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.ppat.1002941 | Primary and Secondary siRNAs in Geminivirus-induced Gene Silencing | In plants, RNA silencing-based antiviral defense is mediated by Dicer-like (DCL) proteins producing short interfering (si)RNAs. In Arabidopsis infected with the bipartite circular DNA geminivirus Cabbage leaf curl virus (CaLCuV), four distinct DCLs produce 21, 22 and 24 nt viral siRNAs. Using deep sequencing and blot hybridization, we found that viral siRNAs of each size-class densely cover the entire viral genome sequences in both polarities, but highly abundant siRNAs correspond primarily to the leftward and rightward transcription units. Double-stranded RNA precursors of viral siRNAs can potentially be generated by host RDR-dependent RNA polymerase (RDR). However, genetic evidence revealed that CaLCuV siRNA biogenesis does not require RDR1, RDR2, or RDR6. By contrast, CaLCuV derivatives engineered to target 30 nt sequences of a GFP transgene by primary viral siRNAs trigger RDR6-dependent production of secondary siRNAs. Viral siRNAs targeting upstream of the GFP stop codon induce secondary siRNAs almost exclusively from sequences downstream of the target site. Conversely, viral siRNAs targeting the GFP 3′-untranslated region (UTR) induce secondary siRNAs mostly upstream of the target site. RDR6-dependent siRNA production is not necessary for robust GFP silencing, except when viral siRNAs targeted GFP 5′-UTR. Furthermore, viral siRNAs targeting the transgene enhancer region cause GFP silencing without secondary siRNA production. We conclude that the majority of viral siRNAs accumulating during geminiviral infection are RDR1/2/6-independent primary siRNAs. Double-stranded RNA precursors of these siRNAs are likely generated by bidirectional readthrough transcription of circular viral DNA by RNA polymerase II. Unlike transgenic mRNA, geminiviral mRNAs appear to be poor templates for RDR-dependent production of secondary siRNAs.
| RNA silencing directed by small RNAs (sRNAs) regulates gene expression and mediates defense against invasive nucleic acids such as transposons, transgenes and viruses. In plants and some animals, RNA-dependent RNA polymerase (RDR) generates precursors of secondary sRNAs that reinforce silencing. Most plant mRNAs silenced by miRNAs or primary siRNAs do not spawn secondary siRNAs, suggesting that they may have evolved to be poor templates for RDR. By contrast, silenced transgenes often produce RDR-dependent secondary siRNAs. Here we demonstrate that massive production of 21, 22 and 24 nt viral siRNAs in DNA geminivirus-infected Arabidopsis does not require the functional RDRs RDR1, RDR2, or RDR6. Deep sequencing analysis indicates that dsRNA precursors of these primary viral siRNAs are likely generated by RNA polymerase II-mediated bidirectional readthrough transcription on the circular viral DNA. Primary viral siRNAs engineered to target a GFP transgene trigger robust, RDR6-dependent production of secondary siRNAs, indicating that geminivirus infection does not suppress RDR6 activity. We conclude that geminiviral mRNAs, which can potentially be cleaved by primary viral siRNAs, are resistant to RDR-dependent amplification of secondary siRNAs. We speculate that, like most plant mRNAs, geminiviral mRNAs may have evolved to evade RDR activity.
| RNA silencing directed by miRNAs, short interfering (si)RNAs and PIWI-interacting RNAs is involved in regulation of gene expression and chromatin states and in defense against invasive nucleic acids such as transposons, transgenes and viruses [1]–[3]. Virus-infected plants accumulate high levels of viral siRNAs (vsRNAs) of three major size-classes: 21-nt, 22-nt and 24-nt [4], [5]. In Arabidopsis thaliana infected with DNA viruses, all four Dicer-like (DCL) enzymes are involved in processing of vsRNA duplexes from longer double-stranded RNA (dsRNA) precursors: DCL4 and DCL1 generate 21-nt class, DCL2 generates 22-nt class and DCL3 generates 24-nt class; 21-nt and 24-nt vsRNAs accumulate at higher levels than 22-nt vsRNAs [6]–[8]. By contrast, in RNA virus-infected Arabidopsis, DCL4-dependent 21-nt vsRNAs and/or DCL2-dependent 22-nt vsRNAs are the most abundant species, whereas DCL3-dependent 24-nt vsRNAs accumulate at much lower levels [7], [9], [10]. This reflects the difference in viral life cycles: DNA viruses transcribe their genomes in the nucleus, whereas RNA viruses are generally restricted to the cytoplasm. Likewise, plant endogenous genes and transgenes that undergo transcriptional silencing spawn predominantly DCL3-dependent 24-nt siRNAs, whereas those that undergo post-transcriptional silencing spawn predominantly DCL4-dependent 21-nt siRNAs and, in certain cases, DCL2-dependent siRNAs [1], [11], [12].
In endogenous and transgene-induced silencing pathways, dsRNA precursors of siRNAs can be generated by RNA-dependent RNA-polymerase (RDR). The Arabidopsis thaliana genome encodes six RDRs, three of which have been implicated in siRNA biogenesis [13]. RDR2 is required for biogenesis of 24-nt heterochromatic siRNAs (hcsiRNAs) mainly originating from repetitive DNA loci including transposons. RDR6 is required for biogenesis of trans-acting siRNAs (tasiRNAs), natural antisense transcript siRNAs and siRNAs derived from posttranscriptionally-silenced transgenes [1]. RDR6 is also involved in production of secondary siRNAs from some protein-coding genes targeted by miRNAs [14], [15]. RDR1 has so far been implicated in viral siRNA biogenesis (see below) and its function in endogenous or transgene-induced silencing is not known. Presumptive single-stranded RNA templates for RDR2 are produced by plant-specific RNA polymerases Pol IV and/or Pol V, but little is known about Pol IV and Pol V transcripts and RDR2-dependent dsRNAs [16]. dsRNA precursors of tasiRNAs originate from Pol II transcripts of TAS genes, which are cleaved by a miRNA::Argonaute (AGO) protein complex [17]–[20]. Either the 3′ cleavage product or the 5′ cleavage product is converted by RDR6 to dsRNA: RDR6 recruitment to only one of the two cleavage products is determined by 22-nt size of the initiator miRNA produced from a bulged hairpin precursor [21]–[23] or a second binding site of the miRNA::AGO complex [17], [19], respectively.
The possible role of RDRs in vsRNA biogenesis has been extensively studied using A. thaliana single, double and triple null mutants for RDR1, RDR2 and RDR6 [8], [24]–[28]. These studies produced rather conflicting results, but in many cases, wild type viruses were shown to predominantly spawn RDR-independent vsRNAs [29]. However, mutant RNA viruses with deletion or point mutation in the viral silencing suppressor gene spawn RDR6- and/or RDR1-dependent vsRNAs [26]–[28]. As a consequence the suppressor-deficient RNA viruses could establish systemic infection only on A. thaliana mutant plants lacking RDR6 and/or RDR1 activity. Nevertheless, suppressor-deficient RNA viruses spawn substantial amounts of RDR-independent vsRNAs. Thus, one of the major precursors of RNA virus-derived vsRNAs is likely a double-stranded replicative intermediate, transiently produced by viral RNA-dependent RNA-polymerase (vRdRP). Primary vsRNAs generated from such precursors may trigger RDR-dependent production of secondary siRNAs.
Plant DNA viruses do not encode a vRdRP. However, the biogenesis of DNA virus-derived vsRNAs does not appear to involve host RDRs. Thus, Cauliflower mosaic virus (CaMV)-derived vsRNAs of all major classes accumulate at comparable high levels in A. thaliana wild-type and rdr1 rdr2 rdr6 triple mutant plants and their long dsRNA precursors are likely generated by Pol II [8]. The lack of RDR-dependent vsRNAs can be explained by the ability of a CaMV silencing suppressor protein to interfere with DCL4-mediated processing of dsRNAs produced by RDR6 [30], [31]. Silencing suppressor proteins of DNA geminiviruses have not been reported to interfere with RDR activity or DCL-mediated processing of RDR-dependent dsRNAs. In A. thaliana null mutants for Pol IV, RDR2, or RDR6 activity, the biogenesis of vsRNAs from Cabbage leaf curl virus (CaLCuV; a member of genus Begomovirus of the family Geminiviridae) was not affected, suggesting that RDR2 and RDR6 are not involved in production of dsRNA precursors of vsRNAs [7]. However, involvement of RDR1 in this process or possible redundancy in activities of distinct RDRs were not investigated so far.
Geminiviruses encapsidate circular single-stranded (ss)DNA of ca. 2.5-to-2.7 kb in geminate virions and accumulate in the nucleus as multiple circular dsDNA minichromosomes. The minichromosomes are both the intermediates of rolling circle replication and the templates for Pol II-mediated bidirectional transcription [32]. Like many members of the genus Begomovirus, CaLCuV has a bipartite genome comprising 2.6 kb DNA-A and 2.5 kb DNA-B [33]. The DNA-A encodes proteins involved in replication (AC1 and AC3), transcription (AC2) and encapsidation (AV1), while the DNA-B encodes BC1 and BV1 proteins with movement functions. A large intergenic region on DNA-A and DNA-B contains a 192 bp common region of nearly identical sequence with the origin of replication and bidirectional promoter elements. By analogy with other begomoviruses [34], the bidirectional promoter is expected to drive Pol II transcription of the leftward (AC1/AC4/AC2/AC3 and BC1) and rightward (AV1 and BV1) genes. In addition, a monodirectional promoter is expected to drive Pol II transcription of a short AC2/AC3 transcript, which is co-terminal with the long AC1/AC4/AC2/AC3 transcript. On both DNAs, the leftward and rightward transcription is terminated by poly(A) signals located in a close vicinity on the virion (sense) and complementary (antisense) strands, respectively. In CaLCuV DNA-A, this juxtaposition of the poly(A) signals creates a ca. 25-nt overlap of the sense and antisense transcripts. Such overlap was proposed to form a dsRNA precursor of primary vsRNAs [35], which may initiate RDR-dependent production of vsRNAs from other regions of the viral transcripts.
Such phenomenon of transitivity has been described for posttranscriptional and transcriptional silencing of a transgene targeted by vsRNAs (virus-induced gene silencing; VIGS) or by primary siRNAs derived from an inverted-repeat transgene. In these cases, RDR6- or RDR2-dependent production of secondary siRNAs outside of the target region was detected, respectively [36], [37]. Notably, posttranscriptional silencing of endogenous plant genes by virus- or transgene-derived primary siRNAs was not associated with secondary siRNA production [36], [38], [39], suggesting that endogenous mRNAs are not good templates for RDRs.
In this study, we used Illumina deep sequencing of short RNAs, combined with blot hybridization and genetic analysis, to investigate the biogenesis of primary and secondary siRNAs. To this end, Arabidopsis wild-type, RDR-mutant and transgenic plants were infected with CaLCuV or its derivatives carrying fragments of an endogenous gene or a transgene. We found that, like most endogenous plant mRNAs, viral mRNAs are not prone to transitivity: the majority of vsRNAs are RDR1-, RDR2- and RDR6-independent primary siRNAs. By contrast, a transgene mRNA targeted by primary vsRNAs is subject to RDR6-dependent production of secondary siRNAs. We also found that silencing of the transgene driven by a CaMV 35S promoter can be triggered by primary vsRNAs targeting an enhancer (but not core promoter) region and this, presumably transcriptional, silencing was not associated with accumulation of secondary siRNAs.
To analyze begomovirus interactions with the host small RNA (sRNA)-generating silencing pathways, we deep-sequenced sRNA populations from mock-inoculated and CaLCuV-infected A. thaliana wild-type (Col-0) plants and CaLCuV-infected rdr1 rdr2 rdr6 triple null mutant plants (rdr1/2/6 in Col-0 background; [8]). The protocol was designed to sequence short RNAs with 5′-phosphate and 3′-hydroxyl groups, which include DCL products. Samples of total RNA extracted from pools of three plants were processed in parallel and the resulting cDNA libraries sequenced in one channel of an Illumina Genome Analyzer, thus allowing quantitative comparison of changes in the profile of host sRNAs upon virus infection and the profile of vsRNAs in wild-type versus mutant plants.
A total number of reads in the high-coverage libraries was ranging from 9.3 to 10.4 million, of which 7.3 million (‘Col-0 mock’), 5.3 million (‘Col-0 CaLCuV’) and 5.0 million (‘rdr1/2/6 CaLCuV’) of 20–25 nt reads mapped to the Arabidopsis thaliana Col-0 or CaLCuV genomes with zero mismatches (Table S1A). Two additional low-coverage libraries with 0.45 million (‘Col-0 mock*’) and 0.43 million (‘Col-0 CaLCuV*’) of 20–25 nt reads with zero mismatches (Table S1A) were obtained in an independent experiment.
In mock-inoculated plants, most of the 20–25 nt sRNAs mapped to the A. thaliana genome (Figure 1A; Table S1A). The 24-nt and 21-nt classes were predominant (35% and 28%, respectively), whereas other size-classes were less abundant (23-nt – 19%; 22-nt – 8%; 20-nt – 7%; 25-nt – 3%) (Figure 1B). This is consistent with the previous studies showing that 24-nt hcsiRNAs and 21-nt miRNAs are the most abundant sRNA classes in A. thaliana [40], [41]. Upon CaLCuV infection, the host sRNA profile was slightly altered in that the 21-nt class became the largest (32%) and the 24-nt class the second largest (28%) (Figure 1B; Table S1A). A similar shift in the host sRNA profile was also detected in the low coverage experiment (Table S1A). By contrast, A. thaliana infection with the pararetrovirus CaMV results in overaccumulation of 24-nt host sRNAs [8]. The biological significance of the opposite effects of geminivirus and pararetrovirus infections on host sRNAs remains to be investigated.
In CaLCuV-infected Col-0 plants, a large fraction of 20–25 nt reads mapped to the virus genome with zero mismatches (ca. 32% and 62% in the high- and low-coverage libraries, respectively; Figure 1A and Table S1A). Notably, the viral DNA-B was the major source of vsRNAs (70% and 85% of 20–25 nt viral reads, respectively; Table S1A). On both DNA-A and DNA-B, vsRNA reads were almost equally distributed between the virion and complementary strands (Table S1A; Figures 2 and S1). Similar to the host sRNAs in infected plants, 21-nt and 24-nt vsRNAs represent the first (42%) and the second (31%) largest fractions of 20–25 nt viral reads, respectively. But unlike the host sRNAs, 22-nt viral reads represent the third largest fraction (18%), while 20-nt, 23-nt and 25-nt classes are significantly underrepresented (Figure 1C). This size-class profile of CaLCuV vsRNAs agrees with our blot hybridization analysis using short probes and confirms the involvement of distinct DCLs in vsRNA biogenesis (Figure S2; [7]).
Interestingly, the host sRNAs of 21-nt and 24-nt classes exhibit a strong bias to 5′-terminal uridine (5′U; 69%) and 5′-terminal adenosine (5′A; 52%), respectively (Table S1A), owing to the preferential association of miRNAs with AGO1 and hcsiRNAs with AGO4 [17], [42]–[44]. By contrast, vsRNAs of 21-nt and 24-nt classes are less strongly enriched in 5′U (46%) and 5′A (32%), respectively, and the second most dominant nucleotide is 5′A for 21-nt class (25%) and 5′U for 24-nt class (32%) (Table S1A). Both the diversity in nucleotide composition and size of CaLCuV vsRNAs and the lack of any strong 5′-nucleotide bias imply the involvement of multiple AGOs in sorting vsRNAs.
Inspection of single-nucleotide resolution maps of 20–25 nt vsRNAs revealed that unique vsRNA species of each major class (21-nt, 22-nt and 24-nt) cover the entire genome of CaLCuV in both sense and antisense polarity as dense tiling arrays without gaps on the circular sequences of 2583 bp DNA-A and 2513 bp DNA-B (Tables S2 and S3). Hence, dsRNA precursors of vsRNAs of each class should cover the entire circular viral DNAs. However, the relative abundance of vsRNAs varies drastically: several large regions of DNA-A and DNA-B are densely covered in both polarities with vsRNA hotspots (defined here arbitrarily as short sequence segments spawning several vsRNA species with more than 300 reads each) (Figure 2 and Figure S1). This implies the existence of several overlapping dsRNA precursors that accumulate at high and low levels. Interestingly, vsRNA hotspots on both virion and complementary strands are interrupted with short sequences that spawn vsRNAs of lower abundance (Figure 2 and Figure S1; Table S2 and Table S3). This implies differential stability of vsRNA duplexes processed consequently from ends of long dsRNA precursors or, alternatively, preferential internal excisions of vsRNA duplexes from certain regions of a long dsRNA. We also found that most vsRNA hotspots contain all the three major size-classes of vsRNAs (Figure S1; Table S2 and Table S3), indicating that same dsRNA precursors are processed by different DCLs. This conclusion is consistent with our genetic analysis coupled with blot-hybridization of DNA virus-derived sRNAs [6], [7] (Figure S2) and sRNA deep-sequencing studies of other viruses [8], [45]–[48].
In DNA-A, the most abundant vsRNAs of both sense and antisense polarities, which include those with more than 1000 reads, originate from the AV1 ORF (Figure 2A and Figure S1A). The left border of this vsRNA hotspot region is at position 331 (Table S2), where the transcription start site can be predicted, i.e. at an optimal distance downstream of the TATA box (TATATAA at positions 228–305) and 9 nts upstream of the AV1 start codon (339–341). The right border of this vsRNA hotspot is at around position 1060 (Table S2), i.e. just upstream of the AV1 stop codon (1092–1094). After a short gap of 55 bp (1061–1116) lacking highly abundant vsRNAs, a large region spanning all the leftward ORFs is also covered with vsRNA hotspots, albeit at lower density than in the AV1 region. In this region, the most abundant vsRNAs originate from the large portion of the AC1 ORF including the nested AC4 ORF and less abundant vsRNAs from the AC2 ORF (Figure 2A; Table S2). Notably, the 25 nt region (1089–1113), in which the rightward (AV1) and the leftward (AC1/AC4/AC2/AC3 and AC2/AC3) viral mRNAs are expected to overlap and potentially form a dsRNA substrate for DCL, is not a vsRNA hotspot. Likewise, the 240 bp intergenic region between the predicted leftward and rightward transcription start sites (at positions 93 and 331, respectively), which contains the bidirectional promoter elements and overlaps the common region (22–213), is also devoid of vsRNA hotspots: it has only two islands covered with vsRNAs of 100–250 reads. Furthermore, the promoter region in front of the predicted transcription start site of AC2/AC3 mRNA (position 1651, downstream of TATATAA at 1683–1677) does not contain any prominent vsRNA hotspots (Figure 2A and Figure S1A; Table S2). Taken together, the promoter and terminator regions of CaLCuV DNA-A are devoid of highly abundant vsRNAs. Thus, the virus may have evolved a mechanism to evade transcriptional silencing which could potentially be directed by vsRNAs.
In DNA-B, two large regions are covered with extreme hotspots containing multiple vsRNA species with more than 1000 reads on both sense and antisense strands. The first is located downstream of the common region and it spans a large portion of the BV1 ORF. The second is located upstream of the common region and it spans a large portion of the BC1 ORF (Figure 2B and Figure S1B; Table S3). Like in DNA-A, the terminator region of rightward (BV1) and leftward (BC1) genes is devoid of vsRNA hotspots. Note that the DNA-B poly(A) signals AATAAA are located at positions 1305–1310 and 1356–1361 of the virion and complementary stands, respectively, and therefore the BV1 and BC1 mRNAs are not expected to overlap. A predicted BC1 promoter region with the TATA-box at positions 2471–2463 (TATATAA) is devoid of vsRNA hotspots and the border of the vsRNA hotspot region corresponds to the predicted transcription start site at 2439. Thus, BC1 mRNA can form one of the strands of a vsRNA precursor. In contrast, a predicted BV1 promoter region with the TATA-box at position 442–447 (TATATAA) is covered with vsRNA hotspots on both strands. This suggests that the region upstream of the BV1 ORF might be actively transcribed. Interestingly, it contains an ORF at positions 319 to 471 (Figure 2B). Such active transcription could in turn lead to production of abundant vsRNAs that can potentially direct transcriptional silencing of the BV1 promoter. This may represent either a host antiviral defense or a viral strategy of gene regulation.
Based on close inspection of cold versus hot spots of viral siRNAs, AU-rich sequences can generally be considered as a poor source of siRNAs, possibly owing to relatively low stability of AU-rich siRNA duplexes processed by DCLs from long dsRNA precursors. Other features of RNA primary or secondary structure which might potentially influence siRNA biogenesis or stability remain to be further investigated.
The Arabidopsis sRNA profile is drastically altered in rdr1/2/6 triple mutant compared to wild-type plants: 24-nt and 23-nt classes are selectively and strongly reduced, mainly owing to the loss of RDR2-dependent hcsiRNAs [40]. Thus, 21-nt class becomes the most predominant, followed by 20-nt and 22-nt classes (Table S1A): these three classes are mainly populated with RDR-independent miRNAs, whereas RDR6-dependent tasiRNAs and secondary siRNAs are much less abundant [41]. By contrast, the CaLCuV vsRNA profile was only slightly altered in rdr1/2/6 compared to wild-type (Figure 1C).
The overall accumulation level of 20–25 nt vsRNAs was higher in rdr1/2/6 than wild-type plants. If normalized by the levels of 21-nt host sRNAs (1.22 million in ‘Col-0 CaLCuV’ versus 1.21 million in ‘rdr1/2/6 CaLCuV’), this ca. 1.5-fold increase is mainly owing to higher accumulation of DNA-B vsRNAs of all the major classes (Table S1A; Figure 1A).
The single-nucleotide resolution maps of vsRNAs from Col-0 and rdr1/2/6 are remarkably similar. The vsRNA hotspots occur in the same regions and the relative abundance of vsRNA species is very similar within most hotspots (Figure 2 and Figure S1; Table S2 and Table S3). For DNA-A, the levels of 20–25 nt vsRNAs derived from the AC2 hotspot region are relatively lower in rdr1/2/6 than in Col-0, whereas those derived from the AV1 region are generally similar in rdr1/2/6 and Col-0 (Figure 2A), with an exception of 24-nt vsRNAs that accumulate at relatively higher levels in rdr1/2/6 (Figure S1A; Table S1A). For DNA-B, the levels of 20–25 nt vsRNAs in most hotspots are 1.5- to 2.5-fold higher in rdr1/2/6 than in Col-0, with an exception of the middle part and the 3′ part of BV1 ORF, in which vsRNA levels are generally similar in rdr1/2/6 and Col-0 or, at some locations in the 3′ part, lower in rdr1/2/6 (Figure 2B). No drastic difference in the relative abundance of vsRNA size-classes along the DNA-B sequence was observed (Figure S2B; Table S3).
Analysis of 5′-terminal nucleotides of vsRNAs revealed no substantial difference between Col-0 and rdr1/2/6 (Table S1A), further supporting that vsRNA biogenesis is not drastically affected by null mutations in RDR1, RDR2 and RDR6.
The above-described deep sequencing findings for vsRNA size-classes, relative abundance and distribution along the viral genome and RDR1/2/6-independence of vsRNA biogenesis were confirmed by blot hybridization analysis of sRNAs from CaLCuV-infected wild-type and rdr1/2/6 mutant plants using several short probes specific to DNA-A or DNA-B (Figure S2 and Figure 3B). In addition, analysis of CaLCuV-infected dcl1 dcl2 dcl3 dcl4 quadruple mutant plants (dcl1/2/3/4) confirmed our previous findings that the majority of vsRNAs are generated by four DCLs [7]. We further established that a mutant DCL1 protein produced from the dcl1-9/caf1 allele in dcl1/2/3/4 plants [8] appears to be capable of generating 21-nt vsRNA from dsRNA precursors derived from vsRNA hotspot regions of DNA-B (Figure S2). Likewise, a major fraction of 21-nt vsRNAs derived from the leader region of CaMV, which is an extreme hotspot of 21-24 nt vsRNA production, requires DCL1 for their biogenesis and residual accumulation of 21-nt vsRNAs was observed in dcl1/2/3/4 [8].
Taken together, our findings indicate that CaLCuV vsRNA biogenesis does not require RDR1, RDR2, or RDR6. However, there appears to be a quantitative difference in relative abundance of dsRNA precursors derived from the vsRNA hotspot regions of DNA-A and DNA-B in wild-type versus rdr1/2/6 plants.
To test if the observed differences in relative abundance of vsRNAs correlate with relative levels of viral transcripts and/or viral DNA, we measured the accumulation of viral long nucleic acids in wild-type and rdr1/2/6 plants by RNA and DNA blot hybridization as well as real time PCR (Figure 3). The results of total RNA (Figure 3A) and polyadenylated mRNA (Figure 3D) analyses revealed that the relative accumulation of viral transcripts positively correlates the relative abundance of vsRNAs in the major hot spot regions. Indeed, AV1 mRNA, the most readily detectable viral transcript, accumulated at slightly higher levels in rdr1/2/6 than wild type plants, whereas accumulation of the less abundant AC2/AC3 mRNA was slightly reduced in rdr1/2/6. This resembles the profile of DNA-A derived vsRNAs and its alteration in rdr1/2/6. Furthermore, accumulation of BC1 and BV1 polyadenylated mRNAs was increased ca. 1.2- and 1.4-fold, respectively, in rdr1/2/6 compared to wild type plants, which correlates with slightly increased accumulation of DNA-B derived vsRNAs in rdr1/2/6. Notably, in addition to viral mRNAs, shorter viral transcripts also accumulate at high levels and appear as a smear on the total RNA blot (Figure 3A). These aberrant RNAs may represent degradation products of viral mRNAs or prematurely terminated viral transcripts. In the case of DNA-B, the aberrant RNAs appear to be much more abundant than BV1 and BC1 mRNAs, since the latter are barely detectable (Figure 3A). This correlates with much higher accumulation of vsRNAs from DNA-B than DNA-A (Figure 1A). The higher abundance of aberrant RNAs transcribed from DNA-B can be explained by higher accumulation of total DNA-B compared to total DNA-A as estimated by Southern (Figure 3C).
