of the expression cassette into the ssu locus was confirmed by PCR using genomic DNA from the wild-type (WT) and transgenic strains of L. donovani (Luc line) as a template. For this purpose, we used primer pairs ssu forward primer F3001 5’-GATCTGGTTGATTCTGCCAGTAG-3’ and 5’ utr (aprt) reverse primer A1715 5’-TATTCGTTGTCAGATGGCGCAC-3’, and primer pairs hyg forward primer A3804 5’-CCGATGGCTGTGTAGAAGTACTCG-3’ and ssu reverse primers 3002 5’-CTGCAGGTTCACCTACAGCTAC-3’.
Promastigotes and amastigotes isolated as described previously [19], were resuspended in HBS buffer (21 mM HEPES, 0.7 mM Na2HPO4, 137 mM NaCl, 5 mM KCl, and 6 mM D-glucose, pH 7.1) supplemented with 25 μM cell-permeable DMNPE-luciferin. After 15 minutes at room temperature, aliquots of this suspension (100 μL/well) were distributed into 96-well white polystyrene microplates. Luminescence was recorded with an Infinite F200 microplate reader (Tecan Austria GmbH, Austria). To measure the luciferase activity of intracellular amastigotes contained within infected cells, the Luciferase Assay System (Promega, Madison, Wis) was used according to the instructions of the manufacture. Luminescence was measured in the Infinite F200 microplate reader immediately after mixing.
Log-phase promastigotes from the control (WT) and A, M, P, S, AM, AP, AS, MP and SP resistant lines were mixed in a 1:1 ratio with log-phase promastigotes from the Luc line (1x106 mixed parasites/mL). Parasite density was microscopically determined every 24 h for a total of 144 h using Neubauer count chambers in order to monitor the growth. Parasite density in the WT line plus Luc line mixture was used as a control. Also, we evaluated the growth of each line in these mixed populations after 144 h of incubation, determining the luminescence as a measure of the cellular density of the Luc line. All growth experiments with the different parasite lines were performed in triplicate. In parallel, the mixed populations at a 1:1 ratio (4x106 mixed parasites/mL) were sub-cultured every 48 h (logarithmic phase) and luminescence measured at the end of the first and second sub-culture.
Promastigotes from the WT and resistant lines were mixed with the Luc line in a 1:1 ratio (4x106 mixed parasites/mL), and the proliferation of resistant parasites exposed to different stress conditions (late stationary growth phase, starvation and heat shock plus pH modification) was compared to the Luc line by measuring the luminescence intensity. The mixture of the WT and Luc lines was used as a control for all experiments.
For experiments studying intracellular amastigotes, mouse peritoneal macrophages were obtained as described previously [20] and plated at a density of 3 x 104 or 3 x 105 macrophages/well in 96-well white polystyrene microplates or 24-well tissue culture chamber slides, respectively, in an RPMI 1640 medium supplemented with 10% hiFBS, 2 mM glutamate, penicillin (100 U/mL) and streptomycin (100 μg/mL). Promastigotes from resistant or WT lines were mixed with the Luc line 1:1 ratio and maintained in culture for 6 days. Afterwards, the mixed populations of stationary phase cultures were used to infect macrophages at a macrophage/parasite ratio of 1:10. Six hours after infection at 35°C and 5% CO2, extracellular parasites were removed by washing with serum-free medium. Infected macrophages were maintained in culture medium at 37°C with 5% CO2 for 24 h and 96 h. To determine the infection index (% infection x amastigotes/macrophages), infected macrophages maintained in 24-well plates were fixed for 30 min at 4°C with 2.5% paraformaldehyde in PBS buffer (1.2 mM KH2PO4, 8.1 mM Na2HPO4, 130 mM NaCl, and 2.6 mM KCl, adjusted to pH 7), and permeabilized with 0.1% Triton X-100 in PBS for 30 min. Intracellular parasites and macrophages were detected by nuclear staining with ProLong Gold antifade reagent plus DAPI. To determine the intracellular proliferation profile of each line, infected macrophages maintained in 96-well plates were lysed and then the luminescence measured using the Luciferase Assay System (Promega).
Eight-week-old male BALB/c mice were purchased from Charles River Breeding Laboratories and maintained in our Animal Facility Service under pathogen-free conditions. They were fed a typical rodent diet and given drinking water ad libitum. These mice were used to collect primary peritoneal macrophages.
All experiments were performed according to National/EU guidelines regarding the care and use of laboratory animals in research. Approval for these studies was obtained from the Ethics Committee of the Spanish National Research Council (CSIC, file CEA-213-1-11).
Statistical comparisons between groups were performed using Student’s t-test. Differences were considered significant at a level of p<0.05.
The firefly luciferase (Luc) [21] has proved to be a useful reporter gene for monitoring gene expression [22] and quantifying Leishmania infections in macrophages and animal models, with the overall aim of probing host-microbe interactions [23, 24]. To assess the feasibility of using bioluminescence as a quantitative indicator of parasite proliferation, studies were performed to correlate bioluminescence with parasite number. For this purpose, the Luc gene was amplified by PCR and cloned into pLEXSY-hyg2. The LUC-expressing vector was electroporated into L. donovani parasites which were then selected in the presence of hygromycin-B. To test whether luciferase activity correlated well with parasite number, 4-fold serial dilutions were prepared and their luciferase activity measured. An excellent linear correlation was observed between the number of transgenic promastigotes and the luminescence intensity (S1 Fig).
Transfectant parasites that overexpress luciferase (from now on, Luc line) were also tested for their ability to infect macrophages. Stationary-phase recombinant promastigotes were used to infect mouse peritoneal macrophages. Intracellular Leishmania infection was observed microscopically after DAPI staining and no significant differences were noted in the infectivity of the Luc line versus the WT line (S2A Fig). Furthermore, an excellent correlation was observed between amastigote numbers and luminescence intensity (S2B Fig). Collectively, these results strongly suggest that the Luc line constitutes a valuable tool for assessing the viability and dynamics of mixed populations.
The promastigote number was evaluated every day for 6 days so that the growth features of each resistant line, or the WT line mixed in a 1:1 parasite ratio with the Luc line, could be studied and compared. We found that all mixed populations showed a similar growth profile as the control (WT+Luc) (Fig 1A). Moreover, the luminescence values were similar for each of these mixed populations after the 6th day of culture (Fig 1B). These results clearly indicate that the Luc line was present in the same ratio in all mixed populations and, therefore, there was no predominance of one line over another line under these conditions.
To evaluate whether there was predominance of any resistant lines over the susceptible Luc line, promastigotes from resistant lines and the Luc line were mixed in a 1:1 parasite ratio and grown without stress or exposure to different stresses. To assess the growth recovery, the luminescence intensity of mixed populations was determined in all cases after 48 h of culture in standard conditions. The WT plus Luc lines (WT+Luc) mixture was used as a control. The total number of mixed parasite cultures shows no significant differences between WT+Luc and the resistant lines+Luc in the different stress conditions, ranging indistinctly between 22-32x107 mixed parasites/mL.
Within the mammalian host, Leishmania promastigotes differentiate into amastigotes and multiply predominantly inside macrophages, where they are exposed to stress, including starvation, acidic pH, high temperatures (heat shock) and ROS and RNS production [11, 26]. To determine whether intracellular amastigotes from resistant lines were able to displace in vitro intracellular amastigotes from the susceptible Luc line, macrophages were infected with mixed populations of promastigotes taken from 6 day-old cultures where, as shown in Fig 1, no differences were observed in proliferation and the Luc line ratio. The infection index and luminescence of intracellular amastigotes were determined at 24 and 96 h post-infection to assess their infectivity and survival rates in mouse peritoneal macrophages. The infection indexes were similar in all cases, with values ranging for 24 h between 183±22 and 234±28, and for 96 h between 182±27 and 250±34. The results showed that in the early stage of macrophage infection (24 h) all the resistant lines, except the M line, had a significant predominance over the Luc line compared to the control (Fig 5A). The different lines showed a range of luminescence between 31 and 63% from SP+Luc and S+Luc populations, respectively, compared to the control (Fig 5A). These results could be due to: i) a higher percentage of metacyclic parasites, ii) tolerance to oxidative stress, and/or iii) tolerance to acidic pH and high temperatures of the resistant lines compared to the susceptibility of the Luc line. Also, in the late stage of infection (96 h), the resistant lines, again with the exception of the M line, were able to fully benefit from their initial advantage (Fig 5B). They showed a range of luminescence between 27 and 68% from SP+Luc and A+Luc populations, respectively, compared to the luminescence produced by the control (Fig 5B). Additionally, there were no significant differences on predominance of resistant lines between 24 and 96 h, with the exception of the S line (p<0.05). These results suggest that the predominance of some resistant lines over the Luc line is the result of a higher infection rate during the initial stage which reflects as a higher survival rate of intracellular amastigotes within macrophages.
Leishmania have successfully adapted to different environments for thousands of years and developed a highly flexible nature. Their survival capacity mainly relies on their ability to suppress oxidative outbursts of the host defense mechanism [15] and on a unique oxidant-protective redox metabolism, where thiols play a key role in antioxidant defenses [14]. In this regard, we have previously described that the resistant lines, except the M and S lines, had higher non-protein thiol levels than the WT line [4], which could contribute to a greater parasite survival rate within the host macrophages.
It has been demonstrated that some L. donovani strains resistant to antimonials have a more variable and markedly higher capacity of in vivo infection compared to antimony susceptible Leishmania strains [16]. Strains resistant to antimony also have a higher metacyclogenic capacity [27] and have specifically evolved extra mechanisms to manipulate their host cells in order to avoid antimony-induced stress [28]. Such adaptations would not only improve the parasites survival capacity when stressed by antimony, but would also favor their survival in drug-free conditions.
Since SbIII, AmB and MIL kill Leishmania through a common cell death pathway to achieve apoptosis, strains resistant to one or more of these drugs could develop tolerance to apoptosis, which would grant them a higher survival rate in macrophages, as we have observed with our Leishmania resistant lines (Fig 5B).
In conclusion, the experiments using our transgenic Leishmania luc line have clearly demonstrated and validated the fact that Leishmania lines experimentally resistant to individual and combinatorial anti-leishmanial drugs have an increased fitness compared to Leishmania susceptible lines, probably as a consequence of their metabolic adaptations which all converge on the higher tolerance to stress conditions, as recently described [25]. Subsequently, they also have a better chance of survival. However, although this approach using promastigotes for assessing the viability and dynamics of mixed populations have important advantages, their use on intracellular amastigotes has some methodological limitations. Therefore, the emergence and spread of drug-resistant parasites in the field will probably result in a greater competitive fitness cost with respect to susceptible parasites, plus negative effects on the chemotherapy strategies used to control leishmaniasis.
|
10.1371/journal.pgen.1006685 | Whole-genome analysis of papillary kidney cancer finds significant noncoding alterations | To date, studies on papillary renal-cell carcinoma (pRCC) have largely focused on coding alterations in traditional drivers, particularly the tyrosine-kinase, Met. However, for a significant fraction of tumors, researchers have been unable to determine a clear molecular etiology. To address this, we perform the first whole-genome analysis of pRCC. Elaborating on previous results on MET, we find a germline SNP (rs11762213) in this gene predicting prognosis. Surprisingly, we detect no enrichment for small structural variants disrupting MET. Next, we scrutinize noncoding mutations, discovering potentially impactful ones associated with MET. Many of these are in an intron connected to a known, oncogenic alternative-splicing event; moreover, we find methylation dysregulation nearby, leading to a cryptic promoter activation. We also notice an elevation of mutations in the long noncoding RNA NEAT1, and these mutations are associated with increased expression and unfavorable outcome. Finally, to address the origin of pRCC heterogeneity, we carry out whole-genome analyses of mutational processes. First, we investigate genome-wide mutational patterns, finding they are governed mostly by methylation-associated C-to-T transitions. We also observe significantly more mutations in open chromatin and early-replicating regions in tumors with chromatin-modifier alterations. Finally, we reconstruct cancer-evolutionary trees, which have markedly different topologies and suggested evolutionary trajectories for the different subtypes of pRCC.
| Renal cell carcinoma accounts for more than 90% of kidney cancers. Papillary renal cell carcinoma (pRCC) is the second most common subtype of renal cell carcinoma. Previous studies, focusing mostly on the protein-coding regions, have identified several key genomic alterations that are critical to cancer initiation and development. However, researchers cannot find any key mutation in a significant portion of pRCC. Therefore, we carry out the first whole-genome study of pRCC to discover triggering DNA changes explaining these cases. By looking at the entire genome, we find additional potentially impactful alterations both in and out of the protein-coding regions. These newly identified critical mutations from scrutinizing the entire genome help complete our understanding of pRCC genomes. Two alterations we find are associated with prognosis, which could aid clinical decisions. We are also able to unveil mutation patterns, signatures and tumor evolutionary structures, which reflect the mutagenesis processes and help understand how heterogeneity arises. Our study provides valuable additional information to facilitate better tumor subtyping, risk stratification and potentially clinical management.
| Renal cell carcinoma (RCC) makes up over 90% of kidney cancers and currently is the most lethal genitourinary malignancy [1]. Papillary RCC (pRCC) accounts for 10%-15% of the total RCC cases [2]. Unfortunately, pRCC has been understudied and there is no current form of effective systemic therapy for this disease. pRCC is further subtyped into two major subtypes: type I and type II based on histopathological features. For many years, the only prominent oncogene in pRCC (specifically, type I) that physicians were able to identify was MET, a tyrosine-kinase receptor for hepatic growth factor. An amino acid substitution that leads to constitutive activation and/or overexpression are two mechanisms of dysfunction of MET in tumorigenesis. Recently, the Cancer Genome Atlas (TCGA) published its first result on pRCC [3], which greatly improves our understanding of the genomic basis of this disease. Several more genes and pathways were identified to be significantly mutated in pRCC. Nevertheless, a significant portion of pRCC cases remain without any known driver. Therefore, we think it is a good time to explore the noncoding regions of the genome using whole-genome sequencing (WGS). Noncoding regions, often overlooked in cancer studies, have been shown to be actively involved in tumorigenesis [4–6]. Mutations in noncoding regions may cause disruptive changes in both cis- and trans-regulatory elements, affecting gene expression. Understanding noncoding mutations helps fill the missing “dark matter” in cancer research. Meanwhile, it is an open question to the degree which these regions harbor significant driver alterations and here we can address this question in the specific context of pRCC.
Multiple endogenous and environmental mutation processes shape the somatic mutational landscape observed in cancers [7]. Analyses of the genomic alterations associated with these processes give information on cancer development and heterogeneity, shed light on the mutational disparity between cancer subtypes and even indicate potential new treatment strategies [8]. Additionally, genomic features such as replication time and chromatin environment govern mutation rate along the genome, contributing to spatial mutational heterogeneity. Last, while studying mutation patterns, landscape and tumor evolution is possible using data from whole exome sequencing (WXS), whole genome sequencing (WGS) gives richer information on mutation landscape and minimizes the potential confounding effects of exome capture process and driver selection.
In this study, we comprehensively analyzed 35 pRCC cases that were whole genome sequenced along with an extensive set of WXS data on multiple levels. We went from microscopic examination of driver genes to analyses of whole genome sequencing variants, and finally, to an investigation of high-order mutational features. We focused on two aims: exploring potential noncoding drivers and better understanding the heterogeneity of the cancer. First, we focused on MET, an oncogene that plays a central role in pRCC, especially in the type I. We found rs11762213, a germline exonic single nucleotide polymorphism (SNP) inside MET, predicted cancer-specific survival (CSS) in type II pRCC. We also discovered several potentially impactful noncoding mutations in MET promoter and its first two introns. The previous TCGA study identified a MET alternate transcript as a driver but without illustrating the etiology [3]. We found that a cryptic promoter from an endogenous long interspersed nuclear element-1 (L1) triggered the alternate isoform expression. Surprisingly, we did not find a significant amount of small structural variations affecting MET. Then we went on to cases not as easily explained as those with MET alterations. We analyzed about 160,000 noncoding mutations throughout the whole genome and found several potentially high-impact mutations in the noncoding regions. Further zooming out, we discovered pRCC exhibited mutational heterogeneity in both nucleotide context and genome location, indicating underlying interplay of mutational processes. We found methylation was the leading factor influencing mutation landscape. Methylation status drove the inter-sample mutation variation by promoting more C-to-T mutations in the CpG context. APOBEC activity, although infrequently observed, left an unequivocal mutation signature in a pRCC genome but not in ccRCC. Also, we discovered samples with chromatin remodeler alternations accumulated more mutations in open chromatin and early-replicating regions. Lastly, we derived evolutionary trees based on the whole-genome mutation calls for each individual sample. The tree topologies varied, reflecting tumor heterogeneity and correlating with the known tumor subtypes.
We began with coding variants in the long known driver MET. The TCGA study of 161 pRCC patients found 15 samples carrying somatic, nonsynonymous single nucleotide variants (SNVs) in MET. By analyzing 117 extra WXS samples (see Methods), we found six more nonsynonymous somatic mutations in six samples (S1 Table). V1110I and M1268T were two recurrent mutations in this extra set. Both of them were observed in the TCGA study as well. Additionally, we found two samples carrying H112Y and Y1248C respectively. H1112Y has been observed in two patients in the original TCGA study cohort and H1118R is a long-known germline mutation associated with hereditary papillary renal carcinoma (HPRC) [9]. Y1248C has been observed in type I pRCC before and the TCGA cohort has a case carrying Y1248H. All mutations occurred in the hypermutated tyrosine kinase catalytic domain of MET. Two out of these six samples were identified as type I pRCC while the subtypes of the rest four were unknown.
Although many MET somatic mutations are believed to play a central role in pRCC, some germline MET mutations have also been associated with the disease. In particular, a germline SNP, rs11762213 (Fig 1A), has been discovered to predict recurrence and survival in a mixed RCC cohort [10]. ccRCC predominated the initial discovery RCC cohort. This conclusion was later validated in a ccRCC cohort but never in pRCC [11]. It was not clear whether this SNP has a prognostic effect in pRCC. Using an extensive WXS set of 277 patients (see Methods; S1 Fig and S1 Table;), we found 14 patients carrying one risk allele of rs11762213 (G/A, minor allele frequency (MAF) = 2.53%). No homozygous A/A was observed. Cancer specific deceases were concentrated in type II pRCC. Among 96 type II pRCC cases, seven patients carried the minor A allele (MAF = 3.65%, Table 1). Cancer-speccific survival was significantly worse in type II patients carrying the risk allele of rs11762213 (p = 0.034, Fig 1B). But we did not find a significant association of this germline SNP with survival in type I patients. We did not observe statistically significant correlation of rs11762213 with MET RNA expression in either tumor samples or normal controls (p > 0.1, two-sided rank-sum test). c-Met pY1235 levels in tumor samples, as measured by Reverse phase protein array (RPPA), were not significantly different between carriers of these two genotypes (p > 0.1, two-sided rank-sum test).
The TCGA study identified a MET alternate transcript as driver [3]. However, the etiology of this new isoform was unknown. Here we found this alternate transcript resulted from the activation of a cryptic promoter from an endogenous L1 element (Fig 1A), likely due to a local loss of methylation [12]. This event was reported in several other cancers [13, 14]. To test its relationship with methylation, we found the closest probe (cg06985664, ~3kb downstream) on the methylation array showed marginally statistically significantly lower methylation level in samples expressing the alternate transcript (p = 0.055, one-sided rank-sum test). Additionally, this event was associated with methylation cluster 1 (odds ration (OR) = 4.54, p<0.041), indicating genome-wide methylation dysfunction. This association was stronger in type II pRCC and the alternate transcript was tightly associated with the C2b cluster (OR = 17.5, p<0.007).
Despite the fact, MET is the most common driver alteration, about 20% presumably MET-driven yet MET wild-type pRCC samples were still left unexplained [3]. That is, they had a characteristic Met-dysregulated gene expression pattern but no obvious Met-associated alteration. Therefore, we scanned the MET noncoding regions. We observed one mutation in MET promoter region in a type I pRCC sample (Fig 1A and S2 Table). This sample showed no evidence of nonsynonymous mutation in MET gene but had copy number gain of MET. Additionally, we observed 6/35 (17.1%) samples carry mutations in the intronic regions between exons 1–3 of MET (Fig 1A and S2 Table). As we describe above we could see that these mutations nearby to regions with methylation dysregulation and the activation of a cryptic promoter. However, we were not able to find a direct statistically significant correlation between the alternative splicing event and these intronic mutations.
We further expanded our scope and ran FunSeq [4, 5] to identify potentially high-impact, noncoding variants in pRCC. First, we identified a high-impact mutation hotspot on chromosome 1. 6/35 (17.1%) samples had mutations within this 6.5kb region (Fig 2A and S2 Table). This hotspot located at the 5’-end of ERRFI1 (ERBB Receptor Feedback Inhibitor 1) and overlapped with the predicted regulatory region. ERRFI1 is a negative regulator of EGFR family members, including EGFR, HER2 and HER3; all have been implicated in cancer. However, due to a limited sample size here, our test power was inevitably low. We did not observe statistically significant changes among mutated samples in mRNA expression level, protein level and phosphorylation level of EGFR, HER2 and HER3.
Another potentially impactful mutation hotspot was in NEAT1. We saw mutations inside this nuclear long noncoding RNA in 6/35(17.1%) samples (Fig 2B and S2 Table). Several studies indicated NEAT1 is associated with various cancers [15, 16]. It promotes cell proliferation in hypoxia [17] and alters the epigenetic landscape, increasing transcription of target genes [18].
Mutations we found all fell into a putative promoter and its flanking region of NEAT1. We noticed NEAT1 mutations were associated with higher NEAT1 expression (Fig 2C, S2A Fig, p < 0.032, two-sided rank sum test). We also found NEAT1 mutations were associated with worse prognosis (Fig 2D, p < 0.041, log-rank test). To further investigate the role of NEAT1 in RCCs, we found NEAT1 overexpression is significantly associated with shorter overall survival in ccRCC (TCGA cohort, p = 0.0132, S2B Fig). Moreover, MALAT1, another noticeable lncRNA in cancer, is tightly co-expressed with NEAT1 in both pRCC and ccRCC (Spearman’s correlation; 0.79 and 0.87 respectively). MALAT1 is located ~50kb downstream of NEAT1 and might share the same regulatory mechanism with it. The Catalogue of Somatic Mutations in Cancer (COSMIC) [19] annotates MALAT1 as a cancer consensus gene, associating it with pediatric RCC and lung cancer. MALAT1 was also reported to be associated with ccRCC [20].
We performed structural variants (SVs) discovery using WGS reads (see Methods and S3 Table, S2C Fig). This SV discovery approach has higher sensitivity and resolution than array-based copy number variation methods, which were employed in the TCGA analysis. This was a large-scale, big-compute calculation that involves mapping more than 100 billion reads (see Methods). In the end, we found 424 somatic SV events, including 170 deletions, 53 duplications, 105 inversions and 96 translocations (S3 Table). The samples clearly split into two categories based on the number of SV events (ranging from 0 to 88): genome unstable (6 samples, >40 events/per samples) and genome stable (29 samples, <10 events/per sample). The unstable category was significantly associated with type II versus type I pRCC (p<0.015, two-tailed Fisher exact test) and enriched in the C2b cluster (p < 0.002, two-tailed Fisher exact test).
We overlapped SVs with curated cancer genes from COSMIC [19]. Somewhat surprisingly, we did not find SVs affecting MET except a single example—one genomically highly unstable sample, TCGA-B9-4116, with deep amplification of MET, showed multiple SVs of various classes hitting MET. To explain this lack of enrichment for small SVs in MET, we postulated trisomy/polysomy 7 is the main mechanism of MET structural alteration rather than small-scale duplication. Moreover, besides duplication, we did not expect to find deletion, inversion or translocation disrupting oncogene MET. These SVs were likely to cause loss-of-function rather than gain-of-function. Indeed, we did not find any breakpoint splitting MET. This was consistent with the putative role of MET as an oncogene, rather than a tumor suppressor.
We next looked for other cancer genes affected by somatic SVs. We found two cases with deletions in SDHB. The median SDHB expression was significantly lower (p<0.0034, one-sided rank sum test), only ~50% compared to cases without alternation (S2D Fig). We validated the deletions affecting SDHB with another SV caller, Lumpy-SV. Besides, we confirmed three cases carrying deletions affecting CDKN2A called by the TCGA array-based method but not the other two cases. Notably, three confirmed cases had significantly lower CDKN2A expression (p<0.0013, one-sided rank sum test) but the unconfirmed two cases did not (S2D Fig). This suggests SV calling from WGS is accurate and predicts CDKN2A expression better. Lastly, we observed several high-impact sporadic events, including duplications in EGFR and HIF1A, and deletions in DNMT3A and STAG2 (S2C Fig).
To further get an overview of the mutation landscape, we summarized the mutation spectra of 35 whole genome sequenced pRCC samples (Fig 3A). C-to-T in CpGs showed the highest mutation rates, which were roughly three to six-fold higher than mutation rates of other nucleotide contexts.
We used principle components analysis (PCA) to reveal factors that explained the most inter-sample variation. The loadings on the first principle component (which explained 12.5% of the variation) demonstrated C-to-T in CpGs contributed the most to inter-sample variation (Fig 3B). C-to-T in CpGs is highly associated with methylation. It reflects the spontaneous deamination of cytosines in CpGs, which is much more frequent in 5-methyl-cytosines [21]. So we further explored the association between C-to-T in CpGs and tumor methylation status. First, we validated the TCGA identified methylation cluster 1 showed higher methylation level than cluster 2 in all annotated regions (S3 Fig, see Methods), prominently in CpG Islands (Odds ratio of sites being differentially hypermethylated: 1.29, 95%CI: 1.20–1.39, p<0.0001). We confirmed this association by showing samples from methylation cluster 1 had higher PC1 scores as well as higher C-to-T mutation counts and mutation percentages in CpGs (Fig 3C). This trend was further validated using a larger WXS dataset as well. Especially, the most hypermethylated group, CpG island methylation phenotype (CIMP), showed the greatest C-to-T rate in CpGs (S3C Fig). Therefore, methylation status was the most prominent factor shaping the mutation spectra across patients.
Furthermore, we explored the functional impacts of the excessive mutations driven by methylation. C-to-T mutations in CpGs we observed in pRCCs were more likely to be in the coding region (OR = 1.54, 95%CI: 1.27–1.85, p<0.0001) and nonsynonymous (OR = 1.47, 95%CI: 1.17–1.84, p<0.001), which indicated they tended to be high-impact mutations. However, C-to-T mutations in CpGs did not show functional bias between the two methylation clusters in noncoding regions (based on FunSeq score distribution).
Recently, 30 somatic mutation signatures were identified; many have putative etiology, revealing the underlying mutational processes and helping understand tumor development [7]. We used a LASSO-based approach (see Methods) to decompose the observed mutations into a linear combination of these canonical mutation signatures in both WGS and WXS samples (S4 Fig). The leading signature was "signature 5" (from reference 7). Interestingly, we found one type II pRCC case out of 155 somatic WXS sequenced samples exhibited APOBEC-associated mutation signatures 2 and 13. APOBEC mutation pattern enrichment analysis (see Method) further confirmed the presence of APOBEC activity (Fig 3D, S4 Table). This sample was statistically enriched of APOBEC-induced mutations (adjusted p-value < 0.0003).
Prominent APOBEC activities were also incidentally detected in three upper track urothelial cancer (UC) samples sequenced and processed in the same pipeline with pRCC samples. UC often carries APOBEC associated mutation signatures and our result is consistent with the TCGA bladder urothelial cancer study [22].
The APOBEC associated signature carrying pRCC case was centrally reviewed by six pathologists in the original study and confirmed to be type II pRCC [3]. Thus, this tumor is likely a special case of type II with genomic alterations sharing some similarities with UC. It had non-silent mutations in ARID1A and MLL2 and a synonymous mutation in RXRA, all are identified as significantly mutated genes in UC but not in pRCC. Potential type II pRCC driver events, for example, low expression of CDKN2A and nonsynonymous alternations in significantly mutated genes of pRCC, were absent in this sample. Noticeably, the four samples with APOBEC activities showed significantly higher APOBEC3A and APOBEC3B mRNA expression level (p < 0.0022 and p < 0.0039 respectively, one-sided rank sum test, S5 Fig). This is in concordance with previous studies of APOBEC mutagenesis in various types of cancer [23].
Consistent with previous studies [12], we failed to detect statistically significant APOBEC activities in an extensive WXS dataset of 418 clear cell RCC (ccRCC) samples, even after subsampling to avoid p-value adjustment eroding the power. Very low levels of APOBEC signatures (<15%) were found in less than 1%(4/418) samples. With a much larger sample size, this result was unlikely to be confounded by detecting power.
Chromatin remodeling genes are frequently mutated in pRCC and many other cancers, including ccRCC [3, 24, 25]. Defects in chromatin remodeling cause dysregulation of the chromatin environment. Open chromatin regions usually show a lower mutation rate, presumably due to more effective DNA repair [26]. Thus, chromatin remodeler alterations could possibly alter the mutation landscape, specifically increasing mutation rate in previously open chromatin regions. To test this, we tallied the number of mutations inside DNase I hypersensitive sites (DHS) inferred from 11 normal fetal kidney cortex samples (The NIH Roadmap Epigenomics Mapping Consortium) [27], which represent normal tissues under physiological conditions. 9/35 samples with disruptive mutations in ten chromatin remodeling genes, cancer-associated genes showed higher genome-wide mutation counts (p < 0.021, one-sided rank-sum test;), partially driven by higher mutation counts in the DHS regions (p < 0.0023, one-sided rank-sum test). The median number of mutations in DHS regions considerably increased by 60% (67.5 versus 108) in samples carrying chromatin remodeling defects. The effect was still significant after normalizing against the total mutation counts (p < 0.019, one-sided rank-sum test, Fig 3E), indicating a significant shift in mutation landscape.
Replication time is known to correlate greatly with mutation rate. Early replicating regions have lower mutation rate compared to late replicating ones. Researchers reason replication errors are more likely to be corrected by DNA repair system in early replicating regions. With defects in chromatin remodeling genes, we observed this trend became less pronounced (p<0.031, one-sided rank-sum test, S6 Fig). This is presumably because dysregulation of the chromatin environment hinders replication error repair by changing the accessibility of newly synthesized DNA chains.
With the richness of SNVs in WGS samples, we can further tackle the mutational process heterogeneity of pRCC by constructing evolutionary trees for the 35 tumors (S7 Fig). These trees were derived from the whole-genome mutation calls and were produced individually for each tumor, with their topology suggesting a temporal ordering to the mutations. We could classify the trees into four groups based on their topology (Fig 4):
In addition, three trees were excluded from the analysis since they had a largest population faction <0.5, which was likely due to low mutation number, high sequence error and/or particularly high copy number variation.
Both topology groups 3 and 4 showed significant clonal evolution, with more distal subclones, and greater heterogeneity, indicated by substantial mutational divergence between populations. These groups were significantly depleted in type I pRCC (p < 0.0034, two-tailed fisher exact test). In contrast, the short branch group (#2) was significantly enriched in type I pRCC (p<0.011, two-tailed fisher exact test, Fig 4B). This suggested type I tumors were more homogenous and showed less complex evolutionary features compared to type II and unclassified samples.
Our study is the first one that comprehensively looked into the noncoding regions of pRCC. Doing so allowed us to tackle an open question in the field of cancer genomics, whether whole genome sequencing adds additional value over whole exome sequencing. We comprehensively analyzed both WGS and an extensive set of WXS of pRCC, scrutinizing local high-impact events as well as giving an overall view of the mutation landscape and evolution. Our work further completed the genomic alteration landscape of pRCC (Fig 4B). Beyond traditionally driver events, we suggested several novel noncoding alterations potentially drive tumorigenesis. We also provided valuable insights to tumor heterogeneity though investigating the mutational patterns, landscape, and evolutionary profiles.
First, we elaborated on previous results of the long known driver MET. In an extended 117 WXS dataset, we found six additional nonsynonymous somatic mutations in the hyper-mutated tyrosine kinase catalytic domain. These somatic mutations were highly recurrent, concentrated on a few critical amino acids. This was in line with MET being an oncogene and supported its central driver status in pRCC. Then we found an exonic SNP in MET, rs11762213, to be a prognostic germline variance in type II pRCC. Previously, rs11762213 was found to predict outcome in a mixed RCC samples, predominated by ccRCC [10]. Later, the result was confirmed in a large TCGA ccRCC cohort [11]. However, it was never clear whether rs11762213 only predicts the outcome in ccRCC or other histological types as well. In this study, we concluded that the minor alternative allele of rs11762213 also forecasts unfavorable outcome in type II pRCC patients. The mechanism of this exonic germline SNP remains unsettled. A previous study proposed it disrupts a putative enhancer of MET [11]. However, researchers could not find significant association between the SNP and MET expression in either tumor or normal tissues. We noticed there is no other gene within 100 kb in both directions of this SNP. Given the significant role of MET in pRCC, we think rs11762213 is affecting survival through MET, although the mechanism unknown.
Similar to ccRCC, type II pRCC is not primarily driven by MET. Not as significantly mutated in ccRCC and type II pRCC, MET nonetheless seems to play a role in cancer development. Our finding on rs11762213 is potentially meaningful in the clinical management of patients with the more aggressive type II pRCC. rs11762213 genotyping could become a reliable, low-cost risk stratification tool for these patients. Also, rs11762213 might become a biomarker for predicting response to Met inhibitors pending further studies.
Interestingly, rs11762213 is prevalent mostly in European and American populations but not in African populations and rare in Asian populations. However, the MAF of rs11762213 among African American patients in our cohort is 2.73%, higher than MAFs in general for African populations observed in 1000 Genome phase 3 dataset (0.2%, with 0% in Americans with African ancestry, ASW) [28] and the ExAC dataset (1.1%, excluding TCGA cohorts) [29]. This implies a possible effect of rs11762213 on pRCC incidence among African Americans that is worth further investigation.
In MET noncoding regions, we first found a cryptic promoter from a retrotransposon in the second intron initiates the alternate transcript, which was classified as a driver by the TCGA study (3). Methylation is a major source of silencing retrotransposon activities in the human genome [12–14]. Indeed, we observed evidence for a local loss of methylation and global methylation dysregulation in samples expressing the alternate transcript. Our finding indicates methylation change might directly drive pRCC growth through MET.
We also discovered mutations associated with the MET promoter and first two introns, where the alternate transcript starts. Although the implication is unknown, our analysis suggests there is a mutation hotspot in MET that calls for further research.
Expanding our scope from coding to non-coding and using FunSeq to group SNVs by functional elements, we found several potentially significant noncoding mutation hotspots relevant to tumorigenesis throughout the entire genome. A mutation hotspot was found downstream of ERRFI1, an important regulator of the EGFR pathway, which may serve as a potential tumor suppressor. EGFR inhibitors have been used in papillary kidney cancer with an 11% response rate observed [30]. These mutations potentially disrupt regulatory elements of ERRFI1 and thus play a role in tumorigenesis. However, likely limited by a small sample size, we were not able to detect statistically significant functional changes in ERRFI1 and related pathways. Another noncoding hotpot was in NEAT1, a long noncoding RNA that has been speculated to involve in cancer. Patients carrying mutations in NEAT1 had significantly higher NEAT1 expression and worse prognosis. High expression of NEAT1 predicted significantly worse survival in ccRCC as well. NEAT1 has been shown to be hypermutated in other cancers and some studies also linked high NEAT1 association with unfavorable prognosis [31, 32]. Lastly, a downstream lncRNA, MALAT1, showed tight co-expression pattern with NEAT1 in both pRCC and ccRCC. MALAT1 is on COSMIC consensus cancer gene list and annotated as related with pediatric RCC [19]. It was also reported to be associated with ccRCC [20].
Next, with more than 100 billion carefully remapped reads from WGS, we generated a high-confident SV dataset for 35 pRCC samples. Our method has great accuracy. In fact, we confirmed the well-known deletion of CDKN2A and found that we predicted its down-regulate expression better than the copy number variation analysis in TCGA study [3].
In terms of overall numbers of SVs, we found the pRCCs clearly split into two categories: the stable category had less than 10 events per sample while the unstable category had all above 40. Moreover, the unstable category was tightly associated with the C2b cluster, which has inferior outcomes [3]. Our SV study also discovered recurrent cases of SDHB deletion and expression data supported our finding. SDHB is a subunit of succinate dehydrogenase. Previous studies indicated the loss of SDHB being a driver event by disturbing tumor metabolic environment [33, 34] Besides SDHB, we also found some other sporadic events involving known tumor drivers.
Somewhat counter-intuitively, we found the absence of MET alterations that involve small deletion or breakage of the MET gene except in one highly unstable sample. Large-scale duplications involving MET, however, have been found (e.g. trisomy 7). This finding can be rationalized by realizing that the oncogenic activity of MET is encouraged by amplification but not by deletion or disruption. Moreover, we postulated that polysomy 7 might be the major mechanism of MET gain and lack of smaller SVs and breakpoints disrupting MET further supports its oncogene role.
WGS provides many times more SNVs compared to WXS. Thus it gives us an opportunity to look into the high-level landscape of mutations in pRCC. Several recent landmark pan-cancer studies lead to the wide recognition of significance and great research interests in cancer mutational processes [7, 8, 26, 35, 36]. DNA mutation is one of the driving forces of cancer development, and understanding the underlying processes and affecting factors that generate the mutations is vital in cancer studies. In particular, we focused on revealing the underlying sources that fuel tumor heterogeneity, a key feature of pRCC.
We identified mutation rate dispersion of C-to-T transitions in CpGs motifs contributed the most to the inter-sample mutation spectra variation. We further pinned down the cause of dispersion by showing the hypermethylated cluster, identified in the previous TCGA study [3], had a higher C-to-T rate in CpGs. Although increased C-to-T in CpGs is likely the result of hypermethylation, we cannot rule out the possibility the change of mutation landscape plays a role in cancer development. For example, C-to-T in methylated CpGs causes loss of methylation, which could have effects on local chromatin environment, trans-elements recruitment and gene expression regulation. In our study, we observed C-to-Ts in CpGs were enriched in coding regions, which suggested they might have a higher functional impact in the cancer genome.
Significant APOBEC activities and consequential mutation signatures were observed in one type II pRCC case. APOBEC activities were known to be prevalent in UCs [22, 23]. We also successfully detected prominent APOBEC signatures in all three UC samples processed in the same pipeline as pRCCs. Intriguingly, despite being considered to have the same cellular origin with pRCC, we were not able to detect meaningful APOBEC activities in ccRCC. This was in agreement with previous studies [12]. APOBEC mutation signature was also found in a small percentage of chromophobe renal cell carcinoma [37], although they are believed to have a different cellular origin. APOBEC activities have been linked with genetic predisposition and viral infection [38]. Given a statistically robust signal in our conservative algorithm, it is plausible that a small fraction of type II pRCCs might share some etiologically and gnomically similarities with UC. Standard treatment for UC differs significantly from the one for pRCC. Pending further research, this finding might suggest actionable clinical implications.
The chromatin remodeling pathway is highly mutated in pRCC [3, 24, 25]. Several chromatin remodelers have been identified as cancer drivers in pRCC. We investigated the relationship between samples with mutated chromatin remodelers and those without such mutations in terms of mutation landscape. We demonstrated pRCCs with defects in chromatin remodeling genes showed higher mutation rate in general, driven by an even stronger mutation rate increase in putative open chromatin regions in normal kidney tissues. This is likely because chromatin remodeling defects disrupt normal open chromatin environment and impede DNA repairing in these regions.
It is known that replication time strongly governs local mutation rate. Early replication regions have fewer mutations. But the difference dissipates when DNA mismatch repair becomes defective [21]. In our study, we found this correlation weakened in samples with mutated chromatin remodeling genes, presumably caused by failure of replication error repair in an abnormal chromatin environment. Through defects in chromatin remodeling genes, a tumor alters its mutation rate and landscape, which might provide it advantage in cancer evolution. Yet, high mutation burden in functional important open chromatin regions also raises the chance that tumor antigens activate the host immune system. Researchers found tumors with DNA mismatch repair deficiency responded better to PD-1 blockage [39]. These tumors also accumulate more mutations in early replicating regions [26]. Thus chromatin remodeler alterations might as well correlate with higher response rate of immunotherapy, which is worth further studies.
Finally, we constructed individual evolutionary trees for all 35 samples. This is the first study inferring tumor evolutionary trees using a large number of SNVs from WGS in pRCC. Benefited from a large number of SNVs, the tree construction became more statistically robust and revealed more details. In general, evolutionary trees gave us the opportunity to observe how pRCC heterogeneity developed over time. They revealed the history of the tumor and how mutations accumulated. We discovered the trees exhibited four major types of topologies, reflecting different levels of heterogeneity. Type II pRCCs showed distinct evolutionary topologies from type I, perhaps indicating an association with greater heterogeneity and different evolving trajectories.
In this first whole genome study of pRCC, we found several novel noncoding alterations that might drive tumor development and we explored the mutational landscape and evolutionary trees to better understand tumor heterogeneity. However, due to a limited sample size, some of our statistical tests were underpowered. As the cost of sequencing keeps dropping and technology for data management and processing continues advancing, we expect to have more whole genome sequenced tumors in the near future [40]. With a larger cohort, we hope to gain enough power to test the hypotheses we formed as well as further explore the noncoding regions of pRCC.
pRCC and ccRCC WXS and pRCC WGS variants calls were downloaded from the TCGA Data Portal (https://gdc-portal.nci.nih.gov/legacy-archive/search/f) and TCGA Jamboree (https://tcga-data-secure.nci.nih.gov/tcgafiles/tcgajamboree) respectively. pRCC RNAseq, RPPA and methylation data (under project ID: TCGA-KIRP) were downloaded from TCGA Data Portal as well. Wavelet-smoothed repli-seq data was obtained as a part of ENCODE project [41–43] and downloaded from UCSC Genome Browser (Also accessible under GSE34399 in the Gene Expression Omnibus). DHS data (fetal, kidney cortex) were obtained from Roadmap Epigenomics Project and are accessible from http://www.genboree.org/EdaccData/Current-Release/sample-experiment/Fetal_Renal_Cortex/Chromatin_Accessibility/.
We downloaded pRCC clinical outcomes from TCGA Data Portal (https://tcga-data.nci.nih.gov/tcga/tcgaDownload.jsp). pRCC samples that failed the histopathological review were excluded [3]. In total, we included 277 patients in our analyses (S1 Fig, S1 Table). For germline calls, the majority of samples, 163 out of 277, were supported by germline SNV callings from two centers (BCM and BI). 100% genotype concordance rate was observed. Also, 162 curated rs11762213 genotypes were in agreement with automated call sets. All calls have alternative allelic fraction of 0.42 to 0.68, supporting heterozygous genotype [11]. Calls from BI all have genotype quality scores >125 and all calls in BCM pass the caller filter. With proved high confidence in the accuracy of genotyping rs11762213 in the germline, we recruited additional 114 samples from single-center (BCM), automated calls to form an extensive patient set (S1 Fig). For somatic SNVs in MET, after excluding cases that were recruited in the TCGA study, we formed an additional set encompassing 117 patients. Five callings were supported by two centers. The rest were supported by single-center (BCM) automated calls.
Cancer-specific survival was defined using the same criteria as described in a ccRCC study [9]. Deaths were considered as cancer-specific if the “Personal Neoplasm Cancer Status” is “With Tumor”. If “Tumor Status” is not available, then the deceased patients were classified as cancer-specific death if they had metastasis (M1) or lymph node involvement (≥ N1) or died within two years of diagnosis. An R package, “survival”, was used for the survival analysis.
We remapped all reads using bwa 0.7.12, which supports split read mapping [44]. Then we used DELLY [45] with default parameters for somatic SV calling. To avoid sample contamination or germline SVs, we filtered our call set against the entire TCGA pRCC WGS dataset, regardless of sample match. We discharged all callings that were marked “LowQual” (PE/SR support below 3 or mapping quality below 20). Last, to further eliminate germline contamination, we filtered out SVs that show at least 0.8 reciprocally overlapping with 1000 Genome Phase 3 SV call set (only 1/425 filtered out).
For Lumpy-SV [46], we ran it with default parameters. We also filtered the results using the 1000 Genome Phase 3 call set and required the SV have both paired-end and split reads supports.
WGS Mutations were extracted with flanking 5’ and 3’ nucleotide context. The raw mutation counts were normalized by trinucleotide frequencies in the whole mappable genome.
To identify signatures in the mutation spectra, we used a robust, objective LASSO-based method. First, 30 known signatures were downloaded from COSMIC (http://cancer.sanger.ac.uk/cosmic/signatures). Then we solved a positive, zero-intercept linear regression problem with L1 regularizer to obtain signatures and corresponding weights for each genome. Specifically, we solved the problem:
minW(∥SW−M∥2+ λ∥W∥)
Where M is the mutation matrix, containing the mutations of each sample in 96 nucleotide contexts. S is the 96×30 signature matrix, representing the mutation probability in 96 nucleotide contexts of the 30 signatures. W is the weighting matrix, representing the contribution of 30 signatures to each sample.
The penalty parameter lambda (λ) was determined empirically using 10-fold cross-validation individually for every sample.λ was chosen to maximize sparsity and constrained to keep mean-square error (MSE) within one standard error of its minimum. Last, we discharged signatures that composite less than 5% of the total detectable signatures.
In total, we collected HumanMethylation450 BeadChip array data for 139 samples that are either methylation cluster 1 or 2. We used an R package “IMA” to facilitate analysis [47]. After discharging sites with missing values or on sex chromosomes, we obtained beta-values on 366,158 CpG sites in total. Then we tested beta-values of each site by Wilcoxon rank sum test between two methylation clusters. After adjusting p-value using Benjamini-Hochberg procedure, we called 9,324(2.55%) hypermethylation sites. These sites had an adjusted p-value of less than 0.05 and mean beta-values in methylation cluster 1 were 0.2 or higher than in methylation cluster 2.
We used the method described by Roberts et al. [23]. For every C>{T,G} and G>{A,C} mutation we obtained 20bp sequence both upstream and downstream. Then enrichment fold was defined as:
Enrichment Fold= MutationTCW/WGA × ContextC/GMutationC/G×ContextTCW/WGA
Here TCW/WGA stands for T[C>{T,G}]W and W[G>{A,C}]A. W stands for A or T. p-value for enrichment were calculated using one-sided Fisher-exact test. To adjust for multiple hypothesis testing, p-values were corrected using Benjamini-Hochberg procedure.
WXS data for APOBEC enrichment and signature analysis was obtained from a processed somatic call set: hgsc.bcm.edu_KIRP.IlluminaGA_DNASeq.1.protected.maf. This dataset includes 155 pRCC samples and three UC samples. We used hgsc.bcm.edu_KIRC.Mixed_DNASeq.1.protected.maf for ccRCC analyses.
We identified chromatin remodeling genes based on its significance in pRCC and function. Our gene list was the intersection of genes in the original TCGA pRCC study [3] molecular feature table with the chromatin remodeling and SNI/SWF pathway gene lists. Our gene set included ten genes: SETD2, KDM6A, PBRM1, SMARCB1, ARID1A, ARID2, MLL2 (KMT2D), MLL3(KMT2C), MLL4(KMT2B), EP300. We defined chromatin remodeling defect as nonsynonymous mutations in these genes. For missense mutations, we additionally filtered out mutations with Polyphen score [48] less than 0.9 (benign). We noticed BAP1 is not in the gene list. However, adding BAP1 into the list did not change the significance of our key tests (p<0.0115 for mutation counts in DHS and p<0.020 for mutation percentage in DHS).
For replication time, in order to avoid cell type redundancy, we only kept GM12878 as the representative of all lymphoblastoid cell lines. Eleven cell types were included in our analysis: BG02ES, BJ, GM12878, HeLaS3, HEPG2, HUVEC, IMR90, K562, MCF7, NHEK, SK-NSH. Wave smoothed replication time signal was averaged in a±10kb region from every mutation. To avoid potential selection effects, we removed mutations in exome and flanking 2bp. Regions overlapping with reference genome gaps and DAC blacklist (https://genome.ucsc.edu/) were removed as well. Last, we picked the median number from 11 cell types at each mutation position.
To test the significance of replication time of noncoding mutations between two groups, we defined the ones have replication time stand above 90 percentile in all pooled mutations as “mutations in early replicating regions”. Then we calculated the percentage of “mutations in early replicating regions” of total mutations for each sample and compared between the two groups using one-sided rank-sum test.
We used PhyloWGS [49] to infer the evolutionary trees for each individual tumor. To mitigate the effects of copy number change, we removed all the SNVs inside the copy number change regions as defined by the assay-based method in the original TCGA study [3]. To be prudent, we filtered SNPs in any region with an absolute log2 tumor to normal copy number ratio larger than 0.3. Additionally, we removed all SNVs with allele frequency higher than 0.6 as they were likely affected by copy number loss.
|
10.1371/journal.pcbi.1007129 | DeepConv-DTI: Prediction of drug-target interactions via deep learning with convolution on protein sequences | Identification of drug-target interactions (DTIs) plays a key role in drug discovery. The high cost and labor-intensive nature of in vitro and in vivo experiments have highlighted the importance of in silico-based DTI prediction approaches. In several computational models, conventional protein descriptors have been shown to not be sufficiently informative to predict accurate DTIs. Thus, in this study, we propose a deep learning based DTI prediction model capturing local residue patterns of proteins participating in DTIs. When we employ a convolutional neural network (CNN) on raw protein sequences, we perform convolution on various lengths of amino acids subsequences to capture local residue patterns of generalized protein classes. We train our model with large-scale DTI information and demonstrate the performance of the proposed model using an independent dataset that is not seen during the training phase. As a result, our model performs better than previous protein descriptor-based models. Also, our model performs better than the recently developed deep learning models for massive prediction of DTIs. By examining pooled convolution results, we confirmed that our model can detect binding sites of proteins for DTIs. In conclusion, our prediction model for detecting local residue patterns of target proteins successfully enriches the protein features of a raw protein sequence, yielding better prediction results than previous approaches. Our code is available at https://github.com/GIST-CSBL/DeepConv-DTI.
| Drugs work by interacting with target proteins to activate or inhibit a target’s biological process. Therefore, identification of DTIs is a crucial step in drug discovery. However, identifying drug candidates via biological assays is very time and cost consuming, which introduces the need for a computational prediction approach for the identification of DTIs. In this work, we constructed a novel DTI prediction model to extract local residue patterns of target protein sequences using a CNN-based deep learning approach. As a result, the detected local features of protein sequences perform better than other protein descriptors for DTI prediction and previous models for predicting PubChem independent test datasets. That is, our approach of capturing local residue patterns with CNN successfully enriches protein features from a raw sequence.
| The identification of drug-target interactions (DTIs) plays a key role in the early stage of drug discovery. Thus, drug developers screen for compounds that interact with specified targets with biological activities of interest. However, the identification of DTIs in large-scale chemical or biological experiments usually takes 2~3 years of experiments, with high associated costs [1]. Therefore, with the accumulation of drugs, targets, and interaction data, various computational methods have been developed for the prediction of possible DTIs to aid in drug discovery.
Among computational approaches, docking methods, which simulate the binding of a small molecule and a protein using 3D structure, were initially studied. Docking methods recruit various scoring functions and mode definitions to minimize free energy for binding. Docking methods have advanced by themselves, and recently, the Docking Approach using Ray-Casting (DARC) model identified 21 compounds by using an elaborate binding pocket topography mapping methodology, and the results were reproduced in a biochemical assay [2]. In addition, studies have examined several similarity-based methods in which it was assumed that drugs bind to proteins similar to known targets and vice versa. One of the early methods is that of Yamanashi et al., which utilized a kernel regression method to use the information on known drug interactions as the input to identify new DTIs, combining a chemical space and genomic spaces into a pharmacological space [3]. To overcome the requirement of the bipartite model for massive computational power, Beakley et al. developed the bipartite local model, which trains the interaction model locally but not globally. In addition to substantially reducing the computational complexity, this model exhibited higher performance than the previous model [4]. As another approach to DTI prediction models, matrix factorization methods have been recruited to predict DTIs, which approximate multiplying two latent matrices representing the compound and target protein to an interaction matrix and similarity score matrix [5, 6]. In this work, regularized matrix factorization methods successfully learn the manifold lying under DTIs, giving the highest performance among previous DTI prediction methods. However, similarity-based methods are not commonly used at present to predict DTIs, as researchers have found that similarity-based methods work well for DTIs within specific protein classes but not for other classes [7]. In addition, some proteins do not show strong sequence similarity with proteins sharing an identical interacting compound [8].
Thus, feature-based models that predict DTI features of drugs and targets have been studied [9–11]. For feature-based DTI prediction models, a fingerprint is the most commonly used descriptor of the substructure of a drug [12]. With a drug fingerprint, a drug is transformed into a binary vector whose index value represents the existence of the substructure of the drug. For proteins, composition, transition, and distribution (CTD) descriptors are conventionally used as computational representations [13]. Unfortunately, feature-based models that use protein descriptors and drug fingerprints showed worse performance than previous conventional quantitative structure-activity relationship (QSAR) models [9]. To improve the performance of feature-based models, many approaches have been developed, such as the use of interactome networks [14, 15] and minwise hashing [16]. Although various protein and chemical descriptors have been introduced, feature-based models do not show sufficiently good predictive performance [17]. For conventional machine learning models, features must be built to be readable by modeling from original raw forms, such as simplified molecular-input line entry system (SMILES) and amino acid sequences. During transformation, rich information, such as local residue patterns or relationships, is lost. In addition, it is hard to recover lost information using traditional machine learning models.
In recent years, many deep learning approaches have recently been developed and recruited for omics data processing [18] as well as drug discovery [19], and these approaches seem to be able to overcome limitations. For example, DeepDTI built by Wen et al. used the deep belief network (DBN) [20], with features such as the composition of amino acids, dipeptides, and tripeptides for proteins and extended-connectivity fingerprint (ECFP) [21] for drugs [7]. The authors also discussed how deep-learning-based latent representations, which are nonlinear combinations of original features, can overcome the limitations of traditional descriptors by showing the performance in each layer. In another study by Peng et al. [22], MFDR employed sparse Auto-Encoder (SAE) to abstract original features into a latent representation with a small dimension. With latent representation, they trained a support vector machine (SVM), which performed better than previous methods, including feature- and similarity-based methods. In another study called DL-CPI by Tian et al. [23], domain binary vectors were employed to represent the existence of domains used to describe proteins.
One way to reduce the loss of feature information is to process raw sequences and SMILES as their forms. In a paper by Öztürk et al., DeepDTA was used to represent raw sequences and SMILES as one-hot vectors or labels [24]. With a convolutional neural network (CNN), the authors extracted local residue patterns to predict the binding affinity between drugs and targets. As a result, their model exhibited better performance on a kinase family bioassay dataset [25, 26] than the previous model, kronRLS [27] and SimBoost [28]. Because their model is optimized by densely constructed kinase affinities, DeepDTA is appropriate to predict kinase affinities not to predict new DTIs with various protein classes. Furthermore, they evaluated their performances on the identical dataset, rather than on independent dataset from new sources or databases.
To overcome the aforementioned problems, here, we introduce a deep learning model that predicts massive-scale DTIs using raw protein sequences not only for various target protein classes but also for diverse protein lengths. The overall pipeline of our model is depicted in Fig 1. First, for the training model, we collected large-scale DTIs integrated from various DTI databases, such as DrugBank [29], International Union of Basic and Clinical Pharmacology (IUPHAR) [30], and Kyoto Encyclopedia of Genes and Genomes (KEGG) [31]. Second, in model construction, we adopted convolution filters on the entire sequence of a protein to capture local residue patterns, which are the main protein residues participating in DTIs. By pooling the maximum CNN results of sequences, we can determine how given protein sequences match local residue patterns participating in DTIs. Using these data as input variables for higher layers, our model constructs, abstracts and organizes protein features. After new protein features are generated, our model concatenates protein features with drug features, which come from fingerprints in the fully connected layer and predict the probability of DTIs via higher fully connected layers. Third, we optimized the model with DTIs from MATADOR [32] and negative interactions predicted from Liu et al. [33]. Finally, with the optimized model, we predicted DTIs from bioassays such as PubChem BioAssays [34] and KinaseSARfari [35] to estimate the performance of our model. As a result, our model exhibits better performance than previous models.
As a normal step of hyperparameter setting, we first tuned the learning rate of the weight update to 0.0001. After the learning rate was fixed, we benchmarked the sizes and number of windows, hidden layers of the drug features, and the concatenating layers with the area under precision-recall (AUPR) on the external unseen validation dataset, which was built with MATADOR and a highly credible negative dataset. Finally, we selected the hyperparameters of the model, as shown in Table A in S1 Text, with the external unseen validation dataset, yielding an AUPR of 0.832 and area under the curve (AUC) of 0.852, as shown in Fig 2. The AUPR value of our model was less than the AUPR of the similarity descriptor; however, that does not mean that our method has lower prediction performance than the similarity method because the size of the validation is too small to evaluate the general performance. In addition, we further examined the effect of fixed maximum protein length on the prediction performance. As shown in Fig A in S1 Text, we confirmed that the prediction performance of our model is not biased to the fixed maximum protein length. Finally, the fully optimized model is visualized as a graph, shown in S1 Fig, respective to our model, the CTD descriptor, and similarity descriptors. In the same manner, we built and optimized models that use other protein descriptors with the same activation function, learning rate, and decay rate.