Real time PCR analysis (Figure 3D) revealed that total viral DNA accumulates at higher levels in rdr1/2/6 compared to wild type plants (ca. 1.4- and 2-fold increase for DNA-A and DNA-B, respectively). However, Southern blot hybridization analysis (Figure 3C) showed that this increase is mainly owing to increased accumulation of viral single-stranded DNA (ssDNA). By contrast, the levels of viral dsDNA, which serves as a template for both transcription and replication, are similar in wild type and rdr1/2/6 plants. Thus, rolling circle and/or recombination-dependent replication mechanisms [32] produce increased levels of viral ssDNA (but not dsDNA) in the absence of RDR1, RDR2 and RDR6. This finding implicates an RDR activity in the regulation of geminiviral DNA replication. Interestingly, homologous recombination-dependent, double-stranded DNA brake (DSB) repair in Arabidopsis involves DSB-induced small RNAs (diRNAs) [49]. RDR2 and RDR6 play redundant roles in the biogenesis of diRNAs, implicating RDR activity in DSB repair.
Our above-described results suggested that CaLCuV vsRNAs are primary siRNAs (i.e. RDR-independent) and that secondary siRNAs (i.e. RDR-dependent) may comprise only a small fraction of vsRNAs (if any). To investigate if primary vsRNAs are capable of triggering production of secondary siRNAs in CaLCuV-infected plants, we used a virus-induced gene silencing (VIGS) vector based on the CaLCuV DNA-A derivative lacking most of the AV1 ORF sequence (positions 350–1032) [50].
When a 354 bp fragment of the A. thaliana Chlorata I (ChlI/CH42; At4g18480) gene ORF is inserted in place of the AV1 ORF, the resulting recombinant virus CaLCuV::Chl knocks down ChlI mRNA levels in all tissues of CaLCuV::Chl-infected A. thaliana plants [7] and causes whitening of newly growing tissues due to the loss of chlorophyll (“chlorata” phenotype; [50]). The recombinant virus spawns abundant 21, 22, and 24 nt siRNAs from the ChlI insert, whose biogenesis does not require RDR6 or RDR2. However, an extensive chlorata phenotype is nearly abolished in rdr6 and dcl4 null mutant plants [7], suggesting that RDR6-/DCL4-dependent secondary siRNAs might be involved in total silencing the ChlI gene. To test this hypothesis we deep-sequenced sRNAs from CaLCuV::Chl-infected Col-0 plants exhibiting an extensive chlorata phenotype.
Of 2.28 million total 20–25 nt reads, 1.58 million mapped to the A. thaliana genome and 0.61 million to CaLCuV::Chl genome (A+B) with zero mismatches. Of the latter reads, 0.45 million originate from the circular CaLCuV::Chl DNA and 0.16 million from DNA-B (Table S1B). This is in contrast to our above observation for wild-type CaLCuV which spawns more abundant vsRNAs from DNA-B.
Inspection of the single-nucleotide resolution map of 20–25 nt sRNAs perfectly matching to a 3298 bp region of the A. thaliana genome, which contains the ChlI gene, revealed that of the 109′098 redundant reads, 109′002 originate from the 354 bp segment (positions 1192–1545) that corresponds exactly to the ChlI segment inserted in CaLCuV::Chl. The remaining sRNAs (91 reads) originate mostly from the ChlI sequence downstream of this segment (Figure 4; Table S1B and Table S4). We conclude that accumulation of secondary siRNAs outside of the vsRNA target region is negligible compared to primary siRNAs. This is consistent with the previous studies that detected no transitivity when endogenous plant genes were knocked down by RNA virus- or transgene-induced silencing [36], [38], [39].
Within the ChlI target region the sRNA profile resembles the global profile of CaLCuV vsRNAs in that the three size-classes are predominant (21-nt – 30%; 22-nt – 25%; 24-nt – 38%). However, the distribution of sRNAs is unequal between the strands: 80% of 20–25 nt reads map to the coding strand, and 21-nt and 22-nt classes derived from the coding strand are equally abundant (28% each). This strong bias is due to a bigger number of sRNA hotspots and higher accumulation levels of sRNA species within the hotspots on the coding strand (Figure 4; Table S4). The significance of this bias for ChlI silencing remains to be investigated.
In A. thaliana, the ChlI gene has a close homolog ChlI-2 (At5g45930), silencing of which is likely required for the chlorata phenotype. To address if potential silencing of ChlI-2 is associated with secondary siRNA production we created a map of ChlI-2 sRNAs (Figure S3A). Of 3′093 reads of 20–25 nt sRNAs matching the ChlI-2 genomic locus with zero mismatches in CaLCuV::Chl-infected plants, 2′987 reads map within the 354 bp VIGS-target sequence and only 104 (ca. 3%) map downstream of the target. Moreover, within the target sequence almost all the reads (2′977) match two sequence stretches of >20 nts in length which are identical in ChlI and ChlI-2 (Figure S3A; Table S4). Thus, similar to ChlI, only small amounts of secondary siRNAs are generated on ChlI-2 target gene. Presently, we cannot exclude that these small amounts of secondary siRNAs are required for total chlorata silencing. As we hypothesized earlier [7], total Chl silencing is likely established in newly emerging leaves by mobile RDR6- and DCL4-dependent Chl siRNAs. Recent studies indicate that 21–24 nt siRNAs act as mobile silencing signals and can direct mRNA cleavage and DNA methylation in recipient cells, even though they accumulate in recipient tissues at much lower levels than in source tissues [51], [52].
Notably, vsRNAs targeting ChlI-2 mRNA at two potentially cleavable sites separated by ca. 100 nts do not trigger any robust secondary siRNA production from the intervening region. This indicates that a two-hit model for the RDR6-dependent biogenesis of tasiRNAs and other secondary siRNAs [14], [19], [53] does not apply for ChlI-2 and ChlI.
Like in the wild-type DNA-A, vsRNAs cover the entire circular CaLCuV::Chl DNA in both orientations without gaps (Table S4). However, vsRNA hotspots are more evenly distributed along the CaLCuV::Chl sequence compared to the wild-type DNA-A: in fact, new hotspots appear in the intergenic region between the transcription start sites as well as in the terminator region (Figure S3; Table S4). This finding was confirmed by blot hybridization (Figure S2, compare CaLCuV wt and CaLCuV::Chl). Furthermore, genetic analysis revealed that production of vsRNAs from any region of CaLCuV::Chl including the ChlI insert does not require RDR6 or RDR2, since vsRNAs of all classes accumulated at similar levels in wild type and rdr2 rdr6 double mutant plants (rdr2/6; Figure S2). The latter finding indicates that RDR6-dependent secondary siRNA production does not occur within the VIGS target region and that potential cleavage of endogenous (ChlI or ChlI-2) and CaLCuV mRNAs at two sites is not sufficient to attract RDR6 activity.
Taken together, our findings for both wild-type and CaLCuV::Chl viruses suggest that dsRNA precursors of vsRNAs originate from the entire circular viral DNAs including “non-transcribed” intergenic regions. Therefore, these precursors might be produced by Pol II-mediated readthrough transcription far beyond the poly(A) signals, thus encircling the viral DNA in sense and antisense orientation. It can be further suggested that such readthrough transcription is more efficient on CaLCuV::Chl DNA-A than wild-type DNA-A, owing to the smaller size and the chimeric configuration of the rightward transcription unit carrying the ChlI segment. This would explain prominent hotspots in the promoter and terminator regions and also much higher production of vsRNAs from CaLCuV::Chl DNA-A than DNA-B, which is not the case for wild-type CaLCuV. Notably, CaLCuV::Chl is an attenuated virus which produces much less severe symptoms than wild type CaLCuV [49]. Whether vsRNA-directed silencing contributes to the attenuated symptom development of this recombinant virus remains to be investigated.
The apparent paucity of secondary siRNAs derived from CaLCuV mRNAs or ChlI and ChlI-2 mRNAs could be explained by two scenarios. In the first scenario, the products of potential vsRNA-directed cleavage of host and viral mRNAs are not optimal templates for RDR activity. In the second one, CaLCuV infection blocks RDR activity and thereby prevents RDR-dependent amplification of siRNAs. To distinguish between these scenarios, we used the CaLCuV VIGS vector for targeting a transgene in the A. thaliana line L2 expressing green fluorescence protein (GFP) under the control of the CaMV 35S promoter and terminator (35S::GFP; [54]; Figure 5). Like other transgenes, 35S promoter-driven GFP transgenes in A. thaliana and N. benthamiana are prone to transitivity in which secondary siRNAs are generated outside of the region targeted by primary sRNAs [36], [38], [55]. An aberrant nature of transgenic transcripts appears to attract RDR activity.
We inserted in the CaLCuV vector a full-length (FL), 771 bp GFP coding sequence (designated ‘CodFL’) or 30-bp sequences of the GFP transgene transcribed region. The latter is defined here as the GFP mRNA region from the transcription start site to the mRNA processing/poly(A) addition site. As depicted in Figure 5, the short inserts included the sequences from within the 5′-untranslated region (5′UTR) (designated ‘Lead’), the beginning, middle and end of the coding sequence (‘CodB’, ‘CodM’ and ‘CodE’), and the 3′UTR (‘Trail’ and ‘PolyA’) and the sequences surrounding the ATG start codon (‘Start’) or the TAA stop codon (‘Stop’). Inoculation of L2 plants with the resulting recombinant viruses by biolistic delivery of viral DNA led to development of local GFP silencing on inoculated leaves followed by systemic GFP silencing on newly-emerging infected tissues (both leaves and inflorescence; Figure S4B). GFP silencing in infected tissues, which was manifested under UV light as red fluorescence areas on otherwise green fluorescent tissues (Figure 5B and Figure S4B), well correlated with knockdown of GFP mRNA levels as measured by real time PCR (Figure S4D).
All the recombinant viruses carrying an insert from the GFP transcribed region induced systemic GFP silencing, although to various degrees (Figure 5B). Furthermore, in all these cases, GFP silencing correlated with accumulation of GFP siRNAs derived from both the short insert/target sequences and the GFP mRNA sequences outside of the target sequence (Figure 5C and Figure S4C). Notably, the 30 bp GFP insert/target sequences generally gave rise to abundant siRNAs of 21-nt, 22-nt and 24-nt classes, resembling those derived from the virus genome and therefore likely originating from the replicating virus carrying the insert rather than from the transgene. By contrast, secondary siRNAs derived from non-target sequences of the GFP transgene were generally represented by a dominant 21-nt class, although 22-nt and 24-nt classes were also detected (Figure 5C; also see below). Furthermore, targeting the GFP sequences upstream of the translation stop codon (Lead, Start, CodB, CodM and CodE) induced the production of abundant secondary siRNAs exclusively from sequences downstream of the target site, whereas targeting the 3′UTR sequences (Stop, Trail and PolyA) resulted in secondary siRNAs from the sequences upstream and downstream of the target site (Figure 5C). Such directionality in secondary siRNA biogenesis resembles that in RDR6-/DCL4-dependent biogenesis of tasiRNAs [17], [18]. Our findings further suggest that, following vsRNA-directed cleavage of GFP mRNA, the 5′-cleavage product might be protected by translating ribosomes from being converted to dsRNA precursor of secondary siRNAs. However, if it contains the translation stop codon, the ribosomes can terminate translation and be released. Thus, following vsRNA-directed cleavage downstream of the stop codon, both 5′ and 3′ cleavage products of GFP mRNA enter the secondary siRNA-generating pathway.
The above findings based on blot hybridization analysis (Figure 5C) were fully validated by Illumina sequencing of sRNAs from L2 plants infected with Lead, CodM, Trail and polyA viruses (Figure 6 and Figure S5; Table S5 and Table S6). In addition, analysis of the deep sequencing data showed that vsRNAs targeting the 3′UTR induce production of much more abundant secondary siRNAs from the region upstream of the target site than from downstream sequences (Figure 6). Interestingly, secondary siRNA hotspots are non-randomly distributed along the GFP transcribed region: in all the four cases the siRNA hotspots occur in the region comprising the 3′ portion of the GFP ORF and the beginning of the 3′UTR. The size-class profile and relative abundance of siRNA species in this siRNA hotspot region are very similar. In the case of Lead and polyA viruses, additional siRNA hotspots occur in the middle of GFP ORF and the 3′UTR, respectively (Figure 6 and Figure S5). Interestingly, vsRNAs targeting the 5′UTR does not induce abundant secondary siRNA production from the region immediately downstream of the target site, which contains the 5′ portion of GFP ORF. This region also appears to be a poor source/target of primary vsRNAs (see CodB in Figure 5). Furthermore, robust production of secondary siRNAs does not appear to depend on the accumulation levels of any major size-class of primary vsRNAs of antisense polarity that have the potential to cleave GFP mRNA and initiate secondary siRNA biogenesis (Figure S5; Table S1, Table S5 and Table S6). We assume that, once initiated by primary vsRNAs, secondary siRNA biogenesis might be reinforced by feedback loops in which certain secondary siRNAs of antisense polarity target the GFP mRNA. Such feedback loops regulate tasiRNA production from TAS1c gene, in which certain tasiRNAs cleave its own precursor transcript to initiate RDR6-dependent production of additional dsRNAs [20], and potentially occur in transgene-induced silencing systems [56], [57].
Contrary to what we observed for the transcribed region, targeting of the GFP non-transcribed regions with short sequences inserted into the CaLCuV VIGS vector did not lead to GFP silencing or secondary siRNA production in systemically-infected L2 plants (Figure 7 and Figure S4). The 30-bp sequences which surround the 35S core promoter elements including the CAAT and TATA boxes (‘CAAT’ and ‘TATA’) and the transcription start site (‘Plus1’), or sequences that occur in a distal region of the 35S enhancer (‘EnhSh’) and just downstream of the mRNA processing/poly(A) addition site (‘Post’) gave rise to abundant siRNAs of the three major classes but no secondary siRNAs were detected outside of the target sequence. Furthermore, insertion of the 90-bp 35S core promoter region (‘Core’) did not result in GFP silencing or secondary siRNA production, despite abundant primary siRNAs targeting this region. However, insertions of the entire 35S enhancer region of 272 bp (‘Enh’) or the full-length promoter of 382 bp (‘ProFL’) resulted in systemic GFP silencing. But also in these two cases no secondary siRNAs were detected outside of the target region (Figure 7). These findings were confirmed by Illumina sequencing of sRNAs from L2 plants systemically infected with Core, Enh and ProFL viruses (Figures 8 and Figure S6; Table S5 and Table S6). In addition, the deep sequencing revealed that, besides extremely low levels of secondary siRNA accumulation outside of the target region, there appear to be almost no secondary siRNA amplification within the target region. Thus, the duplicated 273-bp Enhancer* region shares 94% nucleotide identity with the target Enhancer region, since these sequences originate from two different strains of CaMV, and we found only negligible numbers of reads in the three stretches of the Enhancer* sequence that have mismatches to corresponding stretches of the Enhancer sequence (Figure 8; Table S5, see positions 760–781, 803–837 and 869–905).
Taken together, we conclude that production of abundant secondary siRNAs can be triggered by primary virus-derived siRNAs that target GFP mRNA. Hence, CaLCuV infection does not block amplification of secondary siRNAs likely mediated by RDR activities (see below). This is also supported by our blot hybridization analysis showing that accumulation of RDR6-dependent tasiRNAs is not significantly affected by CaLCuV infection (Figure S2; siR255). Both primary (virus-derived) and secondary siRNAs correlate with efficient GFP silencing. However, targeting of the non-transcribed, 35S enhancer region by primary siRNAs induces efficient GFP silencing without any substantial production of secondary siRNAs. Hence, secondary siRNAs do not appear to be necessary for silencing GFP transgene, at least at the transcriptional level. Previously, transcriptional VIGS through targeting the 35S promoter region of 35S::GFP transgene was observed but its dependence on primary or secondary siRNAs was not tested in that case [58].
To investigate genetic requirements for the biogenesis of GFP secondary siRNAs, the L2 transgenic line was crossed with the Col-0 mutant lines carrying point mutations in RDR6 (rdr6-14; [59]) and DCL4 (dcl4-2; [60]). The resulting homozygous mutant lines L2 x rdr6 and L2 x dcl4 expressed high levels of GFP, similar to those of the parental L2 plants (not shown).
Systemic infection of L2 x rdr6 and L2 x dcl4 plants with the recombinant viruses Lead, CodM and Trail resulted in GFP silencing in all cases, except L2 x rdr6 plants infected with the Lead virus. Consistent with our findings for wild-type CaLCuV (Figure S2) and CaLCuV::Chl ([7]; Figure S2), blot hybridization analysis revealed that the biogenesis of 21, 22 and 24 nt vsRNAs derived from the AC4 ORF region of the three recombinant viruses was not affected in L2 x rdr6 plants lacking RDR6 (Figure 9). By contrast, probes specific for the target transgene revealed a major contribution of RDR6 in secondary siRNA production. In fact, production of secondary siRNAs of all size-classes outside of the target region was nearly abolished in L2 x rdr6 plants infected with Lead, CodM and Trail viruses (Figure 9). For the latter two viruses, accumulation of siRNAs from the insert/target sequence was also reduced: interestingly, the reduced accumulation was observed for siRNAs of sense but not antisense polarity in CodM virus, while siRNAs of both polarities were strongly reduced in Trail virus. By contrast, accumulation of siRNAs from the Lead insert/target sequence was not altered in L2 x rdr6 plants infected with Lead virus (Figure 9). We conclude that RDR6-independent primary vsRNAs represent the majority of siRNAs derived from the Lead sequence, whereas the CodM and Trail sequences also spawn RDR6-dependent secondary siRNAs in addition to primary vsRNAs. These secondary siRNAs could potentially be produced from the transgene and/or the viral insert. We therefore used the probes specific to the viral sequence located just downstream of the insert (CbA1063_s and CbA1063_as), i.e. present in the chimeric rightward viral transcript. The results revealed that, in the case of Lead and CodM viruses, RDR6 is not involved in production of vsRNAs from this region (Figure 9). Thus, the contribution of RDR6 to siRNA production from the CodM insert/target sequence of antisense polarity can be explained by RDR6-dependent siRNA production from the target gene rather than the chimeric virus. However, accumulation of vsRNAs derived from the chimeric transcript region of Trail virus was substantially reduced (24-nt) or nearly abolished (21-nt and 22-nt) in L2 x rdr6 plants. This indicates that, in addition to the transgenic mRNA, the chimeric viral transcript can also be used for RDR6-dependent production of secondary siRNAs. But the insert sequence itself appears to regulate relative contribution of RDR6. Notably, the ChlI insert sequence does not make the chimeric viral transcript prone to RDR6-dependent vsRNA production (Figure S2). It remains to be further investigated why the Trail (but not Lead, CodM or ChlI) sequence makes the viral chimeric transcript prone to RDR6-dependent amplification of secondary siRNAs. Interestingly, this sequence originates from the CaMV terminator/leader region and contains two stretches of AG-repeats (Protocol S1).
It is puzzling that, in the absence of RDR6-dependent secondary siRNAs in L2 x rdr6 plants, the GFP silencing is efficiently triggered by CodM and Trail viruses but not by Lead virus. We speculate that GFP mRNA cleaved by primary siRNAs within its 5′UTR can still be translated, unless it enters the RDR6 pathway converting the coding and 3′UTR sequences to secondary siRNAs. By contrast, primary siRNA-directed cleavage within the coding sequence or 3′UTR would block productive translation and could therefore be sufficient for GFP silencing.
In L2 x dcl4 plants, we detected reduced accumulation of 21-nt primary siRNAs from the viral AC4 region and 21-nt primary and secondary siRNAs from the GFP sequences. Unexpectedly, accumulation of 22-nt and 24-nt primary and secondary siRNAs was increased: this increase was more prominent for secondary GFP siRNAs (Figure 9). This resembles the shift in the profile of RDR6-dependent 21-nt tasiRNAs in this particular mutant background ([60]; Figure 9, see tasiRNA siR255). Thus, a mutated DCL4 protein expressed from the dcl4-2 allele appears to promote processing of RDR6-dependent dsRNAs by alternate DCLs that generate longer siRNAs (i.e. DCL2 and DCL3).
Taken together, our findings confirm a major role of DCL4 in processing 21-nt secondary siRNAs from RDR6-dependent dsRNA precursors derived from the transgene and 21-nt primary vsRNAs from RDR6-independent viral dsRNA precursors. In addition, our results reveal that RDR6-dependent dsRNA can be efficiently processed by alternate DCL activities if the DCL4 protein is mutated by an amino acid substitution in the helicase domain. These alternate DCLs produce primary and secondary siRNAs which are equally potent in GFP silencing, since we did not observe any substantial difference in systemic silencing phenotypes between wild-type and dcl4-2 plants infected with any of the recombinant viruses. This is in line with our previous findings for CaLCuV::Chl-derived primary vsRNAs of distinct classes produced in single, double and triple dcl mutant plants, which could efficiently knockdown ChlI mRNA [7]. Previously, a major role of DCL2 was established for production of secondary siRNAs in a transgene targeted by primary siRNAs from another transgene [11]. Here, in addition to DCL2, we find the apparent involvement of DCL3 which normally generates 24-nt nuclear siRNAs in secondary siRNA production. Thus, a fraction of dsRNA precursors of the GFP transgene-derived secondary siRNAs might be localized in the nucleus. Alternatively, a fraction of DCL3 protein might also be cytoplasmic.
Secondary siRNAs are involved in various silencing pathways in plants, fungi and some animals. In C. elegans, RDR-dependent amplification of secondary siRNAs appears to reinforce silencing triggered by primary siRNAs which are processed by dicer from endogenous or exogenous dsRNA [61], [62]. In plants, some of the endogenous mRNAs targeted by miRNAs spawn RDR6-dependent secondary RNAs, a contribution of which to miRNA-directed gene silencing is not fully clarified [14], [15]. In most cases, plant miRNA-directed cleavage or translational repression is sufficient for robust gene silencing without production of secondary siRNAs [14]. Likewise, most plant mRNAs silenced by transgene- or virus-derived primary siRNAs do not spawn secondary siRNAs. This suggests that plant mRNAs could have evolved to be poor templates for RDR activity. Our study supports this notion by demonstrating that Arabidopsis ChlI and ChlI-2 mRNAs that undergo robust VIGS spawn only small amounts of secondary siRNAs. Furthermore, we demonstrate that geminiviral mRNAs, which can potentially be targeted by highly abundant vsRNAs of antisense polarity (Figure 2), are not templates for RDR1-, RDR2-, or RDR6-dependent siRNA amplification. By contrast, the transgenic GFP mRNA targeted by primary viral siRNAs spawns massive amounts of secondary siRNAs whose production requires RDR6. Our findings suggest that some aberrant feature(s) of the transgenic GFP mRNA possessing non-self UTR sequences may attract RDR6 activity. Notably, the involvement of RDR6 and RDR1 in production of viral siRNAs in RNA virus-infected plants was revealed only by using the mutant RNA viruses carrying deletions or point mutations in viral silencing suppressor genes: unlike wild-type RNA, the mutated viral RNA spawned RDR-dependent vsRNAs. What makes mutant/chimeric viral mRNAs and transgenic mRNAs good templates for RDR activity remains unclear. One possibility is that viral and plant mRNAs could have evolved primary sequence or secondary structure elements that block RDR activity. Such elements may accidentally be disrupted by mutations in the suppressor-deficient RNA viruses. Likewise, transgene transcripts might lack some of the naturally evolved sequence or structure elements.
Our findings suggest that the precursors of geminiviral siRNAs are most likely produced by Pol II-mediated bidirectional readthrough transcription in both sense and antisense orientations on the circular viral DNA. Such transcripts (or their degradation products) can potentially pair viral mRNAs and thus form perfect dsRNAs to be processed by multiple DCLs into vsRNAs. Readthrough transcription far beyond a poly(A) signal is a known property of Pol II. In pararetroviruses, it represents an obligatory mechanism by which a pregenomic RNA covering the entire circular genome is generated. The poly(A) signal of plant pararetroviruses is located at a relatively short distance (e.g. 180 bp in CaMV) downstream of the pregenomic RNA promoter: this allows efficient readthrough transcription at the first encounter by the Pol II complex and termination of transcription at the second encounter [63], [64]. Thus, substantial readthrough transcription can also be expected in geminiviruses which possess relatively short transcription units. Evidence for the existence of readthrough transcripts was obtained earlier for a related geminivirus [34] and is also provided here by deep sequencing showing that vsRNAs of both sense and antisense polarities densely tile along the entire CaLCuV genome including “non-transcribed” intergenic region of both DNA-A and DNA-B. Pol II readthrough transcription downstream of a canonical poly(A) signal of the endogenous A. thaliana gene FCA was recently shown to be repressed by a DCL4-dependent mechanism [12]. In a dcl4 mutant, the increased transcriptional readthrough far beyond the FCA poly(A) signal triggered silencing of a transgene containing the same 3′ region. Notably, the transgene silencing was caused by RDR6-dependent production of very abundant 22-nt siRNAs by DCL2 and less abundant 24-nt siRNAs by DCL3. This siRNA pattern resembles the pattern of GFP transgene-derived secondary siRNAs that we observed in L2 x dcl4 plants (Figure 9). Also in line with our observations, robust siRNA-directed silencing of the transgene and FCA did not spread to a converging gene that overlaps with the FCA readthrough transcript [12], further supporting the notion that most endogenous genes are not prone to RDR6-dependent transitivity.
Arabidopsis thaliana wild-type (Col-0) and rdr2/6, rdr1/2/6 and dcl1/2/3/4 mutant lines used in this study, their growth conditions and infection with wild-type CaLCuV (the DNA-A clone ‘CLCV-A dimer’ [33] and the DNA-B clone pCPCbLCVB.002 [50]) and CaLCuV::Chl (pMTCbLCVA::CH42 and pCPCbLCVB.002 [50]) using biolistic delivery of viral DNA have been described earlier [7], [8]. Using the same protocols, L2 transgenic plants (Line 2; [54]) were grown and inoculated with CaLCuV::GFP viruses.
L2 plants [54] were crossed with the dcl4-2 and rdr6-14 mutants [59], [60]. L2 homozygosity was determined by PCR in the F2 populations using 5′-TTGCTGCAACTCTCTCAGGGCC-3′ and 5′-GATAAATGTGGAGGAGAAGACTGCC-3′ for detecting the presence of the T-DNA and 5′-ACACTCTCTCTCCTTCATTTTCA-3′ and 5′-TCTGCAACACTCTGTCATTGG-3′ for detecting the absence of intact genomic region. RDR6-14 homozygosity was determined by visual observation of the typical epinastic leaf phenotype of the rdr6 mutants and was further confirmed using a dCAPS marker consisting of NcoI digestion of the PCR product obtained using 5′-AAGATTTGATCCCTGAGcCAT-3′ and 5′-GTTCGCCTTGTTTCTTGCTT-3′. DCL4-2 homozygosity was determined by the typical epinastic leaf phenotype of the dcl4 mutants. Homozygosity for L2 and the respective mutations were confirmed in F3 plants following the same procedures.