After the hyperparameters were tuned, we compared the performance based on the independent test datasets with the different protein descriptors, the CTD descriptor (which is usually used in the conventional chemo-genomic model) [13], the normalized Smith-Waterman (SW) score [36], and our convolution method. The results showed that our model exhibited better performance than the other protein descriptors for all datasets, as shown in Fig 3 and Fig B in S1 Text. With the threshold selected by the equal error rate (EER) [37], our model performed equally well with both the PubChem and KinaseSARfari datasets, indicating that our model has general application power. Our convolution method gave the highest accuracy score and F1 score for the PubChem dataset (Fig 3A) [34] and its subsets (Fig 3B–3D) and a slightly lower F1 score for the KinaseSARfari dataset (Fig B in S1 Text) [35]. The CTD descriptor gave the lowest score for any dataset and any metric, which implies that CTD is less informative and less enriched than the other descriptors. Here, we also observed that the model performance using a similarity descriptor for the KinaseSARfari dataset was similar to that of the proposed model. We can interpret this result as the similarity descriptor acts as an informative feature as a local residue pattern at the domain level, not the whole protein complex.
In addition to the comparison between convolution in our model and other protein descriptors, in this section, we compared the performance of our model against recently developed deep-learning-based models. We selected three deep learning models for comparisons, SAE (MFDR, Peng et al, 2016) [22], DBN (DeepDTI, Wen et al, 2017) [7] and CNN (DeepDTA, Ozturk et al, 2018). First, MFDR trains SAE in an unsupervised manner, while proteins are represented by multi-scale local descriptor feature [38] and compounds are represented by PubChem fingerprints as input and output for SAE. With trained deep representations of sparse Auto-Encoder, they performed 5-fold cross-validation by using SVM. As a result, their model gives better performances than previous bipartite local models. Because the authors do not provide the model, we implemented the MFDR model with optimized parameters the author provided in their original paper. We tested the validity of implemented MFDR and confirmed that the implemented model produces reasonably same performance compared to the results from its original work (see Fig C S1 Text). Second, DeepDTI built by Wen et al. is based on DBN [20], which is a stack of restricted Boltzmann machine (RBM). DeepDTI takes amino acid, dipeptide and tripeptide compositions (protein sequence composition descriptors, PSC) as the protein input and ECFP with radius 1, 2 and 3 as the compound input. We used DeepDTI with the code that the authors provided (https://github.com/Bjoux2/DeepDTIs_DBN) and optimized hyperparameters as the authors mentioned. Third, DeepDTA built by Ozturk et al. used stacked CNN on protein sequences and SMILES to predict affinity between target protein and compound. DeepDTA is optimized for Davis [25] and KIBA [26] dataset which contains kinases protein, their inhibitors, and dense affinity values, showing better prediction performances than previous affinity prediction models. We also used DeepDTA with the code from the original work (https://github.com/hkmztrk/DeepDTA) and optimized hyperparameters they provided. For the DTI prediction performance comparison, we activate the last layer with sigmoid function to predict interaction, not affinity, also we changed loss function as binary cross-entropy from mean squared error. It should be noticed that we compared the performance of all three models by training and testing with the same data set we used for a fair comparison.
Results of performance comparison between our proposed model and the three related models are shown in Fig 4, showing that performances (accuracy, F1) of our model (DeepConv-DTI) are better than other models. MFDR which gave high AUC in 5-fold cross-validation shows decreased performances in the independent test dataset. We can speculate that SAE which learns deep representation of DTI in an unsupervised way is not appropriate for a case that datasets are composed of various protein classes. In the case of DeepDTI, DeepDTI takes physicochemical properties (PSC) of whole protein sequence including subsequences or domains which do not participate in the interaction with compounds, resulting in worse performance than our model which extracts local residue patterns. For DeepDTA, DeepDTA also shows worse performances than our model with having a relatively large variance. We interpret the worse performance of DeepDTA as follows. DeepDTA is optimized for a densely constructed dataset with specific protein class, while the training dataset in this comparison covers various protein classes (kinase, protease, ion channel, nuclear receptor, GPCR, etc), not only kinase class. Thus, DeepDTA which is specialized for a specific protein class could not achieve better prediction performance in the generalized protein classes.
In addition to the three models we compared, we also compared our model with DL-CPI [23] built by Tian et al. which used protein domain information. For proteins whose domain information is not in Pfam [39], datasets for training, validation and test are not fully available. Therefore, we independently compared performances between DL-CPI and our model by additionally built the training, validation, and test datasets. Performance comparison results are described in Fig E in S1 Text. We confirmed that the proposed model shows better performance than DL-CPI. Because protein descriptor of DL-CPI is sparse, containing few values in large dimension, which may decrease performances.
In overall, our model shows better performance than previous deep learning models in an independent test dataset from a different database, which contains distinct DTIs, dealing with DTIs with various protein classes and their interacting compounds.
Because we pooled the maximum convolution results by each filter for each window, the pooled results could highlight regions of matches with local residue patterns. Although we cannot measure exactly how those values affect the DTI prediction results, the pooled maximum convolution result will affect the prediction performance by going through higher fully connected layers. Therefore, if our model is capable of capturing local residue patterns, it would give high values to important protein regions, such as actual binding sites.
Examining and validating the convolution results from the intermediate layer showed that our model could capture local residue patterns that participate in DTIs. The sc-PDB database provides atom-level descriptions of proteins, ligands, and binding sites from complex structures [40]. By parsing binding site annotations, we can query binding sites between protein domains and pharmacological ligands for 7,179 entries of Vertebrata. From the queried binding sites and pooled maximum convolution results, we statistically test our assumption that the pooled maximum convolution results cover the important regions, including binding sites. Each window has 128 pooled convolution results, which shows bias in covering some regions. Thus, we randomly generated 128 convolution results 10,000 times for each sc-PDB entry and counted how many of those random results covered each amino acid in the binding sites, which resulted in the construction of normal distributions. For each normal distribution constructed by the randomly generated convolution results, considered a null hypothesis, we executed a right-tailed t-test with the number from the convolution results of our model for each window. Because we did not know which window detects the binding site, we took the most significant p-value (minimum p-value adjusted by the Benjamini-Hochberg procedure [41]). The sc-PDB entry information and p-values of a window for each sc-PDB entry are summarized in the S1 File. We summarize the results of binding site detection from the most significant p-value among windows by significance level cutoff in Fig 5. In addition, we examined sc-PDB entries with the most significant p-values for diverse window sizes. We visualized two high-score sc-PDB entries from two perspectives—the whole receptor-ligand complex and binding site-ligand perspectives—by using UCSF Chimera [42] as shown in Fig 6. To visualize convolution results with a simplified view, first, we selected the top 5 ranked globally max-pooled results among all filters for each window because whole protein sequences are usually covered by convolution results if we select all results. Second, we rendered residues covered by convolution results by the number of covering convolution results. We visualized two sc-PDB entries, 1a7x_1 and 1ny3_1. 1a7x_1, representing the complex of the ion channel, protein Peptidyl-prolyl cis-trans isomerase FKBP1A (FKB1A_HUMAN in UniProt), which has a short sequence length (108), and BENZYL-CARBAMIC ACID [8-DEETHYL-ASCOMYCIN-8-YL]ETHYL ESTER (FKA in PDB ligand) [43]. 1ny3_1 is the complex of the kinase protein, MAP kinase-activated protein kinase 2 (MAPK2_HUMAN in UniProt) with sequence length 400, and ADENOSINE-5’-DIPHOSPHATE (ADP in PDB ligand) [44]. Through the above evaluation, we can confirm that our proposed model is capable of capturing local residue patterns of proteins that are considered important features for DTI prediction, such as actual binding sites.
From the results shown in Fig 6, we can confirm that our model can capture the local residue patterns of proteins that participate in DTIs. Thus, to examine further characteristics of the captured protein local residue patterns, we visualized the protein features from the fully connected layer after the global max-pooling of convolution results. We visualized 1,527 proteins used in the training dataset categorized in various protein classes. Specifically, we visualized 257 GPCRs, 44 nuclear receptors, 304 ion channel receptors, 604 kinases, and 318 proteases. For visualization, we conducted t-distributed stochastic neighbor embedding (t-SNE) for dimension reduction and visualization [45]. t-SNE can map high-dimensional features to low-dimensional ones, such as 2-dimensional features, minimizing information loss during dimension reduction. Surprisingly, although our model is not intended to identify protein classes, it can roughly discriminate protein classes from the intermediate protein layer, as shown in Fig G in S1 Text.
In this work, we built a novel DTI prediction model to extract local residue patterns of whole target protein sequences with CNN. We trained the model with DTIs from various drug databases and optimized the model with an external validation dataset. As a result, the detected local features of protein sequences perform better than other protein descriptors, such as CTD and SW scores. Our model also performs better than a previous model built on DBN. In addition, by analyzing pooled convolution results and statistically and manually comparing them with annotations from sc-PDB entries, we showed that, for some proteins, our model is capable of detecting important regions, including binding sites. Therefore, our approach of capturing local residue patterns with CNN successfully enriches protein features for DTI prediction.
The number of 3D structures in Protein Data Bank [46] is relatively smaller than the number of sequences, limiting 3D structure-based DTI prediction methods. For example, the number of PDB entries for Homo sapiens is 42,745, while the number of protein sequences for Homo sapiens is 177,661 in UniProtKB. However, our method does not depend on the 3D structure of proteins because it considers only protein sequence, rather than classical protein feature descriptors such as the CTD descriptor and normalized SW score. As a result, our method can be more generally applied to predict DTIs than methods needing 3D structures.
Although our model shows improved prediction performance, there is still room for improvement. First, we simply used Morgan/Circular fingerprints, which are binary and have large dimensions. Therefore, we will use more informative chemical descriptors, based on neural networks for DTI prediction, to achieve advanced performance. Second, as shown in a previous study [47], considering 3D structure information is an effective substitution for chemical elaboration. Therefore, in the future, we will elaborate upon our model by considering 3D structure features.
To build the training dataset, we obtained known DTIs from three databases: DrugBank, KEGG, and IUPHAR. To remove duplicate DTIs among the three databases, we unified the identifiers of the compounds and the proteins. For the drugs, we standardized the identifiers of the compounds in the DrugBank and KEGG databases with the InChI descriptor. For the proteins, we unified the identifiers of the proteins as UniProtKB/Swiss-Prot accessions [48]. Among the collected DTIs, we selectively removed proteins of Prokaryota and single-cell Eukaryota, retaining only proteins of Vertebrata. Finally, 11,950 compounds, 3,675 proteins, and 32,568 DTIs were obtained in total. Because all collected DTIs are regarded as positive samples for training and negative DTIs are not defined in the databases above, a random negative DTI dataset is inevitably generated. To reduce bias from the random generation of negative DTIs, we built ten sets of negative DTIs exclusively from the positive dataset. The detailed statistics of the collected training dataset are shown in Table D in S1 Text.
To optimize our model with the most adequate hyperparameters, we constructed an external validation dataset that had not seen DTIs in the training phase. We collected positive DTIs from the MATADOR database [32], including ‘DIRECT’ protein annotations, and all DTIs observed in the training dataset were excluded. To build a credible negative dataset, we obtained negative DTIs via the method of Liu et al. [33]. This method selects candidate negative DTIs with low similarity to known positive DTIs. From the obtained negative dataset, we balanced the negative dataset with the positive dataset, using a negative score (>0.95). As a result, 370 positive DTIs and 507 negative DTIs were queried for the external validation set. The statistics of the external validation dataset are summarized in Table E in S1 Text.
To evaluate our model, we built two independent test datasets from the PubChem BioAssay database [34] and ChEMBL KinaseSARfari [35]; these datasets consisted of results from experimental assays. To obtain positive DTIs from PubChem, we collected ‘Active’ DTIs from the assays with the dissociation constant (Kd < 10μm) [49]. Because we sought to predict whether a drug binds to a protein, among the many types of assays (Potency, IC50, AC50, EC50, Kd, Ki), evaluation of the dissociation constant (Kd) was the most appropriate assay for obtaining positive samples. For the negative samples, we took the samples annotated as ‘Inactive’ from the other assay types. Because there were too many negative samples in the PubChem BioAssay database, we first collected only negative samples whose drug or target was included in the positive samples from the PubChem BioAssay database. Second, we selected as many random negative samples as positive DTIs from PubChem BioAssay. As a result, total 36,456 positive and negative samples were built with 21,907 drugs and 698 proteins. For the performance evaluation, we created three subsets of the PubChem bioassay independent dataset for humans, which consisted of only new compounds, new proteins, and new DTIs. Detailed summaries of the PubChem dataset and its subset are shown in Table F in S1 Text. We also collected samples from KinaseSARfari. KinaseSARfari consists of assays involving a compound that binds to a kinase domain. To obtain positive samples from KinaseSARfari, we considered each assay result with a dissociation constant of (Kd < 10μm) as positive [49]; this value is sufficiently small to be considered positive. In contrast to the PubChem BioAssay, the number of negative samples was similar to the number of positive samples in KinaseSARfari; therefore, we did not sample the negative samples. We collected 3,835 positive samples and 5,520 negative samples with 3,379 compounds and 389 proteins. Detailed statistics of the KinaseSARfari dataset are shown in Table F in S1 Text. In addition, we summarize the portion of the protein class in each dataset in Fig H in S1 Text. Here, we confirmed that the training and the validation datasets were not biased toward a specific protein class.
In our model, we used the raw protein sequence as the input for the protein but did not use the raw SMILES string as the input for the drug. For the drug, we used the Morgan/Circular drug fingerprint, which analyzes molecules as a graph and retrieves substructures of molecular structures from subgraphs of the whole molecular graph [21]. Specifically, we used RDKit [50] to yield a Morgan/Circular fingerprint with a radius of 2 from a raw SMILES string. Finally, each drug can be represented as a binary vector with a length of 2,048, whose indices indicate the existence of specific substructures.
SAE is Auto-Encoder whose distribution of latent representations is regularized with sparsity term [58]. In loss calculation, Kullback-Leibler divergence (KLD) loss between Bernoulli distributions each dimension in latent representation ρ^ and desired sparsity parameter ρ is added to reconstruction loss of Auto-Encoder and ridge loss for weights.
Jsparse(W,b)=J(W,b)+β∑js2KL(ρ||ρ^j)
where
ρj^=1m∑i=1m[aj(2)(x(i))]
During the training of the neural network, KLD acts as a constraint for latent representation following desired sparsity parameter. As a result, for each dimension of latent representation, only a few samples are activated, giving a more reliable representation of original input. In the previous study, MFDR used SAE to build an informative latent representation of DTI, which are composed of multi-scale local descriptors [38] and PubChem fingerprints.
DBN is a generative graphical model proposed by Geoffrey Hinton [20]. DBN is actually a stack of an RBM. RBM consists of visible and hidden units, constructing a bipartite graph. In RBM, probabilistic distribution of visible units is learned in an unsupervised way, with a probabilistic distribution of visible and hidden units
P(v,h|W)=1ZeaTv+bTh+vTWh
and marginal distribution of visible units
P(v|W)=1Z∑heaTv+bTh+vTWh
to maximize the probability of visible units for V in a training set with weight matrix W
argmaxW∏v∈VP(v|W)
In DBN, during stacking of RBMs, hidden units of the previous RBM are fed as visible layers of the next RBM. In addition, RBM adopts contrastive divergence for fast training, which uses gradient descent and Gibbs sampling. In a previous study, DeepDTI, the input concatenation of drug and target protein features, PSC descriptors and ECFP with a radius of 1, 2 and 3, was considered a first visible layer. The authors attached logistic regression to the last hidden units to predict DTIs.
To measure the prediction performance of our deep neural model based on the independent test dataset after the classification threshold was fixed, we obtained the following performance metrics: sensitivity (Sen.), specificity (Spe.), precision (Pre.), accuracy (Acc.), and the F1 measure (F1). See the formulas below:
Sen.=TP/P
Spe.=TN/N
Pre.=TP/(TP+FP)
Acc.=(TP+TN)/(P+N)
F1=(Sen*Pre)/(Sen+Pre)
where TP is true positive, TN is true negative, FP is false positive, FN is false negative, T is positive, and N is negative.
|
10.1371/journal.ppat.1005521 | Endoplasmic Reticulum Stress Induced Synthesis of a Novel Viral Factor Mediates Efficient Replication of Genotype-1 Hepatitis E Virus | Hepatitis E virus (HEV) causes acute hepatitis in many parts of the world including Asia, Africa and Latin America. Though self-limiting in normal individuals, it results in ~30% mortality in infected pregnant women. It has also been reported to cause acute and chronic hepatitis in organ transplant patients. Of the seven viral genotypes, genotype-1 virus infects humans and is a major public health concern in South Asian countries. Sporadic cases of genotype-3 and 4 infection in human and animals such as pigs, deer, mongeese have been reported primarily from industrialized countries. Genotype-5, 6 and 7 viruses are known to infect animals such as wild boar and camel, respectively. Genotype-3 and 4 viruses have been successfully propagated in the laboratory in mammalian cell culture. However, genotype-1 virus replicates poorly in mammalian cell culture and no other efficient model exists to study its life cycle. Here, we report that endoplasmic reticulum (ER) stress promotes genotype-1 HEV replication by inducing cap-independent, internal initiation mediated translation of a novel viral protein (named ORF4). Importantly, ORF4 expression and stimulatory effect of ER stress inducers on viral replication is specific to genotype-1. ORF4 protein sequence is mostly conserved among genotype-1 HEV isolates and ORF4 specific antibodies were detected in genotype-1 HEV patient serum. ORF4 interacted with multiple viral and host proteins and assembled a protein complex consisting of viral helicase, RNA dependent RNA polymerase (RdRp), X, host eEF1α1 (eukaryotic elongation factor 1 isoform-1) and tubulinβ. In association with eEF1α1, ORF4 stimulated viral RdRp activity. Furthermore, human hepatoma cells that stably express ORF4 or engineered proteasome resistant ORF4 mutant genome permitted enhanced viral replication. These findings reveal a positive role of ER stress in promoting genotype-1 HEV replication and pave the way towards development of an efficient model of the virus.
| Hepatitis E virus (HEV) is one of the most common causes of acute and sporadic viral hepatitis. It is a small positive strand RNA virus, which is transmitted through the feco-oral route. Owing to lack of sanitation and unavailibility of safe drinking water, populations of developing and resource starved countries are prone towards HEV infection. Recent reports also indicate HEV induced acute and chronic hepatitis in organ transplant patients. Another peculiar characteristic of HEV is attributed to its ability to cause high mortality (~30%) in infected pregnant women. Even after 30 years of discovery of the virus, little information exists regarding viral life cycle and replication machinery. HEV is divided into seven genotypes. Genotype-3 and 4 viruses infect humans and a few animals (such as pigs, deer, mongeese) and have been reported from industrialized countries. Genotype-3 and 4 viruses have been successfully propagated in the laboratory in mammalian cell culture. However, genotype-1 virus, which is known to infect human and is a major public health concern in south Asian countries, replicates poorly in mammalian cell culture and no other efficient model exists to investigate the viral life cycle. With the goal of developing an efficient laboratory model of genotype-1 HEV, we attempted to identify a permissive cellular condition that would allow efficient viral replication in human hepatoma cells. Here, we report that endoplasmic reticulum stress inducing agents promote genotype-1 HEV replication by initiating cap-independent, internal translation mediated synthesis of a novel viral factor, which we have named ORF4. Further investigations revealed that ORF4 is expressed only in genotype-1 and it acts by interacting with multiple viral and host proteins and cooperates with host eEF1α1 (eukaryotic elongation factor 1 isoform 1) to control the activity of viral RNA dependent RNA polymerase. Moreover, a proteasome resistant ORF4 mutant significantly enhanced viral replication. Thus, our study identifies an optimal condition required for efficient replication of genotype-1 HEV and dissects out the molecular mechanism governing that. Data presented here will be instrumental in developing an efficient model of the virus.
| Hepatitis E is a feco-orally transmitted positive strand RNA virus that causes acute and sporadic hepatitis in human and other animals [1]. It is also emerging to be a major cause of infection in organ transplant patients worldwide [2]. Though self-limiting in normal individuals, a peculiar characteristic of HEV is attributed to its ability to cause high mortality (~30%) in infected pregnant women [3]. The viral genome consists of a 7.2 kb 5’-capped and 3’-polyadenylated RNA, which encodes three known open reading frames (ORF); ORF1 codes for non-structural proteins, ORF2 codes for the major capsid protein and ORF3 codes for an accessory protein that associates with multiple host proteins and is supposed to modulate host signaling pathways [1]. ORF3 also interacts with host tumor susceptibility gene 101 (TSG 101) and plays an essential role in virus release [4, 5]. ORF2 has been observed to bind to the viral genomic RNA [6], induce endoplasmic reticulum (ER) stress [7, 8] and inhibit NFκB activity [9] in human hepatoma cells, suggesting a possible regulatory role of the viral capsid protein. Seven genotypes of HEV have been reported; genotype-1 (g-1), genotype-2 (g-2) exclusively infect human whereas genotype-3 (g-3), genotype-4 (g-4) infect human, pig, deer, mongeese and rabbit. Infection by genotypes 5–7 have not been reported in human. Genotype-5 (g-5), genotype-6 (g-6) infects wild boar and genotype-7 (g-7) is known to infect camel [10, 11]. Little is known about the life cycle of HEV owing to lack of a handy animal or cell culture model. Among the various genotypes, in vitro synthesized genome of g-3 and g-4 HEV replicates well in mammalian cell culture [12]. Attempts at achieving high replication efficiency of g-1 HEV in mammalian cell culture have not been successful [13, 14]. Interestingly, in one of the g-3 HEV infected patients, human ribosomal S17 coding sequence was found to be inserted in the ORF1 region, which conferred growth advantage to the virus [15, 16]. Molecular mechanisms underlying the above observation remain to be explored. Moreover, such an insertion appears to be a very rare occurrence.
Efficient translation and replication are two crucial events in the life of an RNA virus, tight control of which is essential for survival of both virus and its host. These events are under strict surveillance by the host defence machinery as innate antiviral measures, most common being induction of ER stress and unfolded protein response, inactivation of eukaryotic translation initiation factor 2α and shut down of cap-dependent translation [17]. Degradation of viral double stranded RNA by innate immune effectors [18] and autophagy [19] also serves as host defence mechanisms against many viruses. Viruses on the other hand, employ clever strategies to exploit the adversities imposed by the host. Innate immune response is countered by various strategies such as inhibition of type I interferon production [20], manipulation of pattern recognition receptor signaling [21], IRF3 (interferon regulatory factor 3) inhibition [22, 23] and autophagy inhibition [24]. Translational restrictions are overcome by ribosome shunting, reinitiation, stimulation of eIF4F complex assembly, inhibition of elF2α phosphorylation [25], internal ribosome entry site (IRES) mediated translation [26, 27] and execution of both cap-dependent and cap-independent modes of translation [28, 29] depending on the cellular state.
The second step in the RNA virus life cycle pertains to genome replication. Most viruses encode regulatory proteins, RNA and/or miRNA that exploit host machineries to augment viral replication. Hepatitis C, Dengue and Polio viruses activate autophagy [30, 31, 32]. Hepatitis B virus inhibits proteasome activity [33], which leads to increased viral replication. Hence, depending upon the host cellular condition, each virus seems to have evolved suitable survival strategies that permits its optimal growth.
Since g-1 HEV does not replicate efficiently in mammalian cell culture, we wondered whether any particular cellular condition might enhance viral replication. Screening of various compounds known to alter cellular condition revealed a role of ER stress inducing compounds, thapsigargin and tunicamycin in enhancing g-1 HEV replication. Further studies led to the identification of a novel viral protein synthesized from an overlapping reading frame within ORF1, which was named open reading frame 4 (ORF4). The role of ORF4 in viral replication was explored.
In order to identify the influence of a particular cellular condition on HEV replication, Huh7 cells were transfected with wild type capped genomic RNA (WT HEV). 6 days post transfection, viral replication was measured by monitoring the level of sense and antisense RNA and estimating the percentage of cells expressing viral helicase and ORF2. Note that helicase synthesis reflects ORF1 translation from genomic RNA whereas ORF2 is synthesized from the subgenomic RNA (generated after replication). A replication deficient mutant genome (GAA HEV), in which “DD” amino acids of RdRp (a.a. 1551, 1552 in ORF1) were altered to “AA”; was used to ensure specificity of both assays. Quantitative real-time PCR (QRT-PCR) of sense strand RNA level in GAA HEV transfected samples reflected the level of input RNA (quantity of transfected RNA, Fig 1A). Sense strand RNA level was approximately four fold higher in WT HEV RNA transfected sample (compared to GAA HEV), reflecting replication mediated increase. As expected, no antisense RNA was detected in GAA HEV transfected samples but was detectable at basal levels in WT HEV expressing samples. Upon treatment with known ER stress inducers; thapsigargin (TG) and tunicamycin (TUN), sense and antisense RNA levels were further increased by 2–3 fold (Fig 1A). Similarly treated samples were analyzed by immunofluorescence assay to measure the percentage of helicase and ORF2 positive cells (Fig 1B). Representative images are shown (S1 Fig). GAA HEV transfected sample contained 2% helicase positive and no ORF2 positive cells, in agreement with QRT-PCR data, confirming specificity of the assays (Fig 1B). 20% helicase and 5% ORF2 positive cells were detected in DMSO treated WT RNA transfected cells. Thapsigargin and tunicamycin treated samples contained significantly higher percentage of helicase and ORF2 positive cells in WT HEV. Helicase positive cells were absent in thapsigargin and tunicamycin treated GAA HEV sample, ruling out the possibility of increased ORF1 translation by these compounds. We next tested whether thapsigargin and tunicamycin enhanced the replication of g-3 HEV using a Gaussia luciferase secreting replicon of g-3 HEV [16]. There was no increase in luciferase level upon treatment of the replicon expressing cells with thapsigargin and tunicamycin (Fig 1C), suggesting that both compounds had a stimulatory effect only on g-1 HEV replication.
Assuming that the mechanism underlying the observed stimulatory effect of ER stress on g-1 HEV replication is encoded in the viral genome, we analysed the g-1 HEV genome (SAR 55 strain, Genbank ID: AF444002.1) using “ATGpr”, a software to predict potential open reading frames [34]. All known ORFs of HEV were predicted. An unknown ORF of 158 amino acids within ORF1, located in +1 reading frame (with reference to ORF1, 2835–3308 bases from 5’) was also predicted, which was named ORF4 (S1 Table, Fig 1D). An ORF coding for a truncated ORF1 protein was also predicted. In contrast, sequence analysis of other HEV genotypes did not reveal any ORF resembling that of ORF4 (S1 Table). Next, we performed a bioinformatics analysis of several g-1 HEV genomic sequences available in public database to find out whether open reading frame 4 is present in all and whether the ORF4 protein sequence is conserved among the various isolates. Additionally, we analyzed the viral genomic sequence from five new cases of g-1 HEV infection, recently isolated by us at the All India Institute of Medical Sciences, New Delhi, India (Genbank ID: KU168733- KU168737, Fig 1E). All g-1 HEV genomes contain an ORF at the expected position of ORF4, with a suboptimal Kozak sequence starting either at 2832 or 2834 nucleotides, from 5’ end (Fig 1D). Three different patterns were observed with respect to termination of the putative ORF4; in 8 cases, it terminates at 3311 nucleotides (158 amino acids, full length ORF4 protein), in two cases, at 3277 nucleotides (147 amino acids) and in remaining 9 cases, it terminates at 3256 nucleotides (139 amino acids, Fig 1E). ClustalW analysis of the putative ORF4 protein sequence of these isolates revealed ~80% conservation of amino acids (Fig 1E).
To verify ORF4 and/or ΔORF1 expression, an in vitro transcription-translation assay was performed using a TNT kit. Two bands corresponding to unprocessed and probably partially processed ORF1 protein (**) were detected (Fig 2A, top and middle). Two bands corresponding to ~20kDa and ~40kDa (*) were also observed (Fig 2A, top). No such bands were detected in mock. Mutating the initiator methionine codon of ORF1 to Lysine (ATG-AAA substitution, 26 ATG mut ORF1) resulted in disappearance of ORF1 specific bands without affecting 20kDa and 40kDa bands. Similarly, blocking ORF1 translation initiation by inserting a well characterised stem loop forming sequence [35] upstream of the initiator codon of ORF1 (SL ins ORF1) abolished the bands corresponding to ORF1 without impacting 20kDa and 40kDa bands (Fig 2A). Correlating “ATGpr” prediction with above data suggested that 20kDa band may correspond to translation product of ORF4. 40kDa band could be a denaturation resistant dimeric form of ORF4 or an unrelated protein.
In agreement with the above proposition, TNT of ORF4 coding sequence produced 20 and 40 kDa bands (Fig 2A, pSGI ORF4). A peptide based rabbit polyclonal antibody was generated against the putative ORF4 protein in order to identify the unknown bands. Functionality and specificity of the antibody was validated (S2A and S2B Fig) and aliquots of TNT samples were western blotted using this antibody. Only the 20kDa band was detectable by ORF4 antibody in WT ORF1, 26 ATG mut ORF1, SL ins ORF1 and pSGI ORF4 (Fig 2A, bottom).
Two sub optimal Kozak sequences containing initiation codons are present in the ORF4 coding region (Fig 1D). Both were mutated to Lysine (ATG-AAA) in HEV ORF1 construct (ORF4 ATG DM ORF1), followed by TNT to confirm the identity of 20 and 40kDa bands. Both bands were absent in the autoradiogram and western, without affecting ORF1 level (Fig 2A). As expected, inhibiting both ORF1 and ORF4 translation initiation by stem loop insertion and ATG-AAA substitution, respectively, resulted in disappearance of all bands (SL ins ORF1 DM ORF4).
An immunofluorescence assay was conducted using ORF4 antibody to detect its expression in WT g-1 HEV genome transfected Huh7 cells. ORF4 signal was significantly higher in tunicamycin and thapsigargin treated cells compared to the DMSO control (Fig 2B). Specificity of the signal was controlled by using tunicamycin treated DM HEV (mutant g-1 HEV genome, in which ORF4 initiation codons are mutated to Lysine) transfected cells, which failed to show ORF4 signal. Tunicamycin treated 26 ATG mut HEV or GAA HEV RNA transfected cells also expressed ORF4, clearly ruling out any influence of ORF1 translation or replication on ORF4 production, respectively (Fig 2B).
In order to confirm that no ORF4 like protein is expressed in genotype-3 HEV (g-3 HEV), in vitro transcribed genome of a luciferase replicon of g-3 HEV (pSK HEV p6 luc) or WT g-1 HEV was transfected into Huh7 cells, followed by thapsigargin treatment and subsequent immunofluorescence staining using anti-ORF4 or anti-Helicase antibodies. Helicase expression was detectable in both samples whereas ORF4 expression was detectable only in the case of g-1 HEV (Fig 2C).
Next, we analysed ORF4 expression in the five g-1 HEV infected patients, in which ORF4 sequence was conserved (KU168733-KU168737, Fig 1E). ORF4 expression was assessed indirectly by monitoring the level of anti-ORF4 antibody, if any. Purified GST-ORF4 protein was readily detected by serum from all 5 patients (KU168733-KU168737) whereas serum from two healthy (CS1-CS2) individuals were negative (Fig 3A).
A stable cell line of Huh7 constitutively expressing Flag-tagged ORF4 was generated (ORF4-Huh7) to explore the role of ORF4 in HEV replication (Fig 3B). WT HEV or GAA HEV genome was transfected into ORF4-Huh7 and its control (pCDNA5-Huh7). The level of sense and antisense RNA of WT HEV was approximately two fold higher in DMSO treated ORF4-Huh7 cells compared to control (Fig 3C). As expected, GAA HEV mutant was unable to replicate. Tunicamycin treatment increased sense and antisense RNA by two fold in control and four fold in ORF4-Huh7 cells. DM HEV behaved like GAA HEV in control cells in the presence and absence of tunicamycin. In contrast to GAA HEV, DM HEV produced both sense and antisense RNA at levels equivalent to WT HEV in DMSO treated ORF4-Huh7 cells and remarkably, these levels remained unaltered in the presence of tunicamycin (Fig 3C). Similar pattern was obtained in immunofluorescence analysis of helicase and ORF2 positive cells (Fig 3D). Thapsigargin too displayed a pattern similar to tunicamycin (Fig 3D).
ORF1 translation is cap-dependent [1]. However, ORF4 could be translated in the absence of cap-dependent translation (SL ins ORF1, Fig 2A). Considering its location deep inside ORF1, we wondered whether ORF4 synthesis was driven by an internal translation initiation mechanism. Bioinformatics analysis of viral RNA flanking ORF4 region using “Reg RNA” [36] indicated the presence of a putative IRES-like element between 2701–2787 bases (Fig 1D, IRESl). Analysis of same sequence using “IRESite” [37] predicted weak homology with Equine Rhinitis A virus-1 IRES [38]. Secondary structure analysis of 2664–2845 bases encompassing the predicted IRES-like element using “mfold” [39] revealed the presence of three stem loops within 2701–2787 bases (Fig 4A, sequence in cyan). Increase in sequence length (315 bases, 2619–2933 bases) did not alter those stem loops, indicating their stability (S3 Fig). A dual luciferase reporter assay was conducted to evaluate the functionality of the IRES-like element by placing it between Renilla and Firefly coding sequences (Fig 4B). Three consecutive stop codons were introduced downstream of the Renilla coding sequence to ensure termination of cap-dependent translation of Renilla luciferase. 315 bases from HEV genome encompassing the IRES-like element (HIRESl 315) or 468 bases from 3501–3968 nucleotides (negative control for background Firefly activity, HEVcRNA) were inserted downstream of Renilla, preceding the Firefly start site (Fig 4B). Measurement of the Firefly and Renilla luciferase ratios revealed a significantly higher Firefly activity in HIRESl 315 sample (Fig 4C). The core IRES-like element (2701–2787 bases, HIRESl 87) also displayed similar activity (Fig 4C). Next, several mutant constructs were generated in which individual stem loops were destroyed by altering a few nucleotides at a time. Impairing stem loops A, B, C or bulge (A*) did not affect Firefly activity. A moderate and high reduction in Firefly activity was seen in samples containing dual mutations of both A,C and B,C stem loops, respectively (Fig 4C).
Dual mutations of both B and C were introduced into plasmids containing HEV ORF1 and HEV genome (IRESl mut ORF1 and IRESl mut HEV, respectively). In TNT of IRESl mut ORF1 construct, ORF4-specific band disappeared without affecting that of ORF1 (Fig 2A). No ORF4 was detected in cells transfected with IRESl mut HEV RNA upon tunicamycin treatment (Fig 2B). Expectedly, IRESl mut HEV genome replication was significantly reduced irrespective of tunicamycin treatment in pCDNA5-Huh7 cells, which could be restored in ORF4-Huh7 cells, though in a tunicamycin insensitive manner (Fig 3C).
To explore the mechanism(s) by which ORF4 stimulated viral replication, we identified its interaction partners among viral proteins. ORF4 directly interacted with helicase, X and ORF3 proteins of g-1 HEV, evident from Yeast Two Hybrid (Y2H) assay (Table 1). X protein of g-3 HEV also interacted with ORF4, however neither g-3 helicase nor g-3 ORF3 interacted with ORF4 (Table 1). Using overlapping deletions of ORF4, the interaction domain was mapped to 54–122 amino acids for X and ORF3 and 1–124 amino acids for helicase protein of g-1 HEV (Table 2). Coimmunoprecipitation (CoIP) of Huh7 cells transfected with plasmids encoding ORF4 and various g-1 HEV proteins confirmed its interaction with X, helicase and ORF3 (Fig 5A, 5B and 5C). Interestingly, CoIP also demonstrated that ORF4 interacted with g-1 RdRp in Huh7 cells (Fig 5D). No other viral proteins interacted with ORF4 in CoIP (S4 Fig). Since X and ORF3 interacted with the same region of ORF4 and helicase appeared to interact with a broader region/multiple domains of ORF4, we next determined whether X and ORF3 competed or cooperated with helicase for binding to ORF4. In ORF4-Huh7 cells coexpressing helicase and ORF3, though ORF4 associated with helicase and ORF3 and vice versa, ORF3 was not coprecipitated with helicase, indicating that all three were not in the same complex (Fig 6A). However, helicase, X and ORF4 could be coprecipitated, indicating cooperativity among them (Fig 6B).
We next tested whether viral RdRp associated with the X-helicase-ORF4 complex. CoIP in ORF4-Huh7 and its control cells demonstrated that RdRp coprecipitated with X and helicase only in the presence of ORF4 (Fig 6C). Moreover, all four appeared to be part of one complex as helicase and RdRp antibody could coprecipitate X and ORF4 and vice versa (Fig 6C). A pull down assay using purified proteins further confirmed that ORF4 indeed mediated the assembly of a complex consisting of RdRp, helicase, X and ORF4 (Fig 6D). In contrast, ORF3 inhibited assembly of the above complex, probably by competing for binding to ORF4 (Fig 6D, compare lane 2 with 3).
Since many of the g-1 HEV isolates encode a truncated ORF4 protein consisting of 139aa or 147aa (from N-terminus) and our Y2H based mapping of the X, Helicase and ORF3 interaction region of ORF4 was found to be confined to N-terminal 124aa, a pull down assay was performed using a deletion mutant of full length ORF4 protein comprising of N-terminal 124aa (124 ORF4-Flag). As expected, 124 ORF4-Flag could assemble a complex consisting of RdRp, X and Helicase; similar to the full length ORF4 protein (Fig 6E). These data suggest that ORF4 is functionally active in all g-1 isolates.
Since ORF4 interacts with helicase and RdRp, we wondered whether it influenced their activities. A helicase assay using Huh7 purified Helicase-Flag (Fig 7A) and bacterial purified GST-ORF4 (Fig 7B) revealed that ORF4 had no effect on RNA unwinding activity of helicase (Fig 7C). Huh7 purified ORF2-Flag (Fig 7D) was used as a negative control. Helicase assay in the presence of Huh7 purified ORF4 (Fig 7E) produced similar results (Fig 7F), indicating that under our experimental conditions, ORF4 had no effect on dsRNA unwinding activity of viral helicase.
Next, an RdRp assay was performed using Huh7 purified RdRp-Flag (Fig 8A) in the presence of increasing amount of bacterial purified GST-ORF4 or ORF4-Flag. An in vitro transcribed RNA containing 130 bases from 5’-end and 210 bases from 3’-end of g-1 HEV genome was used as template for the assay (Fig 8B). Addition of ORF4 significantly increased double stranded RNA intermediate level (680 bases), reflecting enhanced RdRp activity (Fig 8C). Observed activity was specific to viral RdRp because no signal was obtained in reactions containing ORF2-Flag or GST-ORF4 alone. RdRp assay in the presence of Huh7 cell purified full length ORF4-Flag or 124 ORF4-Flag (1–124 amino acids of ORF4) produced a similar effect (Fig 8D and 8E) confirming that both full length and 124 aa ORF4 indeed enhanced viral RdRp activity.
ORF4 indirectly associated with g-1 RdRp. In an independent study carried out in our laboratory to isolate direct interacting partners of g-1 HEV RdRp by screening a human fetal brain cDNA library using Yeast Two Hybrid (Y2H) technique, 21 host proteins were identified (Table 3). We tested the ability of those proteins to associate with ORF4. Only eukaryotic translation elongation factor 1 α isoform 1 (eEF1α1), Tubulin beta (Tubβ) and actin gamma isoform 1 were found to be common interaction partners of both RdRp and ORF4 (Table 3). Though eEF1α1 interacted with equal strength with both RdRp and ORF4, Tubβ and Actin gamma 1 weakly interacted with ORF4, compared to RdRp (Table 3, compare growth on 3-amino 1,2,4 triazole).
CoIP of Huh7 cells expressing RdRp and ORF4 demonstrated that both eEF1α1 and Tubβ associated with ORF4 and RdRp (Fig 9A). We could not detect actin gamma 1 association with ORF4 in CoIP (S4 Fig, top panel). Fraction of both eEF1α1 and Tubβ appeared to associate with ORF4-RdRp complex because both of them could be detected in samples subjected to two rounds of sequential immunoprecipitation (Fig 9A). Next, eEF1α1 and Tubβ proteins were ablated using shRNA to find out whether either or both bridged the interaction between ORF4 and RdRp. shRNAs were approximately 95% and 80% effective in reducing eEF1α1 and Tubβ protein, respectively (Fig 9B and 9C). CoIP revealed the inability of ORF4 to associate with RdRp in the absence of eEF1α1 though association of ORF4 and RdRp with Tubβ remained unaffected (Fig 9D). In contrast, ablation of Tubβ had no effect on the interaction between ORF4, RdRp and eEF1α1 (Fig 9D).
Next, an RdRp assay was conducted using purified RdRp-Flag from respective shRNA expressing cells. Stimulatory effect of ORF4 on RdRp activity was absent in samples lacking eEF1α1 (Fig 9E). Level of RdRp in Flag-affinity purified sample was verified by western and quantified to ensure that eEF1α1 or Tubβ ablation did not prevent RdRp translation (Fig 9F). Finally, we measured the level of sense and antisense RNA of wild type (W) and GAA mutant (G) HEV in DMSO or tunicamycin treated Huh7 cells expressing EGFP, heEF1α1 or hTubβ shRNA. Lack of eEF1α1 significantly reduced the level of both RNAs in DMSO and tunicamycin treated samples, confirming its essential role in g-1 HEV replication (Fig 9G) whereas absence of Tubβ had no effect.
Measurement of the level of ORF4 protein in the presence of proteasomal inhibitor MG132 and lysosomal acid protease inhibitor NH4Cl revealed its sensitivity to the former (Fig 10A). Degradation of ORF4 by proteasome was further evident from its polyubiquitination status (Fig 10B). ORF4 contains a lysine at 51st amino acid position flanked by two proline residues (hydrophobic amino acids), which represents a putative ubiquitination site. This Lysine was mutated to Asparagine (K51N mut ORF4). Monitoring the level of wild type and K51N mut ORF4 in the presence of cycloheximide (blocks de novo translation) revealed significantly higher stability of the mutant (Fig 10C), confirming that ORF4 is indeed a target of the proteasome.
The K51N substitution was introduced into g-1 HEV genome, followed by transfection of mutant genome into Huh7 cells. Immunofluorescence analysis revealed an increase in the number of ORF4 positive cells in the K51N mutant (K51N HEV, Fig 10D). Measurement of sense and antisense RNA in wild type and mutant genome transfected cells revealed higher level of both RNAs in the K51N mutant, indicative of enhanced replication of proteasome resistant ORF4 encoding genome, which was further increased upon tunicamycin treatment (Fig 10E).
Since a proteasome resistant ORF4 mutant genome could significantly enhance the viral replication, we wondered whether such mutations are prevalent in natural cases of g-1 HEV infection. Analysis of the ORF4 ubiquitination site in g-1 HEV sequences illustrated in Fig 1E revealed that 51st Lysine is absolutely conserved in all. However, in one case (AY204877.1), 50th and 52nd Proline residues were substituted with Serine and Leucine, respectively (Fig 1E). In 6 cases (JF443721.1- JF443726.1), 50th Proline was substituted with Leucine (Fig 1E). The above substitutions are supposed to prevent ubiquitination at 51st Lysine. Bioinformatics analysis of the above seven ORF4 protein sequences using “UbPred” software (predicts potential ubiquitination sites in a protein [40]) also indicated lack of ubiquitination at the 51st Lysine. Thus, viruses containing these sequences should produce a proteasome resistant ORF4 protein, similar to K51N mutation.
Despite lacking ORF4, g-3 HEV replicates better than g-1 virus in mammalian cell culture [12]. We hypothesised that some host protein(s) might be substituting the function of ORF4 in the g-3 virus, allowing it to bypass the dependency on ER stress dependent synthesis of ORF4. Host proteins identified as g-1 RdRp interaction partners (Table 3) were tested for their ability to associate with g-3 RdRp (Table 4). Only 14 out of 21 g-1 RdRp interaction partners associated with g-3 RdRp (Table 4), indicating that g-3 RdRp interaction profile is different from that of its g-1 counterpart. Therefore, it may interact with additional host proteins that did not interact with g-1 RdRp. We further tested the direct and indirect interactions of g-3 RdRp with other proteins of g-3 HEV by Y2H and CoIP assays. No intra-viral interaction partner of g-3 RdRp could be detected in Y2H assay, in agreement with the data obtained for g-1 RdRp (S2 Table). However, CoIP of Huh7 cell extract expressing both g-3 RdRp and g-3 X revealed that both of them coprecipitated with each other, indicating an interaction between them (Fig 11A). No interaction was observed between g-3 RdRp and helicase or g-3 X and helicase (Fig 11B and 11C). Next, a CoIP assay was performed to assess whether g-3 RdRp, X and helicase could assemble a complex. Indeed, all three could be coprecipitated (Fig 11D), indicating that they remain associated with each other. These findings also suggest that some host factor is essential for bridging the interaction between g-3 RdRp, X and helicase.
The current study attempts to address a long standing issue for researchers in HEV biology, pertaining to poor replication of g-1 HEV in cell culture. We show through multiple experiments that a previously unknown viral protein, which we have named ORF4, is essential for proper functioning of RdRp of g-1 HEV. Because ORF4 is synthesized only under condition of ER stress and it is a short-lived protein, replication of viral genome is inefficient in normal cells. Thus, it appears that ER stress, which is probably initiated as an antiviral response by the host, turns out to be the ideal cellular condition for optimal replication of g-1 HEV. Such a mechanism seems to be remarkably suited to the life of the virus, given that individuals under stress such as pregnant women display enhanced sensitivity towards HEV infection. These findings also suggest that a diverse range of diseases which induce hepatic ER stress, may sensitize individuals towards g-1 HEV infection. A study involving clinical assessment of hepatic stress, quantitation of ORF4 expression and viral titre in liver biopsy of different categories of g-1 HEV patients may unravel the correlation between ER stress, degree of ORF4 expression and disease severity.
Interestingly, ORF4 is encoded only by the g-1 HEV. Our bioinformatics analysis did not predict the presence of ORF4 in other HEV genotypes and experimental analysis of g-3 HEV replicon ruled out the possibility of ORF4 expression by g-3 HEV. Moreover, ER stress inducing compounds tunicamycin and thapsigargin did not have any effect on g-3 HEV replication. The above observations gave rise to two important questions: (a) Does ORF4 really play an important role during the natural course of g-1 HEV replication, (b) If ORF4 is indispensible for g-1 HEV replication, how do other genotypes of HEV replicate in its absence. To answer the first question, we analysed all available g-1 HEV genome sequences and not only observed the presence of ORF4 but also observed a very high level of conservation of ORF4 protein sequence among all g-1 HEV isolates (see Fig 1E). Though the C-terminal 19 amino acids of ORF4 were absent in ~50% of the viral genomes, our experimental data demonstrate that these 19 amino acids are dispensible for known functions of ORF4. The N-terminal 124 amino acids of ORF4 are sufficient for interacting with other viral and host proteins (see Fig 6E) and in vitro, it is able to stimulate RdRp activity just like the full length ORF4 (see Fig 8E). Therefore, ORF4 seems to be indispensable for g-1 HEV life cycle.
Since our study indicates that ORF4 most likely acts by interacting with multiple viral and host proteins to assemble a replication complex and promotes g-1 RdRp activity by interacting with host eEF1α1, in order to understand how other genotypes of HEV replicate in the absence of ORF4, we compared the protein interaction profile of g-1 RdRp with that of g-3 RdRp. Though g-1 and g-3 RdRp share ~85% homology at the amino acid level, only 14 out of 21 g-1 RdRp interacting host proteins could interact with g-3 RdRp. Moreover, g-3 RdRp interacted with g-3 X protein in Huh7 cells, indicating that some host protein(s) bridges that interaction. Further, g-3 RdRp, X and helicase assembled a complex, probably mediated by some host protein(s). G-3 X and g-3 helicase may display differential interaction profile than their g-1 counterparts, as observed in the case of g-3 RdRp. Thus, it is worth speculating that host proteins substitute the function of ORF4 in the case of g-3 HEV. Host protein interaction profile of g-3 RdRp, X and helicase needs to be established to identify the host proteins involved in assembling g-3 RdRp, X and helicase complex. Nevertheless, our data provides evidence for the acquisition of a regulatory system that enhances replication in g-1 HEV as it is not seen in any other genotype. ER stress independent constitutive assembly of the viral replication complex might account for the observed better replication efficiency of g-3 virus in mammalian cell culture.
Investigation of the mechanism(s) driving ORF4 synthesis revealed that it is independent of ORF1 translation, which is cap-dependent. Subsequent studies led to the discovery of a RNA regulatory element, which mediated cap-independent translation of ORF4. In vitro as well as in dual luciferase reporter assays, the HEV regulatory element functioned efficiently irrespective of thapsigargin and tunicamycin treatment. However, it was active only under conditions of ER stress in its natural location in the HEV genome, probably because the regulatory element is inaccessible or remains bound to inhibitory factors in the absence of ER stress. Several other viral and cellular IRESs are known to be active only under specific conditions. Notably, an IRES within Human immunodeficiency Virus-1 mediates viral structural protein synthesis during G2/M phase of cell cycle [27] and under conditions of oxidative stress [41]. Human cytomegalovirus latency protein pUL138 is translated by an IRES like element during serum stress [28]. A subset of cellular mRNAs such as c-Myc, Bip, Apaf-1, p53 and XIAP are translated through IRESs under conditions of stress, hypoxia and/or in a cell cycle dependent manner [42].
Though our data demonstrates the coexistence of both cap-dependent and cap-independent modes of translation in g-1 HEV, we have designated the RNA regulatory element as “IRES-like” element because it does not closely resemble other well known IRESs except for weak homology with the ERAV-1 IRES. Identification of IRES trans acting factors and detailed understanding of the mechanism of translation mediated by this element would confirm whether it is a bonafide IRES. Nonetheless, current data adds g-1 HEV to the list of RNA viruses where both cap-dependent and independent modes of translation coexist.
eEF1α1 ablation inhibited basal RdRp activity and antisense RNA synthesis whereas Tubβ ablation had no effect. Stimulatory effect of ORF4 on RdRp activity was also dependent on the level of eEF1α1, indicating crucial role of the latter in viral replication. Notably, eEF1α1 is important for replication and encapsidation of many plant and animal RNA viruses [43]. It binds to RdRp of Tobacco mosaic virus and silencing it inhibits infection [44]. eEF1α1 interacts with p33 protein of Tombus virus and this interaction is essential for viral antisense RNA synthesis [45]. eEF1α1 appears to have a similar role in g-1 HEV antisense RNA synthesis.
Our data demonstrates that ORF4 is degraded by the host proteasome. Early in life, virus focuses on replication and later on switches towards release of progeny. As lack of ORF4 dampens RdRp activity, it is possible that ORF4 performs two important functions in the life of g-1 HEV. At early phase, it promotes viral replication and later on, being a short lived protein, it acts as a regulatory switch to shift from replication to release. On the contrary, proteasomal degradation of ORF4 might also be an anti-viral strategy evolved in the host to restrict virus spread. Our limited analysis involving sequence analysis of 19 g-1 HEV isolates demonstrated conservation of the 51st Lysine. However, the ubiquitination site was lost in 7 sequences owing to alteration of 50th Proline to Leucine, suggesting that viruses in those patients produced a proteasome resistant ORF4. It is noteworthy that 5 out of the 7 sequences were isolated from fulminant hepatic failure (FHF) patients and 2 were acute viral hepatitis patients. However, considering the very limited number of samples, it might be a coincidence that majority of them represented FHF cases. Experimental analysis of replication efficiency of these 7 genomes will further substantiate the role of proteasome resistant ORF4 in g-1 HEV replication. Furthermore, an elaborate study involving more patient samples should be conducted to confirm the above observation. Analysis of correlation between disease severity and appearance of stabilizing (proteasome resistant) mutations in the ORF4 would further establish its role as a pro-viral factor. Identification of stabilizing mutations in ORF4 seems to have an important practical application in developing a more efficient model of HEV. HEV genome harbouring the ORF4 K51N mutation, which displays a high replication efficiency, might be expressed in cell lines stably expressing viral capsid and ORF3 protein to generate a robust system for producing HEV in the laboratory.
In conclusion, the present study provides yet another example of an opportunistic pathogen, which transforms the adversities imposed by the host towards its own benefit. Identification of ORF4 as an essential proviral factor, which is expressed only under conditions of ER stress, likely explains the inability of g-1 HEV to replicate efficiently in mammalian cell culture under standard laboratory condition. A proteasome resistant ORF4 harbouring HEV genome will be useful for establishing an efficient model of g-1 HEV. Our study also suggests that different HEV genotypes may have evolved different molecular mechanisms to exploit the host and successfully complete their life cycles.
HEV ORFs were PCR amplified from pSKHEV2 (genbank: AF444002.1) or pSK HEV p6 luc (genbank: JQ679013.1) plasmids and cloned into the required vectors following standard protocols [46]. HEV genomic RNA was in vitro synthesised, as described [13]; size and integrity was monitored by formaldehyde agarose gel electrophoresis. Huh7 human hepatoma cells were as described in Surjit et al. [4] and it was originally obtained from the laboratory of C.M. Rice [47]. HEK 293T cells were obtained from ATCC (USA). Cells were maintained in Dulbecco’s modified Eagle medium (DMEM) containing 10% Fetal Calf Serum (FCS), 50 I.U./mL Penicillin and Streptomycin, in 5% CO2. Cells were transfected using Lipofectamine 2000 or 3000, following manufacturer’s protocol (Life Technologies, USA) or electroporated. shRNAs were designed using Oligoengine 2.0 software for cloning into pSUPER puro vector, following manufacturer’s guidelines (Oligoengine, USA). Additional details in supplementary methods.
Antibodies against Flag, GAPDH, Myc, Ubiquitin and actin gamma were from Santa Cruz Biotechnology (USA). Antibodies against HA, eEF1α1, Tubulin β were from Sigma (USA). MG132, thapsigargin, tunicamycin, cycloheximide and NH4Cl were from Sigma (USA). Rabbit polyclonal antibodies against HEV ORF2, Helicase and ORF4 were synthesised at Genscript (USA) and validated in our lab (See supplementary methods). All chemicals were added 24 hours post transfection and maintained for 16 hours, or as indicated. Effective concentrations: MG132-25μM; cycloheximide-100μg/ml; thapsigargin-1μM; tunicamycin-10μg/ml; NH4Cl-30μM.
HEV infected, acute liver failure serum samples were obtained from patients registered in the liver clinic of Department of Gastroenterology, All India Institute of Medical Sciences, New Delhi, India. Serum was also collected from two healthy individuals with informed consent. A total 57 samples were tested for anti HEV IgM by ELISA and viral RNA with nested semi quantitative RT-PCR and quantitative real time PCR (primer sequences in S3 Table), respectively. Five samples were HEV IgM and g-1 HEV RNA positive. ORF4 coding and flanking region of these samples were sequenced and data submitted to genbank (ID: KU168733-KU168737). For testing cross-reactivity with purified ORF4 protein in western, serum was diluted 1:5000, followed by incubation with 1:5000 diluted goat anti-human IgG HRPO (Southern Biotech, USA). Viral genomic sequence of patient samples were aligned to HEV sequence (AF444002.1) for comparison.
ORF4 coding region nucleotide sequence of different HEV isolates was obtained from Genbank and translated into protein sequence using MacVector software. ClustalW alignment was done using MacVector. AF444002.1 sequence was considered as reference.
Total RNA was isolated using TRI reagent (MRC, USA), followed by reverse transcription (RT) and QRT-PCR, as described [48]. Random hexamers and HEVAS RP oligo were used in RT for detecting sense and antisense strands, respectively. Primer sequences are provided in S3 Table.
Done as described [4]. Goat anti rabbit alexa Fluor 488 (Molecular probes, USA) secondary antibody was used in IFA. Nucleus was stained with 4’ 6’- diamino-2-phenylindole (Antifade gold, Molecular probes). Images were acquired using a 60X objective in a confocal microscope (Olympus FV1000) and analyzed by Fluoview software. Details in supplementary methods.
A T7 polymerase based TNT kit (Promega, USA) was used for in vitro synthesis of proteins, following manufacturer’s instructions.
A GAL4 based system (Clontech, USA) was used following manufacturer’s instructions. Briefly, Y2H gold strain was transformed using lithium acetate with required BD and AD plasmids, followed by replica plating of 8 random transformants on different selection media to evaluate the activity of reporters. ORF3-TSG 101 interaction was used as a positive control [4].
For screening the Y2H cDNA library of human fetal brain, g-1 RdRp was cloned into pGBKT7 vector and its self activation potential was evaluated in Y2H gold strain (S4A Table). A mate and plate human fetal brain cDNA library (Clontech, USA) was used to screen the interaction partners of g-1 RdRp, following the instructions of the manufacturer. Mating condition and efficiency is mentioned (S4B Table). From evaluation of the diploids obtained after mating to identification of the bonafide interaction partners is summarized (S4C Table). All interactions were confirmed by retransformation of the prey and bait plasmids in pair along with appropriate negative controls (Table 3).