The CaLCuV::GFP viruses EnhSh, CAAT, TATA, Plus1, Lead, Start, CodB, CodM, CodE, Stop, Trail, PolyA and Post were generated by cloning preannealed sense and antisense oligonucleotides (listed in Protocol S1) into XbaI and XhoI sites of the CaLCuV VIGS vector pCPCbLCVA.007 [50]. The CaLCuV::GFP viruses Enh, Core and ProFL were generated by subcloning into XbaI and XhoI sites of pCPCbLCVA.007 the corresponding regions of the L2 T-DNA 35S promoter using PCR with primers listed in Protocol S1 on total DNA isolated from L2 transgenic plants. In all the above derivatives of the CaLCuV VIGS vector the insert sequences are in antisense orientation with respect of the AV1 gene promoter.
For both blot hybridization and Illumina deep-sequencing, aerial tissues of three virus-infected (or mock-inoculated) plants were harvested one month post-inoculation and pooled for total RNA preparation using a Trizol method [7]. sRNA blot hybridization analysis was performed as in Blevins et al. [7] using short DNA oligonucleotide probes listed in Protocol S1. cDNA libraries of the 19–30 nt RNA fraction of total RNA samples were prepared as we described previously [8]. The high-coverage libraries of wild-type CaLCuV were sequenced on an Illumina Genome Analyzer (GA) Hi-Seq 2000 using a TruSeq v5 kit, while the low coverage libraries on a GA-II using Chrysalis v2. The libraries of CaLCuV::Chl and CaLCuV::GFP viruses were sequenced on a GA-IIx using Chrysalis v4 and TruSeq v5, respectively. After trimming the adaptor sequences, the datasets of reads were mapped to the reference genomes of Arabidopsis thaliana Col-0 (TAIR9), CaLCuV (U65529.2 for DNA-A and U65530.2 for DNA-B) and other references using a Burrows-Wheeler Alignment Tool (BWA version 0.5.9) [65] with zero mismatches to the reference sequence. The reference sequences of CaLCuV DNA-A and its derivatives, CaLCuV DNA-B, L2 T-DNA and ChlI/CH42 and ChlI-2 genomic loci are given in Protocol S1. Reads mapping to several positions on the references were attributed at random to one of them. To account for the circular virus genome the first 50 bases of the viral sequence were added to its 3′-end. For each reference genome/sequence and each sRNA size-class (20 to 25 nt), we counted total number of reads, reads in forward and reverse orientation, and reads starting with A, C, G and T (Table S1). In the single-base resolution maps of 20, 21, 22, 23, 24 and 25 nt vsRNA (Tables S2, S3, S4, S5, S6 and S7), for each position on the sequence (starting from the 5′ end of the reference sequence), the number of matches starting at this position in forward (first base of the read) and reverse (last base of the read) orientation for each read length is given. Note that the reads mapped to the last 50 bases of the extended viral sequence were added to the reads mapped to the first 50 bases.
The detailed protocol for high-resolution analysis of long RNA using total RNA and 5% PAGE followed by blot hybridization was described previously [30]. To detect the viral mRNAs AV1, AC2/AC3, BV1 and BC1 (Figure 3A), the membrane was successively hybridized with mixtures of DNA oligonucleotides complementary to each given mRNA (for sequences, see Protocol S1).
Southern blot analysis was performed as in [66]. In short, total DNA from the plants were extracted by CTAB-based protocol. Five µg of total DNA was electrophoresed in 1% agarose gel prepared in 1× Tris-sodium acetate-EDTA buffer. Full-length linear DNA of CaLCuV was loaded as a positive control for Southern hybridization. After EtBr staining, the DNA in the gel was alkali-denatured and transferred to the Hybond N+ nylon membrane (GE healthcare lifesciences). PCR fragments of DNA-A (900 bp obtained with the primers Cb_AV1_qPCR_s and Cb_AC3_qPCR_as) and DNA-B (862 bp Cb_BV1_qPCR_s and Cb_BC1_qPCR_as), which do not contain the common region of the virus, were labeled with [α-32P]dCTP using Rediprime II DNA labeling system (GE healthcare lifesciences) and used as probes. Hybridization with the labeled probe was performed at 65°C for 16–20 hours using PerfectHyb Plus Hybridization Buffer (Sigma-Aldrich) and the membrane was washed thrice at 65°C with 2× SSC/0.5% SDS. The signal was detected after 5 days exposure to a phosphor screen using a Molecular Imager (Typhoon FLA 7000, GE healthcare lifesciences).
Relative accumulation of polyadenylated viral mRNAs and total viral DNA in wild type versus rdr1/2/6 (Figure 3D) was measured using real time PCR as in [8]. For polyadenylated RNA, cDNA was synthesized from 5 µg of total RNA using 100 pmoles of oligo d(T)16 primer. The RNA-primer mixture was heated to 70°C for 10 min and chilled on ice for 5 min. 4 µl of 5× first-strand synthesis buffer (250 mM Tris-HCl [pH 8.3], 375 mM KCl, 15 mM MgCl2, 0.1 M DTT), 2 µl 0.1 M DTT, 1 µl 10 mM deoxynucleoside triphosphate mix and 1 µl (200 U) of Superscript III reverse transcriptase (Invitrogen) were added and incubated at 50°C for 60 min. The reaction was stopped by heating the mixture to 95°C for 5 min. 2 µl of the 10 times diluted reverse transcription reaction mix or 2 µl of total DNA (2 ng) were taken for PCR in LightCycler 480 Real-Time PCR System (Roche applied sciences) using FastStart Universal SYBR Green Master (Rox) mix (Roche) and primers designed using Beacon designer 2 software (PREMIER Biosoft International). PCR primers specific for viral DNAs A and B and each viral mRNA as well as internal controls (18S rDNA and ACT2 mRNA) are given in Protocol S1. Cycling parameters were 95°C for 10 min, followed by 45 cycles: 95°C for 10 s, 56°C for 10 s, 72°C for 20 s. Amplification efficiency of primers was determined by means of a calibration curve (Ct value vs. log of input cDNA/DNA) prepared in triplicate. The Ct values obtained for viral genes were normalized with internal control values and the delta Ct values were obtained. The normalized values for CaLuCV-infected wild type Col-0 were set to 1. To quantify the L2 GFP mRNA levels, poly-dT cDNAs were made as described above. Real-time PCR was performed in 96-well titer plates on an ABI PRISM 7000 SDS apparatus with SYBR GREEN PCR Master Mix (ABI) following manufacturers' recommendations (95°C for 5 min., followed by 40 cycles: 95°C for 30 s, 60°C for 45 s). Primers are given in Protocol S1. Uncertainties were propagated from standard errors for triplicate measurements of cDNA pools (derived from column-purified RNA of 3–4 plants).
|
10.1371/journal.pgen.1005328 | Context-Dependent Functional Divergence of the Notch Ligands DLL1 and DLL4 In Vivo | Notch signalling is a fundamental pathway that shapes the developing embryo and sustains adult tissues by direct communication between ligand and receptor molecules on adjacent cells. Among the ligands are two Delta paralogues, DLL1 and DLL4, that are conserved in mammals and share a similar structure and sequence. They activate the Notch receptor partly in overlapping expression domains where they fulfil redundant functions in some processes (e.g. maintenance of the crypt cell progenitor pool). In other processes, however, they appear to act differently (e.g. maintenance of foetal arterial identity) raising the questions of how similar DLL1 and DLL4 really are and which mechanism causes the apparent context-dependent divergence. By analysing mice that conditionally overexpress DLL1 or DLL4 from the same genomic locus (Hprt) and mice that express DLL4 instead of DLL1 from the endogenous Dll1 locus (Dll1Dll4ki), we found functional differences that are tissue-specific: while DLL1 and DLL4 act redundantly during the maintenance of retinal progenitors, their function varies in the presomitic mesoderm (PSM) where somites form in a Notch-dependent process. In the anterior PSM, every cell expresses both Notch receptors and ligands, and DLL1 is the only activator of Notch while DLL4 is not endogenously expressed. Transgenic DLL4 cannot replace DLL1 during somitogenesis and in heterozygous Dll1Dll4ki/+ mice, the Dll1Dll4ki allele causes a dominant segmentation phenotype. Testing several aspects of the complex Notch signalling system in vitro, we found that both ligands have a similar trans-activation potential but that only DLL4 is an efficient cis-inhibitor of Notch signalling, causing a reduced net activation of Notch. These differential cis-inhibitory properties are likely to contribute to the functional divergence of DLL1 and DLL4.
| Notch signalling relies on binding of a ligand to a Notch receptor, both residing on the surfaces of neighbouring cells. This interaction forwards a signal into the receptor-expressing cell, this way coordinating cells in many biological processes such as the segmentation of the axial skeleton. Mammals possess four Notch-activating ligands–including DLL1 and DLL4 -expressed in diverse, partially overlapping regions. Whether the different ligands trigger quantitatively or qualitatively distinct Notch responses is largely unknown. In order to directly compare both ligands we generated transgenic mice that express DLL1 or DLL4 in identical patterns. These mice uncover that only DLL1 but not DLL4 can mediate regular segmentation of the embryo. In experiments with cultured cells expressing either ligand and Notch, we found that the functional difference observed is unlikely to depend on differences in the activation of Notch. Rather, the unsuspected but strong difference between both ligands in cis-inhibition, i.e. repression of Notch by a ligand expressed in the same cell as the receptor, a process described in the fruitfly but not in mammals and not for DLL4 provides a possible explanation for the divergence in tissues that coexpress ligand and receptor.
| The Notch signalling pathway mediates local interactions between adjacent cells and thereby regulates numerous developmental processes in a wide variety of different tissues throughout the animal kingdom [reviewed in 1–7]. The Notch gene of Drosophila and its vertebrate homologues encode large transmembrane proteins that act as receptors at the surface of the cell. They interact with transmembrane ligand proteins on the surface of neighbouring, signal-sending cells (i.e. in trans) encoded by the Delta and Serrate (called Jagged in vertebrates) genes. Upon ligand binding, the intracellular domain of Notch (NICD) is proteolytically released, translocates to the nucleus, interacts with the transcriptional regulator Suppressor of Hairless ([Su(H)]; CSL proteins in vertebrates) and activates the transcription of downstream target genes [8–14]. Ligands coexpressed with the Notch receptor in signal-receiving cells (i.e. in cis) are capable of interacting with Notch and attenuate the signal strength [15–17, reviewed in 18].
Vertebrates possess several Notch receptors and ligands. The mouse genome encodes four Notch (NOTCH1–4), three Delta (DLL1, DLL3 and DLL4) and two Jagged (JAG1 and JAG2) proteins. Among the DLL proteins, only DLL1 and DLL4 function as Notch-activating ligands [19–21]. As paralogues, DLL1 and DLL4 are similar in sequence (47% identical plus 14% similar amino acids), size and domain structure [22]. Both contain a DSL domain, which is essential for the interaction with Notch [23,24], as well as eight EGF-like repeats in their extracellular domain and have a short intracellular domain with a C-terminal PDZ binding motif. Dll1 and Dll4 are expressed both in discrete and overlapping patterns during embryonic development and in adult tissues of the mouse. In shared expression domains, the two ligands have redundant or different functions depending on the developmental context. An example for full redundancy is the maintenance of the crypt progenitor pool in the adult small intestine. Dll1 and Dll4 are coexpressed in crypt cells [25,26] and individual inactivation of either ligand has no effect on the crypt progenitor cell pool. However, simultaneous deletion of Dll1 and Dll4 leads to a complete loss of the proliferative crypt compartment and intestinal stem cells [27].
Conversely, in foetal arteries where both ligands are expressed in the vascular endothelium [26,28,29] inactivation of Dll1 causes loss of NOTCH1 activation despite the presence of DLL4 [29] suggesting that DLL4 cannot compensate for the loss of DLL1 in fetal endothelial cells. In the adult thymus, Dll1 and Dll4 are both expressed in thymic epithelial cells [26,30]. Here, DLL4 is the essential Notch ligand required for T-lymphopoiesis [31] and T cell development is unaltered in mice lacking DLL1 in the thymic epithelium [32] suggesting that in this context DLL1 and DLL4 are functionally distinct. This conclusion is supported by in vitro studies showing that DLL1 and DLL4 differ with respect to their binding avidity to Notch receptors on thymocytes and to the steady-state cell surface levels required to induce T cell development, DLL4 being the more effective ligand [33,34] as well as by biochemical studies indicating a 10-fold higher Notch binding affinity of DLL4 than DLL1 [19]. Furthermore, DLL4 but not DLL1 can induce a fate switch in skeletal myoblasts and induce pericyte markers [35]. Collectively, these individual reports of context-dependent redundant and distinct functions of coexpressed DLL1 and DLL4 raise the questions of why DLL1 and DLL4 act equally in some processes but differently in others, which mechanism or factor causes their function to vary and whether they are similar enough to replace each other in domains where only one of both DLL ligands is endogenously expressed.
In early mouse embryos, expression of Dll1 and Dll4 is largely non-overlapping. Dll1 is expressed in the paraxial mesoderm beginning at E7.5, in the central nervous system from E9 onwards and later on, at E13.5, in arterial endothelial cells [29,36]. Deletion of Dll1 disrupts somite patterning and causes premature myogenic differentiation, severe haemorrhages and embryonic death after E11 [37,38]. Dll4 is expressed in the vascular endothelium of arteries beginning at E8 [39] but not in the somite-generating presomitic mesoderm, somites or differentiating myoblasts. Inactivation of DLL4 results in severe vascular defects leading to embryonic death prior to E10.5 [39,40].
Here, we address the functional equivalence of DLL1 and DLL4 in vivo and in vitro. We analyse Notch signalling in mice that conditionally overexpress DLL1 or DLL4 on a Dll1 null genetic background and in mice in which Dll1 is replaced by Dll4, focussing on young embryos in which both Notch ligands have discrete endogenous expression domains. We show that DLL4 cannot replace DLL1 during somite segmentation but can partially replace DLL1 during myogenesis and fully replace DLL1 during maintenance of retinal progenitors. Cell culture assays that measure Notch activation by DLL1 or DLL4 demonstrate that DLL4 trans-activates Notch signalling similarly to DLL1 but cis-inhibits Notch signalling much more efficiently than DLL1, partly overruling the activation by interactions in trans. Consistent with these in vitro data, we observe dominant effects on segmentation by DLL4 ectopically expressed in the presomitic mesoderm (PSM). We propose that differential Notch cis-inhibition by DLL1 and DLL4 contributes to the observed tissue-dependent functional divergence of both paralogues, perhaps in combination with other factors not tested in this study.
In order to directly compare the activities of DLL1 and DLL4 in vivo, we generated mice that conditionally express either Dll1 or Dll4 under the CAG promoter from a single-copy transgene insertion in the same genomic locus. We employed an established system for integration of Cre-inducible expression constructs into the Hprt locus, the pMP8.CAG-Stop vector (Fig 1A; [41,42]). The unrecombined pMP8.CAG-Stop construct expresses neomycin phosphotransferase (neor) from the CAG promoter. Cre-mediated recombination of two loxP sites and two mutant loxP2272 (loxM) sites [43] flips the gene of interest and excises neor so that the recombined construct expresses the gene of interest from the CAG promoter. 5’ and 3’ homology regions from the Hprt gene enable homologous recombination of pMP8 constructs into the Hprt locus [44]. We cloned the Dll1 and Dll4 open reading frames into the pMP8.CAG-Stop vector, introduced both unrecombined (i.e. neor expressing) constructs into Hprt-deficient E14TG2a ES cells and used homologous recombinant clones to produce transgenic mice with Cre-inducible Dll1 or Dll4 (alleles termed CAG:Dll1 and CAG:Dll4).
To check activation of DLL1 and DLL4 in embryos, we induced ubiquitous expression of the CAG:Dll1 and CAG:Dll4 transgene by mating our mice with mice carrying a ZP3:Cre transgene that causes site-specific recombination during oogenesis [45]. We crossed CAG:Dll1;ZP3:Cre and CAG:Dll4;ZP3:Cre females with wildtype males to obtain embryos that overexpress Dll1 or Dll4 from the zygote stage on. The transgenes are transcribed bicistronically with an IRES-Venus (Fig 1A) whose expression marks cells in which Cre-recombination activated the transgene. As Hprt is located on the X chromosome, hemizygous male embryos expressed Venus ubiquitously whereas heterozygous female embryos showed mosaic expression due to random X-inactivation (Fig 1B). Analysis of embryo lysates on a Western blot with anti-GFP antibodies demonstrated CAG:Dll1 and -4 transgene activation at similar levels (Fig 1C; S1A Fig; S1 Table). To directly compare DLL1 and DLL4 protein levels, we generated embryonic stem cells expressing HA-tagged Dll1 or Dll4 from single copy insertions of in the Hprt locus. Western blot analysis of three independent clones for each ligand confirmed similar protein levels in all clones (average DLL1-HA, 1.23±0.33; average DLL4-HA, 0.88±0.53; Fig 1D; S1B Fig; S2 Table).
In order to test whether DLL4 can compensate for the loss of DLL1 in mesodermal tissues of early embryos, we mated mice to combine three different transgenes: a) CAG:Dll1 or -4 inducible transgenes; b) two (i.e. homozygous) floxed alleles of endogenous Dll1 [Dll1loxP/loxP; 32] that get inactive upon recombination; and c) a Cre transgene expressed in the primitive streak driven by a promoter derived from brachyury [T(s):Cre; 46]. In offspring bearing the complete set of transgenes (identified by genotyping PCR; see Material and Methods), recombination by T(s):Cre simultaneously inactivates endogenous Dll1 and activates CAG:DLL1 or CAG:DLL4 expression in all mesoderm-derived tissues.
As expected, inactivation of Dll1 throughout the mesoderm resulted in severe somite patterning defects characterised by loss of Uncx4.1 expression (Fig 1Ed), a marker for caudal somite compartments [47,48] whose expression depends on Notch activation [46]. Expression of CAG:DLL1 in such Dll1-deficient embryos restored robust, largely regularly striped expression of Uncx4.1, which expanded into cranial somite compartments in most axial regions and particularly in hemizygous male embryos (Fig 1Eb and 1Ec). This rescue of somitogenesis demonstrates that expression of CAG:DLL1 (from the Hprt locus) is sufficient to substitute for the loss of endogenous DLL1; cranial expansion of Uncx4.1 is reminiscent of ectopic Notch activity [46].
In contrast, expression of CAG:DLL4 in Dll1-deficient embryos restored only very weak and irregular expression of Uncx4.1 and resembles Dll1loxP/loxP;T(s):Cre embryos without CAG:DLL1 overexpression (Fig 1Ee and 1Ef). Only two out of 16 embryos of this genotype displayed regular Uncx4.1 expression in the cranial most somites, which might reflect residual DLL1 activity perhaps due to delayed excision of endogenous Dll1 (Fig 1Eg). The extensively defective segmentation in Dll1loxP/loxP;T(s):Cre embryos with CAG:DLL4 overexpression directly shows a functional difference between DLL1 and DLL4 during early embryogenesis: DLL4 is not able to take over DLL1 function in the paraxial mesoderm during somite formation. Weak and irregular Uncx4.1 expression in some of these embryos suggest Notch activation at low levels.
To further investigate to which degree DLL4 can compensate for the loss of DLL1 during somite patterning and in other developmental contexts, we generated mice that express DLL4 from the Dll1 locus instead of endogenous DLL1. To replace endogenous Dll1 with Dll4, we applied a knock-in strategy inserting a Dll4 mini gene into the first and second exons of Dll1 (Fig 2A). Production of DLL4 protein of the correct size from the Dll4 mini gene was confirmed by Western blot analysis of lysates of CHO cells transiently expressing the Dll4 mini gene (S2 Fig). We generated mice carrying the Dll4 mini gene in the Dll1 locus, referred to as Dll1Dll4ki. As a control, we used the analogous knock-in of a Dll1 mini gene into the Dll1 locus (Fig 2A bottom; Dll1tm2Gos, here referred to as Dll1Dll1ki), which was identical to the Dll4 mini gene with regard to its exon/intron structure, intron sequences and the 5' and 3' UTRs but encoded DLL1. Homozygous Dll1Dll1ki mice were viable and fertile and appeared phenotypically normal indicating that the Dll1 mini gene can functionally substitute the endogenous Dll1 gene [37].
Heterozygous Dll1Dll4ki/+ mice (containing one endogenous copy of Dll1 and one copy of Dll4ki) were viable and fertile and showed no obvious phenotype except for kinky tails (Fig 2Ba–2Bc, arrow; penetrance 89%; n = 48), a phenotype indicative of irregular somitogenesis rarely observed in Dll1Dll1ki homozygotes or Dll1 null (Dll1lacZ) heterozygotes (penetrance 15%; n = 23 and 53, respectively; Fig 2Ba’–2Bc’). In contrast to homozygous Dll1Dll1ki, no homozygous Dll1Dll4ki mice were obtained after birth. At E15.5, Dll1Dll4ki homozygotes exhibited short body axes, truncated tails and were oedematic (Fig 2C; arrow points at tip of tail) resembling foetuses with severely reduced DLL1 function [37].
Correct expression of Dll4 in the presomitic mesoderm (PSM) of Dll1Dll4ki embryos was confirmed by in situ hybridisation using probes specific for the Dll4 ORF or the 3‘UTR (Dll1 exon 11) common to Dll1Dll1ki and Dll1Dll4ki alleles (Fig 2Db, 2Dc and 2Df, black arrowheads). In situ hybridisation with a specific Dll1 ORF probe confirmed the absence of Dll1 transcripts in Dll1Dll4ki homozygotes (Fig 2Dj, red arrowhead). Homozygous Dll1Dll4ki embryos showed strong expression of Dll4 in the neural tube (Fig 2Dc, white arrow), reflecting activation of the Dll1 promoter in this region [50,51]. Northern blot analysis of Dll1Dll1ki and Dll1Dll4ki homozygous embryos indicated equal levels of transcription of the transgenes (Fig 2E). In Dll1Dll4ki/Dll4ki embryos, ectopic DLL4 protein was detected at the plasma membrane of PSM cells (Fig 2Fa–2Fc). Likewise, DLL1 protein was detected at the surface of PSM cells in homozygous Dll1Dll1ki embryos (Fig 2Fj–2Fl), confirming that DLL4 and DLL1 protein is generated from their mini genes and targeted to the plasma membrane in vivo. Taken together, these data show that Dll1Dll4ki mice indeed express Dll4 instead of Dll1 from the Dll1 locus at comparable levels and confirm our previous observation that DLL4 is unable to support proper mouse development in the absence of endogenous DLL1.
Cranial-caudal somite patterning critically depends on DLL1-mediated Notch signalling [38,46,52]. We analysed if DLL4 can functionally replace DLL1 in this process in homozygous Dll1Dll4ki embryos. Unlike embryos that contained at least one wildtype or Dll1Dll1ki allele, homozygous Dll1Dll4ki embryos displayed severely reduced and irregular Uncx4.1 expression (Fig 3A), which indicates disrupted somite patterning and reduced Notch activity in the PSM due to the inability of DLL4 to replace DLL1. Consistent with defective somite formation and the shortened body axis observed in E15.5 foetuses, Dll1Dll4ki/Dll4ki axial skeletons were severely disorganised (Fig 3B). Therefore, expression of DLL4 from the Dll1 locus does not cause a significant rescue of the Dll1 somitogenesis phenotype.
Remarkably, as anticipated by the kinky tail phenotype of heterozygous Dll1Dll4ki/+ adults (Fig 2B), E18.5 Dll1Dll4ki/+ skeletons reveal fusions of dorsal ribs and malformations of individual vertebrae in various regions of the vertebral column (e.g. Fig 3Bd, red arrowheads; 8/12 E18.5 Dll1Dll4ki/+ skeletons displayed apparently irregular vertebrae). Additional examination of seven Dll1Dll4ki/+ adult skeletons uncovered fused ribs and/or irregular segments in the tail of all preparations (S3 Fig). A single Dll1 allele is sufficient to support regular segmentation and Dll1lacZ/+ mice form essentially normal skeletons [53]. Our consistent finding of skeletal irregularities in Dll1Dll4ki/+ mice indicates subtle disturbances during segmentation and suggests that the Dll1Dll4ki allele has a dominant effect on segmentation.
Processes other than somitogenesis in the developing embryo that depend on DLL1–Notch signalling include myogenesis [37] and retinal development [51]. Embryos lacking DLL1 display excessive differentiation of myoblasts, which exhausts the progenitor pool and leads to severely reduced or absent skeletal muscles [37]. Homozygous E9.5 Dll1Dll4ki embryos showed transient upregulation of the myocyte marker Myogenin [54] as also observed in homozygous Dll1lacZ embryos (Fig 3C, arrowheads; [37]). At E15.5, they had significantly less skeletal muscle tissue than wildtype or homozygous Dll1Dll1ki foetuses but clearly more skeletal muscle tissue than Dll1 null mutants (Dll1lacZ) as shown for the intercostal muscles, the diaphragm and forelimbs by anti-MHC antibody staining of sectioned foetuses (Fig 3D–3F, arrowheads). These results indicate that DLL4 can partially substitute DLL1 during muscle cell differentiation and Dll1Dll4ki behaves like a hypomorphic Dll1 allele.
In the embryonic neural retina, Dll1 and Dll4 are sequentially expressed and can both function to maintain proliferating progenitors, while they have different functions in retinal fate diversification [51,55]. In contrast to myogenesis, DLL4 can fully replace DLL1 function in maintaining neuronal progenitors in the embryonic retina. Whereas Dll1 mutants show a striking disruption of the retinal neuroepithelium with formation of rosettes (Fig 4A), due to premature differentiation of retinal progenitors [51], both Dll1Dll1ki/Dll1ki and Dll1Dll4ki/Dll4ki retinas have a normal neuroepithelial organisation with a clear stratification of Chx10+ progenitors and p27+ differentiating neurons (Fig 4B). Moreover, we find that similar numbers of early born retinal neurons [retinal ganglion cells (RGCs) and amacrine cells] are present in Dll1Dll1ki and Dll1Dll4ki retinas (Fig 4C and 4D; n≥4 retinal sections), confirming that DLL1 and DLL4 functions are interchangeable in regulating early retinal neurogenesis. We have further analysed DLL4 expression in Dll1Dll4ki/Dll4ki retinas and found it recapitulates the broader Dll1 expression pattern, with the transgenic protein expressed at similar levels as endogenous DLL4 in the retinal neuroepithelium (compare Fig 4Ea–4Ec with 4Eb–4Ed). Together, these results offer further evidence that the Dll4 transgene is fully functional in Dll1Dll4ki/Dll4ki embryos. The extent of the functional equivalence of DLL1 and DLL4 depends on the developmental context.
To investigate the functional difference between DLL1 and DLL4 in vitro, we performed co-culture experiments by mixing cells expressing NOTCH1 receptor or DLL ligands and measured Notch activation with a reporter in the receptor-expressing cells. Specifically, we used HeLa cells that express both the NOTCH1 receptor (stable HeLa-N1 cells; [10]) and a transient Notch activity reporter based on an RBP-Jk promoter-driven Luciferase [56] with CHO cells stably expressing Flag-tagged DLL1 or DLL4 ligands. To ensure comparability of results, we integrated single copies of Dll1 or Dll4 ORFs under the control of the CMV promoter into the identical genomic locus of CHO cells by adopting a site-directed attP/attB recombination system (Fig 5A top; S4 Fig; [57]). We established CHO cells with a pre-inserted, randomly integrated single attP site (termed CHOattP; uniqueness of this attP site was confirmed by Southern blot analysis; S4A and S4B Fig) and recombined Dll1 or Dll4 ORFs into this site (cell lines termed CHOattP-DLL1 and CHOattP-DLL4; Fig 5A bottom left). Consistent with the expression from the same genomic locus, independent CHOattP-DLL1 (B5, C6) and CHOattP-DLL4 (B5, D3) clones expressed DLL1 and DLL4 protein at similar levels (Fig 5B; n = 4 lysates of each clone; S5A Fig, S3 Table, S5B Fig, S4 Table) and cell surface representation of DLL1 and DLL4 was similar in all lines (~40%; Fig 5C; n≥3 biotinylation assays; S5C and S5D Fig, S5 Table). Likewise, half-lives of DLL1 and DLL4 proteins were similar, DLL4 being slightly more stable (S5E and S5F Fig).
Co-culture of HeLa-N1 with either CHOattP-DLL1 or CHOattP-DLL4 (schematically shown in Fig 5A bottom) led to a >10-fold increase of Notch activity as compared to co-cultures of HeLa-N1 with CHOattP cells that did not express transgenic DLL1 or DLL4 (Fig 5D; n = 3) confirming that all transgenes were functional. DLL4 trended to activate Notch more strongly than DLL1 (including clone CHOattP-DLL4 B5 whose protein level was slightly reduced in Fig 5B); the difference between individual clones was not statistically significant in these experiments and partly significant in similar experiments with other clones (S6A and S6G Fig).
Next, we tested whether coexpression of further factors (LFNG, JAG1) in our cell culture system differently alters Notch activation by DLL1 or DLL4 and thereby provides a plausible explanation for the distinct phenotypes. The glycosyltransferase LUNATIC FRINGE (LFNG), which is expressed in the PSM, is able to modify NOTCH in the trans-Golgi [58,59] and thereby modulates receptor activation. The Notch ligand JAG1 is expressed in forming somites [60,61] and can act as a competitive inhibitor of DLL ligands [62,63]. We performed co-culture assays with HeLa-N1 cells transiently expressing LFNG-HA (S6A–S6C Fig) or with CHOattP-DLL1 and CHOattP-DLL4 cells coexpressing JAG1 (S6D–S6F Fig) and found no statistically significant changes in Notch activation.
Also, different glycosylation patterns of the ligands’ extracellular domain could contribute to differences in their activity. To test this possible influence, we treated co-cultures with tunicamycin to prevent N-glycosylation. Blocking N-glycosylation reduced the activity of DLL4 in cultured cells significantly, but not below DLL1 activity (S6G and S6H Fig), suggesting that distinct N-glycosylation is an unlikely cause for the observed differences between both ligands. Collectively, our results do not reveal a difference in the trans-activation potential of DLL1 and DLL4 that could explain the different segmentation phenotypes of our transgenic DLL1- or DLL4-expressing mice.
We modified the co-culture assay by (transiently) expressing the ligands in the HeLa-N1 cells instead of in the CHO cells (Fig 5E, S7 Fig). In this setting, DLL ligands (expressed in HeLa-N1 cells) can trans-activate Notch in neighboring HeLa-N1+DLL cells (schematically shown in Fig 5Ec or in detail in S7A Fig); in addition, they can interact with Notch expressed in the same cell, i.e. in cis. When co-culturing HeLa-N1 cells expressing DLL1 with empty CHO cells (Fig 5Ea, S7 Fig), activation of Notch signalling was significantly increased as compared to a co-culture of HeLa-N1 cells expressing no transgenic DLL ligand with empty CHO cells (Fig 5Ea’; compare light grey bar “+DLL1” and white bar; n = 6; numbers are normalised to white bar). Intriguingly, in co-cultures of HeLa-N1 cells expressing DLL4 (with empty CHO cells), Notch activation was significantly lower than with DLL1 (Fig 5Ea’; compare dark grey bar “+DLL4” and light grey bar “+DLL1”; n = 6). Given the similar trans-activation potential of DLL1 and DLL4 (Fig 5D; S6A and S6G Fig), a likely explanation for the different levels of Notch activation is a higher cis-inhibitory potential of DLL4 than of DLL1.
In order to facilitate the analysis of cis-inhibition, we repeated the experiments shown in Fig 5Ea and 5Ea’ with the only modification of expressing a DLL ligand in the CHO cells (CHOattP-DLL1) so as to enhance the level of Notch activation (Fig 5Eb and 5Eb’). In the experiments both without any transgenic DLL ligand in HeLa-N1 and with DLL1 in HeLa-N1, co-culture with CHOattP-DLL1 cells caused a >10-fold increase of Notch activation. In the HeLa-N1 cells with DLL4, Notch activation was significantly less, i.e. about 5-fold increased (Fig 5Eb; n = 6; all numbers in Fig 5Ea’ and 5Eb’ are normalised to the left bar in a’, i.e. HeLa-N1 without transgenic DLL co-cultured with empty CHO cells, set to 1). These results support cis-inhibition of Notch by DLL4 resulting in a strong reduction of net Notch activation. In contrast, DLL1 does not cis-inhibit Notch in this assay (no significant difference between white and light grey bar in Fig 5Eb’).
To approximate the setting of the embryonic cranial PSM in which every cell expresses both DLL1 and NOTCH1 [64], we also analysed pure cultures of HeLa-N1 cells expressing either no transgenic DLL or (transient) DLL1 or DLL4 (Fig 5Ec). Expression of DLL1 enhanced Notch activation ~15-fold whereas expression of DLL4 increased Notch activation only <5-fold which was not significantly different from HeLa-N1 cells without transgenic ligand (Fig 5Ec’; n = 6; numbers in Fig 5Ec’ are normalised to the left bar in 5Ec, i.e. culture of HeLa-N1 without transgenic DLL, set to 1). These data show that cis-inhibition by DLL4 partially overrides trans-activation, and reduces Notch activation to <30% in an in vitro setting modeling the arrangement of ligand and receptor molecules in the PSM.
A conceivable alternative explanation of our in vitro results, which show attenuated Notch signalling when NOTCH and DLL4 are coexpressed, could be a reciprocal mechanism, i.e. cis-inhibition of DLL4 by NOTCH1 [18]. To test this possibility, we modified our first Notch activation assay (Fig 5A bottom, 5D) by transiently coexpressing NOTCH1 (NOTCH1deltaC, see Methods) in CHOattP-DLL1 and CHOattP-DLL4 cells. In co-cultures with HeLa-N1 cells containing the reporter (Fig 5Fa and 5Fb, S8 Fig), both ligands activated NOTCH in HeLa-N1 >10-fold irrespective of the presence of NOTCH1 in the CHO cells (Fig 5Fa’ and 5Fb’; n = 3) and we measured no significant difference between trans-activation by DLL1 or DLL4 as before (Fig 5D). These data indicate that NOTCH1 does not cis-inhibit its ligands.
In summary, our cis-inhibition assays (Fig 5E) reveal a functional difference between DLL1 and DLL4 that was not evident in the trans-activation assays (Fig 5D and 5F): DLL4, but not DLL1, is a potent cis-inhibitor of NOTCH1 and cis-inhibition by DLL4 can significantly reduce Notch activation. Our in vitro results are consistent with our in vivo data: they can explain both why DLL4 appears to be a weaker activator of Notch signalling than DLL1 during somitogenesis in our transgenic mice and why transgenic DLL4 has a dominant effect on segmentation in Dll1Dll4ki/+ mice (see Discussion and Fig 6). We propose that in the PSM, DLL1 is a more efficient net activator of Notch than (ectopic) DLL4 because it does not efficiently cis-inhibit Notch.
In order to identify the protein domain that mediates cis-inhibition by DLL4, we cloned chimeric Dll1 and Dll4 ORFs by swapping extracellular domains (resulting in DLL4-ICD+TM/DLL1-ECD, termed DLL4-DLL1ECD, or DLL1-ICD+TM/DLL4-ECD, termed DLL1-DLL4ECD; ICD, intracellular domain; TM, transmembrane domain; ECD, extracellular domain; Fig 5G top). We introduced the chimeric Dll1-4 ORFs transiently into HeLa-N1 cells and performed co-culture assays analogous to the cis-inhibition experiments with non-chimeric DLL1 and DLL4 shown in Fig 5E (Fig 5G; S9 Fig). Measurement of Notch activity showed similarity between DLL1 and DLL4-DLL1ECD as well as between DLL4 and DLL1-DLL4ECD (Fig 5Ga’, 5Gb’ and 5Gc’; n = 6; compare with Fig 5Ea’, 5Eb’ and 5Ec’). Particularly the statistically significant differences between bars in Fig 5Gb’ clearly indicate that DLL4-DLL1ECD enhances, but DLL1-DLL4ECD reduces Notch activation by DLL1. As both chimeric ligands localise to the cell surface (S9B and S9C Fig; S6 Table) and are able to trans-activate Notch in a range similar to DLL1 and DLL4 (S9D Fig) these results show that cis-inhibition is mediated by the extracellular domain of DLL4. This observation is consistent with studies that showed that the DSL domain as well as EGF repeats 4–6 of Serrate are essential for cis-inhibition in Drosophila although these EGF repeats are not well conserved between Serrate and Delta ligands [24,65,66]. Analysis of Notch activation by chimeric proteins in which smaller domains of the extracellular regions are swapped will help to precisely map the cis-inhibitory domain in DLL4.
The presence of several Notch receptors and ligands in mammals offers a multitude of possible receptor-ligand interactions; whether different combinations of receptor and ligand qualitatively or quantitatively vary in their signalling output is largely unknown. In this study, we focus on the mouse Notch ligands DLL1 and DLL4 and find functional differences in vivo, which are particularly apparent in the PSM: DLL4 cannot replace DLL1 during axial segmentation, and a striking dominant segmentation phenotype in Dll1Dll4ki/+ mice hints towards an inhibitory function of ectopically expressed DLL4 in the PSM. We examined the possibility that differential cis-inhibition contributes to the phenotype and our in vitro Notch activation data are indeed consistent with this possibility (Fig 6), but do not exclude that other factors may contribute (see below).
Mesodermal expression of DLL1 and DLL4 from the Hprt locus on a Dll1 mutant background caused different phenotypes providing first hints that DLL1 and DLL4 are functionally different during early embryogenesis: CAG:DLL1 largely rescued the somitogenesis defects (Fig 1Eb and 1Ec) indicating that expression from this heterologous locus is strong enough to rescue the Dll1 null segmentation phenotype. In contrast, CAG:DLL4 expressed from the same locus failed to sufficiently activate Notch (Fig 1Ee–1Eg). Our Dll1Dll4ki knock-in data independently confirm and extend the CAG:Dll4 expression data, corroborating the inability of DLL4 to substitute for DLL1 function in the PSM (compare Figs 1Ee, 1Ef and 3Ae). We also show that the level of redundancy depends on the developmental process: there is essentially no redundancy during segmentation (Fig 3Ac, 3Ae, 3Bc and 3Be), partial redundancy in myoblast differentiation (Fig 3C–3F) and full redundancy in retinal progenitor maintenance (Fig 4). The effects on myogenesis and retinal development confirm that functional DLL4 is expressed from the Dll1Dll4ki allele (Figs 3De, 3Ee, 3Fe and 4B–4D).
Different protein levels of DLL1 and DLL4 are unlikely to account for the different phenotypes observed. Both proteins are expressed from identical genomic sites and the comparison of levels of bicistronic GFP (Fig 1C), transcripts (Fig 2E) and HA-tagged proteins (Fig 1D) confirm similar expression levels. Consistently, we find similar steady state levels, surface representation and half-lives of both ligands in CHO cells (Fig 5B and 5C; S5D–S5F Fig). Furthermore, immunohistochemistry using anti-DLL4 antibodies show similar levels of endogenous DLL4 and Dll1-driven DLL4ki expression in the retina (Fig 4E) as well as similar localisation of ectopic DLL4 and endogenous DLL1 at the cell surface within the PSM (Fig 2F). In our mouse models, we expressed untagged Dll1 and Dll4 transgenes to avoid alteration of protein function by the tag. As a consequence, we were unable to directly compare DLL1 and DLL4 levels in vivo and therefore cannot exclude small differences that may have contributed in part to the observed phenotype; strong differences are not indicated in the controls mentioned above.
Also, it is very unlikely that DLL1 or DLL4 have functions other than interacting with and activating Notch receptors. Although it has been previously suggested that the intracellular domain of DLL1 may influence gene transcription in the signal sending cell [67,68], we were unable to reproduce these in vitro results and showed that overexpression of the intracellular domain of DLL1 does not cause a phenotype in mice [42]. Collectively, the distinct ability to cis-inhibit Notch is a plausible explanation for the context-dependent DLL1-DLL4-divergence.
The ability of vertebrate DLL homologues to cis-inhibit Notch has been suggested before: overexpression of truncated DLL1 proteins lacking the intracellular domain in Xenopus, chicken and mouse embryos show dominant-negative effects on Notch signalling that are likely to be caused by cis-inhibition of Notch [53,69,70]. In primary human keratinocyte cultures, expression of DLL1 (and truncated DLL1T) renders cells unresponsive to Delta signals from neighbouring cells and controls differentiation of stem cells [71]. Our data show for the first time that DLL4 is a strong cis-inhibitor of Notch signalling, far stronger than DLL1. We have examined cis-inhibition in various types of cultures, in NOTCH- and DLL-expressing HeLa cells with and without co-culture of empty or DLL-expressing CHO cells and with chimeric DLL1-4 proteins (Fig 5E and 5G). Furthermore, we have tested cis-inhibition of DLL1 and DLL4 ligands by NOTCH1 (Fig 5F). All those assays consistently show a strong reduction of Notch signalling by DLL4 when coexpressed with NOTCH1.
In our assays, DLL1 had no obvious cis-inhibitory effect (Fig 5Eb’; n = 6), which differs from earlier reports showing that vertebrate DLL1 proteins can cis-inhibit NOTCH1 [20,72–74]. This is likely due to different assay conditions: in these previous studies, DLL1 was derived from different vertebrate species or differently tagged, or different cell systems or higher ligand concentrations were used. In studies in which cis-inhibition of Notch by Delta and Serrate was compared, Delta displayed a relatively weaker cis-inhibitory potential [75,76].
The ability for strong cis-inhibition resides in the extracellular domain of DLL4 (Fig 5G) that physically interacts with the Notch extracellular domain. Possible causes for the higher cis-inhibitory potency of DLL4 as compared to DLL1 include a potentially higher Notch cis-binding affinity of DLL4 as determined for the trans-interaction in vitro [34] or different glycosylation patterns in the extracellular domains of DLL1 and DLL4 (DLL4 contains an additional O-fucosylation site in EGF5 and four additional N-glycosylation sites, three of which reside in the N-terminal domain, which is essential for Notch activation; e.g. [19]; sites predicted by www.cbs.dtu.dk/services/NetOGlyc).
Our in vitro findings provide a possible explanation why DLL1 supports regular somite formation whereas DLL4 with its reduced net Notch activation potential is unable to do so. Heterozygous Dll1Dll4ki/+ mice consistently exhibit kinky tails and irregular vertebrae (Figs 2B and 3Bd; S3 Fig) despite the presence of one wildtype Dll1 allele, which should be able to support regular somitogenesis [53]. This finding strongly supports an in vivo inhibitory effect of DLL4 in the PSM, in which Dll4 is ectopically expressed at physiological levels (similar to the endogenous Dll1 levels; Fig 2E). Skeletal malformations observed in Dll1Dll4ki/+ mice are distinct from phenotypes observed upon mild overexpression of Dll1 in the paraxial mesoderm that include fused or split vertebral bodies and reduction of costal heads of ribs [77]. This supports the view that cis-inhibitory DLL4 acts in a dominant-negative manner partially overruling Notch activation by wildtype DLL1 causing axial skeleton defects in Dll1Dll4ki/+ mice, similar to the effect of a truncated dominant-negative form of DLL1 expressed in the paraxial mesoderm [53]. An alternative explanation for the dominant segmentation effect in heterozygous Dll1Dll4ki/+ mice could be a competition between DLL4 and DLL1 for NOTCH binding sites with DLL4 binding NOTCH more efficiently but activating it less efficiently than DLL1; although DLL1 has not been shown to be a more potent activator of NOTCH in vitro (Fig 5D; S6A, S6D and S6G Fig; [33,34]) we cannot exclude that this is the case in certain cellular contexts.
cis-Inhibition has been demonstrated to play a physiological role during fly development at the dorso-ventral border of the wing imaginal disc [15,16] and in photoreceptor precursors of the eye [17]. In vertebrates, the occurrence of cis-inhibition under physiological conditions is less clear but probable (see previous section). We did not observe apparent phenotypes in Dll1Dll4ki/+ mice that indicate dominant-negative effects of DLL4 outside the PSM. However, we hypothesise that cis-inhibition may occur in the foetal arterial endothelium, where DLL1, DLL4 and NOTCH1 are coexpressed and where loss of DLL1 abolishes NOTCH1 activation [29], possibly due to cis-inhibition by DLL4.
The PSM is particularly well suited to test the functionality of Notch ligands in vivo because DLL1 is the only activating ligand endogenously expressed in this tissue [78] and Dll4 mutants have no somitogenesis phenotype [39,40], so the analysis of Notch signalling is not complicated by the presence of several activators or confounded by composite phenotypes. However, two receptors, NOTCH1 and NOTCH2, are expressed in the PSM and may differ in their response to DLL1 or DLL4 binding.
The situation in myoblasts and other tissues is less clear. Outside the PSM, receptor and ligand expression typically exclude each other so that cis-inhibition can occur only during the short process in which the fate as receptor- or ligand-expressing cells is established [76,79,80]. That way, cis-inhibition may also be responsible for differences between DLL1 and DLL4 observed during myogenesis (Fig 3D–3F). Other reasons may contribute to or cause these differences: Firstly, further Notch receptors and ligands are expressed during myogenesis [81]. The contribution of individual Notch receptors to myogenesis is unknown but their function could vary [82]. Also, different ligands, including DLL1 and DLL4, have been shown to activate different Notch targets depending on the cell type in vitro [83,84]. A future thorough analysis of the functional divergence between DLL1 and DLL4 in myoblasts should aim at identifying the involved receptors and modulators (perhaps by in vitro analyses of myogenic or mesodermal progenitor cells including knock-down of individual factors) in order to understand the mechanisms underlying the observed phenotype.
Secondly, other processes may cause the divergence, e.g. modification of the ligands or receptors by glycosylation (may also play a role in the PSM). Activation of Notch by its ligands can be modulated by Fringe proteins. While glycosylation of Notch by LFNG enhances interaction with DLL1 in C2C12 cells [85] and with DLL4 in T cells in vitro [86], it appears to attenuate Notch signalling in the PSM [64,87]. However, we did not observe any shortcomings of DLL4 in the ability to trans-activate NOTCH1 compared to DLL1 when LFNG was present in the receptor-presenting cell (S6A–S6C Fig). The trans-activation potential of DLL1 and DLL4 could vary under certain conditions in vivo, perhaps depending on the glycosylation status, although our in vitro assays did not reveal any difference. Finally, the different extent of the functional difference between DLL1 and DLL4 observed in the PSM and during myogenesis may reflect the fact that mild changes of DLL1 activity affect the delicate Notch signalling in the PSM more readily than outside the PSM because somite patterning appears to be particularly sensitive to reduced Notch activity [88].
In conclusion, our genetic studies revealed a context-dependent functional divergence of the NOTCH ligands DLL1 and DLL4 in mice and provide a basis for a more extensive mechanistic analysis of this divergence in future studies. These will identify the relevant protein domain(s) and biochemical parameters and contribute to our understanding how different combinations of receptors and ligands determine the outcome of Notch signalling.
Statistical analyses were performed using Prism software (GraphPad). Luciferase measurements were analysed by one-way ANOVA and activities obtained with each protein were compared using Bonferoni’s Multiple Comparison Test with a significance level of 0.05. Means for all three DLL1-and DLL4-HA clones in Fig 1D, cell counts in the retina and cell surface levels of chimeric ligands were analysed using the Student’s t-test.
|
10.1371/journal.ppat.1001053 | HIV-1 Populations in Semen Arise through Multiple Mechanisms | HIV-1 is present in anatomical compartments and bodily fluids. Most transmissions occur through sexual acts, making virus in semen the proximal source in male donors. We find three distinct relationships in comparing viral RNA populations between blood and semen in men with chronic HIV-1 infection, and we propose that the viral populations in semen arise by multiple mechanisms including: direct import of virus, oligoclonal amplification within the seminal tract, or compartmentalization. In addition, we find significant enrichment of six out of nineteen cytokines and chemokines in semen of both HIV-infected and uninfected men, and another seven further enriched in infected individuals. The enrichment of cytokines involved in innate immunity in the seminal tract, complemented with chemokines in infected men, creates an environment conducive to T cell activation and viral replication. These studies define different relationships between virus in blood and semen that can significantly alter the composition of the viral population at the source that is most proximal to the transmitted virus.
| The work described in this report is directed at how HIV-1 viral RNA populations differ between the blood plasma and male genital tract in established infection. This site is of special interest since it is the proximal source of most transmissions of HIV-1. Thus, lessons learned about HIV-1 in the seminal tract are directly relevant to the mechanism of HIV-1 transmission. We have used single genome amplification to generate viral sequences from paired blood and semen samples in men with chronic HIV-1 infection. When compared to viral populations in blood plasma, we observe that virus in the seminal plasma can be equilibrated, clonally-amplified, or compartmentalized. We have also performed a characterization of the cytokine and chemokine milieu in these two compartments. We report a dramatic concentration of immune modulators in the seminal plasma relative to the blood, and these likely enhance the potential for viral replication in this compartment by creating an environment where target cells are kept in an activated state. These data define new and distinct features of virus:host interactions and represent a significant advance in our understanding of HIV-1 replication in the male genital tract.
| Sexual transmission of the human immunodeficiency virus type 1 (HIV-1) is the most common mode of transmission worldwide. During sexual transmission, genital secretions are the most proximal source of the transmitted virus. Thus, an understanding of the virus at these sites is central to understanding the transmission event and the nature of the transmitted virus. In this study we have explored the nature of viral populations in seminal plasma.
Virus enters the male genital tract during primary infection [1]–[5]. Initially, the virus found in the semen is similar, if not identical, to that found in the blood [6], [7]. During primary infection the viral RNA load is elevated in both the blood and the semen [1], [3]. The probability of transmission is related to the level of virus in the blood of the donor [8]–[11] and, based on a small cohort, to the level of virus in the semen [12]. Factors that induce inflammation in the seminal tract, such as sexually transmitted infections (STI), can raise the level of virus in semen [13], and this may contribute to the transmission of HIV-1 by the sexual route [14]. In addition, the endogenous semen-derived enhancer of virus infection (SEVI), a fragment of prostatic acid phosphatase, has been shown to increase infectious viral titers in vitro by several orders of magnitude [15].
The presence of virus in semen raises the possibility that virus found in semen could be the product of replication within the seminal tract. CD4+ T cells are found in semen indicating the presence of target cells that could support replication [16], [17]. SIV-infected macaques have infected cells within the tissues of the seminal tract [18], [19], supporting the possibility for local viral replication. Several studies have examined the relationship between viral populations found in blood and semen and noted differences (i.e. compartmentalization) using discordant drug resistance markers [20]–[24], differences in population markers [25], [26], or phylogenetic analysis [27]–[32].
In this study we have carried out a detailed examination of the viral populations in semen, comparing the env gene in blood plasma and seminal plasma. The men were therapy-naive and chronically infected with subtype C (n = 12) or subtype B (n = 4) HIV-1. We found a varied and complex relationship between these two compartments which suggests multiple types of biological phenomena. There is evidence for the direct import of virus from the blood to the semen, evidence for clonal amplification of a subset of genotypes within the seminal tract, and evidence for sustained replication and distinct evolution of virus within the seminal tract resulting in compartmentalization. The latter two of these phenomena result in seminal plasma viral populations that are distinct from those found in the blood, and thus distinct at the site proximal to transmission. Furthermore, semen is enriched in cytokines which may increase the potential for independent viral replication within the seminal tract.
In this study, we examined viral populations and cytokine/chemokine relationships in paired blood and semen samples collected from men with chronic HIV-1 infection in Lilongwe, Malawi (n = 12) [33], or from the CHAVI 001 clinical study (n = 4). We utilized a cohort of men without urethritis to minimize any potential confounders on viral loads, viral populations, or cytokine profiles. The clinical parameters of their HIV-1 infection are shown in Table 1. There was no evidence of urethritis in the dermatology clinic subjects, although the diagnosed cases of syphilis (1) and trichomonas (4) were treated with appropriate antibiotics (Table 1). In addition, blood and semen samples were obtained from twelve HIV-1-negative men without STIs from North Carolina and from 6 men from Malawi to serve as a control for the cytokine and chemokines analyses.
We used viral RNA extracted from blood plasma and seminal plasma to generate cDNAs to use as a template in the single genome amplification (SGA) protocol of the viral env gene [34]–[36]. The use of viral RNA allows for the examination of contemporaneously replicating virus. A mean of 27 amplicons were analyzed per sample, with a range of 15 to 34 in the blood and 14 to 44 amplicons in the seminal plasma. Sampling this number of genomes provides a 95% chance to detect a subpopulation in the range of 10–15%, and provides reasonable power to estimate the relative proportions of the major variants in the population [35]. The sequence of the entire env gene was determined for each amplicon. The use of SGA precludes PCR recombination as a source of confusion about the relationship between the viral genomes present in each sample [36].