Dual luciferase reporter constructs (Firefly and Renilla, 1μg/well) were transfected into HEK 293T cells at 70% confluency in 48 well plate using lipofectamine 2000. Compounds were added for 16 hours, followed by luciferase assay using Dual Luciferase reporter assay kit (Promega, USA). Firefly luciferase values were divided by that of renilla and plotted. Gaussia luciferase was measured from culture medium using renilla luciferase assay kit (Promega, USA). Viability of same cells were measured using stable tetrazolium salt WST-1 (Roche, USA). Gaussia values were normalised to that of cell viability and plotted. Values are mean ± SEM of three independent experiments done in triplicate.
GST-ORF4 was expressed in E. Coli C-41(DE3) strain (0.1 mM IPTG, 18°C, 16 hours). Soluble protein was bound to Glutathione Sepharose beads, washed and eluted using 20mM glutathione. Eluted protein was Flag-affinity purified following manufacturer’s instructions (Sigma, USA). Final protein was recovered in PBS. Flag-tagged ORF2, helicase, RdRp and ORF4 were purified from Huh7 cells transiently expressing respective proteins by Flag-affinity purification. Silver staining was done using Pierce silver stain kit (Thermo Scientific, USA).
Glutathione Sepharose bound GST-ORF4 was mixed with equal amount of purified RdRp, X, Helicase and ORF3 in CoIP buffer [20mM Tris (pH 7.4), 150mM NaCl, 1mM EDTA (pH 8.0), 1mM EGTA (pH 8.0), 1% Triton X 100, 2.5mM Sodium Pyrophosphate, 1mM β glycerol phosphate, 1mM sodium orthovanadate, protease inhibitor cocktail] and rotated overnight at 4°C. Beads were washed thrice in same buffer, bound proteins eluted in 20mM glutathione, followed by western blotting using indicated antibodies.
Done as described [49] with the modification that the 16 base RNA oligo was labelled with 6 FAM (6-carboxyfluorecein). Additional details in supplementary methods.
Done as described [50] with the modification that DIG-II-UTP was used. Additional details in supplementary methods.
Data are presented as mean ± SEM of at least three independent experiments, analyzed using ‘‘GraphPad Prism” by the Student t test. p < 0.05 was considered significant.
Peripheral Blood samples were obtained from HEV infected and healthy adults with informed consent. Written consent was obtained from each individual. The study protocol was approved by the Ethics committee of All India Institute of Medical Sciences, New Delhi, India.
|
10.1371/journal.pcbi.1005396 | Metabolic regulation is sufficient for global and robust coordination of glucose uptake, catabolism, energy production and growth in Escherichia coli | The metabolism of microorganisms is regulated through two main mechanisms: changes of enzyme capacities as a consequence of gene expression modulation (“hierarchical control”) and changes of enzyme activities through metabolite-enzyme interactions. An increasing body of evidence indicates that hierarchical control is insufficient to explain metabolic behaviors, but the system-wide impact of metabolic regulation remains largely uncharacterized. To clarify its role, we developed and validated a detailed kinetic model of Escherichia coli central metabolism that links growth to environment. Metabolic control analyses confirm that the control is widely distributed across the network and highlight strong interconnections between all the pathways. Exploration of the model solution space reveals that several robust properties emerge from metabolic regulation, from the molecular level (e.g. homeostasis of total metabolite pool) to the overall cellular physiology (e.g. coordination of carbon uptake, catabolism, energy and redox production, and growth), while allowing a large degree of flexibility at most individual metabolic steps. These properties have important physiological implications for E. coli and significantly expand the self-regulating capacities of its metabolism.
| Metabolism is a fundamental biochemical process that enables cells to operate and grow by converting nutrients into ‘building blocks’ and energy. Metabolism happens through the work of enzymes, which are encoded by genes. Thus, genes and their regulation are often thought of controlling metabolism, somewhat at the top of a hierarchical control system. However, an increasing body of evidence indicates that metabolism plays an active role in the control of its own operation via a dense network of metabolite-enzyme interactions. The system-wide role of metabolic regulation is hard to dissect and so far remains largely uncharacterized. To better understand its role, we constructed a detailed kinetic model of the carbon and energy metabolism of the bacterium Escherichia coli, a model organism in Systems and Synthetic biology. Model simulations indicate that kinetic considerations of metabolism alone can explain data from hundreds of experiments, without needing to invoke regulation of gene expression. In particular, metabolic regulation is sufficient to coordinate carbon utilization, redox and energy production, and growth, while maintaining local flexibility at individual metabolic steps. These findings indicate that the self-regulating capacities of E. coli metabolism are far more significant than previously expected, and improve our understanding on how cells work.
| Metabolism is a fundamental biochemical process that converts nutrients into energy and biomass precursors, thus enabling cells to maintain their structures, grow, and respond to their environment. While the topology of metabolic networks is fairly well known, understanding how metabolic behaviours emerge from the dynamic interactions of their molecular components remains one of the main challenges faced by systems biology and is crucial for the development of synthetic biology [1].
The operation of metabolic networks, i.e. the metabolic fluxes, represents the ultimate output of several regulatory mechanisms. Metabolic fluxes are functions of enzyme activities and of the concentrations of reactants, products, and other effectors. While the enzyme activities are the ultimate outcome of gene expression through the hierarchy of transcriptional, post-transcriptional, translational and post-translational regulatory mechanisms, the reactant and effector concentrations are directly regulated at the metabolic level by enzyme activities themselves. Hierarchical regulation, and in particular transcriptional regulation, has attracted much attention because of mature experimental methods, but also because of early examples of flux increase with enzyme induction [2, 3]. These studies suggested an intuitive picture where fluxes mainly depend on enzyme concentrations, themselves mainly dependent on the level of transcript—a view that puts genes and their regulation at the top of a hierarchy of control and that regards metabolism as mostly a consequence of gene expression. An increasing body of evidence, however, indicates that this view of a hierarchical (or “dictatorial”) regulation of metabolism by gene expression is too simplistic. Large-scale 13C-flux analyses revealed that flux distributions in Saccharomyces cerevisiae and Escherichia coli are incredibly robust to the deletion of global transcriptional regulators [4, 5]. Integration of transcript and enzyme abundances with fluxes measured under different environmental conditions indicated that hierarchical regulation is insufficient to explain most of the flux reorganizations [6–9]. Therefore metabolism can no longer be seen as a passive process primarily regulated at the hierarchical level, but rather that it plays an active role in the control of its own operation via a dense network of metabolite-enzyme interactions. However, because hundreds of these interactions simultaneously regulate fluxes, which in turn affect metabolite levels, the system-wide role of metabolic regulation is hard to dissect and so far remains largely uncharacterized.
Mathematical frameworks such as metabolic control analysis [10, 11] were developed to analyze such complexity and improve our understanding of the role of each interaction on the metabolic network. These frameworks are particularly useful when applied to (validated) kinetic models that quantitatively describe the mechanistic interactions between the molecular species and their dynamics. However, developing models that are truly representative of real cell metabolism requires large amounts of experimental data to establish complex rate laws and identify parameters for each interaction [12]. Most models have focused on single pathways or on small sub-systems [e.g. 13, 14–18]. Such models accurately predict the response of those pathways to perturbations and reveal insights on the role of particular regulatory interactions on the metabolic operation [19–21]. Great progress has recently been made to develop larger scale kinetic models using top-down approaches [22–29], hence paving the way towards comprehensive understanding of the role of metabolic regulation at the whole cell level. These large-scale kinetic models highlighted the system-wide impact of local properties on the functioning of metabolic networks, such as an improved metabolic flexibility caused by enzyme saturation [26]. However, these large scale models are typically constructed from whole-genome metabolic reconstructions using generic rate laws, and contain a low level of mechanistic details (in particular are mostly devoid of allosteric regulation). An alternative approach constructed a highly detailed model of an entire cell of Mycoplasma genitalium [30], but unfortunately while this model represents a considerably high level of mechanistic detail in many cellular processes, it entirely lacks metabolic regulation (as it uses dynamic flux balance analysis rather than a mechanistic kinetic model). Hence, many of the properties that emerge from metabolic regulation are not captured by current large-scale models.
In this study, we aim at investigating the role of metabolic regulation on the central metabolic network of E. coli, which constitutes the backbone of its metabolism by providing macromolecular precursors, reducing equivalents, and energy for growth and maintenance. While previous studies typically focused on the role of particular regulatory interactions, we attempt to determine whether more global and generic properties arise from the interplay of the many regulatory interactions that compose metabolic regulation. To accomplish this, a kinetic model of E. coli central carbon and energy metabolism was developed and validated against a large set of existing experimental data. This model includes more mechanistic details than previous ones, and the impact of metabolic regulation on this system was analyzed using local and global methods.
The kinetic model developed in this study represents the central metabolism of Escherichia coli cultivated on glucose under aerobic conditions (Fig 1). This model contains 3 compartments (environment, periplasm and cytoplasm), 62 metabolites, and 68 reactions which represent the main central carbon and energy pathways of E. coli, namely: glucose phosphotransferase system (PTS), glycolysis and gluconeogenesis (EMP), pentose phosphate (PPP) and Entner-Doudoroff (EDP) pathways, anaplerotic reactions (AR), tricarboxylic acid cycle (TCA), glyoxylate shunt (GS), acetate metabolism (AC), nucleotide interconversion reactions (NC) and oxidative phosphorylation (OP). A reaction was also included to account for the consumption of metabolic precursors, reducing equivalents, and energy, and thus linking metabolism to cell proliferation. To account for metabolic regulation, a total of 255 metabolite-enzyme interactions (i.e. where metabolites modulate the reaction rates through thermodynamic or kinetic regulation, such as being substrates, products, allosteric modulators, or other type of inhibitors or activators) were included in the model, amongst which 34 are long-range regulatory interactions (i.e. where certain metabolites, which are not reactants, modulate the rates of these reactions).
Previously published kinetic models of E. coli metabolism were used as scaffolds to construct this model [18, 31, 32]. Both the number of pathways and the level of mechanistic detail were increased in the present model (S1 Table). For instance, this model now couples carbon metabolism with a detailed representation of oxidative phosphorylation, which makes it possible to balance the concentrations of cofactors (ATP/ADP/AMP, NAD(P)(H) and FAD(H2)) and simulate energy and redox metabolism. Previous models accounted for the consumption of metabolic precursors for growth in a decoupled way. This may be enough from the point of view of mass balance, but results in artifacts if used for an understanding of dynamics and regulation. In contrast the present model includes a single reaction to model growth, which ensures that the building blocks are consumed in stoichiometric proportions fixed by the cell composition, and not independently from each other. The rate of this reaction is a function of the intracellular concentrations of all the building blocks. This represents a significant improvement by satisfying the following growth rate properties: i) it monotonically increases with the availability of each building block, ii) it is asymptotically independent of each pool above a saturating concentration, and iii) it approaches zero if any pool approaches zero [21]. These properties were not reflected in the previous models [33].
The present model was calibrated to represent the metabolic state of E. coli cultivated under carbon limitation, a condition frequently experienced by this bacterium in laboratories, in industrial bioprocesses, and likely in its natural environment. To the extent possible, values of the biochemical parameters were taken from experimental determinations available in the literature. Parameters not available in the literature were estimated to reproduce steady-state and time-course experimental data obtained from a unique E. coli strain (the model strain K-12 MG1655) grown under a unique reference condition (M9 minimal medium with glucose as sole carbon source, dilution rate = 0.1 h-1, temperature = 37°C, pH = 7.0, pO2 > 20%) [13, 34–37]. This step is critical since both metabolite concentrations and fluxes depend on environmental conditions and differ between strains [38–40]. While results described below are largely in agreement with other experimental observations, the model was not forced to reproduce them, providing an important validation of the model. Detailed information on the construction and validation of the model is given in the Methods section and Supporting Information (S1 Text). The model is included in Supporting Information (S1 Model) formatted in SBML [41] and COPASI [42] formats, and is available from the BioModels database [43] with accession number MODEL1505110000.
The control properties of E. coli central metabolism in the reference state (see above) were investigated under the metabolic control analysis framework [10, 11]. Flux (CEJ) and concentration (CEM) control coefficients quantify the impact of a small change in the rate of each reaction (e.g. through change in the enzyme concentration E), on each flux (J) and each metabolite concentration (M). Since each metabolic step affects all fluxes and concentrations to some extent, we calculate a metric of its overall control on fluxes and concentrations as the L2 norm of all its flux- and concentration-control coefficients (see Methods), respectively.
The overall flux- and concentration-control by each step in the network is displayed in Fig 2. The main control point is the glucose inflow reaction with a control of 8.7 on fluxes and 5.3 on concentrations, using this summary metric. The system is therefore sensitive to its environment, as expected. Note that this is a direct sensitivity of metabolism to the environment, not through the (hierarchical) action of signal transduction and gene expression, which is not represented in this model; if it were its effect would thus be overlaid (likely with a delay) on the direct effect displayed in our model. Reactions that were identified by previous models as exerting a strong flux control under similar environmental conditions, such as the glucose phosphotransferase reactions or phosphofructokinase [13, 32, 44, 45], showed low control in our model (respectively 0.1 and 0.8). Rather, consistently with experimental evidence (see for example [46, 47–49]), the flux control was predicted to be shared between enzymes of all the pathways, amongst which cytochrome bo oxidase (reaction CYTBO, with 4.7 overall flux control), glucose-6-phosphate dehydrogenase (ZWF, 3.9), glyceraldehyde-3-phosphate dehydrogenase (GDH, 2.6), citrate synthase (GLT, 2.9) and the anabolic machinery (GROWTH, 1.4) are the ones with the largest share. A similar situation was observed for the control of concentrations, which is widely distributed across the network, and with the environment as the strongest control. In fact, a significant correlation (Pearson R = 0.86, P-value = 10−15) can be observed between the overall flux- and concentration-control exerted by each step (Fig 2C), indicating that, in general, enzymes which exert the strongest control on fluxes also exert the strongest control on concentration. A global sensitivity analysis [50] shows that these conclusions are robust with regard to parameter uncertainties (Fig 2A and 2B).
Further analysis confirmed the wide distribution of flux control across all the enzymes, with 97% of the individual control coefficients between -0.3 and 0.3 (Fig 2D). These observations, in agreement with the view that “rate-limitation” is distributed across the network and is variable [51], explain why fluxes are robust to moderate or even large changes of enzyme levels. The fluxes in this reference state are more sensitive to the environment, with 75% of the control coefficients exerted by the glucose supply reaction higher than 0.3 (Fig 2E). Similar conclusions were reached regarding the control of metabolite concentrations, which is distributed across the network (Fig 2B), with 97% of the control coefficients exerted by enzymes between -0.3 and 0.3 (Fig 2F) and 49% of the control coefficients exerted by the glucose supply reaction higher than 0.3 (Fig 2G).
Despite the low control exerted by enzymes over fluxes and concentrations at the network level, a detailed analysis of flux control coefficients reveals generic regulatory patterns between most of the pathways (Fig 3). A general observation is that the control of each pathway resides largely outside of itself.
For example, the control of the partition of carbon through competing pathways is shared between enzymes of each pathway. The glycolytic phosphofructokinase (PFK) exerts a (small) negative control on the PPP and ED fluxes (CPFKZWF=−0.15) and a positive control on the glycolytic flux (CPFKPGI=0.05), while the glucose-6-phosphate dehydrogenase (ZWF) of the PPP and ED pathways exerts a strong positive control on its own flux (CZWFZWF=0.75) and a negative control on the glycolytic flux (CZWFPFK=−0.12). Similar behavior is observed at the main metabolic branch nodes, e.g. between the TCA cycle and the glyoxylate shunt or between the pentose phosphate and Entner-Doudoroff pathways. It is important to note that the fraction of flux diverted to each branch does not depend only on the local enzyme kinetics, contrary to what is sometimes suggested [6], but on several enzymes of each of the competing pathways.
Several feedforward and feedback interactions are also observed between the pathways. For instance, the pyruvate kinase (PYK) controls fluxes through the TCA cycle (CPYKSDH=0.07), and is controlled by some TCA reactions (CLPDPYK=0.06, CSDHPYK=0.19). Similarly, the ATP demand (that can be represented by the ATP utilization for maintenance, ATP_NGAM) is activated by glycolysis (CGDHATP_NGAM=0.26) and exerts in turn a positive feedback control on this pathway (CATP_NGAMPGI=0.12) and a negative control on growth (CATP_NGAMGROWTH=−0.09), as observed in vivo [72]. Interestingly, biomass synthesis (GROWTH) is strongly controlled by the upstream glucose supply (CGLC_INFLOWGROWTH=1.5), with all other control coefficients lower than 0.13. In turn, biomass synthesis exerts a small but global feedback control on most catabolic fluxes.
Those several, intertwined feedback and feedforward interactions stress the high degree of functional organization of the central carbon and energy metabolism. This may be an important feature to maintain the coordination between the different pathways at the cellular level: if the rate of a particular reaction—be it upstream or downstream—is affected by a perturbation, this information will be transmitted from this “sensor reaction” to the entire system, resulting in a global response. Note that this response is sensed in a very short time scale, rather than the slower response that happens after signal transduction and consequent changes in gene expression.
To get a broader picture of the role of metabolic regulation on the coordination of E. coli metabolism, the solution space of this network was explored with and without considering metabolic regulation. Two versions of the model were used: the kinetic version which accounts for metabolic regulation, and a stoichiometric version of the same model which contains only stoichiometric constraints (and is thus similar to a flux balance analysis model). The solution space of each model was explored using a random sampling approach: 600,000 flux distributions were uniformly sampled from the solution space using the stoichiometric model, and steady-states were simulated for 600,000 sets of random enzyme levels using the kinetic model. For each set, enzyme levels (i.e. Vmax) were sampled from a log uniform distribution (between 0.1 and 10 times the enzyme levels of the initial model) to ensure each order of magnitude to be sampled in similar proportions. It is important to mention that cells do not express enzymes levels according to the distribution generated, therefore the distribution of the variables is not expected to provide any information on the probability for a cell to reach a specific state in vivo [50]. Rather, uniformity is used to clearly grasp the functional implications of applying metabolic regulation to the network.
We first investigated the relationship between supply (glucose uptake) and demand (growth), which provides information on the allocation of resources by the metabolic network [52]. Direct sampling of the solution space (Fig 4A) revealed that most of the metabolic states are not efficient in term of resource allocation: most of them correspond to a high glucose uptake rate, but with a low growth rate, because this situation significantly increases the attainable intracellular flux states. Interestingly, the opposite picture is observed when metabolic regulation is applied on this network (Fig 4B): a smaller region of the solution space is reached, where the growth rate is now coupled to the glucose uptake rate.
To evaluate this prediction quantitatively, we gathered from the literature experimental data obtained from 254 growth experiments carried out under similar environmental conditions (glucose as sole carbon source in aerobic conditions) [5, 27, 34, 53–67] (S2 Dataset). These data were collected from “wild-type” and mutant strains obtained by deletion or overexpression of central metabolic enzyme genes or global regulators of gene expression, and cultivated under a wide range of experimental conditions. Therefore, these 254 data represent a very broad range of the metabolic states that can be expressed by E. coli growing on glucose. Importantly, these data were not used for parameter estimation, thereby they constitute an independent validation and provide a robust assessment of the predictive ability of the model. These experimental observations correspond to the region of the solution space less frequently sampled using the stoichiometric model, but they closely match the region sampled by the kinetic model (Fig 4B). This means that the observed physiology of E. coli is closer to the metabolic model that is regulated by metabolite-enzyme interactions (the kinetic model) than it is to a metabolic model that would be regulated by gene expression alone (the stoichiometric model). Hence, metabolic regulation alone, without needing to invoke coordinated expression of genes, seems to be sufficient to explain the emergence of a coupling between anabolic (growth) and catabolic (glucose uptake) fluxes, and thereby appears to be a major determinant of the overall cellular physiology by ensuring an efficient and robust allocation of nutrients towards growth.
We extended the above analysis to determine whether additional couplings emerge from metabolic regulation. Several variables representative of the physiological state of E. coli were calculated for each steady-state reached by the kinetic model, namely: growth and glucose uptake rates, ATP, NADH and NADPH production rates, sum of all intracellular fluxes, sum of all intracellular metabolite concentrations and cost of enzymes (defined as the product of enzyme concentration and number of amino acids of the corresponding enzyme, summed over all reactions, as detailed in Methods). Additional variables derived thereof were also computed: biomass, ATP, NADH and NADPH yields, enzyme cost and ATP production rate per sum of fluxes, and sum of fluxes per glucose consumed. Pairwise relationships between systemic variables and (absolute and relative) fluxes through the main pathways were examined using Spearman correlation and mutual information. The outcome is a correlation matrix which maps the degree of functional coupling between all the variables (Fig 5A). The same patterns were highlighted by both methods, which indicate that these couplings are monotonic since mutual information, but not Spearman correlation, would identify non-monotonic relations.
A systematic positive correlation was predicted between some anabolic and catabolic fluxes and yields: glucose uptake rate, growth rate, NADPH production rate, biomass yield and NADPH yield (with ρ > 0.85), indicating a considerable degree of coordination in the metabolic operation. These variables are negatively correlated with the energetic (ATP and NADH) yields (ρ < -0.85), which is consistent with the fact that the single carbon source, glucose, is used by two competing metabolic processes: energy production and biomass synthesis, reflected in an increase in YATP and increase in biomass yield, respectively. The oxygen uptake rate also correlated positively with ATP and NADH production (with ρ > 0.90), reflecting the important role of oxidative phosphorylation in energy production under aerobic conditions.
The low correlation coefficient between the sum of fluxes and the cost of enzymes (ρ = 0.11) indicates that the sum of fluxes cannot be considered as a proxy for enzyme investment per se. The outcome of predictive analyses based on this assumption (such as the minimization of the sum of fluxes in FBA according to the hypothesis that cells minimize their enzyme levels) should therefore be interpreted with caution.
In general, systemic variables correlated poorly with relative and absolute fluxes from most of the pathways. This is interesting as it shows that while there is coordination between several processes, there is nevertheless a significant degree of flexibility in the intracellular flux distribution. A notable exception was observed for the TCA cycle: its absolute flux is positively correlated with catabolic and energy fluxes (vO2_uptake, vGlc_uptake, vNADH_production, vATP_production, with ρ > 0.84), and the relative contribution of this pathway negatively correlated with anabolic rates and yields (growth and NADPH production rates, ρ < -0.92) and positively correlated with energy yields (ρ > 0.90 for ATP and NADH yields). Thus, the partition of carbon between energy production (ATP and NADPH) and growth (via the synthesis of many anabolic precursors) is predicted to be realized primarily at the level of the TCA cycle and appears to be largely controlled at the metabolic level.
To evaluate these model predictions, additional experimental data on extracellular and intracellular fluxes (growth rate, glucose and oxygen uptake rates, and TCA cycle fluxes through the citrate synthase) were collected from the literature [5, 27, 34, 53, 54, 56, 58–60, 62–65, 67] (S2 Dataset). These data, which were not used to calibrate the model, covered the particular regions highlighted by the kinetic model (Fig 5B–5D). The excellent agreement between the spread of simulated and experimental data strongly supports the existence of the functional couplings predicted by the model. It is important to mention that these couplings are not caused by stoichiometric constraints since they are not observed when the solution space is uniformly sampled using the stoichiometric model (Fig 5E–5G). The results also show that the coordination of gene expression by hierarchical regulatory mechanisms is not an important factor in these couplings since they are still maintained when enzyme levels are changed randomly. In contrast, metabolic regulation brought about by metabolite-enzyme interactions is sufficient to explain their emergence; therefore they represent intrinsic properties of the central metabolism of E. coli.
Interestingly, additional couplings predicted by the model were recently observed in vivo in both prokaryotic (E. coli) and eukaryotic (S. cerevisiae) microorganisms [73]: between the ATP and NADH production rates (ρ = 0.93, Fig 6A), between the sum of fluxes per glucose uptake rate and the ATP yield (ρ = 0.71, Fig 6B), and between the growth rate per sum of fluxes and the sum of fluxes per glucose uptake rate (ρ = -0.85, Fig 6C). Since the central metabolic networks of E. coli and S. cerevisiae are highly conserved, the present results may explain why similar properties are observed in both microorganisms, though this hypothesis requires further investigation.
The results presented above support the view that metabolic regulation reduces the solution space defined by the stoichiometric constraints, as previously suggested [68, 69]. However, the very low probability regions of the solution space might not be captured by random sampling approaches [50]. To test further if metabolic regulation actually shrinks the solution space of E. coli central metabolism, its boundaries were determined with and without considering metabolic regulation by using the kinetic and the stoichiometric models, respectively. Unexpectedly, the boundaries were similar for both models (Fig 7). This indicates that metabolic regulation does not shrink the solution space of the system—and thus does not restrict the metabolic capabilities of E. coli–, at least for the variables considered here. Thus we have to conclude that evolution must have “discovered” this region of parameter space which gives selective advantage to the organism.
It has been shown that metabolic regulation plays an important role in metabolite homeostasis, which prevents osmotic stress and disadvantageous spontaneous reactions by avoiding large changes in metabolite concentrations (for example see [20, 70]). Interestingly, we noticed that, for 86% of the steady-states reached by the kinetic model (when the enzyme levels were chosen at random), changes in total concentration of metabolites are lower than three-fold relative to the calibrated state (47 mM) (Fig 8), but the changes in fluxes are several orders of magnitude higher. This narrow range of predicted intracellular concentrations is physiologically relevant [63, 71]. Since no constraints on metabolite concentrations were included in the model, we conclude that metabolic regulation alone may explain global metabolite homeostasis, while still allowing significant changes in fluxes.
In this study, we investigated the contribution of metabolic regulation on the operation of the central metabolism of E. coli, which provides building blocks, cofactors, and energy for growth and maintenance. We developed, to our knowledge, the first detailed kinetic model of this system that links metabolism to environment and cell proliferation through intracellular metabolites levels. This model, validated by 778 independent flux data from some 266 experiments, allowed the identification of several properties which emerge from metabolic regulation and explain many experimental observations of E. coli’s physiology. The intrinsic, self-regulating capacities of E. coli central metabolism appear to be far more significant than previously expected. The results presented here imply that gene regulation is not required to explain these properties.
Metabolic control analysis showed that the flux and concentration control exerted by single enzymes is low and largely distributed across the network, confirming again the insights of Kacser and Burns [51]. This significantly contrasts with the outcome of previous kinetic models [13, 32, 44, 45], where a few enzymes were predicted to exert most of the flux control, but is in line with much experimental evidence [7, 21, 46, 47, 72]. Our results therefore support the view that the concept of “rate-limiting” steps does not apply to E. coli metabolism, and likely not to the metabolism of other organisms. Its persistence in the literature is a major handicap to understanding metabolism. In fact, the central metabolism is not even self-contained in terms of control due to a large portion of control being exerted by the environment, making E. coli responsive to environmental changes. One of the most striking examples of this phenomenon is manifested in growth controlling most fluxes but being controlled virtually by glucose availability alone. The low control exerted by single enzymes on the system makes the metabolic operation of E. coli robust to fluctuations of enzyme levels that may arise from noise in gene expression or other factors. Moreover, the majority of control resides not within but outside the controlled pathways. The dense, yet highly organized, interactions between pathways allow a rapid and coordinated response of the entire system to perturbations.
Exploration of the solution space indicated that metabolic regulation does not significantly restrict the metabolic capabilities of E. coli, as was previously believed [68, 69]. While the observed behavior of many different E. coli strains and mutants are confined to a small region of the solution space, this is not due to kinetic constraints as it is possible to simulate other behaviors simply by changing parameter values. This apparent paradox can be resolved, of course, if the action of natural selection had favored these behaviors.
The systematic mapping of the relationships between various systemic variables revealed that metabolic regulation is sufficient to explain the emergence of several functional couplings, which are independent from gene regulation (since they are conserved when enzyme levels are changed randomly by orders of magnitude) and cannot be explained by stoichiometric constraints. An important finding is that metabolic regulation alone may be responsible for the coordination of major catabolic, energetic and anabolic processes at the cellular level to optimize growth. Metabolic regulation thus appears to be sufficient to maintain multi-dimensional optimality of E. coli metabolism [65]. Despite this overall coordination, there is a large degree of flexibility at most individual metabolic steps. The role of metabolic regulation in maintaining global homeostasis of intracellular metabolite pools under a broad range of flux states was also verified by the present model. The modeling results were in excellent agreement with experimental data, even quantitatively. E. coli metabolism displays remarkably robust yet simple emergent properties, and these properties have major implications on its overall cellular physiology, e.g. by preventing unnecessary osmotic stress, maintaining the coordination between key processes, and optimizing the allocation of resources towards particular functions such as growth.
The self-regulating capabilities of E. coli central metabolism reflect the evolutionary selection that has been exerted on the ensemble of enzymes (in terms of kinetic and regulatory properties, but not necessarily of expression levels) to realize a network with these properties. Since central metabolism is essential in most organisms and is highly conserved across the three domains of life, it is tempting to speculate that metabolic regulation is responsible for the very similar operation principles observed in different organisms [73].
Of course we do not suggest that hierarchical regulation does not play an important role in the metabolic operation of E. coli, but it is in addition to the properties observed here, since these can operate without it. For instance, the robustness of the flux partition to the deletion of global transcriptional regulators was interpreted as a low control of this partition at the hierarchical level [5], and our results confirm that this robustness lies, to some extent, in metabolic regulation, given the low control exerted by enzymes. However, this conclusion is valid only for moderate changes of enzyme levels (with the notable exception of the flux through the TCA cycle), and other mechanisms (such as hierarchical regulation) are required to explain the robust flux partition. Expanding the kinetic model to incorporate regulation of gene expression will be needed ultimately to understand the interplay between these two regulatory levels [19, 74, 75, 88].
The kinetic model of the central carbon and energy metabolism of Escherichia coli K-12 MG1655 (Fig 1) was developed with the software COPASI (build 45) [42]. This model is briefly described in this section, and additional information can be found in Supporting Information (S1 Text). The model is available in SBML and COPASI formats in Supporting Information (S1 Model), as well as from the Biomodels database [43] with identifier MODEL1505110000.
Values of 56% of the parameters (253 on a total of 449) were directly taken from the literature. Parameters not available in the literature, which do not have a real biochemical estimate (e.g. Michaelis constants of the biomass function), or for which biochemical measurements are generally not representative of intracellular conditions (e.g. Vmax) were estimated to reproduce in the best possible way 276 experimental data obtained from E. coli K-12 MG1655 grown on glucose, under aerobic condition, at a dilution rate of 0.1 h-1. These data were steady state fluxes and metabolite concentrations [13, 34, 36, 37, 82] and time-course concentrations of intracellular metabolites in response to a glucose pulse [35] (S1 Dataset). Parameter estimation was formulated as a constrained optimization problem:
minimize f(p)
subject to g(p)≥c
where p is the parameter vector, f is the objective function which evaluates the deviation between the simulated and measured data, g(p) is the constraint function, and c is the constraint vector. The objective function f was defined as the sum of squared weighted errors:
f(p)=∑i(xi−yi(p)σi)2
where xi is the experimental value of the data point i, with experimental standard deviation σi, and yi(p) is the corresponding simulated value. Constraints were defined on estimated parameters (10−4 mM ≤ KM ≤ 103 mM; 10−2 mM/s ≤ Vmax ≤ 103 mM/s; 10−4 ≤ Keq ≤ 106) to ensure they are kept within a biologically reasonable range. The objective function was minimized with the Particle Swarm Optimization algorithm [83], using the Condor-COPASI system [84] on a pool of 2500 CPU cores. The experimental and fitted data are provided in Supporting Information (S1 Dataset). Values of all the parameters (and the corresponding references for those values taken from the literature) are given in Supporting Information (S1 Text).
Analyses described below were performed using R (v3.0, www.r-project.org) after converting the model into Fortran. All the scripts are provided in Supporting Information (S1 Code).
|
10.1371/journal.ppat.1005248 | An O-Methyltransferase Is Required for Infection of Tick Cells by Anaplasma phagocytophilum | Anaplasma phagocytophilum, the causative agent of Human Granulocytic Anaplasmosis (HGA), is an obligately intracellular α-proteobacterium that is transmitted by Ixodes spp ticks. However, the pathogen is not transovarially transmitted between tick generations and therefore needs to survive in both a mammalian host and the arthropod vector to complete its life cycle. To adapt to different environments, pathogens rely on differential gene expression as well as the modification of proteins and other molecules. Random transposon mutagenesis of A. phagocytophilum resulted in an insertion within the coding region of an o-methyltransferase (omt) family 3 gene. In wild-type bacteria, expression of omt was up-regulated during binding to tick cells (ISE6) at 2 hr post-inoculation, but nearly absent by 4 hr p.i. Gene disruption reduced bacterial binding to ISE6 cells, and the mutant bacteria that were able to enter the cells were arrested in their replication and development. Analyses of the proteomes of wild-type versus mutant bacteria during binding to ISE6 cells identified Major Surface Protein 4 (Msp4), but also hypothetical protein APH_0406, as the most differentially methylated. Importantly, two glutamic acid residues (the targets of the OMT) were methyl-modified in wild-type Msp4, whereas a single asparagine (not a target of the OMT) was methylated in APH_0406. In vitro methylation assays demonstrated that recombinant OMT specifically methylated Msp4. Towards a greater understanding of the overall structure and catalytic activity of the OMT, we solved the apo (PDB_ID:4OA8), the S-adenosine homocystein-bound (PDB_ID:4OA5), the SAH-Mn2+ bound (PDB_ID:4PCA), and SAM- Mn2+ bound (PDB_ID:4PCL) X-ray crystal structures of the enzyme. Here, we characterized a mutation in A. phagocytophilum that affected the ability of the bacteria to productively infect cells from its natural vector. Nevertheless, due to the lack of complementation, we cannot rule out secondary mutations.
| Since its discovery in 1994, Human Granulocytic Anaplasmosis (HGA) has become the second most commonly diagnosed tick-borne disease in the US, and it is gaining importance in several countries in Europe. HGA is caused by Anaplasma phagocytophilum, a bacterium transmitted by black-legged ticks and their relatives. Whereas several of the molecules and processes leading to infection of human cells have been identified, little is known about their counterparts in the tick. We analyzed the effects of a mutation in a gene encoding an o-methyltransferase that is involved in methylation of an outer membrane protein. The mutation of the OMT appears to be important for the ability of A. phagocytophilum to adhere to, invade, and replicate in tick cells. Several tests including binding assays, microscopic analysis of the infection cycle within tick cells, gene expression assays, and biochemical assays using recombinant OMT strongly suggested that the mutation of the o-methyltransferase gene arrested the growth and development of this bacterium within tick cells. Proteomic analyses identified several possible OMT substrates, and in vitro methylation assays using recombinant o-methyltransferase identified an outer membrane protein, Msp4, as a specifically methyl-modified target. Our results indicated that methylation was important for infection of tick cells by A. phagocytophilum, and suggested possible strategies to block transmission of this emerging pathogen. The solved crystal structure of the o-methyltransferase will further stimulate the search for small molecule inhibitors that could break the tick transmission cycle of A. phagocytophilum in nature.
| Anaplasma phagocytophilum is an obligately intracellular bacterium classified in the order Rickettsiales, and is the causative agent of Human Granulocytic Anaplasmosis (HGA) [1]. HGA is characterized by high fevers, rigors, generalized myalgias, and severe headache. It is a potentially life-threatening disease, with 36% of patients diagnosed with HGA requiring hospitalization, 7% needing urgent care, and mortality of ~1% [2]. The incidence of HGA has been increasing steadily, from 348 identified cases in 2000 when it first became reportable to the CDC, to 1761 cases in 2010 [3] and 2,782 reported cases in 2013 [4]. Similar trends are evident in other countries in Europe and Asia [reviewed in [5]]. In addition, A. phagocytophilum infects domestic animals such as dogs, cats, and horses, as well as wild mammals from deer and wolves to various rodents [6].
A. phagocytophilum is transmitted by ticks of the Ixodes ricinus complex, with Ixodes scapularis and Ixodes pacificus being the most important vectors in the USA [7]. Transovarial transmission does not occur in these ticks, and has only been reported in tick species and A. phagocytophilum strains that are not implicated in human disease [8]. The natural transmission cycle involves acquisition of the pathogen from small wild rodents by tick larvae, and transstadial transmission to nymphs and adults that may infect a new mammalian host during a subsequent bloodmeal. Therefore, the ability of A. phagocytophilum to cycle between ticks and mammalian hosts is imperative for bacterial survival in nature [8]. The development of A. phagocytophilum in Ixodes sp. vector ticks remains unknown but has been described in tick cell culture where it is biphasic [9]. The time required for A. phagocytophilum to complete development in ISE6 cells differs from that observed in HL-60 and endothelial cells [9]. Adhesion to ISE6 cells started at 30 min p.i., and by 1 hr p.i the bacteria were attached to the tick cell membrane, initiating the process of endocytosis, which was probably driven by receptor-mediated interactions. By comparison, in HL-60 cell culture, >70% of bacteria were observed binding to host cells in the first 40 min p.i., and at that time, 26% of the bacteria had been internalized [10]. Internalization in tick cells began at 2 hr p.i. and was complete by 4 hr p.i., whereas in HL-60 cell culture, only 45.5% of the bacteria had entered the cells at this time point [10]. Replication by binary fission started by 8 hr p.i. in tick cells [9] whereas in HL-60 cells only a few bacteria had turned into the reticulate form by 12 hr p.i. and initiated replication [10]. Nevertheless, many of the molecular events involved in the infection of mammalian cells are known [11], but much less is known in the tick counterpart [12]. Studies to understand vector-pathogen interactions have focused on tick responses and tick factors important for successful establishment of the pathogen in ticks [13–15], but A. phagocytophilum genes and proteins that are important for development in tick cells and ticks remain largely unidentified. Some studies have examined gene expression of A. phagocytophilum during infection of I. scapularis ticks or I. scapularis ISE6 cells, but have focused on certain periods, such as transmission feeding or late phases of replication [16,17]. As a result, little is known about the proteins, and their modifications, necessary for the early phases of infection of tick vector cells by A. phagocytophilum. Survival in dissimilar hosts such as the arthropod vector and the mammal that present important biological differences requires rapid adaptation of bacteria, and involves proteins and other molecules that are differentially expressed or produced in response to host-specific cues [17]. Thus, the identification of such factors is crucial to our understanding of the biology of this important pathogen.
Analyses of A. phagocytophilum gene expression and proteomics [16] may fail to identify proteins that are not abundant or not directly involved in infection but still play an important role. The intracellular nature of A. phagocytophilum has made it difficult to study the function of genes involved in intracellular invasion and replication using genetic techniques such as homologous recombination. Nevertheless, random mutagenesis of A. phagocytophilum using the Himar1 transposase system [18] has become an important tool to probe gene function in these and related bacteria [19–21]. Here, we analyzed a mutant, referred to as ΔOMT, with a transposition into genomic locus APH_0584 that contains a gene encoding a member of family 3 S-adenosyl methionine (AdoMet or SAM)-dependent o-methyltransferases. Transcription of genomic locus APH_0584 was barely detected in HL-60, HMEC-1 or ISE6 cells during late phases of infection [17]. However, the mutation of this gene rendered the bacteria unable to efficiently colonize I. scapularis (ISE6) cells.
Methyltransferases are involved in important bacterial activities such as cell signaling, cell invasion, and gene expression, as well as in metabolic pathways and pathogenesis [22–24]. They participate in the modification of membrane components, cofactors, signaling and defense compounds [23], and have been linked to virulence in several bacteria [25–27], fungi [28], and viruses [29]. OmpB proteins from several rickettsial pathogens are methylated at multiple residues by lysine methyltransferases [30], although recombinant OmpB produced by E. coli in the absence of a lysine methyltransferase has been shown to mediate adhesion and invasion of HeLa cells in a Ku70-dependent manner [31]. Methylation of glutamic acid residues in the outer membrane protein OmpL32 of Leptospira interrogans is thought to be involved in its virulence and ability to colonize liver and kidney cells in hamsters [32]. To gain insights into its overall structure and the interaction of the A. phagocytophilum o-methyltransferase with cofactors, we solved the crystal structure of the apo-enzyme, the enzyme bound to S-adenosine homocysteine (SAH), to SAH and manganese, and to SAM and manganese. This revealed large differences with the nearest homolog in the PDB (o-methyltransferase from the cyanobacterium Synechocystis sp.; PDB ID: 3CBG). Here, we analyzed the phenotypic and proteomic changes that characterized ΔOMT, and present evidence that the o-methyltransferase is involved in adherence to and necessary for replication of A. phagocytophilum in tick cells.
The ΔOMT, selected and maintained in HL-60 cells, was unable to grow in ISE6 cells. The mutant expressed the Green Fluorescent Protein (GFPuv) from a Himar1 transposon [18] and Southern blot analysis identified a single insertion site (Fig 1A). Digestion of ΔOMT DNA with BglII yielded a single band hybridizing to the probe, suggesting a clonal population (Fig 1A), although EcoRV yielded several smaller bands that were most likely due to incomplete digestion. Recovery of the transposon along with flanking sequences from ΔOMT DNA by restriction enzyme digestion and cloning indicated transposition into aph_0584 (Gene ID: 3930223; o-methyltransferase 3 family member) between nucleotide positions 612707–612706 of the A. phagocytophilum strain HZ genome sequence ([33]; Fig 1B). The single insertion event suggested that the changes in phenotype were due to the disruption of that particular gene.
To determine the mechanism whereby the mutation affected the phenotype of A. phagocytophilum, we compared wild-type and mutant bacteria with respect to their ability to invade and replicate in tick and mammalian host cells, determined the timing of wild-type OMT expression and its localization, identified the protein methylated by the enzyme as well as cofactors, and solved the crystal structure of the OMT. First, we analyzed the growth of the mutant in HL-60 and ISE6 cells and compared it to wild-type bacteria under the same conditions. The ΔOMT was not able to replicate in ISE6 cells and qPCR showed that msp5 (a single copy gene used as a proxy for bacterial numbers) copy numbers decreased significantly over a 12-day period (P = 0.008) (Fig 2). This was in contrast to the behavior of ΔOMT bacteria in HL-60 cells, in which they were able to multiply in a manner comparable to wild-type bacteria (Fig 2, P = 0.504). Only the datasets for the 1:16 dilution in ISE6 and 1:100 in HL-60 are shown, but other dilutions presented the same tendency.
Because of the rapid decline of ΔOMT numbers noticeable already during the first 24 hr of incubation with ISE6 cells, we tested the ability of the mutant to bind to ISE6. There was a significant reduction (>50%) in binding of ΔOMT (t-value = -4.1011; P = 0.0001) to ISE6 cells from 0.3 bacteria per cell in the wild type to 0.12 mutant bacteria per cell (Fig 3A). To support these results, an inhibitor of SAM-dependent methyltransferases was used to reproduce the effects of the lack of methylation brought about by the disruption of omt on binding of A. phagocytophilum to ISE6 cells. Wild-type bacteria were pre-incubated with 20 nM, 30 nM, and 40 nM of adenosine periodate (AdOx) for 1 hr before addition to ISE6 cells and incubated for another hr; untreated bacteria served as controls. All concentrations of AdOx affected the ability of A. phagocytophilum to bind to ISE6 cells significantly (P<0.001) (Fig 3B). In controls, an average of 0.529 bacteria bound per cell, compared to ΔOMT with only 0.156 bacteria per cell. 20 nM AdOx decreased binding to 0.307 bacteria per cell, which was less than the reduction in attachment observed in ΔOMT (Fig 3B). However, as the concentration of AdOx increased to 30 nM and 40 nM, the effects were stronger than in ΔOMT with only 0.093 and 0.086 bacteria bound per cell, respectively, and these differences were statistically significant when compared to the ΔOMT and the 20 nM AdOx concentration (Fig 3B). These results corroborated the effects of the inhibition of methylation due to the mutation of omt on A. phagocytophilum binding. The greater inhibition of bacterial binding when using 30 nM and 40 nM of AdOx was probably due to inhibition of other methyltransferases. To test this hypothesis, we pre-incubated ΔOMT bacteria with AdOx before addition to ISE6 cells, as described for wild-type bacteria. Incubation of the ΔOMT with 20 nM and 40 nM did not significantly decrease binding to ISE6 cells (S1 Fig), suggesting that no other methyltransferases were involved in tick cell invasion.
Because of the effects on binding to tick cells seen in ΔOMT, we tested the expression of the omt gene by qRT-PCR during early stages of wild-type A. phagocytophilum interaction with and development in ISE6 cells, using rpoB and msp5 genes as normalizers. In our discussion, we focused on the fold change of omt normalized to msp5, but normalization against either gene showed the same trend (Fig 4). Up-regulation of the omt gene started at 30 min post-inoculation (p.i.), continued to increase from 3-fold at 30 min to 5-fold by 1 hr p.i., and by 2 hr, the gene reached its maximum expression, showing 34-fold up-regulation compared to bacteria entering HL-60 cells (Fig 4). At 4 hr, omt expression decreased to 0.97-fold change (Fig 4), similar to that seen in bacteria infecting HL-60 cells. Our results are congruent with electron microscopy based studies that showed A. phagocytophilum bound to tick cells between 30 min and 1 hr p.i. and cell entry at 2 hr p.i., the time when we saw maximum omt gene expression [9].
Because omt expression correlated with binding and entry of the bacteria to ISE6 cells, and the mutation of this gene affected the ability of the bacteria to bind to these cells, we investigated the localization of the protein during this step in cell infection. Mouse antiserum against recombinant OMT (rOMT) was produced to label the protein during binding of wild-type A. phagocytophilum to ISE6 (2 hr p.i.), using an immunofluorescence assay (IFA). OMT was detected with mouse anti-rOMT serum followed by incubation with anti-mouse IgG conjugated to AlexaFluor647 (red fluorescence). All bacteria were labeled with dog anti-anaplasma serum followed by incubation with fluorescein isothiocyanate (FITC)-conjugated anti-dog IgG (green fluorescence). Bacteria interacting with ISE6 showed strong OMT expression while bacteria interacting with HL-60 showed only slight expression (Fig 5). This was in agreement with the 34-fold up-regulation of the gene seen by qRT-PCR during adhesion to ISE6 cells (Fig 4). Bacteria incubated with pre-immune serum did not fluoresce red nor did uninfected ISE6 cells incubated with anti-rOMT antibodies followed by TRITC-conjugated anti-mouse IgG, demonstrating that the serum specifically labeled OMT.
We conducted a time course comparison of wild-type versus ΔOMT bacteria to identify the stage at which the infection process failed. To increase the chances of detecting differences between the intracellular development of the ΔOMT and wild-type bacteria, we performed an optimized binding assay in which a sparse monolayer of adherent ISE6 cells growing in MatTek dishes was exposed to numerous bacteria (100–300 bacteria/cell) and washed gently to maximize retention of bacteria bound to the cells. This is in contrast with our previous assay that used suspended ISE6 cells, a lower multiplicity of infection (MOI), and vigorous washes so that only strongly bound bacteria remained, which increased the sensitivity of the assay but made it difficult to track intracellular development of ΔOMT.
The development of the ΔOMT was compared to wild-type bacteria using confocal microscopy of fixed and immunofluorescently labeled ISE6 cells in MatTek dishes after 1 hr of exposure to bacteria, and subsequently on days 1, 2, 3, 4, 5, 7, and 10. With this method there were no differences observed in binding or internalization between the ΔOMT and wild-type bacteria (Fig 6A and 6B). However, by 44 hr, wild-type bacteria started to form morulae, whereas ΔOMT bacteria remained singly within the cells (Fig 6C). By days 3 and 4, the wild-type bacteria had formed large morulae (Fig 6D and 6E) and on days 5–10, wild-type infections had become asynchronous, with bacteria from lysed cells infecting new cells while other cells harbored large morulae (Fig 6F–6H). ΔOMT bacteria, however, never developed morulae during the 10 days of observation. Only single bacteria were observed within the infected cells throughout (Fig 6D–6H), suggesting that the mutant bacteria were unable to replicate and form morulae within the infected cells even though they were successfully internalized. Observation of ΔOMT bacteria by confocal microscopy identified their location as intracellular.
To confirm that in fact the ΔOMT bacteria resided inside the tick cells, we performed a trypsin-protection assay, similar to those used to remove uninternalized bacteria and/or beads from cells to examine binding proteins in Ehrlichia chaffeensis, A. phagocytophilum, and Helicobacter pylori [34–36]. Cultures were used four days after inoculation with either mutant or wild-type bacteria, when wild-type bacteria had formed large intracellular morulae (S2A–S2C, S2B and S2C Fig), whereas mutants persisted as individual intracellular bacteria (S2D, S2E and S2F Fig). Cultures were treated with trypsin once (wild-type and mutant) or twice (mutant only) to remove any extracellular bacteria, and untrypsinized cells scraped off the growth substrate were used for comparison. As expected, neither mechanical scraping (S2A and S2B Fig) nor trypsinization affected the wild-type morulae (arrow heads) already formed within ISE6 cells (S2C Fig). Similarly, when ISE6 cultures that had been exposed to ΔOMT bacteria for four days were scraped off the flask (S2D Fig), trypsinized once (S2E Fig) or twice (S2F Fig), there was no effect on the bacteria, confirming that they were located intracellularly as also indicated by confocal microscopy. As observed in the previous experiment, mutant bacteria remained as individuals (arrows) within the infected cells and were unable to develop to morulae. Additional controls demonstrated that wild-type bacteria adherent to ISE6 cells for 1 hr were removed from the cells by trypsinization (S2G and S2H Fig), another indication that single ΔOMT bacteria visualized four days p.i. had been internalized. Note that host cell nuclei (asterisks) were recognized by the dog anti-A. phagocytophilum serum and subsequently labeled by the secondary FITC-conjugated anti-dog antibody (S2D, S2E and S2F Fig). They were not detected in panels A, B, and C because the brightness of the large wild-type morulae required a shorter exposure during image acquisition than that used to image ΔOMT. Anti-nuclear antibodies have been detected in dogs infected with vector-borne pathogens (Smith et al. 2004), explaining the reactivity of the dog’s antiserum with host cell nuclei.
To globally identify proteins that were differentially represented in the ΔOMT compared with the wild-type bacteria, we used a proteomic approach based on iTRAQ (isobaric tag for relative and absolute quantitation) technology. Peptides in each sample were labeled with different isotopic tags of known mass to quantify the relative abundance of the proteins in each sample. Both ΔOMT and wild-type bacteria were incubated with ISE6 cells for 4 hr at 34°C. Proteins were extracted from bacteria released from host cells, and triplicate samples were analyzed by tandem mass spectrometry (MS/MS) [37]. In each replicate, multiple A. phagocytophilum proteins were identified that appeared to be differentially abundant in the mutant (S1 Text). Of these, 23 A. phagocytophilum proteins (Table 1) were identified as differentially abundant in all replicates: five proteins were less abundant (hypothetical protein APH_0406, major surface protein 4, anti-oxidant AhpCTSA family protein, and ankyrin (GI88607707)), and 19 appeared more abundant (Table 1). Hypothetical protein APH_0406, and major surface protein 4 (Msp4) presented the lowest relative expression ratios (both <0.2), indicating that they were highly expressed in wild-type bacteria during binding to ISE6 cells compared to the mutant (Table 1). Several proteins known to be involved in infection of mammalian cells [38,39], or highly expressed in A. phagocytophilum replicating in human cells [17], were more abundant in the mutant (Table 1). These proteins included several membrane proteins (P44-18ES, an OmpA family protein, P44-1 Outer membrane protein, an OMP85 family outer membrane protein, and hypothetical protein APH_0405) as well as stress response proteins (co-chaperone GrpE, chaperonin GroEL, and chaperone DnaK) (Table 1). This suggested that, unlike wild-type A. phagocytophilum, the ΔOMT failed to respond to interaction with ISE6 cells in a host cell specific manner, and as a result, the repertoire of proteins in its outer membrane remained unchanged. It is also possible that lack of OMT activity disrupted an environmentally responsive regulatory mechanism or sensor that prepares A. phagocytophilum for changes in hosts. Correct quantification of proteins by iTRAQ is problematic (Shirran and Blotting 2010), and to confirm these results, we examined transcription of several genes that were more abundant in ΔOMT than wild-type during bacterial adhesion to tick cells (based on the iTRAQ data) using qRT-PCR. RNA was isolated from ΔOMT and wild-type bacteria purified from HL-60 cells during late stages of infection to investigate transcription before exposure to ISE6 cells, and after a 2 hr incubation with ISE6 cells for comparison. In HL-60 cells, genes encoding OmpA, p44-18ES, and APH_0404 were strongly up-regulated 19-, 267-, and 5-fold, respectively compared to values obtained after 2 hr in ISE6 cells (S3 Fig). Genes encoding APH_0405 and cytochrome C oxidase subunit II were not regulated (1.5 and 1.3 fold average difference, respectively (S3 Fig)), whereas msp4 expression was down-regulated (0.0033 fold average down-regulation in both the ΔOMT and wild-type bacteria) compared to wild-type bacteria in ISE6 cells (S3 Fig). Thus, the transcript levels mirrored the protein expression detected using iTRAQ, and suggested that the ΔOMT was not able to change gene expression to adapt to conditions in ISE6. In HL-60, the ΔOMT and wild-type bacteria had similar transcript levels, indicating that the mutation did not affect the expression of these genes.
An analysis of the pathways affected in the ΔOMT during incubation in ISE6 based on iTRAQ data showed that several of the more abundant proteins were involved in transcription and protein metabolism, indicating that the mutant was metabolically active (Table 2). In our analysis, we only included proteins with a known role in specific pathways, according to information available at the KEGG (http://www.genome.jp/kegg/) pathways website. Hypothetical and porin proteins were not analyzed within specific pathways, since their roles have not been established.
Because we thought it possible that the OMT might modify either bacterial or host cell proteins, Anaplasma and host cell peptides identified by iTRAQ as having a methyl modification were analyzed to identify those that were less abundant in ΔOMT bacteria, and in whole ΔOMT inoculated cell cultures compared to control wild-type samples. Among peptides with a <0.7 ratio of abundance between the wild-type and mutant (S2, S3 and S4 Texts), we identified eight A. phagocytophilum proteins with reduced methylation of eight corresponding peptides (Table 3). Two of the proteins, Msp4 and APH_0406, were less abundant in the mutant, by a ratio of 0.239 for Msp4 and 0.7484 for Aph_0406, and lacked methyl-modifications of specific residues. The affected amino acids were glutamic acid residues (E) in the Msp4 peptide VEVEVGYK (S2 Text), and an asparagine residue (N) in the APH_0406 peptide NVVLGGMLK (S2 Text).
Fifteen tick host cell proteins displayed reduced methylation when inoculated with the ΔOMT as compared to wild-type infected cells, but prolyl 4-hydroxylase alpha subunit (GI:240974259) and flavonol reductase/cinnamoyl-CoA reductase (GI:241703753) were the only two I. scapularis proteins to be both down regulated as well as to present peptides with reduced methylation in all replicates (S3 Table). Since no OMT was detected in ISE6 cells by IFA during infection with wild-type bacteria (Fig 5), these changes are unlikely to be due to a direct effect of the mutation, but probably reflect an absence of replicating A. phagocytophilum.
To test if the A. phagocytophilum proteins identified by iTRAQ as potential substrates were in fact methylated by the OMT, rOMT was produced in E. coli using the complete coding sequence of the gene (aph_0584) cloned into the vector pET29a. The purity of rOMT was verified by gel electrophoresis (SDS-PAGE) and Coomassie blue staining (Fig 7A), and its molecular weight (MW) corresponded to the predicted MW of ~24 kDa for OMT (Fig 7A). We used the SAM-fluoro:SAM methyltransferase Assay to measure the activity of purified rOMT in in vitro methylation assays with potential substrates (http://www.gbiosciences.com/ResearchProducts/samfluoro.aspx). In this assay, the production of highly fluorescent resorufin (expressed as resorufin units, RU) resulting from oxidization of 10-acetyl-3,7,- dihydroxyphenoxazine (ADHP) by hydrogen peroxide generated during the reaction and monitored at an excitation wavelength of 540 nm and an emission wavelength of 595 nm. Two higher molecular weight proteins present in the un-induced E. coli lysate co-eluted with rOMT. Methylation assays using only the rOMT along with all reagents except for the substrate (negative control) did not demonstrate any detectable increase in fluorescence in the presence of these contaminant proteins, indicating that they did not affect the results of the assay (Fig 7B).