Figure 1 depicts the phylogenetic analysis of viral sequences in the blood and semen for subjects C011 and C111. There was a diverse population in the blood and this diversity was fully represented in the semen. Furthermore, the complexity of the sequences in the blood, where no two sequences were identical, was also represented in the semen. We conclude that in these subjects there was no compartmentalization in the seminal tract. If there were local replication of virus in the seminal tract of these subjects it must have represented the full complexity of the virus in the blood. Alternatively, this virus did not replicate in the seminal tract but rather was imported from the blood. In either scenario, the viral populations in the blood and semen were essentially identical, representing well-equilibrated populations.
A different phylogenetic pattern was detected in subjects C007, C009, C012, C019, C070, and C109 (Fig. 2 and Fig. S1). Similar to the subjects where blood and semen populations were well-equilibrated, these subjects had viral populations in the semen that represented the full diversity of the virus in the blood. In addition, the blood populations were highly complex and consistent with a diverse viral population, with no sampling of identical blood sequences with the exception of patient C109. However, there was an additional feature of the viral populations in the semen of these subjects that distinguished them from the virus in the blood. In these subjects sampling of the viral population in semen resulted in examples where identical or nearly identical sequences were observed (Fig. 2, Fig. S1, Fig. S3, and Fig. S4-S15). Patient C109 had a clade of identical/nearly identical sequences that comprised nearly 75% of the entire semen viral population. Similarly, patient C009 had three duplicated viral variants that each comprised <10% of the semen population, indicating a broad range in the amount of sequence duplication that can exist within semen. We term this phenomenon clonal amplification, and because of the nature of the SGA strategy, this cannot be the result of PCR resampling since each amplicon was generated from a single template.
In four of the 12 subjects with subtype C HIV-1 (C047, C083, C018, and C113), a third relationship between the viral populations in the blood and semen was seen. For these subjects there was a deep branch point with high bootstrap support in the phylogenetic tree separating sequences found in the blood from sequences found in the semen (Fig. 3 and Supplemental Fig. S2). In addition to visual inspection of the phylogenetic trees to identify semen clades with long branch lengths with high bootstrap support, the presence of compartmentalized sequences was confirmed with the Slatkin-Maddison statistical [37] and correlation coefficient tests [38] available through Hypothesis testing through Phylogenies (HyPhy) [39]. Previous analyses have revealed that there is no gold standard from the variety of statistical measures available for detecting compartmentalization; therefore, multiple tests are recommended to determine the existence of compartmentalization [40]. Compartmentalization tests were performed with all viral sequences, and after removal of duplicated sequences since amplified variants in the semen can increase both the frequency of compartmentalization calls and the statistical support for those calls (Supplemental Table S2). Thus, compartmentalization of these viral populations was observed in subjects C047, C083, C018, and C113 and indicates an autonomously replicating subpopulation in the seminal tract that followed a distinct evolutionary pathway. As a result of this compartmentalized subpopulation, the virus in the semen was genetically distinct from the virus in the blood. Two subjects (C083 and C018) had compartmentalization of semen-derived sequences without clonal amplification (Fig. 3 and Fig. S2). In addition, two subjects (C047 and C113) had both clonal amplification and compartmentalization of semen-derived sequences (Fig. 3, Fig. S2, Fig. S8, and Fig. S11). Similar to the previous subjects with semen clonal amplification (with the exception of C109 as previously mentioned), there were no duplicated blood sequences. Thus, these data indicate that the male genital tract is capable of supporting complex viral populations, and that compartmentalization and amplification can occur independently.
In addition to the 12 men with HIV-1 subtype C infection, we analyzed blood and semen plasma viral RNA populations from 4 men with subtype B infection (Fig. 4, Fig. S3, Fig. S12-15). Each of the four had identical sequences (clonal amplification) in the seminal plasma that ranged from <10% to one-third of the semen viral population. In contrast, none of the patients had identical sequences in the blood plasma. In addition, three of the men had equilibrated blood and seminal plasma sequences: 700010333, 700010501, and 700011145; whereas, one of the 4 subtype B infected men (700010380) had significant compartmentalization of semen-derived sequences. Thus, clonal amplification and compartmentalization within the seminal plasma is a common feature of HIV-1 of different subtypes.
We carried out an analysis for each subject using Bayesian Evolutionary Analysis by Sampling Trees (BEAST) [41] to estimate the time to most recent common ancestor (TMRCA) of the amplified variants, and/or the TMRCA of compartmentalized variants using maximum likelihood trees. Of note, the topologies of the neighbor-joining and maximum likelihood trees were very similar (data not shown), indicating that these two different phylogenetic methods produced concordant results in their evolutionary models.
As a control to compare the BEAST estimates to known values obtained from previously published sequence data sets, a separate analysis was performed using a subset of published longitudinal env sequences (Fig. S16) [42]. From this data set, we calculated the time of divergence using C2-V5 env sequences obtained from longitudinal plasma samples at 3, 29, 42, 58, 70, and 100 months post-seroconversion; BEAST estimates of 10, 34, 49, 147, 144, and 204 months, respectively, were observed with a high coefficient of determination (R2 = 0.9155). Thus, in the setting of chronic HIV-1 infection, the observed BEAST estimates were similar to the expected values for periods up to several years, but there is a trend to overestimate time periods greater than four years by approximately two-fold.
Next, we determined the TMRCA of amplified and/or compartmentalized variants within seminal plasma. The TMRCA for the oligoclonal amplifications within the seminal compartment ranged from 1 to 375 days, with a mean of 57 days, indicating recent divergence. In contrast to the short evolutionary times observed with the semen variants displaying oligoclonal amplification, the subjects with significant semen compartmentalization had divergence estimates from 1.5 to 9.7 years, with a mean of 5.2 years. If clonally-amplified sequences were used only once, there was negligible effect on the TMRCA of the entire tree, or the TMRCA of amplified or compartmentalized variants (data not shown). Thus, the TMRCA of the clonally amplified variants tends to be relatively short in contrast to compartmentalized variants, which represent more distant divergence. However, we do not know if the rate of evolution in the semen is comparable to the blood adding additional uncertainty to the accuracy of the absolute values generated with the BEAST analysis.
To determine if populations were evolving randomly under neutral evolution, a Tajima's neutrality test was performed using DnaSP [43]. Fifteen of the 16 patients showed no evidence of selection (P values >0.10); however, C019 had a Tajima's D of -2.1 (P value <0.05) implying either population size expansion, or positive selection. Thus, a coalescent model of viral evolution as assumed by BEAST remains valid for the majority of patients. In the case of C019, the violation of a coalescent model was most likely due to the blood compartment (Tajima's D of -1.82, P value <0.05) vs. the semen compartment (Tajima's D of -1.60, P value >0.05). Taken together, these data suggest that BEAST is a robust tool to compare the TMRCA of amplified and compartmentalized variants for the majority of the patients that were analyzed.
To determine if the seminal plasma has a distinct immunologic profile relative to blood plasma, we measured the levels of nineteen cytokines and chemokines in the paired blood and semen samples from 12 of the men with chronic HIV-1 subtype C infection. As a control, we measured cytokines and chemokines from paired blood and semen samples in 12 uninfected men from the US and 6 uninfected men from Malawi without STIs. There were two features of the patterns of cytokines and chemokines (Fig. 5) that are noteworthy. First, a subset of cytokines and chemokines (IL-5, IL-7, IL-8, MIG, IP-10, and MCP-1) were concentrated in the semen of uninfected men with median levels that were 5 to approximately 1000 fold greater than in the blood; none of the remaining cytokines or chemokines was as high as five-fold concentrated in the semen (Fig. 5). Second, for seven of the cytokines and chemokines (IL-1b, IL-4, IL6, IL-7, IL-8, GM-CSF, and MCP-1) there was a significant increase in the semen:blood ratio of HIV-infected subjects compared to the uninfected subjects; conversely, MIG was significantly decreased in the infected subjects. Although our small sample size prevented a robust analysis, there were no cytokine correlates with amplification or compartmentalization of HIV-1 sequences in the semen. Moreover, there were no correlates with cytokine levels, and HIV-1 viral loads, amplification, compartmentalization, or the presence of asymptomatic STIs that were detected in five of the HIV-infected men (data not shown, although the small sample size and the intersubject variability precludes an assessment beyond more general trends).
The seminal compartment is the source of the transmitted virus in a majority of the transmission events for HIV-1. Thus, an understanding of the biology of HIV-1 in the seminal tract is integral to understanding the biology of transmission, and a comparison of blood and seminal sequences is critical to increase our knowledge of viral dynamics. We have used viral sequence populations to examine the dynamic relationship between virus and host in the seminal tract, and identify multiple mechanisms by which HIV-1 populations exist in the male genital tract. A significant limitation of this study is that it is cross-sectional, involving a single time point. Another limitation of the current work is that there are no proviral sequences from semen cells to define the source of the amplified or compartmentalized variants.
Previous work has identified paired blood and semen samples where the viral populations were discordant. In some cases this involved a comparison of viral RNA in blood plasma and seminal plasma, or a comparison of the sequences in viral DNA in blood cells and seminal cells [21]–[26], [28], [29], [32]. While these studies clearly established the potential for the virus to become compartmentalized, in most cases there were two potential limitations intrinsic to the experimental approach: the possibility of recombination of viral sequences during PCR which would introduce artifacts into the phylogenetic analysis, and the analysis of a fairly small number of viral genomes in each population precluding a comparison of the population structure. As a result the phenomenon of compartmentalization has been described as a dichotomous state, i.e. the presence or absence of compartmentalized viral populations. However, in subjects where there is equilibration in the seminal tract over the entire range of complexity in the blood compartment, virus in the semen is most easily explained by the direct import of virus into the seminal plasma from blood, perhaps with no local replication of this population.
The enrichment of cytokines and chemokines in the seminal tract (Fig. 5) likely contributes to an environment that is supportive of HIV-1 replication. Our data, as well as others [44], [45], show that in the absence of HIV-1 infection several cytokines and chemokines are enriched, suggesting that the seminal tract maintains a constitutive state of innate immune activation. This state is exacerbated with HIV-1 infection where the concentration of a broader array of cytokines and chemokines indicates both innate and adaptive responses shaping the environment [46]. Thus, target CD4+ T cells and macrophages are likely to be in an activated state in this environment, enhancing their ability to support viral replication.
In several subjects (C109 and 701010380) there is some evidence for clonal amplification of sequences in the blood, with this being more pronounced in C109. However, we do not know if the mechanism causing selective outgrowth in the blood is the same as that in the seminal tract, and in these subjects it is rare in the blood compared to the detection of clonal amplification in the semen in 12 of 16 men. The detection of clonal amplification within the seminal compartment raises several important questions. First, does amplification represent an initial stage of immunodeficiency? We have detected an example of clonal amplification during primary infection (data not shown) suggesting clonal amplification can occur at any stage of infection. Given that clonal amplification was detected in equilibrated and compartmentalized populations, this also suggests that clonal amplification is not determined by the overall state of immunodeficiency. Second, what is the cellular source where this amplification occurs? At one extreme clonally amplified sequences could be the product of a single cell. This seems unlikely since the seminal tract can support very complex populations in the compartmentalized state consistent with many available target cells, and some of the clonally amplified populations have some population structure suggesting they are the result of multiple rounds of replication (supported by the longer BEAST estimates of TMRCA for some of these populations). The alternative is that clonal amplification occurs in a population of cells that are not infected by diverse viral genotypes. We suggest that either uninfected CD4+ T cells concentrate in specific sites, or are seeded by a single cell that then expands, until the focus of cells becomes infected with a single virus that spreads through this isolated population until the target cells are depleted. This would explain the self-limiting nature of the clonal amplification and explain how several clonal amplifications can occur concurrently. Finally, this process could be at work during compartmentalized virus replication, and thus account for the clonal amplification process also appearing during the replication of a complex compartmentalized population.
A corollary of the isolation of the clonally amplified population is that the complex, compartmentalized population must be sustained by a distinct mechanism. There is likely continued import of virus from blood; however, the amount of locally replicating virus must obscure detection of this imported population. Based on these inferences we propose a model (Fig. 6) to account for virus in semen. An assumption of this model is that viral populations within blood and semen are turning over similarly, and this is supported by a recent report in the literature showing similar decay kinetics of HIV-1 populations in blood and semen in men who initiate antiviral therapy [47]. In addition, our model is distinct from the semen being a viral reservoir, which is associated with reduced levels of viral replication [48]. We suggest that virus in the semen is derived from multiple sources. First, there is direct import of virus from the blood compartment, potentially without replication in the seminal tract, accounting for virus that is fully equilibrated between the blood and seminal tract compartments. Second, there is infiltration of individual infected CD4+ cells or virions into pockets of uninfected target cells that generate local foci of infection in the seminal tract, giving rise to clonal amplification of virus in this compartment. Third, ongoing local immune activation provides an environment that can support sustained, autonomous virus replication giving rise to compartmentalized virus. We estimate that this distinct population can replicate independently for a significant period of time, although lack of information about the rate of evolution in the compartment precludes a detailed analysis of the age of the population.
An alternative interpretation of the appearance of compartmentalization is that there is delayed equilibration between blood and the seminal tract. In this circumstance a change in the population in the blood would not immediately be reflected in the semen, giving the transient appearance of compartmentalization. We do not favor this interpretation since the complexity of the virus in the semen can be quite high giving TMRCA values of months to years. However, the analysis of longitudinal samples in subjects displaying compartmentalization will be required to resolve this issue.
An important unanswered question is the site within the seminal tract where virus undergoes independent replication. A relevant observation in this regard is that vasectomy does not preclude the presence of virus in semen [49], [50], suggesting that production of significant amounts of virus occurs outside of the testis, and implicating the seminal vesicles and prostate. Moreover, distal genitourinary sources other than the prostate have been implicated as the major source of seminal HIV-1 in men without urethritis or prostatitis [51]. In the setting of the blood compartment, disease progression is associated with higher levels of immune cell activation [52]. This may reflect an increasing trend to fail to control viral replication but with a continued response to the presence of viral antigen. We suggest a similar process may occur in the seminal tract and perhaps in other peripheral sites of viral replication. Recent literature reports the existence of clonal amplification of HIV-1 sequence in other compartments, including the CSF [53], breast milk [54], and cervicovaginal lavage fluid [55]. Thus, an influx of activated immune cells into areas where virus suppression is incomplete could lead to sustained viral replication and a distinct evolutionary pathway. In this regard, the presence of activated immune cell infiltrates that have been observed in the seminal tract of SIV-infected macaques [18] provides the likely sites where viral replication could occur in the male genital tract.
A result of independent replication in the seminal tract, both clonal amplification and sustained replication, is to alter the composition of the viral population in the semen relative to that in the blood plasma. These differences can make blood a suboptimal surrogate for the seminal compartment in assessing the relationship of virus in the donor and recipient of a sexual transmission event. Several studies have noted differences between the transmitted virus and the virus in donor blood for subtypes A, C, and D [56]–[60] but not subtype B [35], [58], with differences in either glycosylation patterns, variable loop lengths, or susceptibility to neutralizing antibodies. It will be important to determine if the distinctive features of virus in semen play a role in transmission and/or in defining the nature of the transmitted virus.
The patients infected with HIV-1 subtype C (n = 12) were enrolled through the Kamuzu Central Hospital in Lilongwe, Malawi, between January and March, 1996 [33]. The protocol was approved by the University of North Carolina Committee on the Protection of Human Rights and the Malawi Health Sciences Research Committee. All study participants gave written informed consent and were offered a small payment for their participation. The original study design was a prospective, sequential comparison of two cohorts: HIV-1-infected men with urethritis who had urethral discharge on physical exam and at least five white blood cells per high-power field from a urethral swab, selected from the STI clinic, and HIV-infected men without urethritis on physical exam, selected from the dermatology clinic [33]. Blood and semen samples used in the current study were collected from men attending the dermatology clinic in Lilongwe, Malawi as described previously [33]. For both the STI and dermatology clinics, screening for gonorrhea, trichomonas, syphilis, and chlamydia was performed. In addition, blood and semen samples were obtained from participants with HIV-1 subtype B from the US (n = 4) who were enrolled through the CHAVI 001 clinical core, a multi-center, prospective, observational cohort study of acute HIV-1 infection. IRB approval was awarded by each participating center as well as the Division of AIDS. All study participants gave written informed consent and were offered a small payment for their participation. None of the subtype B infected participants had urethritis on physical exam, and were negative for gonorrhea, Chlamydia, syphilis, and trichomonas infection. Consistent with established infection, all HIV-1 infected Malawi and US patients were confirmed EIA and Western Blot positive at study enrollment. Paired blood and seminal plasma samples from HIV-1 uninfected males for cytokine/chemokine analyses were obtained from the CHAVI 001 clinical core sites in the US (n = 12) and Africa (n = 6).
Cell-free blood plasma and seminal plasma were isolated and frozen as previously described [1]. HIV-1 viral loads from blood and seminal plasma from the Malawi men were determined by quantitative nucleic-acid sequence-based-analysis (NASBA, Organon-Teknika) [33], and by Roche Amplicor vRNA or Abbott RealTime HIV-1 assays for the US men. Virus in the seminal plasma was pelleted by centrifugation prior to RNA isolation to remove the seminal plasma. The blood plasma or the resuspended virus pellet from the seminal plasma was extracted to isolate viral RNA using the QIAMP Viral RNA Mini Kit (Qiagen). For each sample, approximately 10,000 viral RNA copies based on viral load were extracted and eluted. cDNA synthesis was performed using Superscript III Reverse Transcriptase (Invitrogen) with an oligo-d(T) primer as previously described [34]–[36]. To confirm that proviral DNA was not the source of SGA env-derived amplicons from cell-free viral RNA, RT-minus blood (n = 11) and seminal (n = 11) plasma samples were subjected to the SGA protocol; the remaining samples had insufficient volume remaining for the RT-minus control experiment. To preclude PCR recombination and Taq-induced errors, single genome amplification (SGA) of the env gene was performed using limiting dilution [34]–[36], [61]–[63]. PCR amplicons were bidirectionally sequenced. To ensure that sequences arose from single DNA molecules, chromatograms with double peaks, indicating amplification from more than one cDNA template, were excluded. SGA-derived env amplicons with frameshift mutations that resulted in premature stop codons were also excluded. GenBank accession numbers are HM638460 to HM639260.
DNA sequence alignments were performed using clustal W [64]. Phylogenetic trees were generated using a neighbor-joining method (MEGA 4.0) [65]. Pairwise DNA distances were computed using MEGA 4.0. Highlighter plots were generated to visualize sequence differences (www.hiv.lanl.gov). A Tajima's D test for neutrality was performed for each patient using DnaSP [43]. Compartmentalization of viral sequences was assessed by using the Slatkin-Maddison test [37] and correlation coefficient [38] available through HyPhy [39]. Gene flow was determined by the number of migration events compared between semen and blood after 10,000 permutations for the Slatkin-Maddison test. Compartmentalization was defined when P values <0.01 were obtained with the Slatkin-Maddison test using all sequences except the clonally-amplified sequences, of which only one was included, and when concordant results were obtained with the correlation coefficient test. More extreme P values were obtained when all of the clonally amplified sequences were included (Supplemental Table S2).
Nineteen cytokines and chemokines were analyzed by luminex from paired blood and seminal plasma from 12 HIV-1 infected and uninfected subjects as previously described [66]. Concentrations of IL-1b, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12 (p70), IL-13, IFN-g, TNF-a, and GM-CSF were measured using LINCOplex Luminex high-sensitivity 13-plex kits (Millipore) according to the manufacturer's instructions. Concentrations of MIP-1a, MIP-1b, RANTES, MCP-1, MIG, and IP-10 were measured using custom standard sensitivity 6-plex kits (Bio-Rad) according to the manufacturer's instructions. Each sample was assayed in duplicate, and cytokine standards supplied by the manufacturer were run in parallel. Data were collected using the Bio-plex Suspension Array Reader (Bio-Rad) and a regression formula was used to calculate sample concentrations from standard curves. Values that were below the lower limit of detection were reported as the mid-point between the lower level of detection and zero. Semen:blood analyte ratios were calculated for each subject; however, data were excluded if both compartments were below the level of detection. Saturated values were reported as the upper limit of detection. Sensitivity values were adjusted for samples where the volume was limited and had to be diluted before the measurement. Statistical tests comparing analyte levels were non-parametric Mann-Whitney-U test for two groups (compartmentalization vs. equilibration), and Kruskal-Wallis test for three groups (high amplification, low amplification, no amplification). The non-parametric Mann-Whitney test was used to compare semen:blood cytokine ratios between HIV-1 infected and uninfected subjects.
For each subject, blood and semen sequences were aligned using ClustalW 2.0.7 [67]. Markov Chain Monte Carlo Simulation (MCMC) using Bayesian inference was used to resolve a phylogenetic tree with the highest posterior probability to estimate the time of divergence from the most recent common ancestor (MRCA) implemented in BEAST (Bayesian Evolutionary Analysis by Sampling Trees v.1.4.8) [41]. Each independent run had a chain length of 30,000,000 with a sample frequency of 1000. A general time-reversible substitution model was used, with site heterogeneity using a gamma distribution with a proportion of invariant sites and sampling across four categories. Analyses were performed using an HIV-1 generation time of 1.6 days [68]. Rate heterogeneity across codon positions was unlinked, and the mean fixed substitution rate was 2.16×10−5 under a relaxed uncorrelated exponential molecular clock. A coalescent piecewise-constant Bayesian Skyline model with ten groups was used as the tree prior. The MCMC log output of each run was examined in Tracer 1.4 to verify adequate chain mixing and estimated sample sizes of greater than 200 for parameters of interest, and log and tree files with a minimum of two independent runs were combined with a 10% burn-in using LogCombiner 1.4.8. The target tree for each patient was summarized using TreeAnnotater 1.4.8, and visualized in FigTree1.1.2.
|
10.1371/journal.pntd.0002855 | Transcriptional and Proteomic Responses to Carbon Starvation in Paracoccidioides | The genus Paracoccidioides comprises human thermal dimorphic fungi, which cause paracoccidioidomycosis (PCM), an important mycosis in Latin America. Adaptation to environmental conditions is key to fungal survival during human host infection. The adaptability of carbon metabolism is a vital fitness attribute during pathogenesis.
The fungal pathogen Paracoccidioides spp. is exposed to numerous adverse conditions, such as nutrient deprivation, in the human host. In this study, a comprehensive response of Paracoccidioides, Pb01, under carbon starvation was investigated using high-resolution transcriptomic (RNAseq) and proteomic (NanoUPLC-MSE) approaches. A total of 1,063 transcripts and 421 proteins were differentially regulated, providing a global view of metabolic reprogramming during carbon starvation. The main changes were those related to cells shifting to gluconeogenesis and ethanol production, supported by the degradation of amino acids and fatty acids and by the modulation of the glyoxylate and tricarboxylic cycles. This proposed carbon flow hypothesis was supported by gene and protein expression profiles assessed using qRT-PCR and western blot analysis, respectively, as well as using enzymatic, cell dry weight and fungus-macrophage interaction assays. The carbon source provides a survival advantage to Paracoccidioides inside macrophages.
For a complete understanding of the physiological processes in an organism, the integration of approaches addressing different levels of regulation is important. To the best of our knowledge, this report presents the first description of the responses of Paracoccidioides spp. to host-like conditions using large-scale expression approaches. The alternative metabolic pathways that could be adopted by the organism during carbon starvation can be important for a better understanding of the fungal adaptation to the host, because systems for detecting and responding to carbon sources play a major role in adaptation and persistence in the host niche.
| The species of the Paracoccidioides genus, a neglected human pathogen, represent the causative agents of paracoccidioidomycosis (PCM), one of the most frequent systemic mycoses in Latin America. Despite being phagocytosed, the fungus conidia differentiate into the parasitic yeast form that subverts the normally harsh intraphagosomal environment and survives and replicates into murine and human macrophages. It has been suggested that alternative carbon metabolism plays a role in the survival and virulence of Paracoccidioides spp. within host cells. We used large-scale transcriptome and proteome approaches to better characterize the responses of Paracoccidioides, Pb01, yeast parasitic cells, to carbon starvation. We aimed to identify important molecules used by the fungus to adapt to these hostile conditions. The shift to a starvation mode, including gluconeogenesis and ethanol increases, activation of fatty acids, and amino acid degradation are the strategies used by the pathogen to persist under this stress. Our study provides a detailed map of Paracoccidioides spp. responses under carbon starvation conditions and contributes to further investigations of the importance of alternative carbon adaptation during fungus pathogenesis.
| Metabolic adaptability and flexibility are important attributes for pathogens to successfully colonize, infect, and cause disease in a wide range of hosts. Therefore, they must be able to assimilate various carbon sources. Carbohydrates are the primary and preferred source of metabolic carbon for most organisms and are used for generating energy and producing biomolecules [1].
Studies have highlighted the importance of carbon metabolism in fungi [2], [3]. Pathogens such as Candida albicans display sufficient metabolic flexibility to assimilate the available nutrients in diverse niches such as the skin, mucous membranes, blood, and biofilms [4], [5]. The mucosal surface of the lung may provide a more nutrient-limited condition because it is not in direct contact with nutrients from food intake [6]. Additionally, in the lungs, macrophages rapidly phagocytize inhaled microorganisms supported by neutrophils and dendritic cells [7]. Macrophages are considered a glucose- and amino acid-poor environment [8], [9] and may form extremely nutrient-limited conditions causing severe starvation [10]. In C. albicans and Cryptococcus neoformans, alternative carbon metabolism was detected after internalization by macrophages playing a role in fungal survival in these host cells [9], [11], [12]. In contrast, in the case of systemic infections, pathogens can reach different internal organs such as the liver, which is the main storage compartment of glucose in the form of glycogen. The bloodstream, for example, is the major carrier of nutrients, glucose, proteins, amino acids, and vitamins in larger quantities [10]. In this way, metabolic and stress adaptation represent vital fitness attributes that have evolved alongside virulence attributes in fungi [8]–[10], [13].
Alternative carbon metabolism was also described to be important to protozoa and bacteria [14], [15]. Entamoeba histolytica uses an alternative source of energy when the microorganism is exposed to glucose starvation. In the specific pyruvate-to-ethanol pathway in E. histolytica, acetyl-CoA is converted to acetaldehyde, which is then reduced to ethanol [16]. Reduced levels of the long-chain fatty-acid-CoA ligase protein during glucose starvation conditions in E. histolytica may explain a mechanism by which acetyl-CoA is shuttled from the fatty acid metabolism into this pyruvate-to-ethanol pathway. In addition, the glucose starvation modulates the protozoa virulence, based on proteome analysis [15]. The transcriptome and large-scale proteome dynamics were also analyzed in Bacillus subtilis from glucose-starved cells. A direct consequence of glucose depletion on proteins was the switch from glycolytic to gluconeogenic metabolism and elevated abundance of proteins of the tricarboxylic cycle used for energy generation. Genes that are involved in exponential growth, amino-acid biosynthesis, purine/pyrimidine synthesis and the translational machinery were down-regulated in the bacteria cells under glucose starvation [14].
The species of the Paracoccidioides genus represent the causative agents of paracoccidioidomycosis (PCM), one of the most frequent systemic mycoses in Latin America [17]. The genus comprises four phylogenetic lineages (S1, PS2, PS3, and Pb01-like). The phylogenetic analysis of many Paracoccidioides isolates has resulted in the differentiation of the genus into two species: P. brasiliensis, which represents a complex of three phylogenetic groups, and P. lutzii, which represents the Pb01-like isolates [18]–[21]. Paracoccidioides spp. grows as a yeast form in the host tissue and in vitro at 36°C, while it grows as mycelium under saprobiotic condition and in culture at room temperature (18–23°C). As the dimorphism is dependent on temperature, when the mycelia/conidia are inhaled into the host lungs, the transition of the mycelia to the pathogenic yeast phase occurs [22].
One of the first lines of defense faced by Paracoccidioides spp. during host invasion is the lung resident macrophages. Despite being phagocytosed, the fungus conidia differentiate into the parasitic yeast form that subverts the normally harsh intraphagosomal environment and survives and replicates into murine and human macrophages [23]. It has been proposed for PCM and other systemic mycoses that the fungal intracellular parasitism is a major event for disease establishment and progression in susceptible hosts. The survival inside the macrophage may allow fungal latency and/or dissemination from the lungs to several organs such as observed in C. neoformans [24], [25]. In this sense, Paracoccidioides spp. has evolved defense mechanisms to survive under nutritionally poor environments. It has been suggested that alternative carbon metabolism plays a role in the survival and virulence of Paracoccidioides spp. within the host [26], [27], as occurs in C. albicans and C. neoformans [9], [12]. Transcriptional analysis of Paracoccidioides spp. upon internalization by macrophages, as determined by a DNA microarray, consisting of 1,152 cDNA clones, showed that the fungus responds to the glucose-depleted environment found in the macrophage phagosome, by the expression of 119 classified genes, differentially transcribed. Genes involved in methionine biosynthesis (cystathionine β-lyase), oxidative stress response (superoxide dismutase and heat shock protein 60), and cytochrome electron transport system (cytochrome oxidase c) were induced by the fungus. Moreover, Paracoccidioides spp. reduced the expression of genes that are involved in the glycolysis pathway such as the key regulatory phosphofructokinase (pfkA) and genes related to cell wall polysaccharides such as β-glucan synthase (fks) [27]. In addition, studies of the transcriptome profiling from yeast cells of Paracoccidioides spp. derived from mouse liver revealed that the fungus most likely uses multiple carbon sources during liver infection. Genes encoding enzymes, regulators, and transporters in carbohydrate and lipid metabolism were significantly overexpressed. Ethanol production was also detected, indicating that it may be particularly important during infection [28].
Here, we described the response of Paracoccidioides facing carbon starvation using a high-throughput RNA Illumina sequencing (RNAseq) and quantitative proteome NanoUPLC-MSE, a two-dimensional liquid chromatography-tandem mass spectrometry approach. RNAseq is a developed approach to transcriptome profiling that uses deep-sequencing technologies and has already been applied to organisms such as Saccharomyces cerevisiae, Arabidopsis thaliana, mouse, and human cells [29]–[33]. With regard to proteomic analysis, our group has developed detailed proteome maps of the process of the fungus dimorphism, the response to iron and zinc deprivation, the fungus exoproteome, and the response to oxidative stress as well as comparative proteome maps of members of Paracoccidioides phylogenetic species [34]–[39]. In this study, a comprehensive response of Paracoccidioides, isolate Pb01, under carbon starvation was performed by transcriptional and proteomic approaches. To the best of our knowledge, this is the first description of high-resolution transcriptomics and proteomics applied to study the response of Paracoccidioides spp. to carbon starvation. We believe that the obtained data can be relevant in the understanding of the fungal establishment in the host.
Paracoccidioides, Pb01 (ATCC MYA-826), was used in the experiments. The yeast phase was cultivated for 7 days, at 36°C in BHI semisolid medium added to 4% (w/v) glucose. When required, the cells were grown for 72 h at 36°C in liquid BHI, washed with PBS 1×, and incubated at 36°C in a McVeigh/Morton (MMcM) medium with the following composition per 100 mL: KH2PO4 0.15 g; MgSO4.7H20 0.05 g; CaCl2.2H20 0.015 g; (NH4)2SO4 0.2 g; vitamin 1 mL and trace element supplements 0.1 mL). The stock vitamin solution contained, also per 100 mL: thiamine hydrochloride, 6.0 mg; niacin, 6.0 mg; calcium pantothenate, 6.0 mg; inositol, 1.0 mg; biotin, 0.1 mg; riboflavin, 1.0 mg; folic acid, 10 mg; choline chloride, 10 mg; and pyridoxine hydrochloride, 10 mg. The trace element solution contained, per 100 mL: H3BO3, 5.7 mg; CuSO4.5H20, 15.7 mg; Fe(NH4)2(SO4)2.6H2O, 140.4 mg; MnSO4.14H2O, 8.1 mg; (NH4)6Mo7O24.4H2O, 3.6 mg; ZnSO4.7H2O, 79.2 mg, as described previously [40], except for removal of the amino acids. All components except the vitamin supplement were mixed, and the pH was adjusted to 7.0 with 1N NaOH. The vitamin solution was filter sterilized and added after the remainder of the medium had been autoclaved at 121°C for 15 min and cooled.
Paracoccidioides yeast cells were subjected to carbon starvation as following. The Pb01 yeast cells were grown for 72 h at 36°C in liquid BHI added to 4% (w/v) glucose. The cells were harvest and washed three times with PBS 1×. A total of 106 cells/mL were inoculated in modified MMcM medium [40] with 4% (glucose, carbon source) or 0% of glucose (carbon starvation). The cells were incubated at 36°C.
Following Paracoccidioides incubation in carbon starving condition, cells were centrifuged at 1,500× g, frozen in liquid nitrogen, and disrupted by maceration as described in [38]. Briefly, cells were treated with TRIzol reagent (Invitrogen, Carlsbad, CA, USA). The manufacturer's protocol was followed to extract total RNA. The RNA was reversibly transcribed using the high capacity RNA-to-cDNA kit (Applied Biosystems, Foster City, CA, USA). We confirmed the specificity of each primer pairs for the target cDNA by the visualization of a single PCR product following agarose gel electrophoresis and melting curve analysis. The cDNA was quantified by qRT-PCR using a SYBR green PCR master mix (Applied Biosystems Step One Plus PCR System). qRT-PCR analysis was performed in biological triplicate for each cDNA sample as previously described [38]. The data were normalized using the constitutive gene encoding the 60S ribosomal L34 as the endogenous control. In order to analyze the reliability of the normalizer used in our qRT-PCRs we used data obtained from three different housekeeping genes and the software NormFinder (Aarhus University, Aarhus, Denmark). The software identify the most suitable reference genes as previously described in [41]. We used the actin (PAAG_00564), tubulin alpha-1 chain (PAAG_01647) and 60S ribosomal protein L34 (PAAG_00746) genes and the results show that the L34 is the best gene to be used as normalizer in our qRT-PCRs. It was demonstrated by the lower stability value by comparing with actin and tubulin genes after two different runs (Table S1). The 60S ribosomal L34 gene was amplified in each set of qRT-PCR experiments and was presented as relative expression in comparison to the experimental control cells value set at 1. Data were expressed as the mean ± standard deviation of the biological triplicates of independent experiments. Standard curves were generated by diluting the cDNA solution 1∶5. Relative expression levels of genes of interest were calculated using the standard curve method for relative quantification [42]. Statistical comparisons were performed using the student's t test and p-values≤0.05 were considered statistically significant. The specific primers, both sense and antisense, are described in Table S2.
Proteins were fractionated by 12% SDS-polyacrylamide gel electrophoresis, and stained with Coomassie Blue R or transferred to Hybond ECL membrane (GE Healthcare). Membranes were blocked for 1 h at room temperature in a solution containing 10% (w/v) skim milk powder and 0.1% Tween 20 in Tris-buffered saline (TBS-T). The primary polyclonal antibody anti-isocitrate lyase [43] was diluted in blocking solution and incubated with the membrane for 1 h at room temperature. Membranes were washed in Tris-buffered saline and then incubated with alkaline phosphatase conjugated secondary antibodies for another hour at room temperature. Labeled bands were revealed with 5-bromo-4-chloro-3-indolylphosphate/nitroblue tetrazolium and negative controls were obtained with rabbit preimmune. Images from western blots were acquired with ImageMaster 2D Platinum 6.0 (Geneva Bioinformatics, GeneBio). Raw Tiff images were analyzed by densitometric analysis of immunoblotting bands using the software AphaEaseFC (Genetic technologies Inc.). Pixel intensity for the analyzed bands was generated and expressed as Integrated Density Values (IDV).
Following Paracoccidioides growth in the presence or not of carbon for 6 h, cells were treated with TRIzol reagent (Invitrogen, Carlsbad, CA, USA) to obtain RNA molecules, from biological replicates. The cDNAs libraries were prepared from poly(A)-fragment selected mRNA and processed on the Illumina HiSeq2000 Sequencing System (http://www.illumina.com/). As a result, approximately 40 million of reads of 100 bp paired-end sequencing were obtained for each sample. The sequencing reads were mapped to reference the Paracoccidioides genome (Pb01), (http://www.broadinstitute.org/annotation/genome/paracoccidioides_brasiliensis/MultiHome.html), using the Bowtie 2 tool [44]. Mapped reads data were analyzed by the DEGseq package [45]. Briefly, each read was allowed to alignment in just one site of the genome and the reads were counted. The default parameters were used to perform the alignment. The number of mismatches allowed in seed alignment (-N) is 0, and the length of each seed (-L) is 20. The fold change selection method was used for differentially expressed genes selection using a Fisher exact test, and a p-value of 0.001 was considered to select the genes. From the selected genes, the 2-fold change cut-off was considered. Genes with log2 (fold change) higher than 1 or less than −1 were selected and classified as up- and down-regulated genes, respectively. Gene's identifications and annotations were determined from the Paracoccidioides genome database (http://www.broadinstitute.org/annotation/genome/paracoccidioides_brasiliensis/MultiHome.html). The biological processes were obtained using the Pedant on MIPS (http://pedant.helmholtz-muenchen.de/pedant3htmlview/pedant3view?Method=analysis&Db=p3_r48325_Par_brasi_Pb01) which provides a tool to browse and search the Functional Categories (FunCat) of proteins. All scripts can be obtained on request.
Following Paracoccidioides cell incubation in carbon-starved media for up to 48 h, the cells were centrifuged at 1,500× g, resuspended in a 50 mM ammonium bicarbonate pH 8.5 solution and disrupted using glass beads and bead beater apparatus (BioSpec, Oklahoma, USA) in 5 cycles of 30 sec, while on ice. The cell lysate was centrifuged at 10,000× g for 15 min at 4°C and the supernatant was quantified using the Bradford reagent (Sigma-Aldrich) [46]. The samples were analyzed using nanoscale liquid chromatography coupled with tandem mass spectrometry. Sample aliquots (50 µg) were prepared for NanoUPLC-MSE as previously described [47], [48]. Briefly, 50 mM ammonium bicarbonate was added and was followed by addition of 25 µL of RapiGEST (0.2% v/v) (Waters Corp, Milford, MA). The solution was vortexed and then incubated at 80°C for 15 min; 2.5 µL of a 100 mM DTT solution was then added and incubated for 30 min at 60°C. The sample was cooled at room temperature and 2.5 µL of 300 mM iodoacetamide was added followed by sample incubation in a dark room for 30 min. A 10 µL aliquot of trypsin (Promega, Madison, WI, USA) prepared with 50 mM ammonium bicarbonate to 50 ng/uL, was added. The sample was vortexed slightly and digested at 37°C overnight. Following the digestion, 10 µL of 5% (v/v) trifluoroacetic acid was added to hydrolyze the RapiGEST, followed by incubation at 37°C for 90 min. The sample was centrifuged at 18,000× g at 6°C for 30 min, and the supernatant was transferred to a Waters Total Recovery vial (Waters Corp). A solution of one pmol.ul−1 MassPREP Digestion Standard [rabbit phosphorilase B (PHB)] (Waters Corp) was used to prepare the final concentration of 150 fmol.ul−1 of the PHB. The buffer solution of 20 mM ammonium formate (AF) was used to increase the pH. The digested peptides were separated further via NanoUPLC-MSE and analyzed using a nanoACQUITY system (Waters Corporation, Manchester, UK).
Mass spectrometry data obtained from NanoUPLC-MSE were processed and searched using ProteinLynx Global Server (PLGS) version 3.0 (Waters Corp) as previously described [49]. Protein identifications and quantitative data packaging were performed using dedicated algorithms [50], [51] and a search against the Paracoccidioides database (http://www.broadinstitute.org/annotation/genome/paracoccidioides_brasiliensis/MultiHome.html). The ion detection, clustering, and log-scale parametric normalizations were performed in PLGS with an ExpressionE license installed (Waters, Manchester, UK). The intensity measurements were typically adjusted for these components, i.e., the deisotoped and charge state reduced EMRTs that were replicated throughout the entire experiment for the analysis at the EMRT cluster level. STY phosphorylations were set as variable modification. Components were typically clustered with a 10 ppm mass precision and a 0.25 min time tolerance against the database-generated theoretical peptide ion masses with a minimum of one matched peptide. The alignment of elevated-energy ions with low-energy precursor peptide ions was performed with an approximate precision of 0.05 min. One missed cleavage site was allowed. The precursor and fragmention tolerances were determined automatically.
The protein identification criteria also included the detection of at least three fragment ions per peptide, 7 fragments per protein and the determination of at least one peptide per protein. The maximum protein mass was set to 600 kDa and trypsin was chosen as the primary digest reagent. The identification of the protein was allowed with a maximum 4% false positive discovery rate in at least two out of three technical replicate injections. Using protein identification replication as a filter, the false positive rate was minimized because false positive protein identifications, i.e., chemical noise, have a random nature and do not tend to replicate across injections. For the analysis of the protein identification and quantification level, the observed intensity measurements were normalized to the intensity measurement of the identified peptides of the digested internal standard. Protein and peptides tables generated by PLGS were merged and the dynamic range of the experiments, peptides detection type, and mass accuracy were determined for each condition as described in [47] by setting the minimum repeat rate for each protein in all replicates to 2. Normalization was performed with a protein that showed no significant difference in abundance in all injections [52] to accurately compare the expression protein level to carbon and carbon-starved samples.
Paracoccidioides yeast cells were grown for 72 h at 36°C in liquid BHI, washed with PBS 1×, and filtered using a nylon mesh filter to yield small and non-aggregated cells. A total of 5×107 cells/50 mL were inoculated in modified MMcM medium [40] with carbon source (4% glucose) or under carbon starvation (absence of glucose) and were incubated at 36°C. In each time-point, 10 mL of culture were centrifuged at 1,500× g and the supernatants were carefully removed. The cells were ressuspended in PBS 1× up to 500 µl and subjected to 95°C heating for 1 h. The cells were centrifuged, frozen in liquid nitrogen and lyophilized for 24 h. Dry weight was determined. Data are expressed as the mean ± standard deviation of the triplicates of independent experiments. Statistical comparisons were performed using the Student's t test and p-values≤0.05 were considered statistically significant.
The viability was determined by membrane integrity analysis using propidium iodide as dead cells marker as previously described [34], [35]. Briefly, cell suspension (106 yeast cells/mL) were centrifuged and the supernatant was discarded. The cells were stained by addition of the propidium iodide solution (1 µg/mL) for 20 min in the dark at room temperature. After dye incubation, stained cell suspension was immediately analyzed in a C6 Accuri flow cytometer (Accuri Cytometers, Ann Arbor, MI, USA). A minimal of 10,000 events per sample was acquired with the FL3-H channel. Data was collected and analyzed using FCS Express 4 Plus Research Edition software (Denovo Software, Los Angeles, CA, USA).
The concentration of ethanol was quantified by enzymatic detection kit according to the manufacturer's instruction (UV-test for ethanol, RBiopharm, Darmstadt, Germany). Ethanol is oxidized to acetaldehyde by the enzyme alcohol dehydrogenase, in the presence of nicotinamide-adenine dinucleotide (NAD). Acetaldehyde is quantitatively oxidized to acetic acid in the presence of aldehyde dehydrogenase, releasing NADH, which is determined by means of its absorbance at 340 nm. Paracoccidioides Pb01 yeast cells were subjected or not to carbon starvation, and 106 cells were used to assay. Briefly, cells were counted, centrifuged, and lysed using glass beads and bead beater apparatus (BioSpec, Oklahoma, USA) in 5 cycles of 30 sec, keeping the samples on ice. The cell lysate was centrifuged at 10,000× g for 15 min at 4°C and the supernatant was used for enzymatic assay according to the manufacturer's instructions. The concentrations of ethanol were obtained in triplicate.
Following Paracoccidioides growth under carbon source (4% glucose) or carbon starvation (absence of glucose) the cells were centrifuged at 1,500× g, resuspended in a solution containing 20 mM Tris-HCl, pH 8.8, 2 mM CaCl2 [53] and disrupted using glass beads and bead beater apparatus (BioSpec, Oklahoma, USA) in 5 cycles of 30 sec, while on ice. The cell lysate was centrifuged at 10,000× g for 15 min at 4°C and the supernatant was quantified using the Bradford reagent (Sigma-Aldrich) [46]. Formamidase activity was determined by measuring the amount of ammonia formation as previously described [38], [54]. One µg of Paracoccidioides total protein extract prepared as described above was added to 200 µl of a 100 mM formamide substrate solution in 100 mM phosphate buffer containing 10 mM of EDTA, pH 7.4. Samples were incubated at 37°C for 30 min. A total of 400 µl of phenol-nitroprusside and the same volume of alkaline hypochlorite (Sigma Aldrich, Co.) were added on the tube. The samples were then incubated for 6 min at 50°C and the absorbance was read at 625 nm. The amount of ammonia released for each sample was determined by comparing to a standard curve. One unit (U) of formamidase specific activity was defined as the amount of enzyme required to hydrolyze 1 µmol of formamide (corresponding to the formation of 1 µmol of ammonia) per min per mg of total protein.
Isocitrate lyase activity was determined by measuring the formation of glyoxylate as its phenylhydrazone derivative [55]. Glyoxylate-phenylhydrazone formation was determined by measuring the absorbance at 324 nm, using an extinction coeficient of 16.8 mM−1 cm−1, in a reaction mixture containing 2 mM threo-D,L-isocitrate (Sigma Aldrich, Co.), 2 mM MgCl2, 10 mM phenylhydrazine HCl (Sigma Aldrich, Co.), 2 mM dithiothreitol and 50 mM potassium phosphate at pH 7.0. Specific activity was determined as the amount of enzyme required to form 1 µmol of glyoxylate-phenylhydrazone per min per mg of total protein. For both assays, the statistical comparisons were performed using the Student's t test and p-values≤0.05 were considered statistically significant.
To evaluate whether carbon starvation influenced the internalization and survival of Paracoccidioides, the survival rate (viable fungi after co-cultivation) was determined by quantifying the number of colony-forming units (CFUs) recovered from macrophage infection. Macrophages, cell line J774 A.1 (Rio de Janeiro Cell Bank – BCRJ/UFRJ, accession number: 0121), maintained in RPMI medium (RPMI 1640, Vitrocell, Brazil), with 10% FBS (fetal bovine serum, [v/v]) were used in assays. A total of 106 macrophages were seeded into each well of a 24-well tissue culture plate and 100 U. mL−1 of IFN-gamma (murine IFN-γ, PeproTech, Rocky Hill, New Jersey, USA) was used for 24 h at 36°C with 5% CO2 for activation of macrophages [56]. Prior to co-cultivation, Paracoccidioides yeast cells were grown in BHI liquid medium (4% [w/v] glucose, 3.7% [w/v] brain heart infusion, pH 7.2) for 72 h and subjected to both conditions, a carbon source (4% [w/v] of glucose) or carbon starvation (no glucose) at 36°C, in McVeigh/Morton medium (MMcM) for 48 h. The same ratio of 1∶2.5 macrophage∶yeast was used for infection of both, carbon and carbon-starved yeast cells. The cells were co-cultivated for 24 h with 5% CO2 at 36°C to allow fungal internalization. Prior to macrophages lysis, dilutions of the supernatant (culture from co-cultivation) removed by aspiration, were plated in BHI medium supplemented with 5% FBS (fetal bovine serum, [v/v]) incubated in 5% CO2 at 36°C for 10 days. The number of viable cells was determined based on the number of CFUs. Infected macrophages were lysed with distilled water and dilutions of the lysates were plated in BHI medium supplemented with 5% FBS (fetal bovine serum, [v/v]). Colony-forming units (CFU) were determined after growth at 36°C, in 5% CO2, for 10 days. For both experiments, the CFUs data were expressed as the mean value ± the standard deviation from triplicates and the statistical analyses were performed using Student's t test.
To analyze the expression of genes from Paracoccidioides yeast cells infecting macrophages, the same cell line described above was used. Paracoccidioides yeast cells were cultivated in BHI liquid medium (4% [w/v] glucose, 3.7% [w/v] brain hearth infusion, pH 7.2) for 72 h and incubated with macrophages for 24 h, followed by washing with distilled water to promote macrophages lysis. RNAs and cDNAs were obtained as previously described. Specific oligonucleotides were used to amplify the fructose-1,6-biphosphatase, isocitrate lyase and 3-ketoacyl-CoA thiolase genes from Pb01 yeast cells. The negative amplification was obtained when cDNAs were used only from macrophages.
To determine the number of adhered/internalized fungi cells by macrophages, the same macrophage cell line described above was used. A total of 106 macrophages were plated on glass coverslips per 12-well tissue culture plate and 100 U. mL−1 of IFN-gamma (murine IFN-γ, PeproTech, Rocky Hill, New Jersey, USA) was used for 24 h at 36°C with 5% CO2 for its activation, as described above [56]. Prior to co-cultivation, Paracoccidioides yeast cells were grown in BHI liquid medium (4% [w/v] glucose, 3.7% [w/v] brain heart infusion, pH 7.2) for 72 h, and subsequently transferred to McVeigh/Morton medium (MMcM), containing or not a carbon source, for 48 h, at 36°C. The same ratio of 1∶2.5 macrophage∶yeast was used for infection in both conditions. The cells were co-cultivated for 6 and 24 h with 5% CO2 at 36°C, to allow fungal adhesion/internalization. The supernatants were then aspirated, the monolayer was gently washed twice with PBS 1× to remove any non-adhered/internalized yeast cells, and the samples were processed for light microscopy. The glass coverslips were fixed with methanol and stained with Giemsa (Sigma). The cells were observed using the Axio Scope A1 microscope and digital images were acquired using the software AxioVision (Carl Zeiss AG, Germany). A total of 300 macrophages were counted to determine the average number of adhered/internalized fungal cells, as described before [37], [57]. For both experiments, the number of adhered/internalized fungal cells was shown in percentage of the total as the mean value ± the standard deviation from triplicates. The statistical analyses were performed using Student's t test.
We first sought to set up a time-point to analyze the fungus response to carbon starvation at transcriptional and proteomic levels. Hence, we analyzed gene and protein expression of genes known to be regulated under carbon-limited microenvironments in other fungi [9], [12], by using qRT-PCR and western blot assays, in Pb01 (Fig. 1A). Changes in the expression of genes encoding fructose-1,6-biphosphatase, isocitrate lyase, and 3-ketoacyl-CoA thiolase, which are representatives of gluconeogenesis, the glyoxylate cycle, and β-oxidation, respectively, were analyzed in the Paracoccidioides, Pb01, yeast cells subjected to carbon starvation. As depicted in Fig. 1A, carbon starvation promoted the increase of gene expression at 6 h and at 12 h for the treatments. At protein level, the differential expression of isocitrate lyase suggested that Paracoccidioides, Pb01, up-regulated the glyoxylate cycle after 48 h of carbon starvation (Figs. 1B and S1). In this way, the 6 h and 48 h treatments were considered in further transcriptional and proteomic analysis, using a high-throughput RNA Illumina sequencing (RNAseq) and NanoUPLC-MSE, respectively.
The transcriptome analysis was performed using next generation sequencing and the Paracoccidioides, isolate Pb01, genome database (http://www.broadinstitute.org/annotation/genome/paracoccidioides_brasiliensis/MultiHome.html) was used as a reference genome for mapping the reads which were analyzed by DEGseq package [45]. For the global analysis, plotting graphs were performed (Fig. S2).The number of the reads counted for each transcript in carbon and carbon-starved conditions was represented by scattered dots (Fig. S2).The transcripts are represented by dots, which could present a different number of reads in each condition (Fig. S2A). We applied a statistical test to identify differentially expressed transcripts, represented by red dots (Fig. S2B). A significant number of genes were regulated during carbon starvation.
Although 1.5-fold change can be statistically significant [58], a cut-off of 2-fold change was applied to determine the up- and down-regulated transcripts (Tables S3 and S4, respectively) totaling 1,063 differentially expressed transcripts in Pb01 yeast cells under carbon starvation. A biological process classification was performed to gain a general understanding of the functional categories affected by carbon starvation. A total of 64.6% (687 transcripts) were represented by miscellaneous and unclassified categories, and the other 35.4% (376 transcripts) were represented by classified biological categories. The functional classifications and the percentage of up- and down-regulated transcripts in each classified category are shown in Fig. S3.