Recombinant versions of proteins identified by iTRAQ as differentially methylated between the mutant and the wild-type bacteria were also produced in E. coli, purified as done for rOMT, and tested in the in vitro methylation assay. rOMT (40 ng) and four recombinant A. phagocytophilum protein substrates (Msp4, APH_0406, TypA, and P44-16b; 50 ng each) were used in methylation reactions for 4 hr at 34°C. We selected these proteins from eight candidates that yielded the strongest reduction in abundance ratios of <0.60 (Table 3). Production of recombinant preprotein translocase subunit SecA was unsuccessful in One Shot BL21[DE3] chemically competent E. coli (Invitrogen, New York), BL21[DE3] (New England Biolabs, Massachusetts), and in Rosetta 2[DE3] E. coli pLysS (Novagen, Germany), thus it was not pursued further. The number of RU (fluorescence) from known concentrations of resorufin (0 μM, 5 μM, 10 μM, 25 μM, and 50 μM) was determined to produce a standard curve, and the concentration of resorufin produced in each reaction was calculated from the standard curve values. Of the four proteins tested, only rMsp4 resulted in a significant and rapid increase in resorufin production (expressed as resorufin units, RU) when incubated with rOMT (Fig 7B). rAPH_0406, rTypA, and rp44-16b produced high background fluorescence that resulted in high initial readings (~1,500–2,000 RU; 1.6–2.1 μM), but did not continue to accumulate a significant number of RU, and only reached values of ~2,200 RU (2.3 μM) (Fig 7B). By contrast, when rMsp4 was used as the substrate, the fluorescence started at a lower reading (1,200 RU; 1.3 μM) but climbed to higher values (~3,400 RU; 3.6 μM), and reached a plateau at around 210 min after the reaction was initiated (Fig 7B). Kinetics of the enzyme reaction were tested with 60, 80, and 100 ng of enzyme with a constant concentration of 50 ng rMsp4, and in the presence of 80, 100, and 150 ng of rMsp4 with a constant concentration of rOMT at 40 ng. We expected that if Msp4 was the substrate, the velocity of the reaction would increase with increasing concentrations of the enzyme and substrate, which would result in a shorter time for the reaction to reach Vmax (the maximum initial velocity when all enzyme molecules present in the reaction are in complex with the substrate). As predicted, the reaction reached Vmax in less time with higher concentrations of rOMT or rMsp4 (S4 Fig). At the time required for the enzyme to reach Vmax, the rOMT had an activity of 0.13 μM/min with a Km of 5.57x105 M and the reaction reached Vmax after 46 min of initiation (Table 4). Because of the slow reaction kinetics, we suspected that A. phagocytophilum OMT required the addition of specific metal ions to catalyze the reaction, similar other o-methyltransferases [40]. Several concentrations (0.5 mM, 2 mM, 8 mM, and 16 mM) of MnCl2 were added to the methylation reaction containing 100 ng (17.89 ρmoles) of rOMT and 100 ng (16.61 ρmoles) of rMsp4. This was in addition to the 10 mM Mn2+ already included in the kit (GBiosciences, pers. comm.). Addition of Mn2+ resulted in greater fluorescence (higher RU) (Fig 7C) and faster reaction times, reaching peak levels of RU by ~100 min after initiation. With the addition of 16 mM of MnCl2 (17 mM total Mn2+), 98,000 RU (53.8 μM) were reached compared to 20,000 RU (11 μM) when the enzyme and substrate were used alone with the 10 mM Mn2+ supplied in the kit (Fig 7C). Similar decreases in reaction time (130 min) were observed with the addition of 8 mM of MnCl2 (Fig 7C). The higher activity of the enzyme was also evident from the changes in enzyme activity (Km), and in the time to reach Vmax (1.67x104, 1.77 μM/min, and 9:00 min, respectively) (Table 4). In preliminary tests, MgCl2 did not accelerate the activity of the enzyme reaction significantly.
Protein for crystallization experiments was produced and purified by Seattle Biomed, a collaborator within the Seattle Structural Genomics Center for Infectious Disease (SSGCID), and was crystallized as described in the Materials and Methods section. Although this target has 33% sequence identity to its closest neighbor in the Protein Data Base (PDB), phases for the initial X-ray data from the synchrotron could not be determined by molecular replacement (MR). We initially hypothesized that binding of a substrate or co-factor would alter the conformation of the protein to something more amenable to MR. However, even after co-crystallizing the protein with SAH, phases for the X-ray data still remained recalcitrant to being solved by MR. Therefore, we chose to pursue single wavelength anomalous diffraction (SAD) phasing by using high concentration soaks (0.5 M) with sodium iodide solution, as it has previously yielded de novo phases for many other targets from the SSGCID [41]. Iodide-SAD data were collected on our in-house X-ray generator (Table 5) and PHENIX HySS was able to find 72 iodide ion sites during its search, but we were able to identify 118 in the final structure using anomalous difference map peaks with a contour level of 3.5 σ. Phases for the Apo, SAM-Mn2+, and SAH-Mn2+ datasets (Table 5) were then determined by MR, using the SAH-bound structure as a search model.
AnphA.01233.a has a canonical o-methyltransferase fold which consists of a central 7-stranded β-sheet that is flanked on both sides by three α-helices (Fig 8). Since the structure was not solvable by MR, we assayed the PDB for structural homologues using the full-PDB SSM search on the PDBeFold website. The nearest homologue was 3CBG, another o-methyltransferase from Cyanobacterium synechocystis, which had a Cα RMSD of 1.69 Å2. With this much of a difference in structural similarity in the PDB, it is not surprising that MR failed to provide phases. AnphA.01233.a crystallizes as a dimer in both the Apo and SAH-bound crystal forms—the Apo form has one dimer per asymmetric unit, while the SAH-bound form has three dimers per asymmetric unit (Fig 8). The SAH molecule binds at the apex of the β-sheet, and the binding pocket is completely solvent exposed. When aligning a monomer of the Apo- and SAH-bound structures, the RMSD for all Cα carbons is only 0.273 Å2, so no large conformational changes occur due to ligand-binding. However, there is a small movement in a helix, composed of residues 31–40, that moves towards that substrate in the SAH-bound form as compared to the Apo form (S5 Fig).
Enzymatic assays showed that the catalytic activity of the OMT was greatly increased in the presence of the divalent metal ion Mn2+ at >10 mM concentration. Therefore, we chose to attempt co-crystallization experiments with Mn2+ in the presence of both SAM and SAH. Crystals formed readily in multiple initial sparse matrix screen conditions within a week and produced higher resolution data than either of the previous datasets collected in the absence of Mn2+ (Table 5). After molecular replacement and initial refinement of these structures, a positive Fo-Fc map peak at a contour level of 25 σ was observed in both the 4PCA and 4PCL structures, indicating that manganese was bound to the protein in close proximity to the SAH/SAM binding site (Fig 9). The Mn2+ ion is coordinated by the side-chains of D136, D162, and N163 and waters from the solvent (Fig 9). This places the Mn2+ ion within 4.5 Å of the CE methyl group to be transferred from the SAM co-factor to the hypothesized glutamate substrate of Msp4. Interestingly, a glutamic acid residue from a neighboring asymmetric unit, E177, inserts into a catalytic site in the SAM- Mn2+ bound structure. It appears to adopt a slightly different conformation for either chain A or chain B, which contains 2 molecules of OMT per asymmetric unit. In chain A, E177 interacts directly with the manganese ion at a distance of 2.5 Å (Fig 9A), whereas in chain B, interaction of E177 with the Mn2+ ion is mediated by two water molecules (Fig 9B). Since the natural substrate of this enzyme is a glutamic acid residue(s) from Msp4, it is likely that the glutamic acid from Msp4 interacts with the OMT enzyme similarly to this.
In order to understand the possible relationship of the A. phagocytophilum OMT with other members of this family of enzymes, PSI-BLAST was used to search for homologous OMTs in other organisms. Within the order Rickettsiales, only members of the families Anaplasmataceae and Candidatus Midichloria mitochondrii (from the new family “Candidatus Midichloriaceae”) encoded OMTs related to A. phagocytophilum OMT (S6A Fig). However, Δ-proteobacteria encoded OMTs that had even closer homology to A. phagocytophilum OMT, including OMTs from Bdellovibrio bacteriovorus, Gloeocapsa sp., Anaeromyxobacter dehalogenans, and Haliangium ochraceum (S6A Fig). A PSI-BLAST search assigned a better e-value (4e-40) to an OMT from B. bacteriovorus than to the C. M. mitochondrii OMT (2e-36), suggesting that the former enzyme more closely resembled A. phagocytophilum OMT. Furthermore, when the three motif sites detected by MEME (S6B Fig) from the four OMT enzymes were compared, the Δ-proteobacteria OMTs appeared to be more similar to A. phagocytophilum OMT than the C. M. mitochondrii OMT (S6B Fig). Motif 1 of the A. phagocytophilum OMT had 48% identity and 66% similarity to H. ochraceum OMT motif 1, respectively, whereas C. M. mitochondrii OMT motif 1 only showed 33% identity and 53% similarity. A. phagocytophilum OMT motif 2 exhibited 60% identity and 74% similarity with the corresponding motif in B. bacteriovorus OMT compared to values of 52% identity and 67% similarity to the motif regions of C. M. mitochondrii OMT motif 2. A. phagocytophilum motif 3 showed 44% identity and 72% similarity to B. bacteriovorus OMT motif 3 compared to 33% identity and 67% similarity with that motif in C. M. mitochondrii OMT (S6B Fig).
The tertiary structure of Msp4 was predicted using Phyre2 [42] that compares conserved residues of a query protein to the sequence of proteins with known crystallized structures. The predicted tertiary and secondary structures were used to predict the probable positions of the methylated residues. Msp4 was predicted to form a β-barrel typical of porins (S7B Fig), and the glutamic acid residues that are modified by the OMT are predicted to be located at the start of one of the β-strands forming the beta-barrel (S7A and S7B Fig). Furthermore, transmembrane and signal peptide prediction software suggested that the first ~30 aa residues represented a signal peptide to direct transport of the protein from the cytoplasm to the outer membrane (S7C Fig). These residues corresponded to the α-helix at the N-terminus (dark blue) that is probably cleaved before the protein is positioned in the outer membrane (S7B Fig). The protein does not contain predicted transmembrane domains, but it is very likely that its positioning in the outer membrane is similar to that reported for other porins in that the β-barrel spans the membrane, and the portion of the protein with the longest loops is exposed on the outside of the bacteria.
Genetic manipulation of A. phagocytophilum and other members of the Anaplasmataceae is difficult due to the intracellular nature of these organisms, and currently relies on random mutagenesis to study the role of specific genes during pathogenesis in the mammal and development in the tick [18,37,43]. Targeted mutagenesis in the related organism, Ehrlichia chaffeensis, proved ultimately unsuccessful as the transformants obtained were not able to persist in vitro for more than six days [43]. The recent success of targeted mutagenesis in Rickettsia rickettsii resulting in the disruption of a major surface protein gene (ompA) [44] presumed to be a virulence factor without producing a detectable defect provides impetus to develop this method for other Rickettsiales, and serves as a reminder that gene function ultimately must be confirmed by mutational analysis. In this manuscript, we report the effects of the mutation of a specific gene of A. phagocytophilum that we suspect abolished its ability to infect tick cells. However, due to the lack of a complementation system in the Anaplasmataceae, we cannot completely rule out that this change in phenotype was due to secondary mutations. Nevertheless, our conclusions are supported by the effect of the methylation inhibitor AdOx, which mimicked the mutation at a concentration of 30 nM (Fig 3). Previously, Chen et al. [45] described an A. phagocytophilum mutant with an insertion in the dihydrolipoamide dehydrogenase 1 (lpda1) gene at the APH_0065 locus, which altered the inflammatory response during infection of mice by increasing the production of reactive oxygen species [45], but had no effect on in vitro growth. The mutant (ΔOMT) described here was selected in the human cell line HL-60 in which it replicated in a manner comparable to wild-type bacteria (Fig 2). The mutant had an insertion in aph_0584 encoding an o-methyltransferase (OMT) family 3 (GI: 88598384; E.C. 2.1.1.24). The inability of ΔOMT to replicate within tick cells highlighted the distinct mechanisms used by A. phagocytophilum for colonization of mammal and tick hosts.
It is interesting that the search for OMTs similar to that encoded by aph_0584 only identified an OMT in one Rickettsiales organism, i.e., in C. M. mitochondrii, which is outside the family Anaplasmataceae. This intracellular organism develops in the mitochondria of I. ricinus ticks and is a member of a new family thought to be closely related to, but distinct from, the Anaplasmataceae [46]. Phylogenetic analysis showed highest similarity with enzymes from members of the Δ-proteobacteria (S6A Fig) and analysis of the different motifs present in the OMT indicated that A. phagocytophilum OMT is more similar to the OMT from the Δ-proteobacterium B. bacteriovorus than to that from C. M. mitochondrii (S6B Fig). B. bacteriovorus is a predatory bacterium that attacks gram-negative bacteria and, like A. phagocytophilum, presents a biphasic life cycle with an “attack form” that attaches to the host cell and a “dividing form” that occurs only in the periplasm of its host within a vacuole formed by its own proteins as well as host proteins [47]. Like A. phagocytophilum, B. bacteriovorus differentially expresses genes depending on the phase of development during infection [48]. Two OMTs are differentially regulated depending on the phase of infection; one OMT is up-regulated upon entry to Bdelloplast (Bd2861) and another extracellularly (Bd0381) [48]. Whether or not the up-regulation of the OMT in B. bacteriovorus during cell invasion is involved in the methylation of proteins important for entry is not known, however its up-regulation indicates that it may play such a role. It is possible that an ancestor of the families Anaplasmataceae and “C. Midichloriaceae” obtained their OMTs from members of the Δ-proteobacteria by lateral gene transfer. This possibility is supported by the absence of members of this enzyme family in the Rickettsiales, in which the only enzyme that showed slight similarities with the OMT was a bifunctional methyltransferase (m7G46) present in some Rickettsia species, albeit with high e-values of e0.28 –e1.2. These family 3 OMTs existing in Anaplasmataceae and the new family “C. Midichloriaceae” are evidently not required by other members of the Rickettsiales for infection of ticks, which seem to utilize different methyltransferases to carry out similar functions [23].
OMTs of the type encoded by aph_0584 methylate free carboxyl groups on glutamic acid residues of bacterial chemoreceptors [49–51]. Thus, they are involved in environmental sensing, which could also be the case in A. phagocytophilum since a sensor-histidine kinase CckA (aph_0582) is predicted to localize to the membrane and to be involved in signal transduction and regulation of transcription (http://www.uniprot.org/uniprot/Q2GKC9). Notably, CckA is part of the A. phagocytophilum two component system, and is paired with response/regulator transcription factor CtrA, allowing A. phagocytophilum to respond to environmental changes [52]. A lack in the ability of the ΔOMT to respond when transferred from mammalian to tick cells could explain the continued expression of a set of proteins known to be important for infection of mammalian cells, and to be down-regulated in tick cell culture (Table 1; S3 Fig) [17,38]. However, it is unlikely that these phenotypic changes are due to a polar effect on the expression of the sensor-histidine kinase (aph_0582) since the transposon promoter drives transcription in the opposite direction from aph_0582 and the insert is located at a distance of around 1600 bp from that gene (Fig 1B). Furthermore, transcription of the flanking genes is independent of the expression of the omt gene, and they do not appear to be part of an operon since no bands were amplified from the intergenic regions between these genes (S8 Fig). Although methyltransferases modifying glutamic acid residues have recently been identified as being widely conserved in eukaryotes, their targets, poly(A)-binding proteins, are methylated at additional amino acid residues, placing these enzymes in a different class from A. phagocytophilum OMT [53]. Interestingly, the OMT from B. bacteriovorus (Bd0381) which is homologous to A. phagocytophilum OMT (e-value 1e-42) was shown to be up-regulated when the bacteria were extracellular along with several genes involved in chemotaxis and sensing, including a methyl chemotaxis protein (Bd2503), pilS sensor protein (Bd1512), two-component response regulator (Bd0299), and a sensory box histidine kinase (Bd1657) [48]. Whether the B. bacteriovorus OMT, Bd0381, is involved in the methylation of the methyl chemotaxis protein is not known. However, it is possible that this OMT plays a role in environmental sensing, and that acquisition of a gene encoding such an enzyme enabled members of the Anaplasmataceae family to adapt to environmental changes more efficiently.
Analysis of the behavior of ΔOMT in ISE6 cells demonstrated reduced binding of A. phagocytophilum to tick cells (Fig 3A), partly explaining the decrease in bacterial numbers seen as early as 1 day p.i. (Fig 2). However, binding was not completely abolished by either the mutation or treatment with Adox (Fig 3). Therefore, we investigated differences in bacterial internalization and intracellular development. Although the ΔOMT that did bind to tick cells were readily internalized between 1–20 hr (Fig 6A and 6B), the bacteria were not able to form morulae and replicate intracellularly, but persisted as individual bacteria within ISE6 cells for at least 10 days p.i. (Fig 6C–6H). Furthermore, we verified that the ΔOMT bacteria were internalized within ISE6 cells by confocal microscopy and a trypsin-protection assay (S2 Fig). Methylation of outer membrane proteins and virulence factors is increasingly recognized as an essential process during host invasion and infection by several obligately and facultatively intracellular bacteria [22,28,30,32,54–58]. Therefore, we considered that the lack of methylation of glutamic acid residues in Msp4 may have played a role in the reduced adhesion to tick cells and was probably responsible for preventing replication of ΔOMT in tick cell culture. Methylation of proteins that mediate adhesion to host cells has been reported in other members of the Rickettsiales [54,59]. Methylation of R. prowazekii OmpB by lysine methyltransferases was shown to play a role in adhesion to and infection of endothelial cells, and to be important for virulence [59]. Nevertheless, E. coli expressing recombinant OmpB were able to bind to endothelial cells in the absence of methylation [31]. Likewise, E. coli transformed to express recombinant Msp4 were able to adhere to ISE6 cells in the absence of the omt gene (S9A Fig and S5 Text). Furthermore, E. coli transfected only with msp4 construct bound more readily than those harboring both msp4 and omt (S9A Fig). Analysis of the OMT protein sequence using Phobius predicted a non-cytoplasmic location of the enzyme, suggesting that methylation of Msp4 occurred in the periplasm of A. phagocytophilum (S9B Fig). It is likely that OMT is not transported to the periplasm in E. coli and methylation is thus not carried out efficiently. Because E. coli transfected with only msp4 were able to bind to tick cells, methylation of the protein was not essential for adhesion, explaining why disruption of omt only reduced but did not abolish A. phagocytophilum binding to tick cells (Fig 3). Productive infection of cells by A. phagocytophilum requires completion of a multi-step process for efficient invasion and replication to occur. Increased expression of OMT in bacteria bound to ISE6 cells compared to those adhering to HL-60 cells (Fig 5) suggested that physical contact with the tick cell outer membrane induced OMT expression. Induction of OMT expression happened rapidly, and waned as bacteria passed into the cytoplasm.
iTRAQ identified several potential substrates of the enzyme in A. phagocytophilum and I. scapularis cells, two of which included A. phagocytophilum proteins previously shown to be highly expressed during infection of ISE6 cells, i.e., Msp4 and APH_0406 [17]. However, in vitro methylation assays (Fig 7B and 7C) using recombinant versions of potential substrates only confirmed Msp4 (Fig 7B), and identified Mn2+ as the most effective cofactor, indicating that A. phagocytophilum OMT is a metal dependent methyltransferase. The kinetics of the reaction were comparable to those reported for methyltransferases from R. prowazekii and R. typhi in which the linear portion of the reaction curve occupied 50–300 min [30,54]. It is possible that the substrate protein, Msp4, is methylated by OMT as a linear molecule prior to translocation rather than as a folded protein, as used here, and this may further explain the slow in vitro assay kinetics. These results confirmed Msp4 as a substrate of the OMT, but whether the enzyme methylates other proteins awaits further investigation. The other proteins that showed differential methylation in the proteomic analysis, but were not methylated in vitro, are possibly methylated by other methyltransferases present in A. phagocytophilum or in the host cell.
Msp4 is an antigenic protein encoded by a single copy gene that is highly conserved between different strains of A. phagocytophilum [60], as well as in other members of the genus Anaplasma [61]. As a member of the Msp2 superfamily of proteins [62], Msp4 is homologous to A. phagocytophilum Msp2 (P44), which has been shown to facilitate binding to mammalian cells, to be a porin and to be post-translationally modified. It is likely that Msp4 and Msp2 (P44) are structurally and functionally similar but that they have evolved to function in the tick vector and mammal, respectively. The most common non-specific bacterial porins form 16-strand β-barrels configd as trimeric peptide subunits [63,64]. More substrate specific bacterial porins are comprised of 18-, 14-, 12-, or 8-strand β-barrels and in some cases are present as monomers (e.g., the 14 beta-stranded porins OmpG and CymA in Escherichia coli) [63]. Since Msp4 is predicted to contain a 14-stranded β-barrel (S7 Fig), and by homology with the A. marginale Msp4 is likely monomeric [65], we conclude that it is probably substrate specific, which is supported by its activity exclusively in tick cells. It is interesting to note that the glutamic acid residues (E) of Msp4 that appear to be important for A. phagocytophilum development inside tick cells are close to one of the loops on the outside of the channel (S7 Fig). Similarly, L. interrogans OmpL32 contains methylated glutamic acid residues that are important for infection and colonization of kidney and liver cells [32]. However, more research is needed to determine the exact function of these methyl modifications of Msp4. We realize that structure predictions can be unreliable, and ideally the crystal structure of Msp4 in association with that of OMT should be resolved. We also expect that an A. phagocytophilum msp4 mutant would display a similar or even more severely compromised phenotype than the omt mutant, since it is possible that other enzymes participate in methylation of the Msp4. Until such a mutant is available, the predictions serve as a starting point to infer potential implications of this modification for the biology of this bacterium.
The change in phenotype and our proteomic analysis support the conclusion that methylation of Msp4 may be necessary for efficient and productive infection of tick cells. Some porins have been shown to display double functionality, acting also as adhesins and being expressed under specific environmental conditions [64], characteristics that could fit Msp4 [63]. Partial inhibition of adhesion due to the mutation of omt is not surprising, as it is likely that more than one adhesin is involved in binding to tick cells. This has been shown for invasion of mammalian cells by A. phagocytophilum, where three adhesins have been identified, OmpA, Asp14 and AipA [66–68]. Rickettsiaceae possess two additional adhesins besides OmpB and OmpA, named Adr1 and Adr2. These were recently identified by proteomic approaches and also presented putative β-barrel structures [69]. Adhesins may also serve to protect the bacteria from mammalian complement that is abundantly present in the blood, and consequently ingested with the tick blood meal. Rickettsia conorii OmpB β-peptide has been shown to interact with mammalian complement regulatory factor H via the exposed loops extending from the transmembrane β-barrel structure, and a number of bacterial factor H-binding proteins have been identified as adhesins [63,64]. Anaplasma phagocytophilum also evades complement-mediated killing, but it is not known whether this capability is mediated by binding of complement regulatory factors, or by direct interaction with complement [65]. In the cell-culture system used here, such factors would not be relevant due to the absence of active complement.
The most significant phenotypic change due to the mutation in the omt gene was the inability of A. phagocytophilum to replicate and form normal morulae within ISE6 cells (Fig 6C–6H). In ISE6 cells, ΔOMT persisted as individual bacteria, while wild-type bacteria formed large morulae that are distinguishable on day 3 p.i. (Fig 6D). Up-regulation of OMT expression during interaction of A. phagocytophilum wild-type with ISE6 cells (Figs 4 and 5), as compared to the inability of the ΔOMT bacteria to change protein expression (Table 1 and S3 Fig), indicated that this was necessary for normal morphogenesis of A. phagocytophilum in tick cells (Fig 6).
The A. phagocytophilum isolate HZ, which was originally cultured from a New York patient [70], was cultivated in HL-60 cells maintained in RPMI 1640 medium (Lifetechnologies, New York) supplemented with 10% FBS (BenchMark, Gemini Bioproducts, California), and 2mM glutamine at 37°C with 5% CO2 in humidified air [71]. Several transformants were generated to express Green Fluorescent Protein (GFPuv) and were selected and maintained in HL-60 cells as described [18,71]. The ability of the transformants to grow in ISE6, an I. scapularis embryonic cell line, was tested by inoculating purified cell-free bacteria or whole infected HL-60 cells into 25-cm2 flasks containing confluent cell layers of ISE6 cells. Cultures were maintained in L-15C300 supplemented as described, and the pH was adjusted to 7.5–7.7 with sterile 1N NaOH [18,72]. Growth and development of transformants was evaluated by fluorescence microscopy using an inverted Nikon Diaphot microscope (Nikon, New York) to detect A. phagocytophilum expressing GFPuv [73] and by examination of Giemsa stained cell samples spun onto slides.
A transposon mutant (ΔOMT) deficient for growth in ISE6 cells was cultivated in HL-60 cells as described above by passing 3 x 103 infected HL-60 cells into a new flask containing 3–4 x 105 uninfected cells and 20 ml of fresh medium every 5 days. Spectinomycin and streptomycin (100 μg/ml each) were added to the cultures for selection of mutants carrying the aadA resistance gene encoded on the transposon. The number of insertion sites in the mutant population was determined by Southern blots of DNA purified from a 25-cm2 flask of infected HL-60 cells, using the Puregene Core Kit A (Qiagen, Maryland) with an additional phenol-chloroform extraction step, and Phase Lock Gel Heavy (5 Prime, Maryland) to separate phases. DNA concentration was measured with a BioPhotometer (Eppendorf, New York), and DNA extracted from HZ wild-type bacteria served as control. DNA (100 ng) from the mutant and wild-type HZ was digested with BglII and EcoRV and samples were electrophoresed in 1% agarose gels. DNA was transferred and probed as described [73], using digoxigenin-labeled probes specific for gfpuv (PCR DIG Probe Synthesis kit; Roche, Indiana). A plasmid construct, pHIMAR1-UV-SS, encoding the transposon, served as positive control [18].
To determine genomic insertion sites in the mutant population, 5 μg of ΔOMT DNA was digested with BglII, treated with DNA clean & concentrator (Zymo Research, California) and ligated into the pMOD plasmid for electroporation into ElectroMAX DH5α cells (Invitrogen, New York). ElectroMAX DH5α cells containing the transposon were selected on YT plates with 50 μg/ml of spectinomycin and streptomycin. DNA was purified by phenol/chloroform extraction and then sequenced at the BioMedical Genomics Center (University of Minnesota).
The ΔOMT and wild-type strains were grown in 25-cm2 flasks containing HL-60 cells as described above. Bacteria were purified from four flasks containing 25 ml of a >90% infected cell suspension by passing the cell suspension through a bent 27 G needle and filtration of the lysate through a 2 μm pore size filter. Purified bacteria were transferred to two 15 ml tubes, centrifuged at 10,000 x g for 5 min and then resuspended in 3 ml of RPMI medium supplemented as described above. Bacteria were diluted 1:40, 1:100, and 1:400 in 20 ml of uninfected HL-60 cultures, and incubated at 37°C as described above for infected HL-60 cells. Samples of 1.5 ml were taken from each culture every day for a 5-day period, and DNA was extracted as described for DNA samples used in Southern blots. The experiment was repeated in triplicate.
To generate growth curves of ΔOMT and wild-type bacteria in ISE6 cell cultures, bacteria were purified as described above, and centrifuged at 10,000 x g for 5 min at 4°C. Supernatant was discarded, and cell free bacteria were diluted in supplemented L15C300 at ratios of 1:6, 1:12, and 1:24. The experiment was done in triplicate. To assess bacterial growth, DNA was extracted from 1.5 ml of mutant and wild-type cultures of bacteria grown in ISE6 cells every 3 days for 12 days as described before.
The number of bacteria per sample was determined by qPCR using the primers msp5 fwd and msp5 rev (S1 Table) that amplify a fragment of the single copy number msp5 gene. qPCR reactions were performed in an Mx3005p (Agilent, California) cycler, using Brilliant II SYBR Green Low ROX QPCR Master Mix (Agilent, California) under the following conditions: an initial cycle of 10 min at 95°C, 40 cycles of 30 sec at 95°C, 1 min at 50°C, and 1 min at 72°C, and a final cycle of 1 min at 95°C, 30 sec at 50°C, and 30 sec at 95°C. A standard curve was generated using the msp5 fragment cloned into the pCR4-TOPO vector (Invitrogen, New York).
To examine binding of ΔOMT and wild-type A. phagocytophilum to tick cells, we used two different assays to evaluate adhesion to tick cells and subsequent intracellular growth and development. The first assay was carried out under stringent conditions with a low MOI that permitted sensitive assessment of the effect of the mutated omt gene. The second assay (described further below, under “ISE6 infection time point experiment to evaluate intracellular development of the ΔOMT”) was designed to allow maximal, saturating binding so that mutant bacteria could be readily observed inside tick cells. For the first assay, bacteria were purified from 20 ml of one fully infected HL-60 culture and were added to about 2.5x105 ISE6 cells in 50 μl of supplemented L15C300 medium in a 0.5 mL centrifuge tube. To ensure that only activity induced during binding and not cell entry was measured, bacteria were incubated with host cells for 30 min at room temperature, flicking the tube every 5 min to enhance contact between bacteria and cells. The cells were washed twice in unsupplemented L15C300 and centrifuged at 300 x g for 5 min to remove unbound bacteria. The cell pellet was resuspended in phosphate buffered saline (PBS) and spun onto microscope slides for 5 min at 60 x g, using a Cytospin 4 centrifuge (Thermo Shandon, Pennsylvania). Slides were fixed in absolute methanol for 5 min and dried at 50°C for 10 min. Bound bacteria were labeled using an IFA with dog polyclonal antibody against A. phagocytophilum and FITC-labeled anti-dog secondary antibodies. DAPI was used to stain the host cell nuclei, and aid in host cell visualization. The number of bacteria bound to 300 cells was counted for each sample. This was repeated in triplicate, and differences were evaluated using Student’s t-test to assess significance with SigmaStats (Systat Software, California).
To verify that lack of methylation of substrate brought about by the disruption of the omt gene was the cause of reduced binding of the mutant to ISE6 cells, we added adenosine dialdehyde (Adenosine periodate Oxidized, or AdOx (Sigma Aldrich, Missouri)), which inhibits SAM-dependent methyltransferases by increasing the concentration of S-adenosyl-L-homocysteine [29], to wild-type cultures. Wild-type and ΔOMT bacteria were purified from 50 ml infected HL-60 cells as described. Purified wild-type bacteria were incubated with AdOx at 20 nM, 30 nM, and 40 nM final concentrations for 1 hr at 34°C before adding them to 2x105 uninfected ISE6 cells. Controls consisted of wild-type and ΔOMT bacteria purified the same way, but incubated at 34°C for 1 hr without addition AdOx.
The bacteria and ISE6 cells were incubated in 200 μl of supplemented L15C300 medium at 34°C for an additional hr to allow binding. Cells were washed 3 times in supplemented L15C300 medium by centrifugation at 600 xg to remove unbound bacteria. After the final wash, cells were resuspended in 1 ml of supplemented L15C300 medium and 50 μl of the suspension was spun onto microscope slides as described above and fixed in methanol for 10 min. The samples were incubated with dog anti A. phagocytophilum serum and labeled with FITC conjugated anti-dog antibodies. Samples were mounted in Vectashield Mounting medium containing DAPI to aid in host cell visualization (Vector Laboratories, California), and observed using a 100 x oil immersion objective on a Nikon Eclipse E400 microscope. Bacteria were counted as described above for regular binding assays, and differences in the number of bacteria per cell were evaluated using the Student-Newman-Keuls one-way ANOVA on ranks, using SigmaStat.
Two thousand LifeAct-mCherry expressing ISE6 cells [74]were seeded onto the glass portion of MatTek dishes (Ashland, Massachusetts) in 250 μl of medium the day before bacterial challenge. ΔOMT and HZ wild-type bacteria were cultured in five million HL-60 cells until > 95% infected, and bacteria were beginning to be released from cells. Cell-free bacteria were prepared by passing the infected cells through a 25 G needle 5 times, and intact cells were removed by centrifugation at 1,110 x g for 5 min. The supernatant was passed through a 2 μm pore size syringe filter to remove cell debris, and the bacteria were collected by centrifugation at 11,000 x g for 5 min. Bacteria were resuspended in 100 μl fresh culture medium and 20 μl of the suspension was inoculated onto ISE6 cell layers after 170 μL of culture medium had been removed. This resulted in a multiplicity of infection of 100–300 bacteria/cell. Dishes were incubated at 34°C in a water saturated atmosphere of 3% CO2 in air for 1 hr with agitation at 10-min intervals to ensure uniform exposure of cells to bacteria. Unbound bacteria were removed by washing the cells once with 2 ml of medium, and 2 mL fresh medium was added and the cultures returned to the incubator. At each of eight time points (0, 1, 2, 3, 4, 5, 7, and 10 days), culture medium was removed from MatTek dishes and cells were immediately fixed by flooding with 2 mL methanol (1 min) followed by two additional rinses (1 min) with fresh methanol. Cells were then air dried and stored at room temperature until the bacteria were labeled by IFA.
Cells were blocked with 50% FBS in culture medium for 1 hr at room temperature. Bacteria were labeled with dog anti-A. phagocytophilum serum for 1 hr followed by FITC-labeled anti-dog antibodies for 1 hr (each diluted 1:1,000). After each antibody exposure, cells were washed three times in PBS. Cells were mounted in VectaShield (Vector Laboratories) medium with 4',6-diamidino-2-phenylindole (DAPI), and examined and photographed under a cover slip using confocal microscopy as described above.
LifeAct-mCherry-expressing ISE6 cells cultured in MatTek dishes were challenged with ΔOMT or HZ wild-type bacteria at a high MOI as described above, and then exposed to trypsin to confirm the confocal microscopy finding that the ΔOMT bacteria were located inside the ISE6 cells. Four days following challenge, cells were mechanically dislodged using a cell scraper, spun onto slides, air-dried and fixed in methanol. These samples represented non-trypsin controls. Cells in remaining dishes were suspended in 1 ml 0.25% trypsin-EDTA (Gibco) for 3 min, collected by centrifugation (350 x g, 2 min), resuspended in 1 ml culture medium, and 100 μl volumes were spun onto microscope slides. Remaining cells were trypsinized a second time for 2 min, washed in medium and spun onto slides. Samples (mechanically scraped cells, trypsinized 1x, trypsinized 2x) were immunofluorescence-labeled as described, mounted in VectaShield without DAPI and imaged as above (S2A–S2F Fig). As a positive control to demonstrate that trypsin removed wild-type HZ adherent to ISE6 cells, the same trypsinization procedure was followed after exposing ISE6 cells (not expressing LifeAct-mCherry) to HL60-grown HZ bacteria for one hour, and mounting slides with DAPI (S2G and S2H Fig).
Relative expression of omt was measured at different times to determine when it was upregulated. Wild-type A. phagocytophilum HZ was purified from HL-60 cells and inoculated into four 25-cm2 flasks containing either 5 ml of uninfected ISE6 or 2 ml of HL-60 cultures. Bacteria inoculated into HL-60 cell cultures or ISE6 cells were incubated for 30, 60, 120, or 240 min at 37°C or 34°C, respectively. Total RNA was extracted from whole infected cell cultures, using the Absolutely RNA Miniprep Kit (Agilent, California) according to the manufacturer’s specifications. RNA was DNAse treated with 0.5 units TURBO DNAse (Ambion, New York) at 37°C for 30 min. The DNAse treatment was repeated twice and RNA concentrations were measured using a Biophotometer.
Omt expression was normalized against expression of the rpoB and msp5 genes that had been found to be consistently expressed in both cell types during tiling array analysis [17]. qRT-PCR reactions were carried out using Brilliant II QRT-PCR SYBR Green Low ROX Master Mix (Agilent, California) using primers listed in S1 Table. Reaction parameters were as follow: one cycle at 50°C for 30 min, one denaturing cycle at 95°C for 10 min, 40 cycles that consisted of 30 sec at 95°C, 1 min at 50°C, and 1 min at 72°C, and a final cycle of 1 min at 95°C and 30 sec at 50°C. Ct values were established during amplification and the dissociation curve was determined during the final denaturation cycle. Expression of the omt gene was analyzed using the 2-ΔΔct method, and significant differences were determined using Student’s t-test with SigmaStat.
The relative expression of the genes that resulted in a greater abundance of encoded proteins in ΔOMT (Table 1), as well as the expression of msp4 was determined by qRT-PCR, using the primers listed in S1 Table. Total RNA was purified from 8x105 HL-60 cells fully infected with wild-type or mutant bacteria, and from 8.4x105 ISE6 cells fully infected with wild-type bacteria, using the Absolutely RNA Miniprep Kit. qRT-PCR reactions were carried out as described above with msp5 and 23s rRNA used as normalizer genes. Gene expression was determined as described above.
Recombinant OMT (rOMT) protein for antibody production was produced using the pET29a expression vector (Novagen, Germany) by amplifying the entire coding region with the primers rOMT fw and rOMT rv (S2 Table) and pfu DNA polymerase (Promega, Wisconsin). Conditions were as follows: one denaturing cycle at 94°C for 3 min, 10 cycles with a denaturing step at 94°C for 1 min, 40°C for 1 min for annealing, and extension at 72°C for 2 min, then 20 additional cycles with a denaturing step at 94°C for 1 min, 47°C for 1 min for annealing, and extension at 72°C for 2 min, and a final extension step of 5 min at 72°C. The amplified product was digested with the restriction enzymes SalI and EcoRV, followed by ligation into the vector at 15°C overnight. The plasmid was cloned into One Shot TOP10 competent cells (Invitrogen, New York) for replication and purified using the High Pure plasmid isolation kit (Roche, Indiana). Integrity of the plasmid was checked by sequencing with the T7 promoter (5’- TAA TAC GAC TCA CTA TAG GG– 3’) and the T7 terminator (5’- GCT AGT TAT TGC TCA GCG G– 3’) primers at the Biomedical Genomics Center of the University of Minnesota. Plasmids were transfected into BL21(D3) E. coli (New England Biolabs, Massachussets) for expression. BL21(D3) E. coli were inoculated into 100 ml of Superior Broth (AthenaES, Maryland), induced with 200 μM IPTG, and incubated at 37°C overnight with constant shaking. Protein was purified using Ni-NTA Fast Start Kit columns (Qiagen, Maryland). Protein concentrations were measured using the BCA protein assay kit (Pierce, Illinois).
Functional rOMT for enzyme activity assays was produced using the expression vector pET29a as described above but amplifying the entire open reading frame of the gene with the primers rOMTns Fw and rOMTns Rv (S2 Table), using the following conditions: one denaturing cycle at 94°C for 3 min, 10 cycles with a denaturing step at 94°C for 1 min, 45°C for 1 min for annealing, and extension at 72°C for 2 min, then 20 additional cycles with a denaturing step at 94°C for 1 min, 54°C for 1 min for annealing, and extension at 72°C for 2 min, and a final extension step of 5 min at 72°C. The amplified product was digested with NdeI and EcoR1, and ligated into pET29a to produce rOMTns lacking the S-tag present in the plasmid, and cloned into One Shot TOP10 competent cells. After confirming integrity, the plasmid was cloned into Rosetta 2(DE3) pLysS E. coli (Novagen, Germany). Transformed E. coli were incubated in 100 ml of Superior Broth and induced with 1 mM IPTG at 37°C for 5 hr. Proteins were purified with Ni-NTA Fast Start Kit columns, and protein concentration was measured as described above.
Recombinant non-functional rOMT was purified by dialysis in a 3 ml Slide-A-Lyzer Dialysis Cassette 10K MWCO (Thermo Scientific, Illinois) against tris-buffered saline (TBS), over night with one buffer change after 2 hr. Four 6–8 weeks old, female C57BL/6J mice (Jackson Laboratories, Maine) were immunized by subcutaneous (s.c.) injection of 100 μg of recombinant rOMT in TiterMax Research adjuvant (CytRx Co., Georgia). Mice were boosted twice with 100 μg of rOMT without adjuvant at 14 days and 24 days later. To obtain antiserum, blood was collected 10 days after the second booster, and unimmunized mice from the same cohort were used as controls. Serum was frozen at -20°C for future use.
Wild-type bacteria were grown in 50 ml of RPMI1640 containing HL-60 cells until > 90% of the cells were infected. Around 5 x 107 infected cells were used to obtain cell free bacteria by vortexing the infected cells with 60/90 grit silicon carbide (Lortone, Inc., Mukilteo, Washington) for 30 sec followed by filtration through a 2.0 μm pore size filter and centrifugation at 700 x g for 5 min to remove remaining cell debris. The percent of infected cells in the culture was calculated by counting the number of infected and uninfected cells in duplicates Giemsa-stained preparations from the same flask and the total number of cells was determined using a hemocytometer. Cell-free bacteria were incubated for 2 hr at 34°C with 2.5 x 105 ISE6 cells in MatTek chambers (MatTek Corp., Massachusetts) to allow binding and expression of OMT. Unbound bacteria were removed by rinsing cells twice with culture medium. Likewise, bacteria were incubated with 2.5 x 105 HL-60 cells suspended in 500 μl of culture medium for 2 hr at 37°C. Unbound bacteria were removed by washing cells twice and expression of OMT in bound bacteria was analyzed by IFA.
Cell samples were fixed for 10 min in methanol, and incubated with anti-rOMT serum diluted 1:200 in PBS containing 3% bovine serum albumin (BSA) for 2.5 hr at room temperature. Bacteria were labeled with anti-A. phagocytophilum dog serum (diluted 1:1,000). The slides were washed 3 times in PBS and blocked in PBS with 3% BSA for 10 min at room temperature. OMT expressing bacteria were then labeled with anti-mouse antibodies conjugated with AlexaFluor647 (1:500 dilution) (Jackson ImmunoResearch Laboratories, Inc, Pennsylvania) for 1 hr at room temperature. All A. phagocytophilum were labeled with anti-A. phagocytophilum dog serum diluted 1:500 followed by incubation with anti-dog IgG conjugated to FITC, using the same procedure. Tick and HL-60 cell nuclei were labeled using DAPI present in the VectaShield mounting medium. Microscopic images were obtained using an Olympus BX61 disk-scanning unit confocal microscope (Olympus America, Pennsylvania) utilizing a DSU-D2 confocal disk. Confocal images were acquired with a Photometrics Quantem:512SC EMCCD camera (Photometrics, Arizona), and high resolution images were acquired with a QFire color camera (Qimaging, California). Image capture software was Metamorph (Molecular Devices, California). ImageJ (US National Institutes of Health) was used to compile z-projections and Photoshop (Adobe Systems, California) was used for cropping.
HZ wild-type and ΔOMT bacteria were grown in two 50 ml volumes of RPMI1640 medium with HL-60 cells each and the number of infected cells was determined. Host cell free mutant and wild-type bacteria were purified from 2.5 x109 infected cells that were ruptured by repeated passage through a 27 G needle, centrifuged at 600 x g to remove cell debris, and inoculated into a 25-cm2 flask containing 6 x 107 ISE6 cells. This procedure was replicated three times in three independent biological replicates. Cells and bacteria were incubated for 4 hr at 34°C without agitation to allow methylation of possible substrates to be completed, since the time between up-regulation of the gene and completion of the enzymatic reaction was not known. Bacteria were then purified from ISE6 cells as described above. Bacteria were washed in unsupplemented L15C300 medium three times by centrifugation at 16,000 xg for 5 min at 4°C to remove FBS, and the final bacteria pellet was extracted for mass spectrometry.
Protein concentrations were determined by Bradford assay using two aliquots for each sample. All samples were prepared as follows at the Center for Mass Spectrometry and Proteomics at the University of Minnesota: cell pellets were reconstituted with 120 μl of protein extraction buffer [7 M urea, 2 M thiourea, 0.4 M triethylammonium bicarbonate (TEAB) pH 8.5, 20% methanol and 4 mM tris(2-carboxyethyl)phosphine (TCEP)] while on ice. Samples were sonicated at 30% amplitude for 7 sec with a Branson Digital Sonifier 250 (Emerson, Connecticut). The samples were homogenized in a Barocycler NEP2320 (Pressure Biosciences, Inc., Massachusetts) by cycling between 35 k psi for 30 sec and 0 k psi for 15 sec for 40 cycles at 37°C. Samples were alkylated for 15 min at room temperature in 8 mM methyl methanethiosulfonate (MMTS).
In-solution proteolytic digestions were performed as follow: a 200 μg aliquot of each sample was transferred to a new 1.5 ml microfuge tube and brought to the same volume with protein extraction buffer plus 8 mM MMTS. All samples were diluted 4-fold with ultra-pure water, and trypsin (Promega, Wisconsin) was added at a 1:35 ratio of trypsin to total protein. Samples were incubated for 16 hr at 37°C after which they were frozen at -80°C for 30 min and dried in a vacuum centrifuge. Each sample was then cleaned using a 4 ml Extract Clean C18 SPE cartridge (Grace-Davidson, Illinois), and eluates were vacuum dried and resuspended in dissolution buffer (0.5M triethylammonium bicarbonate, pH 8.5) to a final 2 μg/μl concentration. For each iTRAQ 4-plex (AB Sciex, California), two 50 μg replicates for each sample were labeled with iTRAQ reagent (AB Sciex, California). After labeling, the samples were multiplexed together and vacuum-dried. The multiplexed sample was cleaned with a 4 mL Extract Clean C18 SPE cartridge (Mandel Scientific Company Inc., Guelph, Canada) and the eluate was dried in vacuo.
The iTRAQ labeled samples were resuspended in Buffer A (10 mM ammonium formate pH 10 in 98:2 water:acetonitrile) and fractionated offline by high pH C18 reversed-phase (RP) chromatography [75]. A MAGIC 2002 HPLC (Michrom BioResources, Inc., California) was used with a C18 Gemini-NX column, 150 mm x 2 mm internal diameter, 5 μm particle, 110 Å pore size (Phenomenex, California). The flow rate was 150 μl/min with a gradient from 5–35% Buffer B (10 mM ammonium formate, pH 10 in 10:90 water:acetonitrile) over 60 min, followed by 35–60% over 5 min. Fractions were collected every 2 min and uv absorbances were monitored at 215 nm and 280 nm. Peptide containing fractions were divided into two equal numbered groups, labeled “early” and “late”. The first “early” fraction was concatenated with the first “late” fraction by 10 mAU volume equivalents of each fraction from uv = 215 nm and repeated until all fractions were concatenated. Concatenated samples were dried in vacuo, resuspended in 98:2O, H2O:acetonitrile, 0.1% formic acid and 1–1.5 μg aliquots were run on a Velos Orbitrap mass spectrometer (Thermo Fisher Scientific, Inc., Massachussets) as described previously [76] with the exception that the Higher-energy Collisional Dissociation (HCD) activation energy was 20 msec.
The mass spectrometer RAW data (Proteowizard files) were converted as described previously [76]. ProteinPilot 4.5 (AB Sciex, California) searches were performed against the NCBI reference sequence for the I. scapularis (taxon 6945; November 14, 2011) protein FASTA database with (20468 proteins), to which the NCBI reference sequence A. phagocytophilum, HZ (taxon 212042; November 14, 2011; 1267 proteins) and a contaminant database (thegpm.org/crap/index, 109 proteins) was appended. Search parameters were: cysteine MMTS; iTRAQ 8plex (Peptide Labeled); trypsin; instrument Orbi MS (1–3ppm) Orbi MS/MS; biological modifications ID focus; thorough search effort; and False Discovery Rate analysis (with reversed database).
The putative function of differentially expressed proteins identified by iTRAQ was explored using the databases in NCBI (www.ncbi.nlm.nih.gov) to identify conserved domains, Uniprot (http://www.uniprot.org/uniprot/), EMBL-EBI (http://www.ebi.ac.uk/interpro/IEntry?ac=IPR000866), and OMA (http://omabrowser.org). Pathways which involved proteins differentially expressed during infection with the ΔOMT compared to wildtype were identified using the KEGG pathway tool (http://www.genome.jp/kegg/tool/map_pathway1.html).
To identify the peptides that differed in methylation, a text file of data for all peptides was imported into Excel (Microsoft, Washington), and entered into TextWrangler (Bare Bones Software, Massachusetts), which is a text editing tool. Peptides that were less abundant in the mutant when compared to the wild-type, based on the intensities of the reporter ion signals, were selected and the intensities of spectra were visually inspected in the ProteinPilot viewer software (AB SCIEX, Massachusetts) to confirm differences.
A functional version of the OMT without the S-tag (rOMTns) was produced and purified. Recombinant proteins of potential substrates identified by iTRAQ were produced using the pET29a vector. Primers to amplify partial or complete coding sequences were designed with restrictions sites for NdeI and XhoI enzymes (S2 Table), and DNA amplified using pfu enzyme under conditions listed in Table 2. The sequence integrity of purified products was confirmed by Sanger sequencing, and DNA cloned into BL21(D3) E. coli (New England Biolabs) for expression. Proteins were produced in 150 ml of Superior Broth after induction with 200 μM IPTG and used in methylation assays described below. Protein concentrations were measured using the BCA micro protein assay kit (Pierce).
The SAM-fluoro: SAM Methyltransferase Assay (G-Biosciences, Missouri) was used to determine the activity of the enzyme. In this assay, methylation by SAM-dependent methyltransferases is correlated with the production of H2O2 which can be assessed through the production of fluorescent resorufin from 10-acetyl-3,7,-dihydroxyphenoxazine (ADHP). Resorufin production was monitored using an excitation wavelength of 530 nm and an emission wavelength of 595 nm in a Synergy H1 Hybrid microplate reader (Biotek, Vermont) during a kinetic run measuring fluorescence every two min for 4 hr at 34°C. Reactions were carried out in a 96 well EIA/RIA plate flat bottom plate (Costar, New York) covered with a MicroAmp optical adhesive film (Applied Biosystems, New York) to protect samples from evaporation. Assays were performed with 40 ng (7.14 ρmole) of the enzyme and 50 ng (8.31 ρmole) of substrates. The moles of the substrates were calculated from the theoretical molecular weight in kDa, using Zbionet (http://www.molbiol.ru/eng/scripts/01_04.html). Positive controls included the addition of AdoHcy alone or in combination with rOMTns to assay reagents provided by the manufacturer.
To determine enzyme kinetics, the concentrations of the enzyme or the substrate were increased separately. For the first set of reactions, the amount of enzyme was increased to 60 ng (10.71 ρmole), 80 ng (14.29 ρmole), and 100 ng (17.89 ρmole), while the substrate was left at 50 ng. In the second set, the enzyme concentration was left at 40 ng, while the substrate was used at 80 ng (13.29 ρmole), 100 ng (16.61 ρmole), and 150 ng (24.92 ρmole). Each reaction was done in triplicate and the average of the reactions was analyzed. To determine whether the 10 mM Mn2+ included in the kit was a limiting factor for rOMT activity, additional MnCl2 was included at 0.5 mM, 1 mM, 2 mM, 4 mM, 8 mM, and 16 mM concentrations in reactions containing 100 ng of the OMT and 100 ng of the substrate. All reactions were carried out in triplicate using the manufacturer’s recommendations, and the kinetics of the enzyme were analyzed using the Michaelis-Menten equation.
Crystals of OMT from A. phagocytophilum were obtained via the sitting drop vapor diffusion method, where 400 nl of protein solution was mixed with 400 nl of precipitant solution in the sample well and then equilibrated against 80 μl of precipitant in the reservoir well of 96-well Compact 300 crystallization plates (Rigaku Reagents, Washington). For Apo OMT, the protein concentration used was 20 mg/ml and the precipitant solution was 0.2 M MgCl2, 0.1 M TRIS at pH 8.50, and 20% PEG 8000. For SAM-Mn and SAH-Mn bound structures protein at 20 mg/ml was pre-incubated with 2 mM of either SAM or SAH and 10 mM MnCl2 for 1 hr before setting up trays. Final crystallization conditions for SAM-Mn+2 and SAH-Mn+2 were 0.2 M ammonium chloride, 20% PEG 3350 or 0.1 M succinic acid pH 7.0, 15% PEG3350, respectively. All crystallization experiments took place at 16°C.
Crystals were harvested using mounted CryoLoops (Hampton Research, California) and then flash-frozen in liquid nitrogen until data collection. Data for SAH-bound OMT were collected on an in-house FR-E+ Superbright (Rigaku, Washington) rotating anode X-ray generator at a wavelength of 1.54 Å and data for Apo OMT was collected at the LS-CAT 21ID-G beam line at the Advanced Photon Source at a wavelength of 0.9786 Å. Data for the SAM-Mn and SAH-Mn complexes were collect at the LS-CAT 21ID-F beam line at the Advanced Photon Source at a wavelength of 0.9787 Å. All data were indexed, integrated, and scaled using the programs XDS and XSCALE [77]. Data statistics for both datasets are available in Table 5. Phases for structure determination of the SAH-bound OMT were obtained via iodide-SAD using the method previously described by Abendroth et al. [41]. Heavy atom searches, phasing, and density modification were performed using the programs PHENIX HySS [78], Phaser [79], and SOLVE/RESOLVE [80]. Initial model building into density-modified electron density maps was performed using the program ARP/wARP [81]. Phases for Apo OMT were obtained by molecular replacement using the program Phaser, where the SAH-bound OMT structure was used as the search model. Both structures were refined against the reflection data using the programs PHENIX [82] and REFMAC [83] interspersed with rounds of model building using the program Coot [84]. Figs containing molecular graphics were prepared using the program PyMOL (https://www.pymol.org).
To identify the possible origin of the A. phagocytophilum OMT, the corresponding protein sequence available in GenBank for A. phagocytophilum isolate HZ (GI:88607321) was used for a PSI-BLAST. The OMTs with the lowest E-values and highest similarity to the A. phagocytophilum OMT were aligned using ClustalW from MacVector 12.0 (MacVector, Inc, North Carolina). A Minimum Evolution phylogenetic tree of all the OMTs was generated using MEGA 4.0. Conserved motifs within the most closely related non-Anaplasmataceae OMTs, along with the OMT from A. phagocytophilum, were identified using MEME (http://meme.nbcr.net) [85].
Phyre2 [42] was used to determine the putative tertiary structure of the protein (Msp4) that was identified as being methylated by iTRAQ analysis as well as the in vitro methylation assay. The putative localization of the modified residues was determined from the protein sequence and the structure generated from Phyre2. Phobius (http://phobius.sbc.su.se/cgi-bin/predict.pl) was used to determine where the modified residues were located within the membrane of the bacteria since Msp4 is a surface protein [61].
This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. None of the procedures used caused more than momentary pain, and infection with A. phagocytophilum does not cause illness in mice. Animals were euthanized with CO2 before collecting blood for production of serum following current AVMA guidelines. The protocol was approved by the institutional Animal Care and Use Committee of the University of Minnesota (Protocol ID: 1303-30435A).
The NCBI accession numbers for A. phagocytophilum proteins mentioned in the body of the article are the following: O-methyltransferase (GI:88598384); Major Surface Protein 4 (GI:88607879); Hypothetical protein APH_0406 (GI:88607117); OmpA family member (GI:88607566); P44_18ES (GI:88607256); Major Surface Protein 5 (GI:88607263); DNA-dependent RNA polymerase subunit B (GI:88607105); Antioxidant AhpCTSA family protein (GI:88607183); Ankyrin (GI:88607707); P44-1 outer membrane protein (GI:88607426); OMP85 (GI:88607567); Hypothetical protein APH_0405 (GI:88607654); Chaperonine GrpE (GI:88607566); Chaperonin GroEL (GI:88606723); DnaK (GI:88607549); Cytochrome C oxidase subunit II (GI:88607721); TypA (GI:88607727); Outer membrane protein P44-16b (GI:88607043); Preprotein translocase subunit SecA (GI:886078849).
The NCBI accession numbers for proteins from other organism are the following: B. bacteriovorus OMT (GI:426402377); H. ochraceum OMT (GI:262196431); Anaeromixobacter sp. OMT (GI:197124171); C. M. mitochondrii OMT (GI:339319550); and Gloeocapsa sp. (GI:434391334).