The transcriptome analysis showed that transcripts associated with metabolism were the most represented during 6 h of carbon starvation in Pb01 (Fig. S3A). From these, approximately 27% were represented by up-regulated and 17% by down-regulated transcripts (Fig. S3B). Subcategories of metabolism related to amino acid, nitrogen/sulfur, C-compound/carbohydrate, lipid/fatty acid, purines, secondary, and phosphate metabolisms were regulated under carbon starvation stress, and all of them showed a higher number of up- than down-regulated transcripts (Tables S3 and S4). Other categories were also regulated in Pb01 under carbon starvation. The categories associated with protein fate, cell cycle/DNA processing, transcription, and cellular transport were largely represented in the transcriptome (Fig. S3A). The number of transcripts with increased or reduced expression was also investigated for these categories and the results show that, in contrast to metabolism, the number of down-regulated transcripts was generally higher than that of up-regulated transcripts for each category (Fig. S3B). Down-regulated transcripts associated with cell cycle, transcription, cell growth morphogenesis, and signal transduction could reflect the reduced growth of this fungus subjected to carbon starvation, as demonstrated in Fig. 2. Although the reduced growth of Paracoccidioides during carbon starvation, the cells viability, assayed using propidium iodide, was not significantly different from those cultivated with a carbon source (Fig. S4). In the same way, the cells are metabolically active as demonstrated by the high activity of the enzyme formamidase in Paracoccidioides grown under glucose deprivation (Fig. S5).
The cellular transport process was also representative in our transcriptome analysis. The abundance of specific transporters was elevated such as those of copper, hexoses, and monosaccharides (Table S3) indicating that carbohydrate, amino acid and metal-uptake processes are required for Pb01 cells to survival under carbon starvation. Additionally, the abundance of transcripts related to cellular response against ROS (reactive oxygen species) such as superoxide dismutases, catalase and cytochrome c peroxidase were also elevated (Table S3) indicating that Paracoccidioides possibly has evolved the ability to respond to oxidative stress also under carbon starvation.
The proteomic approach was performed using NanoUPLC-MSE as previously described [47], [48]. This method has been shown to improve protein and proteome coverage compared to the conventional LC-MS/MS approach [48]. The resulting NanoUPLC-MSE protein and peptide data generated by PLGS process are shown in Fig. S6, S7 and S8. First, the false positive rates of proteins from carbon and carbon starvation data were 0.34 and 0.27%, respectively. The experiments resulted in 3,327 and 3,842 identified peptides, where 45 and 57% of these were obtained from peptide match type data in the first pass, and 19 and 14% from the second pass [50] to carbon and carbon-starving conditions, respectively (Fig. S6). A total of 17 and 14% of total peptides were identified by a missed trypsin cleavage in carbon and carbon-starving conditions, respectively, whereas an in-source fragmentation rate of the same 4% was obtained for both (Fig. S6). Fig. S7 shows the peptide parts per million error (ppm) indicating that the majority, 94.8 and 95.7%, from identified peptides were detected with an error of less than 15 ppm for carbon and carbon starvation conditions, respectively. Fig. S8 depicts the results obtained from dynamic range detection indicating that a 3-log range concentration and a good detection distribution of high and low molecular weights were obtained for the both conditions.
A total of 421 differentially expressed proteins were identified in our proteomic analysis. As previously described [59], a 1.5-fold change was used as a threshold to determine the up- and down-regulated proteins (Tables S5 and S6, respectively). Approximately 20% of them (86 proteins) were represented by miscellaneous and unclassified categories, and the remaining 80% (335 proteins) were represented by classified biological categories. The biological processes and the percentage of up- and down- regulated proteins in each classified category are shown in Fig. S9.
The proteome analysis showed that proteins associated with metabolism were also the most represented during 48 h of carbon starvation in Pb01 (Fig. S9A). The metabolism was represented by amino acid, nitrogen/sulfur, C-compound/carbohydrate, lipid/fatty acid, purines, secondary, and phosphate metabolisms. All of these subcategories showed more up- than down-regulated proteins (Tables S5 and S6). Interestingly, the nitrogen/sulfur metabolism was detected as up-regulated only at protein level (Table S5). Other categories presented a high number of regulated proteins such as translation, protein fate, energy and cell defense. On the other hand, processes involved with transcription, cellular transport, cell growth/morphogenesis, and signal transduction presented a lower number of regulated proteins in which the majority was down-regulated (Fig. S9A and B).
Thus, a large part of the proteomic response to carbon starvation in Pb01 is involved in an increase of proteins associated with metabolism and reduction of those involved with core cellular processes, in agreement with transcriptome analysis.
The responses of the Paracoccidioides, Pb01, to carbon starvation, as revealed by transcriptome and proteomic analysis, are summarized in Fig. 3, which depicts the metabolic and energy adaptation of the fungus to this stress. Pathways associated with ethanol, acetyl-CoA, oxaloacetate, and consequently glucose production were induced. Moreover, amino acid degradation supply precursors such as pyruvate, oxaloacetate, succinate and also acetyl-CoA for glucose production pathways (Fig. 3).
Specific enzymes related to ethanol production were up-regulated in the absence of carbon sources. The ethanol molecule is derived from pyruvate that, in turn, is not involved directly in oxaloacetate production because the pyruvate carboxylase (PYC) enzyme is down-regulated (Fig. 3). Ethanol measurement was performed, and the results showed that after up to 48 h under carbon starvation, a significantly higher level of ethanol was produced compared with glucose-rich cells (Fig. 4). Regarding the acetyl-CoA molecule, several enzymes associated with its production from pyruvate, via the acetaldehyde precursor, and β-oxidation were also up-regulated. Once produced, acetyl-CoA may be used by the glyoxylate shunt to generate glyoxylate and succinate molecules. This is reinforced by fact that the TCA cycle enzyme isocitrate dehydrogenase is repressed, so the acetyl-CoA pool should be consumed by the glyoxylate cycle. Additionally, succinate can be converted in oxaloacetate by enzymes from the tricarboxylic acid cycle (Fig. 3). The activity of isocitrate lyase, a representative enzyme of the glyoxylate cycle, was also determined confirming our proteomic data and reinforcing our suggested carbon flow. The analysis revealed that a significant higher specific isocitrate lyase activity was obtained after Paracoccidioides yeast cells were subjected to carbon starvation for 48 h (Fig. 5).
The oxaloacetate molecule is a key intermediate of gluconeogenesis. Once produced, gluconeogenic enzymes convert it into glucose. We detected up-regulated specific enzymes that support this suggestion in Pb01 under carbon starvation, such as phosphoenolpyruvate carboxykinase (PEPCK), fructose-1,6-biphosphatase (FBPase), and phosphoglucomutase (PGM)(Fig. 3).
In order to perform additional analysis, we applied a highly stringent criterion (≥5×-fold) to analyze the most induced or repressed proteins in yeast cells upon carbon starvation (Tables 1 and 2). Proteins which were detected only in carbon or carbon starved conditions were considered as down and up-regulated proteins at a high level (Table 1 and 2), respectively [52]. Even at this high cut-off, up-regulated proteins involved in amino acids degradation, in β-oxidation, in ethanol production, among others are present using this high stringent criterion (Table 1), in agreement with the metabolic overview presented in Fig. 3. In the same way, down regulated proteins such as pyruvate dehydrogenase and enzymes involved in fatty acids biosynthesis were detected using this highly stringent fold change criteria (Table 2). Proteins related to cell defense such as thioredoxin reductase, superoxide dismutase and cytochrome c oxidase were also detected among the up-regulated proteins.
To compare similar aspects between transcriptome and proteome data, we sought the same transcripts and proteins detected by both analyses. The transcripts and proteins identities (ID), from the Paracoccidioides, Pb01, database, were shown including their levels of abundance (Tables S7, S8, S9 and S10). Fifty seven identities (IDs) were matched of which 32 and 17 of them presented the same abundance profile, up- or down-regulated in both data, respectively (Tables S7 and S8). In this way, approximately 86% of the matches showed the same pattern of transcript and protein levels. On the other hand, the minority of IDs showed discrepancy in their abundance. Several of the transcripts in these groups were decreased in abundance, while the protein levels were increased and vice – versa (Tables S9 and S10).
A comparative analysis including all transcripts and proteins for metabolism and energy categories from RNAseq and NanoUPLC-MSE analysis was also performed (Fig. 6). Metabolism, which was the most regulated category in our data and energy are considered essential categories for understanding the carbon flow used by Pb01during carbon starvation. The results show that the amino acid, carbohydrate/C-compound, and lipid metabolism were similarly regulated in both approaches, showing the consistency with the suggested carbon flow in Paracoccidioides, Pb01, under carbon starvation (Figs. 3 and 6A). Amino acids and lipids are supposed to be intensively degraded (Tables S3 and S5) suggesting the production of precursors during carbon starvation, which include acetyl-CoA, pyruvate, oxaloacetate and succinate. Furthermore, the percentage of transcripts and proteins related to energy categories such as glycolysis/gluconeogenesis, electron transport/membrane associated energy conservation, and TCA cycle are also similar, in accordance with suggested responses of Pb01 to carbon starvation (Fig. 6B). Thus, the induction of gluconeogenesis, β-oxidation, part of TCA, and glyoxylate cycles was required to compensate for the absence of glucose and depicts the rearrangement of pathways when a carbon source condition is changed.
We investigated the response to macrophages in Paracoccidioides, Pb01, under carbon starvation. We analyzed whether the fungus differentially expresses genes involved in gluconeogenesis, glyoxylate cycle, and β-oxidation pathways after internalization by the J744 A.1 macrophages. The relative expression analysis of transcripts encoding fructose-1,6-biphosphatase, isocitrate lyase and 3-ketoacyl CoA thiolase was performed using qRT-PCR. The Fig. 7A demonstrates that genes encoding isocitrate lyase and 3-ketoacyl CoA thiolase were induced (p≤0.05), suggesting a response of Paracoccidioides to carbon starvation in phagosomes. Furthermore, whether yeast cells under carbon starvation were more susceptible to macrophage killing than cells growing in plentiful glucose was analyzed. Firstly, plating of recovered yeast cells by aspiration of culture supernatant (non-internalized yeast cells) and from lysis of macrophages (internalized yeast cells) was performed (Fig. 7B). The result showed that the number of yeast cells recovered from culture supernatant was not significantly different between carbon and carbon starved yeast cells (Fig. 7B, on the left). On the contrary, the Paracoccidioides, Pb01, yeast cells pre-exposed to carbon starvation were recovered in a lower number than those grown under carbon source (Fig. 7B, on the right). In addition, to verify if the yeast cells were, in fact, more susceptible to macrophage killing we evaluated the average number of adhered/internalized Paracoccidioides, Pb01, using the light microscopy after 6 and 24 h of infection (Fig. S10 and S11). After 24 h of infection, it was observed, in carbon-starved condition, a significant decrease in the number of yeast cells adhered/internalized by macrophages (Fig. S10). Overall, the data suggest that yeast cells pre-exposed to carbon starvation were more susceptible to macrophage killing, reinforcing our suggestion that the carbon starvation can affect the survival of the Pb01 yeast cells inside of the macrophages.
The major focus of this work is directed towards a global view on the responses of Paracoccidioides, Pb01, to carbon starvation using both, high-throughput transcriptome and proteomic analysis. Pb01 yeast cells were able to adapt to carbon starving conditions. The data presented in this study reflected how carbon-starved cells modulate the metabolism by induction or repression of cellular activities. We show that the fungus regulates pathways that lead to glucose production to compensate the effect of stress. The fungus regulates transcripts and proteins that are mainly associated with gluconeogenesis and ethanol production via precursors from β-oxidation, glyoxylate and tricarboxylic acid cycles. Our study presents a detailed response of Paracoccidioides spp. facing carbon starvation and contributes to investigations of the importance of alternative carbon adaptation during fungus pathogenesis.
Changes in the kinetics of expression of representatives of gluconeogenesis, the glyoxylate cycle, and β-oxidation as well as the differential expression of isocitrate lyase at the protein level could establish a better time-point for our transcriptional (6 h) and proteomic (48 h) analysis, using RNAseq and NanoUPLC-MSE, respectively (Fig. 1). The transcriptome and proteomic analysis demonstrated that general metabolism and energy were the most represented regulated categories. Transcripts/proteins classified in energy and cell rescue, defense, and virulence categories were also induced in both approaches although less representative than metabolism (Fig. S3 and S9). Categories involved in the cell cycle, transcription, cellular transport, growth/morphogenesis, and signal transduction were predominantly down-regulated at transcription and protein levels. Although Pb01yeast cells displayed no significant difference in cell viability until 72 h under carbon starvation (Fig. S4), the yeast biomass was significantly reduced in this condition (Fig. 2). In addition, the abundance of specific transporters was elevated such as those related to copper, hexose, and monosaccharide uptake, suggesting a nutrient limitation and a hostile environment as detected in other fungi [60].
Paracoccidioides, Pb01, presented a complex mechanism to respond to nutrient deprivation. The suggestion is that the fungus uses a carbon flow (Fig. 3) through classical biochemical pathways such as glyoxylate cycle, β-oxidation, and gluconeogenesis which is in accordance with previous data on other fungi facing nutrient deprivation [8], [9], [12]. Fungi such as C. albicans and C. neoformans appear to experience a nutrient limited and stressful environment in the context of interaction with host cells. The elevated expression of the glyoxylate pathway and gluconeogenesis genes during Cryptococcus interactions with host tissue and phagocytic cells is similar to the regulation observed in C. albicans. These pathogenic fungi present a niche dependent metabolism, with activation of an alternative carbon source consuming process and the up-regulation of transcripts for enzymes of the glyoxylate cycle, β-oxidation, and gluconeogenesis [9], [12] which was also detected in Paracoccidioides, Pb01. Our data suggest that Pb01 yeast cells facing carbon starvation use the oxaloacetate molecule as a key intermediate of gluconeogenesis. It is possibly supported by β-oxidation and, in part, by glyoxylate and TCA cycles activation. The glyoxylate cycle can allow the fungus to assimilate two-carbon compounds, a relevant aspect to the viability and growth inside macrophages [61], [62]. Studies involving the isocitrate lyase gene, representative of glyoxylate cycle, displayed that this gene is important as a marker for gluconeogenic carbon source utilization and starvation rather than a marker for lipid metabolism [63], [64]. Despite induction of isocitrate lyase and genes required for fatty acid utilization especially after phagocytosis by macrophages [9], this induction may derive from a general stress response due to nutrient or glucose limitation rather than a specific induction from fatty acid utilization [64]. In fact, null mutants to isocitrate lyase in C. albicans, A. fumigatus and C. neoformans and for β-oxidation genes in C. albicans revealed no virulence defects, showing that fatty acids do not provide an essential nutrient source during infection [2], [64]–[67]. In addition, isocitrate lyase activity increased when C. albicans was subjected to carbon starvation or other carbon sources such as acetate, glutamate and peptone as solely carbon source [63] reinforcing its importance as a marker for gluconeogenic carbon source utilization and starvation. Moreover, the isocitrate dehydrogenase enzyme was detected as down-regulated in our proteomic data (Table S6). The activities of isocitrate dehydrogenase and isocitrate lyase enzymes, regulate the flow of isocitrate into either the tricarboxylic acid cycle or the glyoxylate cycle [68], [69]. Here, while the isocitrate dehydrogenase is down, the isocitrate lyase is up-regulated, in accordance with suggested carbon flow in Pb01 through glyoxylate shunt. As the same importance, the increased expression of protein phosphoenolpyruvate carboxykinase confirms that gluconeogenesis process is ongoing (Table S5). Then, the global characterization of the responses of Pb01 to carbon starvation becomes relevant in this context especially by flow of carbon used by the fungus during this stress.
The same transcripts and proteins detected by both analyses were identified (Tables S7, S8, S9 and S10). Approximately 86% of the matches (a total of 57) showed the same pattern of transcript and protein levels. In this way, we believe that the transcriptional and proteome time-points were enough to characterize the global responses of the Paracoccidioides to carbon starvation conditions. Similarly to previous study in A. fumigatus, few times, transcripts and proteins do not follow the same trend of expression that could be explained, for example, by mRNA stabilization process or by active post-transcriptional and translational regulatory mechanisms [70]. In our data, in many metabolic aspects, transcripts and protein levels were correlated. The identities to amino acid degradation, β-oxidation and ethanol synthesis were increased in expression while processes involving the pyruvate molecule were down-regulated (Table S7 and S8) corroborating our hypothesis of a well established response to carbon starved environments (Fig. 3). The pyruvate-acetyl-CoA conversion, for example, is diminished by repression of pyruvate dehydrogenase enzyme. In addition, pyruvate carboxylase is repressed and it is possibly not converted in oxaloacetate. These observations strongly suggest that the available pyruvate would end up in ethanol via acetaldehyde (Fig. 3).
In addition, comparison of regulated molecules in transcriptome and proteome data with focus in metabolism and energy categories can support the use of alternative carbon sources by Pb01 under carbon starvation (Fig. 6). Metabolism of amino acids, lipids, and carbohydrate/C-compound was the most regulated in both used approaches (Fig. 6A). The amino acids and lipids likely are been used as precursors to important molecules involved in alternative carbon metabolism in Pb01 as depicted in Fig. 3. In fact, Paracoccidioides spp. can use a relatively wide range of amino acids and peptides rather than carbohydrates [21]. In addition to the structural importance of lipids, these molecules provide an energy-rich nutrient source. β-oxidation is a common pathway for the utilization of fatty acids [71] in which of the 3-ketoacyl-CoA thiolases enzymes are important [64]. Recent studies have highlighted the relevance of the β-oxidation in response to nutrient or glucose limitation rather than a specific induction from fatty acid utilization. In fact, fatty acids do not provide an essential nutrient source during infection in C. albicans but is important for coupling the glyoxylate cycle and fatty acid β-oxidation during host-pathogen interactions, regulating responses related to carbon starvation [63], [64]. Here, the induction of β-oxidation pathway in Pb01 likely reflects the requirement of this metabolic pathway for carbon starving adaptation, which is consistent with previous data in C. albicans subjected to a poor-nutrition environment [9]. Regarding energy producing pathways, the gluconeogenesis, tricarboxylic acid and glyoxylate cycles are well represented, which reinforces our model of adaptation to carbon starvation conditions (Fig. 6B). Moreover, the electron transport and ethanol subcategories were also shown. Ethanol metabolism was previously described in Pb01 showing evidence for a more anaerobic metabolism of this fungus compared with other isolates of Paracoccidioides [35], [36], [72], and this metabolite has also been described as relevant to pathogenic fungi such as A. fumigatus [70], [73], [74].
The general response to oxidative stress mediated by enzymes provides multiple resistance strategies to Paracoccidioides yeast cells [34]. There was a high increase in production of antioxidant proteins such as thioredoxin reductase, superoxide dismutase and cytochrome c peroxidase that are possibly involved in defense against reactive oxygen species (ROS) during carbon starvation (Tables S5 and 1). In low amounts, ROS are generated continuously as side products of aerobic respiration in the mitochondria and, although potentially cytotoxic, function as signal molecules in cellular processes [75], [76]. In this way, the production of ROS during carbon starvation could be related to the increase in electron transport activity in respiratory chain (Table S5 and 1) as well as to the production of endogenous free radicals in β-oxidation.
The response of Pb01 to macrophage infection shows that the fungus most likely faces carbon starvation in macrophages, because a significant higher expression of genes encoding isocitrate lyase and 3-ketoacyl-CoA thiolase was detected. It is important to note the similarities in the transcriptional profile in inducing alternate carbon metabolism between C. albicans phagocytosed cells and those submitted to carbon starvation (39). In fact, in terms of usable nutrients, the phagosome has been reported to not have a rich environment evidenced by the unsubstantial quantities of glucose, other sugars and amino acids [9], [11], [12], [27], [77]. Here, we showed that Pb01 yeast cells were more susceptible to macrophage killing when were previously starved of carbon (Fig. 7, S10 and S11). This conclusion is based on the fact that the multiplication of Paracoccidioides inside activated macrophages is inhibited [23]. In this sense, we suggested that Paracoccidioides, Pb01, yeast cells pre-exposed to carbon starvation, were more susceptible to macrophage killing, reinforcing that carbon deprivation affects the survival of the Pb01 yeast cells inside of the macrophages.
Taken together, our data suggest that Pb01 changes its metabolism under carbon starvation reprogramming several biological processes to facilitate its maintenance under this condition. These programs are mainly related to gluconeogenesis, β-oxidation and the glyoxylate cycle to compensate for the starved carbon environment. Considering a new perspective, the transcriptome and proteome data could reinforce the responses of this fungus, which is able to survive in the hostile environment during macrophage infection. This study may elucidate potential molecules involved in host-fungus interactions, an important factor related to pathogenic organisms.
|
10.1371/journal.pgen.1002610 | Accurate Prediction of Inducible Transcription Factor Binding Intensities In Vivo | DNA sequence and local chromatin landscape act jointly to determine transcription factor (TF) binding intensity profiles. To disentangle these influences, we developed an experimental approach, called protein/DNA binding followed by high-throughput sequencing (PB–seq), that allows the binding energy landscape to be characterized genome-wide in the absence of chromatin. We applied our methods to the Drosophila Heat Shock Factor (HSF), which inducibly binds a target DNA sequence element (HSE) following heat shock stress. PB–seq involves incubating sheared naked genomic DNA with recombinant HSF, partitioning the HSF–bound and HSF–free DNA, and then detecting HSF–bound DNA by high-throughput sequencing. We compared PB–seq binding profiles with ones observed in vivo by ChIP–seq and developed statistical models to predict the observed departures from idealized binding patterns based on covariates describing the local chromatin environment. We found that DNase I hypersensitivity and tetra-acetylation of H4 were the most influential covariates in predicting changes in HSF binding affinity. We also investigated the extent to which DNA accessibility, as measured by digital DNase I footprinting data, could be predicted from MNase–seq data and the ChIP–chip profiles for many histone modifications and TFs, and found GAGA element associated factor (GAF), tetra-acetylation of H4, and H4K16 acetylation to be the most predictive covariates. Lastly, we generated an unbiased model of HSF binding sequences, which revealed distinct biophysical properties of the HSF/HSE interaction and a previously unrecognized substructure within the HSE. These findings provide new insights into the interplay between the genomic sequence and the chromatin landscape in determining transcription factor binding intensity.
| Transcription factors (TFs) bind DNA to modulate levels of gene expression. TF binding sites change throughout development, in response to environmental stimuli, and different tissues have distinct TF binding profiles. The mechanism by which TFs discriminate between binding sites in a context dependent manner is an area of active research, but it is clear that the chromatin environment in which potential binding sites reside strongly influences binding. This study used the Heat Shock TF (HSF) to study the effect chromatin has upon induced HSF binding. We implemented an experimental technique to quantify all potential HSF binding sites in the genome. These data were incorporated into a modeling framework along with chromatin landscape information prior to HSF binding to accurately predict the intensities of inducible HSF binding sites. DNase I hypersensitivity and tetra-acetylation of H4 were the most influential covariates in the model. The binding data enabled the development of a more complete HSF/DNA interaction model, providing insight into the biophysical interaction of HSF trimer subunits and target DNA pentamers.
| Binding of transcription factors (TFs) to DNA elements is necessary to establish and maintain functional changes in gene expression levels. The mechanism by which these factors seek out and bind to their cognate motif elements remains an area of active investigation (reviewed in [1]). TFs are present at cellular concentrations that allow binding to sites that are degenerate from the consensus sequences, and genomes of eukaryotes are littered with potential degenerate binding sites; however, only a small fraction of potential binding sites are recognized in vivo. Moreover, TF binding sites vary dependent upon cell type and cellular conditions. In vivo, TF binding is potentially dependent upon motif accessibility and the surrounding chromatin landscape. Therefore, determining a comprehensive set of potential genomic binding sites and quantifying the joint effects of DNA sequence and chromatin landscape upon binding intensity remains a challenge.
Experimental approaches to characterize TF binding sites include assays such as ChIP-seq, protein binding microarrays (PBM) [2], iterative rounds of protein-DNA binding and selection with a complex oligonucleotide library [3], or extrapolation from DNase I hypersensitivity regions [4]. However, perhaps the most direct way to determine all potential TF binding sites within a genome is to incubate purified TF and naked sheared genomic DNA in vitro, and then specifically quantify the TF-bound DNA [5]. This in vitro method allows binding sites to be interrogated in their native sequence context without the confounding effects of chromatin and cooperation between chromatin-bound factors.
It is challenging to predict in vivo TF binding accurately even when all potential in vitro binding sites have been characterized, because the chromatin landscape dramatically influences binding and it changes dynamically with development and with alterations in cellular nutrition and environment [6], [7]. Recent TF binding site modeling efforts have considered genomic nucleosome occupancy or DNase I hypersensitivity data to account for the effect chromatin has upon in vivo TF occupancy [8]–[11]. However, these models are limited in that they rely upon genomic accessibility data and TF binding data produced under the same conditions. To date there are no data sets that describe the full set of potential TF binding sites, the chromatin state data prior to binding, and occupied binding sites in vivo, in a single inducible system. Integration of these three data sets would allow one to decouple the effect TF binding has upon chromatin state from the effect pre-existing chromatin state has upon induced TF binding.
The heat shock response of Drosophila is a model system extensively used to study the general functions of sequence specific activators and how they function to regulate transcription (reviewed in [12]). The master regulator of the heat shock genes, Heat Shock Factor (HSF), has a modest affinity for DNA under non-stress conditions [6], [13], [14], and upon stress, HSF homotrimerizes and inducibly binds to a conserved consensus motif at over 400 sites in the Drosophila genome [6], [14]. While over 95% of the HSF binding sites contain an underlying HSF sequence motif element (HSE), the vast majority of predicted genomic HSEs remain HSF–free following heat shock. Therefore, the chromatin landscape most likely plays a prominent role in determining binding of HSF.
Here, we describe an experimental technique, protein/DNA binding followed by high-throughput sequencing (PB–seq), to quantify the binding potential of all binding sites within a genome. We then develop a quantitative model that incorporates HSF PB–seq data, together with HSF ChIP-seq in Drosophila S2 cells [6] and S2 cell chromatin data, that accurately predicts observed in vivo HSF binding profiles. Moreover, our model allows us to quantify the relative importance of the chromatin features influencing HSF binding intensity. Finally, we develop a sequence model that uses HSF PB–seq data to characterizes the relationship between positions within the HSE and provide biophysical insight into the mechanisms by which HSF interacts with its cognate element.