The PDB accession numbers for the crystal structures mentioned in the article are the following: A. phagocytophilum OMT + SAH (4OA5); A. phagocytophilum OMT Apo (4OA8); A. phagocytophilum OMT + SAM + Mn+2 (4PCL); A. phagocytophilum OMT + SAH + Mn+2 (4PCA); and OMT C. synechocystis (3CBG).
|
10.1371/journal.ppat.1007566 | MUC1 is a receptor for the Salmonella SiiE adhesin that enables apical invasion into enterocytes | The cellular invasion machinery of the enteric pathogen Salmonella consists of a type III secretion system (T3SS) with injectable virulence factors that induce uptake by macropinocytosis. Salmonella invasion at the apical surface of intestinal epithelial cells is inefficient, presumably because of a glycosylated barrier formed by transmembrane mucins that prevents T3SS contact with host cells. We observed that Salmonella is capable of apical invasion of intestinal epithelial cells that express the transmembrane mucin MUC1. Knockout of MUC1 in HT29-MTX cells or removal of MUC1 sialic acids by neuraminidase treatment reduced Salmonella apical invasion but did not affect lateral invasion that is not hampered by a defensive barrier. A Salmonella deletion strain lacking the SiiE giant adhesin was unable to invade intestinal epithelial cells through MUC1. SiiE-positive Salmonella closely associated with the MUC1 layer at the apical surface, but invaded Salmonella were negative for the adhesin. Our findings uncover that the transmembrane mucin MUC1 is required for Salmonella SiiE-mediated entry of enterocytes via the apical route.
| The bacterial pathogen Salmonella enterica is one of the most common causes of human foodborne infection affecting millions of people worldwide each year. To establish infection, Salmonella needs to cross the mucus layer and invade intestinal epithelial cells from the apical surface. However, the apical surface of intestinal epithelial cells is covered with a defensive barrier of large glycosylated transmembrane mucins. These large proteins prevent contact between the Salmonella type III secretion needle and the host plasma membrane thereby preventing invasion. We show for the first time that MUC1, one of the intestinal apical transmembrane mucins, facilitates Salmonella invasion. The Salmonella giant adhesin SiiE is the adhesin responsible for engaging MUC1 and the interaction is mediated by glycans on MUC1. We propose that SiiE interacts with MUC1 in a zipper-like manner that involves repetitive domains in both proteins. Adhesin-receptor interactions are essential for bacterial infection of host cells and key factors in determining target tissues and host range of bacteria. The SiiE-MUC1 invasion pathway may explain tropism of different Salmonella strains and provide a novel target for infection intervention and prevention.
| In the gastrointestinal tract, the luminal microbiota is separated from the underlying epithelial cells by a complex system collectively called the mucus layer. The mucus layer consists of soluble gel-forming mucins such as MUC2 and MUC5A that are secreted by Goblet cells, IgA antibodies, host defense peptides, and other anti-microbial components [1]. Another component of the mucus layer are transmembrane mucins, which are large glycoproteins that are expressed on the apical surface of enterocytes and Goblet cells. Transmembrane mucins expressed in the gastrointestinal tract include MUC1, MUC3A, MUC3B, MUC4, MUC12, MUC13, MUC15, MUC17, MUC20 and MUC21 [2]. Transmembrane mucins have a highly glycosylated extracellular domain with potential barrier function, a transmembrane domain and a cytoplasmic tail that links to signaling pathways [3].
MUC1 is the most extensively studied transmembrane mucin and is highly expressed at mucosal surfaces including the stomach and the intestinal tract [4,5]. The MUC1 extracellular domain forms a large filamentous structure with a variable numbers of tandem repeats (VNTR) domain that can protrude 200–500 nm from the plasma membrane [6,7]. The extracellular domain is highly O-glycosylated with complex sugars that frequently terminate with sialic acids or fucose [8]. The human and mouse MUC1 extracellular domains share less than 40% homology while the transmembrane domain and cytoplasmic tail are highly conserved [9].
MUC1 plays an important role in defense against invasive bacterial pathogens such as Helicobacter pylori and Campylobacter jejuni. In vitro experiments with H. pylori and a gastrointestinal cell line showed that the extracellular domain of MUC1 is released and acts as a decoy that prevents bacterial attachment to cells [10]. Overexpression of MUC1 in HeLa cells or HCT116 cells protects against C. jejuni Cytolethal Distending Toxin (CDT) and CDT-treated cells internalize MUC1 into cytoplasmic vesicles or into the nucleus [11]. Expression of MUC1 in HCT116 cells increased adherence of C. jejuni, but invasion was not affected by MUC1 expression. The authors showed that C. jejuni adheres to O-glycan H type 2 sugars that contain a terminal fucose group [11]. In in vivo infection experiments, Muc1 knockout mice showed increased susceptibility to H. pylori and C. jejuni with more severe epithelial damage [10–12], but did not display increased susceptibility to Salmonella Typhimurium infection [11]. In addition to bacterial pathogens, MUC1 (over)expression also reduced infection by adenoviruses and influenza A [13–15].
Salmonella enterica is a food-borne, motile and facultative gastrointestinal pathogen. The non-typhoidal Salmonella (NTS) strains, S. enterica subsp. enterica serovar Enteritidis (S. Enteritidis) and S. enterica subsp. enterica serovar Typhimurium (S. Typhimurium) can cause self-limiting gastroenteritis in a wide range of hosts [16]. Three main routes have been described for Salmonella mucosal invasion: entry through M cells, direct invasion of enterocytes, and uptake through dendritic cells [17]. Salmonella cellular invasion is mediated by a type III secretion system that injects virulence factors into host cells to induce uptake. This process is well-studied for Salmonella invasion of different types of epithelial cells [18]. During intestinal pathogenesis Salmonella encounters the apical surface of intestinal epithelial cells where invasion is less efficient due to a defensive barrier of glycosylated transmembrane mucins. The transmembrane mucins presumably prevent contact of the T3SS with the host plasma membrane. The Salmonella giant adhesin SiiE that is secreted by the TISS and encoded by Pathogenicity Island 4 (SPI4) has been shown to be a key factor in apical invasion of polarized epithelial cells [19,20]. A specific receptor for SiiE that enables Salmonella apical entry has not been identified. In this study, we investigate the role of transmembrane mucin MUC1 during Salmonella invasion into intestinal epithelial cells.
Various human intestinal epithelial cell lines derived from colorectal carcinomas are used to study interactions with enteric pathogens, e.g. Caco-2, HT-29, HT29-MTX and HRT-18 cells. Previously it was shown that Salmonella apical adhesion and invasion is more effective in HT29-MTX cells compared to Caco-2 or HT-29 cells [21]. For this reason, we selected HT29-MTX cells to study the function of MUC1 during Salmonella invasion. To investigate MUC1 expression in HT29-MTX cells, we grew the cells for 5 days to form a confluent monolayer and performed immunofluorescence confocal microscopy with the 214D4 antibody directed against the MUC1 extracellular domain. MUC1 was expressed in a high percentage of HT29-MTX cells and localized to the apical side in an island-like pattern (Fig 1A).
To facilitate confocal microscopy of Salmonella invasion, we used S. Enteritidis containing a plasmid encoding red fluorescent mCherry. S. Enteritidis was grown till late logarithmic phase to achieve a high motile state and co-incubated with confluent HT29-MTX cells for 1 h. In this experimental setup, only the apical surface of the HT29-MTX cells is exposed to Salmonella. As revealed by confocal microscopy, S. Enteritidis invaded and formed clusters predominantly in MUC1 positive cells and did not invade adjacent MUC1-negative cells (Fig 1B). MUC1 seemed to localize to cup-like structures that contained large Salmonella clusters (Fig 1B). Multiple Z-stacks of the image were collected, and an orthogonal view clearly showed apical MUC1 with clusters of Salmonella located underneath (Fig 1C and S1A and S1B Fig). MUC1 did not colocalize with the intracellular Salmonella. Next, we investigated the contribution of the T3SS to apical invasion of HT29-MTX cells with a Salmonella Enteritidis CVI-1 wild type strain and its isogenic mutant invG, that is deficient in T3SS-mediated invasion. While the wild type strain did invade HT29-MTX cells at MUC1-positive loci, no adherent or invaded invG mutant bacteria could be detected (S1C Fig).
To determine if this phenomenon is restricted to S. Enteritidis, we also tested S. Typhimurium, another non-typhoidal Salmonella. We used S. Typhimurium containing a plasmid encoding GFP [22,23] and co- incubated the bacteria with confluent HT29-MTX cells. Comparable to S. Enteritidis, S. Typhimurium invaded and formed clusters in MUC1-expressing cells while MUC1-negative cells remained uninfected (Fig 1D). Again, MUC1-positive cup-like structures containing Salmonella clusters were observed. From these data we hypothesize that MUC1 positively contributes to the apical invasion of different S. enterica serovars into HT29-MTX cells.
To determine the role of MUC1 in S. Enteritidis entry, we firstly designed a CRISPR/Cas9 genome editing strategy to knock out MUC1 in HT29-MTX cells (Fig 2A). Two guide RNAs (gRNAs) were selected in the 5’ end of the MUC1 gene that would generate a 130 bp deletion before the start of the tandem repeats. After transfection with the CRISPR-2xgRNA plasmid, cells were selected with puromycin and a positive single cell clone with the 130 bp deletion was identified (HT29-MTX-ΔMUC1; Fig 2B). Western blot analysis of wild type HT29-MTX cells revealed a reactive band of approximately 460 kDa that most likely represents fully glycosylated MUC1 and a band of about 170 kDa that presumably represents the unglycosylated precursor. Both bands were absent in the ΔMUC1 cells (Fig 2C).
Next, we performed Salmonella apical invasion experiments with the wild type and ΔMUC1 monolayers followed by confocal microscopy. S. Enteritidis invaded and formed clusters in the wild type cells as described above, but the bacteria barely invaded the ΔMUC1 cells (Fig 2D). To quantify this difference, we performed a quantitative invasion assay in which extracellular bacteria are killed with gentamicin and viable internalized bacteria are counted. This assay showed that approximately 41% of the introduced S. Enteritidis invaded the wild type cells, while about 11% of the bacteria invaded ΔMUC1 cells (p<0.01) (Fig 2E). The quantitative assay yielded a less pronounced effect compared to the microscopy. We hypothesize that this discrepancy is due to false positives in the quantification assay, a known problem associated with such assay that might be more pronounced during invasion at the apical surface. These data indicate that MUC1 does not form a barrier for Salmonella invasion but facilitates apical entry into HT29-MTX cells.
As Salmonella utilizes different strategies for apical or (baso)lateral invasion into intestinal epithelial cells, we also carried out invasion assays with 2-day grown non-confluent HT29-MTX WT and ΔMUC1 cells. These cells formed small islands with exposed lateral sides on the rim. MUC1 was only expressed in a limited number of cells in the center of the island. S. Enteritidis invaded both wild type and ΔMUC1 cell islands but showed a very different pattern. The large S. Enteritidis clusters described above were found exclusively in MUC1-expressing cells. A high percentage of ΔMUC1 cells was infected with S. Enteritidis through the lateral route, but only small red puncta of single bacteria could be observed (Fig 2F). In the apical invasion experiments with the confluent monolayers such small puncta were not observed. A quantitative invasion assay with non-confluent cells showed no difference in S. Enteritidis invasion between wild type and ΔMUC1 cells (Fig 2G). The formation of Salmonella clusters during apical invasion was previously described and only occurs in areas of membrane ruffle formation and macropinocytosis [24]. Our data suggest that MUC1 is associated with this apical process and that lateral invasion of Salmonella that is independent of MUC1 results in a different outcome. Collectively, these data demonstrate that MUC1 is an essential component for apical invasion of S. Enteritidis into confluent HT29-MTX cells but is not required for lateral entry of S. Enteritidis into non-polarized cells. Apical invasion through MUC1 appears to be very efficient and results in uptake of large clusters of Salmonella while individual bacteria internalize during lateral invasion.
To further substantiate the MUC1 contribution to Salmonella invasion, we performed apical invasion assays with early confluent monolayers of HT-29, Caco-2 and HRT-18 intestinal epithelial cell lines. Confocal microscopy showed that, under these conditions, HT-29, Caco-2 and HRT-18 cells express relatively low levels of MUC1 compared to HT29-MTX cells. Immunoblot analysis to determine MUC1 expression level in all cell lines indicated clear MUC1 bands in HT29-MTX and HT-29 cells, whereas MUC1 expression was below detection level in the other two cell lines after 5 days of culture (Fig 3A). Invasion assays showed that Salmonella did invade and form clusters in the MUC1-positive cells in the different cell lines although invasion levels seemed lower than detected for HT29-MTX cells (Fig 3B). MUC1 is reported to be upregulated under inflammatory conditions, for example after exposure to IL-6 [25,26]. Treatment of the early confluent monolayers with IL-6 strikingly increased MUC1 expression in HT29-MTX cells and HT-29 cells, but no difference was observed for Caco-2 and HRT-18 cells (Fig 3A). These results demonstrate variable MUC1 expression between cell lines and conditions and explain differences in apical invasion behavior of the pathogen between different intestinal epithelial cell lines.
Alpha-2,3-linked sialic acids are essential for apical invasion of S. Typhimurium into MDCK cells [20]. The tandem repeats of the MUC1 extracellular domain are highly glycosylated with complex O-linked glycans on which sialic acids are frequently present [8]. To investigate if sialic acids on MUC1 are involved in apical invasion of S. Enteritidis into HT29-MTX cells, we removed α2,3-, α2,6-, and α2,8-linked sialic acids by neuraminidase treatment prior to the addition of Salmonella. Confocal microscopy showed that neuraminidase treatment of the HT29-MTX wild type cells resulted in a reduced S. Enteritidis invasion into wild type cells and a disappearance of Salmonella clusters (Fig 4A). Neuraminidase treatment did not affect Salmonella apical invasion into ΔMUC1 cells (Fig 4B). The quantitative invasion assay confirmed the significant reduction in S. Enteritidis invasion into wild type cells after neuraminidase treatment, whereas no significant difference was observed for ΔMUC1 cells (Fig 4B). Again, we did see that the quantitative assay yielded a less pronounced effect compared to the microscopy. Together these data clearly indicate that sialic acids on the extracellular domain of MUC1 are crucial for apical entry of S. Enteritidis into intestinal epithelial cells.
S. enterica binding to receptors on epithelial cells is mediated by different adhesins. SiiE is a giant adhesin that mediates binding of Salmonella to the apical side of epithelial cells [19,27]. To determine if SiiE is the adhesin that contributes to binding of MUC1, we generated a S. Enteritidis siiE knockout strain. This strain demonstrated the same growth rate as the wild type strain. Wild type and siiE knockout bacteria were used in apical invasion assays with confluent HT29-MTX wild type cells. As shown by microscopy, the siiE knockout bacteria failed to invade HT29-MTX wild type cells (Fig 5A). To exclude the possibility that deletion of siiE affects Salmonella motility, we also performed a lateral invasion assay with non-confluent cells. Under these conditions, the wild type bacteria invaded from both the lateral side and the apical surface and the latter was associated with internalization of large bacterial clusters. The siiE knockout bacteria did invade the cells but no large bacterial clusters were detectable (Fig 5B). These data suggest that invasion through MUC1 was completely abolished in the siiE mutant bacteria. Quantification of Salmonella infection showed that invasion of the siiE mutant into HT29-MTX wild type cells was significantly reduced compared to S. Enteritidis wild type bacteria (p<0.05) (Fig 5C). Invasion of the siiE mutant into ΔMUC1 cells was low and not significantly altered (Fig 5C). Together, these data demonstrate that both MUC1 and SiiE are essential components for efficient apical entry of intestinal epithelial cells by Salmonella.
It has been reported previously that Salmonella secretes SiiE into the culture medium, but that the adhesin remains attached to the bacterial cell wall when contacting host cells [19]. In order to further investigate the function of SiiE during MUC1-mediated apical invasion we performed immunofluorescence confocal microscopy with an antibody directed against SiiE (kindly provided by Michael Hensel). As in our previous invasion experiments, Salmonella invaded MUC1-positive cells and the bacteria could be observed underneath the MUC1 surface. While SiiE-positive bacteria were closely associated with the MUC1 layer, Salmonella that had invaded further into the epithelial cells were negative for SiiE (Fig 6A, S2 Fig). SiiE-positive Salmonella associated closely with the MUC1-positive cups (Fig 6B). To visualize the process of SiiE-MUC1 mediated Salmonella invasion in more detail, we generated a 3D projection and an accompanying movie that includes a rotation (Fig 6C and S1 Movie). Using Imaris software, we determined the position of SiiE-positive and Salmonella-positive spots relative to the MUC1 surface layer in 5 independent Z-stacks. The SiiE-positive spots closely associated with the MUC1 layer, while the Salmonella spots localized close to the MUC1 layer and deeper into the cells (Fig 6D), conforming that invaded Salmonella are negative for SiiE. Together, our data show that glycosylated MUC1 acts as a receptor for Salmonella SiiE and enables SiiE-dependent apical invasion into the intestinal epithelium (Fig 7A).
Excellent work has shown that Salmonella T3SS enables cellular invasion, but bacterial entry into intestinal epithelial cells via the apical route appear less efficient due to a defensive barrier formed by transmembrane mucins. We demonstrate for the first time that Salmonella utilizes the transmembrane mucin MUC1 as a receptor to breach the apical barrier of intestinal epithelial cells. Salmonella efficiently invaded the apical surface of a monolayer of intestinal cells with high MUC1 expression whereas apical invasion into MUC1 knockout cells and cell lines that expressed low levels of MUC1 was severely reduced (Figs 2 and 3). Lateral invasion of non-confluent cells is not hampered by a defensive mucin layer and does not require MUC1 for entry (Fig 2). Enzymatic removal of sialic acids from intestinal epithelial cells reduced Salmonella invasion to MUC1 knockout levels, showing that sialic acids are instrumental for the MUC1-Salmonella interaction (Fig 4). MUC1-mediated invasion was abolished after the knock out of the Salmonella giant adhesin SiiE (Fig 5). SiiE-positive Salmonella closely associated with the MUC1 layer on the surface and invaded Salmonella were negative for SiiE (Fig 6). We propose that the SiiE-MUC1 invasion route is crucial to breach the epithelial mucin barrier and enable Salmonella apical invasion into MUC1-expressing enterocytes. A T3SS-1 mutant was unable to adhere or invade through MUC1, demonstrating that both the SiiE-MUC1 interaction and a functional T3SS are instrumental for successful Salmonella apical invasion.
Our data provide evidence that glycosylated MUC1 serves as a receptor for intestinal infection in several non-typhoid Salmonella enterica serovars. Different transmembrane mucins are expressed on the apical surface of intestinal epithelial cells and are generally considered to form a defensive barrier. The large filamentous glycoprotein MUC1 is widely expressed in mucosal tissues. Within the gastrointestinal tract, it is expressed in the stomach, duodenum and colon [2]. Previously, it was shown that MUC1 plays an important role in defense against the common enteric pathogens H. pylori and C. jejuni [10–12]. In contrast, our results showed that Salmonella exploits MUC1 for apical invasion into enterocytes. It is conceivable that other bacteria also make use of MUC1 for adhesion and invasion into host cells. For example, it has been reported that knockdown of MUC1 decreases invasion of Staphylococcus aureus into human corneal-limbal epithelial (HCLE) cells [28]. On the other hand, knockdown of the very long transmembrane mucin MUC16 increased invasion of S. aureus, suggesting that some mucin types have a barrier function for this bacterium. As the MUC protein family consists an array of transmembrane proteins with different extracellular and intracellular domains, it can be imagined that pathogens have evolved distinct strategies to breach the mucus layer and colonize the host.
The importance of MUC1 for Salmonella invasion via the apical route was further underpinned by the variable Salmonella entry in several commonly used intestinal epithelial cell lines. Salmonella apical infection efficiency clearly coincided with the level of MUC1 expression by these cell types. In non-polarized cells, Salmonella was capable of lateral invasion via a MUC1-independent route. These findings indicate that Salmonella can invade a single cell type through different mechanisms, dependent on the cell polarization status. We hypothesize that the SiiE-MUC1 interaction precedes contact of the T3SS with the host plasma membrane (see below).
Salmonella has been reported to exploit different mechanisms to invade the intestinal mucosa: (i) invasion through enterocytes, (ii) entry through M cells and (iii) uptake by mucosal dendritic cells [17]. It is conceivable that these seemingly redundant invasion mechanisms serve to colonize different regions of the gut and/or different species. Mouse and rabbit studies show that entry through M cells is a major route of Salmonella invasion [29,30]. In vivo studies in rabbits, calves and pigs have emphasized the importance of enterocyte invasion in addition to M cells [31–33]. To our knowledge, it is not known if M cells express MUC1. Moreover, MUC1 differs in composition and expression between species which theoretically may impact the preferred route of Salmonella invasion and tropism. For example, mouse Muc1 shares less than 40% homology with human MUC1 in the extracellular domain [9] and the pattern of protein glycosylation along the GI tract is opposite between mice and men. In humans, fucose decreases and sialic acid increases from ileum to colon [34], whereas in mice sialylated structures are the dominant terminal sugars in the small intestine and terminal fucose is more prevalent in the colon [35]. These differences may explain the need for different invasion routes but also make it complex to extrapolate results and to identify the predominant route of Salmonella invasion in different niches and species.
We identified the bacterial SiiE protein as the primary Salmonella adhesin involved in MUC1-mediated apical invasion. Genetic inactivation of the siiE gene reduced Salmonella entry from the apical side (Fig 5). The SiiE protein is expressed by most S. enterica serovars, but is a pseudogene in some highly host-adapted serotypes including S. Typhi [36]. The protein is encoded by Salmonella Pathogenicity Island 4 (SPI4) and secreted through a type I secretion system (T1SS) [19]. Association of SiiE with the bacterial surface is regulated by contact with polarized epithelial cells [37] and the protein interacts with MDCK cells in a lectin-like manner [20]. The sensitivity of the SiiE-mediated invasion for neuraminidase treatment (Fig 4) suggests that sialic acids on MUC1 are likely the primary target of SiiE. In colon, MUC1 is decorated with core 3 and core 4 glycostructures, with terminal fucoses or sialic acids [2,8] and this glycosylation pattern can be altered during S. Typhimurium infection [38]. In addition, MUC1 is upregulated under inflammatory conditions (Fig 3) [25,26] thereby increasing receptor availability. SiiE has been shown to contribute to Salmonella infections in mice. S. Enteritidis and S. Typhimurium strains lacking the SPI4 locus show a defect in intestinal colonization but are capable of establishing systemic infection [39].
A crucial aspect in the SiiE-MUC1 interaction is likely the structural architecture of both molecules. The SiiE protein is a giant adhesin (595 kDa) that contains an array of tandem repeats [37]. The extracellular domain of MUC1 also contains a (glycosylated) tandem repeat region. It can be imagined that the SiiE-MUC1 interaction involves sequential binding of the tandem repeats of both the adhesin and receptor. Such an interaction perfectly fits the need to breach the mucin barrier to correctly position the type III secretion system (T3SS) with its injectable virulence factors that are essential to initiate Salmonella invasion. It is important to consider that the MUC1 extracellular domain extends far from the plasma membrane and forms a barrier that is about 200–500 nm in size (6, 7). The T3SS injectosome, a needle-like machinery that is essential for successful invasion, is 80 nm in length [40]. Therefore, Salmonella needs additional virulence factors to breach the transmembrane mucin barrier and position the needle closer to the cell surface. SiiE has a rod-like structure and is 175 ± 5 nm in length [37]. Sequential interaction of the 53 highly conserved SiiE bacterial immunoglobulin (BIg) domains with the 42 highly glycosylated tandem repeats on MUC1 is likely a crucial first step to allow the T3SS system to initiate the invasion process. This scenario is supported by reports that increased deletion of the Salmonella SiiE Blg domains reduces bacterial invasion [41]. The mucus-binding protein MUB of Lactobacillus reuteri also contains multiple Blg-like repeats that bind intestinal mucus [42,43]. We hypothesize that SiiE interacts with MUC1 in a zipper-like manner (Fig 7B). Interaction of a single Blg domain with a single sialic acid is probably of low affinity, but multivalency of the adhesin-sugar interactions will provide strong avidity. We observed that during Salmonella invasion, MUC1 is localized to cup-like structures that contain large clusters of Salmonella. It remains to be investigated if MUC1 plays an active or a passive role in cup formation and bacterial uptake. We hypothesize that the SiiE-MUC1 interaction positions the T3SS needle close enough to the cell surface to inject its virulence factors and induce ruffle formation and Salmonella uptake. It is tempting to speculate that a cycle of SiiE release and reattachment of a novel molecule would enable the bacteria to move closer to the epithelial cell surface. Our hypothesis extends the previously proposed model for the function of SiiE in Salmonella apical invasion by providing evidence that MUC1 serves as the receptor for the SiiE adhesin.
In conclusion, we demonstrate for the first time that the transmembrane mucin MUC1 facilitates Salmonella apical entry into intestinal epithelial cells and serves as a receptor for the Salmonella giant adhesin SiiE. The exploitation of MUC1 by Salmonella underlines the importance of transmembrane mucins as targets in bacterial pathogenesis.
Salmonella enterica serovar Enteritidis (S. Enteritidis) (strain 90-13-706, CVI, Lelystad) was transformed with plasmid pTVmCherry carrying the mCherry gene of Discosoma sp. optimized for bacterial expression (generously provided by J.M. Wells, Wageningen University). S. Enteritidis strains CVI-1 and CVI-1 invG were described previously [44]. Both strains were transformed with the pJET-mCherry plasmid (a kind gift from Mark Wösten, Utrecht University). Salmonella enterica serovar Typhimurium (S. Typhimurium) SL1344 carrying plasmid pMW85 expressing GFP from a PpagC promoter has been previously described [22,23]. Salmonella strains were routinely cultured at 37°C on LB agar plates containing kanamycin at 50 μg/ml or ampicillin at 100 μg/ml in 10 ml of LB broth while shaking (160 rpm). Escherichia coli DH5α used for cloning was grown on LB plates with the appropriate antibiotics.
The S. Enteritidis siiE knockout strain was constructed by deletion of the siiE gene using the lambda red homologous recombination system [45] combined with flanking regions similar to a strategy that was previously published [19]. The two regions flanking the SiiE gene were amplified with primers (KS290 5’- GTAGCATGCCAAAGGTATAGAACTCAAAAAGG-GTATCTGGA-3’ and KS291 5’- TACGGATCCACTCTCAAGGTGTATCTAATCGTTTAGT-3’ and KS292 5’-GTAGGATC-CCTCACCTTTGGGTGAGGGGGTTTAC-3 and KS293 5’-TACGTC-GACCTTCTGAGATAAAAATATTCCTGTTCTTCT-GTCC-3’) using the wild type S. Enteritidis genome DNA as a template and fused the respective sites of the spectinomycin adenylyltransferease-encoding gene [46] using overlap extension PCR. The final product was ligated into the pKO3 vector carrying a chloramphenicol resistance gene and electroporated into S. Enteritidis. After recovery in SOC medium (1h, 30°C), the bacteria were plated on LB plates with chloramphenicol and incubated at 43°C. The next day, single colonies were suspended into 1 ml of LB broth, serially diluted, and immediately plated on 8% sucrose-spectinomycin plates and incubated at 30°C. Gene replacement was verified by PCR with primers KS295 5’- GTTCATGGTCAGGGCGTTAT-3’ and 452 5’-GGCTG-CTCAAACTATACCAC-3’, 451 5’-AGGAGTATTTAAGCGAAGCAC-3’ and KS296 5’-GGAAATACGGCCAGAGACAAT-3’ and KS295/KS296.
The human gastrointestinal epithelial cell lines HT29-MTX (a kind gift of Dr. Thécla Lesuffleur) [47], HT29-MTX-ΔMUC1 (derived from HT29-MTX wild type, this study), HT-29 (ATCC-HTB-38), Caco-2 (ATCC-HTB-37) and HRT-18 (ATCC-CCL-244) were routinely grown in 25 cm2 flasks in Dulbecco’s modified Eagle’s medium (DMEM) containing 10% fetal calf serum (FCS) at 37°C in 5% CO2. For Salmonella quantitative invasion assays, HT29-MTX wild type and ΔMUC1 cells were split into 12-well plates and grown for 5 days prior to addition of Salmonella. For microscopic cell imaging, cells were cultured on circular glass coverslips in 24-well plates.
To generate a ΔMUC1 cell line, we used the pCRISPR-hCas9-2xgRNA-Puro plasmid [48] that encodes Cas9 with 2 guide RNAs to generate a 130 bp deletion in the 5’ end of the MUC1 gene before the VNTR region. The pCRISPR plasmid was digested with SapI and simultaneously dephosphorylated with alkaline phosphatase (FastAP; ThermoFisher). Guide RNA primer sets A (KS36 5’-ACCGGGTCATGCAAGCTCTACCCC-3’ and KS37 5’- AACGGGGTAGAGCTTGCATGACCC 3’) and B (KS131 5’-CCGGACATCCTGTCCCTGAGTGGG-3’ and KS132 5’- AAACCCACTCAGGGACAGGATGTC-3’) were phosphorylated with T4 polynucleotide kinase (ThermoFisher) at 37°C for 30 min and annealed by cooling down from 85°C to 25°C at 0.1°C/sec. Annealed primer sets were ligated into the SapI-digested pCRISPR plasmid as confirmed by sequencing with primers KS46 5’- GTTCACGTAGTGCCAAGGTCG-3’ and KS47 5’-GAGTCAGTGAGCGAGGAAGC-3’, resulting in plasmid pCR11. Two-day grown HT29-MTX cells were trypsinized from a 25 cm2 flask and transfected in suspension with 2 μg of pCR11, pCRISPR-empty or no plasmid using Fugene (Promega) according to the manufacturer’s instructions. Cells were cultured in DMEM + 10% FCS for two days, after which 5 μg/ml puromycine (Life Technologies) was added to the medium to select for positively transfected cells. Cells were maintained in medium with puromycine until all negative control cells had died. Single cell cloning was performed by serial dilution and single cell clones were tested for the MUC1 deletion by PCR with primers KS133 5’-CAGTCCTCCTGGTATTATTTCTCTGGTG-3’ and KS134 5’- CAGGTGGCAGCTGAACCTGAAG-3’. The absence of MUC1 protein in the HT29 MTX-ΔMUC1 cell line was confirmed by immunoblot with mouse monoclonal antibody 214D4 (a kind gift from Dr. John Hilkens; CD227, Nordic MUBio) directed against the MUC1 tandem repeats.
Cells were grown on cover slips (8 mm diameter #1.5) in 24-well plates and for bacterial invasion studies, Salmonella was grown as described below and added at a multiplicity of infection (MOI) of 60 for 1 h. Then cells were washed twice with Dulbecco’s Phosphate Buffered Saline (DPBS, D8537, Sigma) and fixed with 4% cold paraformaldehyde in PBS (Affimetrix) for 30 min at room temperature. Cells were rinsed twice with DPBS before they were permeabilized in binding buffer (0.1% saponin, Sigma and 0.2% BSA, Sigma in DPBS) for 30 min. Next, coverslips were incubated with α-MUC1 antibody 214D4 at 1:150 dilution and/or α-SiiE rabbit polyclonal antiserum (1:200; a kind gift of Michael Hensel, University of Osnabrück) for 1h followed by 4 washing steps with binding buffer. The cover slips were incubated with Alexa Fluor-488/568-conjugated goat α-mouse IgG secondary antibodies (1:200; A11029, A11031; ThermoFisher), goat α-rabbit IgG secondary antibodies (1:200; A11034, A11036; ThermoFisher) and Alexa Fluor-647-conjugated donkey α-mouse IgG secondary antibody (1:200; 715-605-151; Jackson ImmunoResearch) and DAPI at 2 μg/ml (D21490, Invitrogen) for 1h. Coverslips were washed 3 times with DPBS, once with MilliQ, dried and embedded in Prolong diamond mounting solution (ThermoFisher) and allowed to harden. Images were collected on a Leica SPE-II confocal microscope using a 63x objective (NA 1.3, HCX PLANAPO oil) controlled by Leica LAS AF software with default settings to detect DAPI, Alexa488, Alexa568 and Alexa647. Axial series were collected with step sizes of 1 μm. Object centroid positions in Z were determined in IMARIS 8.2 (Bitplane, UK) on images that were deconvolved using the 3D automated deconvolution package in NIS elements (NIKON). Salmonella and SiiE segmented objects were created using the spot detection wizard and MUC1-labeled positions were isolated using the surface detection wizard in 5 image stacks. Centroid Z positions of the of the objects were extracted in the Vantage module and exported to Excel (Microsoft, Redmond USA) for plotting.
Salmonella overnight cultures (16 h) were diluted 1:30 in LB broth and grown for 3.5 h at 37°C till late logarithmic phase. After adjustment of the OD600 to 0.24, 1 ml of bacterial culture was centrifuged at 8000 rpm for 2 min. The bacterial pellet was resuspended in 1 ml DPBS, immediately 30 μl of bacterial suspension was added to cells in a 12-well plate in DMEM without FCS (MOI 15) and incubated for 1 h at 37°C. Cells were rinsed twice with DPBS, placed in DMEM without FCS containing 300 μg/mL of gentamicin, and incubated for 1 h at 37°C to kill extracellular bacteria. Cells were then washed twice with DPBS and lysed with lysis buffer (0.1% Triton X-100 in DPBS) for 5 min at 37°C. Serial dilutions were made and plated on LB agar plates containing 50 μg/mL kanamycin. The next day, colonies were counted.
Protein samples were prepared from cells cultured in 6-wells plates. The cells were washed once with DPBS and 200 μl of lysis buffer was added to each well to detach the cells. Then 100 μl of 3x Laemmli sample buffer was added to the lysate followed by boiling for 5 min at 100°C. For immunoblotting of large mucin proteins, a Boric acid-Tris system was developed. A 5% mucin gel (12.5% 40% Acryl/Bis acryl, Biorad 161–0144; 26% 1.5 M Tris pH 8.8, Invitrogen; 10% SDS, Invitrogen; 10% ammonium persulfate, Invitrogen; 0.1% TEMED, ThermoFisher) was made in a Mini Protean II chamber (Bio-rad) using 1.5 mm spacer plates. Boiled protein lysates were loaded onto the gel, and run in Boric acid-Tris buffer (192 mM Boric acid, Merck; 1 mM EDTA, Merck; 0.1% SDS, to pH 7.6 with Tris) at 25 mA for 1.5 h. Protein was transferred onto a nitrocellulose membrane using a wet transfer system with transfer buffer (25 mM Tris; 192 mM glycine, Merck; 20% methanol, Merck) for 3 h at 90 V at 4°C. Subsequently, the membranes were blocked with 5% BSA in TSMT (20 mM Tris; 150 mM NaCl, Merck; 1 mM CaCl2, Sigma; 2 mM MgCl2, Merck; adjusted to pH 7 with HCl; 0.1% Tween 20, Sigma) overnight at 4°C. The next day, the membrane was incubated with 214D4 antibody at a dilution of 1:150 in TSMT containing 1% BSA for 1 h at RT, washed 2 times with TSMT, 2 times with DPBS and incubated with α-mouse IgG secondary antibody (A2304, Sigma) diluted 1:8000 in TSMT with 1% BSA for 1 h at RT. For detection of actin, cell lysates were loaded onto a 10% SDS-PAGE gel, transferred to PVDF membranes and incubated with α-Actin antibody (1:5,000; bs-0061R, Bioss) and α-rabbit IgG (1: 10,000; A4914, Sigma). Blots were developed with the Clarity Western ECL kit (Bio-Rad) and imaged in a Gel-Doc system (Bio-Rad).
Desialylation of HT29-MTX and ΔMUC1 cells was achieved by incubating cells grown in a 24-well plate with 250 mU/mL neuraminidase (Sigma) in DPBS for 2 h at 37°C. Cells were then washed twice with DPBS and used in Salmonella invasion assays.
|
10.1371/journal.pgen.1000920 | Replication Timing of Human Telomeres Is Chromosome Arm–Specific, Influenced by Subtelomeric Structures and Connected to Nuclear Localization | The mechanisms governing telomere replication in humans are still poorly understood. To fill this gap, we investigated the timing of replication of single telomeres in human cells. Using in situ hybridization techniques, we have found that specific telomeres have preferential time windows for replication during the S-phase and that these intervals do not depend upon telomere length and are largely conserved between homologous chromosomes and between individuals, even in the presence of large subtelomeric segmental polymorphisms. Importantly, we show that one copy of the 3.3 kb macrosatellite repeat D4Z4, present in the subtelomeric region of the late replicating 4q35 telomere, is sufficient to confer both a more peripheral localization and a later-replicating property to a de novo formed telomere. Also, the presence of β-satellite repeats next to a newly created telomere is sufficient to delay its replication timing. Remarkably, several native, non-D4Z4–associated, late-replicating telomeres show a preferential localization toward the nuclear periphery, while several early-replicating telomeres are associated with the inner nuclear volume. We propose that, in humans, chromosome arm–specific subtelomeric sequences may influence both the spatial distribution of telomeres in the nucleus and their replication timing.
| Functional telomeres are essential for genome stability. While replication of telomeres has been extensively studied in model organisms such as the baker's yeast, little is known about the mechanisms that govern the replication of human telomeres. In this study, we have determined the timing of replication of telomeres of individual human chromosomes and its association with potential modulating factors such as particular subtelomeric sequences, the presence of heterochromatic regions, and nuclear localization. We have found that native telomeres associated with D4Z4 sequences—a macrosatellite naturally located in the subtelomeric regions of 4q, 10q, and acrocentric chromosome extremities—replicate later than others. We also present descriptive and experimental evidence indicating that nuclear localization influences the timing of telomere replication. These results contribute to our understanding of telomere metabolism in humans.
| Cell proliferation potential is a critical attribute that directly influences embryogenesis, development and growth. For instance, insufficient proliferation capacity compromises organogenesis, tissue regeneration and repair, while unrestrained cell proliferation promotes cancer progression [1]. The human chromosome structures that have been most directly linked to cell proliferation control are telomeres [2].
Telomeres are specialized nucleoprotein complexes found at the ends of linear chromosomes. In vertebrates, they consist primarily of thousands of double stranded hexameric repeats (5′-T2AG3-3′) that end in a 3′ G-rich protruding single strand. The double strand region is directly bound by specific telomeric factors (TRF1 and TRF2), while the 3′ overhang is bound by POT1. Interactions of these proteins with three other telomeric proteins (TIN2, TPP1 and RAP1) constitute the shelterin/telosome complex, which is required for telomere function [3], [4].
Telomeres protect chromosome ends from degradation and fusion. They ensure the complete replication of chromosomes by creating a buffer of expendable sequences. Because of both the end replication problem, following which conventional DNA polymerases cannot completely replicate the ends of linear molecules [5] and the post-replication processing required to form a new functional telomere [6], telomeres shorten with every genome replication cycle. In the absence of a mechanism to add telomere repeats to the 3′ end, telomeres shorten with cell division until they reach a critical length, incompatible with proper telomere function [2]. A checkpoint signal is then triggered and cells enter senescence. Cell proliferation capacity is thus determined by initial telomere length and telomere shortening kinetics [2]. The latter is highly variable among human cell lines and ranges from 30 to 300 bp/cell division [7]–[9]. In vivo, telomere shortening in haematopoietic tissues has been estimated about 25–35 bp/year, although this pace is accelerated during the first years of life and also under stress or pathological conditions [10]–[16].
Telomerase, the dedicated reverse transcriptase that adds telomeric repeats de novo to the 3′ end, is highly active during development and its activity persists in stem cell compartments, where it ensures the cell replication potential of highly proliferative tissues [17]–[19]. However, as suggested by the telomere shortening that occurs with aging, the levels of telomerase are limiting [18], [20]. Also, mutations that prevent full telomerase activity accelerate telomere shortening and cause the premature appearance of aging phenotypes [21]–[23]. Accelerated telomere shortening might also result from difficulties during telomere replication. For instance, it has been shown that mutations in the gene coding for the WRN exonuclease/helicase compromises the replication and integrity of the telomere G-rich strand [24], [25]. Telomere replication defects may thus contribute to the aging phenotypes observed in Werner syndrome patients [26], [27].
Very little is known regarding the control of telomere replication in human cells [28]. The bulk of human telomere sequences replicate all through S-phase [29], [30]. This is seemingly different from what is observed in budding yeast where telomeres replicate in concert late in S phase [31], [32], although it is not known whether replication timing for individual human telomeres is spatially or temporally controlled. In a recent study, it was shown that telomeres in the muntjac deer display defined timings of replication and that telomeres on long and short arms replicate asynchronously [33]. This finding suggests that the firing of subtelomeric origins of replication in this species is subjected to chromosome arm-specific control mechanisms.
We have used the ReDFISH-based approach, described previously by Zou et al for the muntjac [33], to determine the timing of replication of individual telomeres in human cells. Our observations indicate that both chromosome arm-specific subtelomeric composition and nuclear localization influence the timing of telomere replication in humans.
We used human primary fetal lung fibroblasts (IMR90) and the ReDFISH approach, which is a modified version of the CO-FISH technique [34] (Figure 1A–1C), to characterize the timing of replication of individual telomeres. At least 30 metaphases were analyzed for each hour of a BrdU/C pulse to determine the percentage of individual telomeres being replicated and to calculate the mean replication timing (mrt) for each chromosome arm.
Global analysis showed that telomere replication takes place during the whole S-phase with a peak (about ¼ of all telomeres) during the fourth hour after S-phase initiation (mrt: 3.27; Figure 1D). This kinetics of bulk telomere replication has already been observed using density-labeling methods [35] or BrdU-based detection [30], [36]. However, our results indicate that single telomeres replicate in less than one hour since 1 hour BrdU pulses are sufficient to reveal perfectly detargeted sister telomeres (i.e. telomeres on homologous sister chromatids that are exclusively recognized by either G-rich or C-rich specific probes, Figure 1C). Moreover, telomeres located at specific chromosome ends tend to preferentially replicate during a defined window of the S-phase, with some telomeres replicating rather early and others replicating late (Figure 1E). For instance, 50% of telomeres on the 19q arm replicate during the first two hours (mrt: 2.33, Figure 1F), whereas around 70% of telomeres on the 4q arm replicate during the last two hours (mrt: 4.45, Figure 1F). Both mean replication timings are significantly different from the mean replication timing of 6p, a mid-S replicating telomere (mrt: 3.33, Fisher exact tests: 19q vs 6p, p = 6.9×10−8, 4q vs 6p, p = 1.2×10−9 and 4q vs 19q, p = 2.2×10−16. Significance threshold: p<0.0025). As observed in muntjac cells [33], telomeres on short and long arms of the same chromosomes show no coordinated replication.
There is no obvious correlation between the reported replication pattern of the last R/G (R = reverse, G = giemsa) cytogenetic bands on each chromosome arm, revealed also by incorporation of BrdU [37]–[39], and the pattern of replication for single telomeres observed here. There is a weak, albeit not significant, correlation of single telomere replication timings and the mean replication timings reported for the most distal chromosome-specific regions included in BAC (bacterial artificial chromosome) arrays (Figure S1) [40] suggesting some synchronicity between telomeres and distal subtelomeric regions (Spearman's rank correlation test: p = 0.093, significance threshold p<0.015).
To understand which factors regulate the replication timing of individual telomeres, we examined the impact of telomere length and telomerase expression. Indeed, recent work in budding yeast suggests that telomere length could influence the timing of replication. Particularly, a shortened telomere tends to replicate earlier [41] while the bulk of telomeres replicate rather late [42]. Using telomere Q-FISH followed by subtelomeric FISH as described previously [9], we measured relative telomere fluorescence intensities (which indicate telomere length relative to the mean telomere length of the cell) specifically associated with polymorphic chromosome arms. We used the same subtelomeric FISH approach after ReDFISH to determine the replication timing of telomeres of the same chromosome pairs (Figure 2A). Although significant telomere length differences exist between some alleles, their telomeres replicated with similar timings (7p, 8p or 16p) (Figure 2B). Also, some allelic telomeres, like those on 9q, showed differences in mean replication timing although no difference in telomere length was observed (Figure 2B). A global comparison of telomere lengths and mean replication timings through a correlation analysis confirmed that no relationship exists between both variables (Figure 2C).
To further corroborate this observation, we examined the replication profile in IMR90 cells expressing the catalytic subunit of human telomerase (hTERT). hTERT is limiting for telomerase activity in most human fibroblasts and is often sufficient to increase their replication potential [43]. However, some cells spontaneously increase the expression of p16INK4a by mechanisms that are unknown and such cells senesce even in the presence of telomerase activity. IMR90+hTERT cells fall into this category [44], preventing us from obtaining enough analyzable material for our studies. We therefore expressed in these cells TIN2, another telomeric factor [45]. TIN2, through its interaction with TRF1, exerts a negative control on telomere length. In our cells, however, telomeres were stabilized above 10 kb with individual telomere lengths being largely homogenized, as expected for a cell line expressing hTERT alone (Figure 3A) [9]. IMR90+hTERT+TIN2 cells grew vigorously, allowing us to perform the same study in cells that had longer and much more homogeneous telomeres than primary cells. The length of S-phase, as indicated by FACS analysis, is somewhat shorter in telomerized IMR90 cells, since it lasts 5.5 hours instead of a little more than 6 hours in the parental cell line (not shown). Concurrently, telomere replication peaks earlier (mrt: 2.61) and very few telomeres are seen replicating during the last pulse (Figure 3B). Remarkably, however, the relative timings of replication for single telomeres in this cell line were very similar to the one observed in unmodified cells (Figure 3C). A statistical analysis (Figure 3D) showed a highly significant positive correlation (Spearman's rank correlation test: p<0.0001, significance threshold p<0.01), indicating little variation in the relative timings of telomere replication between both cell lines. These results strengthen the conclusion that telomere length has no visible impact on telomere replication timing in humans. Interestingly, the mean replication timings for single telomeres in telomerized IMR90 cells appear to be significantly correlated to the mean replication timings of the most distal chromosome-specific sequences reported by Woodfine et al [40] (Spearman's rank correlation test: p = 0.0062) (Figure S1).
Subtelomeric regions show extensive segmental polymorphisms, which can reach several hundreds of kilobases [46]–[48] and could directly impact replication origin firing and/or replication fork speed. In the experiment to determine the relative lengths of allelic telomeres, we used subtelomeric probes recognizing segments (around 30–40 kilobases long) that are present or absent at chromosome extremities and are located very close to the telomere tract (10 to 20 kb) [46], [48], [49]. As shown in Figure 2B, we found that extremities corresponding to allelic locations but carrying segmental polymorphisms (alleles labeled A and B on 1q, 7p, 8p, 9p and 16q in Figure 2A) tend to replicate during the same time window, like do homologous sequences elsewhere in the genome [50]. On the other hand, telomeres on different chromosome pairs but associated with subtelomeric regions of similar segmental composition (compare for instance 1pA and 7pA to 8pA and 9qB) may have different timings of replication. This indicates that segmental polymorphisms do not account for the differential replication timing of individual telomeres.
To determine whether telomeres with identical chromosome positions in the genome tend to replicate at similar times in different individuals, we studied the telomere replication pattern in the foreskin fibroblast cell line HCA2, which expresses an exogenous copy of hTERT and therefore replicates indefinitely, like the IMR90+hTERT+TIN2 cells. In fact, both the length of the S-phase and the global timing of telomere replication are indistinguishable between both cell lines (Figure 4A). Even more remarkably, the ranking of mean replication timings for individual telomeres was very similar (Figure 4B), as indicated again by a statistically significant positive correlation coefficient (Figure 4C). Also striking is the observation that chromosome extremities potentially carrying extended subtelomeric segmental variations in both cell lines harbor similar replication timings (Figure 4D), strengthening the idea that these genetic polymorphisms do not have a major effect on telomere replication timing. Again, the mean replication timings for single telomeres in telomerized HACA2 cells appear also to be correlated to the mean replication timings of the most distal chromosome-specific sequences (Figure S1).
Interestingly, the most conspicuous, albeit limited, variations in replication timing between IMR90 and HCA2 cells concern telomeres that replicate in the first two thirds of the S-phase, while less variation is apparent amongst telomeres that replicate later (Figure 4C). Since the incidence of subtelomeric polymorphisms is equally distributed among early and late replicating telomeres, this observation suggests that differences in replication timing because of genetic variations might be more easily superseded by factors causing telomeres to replicate late in the S-phase. We therefore conducted a closer examination of late replicating telomeres.
It has been suggested that transcriptionally inactive heterochromatic regions tend to replicate during the second part of the S-phase [51]. Also, in females, both X chromosomes display a different replication pattern according to their heterochromatic state. Active X chromosomes behave like autosomal chromosomes, bearing early and late replicating bands, while inactive X (Xi) shows a pattern of late replication that generally encompasses the entire chromosome [38]. Our analysis of telomere replication by ReDFISH revealed that, on the X chromosomes of IMR90 cells, telomeres on the short arm replicate during the middle of S-phase, rather synchronously as expected for homologous chromosomes. Replication of telomeres on the long arm presented a bimodal distribution with one peak of replication in the middle of S-phase and another peak at the end of that phase (Figure S2). The late profile of BrdU incorporation observed all along the chromosome that also presented a late replicating telomere suggested that this chromosome is Xi (not shown). However, since both features depend on the same phenomenon (BrdU incorporation during replication), confirmation that this is a bona fide Xi requires a replication-independent criterion. Unfortunately, detection of Xi-specific heterochromatic marks (such as particular histone modifications) was precluded by the type of chromosome fixation (ethanol/acetic acid) used in the ReDFISH approach. In the male HCA2+T cells, the Yp telomere replicated early in S-phase while the Yq telomere displayed a much later replication pattern (Figure S2) (mrt: 2.49 and 3.50, respectively, Fisher exact test: p = 9.9×10−8), suggesting that the replication timing of this telomere might be influenced by the constitutive heterochromatic region found on the Yq arm. On the other hand, although the Xq telomere shows a peak of replication coincident with the Xp telomere, their calculated mrts are significantly different (2.61 and 3.12, respectively, Fisher exact test: p = 8.7×10−6) (Figure S2) indicating that Xq replicates later than Xp in this male fibroblast cell line. On the other hand, the comparison between the mean replication timings of Xq and Yq telomeres fails to show a significant difference confirming that both telomeres have a late replicating pattern.
Amongst the telomeres that consistently replicated late during S-phase in all cells examined are those located on the short arms of acrocentric chromosomes (Figure 1E, Figure 3C, Figure 4B, and Figure 5A). Together with rDNA clusters, acrocentric regions carry both β-satellite sequences and D4Z4 repeats [52]. These two last kinds of repeats are also found at two other extremities, 4qter and 10qter. As shown in Figure 5B, both 4q and 10 replicate late in IMR90, a behavior also observed in IMR90+TT and HCA2+T cells (Figure 3C and Figure 4B). These observations suggest that the presence of satellite-like repeats at subtelomeric positions may influence the timing of telomere replication. However, while telomeres on acrocentric short arms have been detected as associated with the nucleolus [53], both 4qter and, although much less consistently, 10qter have been reported as being associated with the nuclear periphery [54]–[56], suggesting that nuclear localization could also influence the timing of telomere replication. We therefore tested both the replication timing and nuclear localization of newly created telomeres carrying a defined subtelomeric composition.
To address the specific contributions of subtelomeric elements with regard to nuclear localization and telomere replication, we artificially tagged telomeres in C33A human cells with DNA molecules that carry either multiple D4Z4 repeats, a single D4Z4 repeat, 4 β-satellite repeats or both (Figure 6A). Upon chromosome integration, such constructs lead to non-targeted (random) de novo telomere formation [57]. We have previously shown that in this cell line, polyclonal populations of stably transfected cells are representative of pools of independent clones of tagged telomeres allowing us to perform analyses on populations [56], [58], [59]. Also, the presence of particular subtelomeric sequences does not bias the chromosome integration sites of the seeding constructs [56], [58].
As shown in a previous study [56], [58], a single D4Z4 repeat, alone or inserted together with β-satellite repeats, confers to a chromosome extremity a more peripheral position within the nucleus while multiple copies of D4Z4 repeats (Figure 6B and 6C), or several β-satellite repeats (Figure 6C) alone, do not. We then determined the replication timing of all types of telomere seeded extremities and found that those carrying only one D4Z4 repeat (and bearing a more peripheral localization in the nucleus) replicate later than the others, suggesting that nuclear localization influences telomere replication timing (Figure 6D and 6E). On the other hand, the mean replication timing of telomeres connected to β-satellite sequences alone is significantly higher than the mean replication timing of control telomeres (p = 0.0006, significance threshold p<0.003). This effect is independent of nuclear localization, thus allowing the conclusion that β-satellite sequences by themselves cause a delay in telomere replication.
Given the above results, we examined by immuno-FISH and 3D imaging the nuclear localization of native chromosome ends carrying telomeres that replicated either early or late in IMR90 primary cells. Our results, illustrated in Figure 7, indicate that the late replicating extremities 2pter, 3pter, 4qter, 6qter and 12qter have a clear tendency to localize at or near the nuclear periphery, whereas the early replicating extremities 1p, 5p, 12p and 17q are found in the inner part of the nuclear volume (Figure 7). Furthermore, a correlation analysis (p<0.0002) clearly indicates that there is a direct relationship between the mean replication timing of a telomere and its mean volume ratio determined by immuno-3D (Figure 7E). Together, our data suggest, for the first time, a strong association between telomere replication timing and nuclear localization.
We characterized the replication timing of single telomeres in normal diploid human cells, either primary or immortalized by ectopic expression of telomerase. In agreement with previous studies [29], [30], we found that bulk telomeres replicate throughout the S-phase. Our results further indicate that single telomeres on specific chromosome ends tend to replicate during defined times in the S-phase and that this timing is conserved between homologs and among individuals. Contrary to findings in the budding yeast [31], [32], telomere length does not have a major impact on telomere replication timing. However, given the length of the S-phase and the inherent imprecision of the methodology used, it remains possible that subtle influences introduced by the length of telomeres and/or the presence of telomerase activity may have been overlooked.
Occasionally, small differences were detected, both between homologs and among individuals, which could be explained by variations either in the DNA sequence or the epigenetic status of these extremities. Nevertheless, our study also shows that the segmental polymorphisms (which may span up to hundreds of kilobases) occurring very close to telomeres [47], [48], [60], [61] do not exert a major influence in the replication timing of allelic telomeres. The subtelomeric duplications f7501 and ICRF10 revealed in these experiments are present in about 15 chromosome extremities, a dozen of which are potentially polymorphic. These sequences are located quite close to the telomere tract and their presence or absence indirectly indicate the presence or absence of other subtelomeric segments with which they are commonly associated. For instance, in the cell line we examined (IMR90), the presence or absence of ICRF10 on chromosome 8p (Figure 2A) implies the presence or absence, respectively, of at least three other (more proximal) segments in that extremity (see [48]). This signifies that both alleles differ from each other in their subtelomeric region by at least 120kb [46]. Whether or not this distance is sufficient to introduce a difference in the replication timing for both telomeres (either by delaying the arrival of the replication fork to the telomere or by introducing a new origin of replication) remains to be explored. Nevertheless our experiments do suggest that such polymorphisms may occur without inducing major differences in telomere replication timing. On the other hand, some of the observable differences affect chromosome extremities without (known) segmental variation at subtelomeres, suggesting that other factors are at play.
Previous studies on the replication timing of specific subtelomeric regions (for instance 22q [62] and 16p [63]) suggested that particular telomeres replicate late. The present study did not detect such trend for these particular ends in the cell lines examined. Moreover, these two telomeres are among the earliest to replicate in S phase. The aforementioned studies used subtelomeric probes and interphase nuclei FISH to follow the duplication of signals during S phase progression. However, duplication of signals depends not only on replication of that particular segment but also on the resolution of sister chromatids. This step seemingly follows a different pathway at telomeres [64], which might explain why telomeres placed nearby other sequences may influence (i.e.: delay) the appearance of distinct FISH foci in interphase nuclei after replication. This interpretation is supported by the observation that duplication of telomeric signals in interphase nuclei only occurs during the second half of the S-phase [65], while by this time, as shown here, almost half of telomeres have already replicated. It is clear that the ReD-FISH approach, although laborious and time consuming, has allowed to define in a more precise way the timings of replication of single telomeres in human cells.
One striking feature of the telomere replication pattern in human cells is the late replication timing of telomeres associated with satellite-like repeats, i.e. the short arm of the acrocentric chromosomes as well as 4qter and 10qter extremities. Our experiments using newly created tagged telomeres indicate that the presence of β-satellite sequences, which have high heterochromatinization potential and are late replicated when in their natural context [29], caused the nearby telomere to replicate significantly later than control telomeres. Strikingly, the presence of a single D4Z4 repeat, which is sufficient to increase the association of a telomere with the nuclear periphery, caused the nearby telomeres to replicate much later in the S-phase than the control telomeres and as late as the acrocentric telomeres in the same cell line. Both effects, peripheral nuclear localization and late replication, are no longer detected when multiple D4Z4 repeats are inserted next to the telomere, further supporting the connection between subnuclear localization and telomere replication timing.
The reason why a single D4Z4 is able to mediate the association of a chromosome extremity to the periphery, while multiple copies of this repeat are not, remains mysterious. However, this observation is in agreement with the fact that the presence of multiple copies of D4Z4 at other locations, such as 10q and acrocentric telomeres, is not sufficient to increase the association of these extremities with the nuclear periphery [54]–[56]. As discussed in a previous work [56], the explanation for this apparent paradox may rely on the function of a putative region centromeric to the D4Z4 repeats and only present on 4q extremities. Independently of the mechanism involved in this perinuclear association, our experiments clearly point to a tight relationship between the peripheral localization of a telomere and its late replication behavior. On the other hand, the effect of β-satellite sequences appeared to be independent of nuclear localization. It is theoretically possible that a biased genomic integration of such constructs could have placed the newly created telomeres in a context where replication is intrinsically delayed. However, as shown previously within the limit of resolution of multi-FISH analyses [56], the telomere seeding strategy used here does not lead to a biased distribution of telomere seeds in C33A cells, supporting the contention that the observed effects are directly connected to the presence of particular juxtatelomeric elements carried by the constructions.
In yeast, the well-documented association of telomeres with the nuclear envelope [66] appears to play important roles in telomere metabolism, including length regulation [67], silencing [68] and repair [69], [70]. In humans, telomeres are supposed to be randomly distributed within the nucleus [71], but there have been reported exceptions, such as the nuclear peripheral localization of 4q [54], [55] and the perinucleolar localization of telomeres on the short arms of acrocentrics [53]. Strikingly, we found that other non-D4Z4 associated chromosome extremities are also naturally localized at the nuclear periphery in unperturbed IMR90 cells, adding four more exceptions (2p, 3p, 6q and 12q) to the list of telomeres with preferential nuclear localizations. Remarkably, all these telomeres replicate late in the diploid fibroblasts examined.
Thus, our observations point to a relationship between telomere nuclear localization and telomere replication timing. Nuclear localization has been suggested to affect replication timing of other regions of the genome [51], [72], [73]. Also, recent studies have demonstrated that genome-wide interactions with the nuclear lamina implicate late replicated sequences [74]. Close examination of the subtelomeric chromosome specific sequences available for the extremities examined here (http://genome.ucsc.edu/cgi-bin/hgGateway) revealed that only 12q present a particular enrichment in LADs (lamina-associated domains) [74]. However, actual subtelomeric regions are most often not included in human genome sequence assemblies, either because they are poorly characterized or because their duplicated nature makes their chromosome assignment quite difficult. It is worth noting here that at least the last 120 kilobases of the subtelomeric region of the 6q chromosome are duplicated on other extremities, including 1p [48], whose telomere, contrary to that one on 6q, replicates early and is not associated with the nuclear periphery.
Our results also indicate that single telomere replication timing in human diploid fibroblasts is mostly determined by chromosome-specific features, perhaps at the level of large chromosome domains, as suggested recently [72]. Although telomere replication timings do not appear to be correlated with the replication timings of large cytogenetic bands, we did find a correlation between the mean replication timings for single telomeres and the timing of replication reported for the most distal chromosome-specific sequences present in a BAC-array [40]. This correlation, albeit weak (and only statistically significant when data from telomerized diploid fibroblasts were used, perhaps reflecting the fact that the BAC study was conducted in an EBV-transformed lymphocyte cell line [40]), suggests that telomere replication may be, at least partially, synchronized with chromosome-specific subtelomeric sequences.
Finally, our data conclusively show that telomere replication timing may also be influenced by the presence of relatively small telomere-associated sequences, such as β-satellite sequences or one repeat of the macrosatellite D4Z4, which also confers a peripheral nuclear localization to the chromosome end. It has been recently demonstrated that this D4Z4 sequence behaves both as an A-type lamin-, CTCF-dependent peripheral tethering element and as an insulator [56], [58]. It remains to be determined which of these two properties confer late replication.
Together, our study allowed an original and certainly informative glimpse into the mechanisms regulating telomere replication timing in human cells. Our results suggest that the links between replication timing and high-order genome organization, also observed in other organisms, may have been conserved throughout evolution [28], [41], [75], [76].
The human cell lines used here include the fetal lung fibroblast IMR90 (46XX, obtained from ATCC) and the foreskin fibroblast HCA2 (46XY, obtained from James Smith, Baylor College). Both cell lines were immortalized by transduction of a pBABE-derived retrovirus carrying hTERT. IMR90+hTERT was also transduced with another retrovirus carrying TINF2. We also constructed derivatives of the well-known cervical carcinoma cell line C33A [58]. The different constructs carry a hygromycin resistance gene fused to the herpes simplex virus type 1 thymidine kinase suicide gene and an eGFP reporter gene, each driven by a CMV promoter (see Figure 6A for description of the constructs). A telomere seed is added downstream of the integrated sequences in order to create a de novo telomere after random integration followed by a telomeric fragmentation [57]. A single D4Z4 repeat (black box) or 8 tandem copies of the repeat were cloned between eGFP and the telomere seed in T construct (T1X) as previously described [56]. At different loci, D4Z4 co-segregates with β-satellite sequences. A PCR-amplified fragment of 1.4 kb corresponding to the distal sequence of the 4q35 subtelomere and encompassing 4 β-satellite elements (Boussouar et al., manuscript in preparation) was also cloned in the T construct, either downstream of the eGPF reporter or downstream of the D4Z4 element. After transfection and telomeric fragmentation, the cells were grown and selected for antibiotic resistance in polyclonal batches. The location of newly created telomeres (90% of the cells in a given population) differs between cells [56].
For replication studies, all cell lines were synchronized by a double thymidin/aphidicholin block. Briefly, cells were incubated for 16 h in 2 mM thymidine, released in S after 3 washes of PBS pre-warmed to 37°C and 8 to 12 h later, depending on the cell line, treated again with 1µg/ml aphidicolin for 16 h prior to release in S after three washes in pre-warmed PBS.
To precisely determine the length of the S-phase, cells were analyzed by FACs. Every hour after release from the aphidicolin block, cells were trypsinized, centrifuged, resuspended in 0.5 ml of PBS, fixed by drop-wise addition of 1.5ml ice-cold 100% ethanol and stored at 4°C. Subsequently, cells were centrifuged and resuspended in a staining solution containing 30 µg of propidium iodide and 200 µg of RNaseA per ml. Flow cytometry was performed using a Becton Dickinson FACSort flow cytometer. The data was analyzed using FlowJo software.
Replicative Detargeting (ReD) FISH is a modification of the CO-FISH procedure [34] and was performed as described previously [33] with some modifications. Briefly, after release from the aphidicolin block, 6 to 8 pulse-chase additions of BrdU/BrdC were made depending on the length of the S-phase. For each pulse, cells were incubated for 1 hour in the presence of 10 µM BrdU and 3.3 µM BrdC, then washed 3 times with pre-warmed PBS before new media was added. 7 to 10 hours after release from the aphidicolin block, cells were arrested in mitosis with 1.5 hour incubation in colcemid (0.1 µg/ml) before 40 min hypotonic shock in 0.8 g/L sodium citrate at 37°C and fixed in ethanol/acetic acid. Metaphase spreads were obtained by dropping suspensions of fixed cells onto clean glass slides and were rapidly used for hybridization. Spreads were denatured at 80°C for 4 min in the presence of a Cy3-(CCCTAA)3 PNA probe (Applied Biosystems, 50 nM in 70% formamide, 25 mM Tris pH 7.4) and incubated at room temperature for 2 hours. After washes and ethanol dehydration, the slides were put in contact for 2h with a second LNA probe 5′-(6-Fam)GGGtTAGGGttAgGGTTAGGgttAgGgttTAGGgTTA (6-Fam)-3′ - where small letters correspond to positions with locked nucleic acids - (Proligo-France, 10 mM in 50% formamide, 2xSSC) followed again by washes and ethanol dehydration. Preparations were mounted in Vectashield (Vector) with DAPI (1 µg/ml) and visualized with a Zeiss UV microscope equipped with appropriate excitation/emission filters for each color. Images were captured with a HQ-Coolsnap camera (Photometrics) using the IPlab software. When required, coordinates for all metaphases were recorded in order to retrieve them after a second (subtelomeric) FISH.
The Q-FISH procedure was carried out exactly as described, using a Cy3-(CCCTAA)3 PNA probe [9]. Metaphase spreads, prepared the day before, were fixed with formaldehyde (Sigma, 3.7%) and digested with pepsin (Sigma, 1 mg/ml). Spreads were denatured at 80°C for 4 min in the presence of a Cy3-(CCCTAA)3 PNA probe (Applied Biosystems, 50 nM in 70% formamide, 25 mM Tris pH 7.4) and incubated at room temperature for 2 hours. After washes and ethanol dehydration, preparations were mounted in Vectashield (Vector) with DAPI (1 µg/ml) and visualized with a Zeiss UV microscope equipped with appropriate excitation/emission filters. Images were captured with a HQ-Coolsnap camera (Photometrics) using the IPlab software. When required, coordinates for all metaphases were recorded in order to retrieve them after a second (subtelomeric) FISH.
After CO-FISH or Q-FISH hybridizations, slides where washed 3 times in SSC 2X and dehydrated for subsequent subtelomeric FISH to distinguish homologues. Cosmids carrying subtelomeric regions, f7501 and ICRF10, were obtained from Barbara Trask (Human Genome Center, Lawrence Livermore National Laboratory) and Gilles Vergnaud (IGM, Orsay, France), respectively. Cosmid f7501 contains 36-kb portion of chromosome 19 including three members of the olfactory receptor (OR) family [49]. Cosmid ICRF10 carries minisatellite DNF92 (GenBank accession number Y13543) [46], [48]. Subtelomeric probes f7501 and ICRF10 correspond to two different segments of around 35 kb located close to the telomere tract on around 15 different chromosome extremities. The presence of one segment is exclusive of the other and both segments are typically associated with particular arrangements of other, more centromeric, segments. These probes were used for hybridization on metaphase preparations that had already been analyzed in either Q-FISH or ReDFISH experiments, thus allowing to distinguish between allelic chromosome extremities and to conduct allele-specific telomere fluorescence measurements (as described in [9]) or replication timing scorings (this paper).
One g of cosmid DNA was labeled with biotin-16-dUTP (Roche) using the Nick translation kit (Vysis) following manufacturer's instructions. For hybridization, 50 ng probe per slide was precipitated in the presence of 100 µg single-strand salmon sperm DNA and 20 µg COT-1 (Invitrogen), dissolved in 25 l hybridization mix (50% formamide, 10% dextran sulfate, SSC 2X) and pre-hybridized for 1 h at 37°C. Slides with metaphase spreads were treated with 0.1 g/ml RNase A in SSC 2X for 1 h at 37°C and washed three times in SSC 2X, 5 min each, prior to denaturation in 70% formamide/SSC 2X at 70°C for 2 min. Denaturated slides were dropped in ice-cold SSC 2X, dehydrated in a series of ice-cold ethanol baths, treated with Proteinase K (100 ng/ml in 20 mM Tris pH 7.4 and 2 mM CaCl2) for 8 min at 37°C, dehydrated again and hybridized over night at 37°C. The next day slides were washed three times, 3 min each in 50% formamide/SSC 2X, five times, 2 min each in SSC 2X, and once in BN (0,1 M sodium bicharbonate, 0,05% NP-40) all at 45°C. Slides were blocked with 5% milk in BN for 15 min. Biotinylated probes were detected with three layers of antibodies, each 30 min at 37°C, as follows: fluorescein avidin D (Vector A-2001, 1/400), biotinylated anti-avidin (Vector BA-0300, 1/100) and fluorescein avidin D, all diluted in blocking buffer. After each antibody layer three 2-min washes in BN at 42°C were done. Slides were mounted in Vectashield (Vector) with 0.2 g/ml DAPI.
To detect newly seeded telomeres in the C33A cell derivatives after ReD-FISH, a labeled pCMV vector was used as a probe and labeled with the DIG-Nick Translation Kit (Roche Diagnostics). All probes were denatured at 80±1°C for 5 minutes before hybridization. Conditions for slide preparation, hybridization and immunodetection have been described [56]. For detection, we used mouse anti-DIG antibodies (Roche Diagnostics), diluted 1/200, followed by incubation with secondary donkey antibodies coupled with ALEXA 488 fluorochrome, directed against this epitope and diluted 1/500 (Molecular Probes). Chromosomes were counterstained with DAPI antifade (0.125 µg/ml) (Cytocell).
Metaphases were retrieved thanks to the recorded coordinates and original images were annotated.
For each pulse, 30–50 metaphases were captured and analyzed. For each metaphase, a karyotype was carried out and chromosome ends with detargeted telomeres were identified. For each BrdU/C pulse, the average percentage of detargeted telomeres in the population of metaphases was calculated for each pair of chromosome ends. The addition of these percentages from all 6 BrdU pulses spanning the entire S phase was adjusted to 100%. The mean replication timing of single telomeres was calculated as follows: mrt = [Y]/f, where Y corresponds to the pulse (0.5, 1.5, …) and f is the percentage of telomeres seen replicating during that pulse. For the graphic representation of replication timings of individual telomeres, percentages of replication were grouped in early S (pulses 1+2), mid-S (pulses 3+4) and late S (pulses 5+6) for each telomere and telomeres were ordered according to their mrt.
Ten µg of genomic DNA was digested overnight with HinfI and RsaI enzymes, 50U each, and restriction fragments were separated through a pulsed-field electrophoresis agarose gel (1% in TBE 0.5X) in a CHEF apparatus (BioRad) set to 200 V for 15 h with a pulse ramp between 0.2 and 13 s. After staining with ethidium bromide, DNA was nicked by a UV-crosslinker (Stratagene) at 180,000 J/cm2, denatured, and transferred by capillary alkaline transfer onto Biodyne B Nylon membrane (Pall) for hybridization with a radioactively labeled TAA(CCCTAA)4 oligonucleotide. Signals were detected in a phosphorimager apparatus.
To determine the localization of telomeres in the nucleus, PAC clones recognizing the terminal region of chromosome arms 1p, 2p, 3p, 4q, 5p, 6q, 12p, 12q, 1 and 17q [77] were labeled with the DIG-Nick Translation Kit (Roche Diagnostics). All probes were denatured at 80±1°C for 5 minutes before hybridization. Conditions for slides preparation, hybridization and immunodetection have been described [56].
For detection, we used mouse anti-DIG antibodies (Roche Diagnostics) and goat anti-Lamin B antibodies (M-20, Santa-Cruz), diluted 1/50, followed by incubation with secondary donkey antibodies coupled with different ALEXA fluorochromes, directed against these epitopes and diluted 1/300 (Molecular Probes). Nuclei were counterstained with DAPI antifade (0.125 µg/ml) (Cytocell).
Images were acquired with the confocal scanning laser system, LSM510, from Zeiss (Germany). A 63× Plan-APOCHROMAT, oil immersion, NA 1.40 objective (Zeiss) was used to record optical sections at intervals of 0.48µm. The pinhole was set the closest to 1 Airy with optical slices in all wavelengths with identical thickness (0.8µm). Images were averaged 4 times to improve the signal to noise ratio. Generated .lsm files had a voxel size of 0.1µm×0.1µm×0.48µm and were processed through the Imaris software (Bitplane AG). After 3D analysis, where at least 50 nuclei were examined, data sets are presented as the distribution of FISH signals between three concentric zones of equal volume or as the mean ratio between two volumes.
The R package was used for comparisons using Fisher exact tests and Spearman's rank correlation coefficient calculations. For multiple comparisons, corrections for significance thresholds were applied depending on the number of comparisons actually carried out (p<0.05/k; Bonferroni): k = 20 for Fisher exact test comparisons using diploid fibroblast data and k = 15 for comparisons using C33A data.
|
10.1371/journal.pgen.1003799 | N-acetylglucosamine (GlcNAc) Triggers a Rapid, Temperature-Responsive Morphogenetic Program in Thermally Dimorphic Fungi | The monosaccharide N-acetylglucosamine (GlcNAc) is a major component of microbial cell walls and is ubiquitous in the environment. GlcNAc stimulates developmental pathways in the fungal pathogen Candida albicans, which is a commensal organism that colonizes the mammalian gut and causes disease in the setting of host immunodeficiency. Here we investigate GlcNAc signaling in thermally dimorphic human fungal pathogens, a group of fungi that are highly evolutionarily diverged from C. albicans and cause disease even in healthy individuals. These soil organisms grow as polarized, multicellular hyphal filaments that transition into a unicellular, pathogenic yeast form when inhaled by a human host. Temperature is the primary environmental cue that promotes reversible cellular differentiation into either yeast or filaments; however, a shift to a lower temperature in vitro induces filamentous growth in an inefficient and asynchronous manner. We found GlcNAc to be a potent and specific inducer of the yeast-to-filament transition in two thermally dimorphic fungi, Histoplasma capsulatum and Blastomyces dermatitidis. In addition to increasing the rate of filamentous growth, micromolar concentrations of GlcNAc induced a robust morphological transition of H. capsulatum after temperature shift that was independent of GlcNAc catabolism, indicating that fungal cells sense GlcNAc to promote filamentation. Whole-genome expression profiling to identify candidate genes involved in establishing the filamentous growth program uncovered two genes encoding GlcNAc transporters, NGT1 and NGT2, that were necessary for H. capsulatum cells to robustly filament in response to GlcNAc. Unexpectedly, NGT1 and NGT2 were important for efficient H. capsulatum yeast-to-filament conversion in standard glucose medium, suggesting that Ngt1 and Ngt2 monitor endogenous levels of GlcNAc to control multicellular filamentous growth in response to temperature. Overall, our work indicates that GlcNAc functions as a highly conserved cue of morphogenesis in fungi, which further enhances the significance of this ubiquitous sugar in cellular signaling in eukaryotes.
| In stark contrast to most fungal pathogens, thermally dimorphic fungal pathogens cause systemic infections in immunocompetent humans. Thermally dimorphic fungi grow in the soil as a multicellular filamentous form specialized for replication in this particular environmental niche. Upon infection of a human, these fungi transition to a parasitic cell type that is adapted for replication and pathogenesis within a mammalian host. In this work, we examined factors that are important for growth of the infectious, environmental form of thermally dimorphic fungi. We discovered that N-acetylglucosamine (GlcNAc), a ubiquitous carbohydrate with cellular roles across all kingdoms of life, stimulated a switch to the environmental form for two thermally dimorphic fungal pathogens, Histoplasma capsulatum and Blastomyces dermatitidis. Analysis of how fungal cells respond to GlcNAc revealed that these fungi possess two GlcNAc transporters that are important for controlling their ability to switch between infectious and parasitic states. Overall, our work begins to elucidate the pathways that promote growth in the infectious form of these organisms, which is critical to our understanding of environmental signals that promote disease transmission of thermally dimorphic fungi.
| Cellular differentiation is an essential process for the development and growth of complex multicellular eukaryotic organisms. Similarly, many unicellular eukaryotic organisms undergo a program of cellular differentiation to produce a new cell type that is specialized for survival in a distinct environmental niche. In response to environmental stimuli, the family of thermally dimorphic fungal pathogens undergoes a program of cellular differentiation to transition between a saprophytic soil form and a parasitic host form [1]. The soil form is comprised of multicellular filaments that produce infectious spores. The parasitic form for the majority of thermally dimorphic fungi consists of a unicellular yeast form that is capable of evading host immune defenses. Temperature is the predominant environmental cue that promotes cellular differentiation of thermally dimorphic fungi; however, additional factors including CO2, reactive oxygen species, and steroid hormones are also thought to influence morphogenesis [2]–[5].
The ability of thermally dimorphic fungi to transition between two distinct morphological states in response to environmental stimuli is important for the maintenance of their disparate lifestyles as soil saprobes and mammalian pathogens [6]. Thermally dimorphic fungal pathogens, such as Histoplasma capsulatum and Blastomyces dermatitidis, grow in the soil in a filamentous or “mold” form that produces vegetative spores known as conidia. Upon inhalation of hyphal fragments or conidia by a human host, and subsequent growth at mammalian body temperature, H. capsulatum and B. dermatitidis transition into a budding yeast form capable of growth and pathogenesis in mammals. Since thermally dimorphic fungi can persist in mammals after an acute infection is resolved, it is thought that the parasitic host form returns to the soil after the death of infected animal hosts, thus facilitating a transition to the filamentous form and serving to maintain an infectious reservoir [7]–[9]. Maintenance of an environmental reservoir is crucial for the pathogenic lifestyle of thermally dimorphic fungi since these organisms are not directly transmitted between mammalian hosts [10].
Recently, some progress has been made in understanding the molecular mechanisms by which environmental signals stimulate morphological transitions in thermally dimorphic fungi. Comparative gene expression profiling of H. capsulatum has revealed that approximately 15% of predicted genes are differentially expressed between the two morphological forms (multicellular filaments and unicellular yeast cells), indicating that a significant fraction of transcripts exhibit yeast-phase or filamentous-phase enrichment expression patterns [11]–[13]. In addition, forward genetic screens have identified regulators of yeast- and filamentous-phase growth in thermally dimorphic fungi, namely, the Ryp (required for yeast-phase growth) master regulators (Ryp1, Ryp2, and Ryp3), the histidine kinase Drk1 (dimorphism-regulating kinase 1), and the GATA-family transcriptional regulator Sre1/SreB [12], [14]–[18]. However, it remains to be elucidated how these regulators sense and integrate environmental signals into phenotypic and genotypic outputs.
One challenge in studying the molecular events involved in the morphogenesis of thermally dimorphic fungi has been establishing a robust and synchronous morphological phase transition in vitro. For example, in laboratory cultures of H. capsulatum, the conversion between yeast cells and filaments is recapitulated by switching the temperature from 37°C to room temperature (RT) [5], [19]. The switch is bidirectional, so H. capsulatum filaments can be switched to yeast cells by shifting the temperature in the opposite direction (RT to 37°C). Under these laboratory conditions, temperature is sufficient to promote morphogenesis; however, the interconversion between yeast cells and filamentous cells in the laboratory is slow and asynchronous, suggesting that other environmental cues that promote morphogenesis are missing from in vitro culture. Establishing a robust and synchronous phase transition of thermally dimorphic fungi in vitro would allow an easier and more robust examination of the temporal series of events that occurs during morphogenesis, and permit identification of factors important for cellular differentiation.
To identify potential inducers of morphogenesis in the thermally dimorphic fungi, we contemplated developmental pathways in other fungi. The yeast-to-filament transition is well studied in the human commensal fungus Candida albicans, and can be induced by a variety of signals [20]. For example, the ubiquitous monosaccharide N-acetylglucosamine (GlcNAc) is known to promote filamentation through an unidentified pathway in C. albicans [21]. Fungi are likely to encounter environmental sources of GlcNAc since this carbohydrate is a major component of insect exoskeletons and bacterial peptidoglycan. Additionally, the innermost layer of the fungal cell wall is composed of chitin, which is a polymer of β1–4 linked GlcNAc that undergoes turnover during the remodeling of the cell wall that accompanies cell division. Thus, GlcNAc derived from chitin that is released by growing fungal cells is also likely to be available extracellularly. GlcNAc is an interesting monosaccharide to promote cell fate determination in fungi as it has been implicated as a conserved signaling molecule across all kingdoms of life including its ability to promote morphogenesis in bacteria [22] and also function as a dynamic intracellular signaling modification akin to phosphorylation in metazoans (O-GlcNAc signaling) [23].
Herein we describe the role of GlcNAc as a potent inducer of the yeast-to-filamentous phase transition at RT in thermally dimorphic fungi. Culturing H. capsulatum and B. dermatitidis yeast cells in the presence of exogenous GlcNAc promoted a rapid and more synchronous phase transition of yeast cells to filaments. GlcNAc also promoted faster growth of differentiated H. capsulatum filaments at RT, indicating that GlcNAc influences both the morphogenesis and growth rate of H. capsulatum filaments. In addition to implicating GlcNAc as a critical signal for filamentation, these studies allowed us to examine the temporal regulation of the H. capsulatum transcriptome during morphogenesis in a synchronous population of cells. The resulting analysis provided the first view of transcriptional changes of a thermally dimorphic fungus undergoing yeast-to-filament differentiation and revealed candidate genes that may play roles in establishing and maintaining the filamentous growth program. Furthermore, we found that GlcNAc-promoted filamentation of H. capsulatum is dependent on two genes (NGT1 & NGT2) that encode putative GlcNAc major facilitator superfamily (MFS) transporters. These proteins have homology to C. albicans Ngt1, the only previously characterized eukaryotic GlcNAc transporter [24]. We show that H. capsulatum Ngt1 and Ngt2 can each serve as a GlcNAc transporter. Additionally, NGT1 and NGT2 were required for efficient yeast-to-filament conversion even in the absence of exogenously added GlcNAc. These data suggest that the ability to sense and respond to endogenous GlcNAc through Ngt transporters could be a critical regulatory step during filamentous growth. Finally, taken together with previous work in Candida species, our results indicate that GlcNAc functions as a highly conserved cue to signal morphogenesis in the fungal kingdom.
To determine whether GlcNAc can stimulate morphogenesis of thermally dimorphic fungi, we grew H. capsulatum and B. dermatitidis yeast cells in liquid culture in the presence or absence of exogenous GlcNAc (HMM/100 mM GlcNAc, see Materials and Methods for complete media descriptions and note that HMM medium contains residual 10 mM glucose from the F12 nutrient supplement even before further supplementation of the sugar source) at 37°C or at RT. GlcNAc did not affect yeast-phase morphology at 37°C (Figure 1 A, B), but it triggered a remarkably rapid transition of yeast cells to filaments at RT (Figure 1 C, D). This robust filamentation was in stark contrast to the usual laboratory transition experiment in glucose medium, where it takes weeks to yield a large, homogenous population of filaments from yeast cells (Figure 1 C, D). To assess the concentration-dependence of GlcNAc-enhanced filamentation, we plated serial dilutions of H. capsulatum yeast cells at RT on standard glucose medium containing increasing concentrations of GlcNAc. The enhanced filamentation of H. capsulatum at RT in response to GlcNAc occurred at micromolar concentrations, as evidenced by larger colony diameter and fuzzy colony morphology that was in contrast to yeast cells grown in these media conditions at 37°C, which exhibited no morphological changes (Figure 2 A, B). Since these concentrations are too low for GlcNAc to be utilized as the major carbon source, these data suggested that H. capsulatum cells may be sensing GlcNAc, or one of its catabolic byproducts, to promote morphological differentiation. To confirm that supplementing cultures with an additional carbon source was not sufficient to promote filamentation, we grew H. capsulatum yeast cells in equimolar amounts of glucose or GlcNAc and monitored their conversion to filaments at RT. Cells grown in additional glucose did not show enhanced filamentation at RT in comparison to GlcNAc-grown cells (Figure S1), indicating that simply providing an additional carbon source during aerobic growth is not sufficient to promote morphogenesis of H. capsulatum. Furthermore, the ability of GlcNAc to promote filamentous growth in H. capsulatum was a unique property of GlcNAc as other carbohydrates, including fructose and the amino sugar glucosamine (GlcN), did not enhance morphogenesis at RT (Figure 2 C, D).
In addition to the ability of GlcNAc to promote a faster yeast-to-filament transition, we also examined the effect of GlcNAc on yeast and filamentous cell growth rates at RT. First, we examined the growth of yeast cells at RT before they converted to filaments in GlcNAc and glucose media. We observed that yeast cells grew at a very slow rate at RT prior to conversion to filaments irrespective of whether GlcNAc was present in the medium (Figure S2 A, B). Thus, GlcNAc does not affect the growth rate of yeast cells at RT before their conversion to filaments. However, monitoring the rate of increase in diameter of filamentous colonies revealed that GlcNAc did augment the growth rate of filaments at RT (Figure S2C) even at low concentrations (10 mM GlcNAc). These data suggested that in addition to promoting a faster yeast-to-filament transition, GlcNAc also enhanced the growth rate of filamentous cells. Overall, these conversion and growth experiments show that GlcNAc promotes a specific, rapid, and synchronous switch of thermally dimorphic fungal yeast cells to filaments at RT and that growth in the filamentous form is stimulated by GlcNAc.
The ability of GlcNAc to promote a faster and more synchronous transition of H. capsulatum yeast cells to filaments at RT enabled us to examine the transcriptome of H. capsulatum cells as they underwent the transition from the yeast form to filaments using whole-transcriptome microarray profiling. Previous transcriptional profiling experiments that defined yeast- and filament-enriched transcripts in H. capsulatum have profiled the transcriptomes of fully differentiated yeast cells or filaments grown at 37°C or RT, respectively [11]–[13], which is distinct from deciphering the dynamic changes in transcript expression patterns in cells undergoing a morphological transition. To identify transcripts that are regulated during morphological differentiation as well as to begin to understand how H. capsulatum responds to GlcNAc to promote filamentation, we monitored the transcriptome of yeast cells as they began to form filaments at RT in the presence and absence of GlcNAc. We grew yeast cells at 37°C in standard HMM glucose medium to early-log phase (t = 0), resuspended the cells in fresh glucose or GlcNAc medium (HMM/100 mM GlcNAc), and then shifted the yeast cells from 37°C to RT (see Figure 3A) to monitor transcriptional changes that occurred during the yeast-to-filament transition. For technical reasons, two time-courses were performed: the first (1 h, 4 h, 24 h) to capture early transcriptional changes, and the second (4 d, 7 d) to capture later transcriptional changes (see Figure 3A). As a point of comparison, yeast cells were also grown at 37°C in GlcNAc (HMM/100 mM GlcNAc) or glucose medium for the duration of the time-course; these samples allowed us to identify genes that were induced by growth in GlcNAc at both RT and 37°C, independent of temperature. At each timepoint, RNA was harvested and cellular morphology was examined by microscopy. Morphological examination of yeast cells shifted to RT revealed that GlcNAc-grown cells were predominantly filamentous after 4 days of growth at RT while glucose-grown cells remained predominantly yeast-like for the duration of the time-course (Figure S3A). Yeast cells grown at 37°C in either GlcNAc or glucose medium remained as budding yeast for the duration of the time-course (Figure S3B).
Gene expression ratios from cells grown at RT in the presence of GlcNAc or glucose were subjected to k-means clustering, which revealed distinct cases of dynamically controlled transcript expression during morphogenesis. Although a small number of genes were induced or repressed only in glucose or in GlcNAc medium (e.g., Groups 1, 4, 8, and 9, respectively; Figure 3B), the majority of genes showed similar patterns of expression over replicate time courses in the presence of both sugars (Figure 3B). Namely, early repressed, late repressed, early induced, late induced, and GlcNAc induced transcript expression patterns emerged as yeast cells transitioned to filaments in GlcNAc and glucose media (Figure 3B). Overall, these data highlighted the complex reprogramming that yeast cells employ during cellular differentiation.
Closer examination of the identities of temporally regulated transcripts during morphogenesis uncovered genes that may play a role in establishing the filamentous growth program. Factors involved in fatty acid biosynthesis (FAS1, FAS2, ACC1, & OLE1) [25], which could serve to alter plasma membrane fluidity in response to a change in temperature or cell wall structure, were upregulated in glucose- and GlcNAc-grown yeast cells transitioning to filaments (Figure 4A). We also found genes upregulated during cellular differentiation that could play a role in signal transduction including HMK1, which is predicted to encode a mitogen activated protein kinase, RYP4 and CPH1, which are predicted to encode Zn2C6 and C2H2 transcription factors, respectively, and PHK1, a predicted two component sensor kinase. Notably, the S. cerevisiae homolog of HMK1 (named KSS1) is involved in signal transduction pathways that control filamentous growth [26]. Interestingly, many of the transcripts depicted here (Figure 4A) that were upregulated during morphogenesis are not strongly enriched in fully differentiated filaments, further suggesting that these transcripts may play a role in establishing, but not maintaining, the filamentous growth program.
To put the transcriptional expression patterns observed during morphogenesis in context with what is already known about transcript levels in fully differentiated filaments, we examined the expression patterns of canonical filament-phase specific (FPS) transcripts (MS95, TYR1, HYD1, EFG1, NIR1, and FBC1 [11], [12]) during morphogenesis. Transcript levels of MS95 were upregulated immediately (t = 1 h) upon shifting yeast cells to RT in both GlcNAc and glucose growth conditions, suggesting that MS95 is responsive to the decrease in temperature, either at an extremely early point in the yeast-to-filament transition or even before that developmental program has initiated (Figure 4B). MS95 is homologous to the C. albicans DDR48 gene, which is a stress-response gene that is important for filamentous growth in C. albicans [27]. In contrast to MS95, upregulation of the predicted tyrosinase-encoding TYR1 transcript in yeast cells transitioning to filaments was not observed at early timepoints after temperature shift (1 h or 4 h; Figure 4B), signifying that TYR1 may be induced once the filamentous growth program has initiated, and is not directly regulated by temperature. To further examine the transcriptional expression kinetics of TYR1 during filamentation, we introduced a GFP reporter construct in which expression of GFP was controlled by approximately 1 kilobase (kb) of the TYR1 promoter (PTYR1 - GFP) into H. capsulatum and monitored levels of GFP by confocal microscopy as yeast cells transitioned to filaments at RT in glucose or GlcNAc medium. In accordance with our microarray data, yeast cells converting to filaments in GlcNAc medium exhibited earlier PTYR1 - GFP expression (robust GFP signal detected by 36 h) than glucose-grown cells (Figure S4). Furthermore, PTYR1 - GFP expression was not detected in yeast cells grown at 37°C (data not shown), nor was PTYR1 - GFP expression immediately observed upon the transition of yeast cells to RT (t = 12 h or 20 h, Figure S4). Together, these data confirm the enhanced rate of filamentation in GlcNAc medium, and indicate that the TYR1 transcript is turned on during the filamentous growth program and not directly in response to changes in temperature.
The FPS HYD1, EFG1, NIR1, and FBC1 transcripts, in contrast to MS95 and TYR1, were not robustly upregulated in either glucose- or GlcNAc-grown yeast cells transitioning to filaments despite being highly enriched in fully differentiated glucose-grown filaments (Figure 4B). Thus, HYD1, EFG1, NIR1, and FBC1 transcripts may not be involved in establishment of the filamentous state, and may instead play a role in maintenance of this cell type. We also examined the temporal expression patterns of canonical yeast-phase specific (YPS) transcripts in our transcriptional time-course with the expectation that we would see expression of YPS transcripts such as CBP1, SID1, SSU1, YPS21, CATB, and MFS2 [11], [28]–[30] downregulated as the yeast cells converted to filaments. SSU1, YPS21, CATB, and MFS2 transcripts behaved as we expected, being unchanged or downregulated in expression upon shifting yeast cells to RT (Figure 4B). CBP1 and SID1, however, were unexpectedly more upregulated in GlcNAc- versus glucose-grown yeast cells transitioning to filaments (Figure 4B; upregulation of CBP1 was confirmed by qRT-PCR (data not shown)). CBP1, a gene of unknown molecular function, and SID1, a monooxygenase involved in siderophore biosynthesis, are required for the virulence of H. capsulatum yeast cells [31]–[33]. The upregulation of CBP1 and SID1 transcripts in GlcNAc-grown filaments may reflect a previously unappreciated function for these transcripts in the biology of filamentous cells. Additionally, since GlcNAc-stimulated filaments also grow more rapidly than glucose-grown filaments (see Figure S2C), these data could reflect a correlation between increased growth rate of cells and increased expression of CBP1 and SID1. The unexpected expression patterns of these previously identified phase-specific genes highlight the importance of investigating the temporal regulation of transcripts during morphogenesis.
In addition to the aforementioned transcripts temporally regulated in both GlcNAc- and glucose-grown yeast cells transitioning to filaments, we also identified a class of transcripts robustly induced only in GlcNAc medium (Group 1, Figure 3B). This group of genes was of interest, as it represents a means to understand the mechanism of GlcNAc-promoted filamentation. Further analysis of GlcNAc-induced genes using hierarchical clustering identified a subset of genes that is robustly induced in GlcNAc but not glucose medium at RT (Figure 5A). Whereas many of the GlcNAc-induced transcripts are of unknown function, we noted that OGA1 was particularly intriguing because it could play a signaling role in GlcNAc-promoted filamentation. OGA1 encodes a putative O-GlcNAcase with homology to the human OGA enzyme that cleaves the O-GlcNAc signaling modification from serine and threonine residues [23]. This modification is involved in a variety of signaling processes in metazoan cells [34], but has not been functionally investigated in fungal cells. OGA1 has not previously been identified as a filament-enriched transcript [12] and is induced during the rapid and synchronous transition to filaments in GlcNAc medium. Therefore, we hypothesize that it could play a role in the robust morphologic transition to filaments that occurs in GlcNAc-treated cells. Of note, the H. capsulatum homologs of GIG1 and the galactose catabolic genes were not found to be induced by GlcNAc in H. capsulatum as they are in C. albicans [35], indicating that their regulation may not be significant for GlcNAc-promoted filamentation of thermally dimorphic fungi.
GlcNAc also induced the expression of putative GlcNAc utilization genes in H. capsulatum, including a GlcNAc transporter (NGT1), GlcNAc hexokinase (HXK1), GlcNAc-6-phosphate deacetylase (DAC1), and GlcN-6-phosphate deaminase/isomerase (NAG1; Figure 5 A, B and see Figure S5). These genes were induced in GlcNAc at both RT and 37°C independent of cellular morphology (Figure 5B). H. capsulatum Ngt1, Hxk1, Dac1, and Nag1 are the best BLASTP hits to the functionally characterized C. albicans GlcNAc utilization machinery (see Materials and Methods) [24], [36], [37]. Analysis of the genomic positions of H. capsulatum NGT1, HXK1, DAC1, and NAG1 genes showed that they are clustered in a 75 kb region in the H. capsulatum genome similarly to the C. albicans NAG1, DAC1, and HXK1 GlcNAc catabolic genomic cluster [38] (see Figure S6 and Table S6). We identified a fifth putative GlcNAc utilization gene (HISTO_ZL.Contig1131.fgenesh_plus.101.final_new) by its location in the H. capsulatum GlcNAc utilization genomic cluster (Figure S6 and Table S6) and its transcriptional induction by GlcNAc (Figure 5B). This gene, which we named GIT7 (GlcNAc-Induced Transcript 7), is predicted to encode a β-hexosaminidase domain (Panther Accession: PTHR30480). Despite the conservation of Git7 across many fungal species, including C. albicans, Aspergillus fumigatus, and B. dermatitidis, the biological function of Git7 in GlcNAc utilization is unknown. Together, these data suggest that the H. capsulatum genome encodes and transcribes all of the genes known to be necessary for GlcNAc catabolism as well as an additional uncharacterized gene, GIT7, which could be involved in GlcNAc utilization.
After identifying H. capsulatum Ngt1, we also noticed a second previously uncharacterized gene (HISTO_ER.Contig17.eannot.1311.final_new) with strong sequence homology to the C. albicans Ngt1 transporter by BLASTP analysis (E = 8.2×10−90). Since this gene is predicted to encode a second Ngt-like transporter in H. capsulatum, we named it Ngt2. Similarly to the C. albicans Ngt1 [24], Ngt2 is predicted to be a MFS transporter (Interpro Accession: IPR011701) with 12 transmembrane-spanning regions. However, in contrast to NGT1, NGT2 mRNA is not robustly induced by GlcNAc (Figure 5B) nor is NGT2 clustered in the genome near other H. capsulatum GlcNAc utilization genes (see Figure S6 and Table S6). The presence of two putative GlcNAc transporters in the H. capsulatum genome (NGT1 & NGT2) is surprising as C. albicans requires only one transporter, Ngt1, to transport GlcNAc across its cell membrane [24]. To determine the prevalence of multiple GlcNAc transporters across fungal species, we assessed the phylogenetic conservation of H. capsulatum Ngt1 and Ngt2. We used Bayesian analysis to build a phylogenetic tree with aligned BLASTP homologs to H. capsulatum Ngt1 and Ngt2 (E≤1×10−5) from 20 Ascomycetes and Basidiomycetes species with sequenced genomes (Figure S7). From this analysis, an Ngt1/Ngt2 clade was identified and the phylogenetic model was simplified and then rebuilt to include only species with Ngt1 or Ngt2 orthologs (Figure 6 and Table S5). Ngt1 orthologs were found throughout the Ascomycetes; notably, however, the model yeast S. cerevisiae lacks a homolog to Ngt1 and experimentally has been shown to be unable to efficiently transport GlcNAc across its cell membrane [24]. Conversely, the presence of Ngt2 in fungi is more restricted than the occurrence of Ngt1, with Ngt2 orthologs found only in the Onygenales (H. capsulatum, B. dermatitidis, Coccidioides spp., Trichophyton verrucosum, and Uncinocarpus reesii) and Eurotiales (Penicillium marneffei and Aspergillus spp.) orders of Ascomycetes (Figure 6). Many of the species that have multiple identifiable GlcNAc transporters (i.e., Ngt1 and Ngt2) are human fungal pathogens (H. capsulatum, B. dermatitidis, Coccidioides spp., T. verrucosum, A. fumigatus, and P. marneffei), which is interesting given that the majority of characterized Ascomycetes are plant pathogens or plant saprobes [39].
To investigate whether H. capsulatum Ngt1 and Ngt2 are capable of transporting GlcNAc, we expressed H. capsulatum NGT1 and NGT2 in the C. albicans ngt1Δ strain, which is unable to grow in medium where GlcNAc is the sole carbon source [24]. We overexpressed either H. capsulatum NGT1 or NGT2 in the C. albicans ngt1Δ mutant and examined the ability of these strains to grow in GlcNAc medium as compared to vector control strains. Expression of either H. capsulatum NGT1 or NGT2 in the C. albicans ngt1Δ mutant conferred growth on GlcNAc medium (Figure 7A), indicating that both H. capsulatum Ngt1 and Ngt2 can mediate GlcNAc transport independently. As expected, complementation of the C. albicans ngt1Δ mutant with either H. capsulatum NGT1 or NGT2 did not affect growth on other carbon sources including glucose and galactose (Figure 7A).
To probe whether NGT1 and NGT2 function in GlcNAc transport and utilization in H. capsulatum, we attempted to disrupt each gene, but were unsuccessful since targeted gene disruption in H. capsulatum is highly inefficient. Thus we used RNA interference (RNAi) to deplete levels of NGT1 and NGT2 transcripts. We confirmed knockdown of NGT1 and NGT2 transcripts by qRT-PCR and noted that wild-type levels of NGT1 mRNA were dependent on NGT2 whereas depleting levels of NGT1 had little effect on NGT2 transcript levels (Figure S8 A, B). Nucleotide regions that were unique to either NGT1 or NGT2 were chosen for targeting by RNAi to minimize the possibility of cross-silencing of NGT1 by the NGT2 RNAi construct, or vice versa. With NGT1 and NGT2 RNAi strains in hand, we evaluated the ability of these strains to grow in medium where GlcNAc was the only carbohydrate source. We grew vector control, NGT1, and NGT2 RNAi strains in minimal medium (3M) using either GlcNAc or glucose as the only carbohydrate source. As predicted, NGT1 and NGT2 RNAi strains exhibited no growth defects in glucose medium, indicating that NGT1 and NGT2 do not affect overall fitness of H. capsulatum at 37°C (Figure 7B). However, in medium where GlcNAc was the only carbohydrate source, both NGT1 and NGT2 RNAi strains were unable to achieve the same growth density as control cells (Figure 7C). In contrast, the baseline level of growth exhibited by H. capsulatum in minimal medium containing no exogenous sugar source was indistinguishable for NGT1 and NGT2 RNAi strains and controls (the carbon source for these cells is presumed to be the proline and cystine present in 3M minimal media; Figure 7D).
Given that H. capsulatum cells respond to GlcNAc by upregulating GlcNAc utilization genes (Figure 5B), we predicted that NGT1 and NGT2 RNAi strains would be defective in the upregulation of NAG1, DAC1, and NGT1 transcripts in response to GlcNAc. We first examined the responsiveness of NGT1 and NGT2 expression levels to GlcNAc. Whereas NGT1 is induced approximately 13 fold by GlcNAc in control cells, GlcNAc induction of NGT1 is reduced in strains that target either NGT1 or NGT2 by RNAi (Figure 8A). In contrast, as observed by microarray and qRT-PCR, the expression level of NGT2 was only minimally affected by GlcNAc (Figure 5B and 8B), and was not dependent on NGT1 (Figure 8B). Maximal induction of NAG1 and DAC1 transcripts by GlcNAc was abolished in NGT1 and NGT2 RNAi strains (Figure 8 C, D), suggesting that Ngt1 and Ngt2 play a role in GlcNAc catabolism and utilization in H. capsulatum. Interestingly, we also noticed that NGT1 RNAi strains exhibited a slight defect in upregulation of the GlcNAc catabolic genes NAG1 and DAC1 even in the absence of adding exogenous GlcNAc to the medium (Figure S9 A, B). These data support the idea that cells might scavenge endogenous GlcNAc that becomes available as cells divide (i.e., from cell wall turnover during growth in glucose as the major carbon source) and indicate that NGT1 is required for this process. In sum, these data indicated that Ngt1 and Ngt2 most likely contribute to GlcNAc catabolism in H. capsulatum by mediating GlcNAc transport.
In addition to examining the role of NGT1 and NGT2 in regulating genes involved in GlcNAc catabolism, we also examined whether they regulate the expression of GlcNAc-induced transcripts (Figure 5A) that do not have a predicted role in GlcNAc catabolism. We examined the upregulation of PTR2 and OGA1 transcripts in response to GlcNAc in the NGT1 and NGT2 RNAi strains by qRT-PCR. We observed that OGA1 expression was dependent on NGT1 and NGT2 while PTR2 expression was dependent only on wild-type levels of NGT2 (Figure S10). Similarly to the GlcNAc catabolic genes NAG1 and DAC1, OGA1 expression was dependent on NGT1 even in the absence of exogenous GlcNAc (Figure S9 C, D), providing more evidence that Ngt1 influences the expression of GlcNAc-induced transcripts even when glucose is the major carbon source.
In C. albicans, GlcNAc-mediated filamentation is dependent on Ngt1, but independent of Hxk1 [24], [40]; Hxk1 is required for catabolism of GlcNAc as a carbon source and utilization of GlcNAc in glycan biosynthesis (see Figure S5). These data indicate that GlcNAc is likely being recognized intracellularly by C. albicans as opposed to being catabolized or utilized in glycan biosynthesis to promote filamentous growth. Furthermore, in H. capsulatum we found that Ngt1 and Ngt2 were necessary for the induction of transcripts that were upregulated during the GlcNAc-promoted yeast-to-filament transition. Thus, to investigate the role of H. capsulatum Ngt1 and Ngt2 in morphogenesis, we first confirmed that expression of H. capsulatum NGT1 or NGT2 in the C. albicans ngt1Δ strain restored filamentation in response to GlcNAc (Figure S11). Next, we evaluated whether the H. capsulatum NGT1 and NGT2 RNAi yeast cells were defective in GlcNAc-induced filamentation. Vector control, NGT1, and NGT2 RNAi yeast cells grown at 37°C were inoculated into either HMM glucose medium or HMM glucose medium supplemented with 10 mM GlcNAc and switched to RT to monitor their conversion to filaments over time by live-cell imaging. Depletion of NGT1 or NGT2 transcripts resulted in a dramatically slower conversion of yeast cells to filaments in GlcNAc medium as compared to vector control strains (Figure 9), indicating that NGT1 and NGT2 mediate GlcNAc-promoted filamentation. To investigate whether NGT-mediated GlcNAc filamentation was dependent on the catabolism or utilization of GlcNAc by H. capsulatum cells, we depleted transcript levels of the GlcNAc kinase, HXK1, using RNAi (Figure S8C). Depletion of HXK1 severely reduced the ability of H. capsulatum cells to grow in minimal medium (3M) containing GlcNAc as the only carbohydrate source (Figure S12A), demonstrating that HXK1 functions as a GlcNAc kinase during catabolism. Next, we evaluated the ability of HXK1 RNAi yeast cells to filament in response to GlcNAc and noted no defect in the ability of HXK1 RNAi yeast cells to transition to filaments in response to GlcNAc as compared to vector control cells (Figure S12 B, C). Thus, in H. capsulatum NGT-mediated filamentation in response to GlcNAc does not require GlcNAc catabolism or utilization.
Unexpectedly, NGT1 and NGT2 RNAi yeast cells also exhibited a defect in the ability to convert to filamentous cells at RT in glucose medium (Figure 10), indicating that Ngt1 and Ngt2 play a general role in mediating the yeast-to-filament transition at RT in H. capsulatum. We hypothesized that the GlcNAc polymer chitin found in fungal cell walls could be a source of endogenous GlcNAc for H. capsulatum cells; thus, we examined whether chitin could stimulate morphogenesis similarly to GlcNAc. Indeed, H. capsulatum yeast cells grown in the presence of chitin more robustly converted to filaments at RT (Figure S13), suggesting a potential endogenous source of GlcNAc that H. capsulatum cells could be monitoring. Taken together, our experiments indicate that exogenously added GlcNAc stimulates a morphogenetic pathway in H. capsulatum that facilitates temperature-dependent filamentous growth in a Ngt1- and Ngt2-dependent manner. Additionally, our data reveal that Ngt1 and Ngt2 are required for efficient filamentation in the absence of GlcNAc supplementation, likely due to the role of endogenous GlcNAc in the regulation of morphogenesis.
Here we demonstrated that the ubiquitous amino sugar GlcNAc robustly promotes morphogenesis of the thermally dimorphic fungal pathogens, H. capsulatum and B. dermatitidis. Historically, temperature has been thought of as the only signal necessary to induce morphogenesis of thermally dimorphic fungi; however, the discovery that exogenous GlcNAc represents a secondary signal important for efficient yeast-to-filament conversion indicates that combinatorial signals are integrated by thermally dimorphic fungi to cue morphogenesis. In addition to its role in promoting morphogenesis, GlcNAc also stimulated faster growth of differentiated H. capsulatum filamentous cells at RT. This was surprising as GlcNAc does not appear to be the optimal carbon source for H. capsulatum (glucose is a more efficiently utilized carbon source by H. capsulatum yeast cells in vitro), and suggests that GlcNAc stimulates filamentous growth by a mechanism independent of metabolic flux. In trying to understand the mechanism of GlcNAc-promoted filamentation, we focused on genes that were transcriptionally co-regulated in response to GlcNAc. Interestingly, this analysis led to the observation that GlcNAc-promoted morphogenesis of H. capsulatum is dependent on two GlcNAc transporters, Ngt1 and Ngt2. Furthermore, Ngt1 and Ngt2 were necessary for efficient yeast-to-filament morphogenesis even in the absence of exogenous GlcNAc, suggesting that Ngt1 and Ngt2 may function as part of a general autoregulatory mechanism in H. capsulatum, presumably dependent on endogenous GlcNAc (i.e., GlcNAc that could be turned over from the remodeling of chitin in the cell wall that accompanies cell division), that serves to control multicellular filamentous growth.
The kinetics and synchrony of the temperature-induced morphologic switch were greatly enhanced by GlcNAc supplementation, which allowed more robust profiling of the yeast-to-filament transition. Classifying transcript expression patterns based on their temporal regulation during the yeast-to-filament transition of H. capsulatum yielded distinct categories of regulated transcripts. Surprisingly, we noticed similar expression patterns for regulated transcripts in H. capsulatum yeast cells transitioning to filaments at RT in GlcNAc and glucose medium in spite of their disparate cellular morphologies. This could indicate that many of the transcriptional changes that occur during morphogenesis are induced by temperature as opposed to after initiation of the morphologic program. Alternatively, the transcriptional changes we described may not be sufficient to establish morphogenesis, and key transcriptional or post-transcriptional controls of morphogenesis could remain to be identified. Some genes that were induced during the yeast-to-filament transition, but not during static, long-term growth of the filamentous form, included genes that are known to influence filamentation in other, better characterized fungi, making these genes attractive candidates for conserved regulators of filamentous growth across dimorphic fungi. Of note, the transcription factor Cph1 (alias Ste12 in S. cerevisiae), which we found upregulated during the yeast-to-filament transition in H. capsulatum, is required for filamentous growth in both C. albicans [41] and S. cerevisiae [42]. Intriguingly, we also noted the upregulation of a mitogen-activated protein kinase, HMK1, during H. capsulatum morphogenesis. Hmk1 homologs (Kss1 and Cek1 in S. cerevisiae and C. albicans, respectively) function directly upstream of the Cph1 and Ste12 transcription factors in the pathways controlling filamentous growth in C. albicans [43] and S. cerevisiae [26]. In addition to implicating genes in the regulation of morphogenesis in thermally dimorphic fungi, our temporal transcriptome analysis revealed surprisingly dynamic expression patterns for genes that were originally characterized as showing differential expression in static samples of either yeast or filaments. We hypothesize that this reflects unanticipated roles for these transcripts in other aspects of fungal cell biology, highlighting the need to profile expression patterns of dimorphic fungal transcripts over a variety of cellular conditions and timepoints to fully characterize phase-specific changes in gene expression.
From our transcriptional profiling data, we chose to focus on the class of genes transcriptionally co-regulated by GlcNAc to begin to understand the mechanism underlying GlcNAc-stimulated morphogenesis. Our work indicated that the H. capsulatum GlcNAc transporters, Ngt1 and Ngt2, are required for efficient GlcNAc-mediated filamentation. Whether Ngt1 and Ngt2 ultimately mediate GlcNAc-promoted filamentation by directly sensing GlcNAc or controlling GlcNAc transport for intracellular sensing is unclear. We found that micromolar amounts of GlcNAc were sufficient to promote filamentous growth in H. capsulatum even in the presence of 100 mM glucose, suggesting that GlcNAc does not need to be the main carbon source to exert its effects. This also indicates that thermally dimorphic fungi are exquisitely sensitive to levels of extracellular GlcNAc such that low levels of this amino sugar are sufficient to promote robust filamentous growth. Furthermore, we found that knocking down HXK1, the GlcNAc kinase which phosphorylates intracellular GlcNAc as the first enzymatic step necessary for utilizing GlcNAc in glycan biosynthesis or catabolism (see Figure S5), does not alter the ability of H. capsulatum to filament in response to GlcNAc. Thus, we favor the idea that thermally dimorphic fungi are remarkably sensitive to levels of GlcNAc, such that low levels of this amino sugar are sufficient to promote robust filamentous growth in conjunction with the appropriate temperature signal. External GlcNAc might be sensed by Ngt1/Ngt2, or alternatively, GlcNAc could be transported into the cell and sensed by an intracellular mechanism. To date, no GlcNAc sensing mechanism has been identified in fungi and thus, how GlcNAc fits into the complex regulatory networks that control fungal morphogenesis is unknown.
The identification of GlcNAc-induced transcripts may provide clues into a mechanism for GlcNAc-promoted filamentation. Most notably, we identified OGA1 as a GlcNAc-induced transcript in H. capsulatum and demonstrated that Ngt1 or Ngt2 was necessary for its upregulation in response to GlcNAc. OGA1 shares homology with the metazoan O-GlcNAcase (OGA) enzyme that functions as a regulator of the dynamic intracellular metazoan signaling modification termed O-GlcNAcylation [34]. In metazoans, O-GlcNAcylation is a ubiquitous post-translational modification, akin to phosphorylation, that regulates basic cellular processes such as cellular development, transcription, protein turnover, and the cell cycle [34]. It is unknown whether fungi modify proteins with O-GlcNAc, let alone utilize O-GlcNAc as a signaling modification. Future work will examine whether the O-GlcNAc modification exists in fungi, and ultimately, whether the putative H. capsulatum OGA1 enzyme alters O-GlcNAcylation levels to influence morphogenesis. Interestingly, while homologs of genes that mediate O-GlcNAc signaling in metazoans (OGT and OGA1) can be found in H. capsulatum, they are conspicuously absent from S. cerevisiae and C. albicans [44], making H. capsulatum a useful model system in which to study O-GlcNAc signaling. Overall, it will be important to understand how GlcNAc controls cells fate determination in fungi, as this may contribute to our understanding of the role of GlcNAc in cell signaling and developmental processes across all kingdoms of life.
Our study and other recent work in fungi provide insights into the regulation of intracellular GlcNAc levels. In most cells, GlcNAc is thought to primarily exist as UDP-GlcNAc [45], [46], which is the universal nucleotide-sugar donor for glycan biosynthesis. Most eukaryotic cells have long been thought to salvage GlcNAc from glycans in lysosomal compartments and synthesize UDP-GlcNAc de novo via the hexosamine pathway [45], as opposed to salvaging free, extracellular GlcNAc across their plasma membranes using dedicated carbohydrate transporters [47]. However, the recent discovery of functional plasma membrane GlcNAc transporters in fungi suggests that at least some eukaryotic cells can take up extracellular GlcNAc via a transporter-mediated process. Interestingly, homologs of Ngt1 can be found in some metazoans, including humans [24], and it remains to be investigated whether these homologs represent functional GlcNAc transporters. The implications of extracellular, transporter-mediated uptake of GlcNAc in eukaryotic cells changes our understanding of the overall cellular flux and regulation of the essential metabolite UDP-GlcNAc.
Our work highlights that some fungi possess multiple MFS GlcNAc transporters, Ngt1 and Ngt2, potentially allowing these organisms to more precisely control levels of intracellular GlcNAc and thus, UDP-GlcNAc. MFS transporters can exist either as oligomers or monomers in the plasma membrane; however, the oligomeric state for most characterized MFS transporters is unknown [48]. We hypothesize that the H. capsulatum Ngt1 and Ngt2 GlcNAc transporters could be acting (1) in cooperation as a heterooligomeric complex to transport GlcNAc; (2) with complementary functions, e.g., one directly senses GlcNAc and the other transports GlcNAc; or (3) as transporters with different affinities for GlcNAc. Consistent with this latter hypothesis, it was recently proposed that yeast utilize dual-transporter systems with differing affinities for the same substrate to optimize nutrient homeostasis when environmental resources fluctuate [49].
Interestingly, only a subset of fungi that have Ngt1 transporters also have Ngt2. Additionally, in many of the fungi with both transporters, Ngt1 is located in the same genomic regions as other GlcNAc catabolic genes, whereas that is not the case for fungal species that only have Ngt1. Perhaps this differential genomic location reflects the need for alternate transcriptional regulation of GlcNAc utilization genes in organisms that have both Ngt1 and Ngt2. In support of this idea, our data indicates that GlcNAc utilization genes (NGT1, HXK1, NAG1, and DAC1) in H. capsulatum are not repressed by glucose, which is in contrast to C. albicans GlcNAc utilization genes that are highly repressed by glucose. Furthermore, many of the fungi with identifiable Ngt2 transporters are known human fungal pathogens (H. capsulatum, B. dermatitidis, Coccidioides spp., T. verrucosum, A. fumigatus, and P. marneffei), suggesting that Ngt2 might play a role in pathogenesis. Of note, GlcNAc can be utilized as a carbon source in vivo by some mammalian pathogens and commensals including the parasite Leishmania major that resides in macrophage phagolysosomes [50], the commensal bacteria Escherichia coli that resides in the gastrointestinal tract [51], and the bacterial pathogen Salmonella enterica serovar Typhimurium, which is found intracellularly within macrophage vacuoles [52]. It is hypothesized that these microbes, each of which occupies a very different niche within its host, are able to acquire host GlcNAc intracellularly from glycans being recycled within cellular lysosomal and endosomal compartments [45] or extracellulary from mucins [53]. It will be interesting to determine whether Ngt1 and Ngt2 play a role in nutrient acquisition during H. capsulatum macrophage colonization and growth.
One of the most intriguing observations from this work is that Ngt1 and Ngt2 are necessary for efficient morphogenesis in the absence of exogenous GlcNAc. Thus, we hypothesize that Ngt1 and Ngt2 may be sensing levels of extracellular, endogenous GlcNAc to monitor population density and/or signal filamentous growth. It has long been appreciated that microbial cells can regulate their growth in response to changing local environmental conditions as well as fluctuations within their own population density via autoregulatory factors [54]. In fungi, a handful of autoregulatory small molecules and peptides that control density-dependent growth or morphology, interspecies communication, and biofilm formation have been proposed [55]. As the building block of the fungal cell wall polysaccharide chitin, GlcNAc fits the definition of a small molecule that could serve as an autoregulatory factor as its extracellular concentration would be proportional to the number of actively dividing cells due to the extensive remodeling of chitin that accompanies fungal cell division. Since chitin appears to be a more broadly distributed component of the cell wall in filamentous fungi [56] as compared to yeast cells, which primarily accumulate chitin around septa and bud scars [57], it is compelling to speculate that GlcNAc levels could regulate multicelluar filamentous growth. Furthermore, GlcNAc may trigger other changes beyond morphogenesis as it has been implicated as a signal in interspecies communication in the Gram-negative bacterial pathogen Pseudomonas aeruginosa. P. aeruginosa utilizes a two-component response regulator to sense environmental GlcNAc (one source proposed is GlcNAc shed from peptidoglycan of Gram-positive bacteria) to control the production of an antimicrobial factor [58], [59].
Notably, a major conclusion of our work is that widely diverged fungal species are capable of responding to GlcNAc to initiate filamentous growth. GlcNAc-induced filamentation has been observed previously [60] in some members of the Saccharomycetes fungal class (including C.albicans, Candida lusitaniae, and Yarrowia lipolytica), which are much more closely related to each other than to thermally dimorphic fungi (including H. capsulatum and B. dermatitidis from the Eurotiomycetes class) [61]. Equally noteworthy is that C. albicans and thermally dimorphic fungi occupy disparate environmental niches: C. albicans colonizes the mammalian gut and forms filaments at mammalian body temperature during invasive, pathogenic growth whereas thermally dimorphic fungi form filaments in the soil at the ambient environmental temperature. Thus, in spite of the distinct biological milieu occupied by these organisms, filamentation of fungal cells in response to exogenous GlcNAc appears to be deeply conserved, and therefore is likely to play a fundamental role in fungal biology.
Finally, we note that the life cycle of thermally dimorphic fungal pathogens, which includes the ability to switch between a parasitic form (disease-causing state) and a multicellular filamentous form (infectious state), is crucial to their pathogenesis and infectivity. Temperature is the best characterized cue that governs this reversible morphogenesis; however, as we demonstrated with GlcNAc, additional signals facilitate the efficient morphogenesis of thermally dimorphic fungi. It is critical to define the regulatory networks that integrate multiple environmental cues (i.e., GlcNAc and temperature) into a morphogenetic program for the cellular differentiation of these organisms. Ultimately, understanding how H. capsulatum yeast cells transition to filaments will provide insight into the establishment and maintenance of the infectious environmental reservoir of this human fungal pathogen.
Histoplasma capsulatum strains G217B (ATCC26032), G217Bura5Δ (WU15), both gifts from the laboratory of William Goldman, University of North Carolina, Chapel Hill, were grown in HMM (Histoplasma-macrophage medium) broth or plates [62]. Blastomyces dermatitidis strain SLH14081 (gift of Bruce Klein, University of Wisconsin, Madison) was grown in HMM broth. HMM medium, which contains 110 mM glucose, was supplemented when necessary with uracil (Sigma-Aldrich) (200 µg/ml) or as indicated with GlcNAc (Sigma-Aldrich) (referred to as HMM medium supplemented with the mM concentration of GlcNAc indicated throughout the text). In some experiments, GlcNAc was used as the major carbohydrate source in HMM broth and plates by substituting 100 mM GlcNAc in place of the usual major carbon source, 100 mM glucose, to provide an equal molarity of carbon source (referred to as “HMM/100 mM GlcNAc” throughout the text); however all HMM medium retains 10 mM glucose from the Gibco's F12 nutrient supplement (Life Technologies) that is used to make HMM medium [62]. H. capsulatum and B. dermititidis cultures were grown at 37°C under 5% CO2 for yeast-phase growth or at room temperature (RT) for filamentous-phase growth with continuous shaking of liquid cultures on an orbital shaker.
For the microarray time-course study, G217B yeast cells were grown at 37°C in HMM medium and subjected to passage at 1∶25 dilution into HMM medium. After 1 day of growth to early log phase, a portion of cells were harvested for the t = 0 timepoints and the remaining cells were washed in PBS and then resuspended with no dilution into 200 mL HMM (contains 110 mM glucose) and HMM/100 mM GlcNAc (contains 10 mM glucose and 100 mM GlcNAc) media for t = 1 h, 4 h, and 24 h timepoints or resuspended with a 1∶10 dilution into 200 mL HMM (contains 110 mM glucose) and HMM/100 mM GlcNAc media (contains 10 mM glucose and 100 mM GlcNAc) for t = 4 d and 7 d timepoints. Cultures for all timepoints were allowed to continue to grow at 37°C under 5% CO2 for yeast-phase growth or transferred to RT for filamentous-phase growth. For the endpoint microarray experiment, G217B yeast cells were grown for 2 days at 37°C in HMM medium and filamentous cells were grown for 4–6 weeks with passaging 3 times (1∶5 dilution) into fresh HMM medium at RT before reaching a sufficient density of cells for harvesting. At each indicated timepoint, cultures were harvested and processed as described below.
For quantitative reverse transcriptase PCR (qRT-PCR), strains were grown at 37°C in 5 mL HMM medium to log phase. Cultures were synchronized to reach early-log phase (OD600 = 4.0–6.0) the day of the experiment. Cells were washed once in PBS, and a 1∶10 dilution of each strain was inoculated into HMM (110 mM glucose) and HMM/100 mM GlcNAc (contains 10 mM glucose and 100 mM GlcNAc) media. After 2 days of growth, cultures were harvested by centrifugation and total RNA was harvested using a guanidine thiocyanate lysis protocol as previously described [11].
A 457 bp region of NGT1, 517 bp region of NGT2, and a 464 bp region of the HXK1 coding sequences were amplified using G217B cDNA and oligonucleotides OAS2880-81, OAS2876-77, or OAS4193-94, respectively. All primer sequences are included in Table S7. Using Gateway cloning and the entry vector pDONR/zeo (Life Technologies), these PCR products were used to generate BAS662 containing a hairpin repeat of NGT1 and BAS1198 containing a hairpin repeat of HXK1 in backbone vector pFANTAi4 (gift of Bruce Klein, University of Wisconsin, Madison; [63]) or BAS643 containing a hairpin repeat of NGT2 in vector pSB23 (a Gateway-compatible destination plasmid derived from pCR186, which was a gift from Chad Rappleye, Ohio State University). The vector control (BAS506), BAS662 (NGT1 RNAi), and BAS1198 (HXK1 RNAi) constructs were integrated into H. capsulatum strain G217Bura5Δ by the use of an Agrobacterium-mediated gene transfer method as described previously [12], [64]. The episomally-maintained vector control (BAS538) and BAS643 (NGT2 RNAi) were electroporated into G217Bura5Δ as previously described [31]. The method chosen for RNAi plasmid maintenance (i.e., episomal versus integrating) was determined empirically by assessing which method gave the strongest and most consistent decrease of NGT1 and NGT2 expression.
Yeast form cultures of SLH14081, G217Bura5Δ, vector control, NGT1, or NGT2 RNAi strains were grown to early log phase at 37°C in HMM medium, washed once in PBS and sonicated for 3 s to disperse clumps. For the H. capsulatum G217Bura5Δ and B. dermititidis SLH14081 wild-type time-course, 10 µl of 5×106 yeast cells/mL (G217Bura5Δ) or 10 µl of 1×106 yeast cells/mL (SLH14081) were loaded into 28 mm×120 µm M04S CellASIC microfluidic cell culture plates (Millipore) in HMM (contains 110 mM glucose) or HMM/100 mM GlcNAc (contains 10 mM glucose and 100 mM GlcNAc) medium and transferred to RT to monitor conversion to filamentous cells or to 37°C to monitor yeast form growth. For the NGT and HXK1 RNAi time courses, 10 µl of 5×106 yeast cells/mL of H. capsulatum NGT1 RNAi, NGT2 RNAi, HXK1 RNAi, or vector control strains were loaded into 28 mm×120 µm M04S CellASIC microfluidic cell culture plates (Millipore) in HMM medium (contains 110 mM glucose) and HMM medium supplemented with 10 mM GlcNAc (contains 110 mM glucose and 10 mM GlcNAc). At each indicated timepoint, cell morphology was examined using live-cell differential interference contrast (DIC) microscopy with 65, 1.2 µm-thick Z-stack images acquired using a Yokogawa CSU-X1 spinning disk confocal mounted on a Nikon Eclipse Ti inverted microscope with an Andora Clara digital camera and a CFI APO TIRF 60× oil or PLAN APO 40× objective. Images were acquired by and processed in NIS-Elements software 4.10 (Nikon). For the NGT RNAi yeast to filament transition experiment, the number of yeast cells remaining at each timepoint was scored by counting the total number of yeast and filamentous cells visible in a maximum intensity projection of the Z-stack image covering a 37.5×37.5 µm area. Images with no visible yeast cells (i.e., only filamentous cells present) were scored as 0% yeast cells.
Yeast form cultures of G217B were grown to early log phase in HMM medium at 37°C. 10-fold serial dilutions of G217B yeast cells were spotted onto HMM solid medium containing 110 mM glucose in the absence (110 mM glucose only) or presence of 0.1 mM, 0.25 mM, or 1 mM GlcNAc and HMM/100 mM GlcNAc (contains 10 mM glucose and 100 mM GlcNAc) and transferred to RT to monitor filamentous growth or grown at 37°C to monitor yeast phase growth. To monitor growth on various carbon sources, cells were spotted onto HMM solid medium containing 110 mM glucose in the absence or presence of 1 mM GlcNAc, 1 mM fructose (Sigma-Aldrich), or 1 mM glucosamine (Sigma-Aldrich). Cells were also analyzed in liquid medium by inoculating a 1∶10 dilution of G217B yeast cells into 10 mL of HMM medium containing 110 mM glucose and supplemented with 10 mM GlcNAc or 10 mM glucose (Sigma-Aldrich) and then incubated at RT to monitor the transition to filamentous cells or at 37°C to monitor yeast phase growth. Dilution series on plates as well as liquid cultures were monitored for growth and morphology by light microscopy between 6 and 14 days after inoculation.
H. capsulatum homologs to the C. albicans GlcNAc catabolic genes (NGT1 = orf19.5392; HKX1 = orf19.2154; DAC1 = orf19.2157; and NAG1 = orf19.2156; gene identities from http://www.candidagenome.org/) were identified by BLASTP [65]. For phylogenetic analysis, homologs of H. capsulatum Ngt1 and Ngt2 in each indicated fungal species were identified by BLASTP using a cut-off value of E≤1×10−5. After alignment of protein sequences with MUSCLE [66], an unrooted phylogenetic model was generated using MrBayes [67]. NCBI protein accession numbers are given in Table S5 and Figure S8.
Cells were harvested by centrifugation or filtration and total RNA was isolated using a guanidine thiocyanate lysis protocol as previously described [11]. Fluorescently labeled cDNA was synthesized by incorporating amino-allyl dUTP during reverse transcription with Superscript II (Life Technologies) of 15 µg total RNA with oligonucleotide-dT and random hexamers used as primers. Cy3 or Cy5 dyes (GE Life Sciences) were coupled to the amino-allyl group as described previously [68]. For each time-course sample, cDNA was coupled to Cy5 and a reference cDNA pool was made by combining RNA from t = 0 and all late time course samples, which was coupled to Cy3. For end point microarray experiments (i.e., established yeast samples compared to established filamentous samples), G217B yeast cDNA was coupled to Cy5 and filament cDNA was coupled to Cy3. Samples were hybridized to H. capsulatum G217B 70-mer oligonucleotide microarrays. Each microarray contained one or two 70-mer oligonucleotides for each predicted gene in the G217B genome (11,088 gene predictions and a total of 14,820 oligonucleotides per array). Arrays were scanned on a GenePix 4000B scanner (Axon Instruments/Molecular Devices) and analyzed using GenePix Pro, version 6.0 (Molecular Devices), NOMAD 2.0 (http://derisilab.ucsf.edu/microarray/software.html), Cluster 3.0 [69], and Java Treeview 1.1.4r4 (available at http://jtreeview.sourceforge.net). To eliminate elements with low signal, we analyzed only elements for which the sum of the medians for the 635 nm and 532 nm channels was ≥150 intensity units. Gene expression data was filtered for 80% completion and a cut-off value for change in gene expression of >2.0 (log2) for all clusters is shown unless otherwise indicated. Hierarchical or k-means clustering were used as indicated for unsupervised clustering. For k-means clustering, k = 10 and n = 100 parameters were empirically chosen. All microarray data have been deposited at Gene Expression Omnibus (GEO) database at the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/geo/) and are available through the accession number GSE48044.
Yeast form cultures of NGT1 RNAi, NGT2 RNAi, HXK1 RNAi and corresponding vector control strains were grown in HMM medium at 37°C and synchronized to mid log phase for the day of the experiment. Cells were washed in PBS and then resuspended in 3M minimal medium [62] containing no carbohydrate (without glucose) and grown overnight to starve cells. After starvation, cells were inoculated to an OD600 = 0.6 into 3M medium containing equal molarities of glucose or GlcNAc (55 mM), or no carbohydrate. Growth was monitored by measuring the OD600 at each indicated timepoint.
Total RNA was treated with DNase I (Promega). cDNA was synthesized using 3.3 µg of DNase I-treated RNA, Affinityscript reverse transcriptase (Agilent), and oligo(dT) and random hexamer primers. qRT-PCR was performed using 1∶100 dilutions of cDNA, 0.8× FastStart Universal SYBR Green Master Mix (Roche), and 200 nM primers. Reactions were performed using an Mx3000P qPCR system (Agilent) with the Comparative Quantitation program. Cycling parameters were 95°C for 10 min and then 40 cycles of 95°C (30 s) and 55°C (1 min); cycling was followed by dissociation curve analysis. Reactions were analyzed using MxPro software (Agilent). Primer sequences of each qRT-PCR probe are included in Table S7.
Coding sequences of H. capsulatum NGT1 and NGT2 were codon optimized by gene synthesis (GENEWIZ) for expression in C. albicans using the most frequent S. cerevisiae codon and excluding use of the CUG codon (to avoid translation of CUG as serine instead of the canonical leucine which occurs C. albicans [70]). Plasmids for complementation were constructed using PCR and homologous recombination in S. cerevisiae [71]. H. capsulatum codon optimized NGT1 and NGT2 were put under the control of the C. albicans TDH3 promoter and ACT1 terminator followed by the C. albicans URA3 gene. A vector control construct was made identically except it lacked NGT1 or NGT2 coding sequence. All primers are listed in Table S7. Complementation fragments containing ∼350 bp of homology to RPS10 were excised by digestion with PmeI and then integrated into the C. albicans ngt1Δ strain at the RPS10 locus via homologous recombination. Strains were verified by PCR to contain the appropriate NGT gene and then 10-fold dilutions of cells were spotted onto solid Yeast Nitrogen Base minimal medium containing 50 mM of GlcNAc, glucose, or galactose. Growth results were reproducible with separate isolates obtained from independent transformations. C. albicans strains used in this study are listed in Table S8.
Supplementary Materials and Methods can be found in Text S1.
|
10.1371/journal.pcbi.1002088 | The Eps8/IRSp53/VASP Network Differentially Controls Actin Capping and Bundling in Filopodia Formation | There is a body of literature that describes the geometry and the physics of filopodia using either stochastic models or partial differential equations and elasticity and coarse-grained theory. Comparatively, there is a paucity of models focusing on the regulation of the network of proteins that control the formation of different actin structures. Using a combination of in-vivo and in-vitro experiments together with a system of ordinary differential equations, we focused on a small number of well-characterized, interacting molecules involved in actin-dependent filopodia formation: the actin remodeler Eps8, whose capping and bundling activities are a function of its ligands, Abi-1 and IRSp53, respectively; VASP and Capping Protein (CP), which exert antagonistic functions in controlling filament elongation. The model emphasizes the essential role of complexes that contain the membrane deforming protein IRSp53, in the process of filopodia initiation. This model accurately accounted for all observations, including a seemingly paradoxical result whereby genetic removal of Eps8 reduced filopodia in HeLa, but increased them in hippocampal neurons, and generated quantitative predictions, which were experimentally verified. The model further permitted us to explain how filopodia are generated in different cellular contexts, depending on the dynamic interaction established by Eps8, IRSp53 and VASP with actin filaments, thus revealing an unexpected plasticity of the signaling network that governs the multifunctional activities of its components in the formation of filopodia.
| Cells move and interact with the environment by forming migratory structures composed of self organized polymers of actin. These protrusions can be flat and short surfaces, the lamellipodia, or adopt an elongated, finger-like shape called filopodia. In this article, we analyze the ‘computation’ performed by cells when they opt to form filopodia. We focus our attention on some initiators of filopodia that play an essential role due to their interaction with the cell membrane. We analyze the formation of these filopodia initiators in different genotypes, thus providing a way to rationalize the behaviors of different cells in terms of tendency to form filopodia. Our results, based on the combination of experimental and computational approaches, suggest that cells have developed molecular networks that are extremely flexible in their capability to follow the path leading to filopodia formation. In this sense the role of an element of the network, Eps8, is paradigmatic, as this protein can both induce or inhibit the formation of filopodia depending on the cellular context.
| Filopodia, actin-rich, finger-like structures that protrude from the cell membrane of a variety of cell types, play important roles in cell migration, neurite outgrowth and wound healing [1]. Filopodia are characterized by a small number of long and parallel actin filaments that deform the cell membrane, giving rise to protrusions. In order for filaments to grow to the characteristic length observed in filopodia, capping proteins, specialized molecules that inhibit actin polymerization, need to be locally inhibited or sequestered and nucleation of new filaments needs to be favored. Furthermore, individual actin filaments are not sufficiently stiff to deform the cell membrane [2]. Proteins, such as VASP-family proteins are thought to be required to promote the initial transient association of actin filaments as they directly [3] or indirectly antagonize capping proteins [4], capture barbed ends [5] and cross-link actin filament [4], [5]. Furthermore, they can act as processive filament elongators especially upon high-density clustering, at least in vitro [4], [6], [7]. Actin filaments are then further stabilized by other crosslinkers, such as fascin, thus permitting the formation of bundles of sufficient stiffness to overcome buckling and membrane resilience [8]. Thus, in a simplified view, capping proteins can be seen as inhibitors, while bundling proteins are among the necessary components of filopodia formation. Consistently with this picture, removal of Capping Protein (CP) causes an increase in the number of filopodia [9]. Vice versa, cells devoid of the actin crosslinker fascin display a reduced amount of filopodia [8].
This simple rule does not seem to apply easily to the actin remodeler Eps8, which plays complex roles in filopodia formation reflecting its diverse biochemical functions. Eps8 can efficiently cap barbed ends when bound to Abi-1 [10], while it crosslinks actin filaments, particularly when it associates with IRSp53 (Insulin Receptor Tyrosine Kinase Substrate of 53 KD) [11], [12], [13], [14], a potent inducer of filopodia via its ability to bind actin filaments and deform the plasma membrane (PM) through its IMD domain [15]. Consistent with its dual function, the role of Eps8 in filopodia formation is cell context-dependent. In HeLa and other epithelial cell lines, the ectopic expression of Eps8 in the presence of IRSp53 promotes the formation of filopodia, while its removal reduces them [12]. The opposite behavior is observed in primary hippocampal neurons, where genetic removal of Eps8 increases the formation of axonal filopodia [16].
In order to rationalize the information described so far, we propose that the process of filopodia formation proceeds in a step-wise fashion. During an initial phase, multiple and simultaneous binding reactions (primarily involving cappers, bundlers and filamentous actin) lead to the formation of pre-existing filaments into bundles. In a second phase, elongation of these bundled filaments is required to support the extension of filopodia. Hitherto, efforts in modeling filopodia formation have focused on the structure and physical properties of filopodia [17], [18], [19], [20] as well as into the role of specific proteins in modulating the characteristic of individual filopodia [21]. More recently, some models have started to couple a detailed biophysical description of filopodia dynamics with some of the molecules involved with capping and bundling [22], [23]. However, it is extremely challenging to treat with the same model filopodia formation in terms of theory of elasticity or stochastic simulations while keeping track of the full behavior of the complex protein-protein interactions underlying the formation of bundled filaments. Particularly, the effect of modifications (e.g. gene deletions or over-expression) affecting the network has never been approached so far with computational methods.
Here, we combined computational models, in-vitro and in-vivo experiments and describe in mathematical terms the behavior of the protein-protein interaction network underlying the formation of bundled filaments using a minimal but biologically relevant module, centered on the IRSp53/Eps8/VASP pathway, with the aim of defining general principles governing the formation of filopodia in different cellular contexts.
In this section, we first introduce the topology of the network underlying filament bundling (Fig. 1). Together with the network that we intend to model, we also enlist the assumptions adopted to translate the network in mathematical formalism. We then discuss the determination of the parameters used for the simulations.
In-vivo, capping proteins block most of the barbed ends preventing uncontrolled filament elongation [24]. Additionally, most of the G-actin available for polymerization is bound to profilin, a monomeric actin binding protein that promotes the exchange of ADP to ATP and decreases the affinity of monomeric actin for filament pointed ends and spontaneous filament nucleation [25]. Accordingly, in our model, polymerization occurs at barbed ends only (equations in Table 1 in Text S1). Under these conditions, the rate of polymerization is proportional to the total G-actin concentration and to the number of free barbed ends. While local G-actin concentration can vary due to local polymerization and depolymerization fluxes [26], [27], the total concentration of G-actin in cells is maintained buffered through mechanisms involving ATP turnover and actin sequestering proteins [28], and thus we treat it as a fixed parameter in our model. This choice is particularly suited to our analysis, which aims to reproduce steady state behaviors and not transient dynamics. We used a concentration of 10 µM of G-actin available for polymerization in cells as estimated in [27], [29].
As for depolymerization, we introduce dissociation of monomers from barbed ends. Since for our purposes a simplified description of actin polymerization suffices, we ignore pointed ends dynamics, while, following a formalism presented in [30] we include a turnover for actin proportional to the total amount of F-actin. Notably, even if we explicitly account for pointed ends polymerization and depolymerization together with a variable amount of G-actin, the results of the model are qualitatively similar (unpublished results). Finally, since the model is based on ordinary differential equations, we do not explicitly take into account individual filaments with variable amounts of actin, but identify a bulk of polymerized actin, F-actin (Fa).
In the cell types we examine, two cappers, CP and the Eps8:Abi-1 complex, play important roles in filopodia formation. We thus explicitly introduce these two molecular species and their interaction with barbed ends in the network (Fig. 1).
Cells tightly control polymerization by maintaining most barbed ends capped, since uncapped filaments in cellular extracts would elongate due to G-actin concentrations higher than the critical concentration for barbed ends [31]. Thus, in our model we assume that, at the steady state, the nucleation and depolymerization of filaments results in a fixed total number of barbed ends and that the concentration of capping proteins (CP and the complex Eps8:Abi1) is sufficiently high to cap most of them.
The behavior of the system “out of steady state” (e.g., bursts of polymerization giving rise to the growth of individual filopodia) is not analyzed experimentally and thus, as anticipated, will not be reproduced by the simulations. We use the model only to reproduce changes in the steady state behavior of the network in various genetics backgrounds where components of the network are either deleted or over-expressed. Finally, we purposely avoided including the anti-capping activity of VASP family members as its role in filopodia formation is still unclear [32], and little is known as to whether this activity is regulated upon binding of these proteins to IRSp53.
The formation of filopodia requires a number of other components in addition to those included in the model, most importantly fascin. However, we argue that the filopodia initiation complexes Eps8:IRSp53:Fa and VASP:IRSp53:Fa play a critical role likely in the initial phase of filopodia formation when filaments must be congregated in close proximity to the plasma membrane. These two filopodia initiation complexes share the critical and unique property to be anchored, primarily through IRSp53 and its membrane curvature sensing IMD module, to the plasma membrane, and thus show a high affinity for convex membrane curvature [1], [15]. Under these conditions, we hypothesize that the two complexes are ideally located to facilitate the “convergence” of actin filaments by promoting their bundling at the PM-oriented barbed ends. Notably and consistently with our hypothesis, actin filaments bundles have been recently proposed to be necessary for efficient protrusion by filling the space and providing mechanical support to the initial membrane deformation induced by IRSp53 that precedes the extension of filopodia [38]. Based on these considerations, we propose the “initiation of bundling” at the PM as the critical step in filopodia initiation, which is primarily due to the activity of Eps8:IRSp53 and VASP:IRSp53 and their ability to form initiation complexes with F-actin, upon which we focus our attention. Further supporting the important role of IRSp53-complexes in filopodia formation, theoretical studies show that membrane-bound protein complexes that have convex curvature and enhance actin polymerization, are able to initiate membrane protrusions [39]. As such, in our model we limit our analysis to the formation of Eps8:IRSp53:Fa and VASP:IRSp53:Fa, from now on abbreviated as FIC for “filopodia initiation complexes”.
In a given cell population, the concentrations of the two FIC are expected to be distributed according to a normal (Gaussian) distribution centered around a mean value. Notably, only some of the cells of a population will develop filopodia, whereas others will not, accounting for the observation that filopodia formation shows a threshold behavior [40]. Recent models [39], [41] allow us to rationalize the threshold behavior based on a positive feedback loop triggered by FIC localized at the plasma membrane. When the mean concentration of FIC increases over a threshold value, they induce the spontaneous initiation of membrane protrusions through the following positive feedback mechanism: a local higher concentration of initiation complexes induces a higher local actin polymerization and protrusive force, which creates a local membrane protrusion and drives the accumulation of even more complexes since they are attracted to the convex curvature at the protrusion tip. Filament elongation and anti-capping activities might also involved in this second step following the formation of FIC. Importantly, as explained above, both the FIC considered here belong to the class potentially involved in the loop, i.e. they have both convex curvature (IRSp53) and promote actin polymerization against the plasma membrane, by increasing filament stiffness through their bundling activity. Accordingly, we hypothesize, following this model, that only the fraction of cells that reaches the threshold value of initiators concentration can activate the feedback loop and develop filopodia, as shown for a generic system in Fig. 3A.
We can compute the fraction of cells that crosses the threshold for filopodia formation as a function of the mean value of FIC in the cell population, assuming that this latter has a normal distribution of FIC. The resulting fraction of cells developing filopodia has an Error-function (Erf) dependence on the average concentration; it increases linearly as the average concentration increases around the threshold value, and saturates far above or below (Fig. 3B). This result suggests that there is a regime where the average concentration of FIC is linearly proportional to the fraction of cells that develop filopodia. Following this line of reasoning, we focused on a deterministic model that computes the average amount of initiation complexes present in the different genotypes.
To compare filopodia formation among different cell types, rather than measuring the percentage of cells that develop filopodia in a given genotype we normalized their value relative to the wild type (WT). The resulting ‘relative filopodia index’ (RFI), is the fraction of cells forming filopodia at steady state in a population of cells functionally interfered for the gene of interest (e.g., X), divided by the fraction of filopodia forming cells transfected with scrambled RNAi oligo:Accordingly, in the model we did not simply calculate the concentrations of filopodia initiation complexes, but a ‘filopodia initiation index’ (FII) defined as the concentrations of filopodia initiation complexes Eps8:IRSp53:Fa and VASP:IRSp53:Fa, normalized by their concentration in wild type cells:Throughout the manuscript, we will compare these two quantities to test the capability of the model to reproduce experimental data and predict new results.
To perform numerical simulations of filopodia formation in HeLa cells and neurons, we need to know the concentrations of the different species. Previous measurements showed that Eps8 and Abi-1 are present in similar concentrations in the two cell types, while Abi-2 is less concentrated in HeLa cells [16]. We then determined IRSp53 concentration through quantitative immunoblotting, and found that it is expressed at similar concentrations in both cell lines (Fig. S2A). As for VASP, we measured its concentration in HeLa cells to be in the submicromolar range (Fig. S2B). We could not directly measure the concentration of VASP in neurons due to the lack of antibodies equally effective against the mouse and human protein. However, reports in the literature show that Mena and EVL, the other two proteins in the VASP-family, are specifically expressed in brain at micromolar concentrations and that the three members of the family show high and overlapping expression levels in developing brain [42], [43], [44], [45]. Accordingly, we used a concentration of VASP-family protein higher in neurons than in HeLa cells. As for kinetic parameters for the various binding reactions, they were derived from the literature or measured directly (Fig. 2, Fig. S2 and Table 2 in Text S1).
Finally, when a protein was over-expressed, we assumed its concentration was increased 10 fold over its wild type values. For knockdown experiments via RNAi, we assumed that the protein concentration was reduced to 1/10th.
After setting the topology of the network, having established the values of key parameters and identified an output that can be compared with the formation of filopodia, we utilized our model to explain the fundamental observation that removing Eps8 decreases filopodia formation in HeLa cells, but causes an increase in filopodia formation in neurons.
In HeLa cells, genetic experiments measuring filopodia formation were done under conditions of IRSp53 over-expression (a condition that we define as WT), which in the model translates with concentrations of IRSp53 10 times larger than concentrations of Eps8 and VASP. The RFI was then measured in wild type and in cells in which we individually knocked down Eps8 or Abi-1 or functional interfered with VASP proteins or both with Eps8 and VASP simultaneously [12]. We then compared the fold increase in RFI measured in these cells with the fold increase of the FII in the model and found a good agreement (Fig. 4A). According to our model, in HeLa cells over-expressing IRSp53 the majority of Eps8 is bound to IRSp53 and filamentous actin, and very little is capping barbed ends (compare the red bar in the first two panels of Fig. 4B). Similarly, in HeLa cells no Abi-1 or Abi-2 co-immunoprecipitated with Eps8 [16]. We used the model to have an inside view of what happens to filopodia initiators and other protein complexes in the various genetic mutants after RNAi interference of the individual proteins of the network.
Simulations show that Eps8 knock down caused a reduction in the amount of Eps8:IRSp53:Fa (Fig. 4B, compare red and orange bars in the second panel) leading to a decrease in the total amount of filopodia initiation complexes (Fig. 4A). Although VASP and Eps8 compete for the binding with IRSp53, in our model removal of Eps8 did not significantly increase the amount of VASP-family proteins bound to it (see IRSp53:VASP:Fa, where “VASP” includes VASP-family proteins, in Fig. 4B, red and orange bars in the third panel). We confirmed this prediction by immuno-precipitating VASP in WT HeLa and HeLa cells knocked down for Eps8 (Fig. 4C) and verifying that the amount of IRSp53 bound to VASP remained constant. Simulations suggest that VASP's role is very similar to that of Eps8: indeed, functional removal of VASP caused a decrease in VASP:IRSp53:Fa (Fig. 4B, red and green bars in the third panel) and filopodia initiators in general (Fig. 4A). As VASP and Eps8 are redundant activators of IRSp53, the simultaneous down-regulation of both causes an increased reduction in filopodia formation, as predicted by the model (Fig. 4A).
As for the capping activity of Eps8 in this cell line, our simulations suggest that it does not play an important role. The complex Eps8:Abi-1 is very scarce and the removal of Abi-1 did not affect the amount of filopodia initiators (Fig. 4A and red and blue bars in the second and third panels in Fig. 4B).
Thus, our model supports the idea that the primary capping protein in HeLa cells is CP, and that Eps8 acts almost exclusively as a bundling protein downstream of IRSp53.
In hippocampal neurons, removal of the different activators of IRSp53 leads to drastically different effects [16]. Functional interference with all VASP-family proteins inhibits filopodia formation similarly to what observed in HeLa cells after simultaneous ablation of Eps8 and VASP [12], [46] (Fig. 5A, red and purple bars). However, in neurons, but not in HeLa cells, the removal of Eps8 alone causes a large increase in the formation of filopodia along the neuronal shaft (Fig. 5A, red and yellow bars) [16], [46]. We used our model to understand the reasons behind this apparently paradoxical behavior.
The network described in Fig. 1 applies to both HeLa and hippocampal neurons; therefore we used the same set of equations and parameters for both cell types, with the noticeable exception of the concentrations of some proteins, Table 2 in Text S1. In hippocampal neurons, in fact, Abi-2 is expressed at much higher levels than in HeLa [16]. Similarly, all members of the VASP-family proteins are specifically and abundantly expressed in neurons and are presumably in excess with respect to IRSp53 as explained above. Moreover, at variance with respect to the experiments performed in HeLa, the analysis of axonal filopodia was conducted under conditions in which IRSp53 was not ectopically elevated. Accordingly, for neurons in the model we used a value of IRSp53 10 times smaller than in HeLa cells, and Abi-1 (which accounts for the presence of Abi-2) and VASP (which accounts for all VASP family members) were increased by a factor 5 (see Table 3 in Text S1).
The fold change in FIC derived from the simulations of our model were consistent with the experimental results obtained in WT, and Eps8 null hippocampal neurons either in the absence or the presence of a VASP dominant negative, which impairs the functional activity of all VASP family members [16] (Fig. 5A). A deeper analysis of the model's behavior allowed us to rationalize the phenotypes in molecular terms. Simulations of WT hippocampal neurons under condition of limiting IRSp53 (i.e. endogenous levels of the protein) suggest that a significantly higher fraction of Eps8 is bound to Abi-1 or Abi-2 compared to HeLa cells, to form the capping-active Eps8:Abi-1/2 complexes (compare red bar of the first panel in Fig. 4B with red bar of the first panel in Fig. 5B). Consistent with this notion, we previously reported that Eps8 binds a significant amount of Abi-1 and Abi-2 in neurons but not in HeLa cells [16]. Since a minimal fraction of Eps8:IRSp53 is bound to filamentous actin, the major filopodia initiator in neurons consists of VASP-family proteins bound to IRSp53 and Fa (compare red bars of panels two and three in Fig. 5B). Having defined the WT condition in hippocampal neurons, we set to analyze the change in steady state caused by the removal of Eps8.
In our simulations, removal of Eps8 increases the total amount of uncapped ends, causing an increase in the amount of filamentous actin (not shown). Moreover, we also observe an increase in the formation of the VASP:IRSp53 complex, due to the competition between VASP-family proteins and Eps8 for the scarce amount of IRSp53 available. As VASP:IRSp53 binds to filamentous actin with higher affinity than Eps8:IRSp53, the model predicts an increase in initiator complexes (compare red and orange bars in the third panel of Fig. 5B), which gives rise to a fold change in FIC for Eps8 knock out similar to what experimentally observed (Fig. 5A). We confirmed this result by immunoprecipitating IRSp53 in WT and Eps8 knock out neurons and observed that a higher amount of VASP was recovered in the knock out neurons (Fig. 5C). In our model, the increase in filopodia initiators due to Eps8 removal is reversed by the simultaneous functional interference with VASP-family proteins (Fig. 5A and orange and purple bars in panel four of Fig. 5B) consistent with what was experimentally measured [16].
We conclude that the role of Eps8 in neurons is more complex than in HeLa cells: in the former cells, it contributes to capping and competes with VASP-family proteins for the formation of filopodia initiators.
To further validate the model, we used it to make quantitative predictions about novel phenotypes. CP removal has been reported to cause an increase in filopodia formation in multiple cell-lines with high quantities of VASP-family proteins [9], but not in cell lines genetically devoid of VASP. The lack of filopodia formation in these latter cells was interpreted as an indication that VASP-family proteins are required for filopodia formation following the removal of capping proteins. This interpretation is in agreement with our model, according to which VASP induces filopodia formation via the initiator VASP:IRSp53:Fa. Our experiments also support this view, as we showed that VASP in complex with IRSp53 can induce filopodia formation in-vivo and formation of actin bundles in-vitro. However, in our model, VASP is not the only source of filopodia initiators. Eps8:IRSp53:Fa is also capable of inducing filopodia formation independently of VASP. Thus, we reasoned that in a setting where VASP cannot contribute to filopodia formation, CP removal should still lead to an increase in the fraction of cells producing filopodia via the parallel pathway provided by Eps8:IRSp53.
To test this prediction we analyzed the change in filopodia formation induced by CP removal in fibroblasts genetically devoid of VASP and MENA and expressing undetectable levels of EVL (MVD7 cells) [47]. We first measured the concentrations of IRSp53, Eps8 and Abi1, as compared to the concentrations measured in HeLa, and we found that MVD7 cells have less Abi1, more Eps8 and roughly the same concentration of IRSp53 (Fig. S2C and Table 2 in Text S1). Next, as these cells do not normally produce filopodia, we over-expressed IRSp53 (a condition called WT, in analogy to what done with HeLa cells) to induce these structures in a sizeable fraction of cells in the population, and we calculated the IRSp53-dependent relative filopodia index of CP knocked down cells with respect to scrambled siRNA-transfected cells (Fig. 6A–B). Using the calculated concentrations of the relevant proteins of MVD7 cells, while keeping the same binding parameters employed in HeLa (Table 2 in Text S1), the model predicted an increase of FII due to CP removal (Fig. 6C) as compared to the WT. The prediction was verified in-vivo by down-regulating CP via RNAi. Of note, the agreement between FII and RFI is quantitative.
According to the model, the increase in uncapped filaments leads to an increase in filamentous actin, and as a consequence to an increase in IRSp53:Eps8:Fa filopodia initiation complex (compare red and blue bars in Fig. 6D second panel). The increase in uncapped filaments also causes the amounts of Eps8:Abi-1 capping filaments to increase (compare red and blue bars in Fig. 6D first panel), but this was insufficient to compensate for the loss of CP due to the low amounts of Abi-1 present.
Our model predicts that a similar effect should also be observed in HeLa cells over-expressing IRSp53 (Fig. S3), where VASP is present but no longer capable of forming new initiation complexes since almost all VASP molecules are already present in VASP:IRSp53:Fa complexes (Fig. 4B third panel). Consistently, upon down-regulation of CP in HeLa cells via RNAi, we observed an increase in the measured filopodia index similar to that predicted in-silico (Fig. S3B). According to the model, CP removal does not increase the amount of VASP:IRSp53:Fa, already maximal, but increases both Eps8:Abi1:N and Eps8:IRSp53:Fa, this last having a stronger effect than the previous on filopodia formation (Fig. S3C).
We finally asked how much of these results were dependent on the precise choice of parameter values since, although most of them are experimentally measured, other parameters are not (Table 2 in Text S1). A sensitivity analysis showed that our results are largely independent on parameter values in both cell types (Fig. S4).
A number of proteins that regulate filopodia formation have multiple biochemically-diverse functions. For examples, VASP family proteins bundle filaments and protect barbed ends from cappers, Formins nucleate new linear filaments and protect barbed ends from capping, IRSp53 binds and bundles filaments and deforms the PM. Coherently, all those different roles act in concert to promote the formation of plasma membrane-linked, actin filament and bundles required to induce and/or sustain filopodia initiation or elongation. Eps8, instead, exerts actin-related biochemical roles that produce opposite biological effects (capping of filament ends that limits filament elongation, while crosslinking that promotes filament bundling) on filopodia formation. Our mathematical model shows that the dual function of Eps8 as a capper or bundler as function of the different Eps8 complexes can explain the seemingly paradoxical effects of Eps8 down-regulation in filopodia formation in different cell types. Thus we propose that Eps8 represents a molecular switch in the transduction of signaling, either directing the cells towards a reduction or an increase of filopodia, depending on the molecular context.
The model we propose is noticeably simple and yet it successfully reproduces experimental data and even predicts the outcome of new experiments. The biological system and experimental setting we employed justified some of the simplifications of the model. For example, since the phenotype we reproduce is a RFI that describes the time-averaged ability of cells to form filopodia, an overly detailed description of the physical process underlying filopodia formation is not required. For this reason, we did not take intra-cellular spatial localization into account and used values of total protein concentrations considering a cell as a well-stirred system.
Other simplifications concern the molecular players of our network. First, the model focuses only on a subset of well characterized molecules that are involved in generating filopodia, while it lacks some key components that have also recently been implicated in filopodia formation, such as formin or Myosin transporters, like Myosin X, or fascin. This choice was based on recent experiments that have revealed that multiple and independent mechanisms of filopodia formations may concomitantly operate. Indeed, it was recently shown that in neurons filopodia could be formed even after the ablation of all three VASP family members upon expression of Myosin X or the activated Formin, mDia2 [46]. This result was the basis to exclude the above-mentioned molecular pathways from our model.
Secondly, we neglected some components even within the pathway that we considered explicitly, as in the case of the cross linker fascin, which was shown to be essential for filopodia stabilization [8]. In this case, we hypothesize that diverse crosslinking proteins or protein complexes may all be required and act in a hierarchical and coordinated manner to promote filopodia formation. Under this scenario, complexes formed between filamentous actin, IRSp53 and its binding partners Eps8 or VASP may serve as the “initiators” of filopodia by promoting the “convergence and bundling” of actin filaments close to the barbed ends oriented toward the PM mainly by virtue of the established properties of IRSp53 to sense membrane curvature and promote convex membrane deformation. Such a mechanism for the initiation of membrane protrusions driven by actin polymerization was proposed from theoretical analysis [39], [41]. The good agreement found in our study between the experimental results and those obtained with our modeling supports this mechanism for filopodia initiation, and suggests that this hypothesis is worth further investigation.
Thirdly, we have not included all the known biological roles of the proteins under consideration. In particular, recent work showed that VASP may also act as an anti-capper and promote in a processive manner filament elongation [4], . Notably, these latter activities become significant mainly upon high-density clustering of VASP. Although we have not explicitly introduced these additional biochemical properties of VASP, they are partly intrinsic in our model as they may occur in later critical steps of filopodia extension. Our model, indeed, addresses what might be the very first step of filopodia fomation that requires the deformation of the plasma membrane and its coupling with the generation of actin filament bundles to support extension. This event is possibly initiated by proteins, such as IRSp53, and promoted, in a feedback loop fashion (as proposed in [39], [41]) by the bundling of pre-existing filaments. Within this context IRSp53 and VASP may act synergically (as a physically tethered complex) to cause filament bundling, and increased barbed-end polymerization, thus increasing the local protrusive force acting on the membrane. The membrane-curvature sensing domain of IRSp53 completes the positive feedback loop by causing IRSp53:VASP aggregation at the tips of emergent membrane deformations. For the filopodia to grow beyond this initiation stage, the actin filaments must remain uncapped to elongate in a processive fashion and further stabilized into tight bundles by actin cross-linkers, such as fascin.
VASP is likely essential also in this “second phase” by sliding from the side to filament tips. Here, upon clusterization possibly promoted by IRSp53-bound to the deformed plasma membrane, VASP may elongate actin filaments while protecting them from capping. According to this hypothesis, the initial recruitment of VASP from IRSp53 has a dual role – both as a filament crosslinker, bundling actin filaments into sufficiently stiff bundles, and recruiting VASP in close proximity to sites of membrane deformation.
Finally, The RFI represents the fraction of cells that develop filopodia, and thus the probability that cells with a certain genetic background can develop filopodia. The formation of filopodia has been proposed to be triggered by initiators of filopodia at the plasma membrane, by a positive feedback loop [39],[41]. Based on this model, we find that the probability of forming filopodia is linearly proportional to the average concentration of filopodia initiation complexes (FII), the molecular species that contain both F-actin and IRSp53. The model shows that the “probability of forming filopodia” becomes significant only above a threshold value of the filopodia initiator complexes, and it saturates as the amounts of filopodia initiator complexes increase further (Fig. 3B). Interestingly, our finding that real cells obey the above linear relationship suggests that the physiological amounts of initiators in the WT cells are kept close to the threshold value. In this regime, small changes to the concentration of these initiators can both easily induce a consistent increase or decrease of the probability of developing filopodia, thereby determining precisely the number of filopodia forming cells in a population. This sort of behavior is in agreement with previous observations [33] that indicate that cells are naturally positioned close to the filopodia-formation threshold.
Collectively, our systems analysis and experimental results provide a cogent molecular and mathematical framework to account for how the multifunctional activities of the components of this network, with particular emphasis on Eps8 and VASP family of proteins, are controlled in different cellular context. The variable formation of distinct protein complexes either exerting capping activities or promoting filament bundling is key in determining the final biological output through quantitative relationships that only a systems biology approach could reveal. It is of note, for example, that the diverse combinatorial arrangement of a limited numbers of components ensures a level of unexpected plasticity of the network, so that seemingly opposite actin-related activities (from capping to anticapping and filament bundling) can be properly coordinated, ultimately differentially controlling the promotion of filopodia. One implication of these finding is that filopodia may not be considered entities governed by different and entirely independent molecular pathways [46]. Rather, the formation of these structures is finely regulated by a unique network connecting numerous molecular, presumably interchangeable and functionally redundant, players through distinct multi-protein complexes. In this context, our study shows clearly the potential of differentially expressing components of the network in terms of filopodia formation, as HeLa, MVD7 fibroblasts and neurons differ for the total concentration of 3 proteins of the network, and yet the effect in terms of filopodia is dramatic. The experiments performed in VASP-family-deficient MVD7 cells further show how the dynamic interplay of the components of the network underlying filopodia formation makes the system robust even to drastic changes, such as the absence of apparently essential components. In conclusion, our results suggest that the outer layer controlling filopodia formation plays a critical role to make the machinery controlling filopodia formation at the same time adaptable and capable of responding to different extracellular stimuli and environmental conditions.
Cytomegalovirus (CMV)-promoter-based and elongation factor-1 (EF1) promoter-based eukaryotic expression vectors and GST bacterial expression vectors were generated by recombinant PCR. Myc–IRSp53 was a gift from S. Krugman (The Babraham Institute, Babraham Research Campus, Cambridge, UK). All constructs were sequence verified. The antibodies used were: monoclonal anti-Eps8 (Transduction Laboratories, Lexington, KY); rabbit polyclonal anti-GST, anti-Myc 9E10 (Babco, Berkeley, CA); anti-Flag M2 (Sigma-Aldrich, St Louis, MO); rabbit polyclonal anti-VASP (Immunoglobe, Himmelstald, Germany); monoclonal anti-Abi-1 was previously described [48] and monoclonal anti-IRSp53 [12]. HeLa knocked down for CP or control cells were obtained by transfecting cells with short hairpin loop oligos targeting human CP gene (AACCTCAGCGATCTGATCGAC) or scramble oligos (AACCTCAGCGATCTGATTGAC) respectively. For MVD7 cells, we used two Stealth RNAi oligos (Invitrogen) targeting murine CP (T1 = GAACCUCAGCGAUCUGAUCGACCUG; T2 = GAAGCACGCUGAAUGAGAUCUACUU) in combination with the appropriate scrambled oligo (scr T1s = GAACCUCAGUGAUCUGAUUGACCUG; scr T2s = AAGUAGAUUUCAUUAAGCGUGCUUC), as control.
His–Eps8 FL and His-IRSp53 were obtained as previously described [12]. Recombinant VASP was expressed as GST fusion protein in the BL21 Escherichia colistrain(Stratagene, Cedar Creek, TX) and affinity purified using GS4B glutathione–Sepharose beads (Amersham Pharmacia Biotech, Piscataway, NJ). Eluted proteins were dialyzed in 50 mM Tris–HCl, 150 mM NaCl, 1 mM DTT and 20% glycerol. GST–VASP was cleaved from the GST using the PreScission protease (Amersham Pharmacia Biotech, Piscataway, NJ) according to the manufacturer's instructions. Actin was isolated from rabbit muscles and purified in the Ca–ATP–G-actin form by Sephadex G-200 chromatography in G buffer (5 mM Tris–Hcl at pH 7.8, 0.1 mM CaCl2, 0.2 mM ATP, 1 mM DTT and 0.01% NaN3).
Monomeric G-actin was polymerized as previously described [12]. F-actin was mixed with varying concentrations of recombinant and purified proteins (as described in the text) in F-buffer and incubated at room temperature for 30 min. Actin was then labeled with rhodamine–phalloidine and 0.1% DABCO and 0.1% methylcellulose were added to the mixture. The samples were mounted between a slide and a coverslip coated with poly-lysine and imaged by fluorescence microscopy.
HeLa cells, Cos7 cells and Hippocampal neurons were cultured as described in [10], [12] and [16], respectively. VASP-family deficient cells (MVD7) were a kind gift from F. Gertler and were cultured as described [47]. HeLa, Cos7 and MVD7 cells seeded on gelatin and were transfected with the indicated expression vectors using FuGene (Invitrogen, Carlsbad, CA), according to the manufacturer's instructions. After 24 h, cells were processed for epifluorescence or indirect immunofluorescence microscopy. Briefly, cells were fixed in 4% paraformaldehyde for 10 min, permeabilized in 0.1% Triton X-100 and 0.2% BSA for 10 min and then incubated with the primary antibody for 45 min, followed by incubation with the secondary antibody for 30 min. F-actin was detected by staining with rhodamine–phalloidine at a concentration of 6.7 U ml−1.
Hela: Epitope-tagged IRSp53 expressing or control cells seeded on gelatine were transfected with CP or control oligos using Oligofectamine (Invitrogen, Carlsbad, CA), according to the manufacturer's instructions.
MVD7: Epitope-tagged IRSp53 expressing or control cells seeded on gelatine were subjected to a double transfection protocol with CP or control oligos using Lipofectamine RNAiMAX (Invitrogen, Carlsbad, CA), according to the manufacturer's instructions.
Images were obtained by applying the Adobe Photoshop filter ‘find edges’ to outline the cell contour. The average total length of protrusions per cell extending from the cell soma was calculated using ImageJ program in at least 30 different cells in triplicate experiments and expressed as fold increase with respect to the average total length of protrusions in control cells. Similarly, the number of branches per cell was manually counted and expressed as above.
Standard procedures in vitro binding, cell lysis and coimmunoprecipitation were as previously described [12].
We solved the equations at steady state using XPP-AUTO or MATLAB.
For numerical analysis in MATLAB, we used the SBtoolbox2 [49]. To do this, we translated the set of equations in SBmodel files.
The steady state of the system was found using the SBsteadystate function, a function that numerically calculates the eigenvalues of the Jacobian matrix of the system.
Exploration of the parameter space in the model was carried out either manually using the SBtoolbox2, or through optimization algorithms found in the SBPD package [49].
|
10.1371/journal.pbio.1001073 | N-Terminal Acetylation Inhibits Protein Targeting to the Endoplasmic
Reticulum | Amino-terminal acetylation is probably the most common protein modification in
eukaryotes with as many as 50%–80% of proteins reportedly
altered in this way. Here we report a systematic analysis of the predicted
N-terminal processing of cytosolic proteins versus those destined to be sorted
to the secretory pathway. While cytosolic proteins were profoundly biased in
favour of processing, we found an equal and opposite bias against such
modification for secretory proteins. Mutations in secretory signal sequences
that led to their acetylation resulted in mis-sorting to the cytosol in a manner
that was dependent upon the N-terminal processing machinery. Hence N-terminal
acetylation represents an early determining step in the cellular sorting of
nascent polypeptides that appears to be conserved across a wide range of
species.
| The eukaryotic cell comprises several distinct compartments, called organelles,
required to perform specific functions. The proteins in these compartments are
almost always synthesised in the cytoplasm and so require complex sorting
mechanisms to ensure their delivery to the appropriate organelle. Of course, not
all proteins need to leave the cytoplasm since many remain there to perform
cytoplasmic functions. It is well known that many proteins are modified by
acetylation of their amino-terminus at a very early stage in their synthesis. We
have discovered a profound difference between the likelihood of such a
modification on cytoplasmic proteins and on those destined for one of the major
organelles, the endoplasmic reticulum (ER): whereas cytoplasmic proteins are
typically acetylated, those bound for the ER are largely unmodified. Moreover,
when specific ER proteins were engineered to induce their acetylation we found
that their targeting to the ER was inhibited. Our data suggest that N-terminal
acetylation is a major determinant in protein sorting in eukaryotes.
| The mechanism of translational initiation dictates that eukaryotic proteins are
synthesized with an amino-terminal methionine residue. In 80% of yeast
proteins studied, the initiating methionine is removed to reveal a new
amino-terminal residue [1], and some 50% of proteins have their amino-terminal
residue acetylated [2],[3]. Hence rather few proteins possess an unmodified
N-terminus. However, while N-terminal processing is widespread, its biological
significance is not well understood. It has been suggested to contribute to
differential protein stability and has recently been shown to function as a degron
for certain cytosolic proteins [4],[5], while in a small number of cases the processed N-terminus
is known to contribute directly to protein function [6]–[9].
Methionine cleavage is catalysed by methionine aminopeptidases (MetAPs) that act
co-translationally as the N-terminus emerges from the ribosome [1],[10]. MetAPs exhibit substrate
specificity and are strongly influenced by the residue at position 2 (P2), with
cleavage favoured by P2 residues with small side chains such as glycine, alanine, or
serine [11],[12]. Yeast and
humans each possess two MetAPs (MetAP1 & 2), and while yeast can tolerate the
loss of either enzyme, the double mutant is lethal demonstrating that methionine
processing is a vital function [13]. Interestingly, MetAP2 is the target for the potent
anti-angiogenic compound fumagillin that exhibits anti-tumourigenic properties [14],[15].
Protein N-termini can also be modified by acetylation of the free α-amino group
by N-α-acetyl transferases (NATs). Five distinct NATs have been identified with
different substrate specificities. NatA normally acetylates N-terminal G, S, A, and
T residues exposed by MetAP cleavage, whereas NatB acetylates methionine residues
that are followed by either D, E, or N at P2 [3],[16],[17]. NatC acetylates certain
methionines with either L, I, W, or F at P2, but other sequence elements influence
processing in this case [18]. NatD appears to be specialised for histone N-acetylation
[19] and finally
NatE acetylates substrates with Leucine at P2 and Proline at P4 [20].
While most proteins remain in the cytoplasm after synthesis, others are targeted to
different compartments. Those destined for the secretory pathway typically possess
an N-terminal signal-sequence which directs them to the endoplasmic reticulum (ER)
[21]. These
proteins are translocated into the lumen of the ER, via the Sec61 translocon,
whereupon their signal-sequence is removed by signal peptidase [22]. A subset of membrane
proteins can be targeted to the ER via non-cleaved internal signal anchor or
C-terminal trans-membrane segments, which act as both targeting and
membrane-integration signals.
N-terminal signal sequences are degenerate in primary structure but are typically
15–30 residues long, and usually comprise charged/polar residues, followed by
6–15 hydrophobic residues and a polar C-terminal region containing the
cleavage site for signal peptidase [23],[24].
In yeast, there are two pathways by which secretory proteins are targeted to the ER.
The co-translational pathway is mediated by Signal Recognition Particle (SRP), which
recognises a signal sequence emerging from the ribosome and targets the
ribosome-nascent chain (RNC) complex to the translocon via SRP-receptor (SR) [25],[26]. The targeted
ribosome then binds tightly to the cytosolic surface of Sec61p allowing the
elongating polypeptide chain to be delivered directly into the translocation channel
[27]–[29]. The alternative “post-translational” pathway
is independent of SRP/SR [30] and targets full-length polypeptides in a reaction that
requires cytosolic chaperones that maintain precursors in a translocation-competent
conformation [31]–[33]. Translocation occurs via the same Sec61-channel, but in
this case, targeting requires the essential integral membrane protein Sec62p that
interacts with precursor and may constitute a specific receptor [34]. Mammalian
cells possess a homologue of SEC62, but this mode of translocation remains poorly
characterized in metazoans [35],[36].
Properties of the signal sequence, and in particular the hydrophobicity of the
central core, determine which pathway a substrate will access, with more hydrophobic
signal sequences utilizing the SRP pathway [30].
Cleavage of the signal sequence reveals a novel N-terminus for the mature
translocated protein, which is located in the ER lumen and so inaccessible to the
N-terminal processing enzymes. The processing status of the initiating methionine of
signal sequences has largely been ignored, particularly as such N-termini are not
detected in proteomic analyses. We therefore decided to investigate the N-terminal
processing of signal sequences using a combination of bioinformatic and experimental
approaches and find that N-terminal modification is incompatible with targeting to
the ER.
Signal sequence recognition and N-terminal processing both occur co-translationally
as the nascent chain emerges from the ribosome [10],[37]. We therefore decided to
investigate whether secretory proteins might be subject to N-terminal processing in
a similar manner to their cytosolic counterparts. As the P2 residue is the major
determinant of N-terminal processing, we first surveyed the amino acid frequency at
this position for signal sequence-containing proteins versus cytosolic proteins
(Figure 1A). Surprisingly,
we found a significantly different frequency distribution between the two sets
(p<0.0001, according to the χ2 test with 18
degrees of freedom). Lysine, leucine, and arginine were most frequent at P2 in
signal sequences but were rarely found at this position in the cytosolic set.
Conversely, while serine and alanine were most frequent at P2 in cytosolic proteins,
these were less evident in signal sequences. A clear pattern emerged when the ratio
of frequencies were compared between the two classes of proteins (Figure 1B); small and acidic
residues were strongly biased towards cytosolic proteins, whereas large and basic
ones were favoured in signal sequences. The frequency of small residues at P2 in
cytosolic proteins predicts that ∼72% of these proteins would be
substrates for MetAP cleavage (Figure
1C), in good agreement with empirical data from proteomic studies [2]. In contrast
only 23% of signal sequences would be predicted to be MetAP substrates (Figure 1C). Hence our data reveal
that for signal sequences there appears to be a strong selection for P2 residues
that would maintain the original N-terminal methionine.
We next addressed whether this bias was of functional significance for ER
translocation. The signal sequence of Carboxypeptidase Y (CPY) [38] begins with “MK”
and so, like most secretory proteins in our analysis, is predicted to remain
unprocessed. Rather than mutating the native P2 residue we chose to insert one of
seven different amino acids between the initiator methionine and the following
lysine residue (Figure 2A). We
then assessed the translocation efficiency of these mutants in vivo by monitoring
their ER-dependent glycosylation (Figure 2B). Insertion of arginine or valine had no effect on the
efficiency of translocation, demonstrating that an insertion at this position does
not inherently perturb signal sequence function. However, the other five insertions
tested all resulted in translocation defects indicated by the accumulation of the
cytosolic precursor form of preproCPY (ppCPY). The most significant defects were
observed for glycine, serine, and glutamate, which are three of the four residues
most biased in their frequency distribution towards cytosolic proteins (Figure 1B). Thus the bias observed
in our bioinformatic analysis correlates with defects in translocation, thereby
implying an important role for P2 in a functional signal sequence.
The inhibitory effects of these various P2 residues might reflect either some simple
perturbation of the signal sequence or their predicted impact on N-terminal
processing. We reasoned that if processing alone were responsible for the effects,
then inhibiting MetAP activity might restore translocation of the mutant proteins.
We therefore analysed translocation in wild-type and Δmap1
cells in the presence of the Map2p inhibitor fumagillin (Figures 2C and S1). In
wild-type (MAP1) cells, fumagillin had little or no effect on the
translocation of native (MK) CPY nor the translocation defects observed for the
various insertion mutants. Similarly, the absence of Map1 alone
(Δmap1) had no discernible effect on any of the
translocation substrates. In contrast, when Δmap1 cells were
treated with fumagillin we found almost complete restoration of translocation for
the MA, MC, MG, and MS mutants. All four are predicted substrates for Met-cleavage,
and our data demonstrate that their inhibitory effects are entirely dependent upon
MetAP activity. In contrast, ME is not a substrate for MetAP and we found that the
translocation defect for this mutant persisted under these conditions. The effect of
fumagillin was therefore substrate-specific, correlating precisely with the known
specificity of MetAPs [11]. We therefore conclude that MetAP-dependent cleavage of a
signal peptide's initiating methionine has a strong inhibitory effect on the
translocation of CPY.
In our analysis, the ME and MS mutations had the strongest effects on translocation
(Figure 2B) and these P2
residues displayed extreme bias against their occurrence in natural signal sequences
(Figure 1B). While
“ME” is not a substrate for MetAP, it is known to promote
N-α-acetylation of the N-terminal methionine by NatB [6]. Likewise, the P2 serine, once
revealed by MetAP, is predicted to be N-α-acetylated by NatA. We therefore
tested whether acetylation might be the key determinant affecting translocation by
analysing translocation efficiencies in either NatA(Δard1) or
NatB(Δnat3)-deficient strains (Figure 3). In Δard1 cells,
translocation of MS-CPY appeared largely restored while the ME mutant remained
unaffected. The converse was observed in the Δnat3 strain.
Importantly, the ability of the different Nat mutants to rescue
precursor translocation matched precisely the substrate specificities of NatA and
NatB for MS and ME, respectively. Moreover, the observation that inhibition of MAP
activity specifically rescues the translocation of NatA substrates is entirely
consistent with methionine cleavage being a prerequisite for NatA-dependent
acetylation. Thus, it is the N-α-acetylation of these substrates that is the
major determinant in the inhibition of translocation in vivo.
We next examined the effect of mutants predicted to induce acetylation of two
independent ER translocation substrates, namely Pdi1p and prepro-alpha factor
(ppαF) (Figure 4A and 4B).
The signal sequence of Pdi1p begins MK and hence is not predicted to be a substrate
for MetAP or N-acetylation [2]. MSK and MEK mutations both led to accumulation of
non-translocated precursor and a reduction of fully translocated glycosylated Pdi1p
at steady state. Furthermore, analysis by mass-spectrometry confirmed that the MSK
mutant of pPdi1 was methionine-processed and N-acetylated in vivo, as predicted
(Figure
S2). No peptides corresponding to an unmodified N-terminus were
detected.
Wild-type ppαF, which begins MR, is efficiently translocated and secreted. In
contrast an MS mutant, which is a predicted substrate for NatA, accumulated in
cells, as the non-translocated precursor. Hence, the inhibitory effect of
acetylation appears widespread and not restricted to CPY.
Next we sought to reconstitute this phenomenon in vitro using ppαF. We translated
both wild-type (MR) and MS mutant forms of ppαF in reticulocyte lysate and then
incubated these precursors with yeast microsomes (Figures 4C and S3). We
observed microsome-dependent translocation and glycosylation of wild-type ppαF
but found no evidence of translocation of the MS mutant. Thus the inhibitory effect
of the P2 Serine can also be reconstituted in vitro.
Our data thus far indicate that MS-ppαF would be acetylated following processing
by MetAP. To verify this directly we performed in vitro translations in the presence
of 1-[14C]-acetyl-CoA and detected incorporation of radiolabel
into MS-ppαF but not wild-type (Figure 4D). For this experiment, we utilised a ppαF variant where
all lysines have been mutated to arginine; hence, the only primary amine potentially
available for acetylation is the N-terminal αNH2 group. These in
vitro data demonstrate directly that the MS mutant form of ppαF is indeed
acetylated as predicted and support our hypothesis that N-terminal acetylation
inhibits ER translocation.
Charge distribution across the signal sequence has been shown to affect translocation
efficiency [39]. N-α-acetylation of the signal peptide would reduce the
overall positive charge of the N-terminus by +1, and therefore one potentially
trivial explanation might be that it is the loss of positive charge, rather than
acetylation per se, that inhibited translocation. However, we can exclude this
possibility given that the insertion of an additional arginine residue at position 3
(MSRR), which restores the overall charge of the N-region following
N-α-acetylation, also failed to translocate (Figure S3).
We next wished to assess the stage at which the translocation of an acetylated MS
substrate is blocked. We incubated in vitro translated wild-type (MR) ppαF with
yeast microsomes in the absence of ATP, which permits targeting to Sec61, but not
subsequent translocation. Using site-specific photocross-linking probes incorporated
into the signal sequence, we could detect a complex spectrum of uv-induced adducts
as has been reported previously (Figure
4E; [40]). An adduct of ∼50 kD could be readily immunoprecipitated
with Sec61p antisera, indicating the engagement of precursor with the translocon. In
striking contrast, the MS mutant completely failed to crosslink with Sec61p. Hence
we conclude that targeting arrests at a step prior to the interaction of the
precursor with the translocon.
There are two pathways by which secretory precursors can be targeted to the ER; some
precursors follow a post-translational Sec62p-dependent pathway, while substrates
with more hydrophobic signal sequences utilise a co-translational SRP-dependent
mechanism [25],[30]. As CPY, Pdi1p, and ppαF are all translocated
post-translationally, we therefore sought to compare the behaviour of an
SRP-dependent substrate. We chose the well-characterised SRP-dependent substrate
OPY, a variant of CPY in which the endogenous signal sequence is replaced with that
of Ost1p [41]. The
OPY signal sequence begins MR, and so should remain unprocessed, enabling us to
perform a precisely parallel mutational analysis to that for CPY (see Figure 2C). In striking contrast
to CPY, we found that the introduction of various processable residues at P2 had no
effect on the translocation of OPY (Figure 5A and 5B). Thus the observed inhibitory effect of an MS mutation
on translocation can be suppressed in the context of an SRP-dependent signal
sequence. This property was not limited to the Ost1p signal sequence;
co-translational translocation of the SRP-dependent substrate DHC-αF
[30],[42] into yeast
microsomes using a yeast translation extract was also unaffected by the
incorporation of a potentially acetylatable serine residue at P2 (Figure S4).
Moreover, the well-characterized SRP-dependent substrates Sec71 and Dap2 (DPAP B)
[30],[43] have P2
residues of S and E, respectively, entirely consistent with our finding that NAT
substrates can be tolerated by the SRP pathway.
These data suggest either SRP can successfully target an acetylated substrate or
alternatively such substrates might not be processed as expected. Therefore, to
address this point we assessed whether or not the Ost1p signal sequence was
N-terminally processed. We tested the MS mutant for the presence of any unmodified
N-termini using a biotinylation assay to detect free α-NH2 groups in
a protein completely lacking lysine residues. We observed no difference in the
efficiency of biotinylation between wild-type (MR) and mutant (MS) suggesting that
in the context of an SRP-dependent signal sequence, and contrary to expectation, the
MS amino-terminal was not acetylated (Figure 5C).
This effect of SRP might go some way to explain the small, but not insubstantial,
minority of secretory proteins predicted to be processed in our bioinformatic
analysis. Consistent with this idea, we found that average peak hydrophobicity of
signal sequences among this minority was significantly greater than for the majority
subset of sequences (Figure S5). Overall, more than 99% of signal sequences were
either not predicted to be acetylated or were sufficiently hydrophobic to interact
with SRP.
Having validated the biological significance of the bias observed in our
bioinformatic study, we extended our analysis from yeast to higher eukaryotes (Figure 6). The pattern observed in
nematodes and insects was remarkably similar to that seen in yeast, with
∼70% of signal peptides predicted to retain an unprocessed methionine
compared to only 20% for the proteome as a whole [2]. The trend was similar in
humans and plants, albeit less pronounced, with ∼50% of secretory
N-termini predicted to remain unprocessed compared to 15% for the proteome as
a whole [2].
Thus this phenomenon appears not to be restricted to fungi but is very widely
conserved.
Here we describe the striking observation that yeast signal sequences display a
profound bias against N-terminal processing. The bias is precisely converse to that
observed in cytosolic proteins where N-terminal processing is highly favoured.
Moreover, we show that this bias is of functional significance as introduction of
residues at position 2 which promote N-terminal processing inhibits translocation in
to the ER. Importantly this inhibition can be reversed by blocking N-terminal
processing, confirming that it is the processing itself that leads to the block in
translocation. The bias against N-terminal processing is not restricted to yeast but
is also observed across eukaryotes, suggesting this is a widely conserved
phenomenon.
It is possible that other factors distinct from N-terminal processing might affect
the observed bias in amino acid frequency at position 2. We considered the potential
effect of the Kozak consensus sequence that favours a G at the +4 position
(corresponding to the first base of codon 2) in genes optimised for translation
efficiency [44].
However, while this might contribute to the bias observed among cytosolic proteins,
it is unlikely to be the dominant feature since it does not explain the predominance
of Serine at position 2. Furthermore, the Kozak consensus does not have such a
strong effect in yeast and it has recently been reported that the effect of the
+4 position may be more important in promoting N-terminal modification than in
influencing initiation efficiency [45].
A second possible factor influencing the P2 frequency distribution could be the
previously reported bias for an adenine-free stretch within the signal-sequence
coding region of a secretory mRNA, which is important for its nuclear export [46]. However, this
also seems an unlikely explanation as lysine, with its A-rich codon (AAA/AAG), is
actually more frequent at position 2 of signal sequences as compared to cytosolic
proteins. Critically, however, both translation initiation and nuclear mRNA export
operate independently of N-terminal processing and so would not lead to
translocation defects that could be reversed by N-terminal processing mutants, as we
observe. Furthermore the restoration of translocation in such processing mutants
shows a precise substrate dependency, ruling out rescue of translocation by some
indirect effect. Hence, while we would not completely exclude a minor role for the
Kozak consensus or mRNA elements in influencing the P2 residue of signal sequences,
the strong correlation and clear functional effects make a bias against N-terminal
processing the simplest and most likely explanation of the relative P2 residue
frequency.
A trivial explanation for the inhibitory effect of acetylation could be the change in
charge distribution across the signal sequence, which is known to be important for
targeting [39].
However, this appears unlikely, firstly as insertion of an additional positively
charged residue to counteract the loss of the +1 charge following acetylation
of the free amino terminus did not restore translocation (Figure S3).
Secondly, translocation of the ME CPY mutant can be restored in a strain lacking
NatB activity (Δnat3), which results in the same net N-terminal
charge as is present in the acetylated MS CPY mutant, which fails to translocate
(Figure 3). Hence simple
charge distribution alone cannot explain the inhibitory effects of
N-acetylation.
Overall our data indicate that N-acetylation inhibits ER translocation and that most
secretory proteins avoid this by virtue of a P2 residue that prevents processing.
Interestingly, SRP-dependent substrates appear to evade this effect as SRP blocks
N-terminal N-acetylation even in the presence of a P2 residue predicted to be a NatA
substrate. SRP and NatA are both thought to contact the ribosome via the same site
(ribosomal protein Rpl25/L23a) [47]–[49]. Hence competition for this site would provide a
potential mechanistic explanation for this phenomenon. This finding also predicts
that while the P2 residue is the major determinant of N-acetylation by NatA, there
are scenarios where N-acetylation does not occur, despite the presence of an
appropriate P2 residue. Empirical evidence for this prediction was recently provided
by the global analysis of N-acetylation of the drosophila proteome [50].
Comparison of predicted N-terminal processing of signal sequences across other
species indicates an almost identical bias for nematodes and drosophila as seen in
yeast. In plants and humans, the bias is still present but is less marked.
Interestingly, a bias against predicted N-terminal processing (73%) has also
been noted for prokaryotic signal sequences [51]. Hence the bias against
processing of signal sequences appears widespread and not restricted to yeast.
Current dogma suggests that the SRP-dependent targeting pathway is more pervasive in
mammals. As SRP appears to allow substrates to evade the effects of acetylation,
this may well explain why the bias against N-terminal processing is less pronounced
in humans. Nevertheless, homologues of the SRP-independent pathway components Sec62
and Sec63 are present in mammals and form complexes with the Sec61 translocon [35],[36]. Furthermore,
both mammalian and drosophila Sec62 can functionally replace their yeast counterpart
[52],[53]. These
observations, combined with our observed bias against N-terminal processing in these
organisms, suggest that although SRP-dependent targeting is perhaps more dominant,
Sec62-dependent translocation still likely occurs. Identification of substrates for
this pathway remains an important question to be addressed in the future.
What might be the reason as to why secretory and cytosolic proteins have a precisely
converse bias for N-acetylation? Cytosolic proteins, once synthesized, typically
fold rapidly to their final tertiary structure in the cytoplasm. In contrast,
secretory precursors must reach the translocon in an unfolded state in order to be
competent for translocation. Post-translationally translocated substrates achieve
this by their interactions with cytosolic chaperones that prevent their folding
within the cytoplasm [32]. SRP-dependent substrates are targeted co-translationally
and so reach the translocon as short nascent chains, thus eliminating the
possibility of folding in the cytoplasm. It is not known what causes translocation
substrates to recruit these chaperones, but our data allow us to propose a model in
which acetylation determines the fate of nascent polypeptides. We speculate that
acetylation identifies nascent polypeptides, very early in their synthesis, as being
destined to fold in the cytoplasmic compartment. Most secretory proteins are
unmodified and so would be delayed in their folding sufficiently to facilitate their
functional interaction with the translocon. This would be entirely consistent with
our finding that acetylation blocks secretory substrate interaction with Sec61,
arresting the protein in the cytosol.
Not all proteins that fold and remain in the cytosol are acetylated. It may be that
such modification would be incompatible with function, but it might also be that
such proteins have more complex folding requirements; for example, they might be
required to fold more slowly, perhaps relying on the recruitment of specific
cytosolic chaperones.
An alternative biological explanation for this phenomenon could relate to a
proofreading step for Sec62-dependent substrates. Unlike their SRP-dependent
counterparts, Sec62-dependent signal sequences are only modestly hydrophobic [30]. It is quite
likely, therefore, that globular cytosolic proteins may contain internal regions of
similar hydrophobicity, which upon folding form the hydrophobic core of such
proteins. Clearly, it is critical that these proteins do not translocate into the ER
and become mis-sorted. Entirely consistent with this idea, it has been shown that
randomly selected regions of the mature domains of both CPY and invertase (Suc2) can
promote translocation, albeit inefficiently, when positioned at the N-terminus [54],[55].
A requirement for a free N-terminus proximal to the hydrophobic region could provide
a mechanism to prevent internal regions of non-secretory proteins engaging the
translocation machinery. Modification of the N-termini of cytosolic proteins would
also help prevent mis-sorting.
Internal ER targeting sequences of course exist, but they tend to be trans-membrane
domains which act as signal anchor sequences; hence they are much more hydrophobic
and thus promote targeting via the SRP pathway [30].
In summary, our finding that N-terminal processing inhibits ER translocation of
secretory proteins identifies a non-acetylated N-terminus as a hitherto
unappreciated yet general feature of signal sequences, which is necessary to promote
efficient targeting of substrates to the ER translocon.
The set of S. cerevisiae signal sequence-containing proteins was
obtained from the signal peptide database (SPdb) v 5.1 [56]. This set of 291 sequences
was manually filtered for duplicates, dubious ORFs (as defined by SGD), and
proteins known to be localized to mitochondria, to yield a final filtered set of
277 ORFs. For a complete list of ORFs, see Table S1.
The P2 amino acid frequency distribution did not differ significantly between
the filtered and unfiltered sets
(χ2 = 5.17, 19 df).
Graphical and statistical analysis was performed using Prism 4.0 (GraphPad).
MetAP cleavage was assumed for P2 residues A, C, G, P, S, V, and T [11],[12]. The yeast
cytosolic dataset (Table S2) was generated by random selection
from SGD of proteins with known cytosolic localization. Prediction of
N-acetylation was performed as described previously [2]; where appropriate, the P3
residue was also taken into consideration. MN, which is only predicted to lead
to N-acetylation in 55% of cases [2], was scored as acetylated.
Human and Caenorhabditis elegans signal sequence datasets were
also obtained from the signal peptide database (SPdb) v5.1 [56]. Drosophila
melanogaster and Arabidopsis thaliana datasets
were obtained from the signal peptide website (www.signalpeptide.de,
accessed March 2010). Peak hydrophobicity was determined by Kyte-Doolittle using
a window size of 11 [30],[57].
Yeast strains in this study are listed in Table S7. GFY3 was constructed by mating
Δpep4 and Δprc1 strains,
sporulation of the diploid, and selection of tetrads, which had three
G418-resistant spores; spores were scored for null mutations by PCR and western
blotting. GFY7 was made by PCR amplification of the pFA6a-His3MX6 module [58] with
appropriate primers (Table S8); the PCR product was used to
transform GFY3 and His+ colonies selected. GFY11 and GFY12 were
made by PCR amplification of pAG26 [59] with appropriate primers
(Table
S8); the PCR products were used to transform
Δprc1 followed by selection on Hygromycin B. All
deletions were confirmed by PCR. Yeast strains were grown in either YPD
(1% yeast extract, 2% peptone, and 2% glucose) or YNB
(0.67% yeast nitrogen base, 2% glucose, and appropriate
supplements) at 30°C, with the exception of pulse-labelling of MWY63
(sec61-3), which was grown at 30°C, then shifted to
17°C for 2 h.
The constructs which express ppCPY and ppOPY with position 2 insertion mutations
of the signal sequence listed in Table S9 were made using the respective pairs
of primers (Table S8) to perform site-directed mutagenesis of pMW346 or pOPY,
respectively. pGF22, the PsiI/SphI fragment of
pA11-k5, was cloned into pEH3 to replace this portion of wild-type ppαF and
thus making a lysine-free ppαF. pGF24 and pGF25 were constructed by PCR
(Table
S9) of the Ost1 signal sequence from pOPY and pOPY-S, respectively.
The PCR products were digested with
EcoRI/HincII and cloned into pGF23 (Table S9)
to replace the ppαF signal sequence with that of Ost1 or the serine mutant
version, respectively. PDI1 was amplified from genomic DNA with
appropriate primers (Table S8) that introduce a single C-terminal
c-myc-tag. The PCR products were digested with
PsiI/BamHI and were then ligated into
BstZ171/BamHI sites of pMW346, placing the
PDI1-myc ORF under the control of the PRC1
promotor. pPPαF-2myc constructs were generated in a similar manner except
that they contain two c-myc-tags and the PCR products generated were digested
with BstZ171/BamHI.
Yeast cells expressing wild-type CPY or signal sequence mutants (Table S7)
were grown in YNB medium with appropriate supplements to an
OD600nm = 0.2, where stated cells were treated
with 3 µM Fumagillin (Fluorochem) for 30 min at 30°C prior to
radio-labelling. Pulse-labelling was initiated by addition of 10 µCi of
[35S] Methionine/Cysteine mix (Perkin Elmer) per
OD600nm units of cells for 5 min at 30°C (20 min at 17°C
for sec61-3). Labelling was terminated by addition of ice cold
sodium azide to a final concentration of 20 mM. For each sample 5 or 10
OD600 units of cells were harvested.
Radiolabelled yeast cells were spheroplasted prior to addition of lysis buffer
(1% SDS, 50 mM Tris-HCl, pH 7.4, and 5 mM EDTA) and then incubated at
95°C. Samples were then diluted with 5 volumes of immuno-precipitation
buffer (62.5 mM Tris-HCl, pH 7.4, 1.25% (v/v) Triton-X-100, 190 mM NaCl,
6.25 mM EDTA), pre-cleared for 1 h, and then antiserum (anti-CPY or anti-αF
[60],[61]) added to
the supernatant. After 1 h, immune complexes were recovered with Protein A
sepharose for a further hour and then washed extensively prior to elution with
SDS-PAGE sample buffer. Samples were then analysed by SDS-PAGE and visualised
either by phosphorimaging or autoradiography. Quantification was performed with
Aida image-analyzer software (Raytek). Subsequent statistical analysis was
performed using Prism 4.0 (GraphPad). Samples for scintillation counting were
dissociated from the sepharose with 3% SDS for 5 min at 95°C.
Dissociated protein was dried onto Whatman glass GF/A filter discs and placed in
4.5 mL of scintillant and counted in a Tricarb 2100TR liquid scintillation
counter (Packard).
Templates for transcription of various ppαF mRNAs were generated by PCR from
plasmids pEH3 or pGF22 using appropriate primers (Table S8)
and transcription carried out with SP6 polymerase. Transcriptions of OpαF
mRNAs were from pGF24 or pGF25 for MR and MS OpαF, respectively, and were
carried out with T7 polymerase. Translations were performed in rabbit
reticulocyte lysate system (Promega) for 30 min with the inclusion of either
2.04 µCi [35S] Methionine or 0.04 µCi
1-[14C]-Acetyl Coenzyme A (Perkin Elmer) per 10
µL of reaction. Translation was terminated by addition of 2 mM
cycloheximide.
Co-translational translocation of DHC-αF into yeast microsomes was
performed using translation extracts from a strain over-expressing SRP, as
described previously [42].
Preparation of yeast microsomes from a Δpep4 strain was
carried out as previously described [62]. For translocation
assays; 10 µL of translation reaction was incubated with 2 µL
microsomes for 20 min at 30°C.
Wild-type and MS K5K14ppαF were translated in rabbit reticulocyte lysate as
above but in the presence of ε-4-(3-trifluoro-methyldiazirino) benzoic acid
(TDBA)-lysyl-tRNA and then used for photocross-linking assays as described [63].
Briefly, translations were terminated with 2 mM puromycin for 10 min at
30°C, and then treated with 0.5 mg/mL RNase A for 5 min on ice prior to
depletion of ATP from the translation reaction and yeast microsomes by treatment
with hexokinase/glucose. The microsomes and translation reaction were then
combined, allowing targeting to occur for 15 min at 30°C. Microsomes were
re-isolated by centrifugation and resuspended in membrane storage buffer.
Samples were irradiated with uv light (365 nm, 15 mW/cm2) twice for 5
s and then precipitated with ethanol and analysed directly or following
denaturing immuno-precipitation with Sec61 antiserum [64].
In vitro translations (20 µL scale), programmed with lysine-free OpαF
mRNAs, were performed as above in the presence of [35S]
methionine. Proteins were sequentially precipitated with ammonium sulphate, then
ethanol. The samples were then denatured in PBS+1% SDS for 10 min at
65°C. Free N-termini were modified by treatment with 1 mM
sulpho-NHS-SS-Biotin (Pierce) for 20 min at 37°C. After removal of free
biotinylation reagent by acetone precipitation, samples were resuspended in
PBS+0.1% SDS and then biotinylated proteins recovered on
immobilized-streptavidin beads (Pierce). Beads were washed 5 times with
PBS+0.1% SDS and bound protein eluted in SDS-PAGE sample buffer.
|
10.1371/journal.pgen.1008075 | Genomic inversions and GOLGA core duplicons underlie disease instability at the 15q25 locus | Human chromosome 15q25 is involved in several disease-associated structural rearrangements, including microdeletions and chromosomal markers with inverted duplications. Using comparative fluorescence in situ hybridization, strand-sequencing, single-molecule, real-time sequencing and Bionano optical mapping analyses, we investigated the organization of the 15q25 region in human and nonhuman primates. We found that two independent inversions occurred in this region after the fission event that gave rise to phylogenetic chromosomes XIV and XV in humans and great apes. One of these inversions is still polymorphic in the human population today and may confer differential susceptibility to 15q25 microdeletions and inverted duplications. The inversion breakpoints map within segmental duplications containing core duplicons of the GOLGA gene family and correspond to the site of an ancestral centromere, which became inactivated about 25 million years ago. The inactivation of this centromere likely released segmental duplications from recombination repression typical of centromeric regions. We hypothesize that this increased the frequency of ectopic recombination creating a hotspot of hominid inversions where dispersed GOLGA core elements now predispose this region to recurrent genomic rearrangements associated with disease.
| Human chromosome 15 derived from a fission event that occurred in the ancestor of great apes. Following inactivation of the ancestral centromere at 15q25 a dispersal of segmental duplications took place, providing templates for ectopic recombination and predisposing the region to genomic instability. Different disease-associated microdeletions and chromosomal markers have been described with breakpoints mapping within these segmental duplications. To gain insight into the instability at 15q25, we sought to analyze this region in human and nonhuman primates using multiple genomics techniques and demonstrated the presence of two independent inversion events that occurred during great apes evolution. One of these inversions is still polymorphic in humans and may cause, in conjunction with a GOLGA core duplicon—a ~14 kbp chromosome 15 repeat, susceptibility to non-allelic homologous recombination leading to pathogenic recurrent rearrangements. Our results support the existence of a strong relationship between inversions and core duplicons and reinforce the hypothesis that GOLGA repeats play a fundamental role both in disease and evolution.
| Human chromosome 15 was generated by the chromosome fission of an ancestral submetacentric chromosome in the ancestor of great apes. Macaque chromosome 7 represents the ancestral state, as more distantly related organisms have the same configuration [1]. The ancestral centromere at 15q25 inactivated and lost any centromeric satellites, whereas a large cluster of segmental duplications persisted [2]. Following inactivation of the ancestral centromere, the constraint against recombination in this area was likely weakened [3–5], providing an environment permissive to non-allelic rearrangements that promoted the dispersal of the segmental duplications.
The 15q25 locus approximately corresponds to the position of the ancestral centromere [2] and is an unstable region of the human genome enriched in segmental duplications containing the GOLGA core duplicon, a ~14 kbp chromosome 15 repeat [6]. Cores represent ancestral duplications where additional duplication blocks have been formed around, and correspond to the expansion of gene families, some of which show signatures of positive selection [7]. GOLGA belongs to the golgin subfamily of coiled-coil proteins associated with the Golgi apparatus. These genes appear to have roles in membrane traffic and Golgi structure, but their precise function is in most cases unclear. GOLGA encodes a primate-specific gene family that expanded over the last 20 million years [8, 9]. Human chromosome 15 contains nearly 40 copies of the GOLGA core element [6], dispersed to multiple locations across the long arm of chromosome 15. GOLGA is one of 14 core duplicons associated with the burst of interspersed segmental duplications in the human–great ape ancestral lineage [10, 11] and the most enriched sequence associated with segmental blocks promoting evolutionary rearrangements in primates [12–14] and disease instability, including Prader-Willi/Angelman syndromes, 15q13 microdeletions and 15q24 microdeletions [13, 15–18].
The 15q25 region represents a high-risk locus for pediatric neurologic disease with variable outcomes [2, 19–29]. Different microdeletions and chromosomal markers with inverted duplications of chromosome 15 all have breakpoints mapping within a 3.3 Mbp region containing three blocks of segmental duplications of 350 kbp, 560 kbp and 115 kbp in size. The middle block contains a gap in the last release of the human reference genome (GRCh38/hg38) suggesting the possible presence of different structural haplotypes in this locus. We characterized the organization of this region in human and nonhuman primate genomes by conducting a detailed analysis by fluorescence in situ hybridization (FISH), single-cell strand-sequencing (Strand-seq), high-quality finished sequencing using PacBio single-molecule, real-time (SMRT) sequencing technology, and optical mapping (Bionano) in order to understand the extent of human genetic variation, its origin, and impact on disease.
Three duplication blocks, containing GOLGA repeats, map at the 15q25 region (S1 Table), with the middle one (block B) containing a gap in the reference genome (Fig 1A), suggesting that alternative structural configurations might exist within the human population. To gain insight into the instability associated to disease and evolutionary rearrangements at 15q25 we sought to characterize the organization of this region in more detail in human and nonhuman primate genomes. Using FISH experiments in interphase nuclei, we tested 22 HapMap individuals from different populations for the presence of two putative inversions: the proximal inversion of 1.5 Mbp between duplication blocks A and B (chr15:82534139–84045983) and the distal inversion of 600 kbp between duplication blocks B and C (chr15:84596420–85169772). All individuals tested for both distal and proximal inversions were in direct orientation (Fig 1B; S2 Table).
In order to gain more information regarding the presence and frequency of the 15q25 inversions, we investigated published Strand-seq data from 47 libraries from a pool of 353 separate cord blood and bone marrow donors [30]. In Strand-seq libraries, inversions cause a segmental change in strand orientation at the inverted locus, which allows inversions to be directly visualized and genotyped in single-cell data [30]. One out of 22 informative libraries was heterozygous for the proximal inversion, while all the others were in direct orientation. All 22 libraries were in direct orientation for the distal region (Fig 2). In total, we tested 88 chromosomes (44 chromosomes by FISH analyses and 44 chromosomes by Strand-seq) and showed that only one out of the 88 chromosomes was in inverted orientation for the proximal region (inversion allele frequency of 1.14%) (Table 1).
In order to close the gap in the reference genome we generated a map of contiguous clones from the CH17 BAC library from a hydatidiform mole-derived (haploid) human cell line (CHM1hTERT) [31]. Using BAC-end sequence pair mapping, we constructed a contiguous set of four BAC clones (S3 Table) and then performed SMRT sequencing. We generated a 657 kbp sequence contig spanning the B block of segmental duplications and closed the gap in this region. Miropeats analysis of the CH17 contig versus the human hg38 reference showed that the hydatidiform mole is in direct orientation for the proximal inversion and highlighted the presence of 64 kbp redundant sequence, containing GOLGA repeats, that was represented twice within the reference (S1 Fig).
In order to investigate the ancestral configuration of the 15q25 region, we compared the orientation of the proximal and distal regions in human with other nonhuman primate species. We tested for the presence of the proximal inversion between duplication blocks A and B by FISH analysis of cell lines from eight chimpanzees (Pan troglodytes), four gorillas (Gorilla gorilla), four orangutans (Pongo pygmaeus), and one macaque (Macaca mulatta) (Fig 1B; S2 Table). Moreover, we analyzed Bionano optical mapping data of DNA from one chimpanzee (Pan troglodytes), one gorilla (Gorilla gorilla) and one orangutan (Pongo abelii) (Fig 3; S2 Table). Chimpanzee, orangutan and macaque were found to be inverted when compared to the human reference genome orientation suggesting that this represents the likely ancestral state, while all gorillas were in direct orientation, similar to humans. We conclude that the proximal inversion likely occurred in the human–African great ape ancestor and the chimpanzee configuration may represent incomplete lineage sorting of the ancestral state or the inversion may have occurred at multiple times during great ape evolution as a result of recurrent mutation events involving the duplicated sequences.
Next, we tested for the presence of the distal inversion between duplication blocks B and C by FISH analysis of eight chimpanzees (Pan troglodytes), four gorillas (Gorilla gorilla), four orangutans (Pongo pygmaeus), and one macaque (Macaca mulatta). We found this inversion to be widely polymorphic within the chimpanzee population, while all other nonhuman species were inverted in the homozygous state (Fig 1B; S2 Table). Bionano optical mapping data of DNA from one gorilla (Gorilla gorilla) and one orangutan (Pongo abelii) show that these individuals are inverted in the homozygous state while chimpanzee (Pan troglodytes) is in direct orientation for both haplotypes (Fig 3; S2 Table). These data suggest that the inversion occurred in the human–chimpanzee ancestor and is still polymorphic in chimpanzee with a 39% allele frequency (Table 1).
Given the central role of the duplications in both microdeletions and the evolution of inversions [13, 32–38], we compared the duplication architecture among primate species. Using BAC-end sequence pair mapping, we selected three clones from the CH276 orangutan BAC library and one clone from the CH251 chimpanzee library, which spanned the 600 kbp distal inversion breakpoint between blocks B and C, and then sequenced them using PacBio SMRT sequencing (S3 Table; S2 Fig). In orangutan we generated a ~400 kbp sequence contig and compared this with the human reference assembly. In addition, to confirm the presence of the inversion between duplication blocks B and C, we identified that the inversion would create an ancestral B/C block-hybrid. Sequence analysis demonstrates that this block is missing one copy of the GOLGA core duplicon identified in the human B block and two copies in the human C block (S2A Fig). We queried GenBank and identified two additional clones (250 kbp sequence contig) from CH250 macaque BAC library spanning this ancestral B/C block. Comparison of the B/C block-hybrid between orangutan and macaque shows that they both have two copies of GOLGA repeats in this region and, therefore, are missing a total of three copies with respect to the human orthologous regions (S2B Fig). The ancestral orientation of the duplication block is itself inverted in the macaque relative to orangutan (S2B Fig), suggesting a restructuring of the duplication architecture of the region during primate evolution. Finally, Miropeats analyses of a CH251 clone from a chimpanzee (Clint) in homozygous direct orientation for the distal inversion between B/C blocks shows that human and chimpanzee are collinear and both have two copies of GOLGA repeats for this region (S2C Fig).
To further investigate the copy number of the GOLGA core duplicons in humans and primates, we performed a BLAT analysis using GOLGA2P10 and GOLGA6L5P exon sequences. We generated a map of GOLGA repeats in the 15q25 human region, which allowed us to identify 24 repeats in the latest human reference genome assembly (S1 Table; S3 Fig). We performed the same analysis on primate reference genomes (Clint_PTRv2/panTro6, gorGor4.1/gorGor4, Susie_PABv2/ponAbe3 and BCM Mmul_8.0.1/rheMac8) and found 11 repeats in chimpanzee and gorilla and 13 in orangutan and macaque (S1 Table). However, the exact number of copies could not be determined due to the presence of gaps in the assembly region (3 gaps in chimpanzee, 67 in gorilla, 1 in orangutan and 20 in macaque). To determine the extent to which gaps affected the observed difference in copy number we performed a parallel analysis using four assemblies all built using the same PacBio sequencing technology and FALCON assembly method (chimpanzee Clint_PTRv2/panTro6, gorilla Susie3, orangutan Susie_PABv2/ponAbe3, and human PacBioCHM1_r2). Using whole genome alignments of these four assemblies to GRCh38 we found that of the 24 GOLGA repeats in the 15q25 region, 21 are found in the human de novo assembly (CHM1) which is twice the number of copies compared to other primate species (9 in chimpanzee, 10 in gorilla and 8 in orangutan) (S4 Table). The high recovery rate of GOLGA repeats in CHM1 as compared to GRCh38 (21/24) suggests that the large majority of these sequences are resolved in de novo PacBio assemblies and that the reduced copy number in Clint_PTRv2/panTro6, Susie3, and Susie_PABv2/ponAbe3 is not due to gaps in their respective assemblies.
Three different classes of 15q25 microdeletions associated with developmental delay and intellectual disability have been described [19–24]. The microdeletions can be classified based on recurrent breakpoints, with breakpoints between duplication blocks A and B (A-B deletions), B and C (B-C deletions) and A and C (A-C deletions). The three blocks of segmental duplications that mediate the rearrangements are 350 kbp, 560 kbp and 115 kbp in size. Each block contains at least one copy of the GOLGA core duplicon (Fig 4; S1 Table).
Using single-nucleotide polymorphism (SNP) microarrays, we analyzed a patient with global developmental delay harboring a 15q25 deletion and mapped the disease-critical region to a 1.6 Mbp region spanned by segmental duplication blocks A and B (Fig 4). To refine the breakpoints with greater precision, we performed whole-genome sequencing (WGS) of the 15q25 deletion sample using Illumina HiSeq X Ten (150 bp PE reads; 26.2X coverage for whole chromosome 15) and aligned the sequences to the human reference. Using singly unique nucleotide (SUN) variants that allowed us to discriminate between the paralogous copies [39], we narrowed the deletion breakpoints to a 20 kbp (chr15:82478935–82498934) segment within duplication block A and a 101 kbp (chr15:84404774–84506312) segment within block B. Blast2seq analysis of the sequence mapping within the breakpoints intervals shows that the longest alignments with 99% similarity correspond to two sequences of 9.4 kbp (blast2SeqA) and 9 kbp (blast2SeqB), mapping at GOLGA2P10 sequences, in inverted orientation (Fig 4; S1 File).
Centromeres and their neighboring pericentromeric chromatin are well-established cold spots of meiotic crossover activity [3–5]. We hypothesized that inactivation of the ancestral centromere at 15q25 reduced the strength of recombination suppression at this locus, leading to higher rates of rearrangement and genome instability. Three lines of evidence support this interpretation. First, there are active meiotic recombination hotspots within the 15q25 region of the human genome [40]. Second, GOLGA repeats within the 15q25 region engage in frequent bouts of interlocus gene conversion [41]. Both of these observations point to homology-driven repair activity within the 15q25 region. Third, broad-scale recombination rates over the 15q25 region are elevated relative to the recombination levels of active centromeres and pericentromeric regions in the human genome. We observe a similar relaxation of recombination rate suppression in the inactivated 15q25 ancestral centromere region in bonobo, chimpanzee, and gorilla. We caution that recombination map quality is likely reduced across the structurally complex 15q25 locus, and the absence of informative centromeric markers precludes estimates of recombination rate within these gapped regions on the assembly. However, these qualitative findings suggest that the repositioning of the chromosome 15 centromere in the common ancestor of great apes weakened the recombination-suppressive environment that defines centromeres and set the stage for recurrent homology-driven rearrangements at the human 15q25 locus.
In this study, we sought to better understand the mechanisms leading to the genomic instability of the 15q25 locus by characterizing evolutionary and contemporary rearrangements by FISH, single-cell Strand-seq, PacBio SMRT sequencing, and Bionano optical mapping analyses. Chromosome 15q25 harbors a complex genomic region with three large blocks of segmental duplications containing several copies of the GOLGA core duplicon, previously shown to be involved in several recurrent pathogenic and evolutionary rearrangements on chromosome 15 [2, 12–14]. The middle block of segmental duplications contains a gap in the current release of the human reference assembly. Using SMRT technology, we sequenced and de novo assembled a tiling path of four BAC clones (657 kbp region) across this medically relevant region from the library of a hydatidiform mole and showed that the gap was flanked by two identical copies of GOLGA core duplicons that might have confounded mapping and assembly of the region. Alternatively, this might be a biological difference in structural haplotypes and/or copy number of GOLGA between human individuals.
Microarray analysis of 15q25 microdeletions in previous [19–24] and current studies refined breakpoint locations to segmental duplication blocks A, B and C; however, probe cross-hybridization prevented further narrowing of breakpoint locations within the duplications. All three segmental duplication clusters have multiple regions of high sequence identity—the most significant in direct orientation has 99% identity across 59 kbp. Here we performed WGS of a 15q25 microdeletion between duplication blocks A and B. SUN variant mapping allowed us to differentiate the segmental duplication paralogous copies [39] and revealed that the breakpoints map precisely to the GOLGA core duplicon sequences, which are organized in a palindromic configuration. Previous studies of many different chromosome 15 rearrangements, including several recurrent microdeletion/duplication syndromes and more complex rearrangements such as inverted duplications and triplications of chromosome 15 [13, 34], have shown that the breakpoints of all of these appear to coincide precisely with the location of the duplication family containing the GOLGA gene. An example is the 2 Mbp microdeletion encompassing the 15q13.3 region associated with intellectual disability, schizophrenia, autism and epilepsy [13]. WGS of two idiopathic autism patients carrying de novo 15q13.3 microdeletions showed that the two probands have different breakpoints but in both cases they map to GOLGA sequences [13].
Taken together these results suggest that despite the presence of large segmental duplications in direct orientation, known to be a predisposing factor for non-allelic homologous recombination (NAHR) leading to deletion/duplication events, palindromic GOLGA repeats seem to be preferential sites for NAHR promoting disease-related instability of chromosome 15. The presence of GOLGA core duplicons at multiple disease-associated rearrangements [13, 15–18] and evolutionary breakpoints [12–14] indicate the high level of genomic instability driven by these sequences.
The chromosome 15q25 locus approximates the position of the ancestral centromere, which became inactivated about 25 million years ago [2]. This inactivation followed a noncentromeric chromosomal fission of an ancestral chromosome that gave rise to human and great apes chromosomes 14 and 15 [2]. The duplications flanking the ancestral centromere were formed within a pericentromeric context where recombination was almost absent [3–5]. Our recombination and evolutionary analyses support the hypothesis that following inactivation of the ancestral centromere, the constraint against recombination in this area was relaxed, rendering the locus permissive to NAHR-mediated rearrangements. We speculate that such events ultimately led to two local inversions specific to humans and African great apes. These inversions may have ultimately helped to disperse the GOLGA core elements that are now mediating pathogenic microdeletions (Fig 5). To test this hypothesis, we performed a BLAT analysis on the latest releases of the chimpanzee, gorilla, orangutan and macaque genome assemblies and found that human has nine copies of the core duplicon in excess compared to macaque.
The distal inversion is polymorphic within the chimpanzee population (39% allele frequency), while all the other species are configured in the opposite orientation compared to human suggesting that the inversion occurred in the human–chimpanzee ancestor and is highly polymorphic in chimpanzee. These findings are strikingly reminiscent of the 17q21.31 chromosome inversion [42], which is flanked by LRRC37 core duplications on either side of the inversion region and is highly polymorphic in multiple species, especially in chimpanzee (56% allele frequency).
The proximal inversion is 1.5 Mbp in size and likely occurred in the human–African great ape ancestor. The orthologous region in chimpanzee, however, is in the opposite orientation as that of human, suggesting that the region either flipped back to the ancestral orientation in the chimpanzee lineage or the chimpanzee configuration may represent incomplete lineage sorting of an ancestral state. FISH and Strand-seq analyses of 44 human samples show that this inversion is still polymorphic in humans, with a minor allele frequency of 1.14%. The inversion corresponds to the exact same region that is deleted in patients with intellectual disability, suggesting that the inversion may represent a premutation state to pathogenic rearrangements, as previously observed for other regions of the genome, such as the 17q21.31, 7q11.23 and 8p23.1 loci [42–44]. The inversion could change the orientation and composition of GOLGA repeats (depending on breakpoints), and this could impact the likelihood of a recombination event generating a deletion.
Interestingly, chromosome 15q25 is also one of the hotspots of neocentromere appearance in clinical cases [45]. The majority of neocentromeres reported in clinical samples rescue acentric chromosome fragments associated with duplications or chromosomal rearrangements found in patients with developmental disabilities [46]. From a literature review we identified several recurrent rearrangements, consisting of tetrasomies from 15q25→qter due to analphoid chromosomal markers with a neocentromere at 15q25 (S5A Fig) [2, 25–29, 46–63]. Ventura and colleagues performed detailed mapping of the ancestral centromere as well as a neocentromere mapping to chromosome 15q25 and established that the ancestral centromere and the neocentromere map to two different clusters of segmental duplications separated by a 1.5 Mbp single-copy region [2]. Notably, we show that this region corresponds exactly to the 1.5 Mbp proximal inversion that occurred in the human–African great ape ancestor after inactivation of the ancestral centromere. Following inactivation, an increased frequency of ectopic rearrangements at 15q25 might have resulted in evolutionary inversions that led to a duplicative transposition of GOLGA core elements from one breakpoint to the other (from segmental duplication block A to B).
The existence of human individuals heterozygous for the proximal inversion led us to hypothesize that inversions may be a driving force in the formation of 15q25 neocentric invdup marker chromosomes. FISH mapping of the breakpoints of invdup markers from both 8p and 15q suggests that between the two duplicated symmetrical arms is found an unduplicated region containing proximal sequences contiguous with one of the arms (S5B Fig) [2, 44]. Such an arrangement is consistent with meiotic recombination between chromosomes that are heterozygous for a polymorphic inversion flanked by inverted segmental duplications. During cell division, asynapsis at the inverted region may promote the refolding of one chromosome onto itself, allowing intrachromatid synapsis and NAHR between two GOLGA repeats (S5C Fig). This would favor the formation of the 15q25→qter inverted-duplicated chromosomal markers, similar to what has been previously shown for the 8p23.1 inversion polymorphism [44]. In conclusion, our findings highlight the intimate relationship between inversions and core duplicons and reinforce the hypothesis that GOLGA repeats played a fundamental role in shaping the architecture of chromosome 15 in humans and great apes and continue to predispose it to disease-causing rearrangements.
In conclusion, we propose the following model. Core duplicons in 15q25 were formed within a pericentromeric context, and sequences in the same region have continued to undergo sequence exchange/duplication within the human lineage long after the centromere became inactivated about 25 million years ago. This had significant implications for genomic stability in the region. It is highly likely that following inactivation of the ancestral centromere, the constraint against recombination in this area was relaxed. This would have increased the frequency of ectopic rearrangements, accelerating the dispersal of the linked GOLGA duplicons and leading to two local inversions specific to humans and African great apes. Our findings highlight the intimate relationship between inversions and core duplicons and reinforce the hypothesis that GOLGA repeats played a fundamental role in shaping the architecture of chromosome 15 in humans and great apes and continue to predispose it to disease-causing rearrangements.
The study was approved (Prot. 117CE;6/11/2017) by the local Ethics committee of the IRCCS "Casa Sollievo della Sofferenza" Hospital, San Giovanni Rotondo (FG). Written, informed consent was received.
Interphase nuclei were obtained from lymphoblast and fibroblast cell lines from 22 human HapMap individuals (Coriell Cell Repository, Camden, NJ, USA), eight chimpanzees (Pan troglodytes), four orangutans (Pongo pygmaeus), four gorillas (Gorilla gorilla) and one macaque (Macaca mulatta) (S2 Table). FISH experiments were performed using human fosmid (n = 6) clones (S6 Table) directly labeled by nick-translation with Cy3-dUTP (Perkin-Elmer), Cy5-dUTP (Perkin-Elmer) and fluorescein-dUTP (Enzo) as described by Lichter et al. [64], with minor modifications. Briefly, 300 ng of labeled probe were used for the FISH experiments; hybridization was performed at 37°C in 2xSSC, 50% (v/v) formamide, 10% (w/v) dextran sulphate and 3 mg sonicated salmon sperm DNA, in a volume of 10 mL. Posthybridization washing was at 60°C in 0.1xSSC (three times, high stringency, for hybridizations on human, chimpanzee, gorilla and orangutan) or at 37°C in 2xSSC and 42°C in 2xSSC, 50% formamide (three times each, low stringency, for hybridizations on macaque). Nuclei were simultaneously DAPI stained. Digital images were obtained using a Leica DMRXA2 epifluorescence microscope equipped with a cooled CCD camera (Princeton Instruments). DAPI, Cy3, Cy5 and fluorescein fluorescence signals, detected with specific filters, were recorded separately as grayscale images. Pseudocoloring and merging of images were performed using Adobe Photoshop software. Proximal and distal inversions were interrogated using two probes within the putative inversion region and a reference probe outside, as previously described [37].
SNP array-based copy number variant (CNV) analysis was performed on genomic DNA extracted from peripheral blood lymphocytes of the patient and parents, after obtaining written informed consent, using the CytoScan HD Array (Affymetrix, Santa Clara, CA, USA) as previously described [65]. Data analysis was performed using the Chromosome Analysis Suite software version 3.1 (Affymetrix, Santa Clara, CA, USA). A CNV was validated if at least 25 contiguous probes showed an abnormal log2 ratio. The clinical significance of each CNV detected was assessed by comparison with public databases such as the Database of Genomic Variants (DGV; available online at: http://dgv.tcag.ca/), DECIPHER (https://decipher.sanger.ac.uk/), and ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/). We also checked an internal database of 3,500 patients studied by SNP array in our laboratory since 2010 with a diagnosis of syndromic/non-syndromic neurodevelopmental disorders. Finally, to predict the pathogenic role of the identified microdeletions/microduplications, we followed the American College of Medical Genetics guidelines [66].
Read-depth profiles were generated by extracting the first 36 bp of each read from a BWA-MEM aligned BAM file and aligning these reads to the hg38 genome at all possible positions using mrsFAST-Ultra [67]. Read-depth profiles were converted to copy number estimates at edit distances of 2 and 0 to define total and locus-specific (SUN) copy number estimates at ~93% of confidence, as described in Sudmant et al. 2010 [39]. Coverage for chromosome 15 is 26.2X for the full reads and ~6.3X for the first 36mer. Sequence read-depth corresponding to SUN variants was then used to refine the microdeletion breakpoints as previously described [39].
The inversion status of Strand-seq libraries generated from a pooled cord blood sample comprising 47 unrelated donors (described in detail in: [30]) was assessed at the 15q25 locus. Briefly, Strand-seq sequence data were aligned to GRCh37/hg19, BED-formatted for upload to the UCSC Genome Browser, and analyzed using the open-source ‘Invert.R’ software (https://sourceforge.net/projects/strandseq-invertr/). Only cells that inherited chromosome 15 in the WW (W, Watson; reverse or minus strand) or CC (C, Crick; forward or plus strand) were analyzed (n = 22) to ensure homozygous inversions were fully captured. Libraries were tested for a segmental change in strand orientation at the putative inversion loci (coordinates lifted to GRCh37/hg19; proximal inversion at chr15:83202890–84714735 and distal inversion at chr15:85139055–85713003), and genotypes were confirmed from the Invert.R results [30].
To estimate the copy number of GOLGA repeats in all the tested species we performed a blat analysis. We downloaded GOLGA2P10 and GOLGA6L5P exon sequences from the UCSC Genome Browser and used the blat tool to compare them with the human reference (GRCh38/hg38). We also performed the same analysis for chimpanzee (Clint_PTRv2/panTro6), gorilla (gorGor4.1/gorGor4), orangutan (Susie_PABv2/ponAbe3), and macaque (BCM Mmul_8.0.1/rheMac8).
We then analyzed the copy number of GOLGA in 15q25 locus using PacBio-based assemblies of human CHM1 (https://www.ncbi.nlm.nih.gov/assembly/GCA_001297185.1), chimpanzee (https://www.ncbi.nlm.nih.gov/assembly/GCF_002880755.1), gorilla (https://www.ncbi.nlm.nih.gov/assembly/GCA_900006655.3), and orangutan (https://www.ncbi.nlm.nih.gov/assembly/GCF_002880775.1). Contigs from PacBio assemblies were aligned to the human reference using Mashmap 2.05 with default parameters. Three filtering steps were then applied to the alignments. First, alignments were filtered such that contigs were only mapped to one location in GRCh38. Second, remaining alignments were intersected with the 24 regions found in the BLAT analysis using BEDTools6. Finally, these intersections were filtered to only those with at least 90% overlap with one of the 24 defined GOLGA regions. After these steps, the GOLGA copy number was estimated by counting the number of GOLGA regions that were intersected by each primate assembly.
DNA was isolated from CH17, CH251 and CH276 BAC clones (S3 Table) as previously described [68]. PacBio (Pacific Biosciences, Inc., Menlo Park, CA, USA) SMRTbell libraries were prepared and sequenced using RS II P6-C4 chemistry. We performed de novo assembly of pooled BAC inserts (5–6 BACs per pool) using the Canu assembler [69] followed by consensus calling using Quiver [68]. PacBio assemblies were reviewed for misassembly by sequencing to a minimum coverage depth of 200X and visualizing read depth of PacBio reads in Parasight (http://eichlerlab.gs.washington.edu/jeff/parasight/index.html) using coverage summaries generated during the resequencing protocol [68]. As a final validation, we mapped publically available BAC end sequences to high-quality finished clone inserts to confirm order and orientation. Human, chimpanzee, orangutan and macaque assemblies, including PacBio sequenced clones from CH17, CH251, CH276 and CH250 BAC libraries, were assembled with Sequencher and compared to the human reference genome using Miropeats [70] and BLAST [71]. Duplication analysis using whole-genome shotgun sequence detection (WSSD) was performed as previously described [72].
High molecular weight DNA from one chimpanzee (Clint), one gorilla (Kamilah), and one orangutan (Susie) were used to construct Bionano optical maps (S2 Table). The chimpanzee and orangutan maps were constructed as previously described by Kronenberg et al. [73]. To label the chimpanzee and orangutan genomes, two different enzymes were used to ensure contiguous coverage. The two labelling enzymes used (Nt.BspQI and Nb.BssSI) are nickases, and each enzyme makes single-stranded nicks at specific recognition motifs along the genomes. However, whenever two nick sites in opposite strands are in close proximity, double-stranded breaks would be created in the DNA. Such double-stranded breaks—also known as fragile sites—would create permanent disruptions in the DNA molecules and the assembled contigs. Therefore, to bridge these fragile sites, in a separate labelling experiment, a second enzyme that recognizes a different motif site was used. The gorilla maps were constructed using DLE-1 non-nicking enzyme, which is a newer enzyme that directly labels its recognition sites without creating any nicks. Thus, for that experiment, no fragile sites were created, and we achieved continuous coverage across the genome. Briefly, labelling and staining of the DNA were performed according to a protocol developed by Bionano Genomics. Labelling was performed by incubating 750 ng genomic DNA with 1X DLE-1 Enzyme (Bionano Genomics) for 2 hours at 37°C, followed by 20 minutes at 70°C, in the presence of 1X DL-Green and 1X Direct Labelling Enzyme (DLE-1) Buffer. Following proteinase K digestion and DL-Green cleanup, the labelled DNA was mixed with 1X Flow Buffer, in the presence of 1X DTT, and left to incubate overnight at 4°C. Staining was performed by adding 3.2 μl of a DNA stain solution for every 300 ng of pre-stained DNA and incubating at room temperature for at least two hours before loading onto the Bionano Chip. Loading of the chip and running of the Bionano Genomics Saphyr System were all performed according to the Saphyr System User Guide (https://bionanogenomics.com/support-page/saphyr-system/).
Fine-scale recombination rates for human, chimpanzee, bonobo, and gorilla were obtained from previously published sources [74, 75]. Recombination rates were averaged over 500 kbp windows and plotted as a function of distance to active, annotated centromeres on hg38 and the ancestral centromere at 15q25. Recombination data were smoothed using locally weighted smoothing with α = 0.05 for ease of visualization. The coordinates of the ancestral 15q25 centromere were delineated by clones RP11-152F13 and RP11-635O8 [2]. The positions of known recombination hotspots [40] and documented sites of interlocus gene conversion [41] were used to substantiate evidence for homology-driven repair activity at 15q25.
Sequencing data from Illumina HiSeq X Ten can be found at the Sequence Read Archive (SRA) under BioProject ID PRJNA493749, as BAM file (GRCh38/hg38). PacBio SMRT sequences of CH17 clones can be found under BioProject ID PRJNA514724. Complete sequences of primate BAC clones (S3 Table) can be found as NC_036894.1, NC_036918.1, AC275844.1, AC212984.3, AC210775.3 and AC211297.2.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.