We performed an in vitro binding experiment with purified HSF (Figure S1) and naked, sheared genomic Drosophila DNA, to derive an accurate set of potential HSF binding sites in the Drosophila genome. HSF–bound DNA was specifically eluted and detected by high throughput sequencing (Methods). The HSF PB–seq experiment yielded 68% of the sequence tags within peaks. In contrast, typical ChIP-seq protocols are more inefficient and the majority of DNA (60% to >99%) sequenced is uninformative background DNA [15].
Peak calling revealed 3952 HSF–binding peaks (p<0.01; 2848 peaks were common to both experimental replicates), which include 60% of the previously identified high-confidence HSF binding peaks in vivo [6]. The naïve expectation is that every in vivo HSF peak should have a corresponding in vitro peak, but it is not surprising to observe an incomplete overlap of in vivo by in vitro peaks, for various reasons. As will be discussed, binding sites detected in vivo but not in vitro tend to be more degenerate and have higher DNase I accessibility. Additionally, in vivo binding sites that are dependent upon cooperative interactions with pre-bound chromatin factors, long range DNA interactions, post-translational modifications of HSF [16], higher-order chromatin structure, or bridging protein interactions [17] will not be detected in the current form of PB–seq.
Underlying the in vitro binding peaks, we detected 3735 clusters of HSF binding site HSE sequences (2896 in peaks common to both replicates) at 20% HSE False Discovery Rate (FDR). We used clusters of co-occurring sites due to the uncertainty in HSE detection (see Methods). Furthermore, the majority, 3389 clusters (2586 in peaks common to both replicates) are not detectably bound in S2 cells in vivo. Figure 1 shows two examples of in vitro binding sites flanking the Cpr67B gene that are not bound in vivo. Moreover, the in vitro binding data quantifies differences in the in vitro and in vivo HSF binding intensity, such as the peaks within each of the promoters for Hsp23 and Hsp26 (Figure 1).
The PB–seq experiment allows for an estimate of the relative binding intensity of each HSE, based on the number of sequence tags associated with it. To compute the dissociation constant (Kd) values it is necessary to have estimates for both the fraction of bound and free HSE in the PB–seq experiment. Since the PB–seq data only provides information on the bound fraction, we needed to determine the absolute Kds for two HSEs that are found within the PB–seq data in order to provide enough information to estimate the free fraction (see Methods).
To generate the HSF/HSE Kd measurements, we performed electrophoretic mobility shift assays (EMSA). The EMSAs were performed with purified HSF and HSEs that are only modestly degenerate from the consensus. We found that HSF binds to the first HSE with ∼42.6 pM interval: 36.8–49.4 pM; Figure 2A and 2C) and the second HSE with ∼224 pM affinity (95% confidence interval: 181–276 pM; Figure 2B and 2D). The resulting two absolute Kd values enabled us to transform PB–seq read depths into absolute Kd values (Figure 2E and Methods). We confirmed the transformation of the relative Kd values to absolute Kds by performing band shifts with genomic HSEs of different predicted Kd values (Figure S2). The experimental verifications of the measurements are within the estimated error of the EMSA confidence interval and the variability between PB–seq replicates (Figure S3).
Taken together, these measurements allow us to characterize the binding energy landscape for HSF across the entire Drosophila genome, in the absence of chromatin. Our estimated Kd values for isolated HSEs in the Drosophila genome ranged from 40–400 pM (Figure 2E). These in vitro binding results demonstrate the feasibility and efficiency of combining high-throughput detection methods with classic EMSA and competition experiments to quantify the binding energy for the comprehensive set of potential genomic binding sites for a sequence-specific TF.
Our data reveals substantial differences between in vivo and in vitro binding intensities (Figure 3A), underscoring the role of chromatin in determining in vivo binding site selection and affinity. We found DNase I hypersensitivity was the most important predictor of HSF binding; therefore, we scaled the in vivo and the in vitro read counts so that they were approximately equal at in vivo sites with high DNA accessibility (Methods, Figure S4). After this normalization, we partitioned the binding sites that were detectable in vitro into classes: “unaffected” sites, bound at comparable affinities in vivo and in vitro (55 red points in Figure 3A; 2% of all sites); “suppressed” sites, with reduced, but detectable, in vivo intensity (365 green points; 13%); and “abolished” sites, below the in vivo threshold for detection (2223 blue points; 76%). In addition, sites not detectable in vivo or in vitro were labeled “background” (249 gray points; 9%), and sites with stronger relative in vivo intensity compared to in vitro were labeled “enhanced” (4 black points; 0.1%).
PB–seq data reveals potential HSF binding sites, providing the opportunity to model the effect that non-stressed chromatin landscape has upon induced HSF binding intensity. There is a wealth of chromatin data available for S2 cells during unstressed conditions [18], [19], and heat-shock induced binding sites of HSF in S2 cells are also known [6]. We used DNase I hypersensitivity data [18], MNase data [19] and ChIP-chip data for 9 factors and 21 histone modifications for unstressed Drosophila S2 cells (Table S1) [18] to predict the intensity of inducibly bound in vivo HSF–bound sites (Figure 4A, Figure S5 and Figure S6). For our statistical model, we selected a rules ensemble [20], a linear regression model in which some terms are combinations of covariates known as “rules”. This approach allowed us to capture fairly complex interactions between covariates. For example, a rule might apply when H3K27ac and DNase I hypersensitivity both exceeded designated thresholds (value ranges can also be expressed). Each rule's coefficient is added to the predicted value if, and only if, the rule applies. When there is only one covariate, the model reduces to a linear regression.
The Pearson's correlation coefficient between HSF ChIP-seq data for the model incorporating all these data sets was r = 0.62 (Figure S6 and Figure S7). As the large number of covariates brings with it some danger of overfitting, we tested combinations of the four classes of covariates: DNase I hypersensitivity, MNase, histone modifications/variants, and non-histone factors (Figure 4B, Figure S6, Figure S7). Of notice, the correlation of the linear regression model that incorporates DNase I data was r = 0.64 on the test data (Figure 4B and Figure S7B). Our study is consistent with a previous study that obtained r = 0.65 for actual and inferred TF binding intensities using a DNase I dependent model [8].
Other covariate classes produce similar, but lower, correlations. The rules model using histone modifications and histone variants yielded r = 0.57 (Figure 4B and Figure S7), while a rules model incorporating non-histone bound chromatin factors yielded r = 0.58 (Figure 4B and Figure S7). Combining covariate classes further improves the correlation to as much as r = 0.70 (Figure S6 and Figure S7). We also examined the Receiver Operator Curves (ROC) for the different covariate combinations (Figure S8) and found concordant results. If we assume that the PB–seq, genomic ChIP, DNase I-seq, and MNase-seq experiments are maximally resolved and sensitive, with no experimental noise, an approximate upper bound is given by r = 0.90, as observed for two HSF–ChIP-seq replicates [6]. Notably, the higher resolution of the DNase I-seq data, compared to the ChIP-chip data, may be why DNase I-seq alone is strongly predictive in the linear regression model and most influential in the rules ensemble models. Notably, we used the chromatin landscape prior to induced TF binding to predict binding intensity, whereas previous models have used the chromatin landscape present when the TF is bound in order to infer binding intensity [8] or infer binary binding events [10], [11] (see Discussion).
Our data and modeling indicated that the presence of active chromatin features, such as histone acetylation and DNase I hypersensitivity, had a significant influence on the predictive power of the model, while repressive features had minimal influence (Figure S9). DNase I hypersensitivity was a strongly predictive covariate in the model when used in a simple linear regression model (Figure 4), or in combination with histone modification and non-histone factor covariates in the rules (Figure S9E–S9G, S9J, S9K, and S9M). Tetra acetylation of H4 and H3K9ac were the most informative histone marks in the model that used histone variants and histone modifications as covariates (Figure 5A). GAGA associated factor (GAF), which has a proposed role in permitting HSF binding [21], was the most influential factor in the HSF binding prediction model that considered all chromatin-binding factors (Figure 5B).
The analysis above indicates that DNA accessibility, as measured by DNase I hypersensitivity, is a primary determinant of binding intensity. Previous studies have similarly shown that TF binding sites correlate strongly with DNase I hypersensitive sites [8], [10], [11], [22]. For instance, histone acetylation causes local chromatin decondensation by reducing the ionic interactions between lysine residues and DNA and promotes accessibility, but the extent to which combinations of histone marks and TFs act together to dictate chromatin accessibility is not known. Therefore, it is of interest to see whether DNA accessibility can be predicted from specific features of the chromatin landscape, such as histone modifications and non-histone chromatin bound factors. In addition, accurate predictions of DNA accessibility would be of practical use, because direct measurements are often not available.
To address this question, we applied our rules ensemble framework to predict DNase I hypersensitivity (the best available proxy for DNA accessibility) from ChIP-chip data for histone features, non-histone chromatin bound factors, MNase data and combinations of these covariate pools (Figure 6). Tetra-acetylation of H4 and H3K9 acetylation were most influential in the model that uses histone modifications, bulk histone and histone variant intensities (Figure S10E); the correlation coefficient for this model using the test data is 0.51 (Figure S11B). The model that uses non-histone factor ChIP-chip data obtains a correlation of 0.52 (Figure S11B), which is consistent with TFs having characteristic DNase I hypersensitivity footprints [10], [11]. The model that combines both histone data and non-histone data into a rules model performs the best on the test set, with a correlation of 0.60 (Figure S11B). Repressive histone marks appear to contribute little to generating the DNase I hypersensitivity pattern (Figure S10) and the lack of active chromatin marks appears to be sufficient to package DNA into inaccessible units. These models reinforces the notion that the biochemical composition of chromatin permits DNase I hypersensitivity and quantifies the contributions individual modifications, and combinations thereof, make to DNase I hypersensitivity (Figure S11). As more and higher-resolution genome-wide data becomes available, these models will be refined.
PB–seq provides the opportunity to model the sequence-dependent binding preferences of a purified TF genome-wide and independent of chromatin or other factors. In the case of HSF, the consensus binding site is well characterized and consists of three pentamers, ÒAGAAN NTTCT AGAANÓ, (here denoted pA, pB, and pC), each bound by a monomer of the HSF homotrimer. Note that the consensus sequences for pA and pC are identical, while the one for pB is their reverse complement. Of course, the consensus HSE is a crude summary that ignores subtleties in the base preferences at each position. A position-specific scoring matrix (PSSM) provides a somewhat improved description but still ignores dependencies between positions within the binding site. We sought to use genome-wide binding sites from PB–seq to produce an improved model for the sequence preferences at HSEs.
We began by computing the mutual information for all pairs of HSE positions based on the identified in vitro binding sites. We found negligible evidence of correlated base preferences between positions, but we did observe that some pentamers within PB–seq peaks adhered closely to the consensus motif while others did not. This led us to formulate a probabilistic model that allows each pentamer in an HSE to closely match the consensus (“strict”) or diverge from it more substantially (“relaxed”), and considers all possible combinations of pentamer composition (Figure S12). More specifically, we described each of the three pentamers using a two-component mixture model, with a latent variable indicating “strict” or “relaxed” binding preferences, and estimated the joint distribution of these three latent variables from the data.
The model parameters—the position-specific nucleotide probabilities and prior distribution for the combinations of strict/relaxed pentamers—were estimated from the data by maximum likelihood using an expectation maximization algorithm (see Methods). In fitting the model, we considered only the 1309 isolated HSEs, sequence elements that were at least 200 base pairs away from any other degenerate HSE motif, to avoid complications arising from overlapping HSEs. The model fit the data substantially better than did a simple PSSM (lnL = −15442 vs. lnL = −15673 for the PSSM; Akaike information criterion [AIC] = 15636 vs. AIC = 15763 for the PSSM), suggesting that it effectively captures important dependencies between positions.
Based on the estimated model parameters, we computed a posterior probability distribution over all combinations of pentamer stringency and order for each HSE (Methods; Figure 7B). These values were averaged across HSEs to obtain expected genome-wide fractions of HSEs having each of the strict/relaxed pentamer combinations. We found that binding sites with strict pB and pC, and relaxed pA, were most frequent (an expected 38% of sites), indicating that this configuration is preferred (Figure 7B). The next most frequent configurations were a relaxed pB flanked by a strict pA and pC (33%), and a strict pA and pB combined with a weak pC (29%). Interestingly, combinations of three strict pentamers occur at negligible frequency. Indeed, only 5 out of 1309 isolated genomic HSEs matched the consensus sequence exactly, while 148 differed from it by a single mismatch. Configurations with at most one strict pentamer were also rare. Together, these results indicate that the biophysical interactions of the pentamers within the binding sites are critically dependent upon their composition and position relative to the other pentamers in an HSE.
While the three estimated strict pentamer matrices were similar (Figure 7A top), the relaxed matrices showed substantial differences with respect to each other (Figure 7A bottom). For example, the relaxed pA matrix indicates that 70–80% of HSEs containing a weak pA have the consensus base at positions two, three and four. In contrast, position 12 in pC (the analog of position 2 in pA) almost invariably contains a G in all HSEs, while positions 7 and 8 in pB (analogous to positions 3 and 4 in pA) have only modest base preferences in HSEs containing a weak pB. This analysis indicates that each monomeric HSF/pentamer interaction has distinct biophysical properties within the context of the broader HSF/HSE interaction.
We also devised a simplified model, with a single strict matrix shared by all three pentamers, and a single relaxed matrix obtained by applying a “dampening” factor to the strict matrix (Figure S13, Methods). This model further supports the strict/relaxed pentamer split (lnL = −15908 vs. lnL = −16048 for a single-monomer PSSM; and AIC = 15952 vs. AIC = 16078), although both the full model and the full PSSM fit the data better (lower AIC). Moreover, not only was the simplified model still able to reproduce the posterior distributions over pentamer configurations of the full model, but it was also able to replicate synthetic patterns from simulated data (Figure S14).
Finally, the preference for single pentamer degeneracy was also observed independently by comparing the pentamer-specific KL-divergence in PSSMs obtained from subsamples of HSF bound peaks (Figure S15; Methods).
The PB–seq technique combined with EMSA and competition assays provides a straightforward, yet versatile and powerful framework for characterizing all potential binding sites in a genome, regardless of tissue specificity, developmental stage, or environmental conditions. Comparing in vitro and in vivo binding profiles, in the context of pre-induction genomic chromatin landscape, revealed DNase I hypersensitivity, H4 tetra-acetylation, and GAF as critical features that modulate cognate element binding intensity in vivo. Furthermore, DNase I sensitivity was found to be strongly influenced by high GAF occupancy and histone acetylation, while repressive factors were minimally influential in the statistical models. Finally, the full set of potential genomic binding sites provided a rich data set that was used to build more detailed sequence models, which tease apart substructure and features that are lost with traditional PSSM models.
One initially surprising observation from our study was that 40% of the in vivo HSF peaks were not found in vitro. We believe that the limited dynamic range for quantifying in vitro binding affinity may be responsible for the lack of detectable in vitro peaks. Although we quantify in vitro binding over an order of magnitude (40–400 pM), the experimental concentrations of HSF and genomic DNA and our depth of sequencing do not permit the detection of lower affinity HSF binding sites. For instance, only eleven sequence tags would be predicted to underlie a hypothetical 5 nM HSF binding site, and these would not be distinguishable from background. Upon further examination, we find that the composite HSE representing those in vivo binding sites that were not found in vitro is more degenerate than those found using both assays (Figure S16A). Moreover, the in vivo sites that were not found using PB–seq were also more accessible in vivo (Figure S16B), in support of our hypothesis. Performing PB–seq at a range of protein and DNA concentrations, or increasing sequence coverage would expand the dynamic range of quantification by PB–seq.
Other possible explanations for this observation include cooperative interactions with pre-bound chromatin factors, long-range DNA interactions, post-translational modifications of HSF, higher-order chromatin structure, or bridging protein interactions. The influence of DNA modifications and immediate flanking sequence do not contribute to this disparity, since we use large fragments of purified genomic DNA. Bridging protein interactions [17], which do not involve HSF directly binding to DNA, appear not to be responsible for our results because 95% of in vivo peaks encompass at least one HSE near the peak center [6]. However, if other proteins were cooperating with HSF in vivo to enhance HSF binding intensity at low affinity binding sites, then some of these peaks may not be observed in vitro. Since our PB–seq experiment used recombinant HSF in the binding experiments, we would also not capture differences in binding site affinities that are due to post-translational modifications of HSF [16]. To overcome these potential limitations, PB–seq could be adapted to include known bridging/cooperative factors and proteins could be purified from in vivo sources to capture indirect or modification-dependent interactions.
The notion that motif accessibility is driving inducible TF binding in vivo is supported by independent studies of distinct TFs: STAT1, HSF, glucocoticoid receptor (GR), and GATA1 [6], [22]–[24]. These studies show that the chromatin landscape prior to TF binding influences inducible TF binding. In the first study, it was found that a large fraction of STAT1 induced binding sites contained H3K4me1/me3 marks prior to interferon-gamma (IFN-γ) induced STAT1 binding [23]. Our group previously found that inducible HSF binding sites are marked by active chromatin compared to sites that remain HSF–free [6]. A more recent study has shown that inducibly bound GR sites are marked by DNase I hypersensitive chromatin prior to GR binding [22]. Likewise, the permissive chromatin state at GATA1 binding sites is established even in GATA1 knock out cells [24]. While these correlations are instructive, no previous attempt has been made to model inducible TF binding using biological measurements of chromatin landscape present prior to TF binding. Recent models have successfully inferred TF binding profiles using DNA sequence and chromatin landscape data, generated at the same time the TF is bound [8]–[11]. However, these models do not distinguish between the influence TFs have upon local chromatin and the chromatin features that permit TF binding. In contrast, we modeled the changes between HSF in vitro binding (PB–seq) and in vivo binding (ChIP-seq) landscapes as a function of the non-heat shock chromatin state. This produced a quantitative model describing the important features that modulate the in vivo HSF binding intensity. Moreover, the use of our rules ensemble model enabled the capture of potential interactions between these chromatin features.
Our study reveals that DNase I hypersensitivity and acetylation of H4 and H3K9 are strong predictors of inducible HSF binding intensities, however the molecular events and factors that precede TF occupancy to maintain accessible chromatin remain poorly characterized. For instance, the degree to which pioneering factors or flanking DNA sequence, individually or in combination, maintain or restrict accessibility remains unclear. A recent study highlights the biological consequences of maintaining the inaccessibility of TF binding sites, in order to repress expression of tissue-specific transcription factors in the wrong tissues. The authors found that ectopic expression of CHE-1, a zinc-finger TF that directs ASE neuron differentiation, in non-native C. elegans tissue is not sufficient to induce neuron formation [25]. However, combining ectopic CHE-1 expression with knockdown of lin-53 did modify the expression patterns of CHE-1 target genes in non-native tissue, effectively converting germ line cells to neuronal cells [25]. LIN-53 has been implicated in recruitment of deacetylases, and deacetylase inhibitor treatment mimics lin-53 depletion, suggesting that LIN-53 is actively maintaining CHE-1 target sites inaccessible in germ cells.
Alternatively, functional TF binding sites could be actively maintained in the accessible state. HSF binding within ecdysone genes has a functional role in shutting down their transcription [14], and activating ecdysone-inducible genes containing inaccessible HSEs causes chromatin changes that are sufficient to allow HSF binding [6]. In this special case of HSF–bound ecdysone genes, active transcription and the corresponding histone marks are mediating access to HSEs, in order for HSF to bind and repress transcription upon heat shock. A more recent study has shown that activator protein 1 (AP1) actively maintains chromatin in the accessible state, so that GR can bind to cognate elements [26].
Although TF accessibility to critical genomic sites appears to be actively maintained, many binding sites may be a non-functional result of fortuitous TFBS recognition. It has long been hypothesized that the binding affinities for TF/DNA interactions are sufficiently strong to allow promiscuous binding at the cellular concentrations of TFs and DNA [27], [28]. There are roughly 32,000 HSF molecules per tetraploid S2 cell [29] and the dissociation constants for trimeric-HSF/HSE interactions are in the picomolar range (Figure 2E); therefore much of the in vivo HSF binding may be non-functional promiscuous binding. Additional investigation will further illuminate the role of chromatin context in TF binding and the mechanisms by which programmed developmental or environmental chromatin changes permit or deny TF binding.
Elucidating the rules that govern accessibility is essential for predicting in vivo occupancy of TFs. Diverse transcription factors [7], from a broad spectrum of organisms [22], bind their sequences based on site accessibility. We found that chromatin accessibility as measured by DNase I hypersensitivity could be inferred using ChIP-chip data for various histone modifications and transcription factors. Although our model can infer accessibility based on chromatin composition, the mechanism by which accessibility originates is not addressed. Previous studies have shown that activators, such as HSF, glucocorticoid receptor, and androgen receptor bind to their cognate sites and direct a concomitant increase in local acetylation, DNase I hypersensitivity, and nucleosome depletion [6], [22], [30], [31]. Androgen receptor also acts to position flanking nucleosomes marked by H3K4me2 [31]. These post-TF binding chromatin changes that occur are the result of acetyltransferase and nucleosome remodeler recruitment, both of which functionally interact with activators. For instance, both GR and GATA1 interact with the nucleosome remodeling complex Swi/Snf [32], [33]. Concomitant increases in locus accessibility likely allow large molecular complexes such as RNA Pol II and coactivators to access the region that in turn can reinforce and maintain active and accessible chromatin.
Thorough biophysical characterization of TF binding site properties is critical for accurate predictions of TF binding sites, underscoring the need for more complete models of TF binding. While the commonly used PSSM model makes the assumption of base independence, recent work has revealed that richer models providing for interactions between positions are necessary [34], [35]. Our model captures critical features of the HSF/HSE interaction that are lost with simpler computational models, namely the interdependencies between the sub-binding sites of each HSF monomer. Consistent with our model, a series of in vitro experiments with S. cerevisiae, D. melanogaster, A. thaliana, H. sapien and D. rerio HSFs indicate that HSF from each of these species can bind to discontinuous HSEs containing canonical pentamers that contain intervening five base pair gaps [36], [37]; interestingly, however, C. elegans HSF strictly binds to continuous HSEs that do not contain gaps [36]. The complex interactions between positions within a binding site are a critical aspect of inferring whether a polymorphism or mutation affects TF binding. These features should prove useful in providing degenerate HSE sequences for optimal co-crystallization of trimeric HSF and DNA and inferring changes in DNA sequence that affect HSF binding within and between species.
In conclusion, the data and models presented here reinforce both the importance of chromatin landscape in modulating in vivo TF binding intensity and how genome wide, chromatin free, binding assays contribute to the understanding of TF sequence binding specificity.
Drosophila HSF was N-terminally tagged with glutathione s-transferase and a tobacco etch virus (TEV) protease cleavage site. The C-terminus of the recombinant HSF was fused to the 3xFLAG epitope. Recombinant HSF was purified from E. coli with glutathione resin as previously described [38], with the following modifications: HSF–3xFLAG elution was achieved by addition of 6xHistidine tagged TEV protease and TEV protease was cleared from the HSF preparation using a Nickel-NTA column. Densitometry was used to show that the HSF protein preparation was 40% full length HSF–3xFLAG, and known amounts of bovine serum albumin (BSA) were used to quantify the HSF (Figure S1).
Serial two-fold dilutions of recombinant HSF, from 3 nM (1.5 nM for the 221 pM HSE) to 23.3 pM, was incubated with 200 attomoles of radiolabeled dsDNA containing modestly degenerate HSEs (chrX:3380775–3380824 (224 pM), chr2L:5009892–500994 (42.7 pM), chr2R:3529792–3529841 (308 pM), chr3L 13470978–13471009 (221 pM), and chr3L:4073542–4073591 (97.5 pM)) and allowed to come to equilibrium for 30 minutes in a total of 10 µl of 1xHSF binding buffer (20 mM HEPES pH 7.9, 10% glycerol, 1 mM EDTA, 4 mM DTT, 3 mM MgCl2, 100 mM NaCl, 0.1% NP-40, and 300 µg/ml BSA) at room temperature. Binding reactions were loaded in a 3% agarose TBE (10 mM Tris-HCl pH 8.0, 25 mM boric acid, and 1 mM EDTA) gel and electrophoresed at 50 Volts for 2 hours. The HSF–bound probe and free probe were quantified by densitometry and the dissociation constant, Kd = ([A][B])/[AB], was estimated using a non-linear least squares method on the function [AB]/[A]total = [B]/([B]+Kd) where [AB]/[A]total is the measured shifted fraction and [B] is the free HSF trimer concentration.
We incubated 600 pM HSF and 2500 ng genomic DNA (sonicated to 100–600 bp fragment size as previously described [6]) in 1500 µl final volume of 1xHSF binding buffer and let it come to equilibrium for an hour at room temperature. We added 20 µl ANTI-FLAG M2 affinity gel for 10 minutes and washed 8 times with 1xHSF binding buffer to remove unbound DNA, 3xFLAG peptide was added to a final concentration of 200 ng/µl to specifically elute HSF and HSF–bound DNA. The mock IP was done in the absence of recombinant HSF. We attribute the in vitro binding assay's low background to the design of the experiment. Since recombinant C-terminally 3xFLAG tagged HSF was used, the HSF–associated DNA could be specifically eluted by the addition of excess 3xFLAG peptide. In contrast, standard ChIP protocols rely on non-specific elution of all protein and DNA that binds the resin.
The sample preparation was as previously described [6], except that 15 rounds of amplification were performed in this case.
The PB–seq reads were aligned to the Drosophila Genome (BDGP R5/dm3) using BWA (v 0.5.8c) [39]. We obtained 5,052,425 uniquely aligned reads for replicate one, 4,694,846 for replicate two and 5,410,049 for the mock. Files that contain raw sequence data and uniquely aligned reads were deposited into NCBI's Gene Expression Omnibus (GEO) [40], accession number GSE32570.
We called peaks using MACS (v 1.3.7.1) [41], both for each individual replicate and for the merged set, using a tag size of 55 bp, a starting bandwidth of 100 bp and an appropriate genome size. After experimenting with several p-value thresholds, we selected a value of p = 0.01, which achieved a good tradeoff between maximizing the number of called peaks and ensuring consistency between replicates. Our results were largely unaffected by the ‘mfold’ parameter (the threshold for fold enrichment relative to background for inclusion in the peak model), so we left this parameter at its default value.
To improve our sensitivity in binding site detection, we made use of an ensemble of position weight matrices (PSSMs), rather than a single matrix. We sampled 10,000 sets of 100 peaks and used the program MEME [42] for motif discovery in each set. As input, MEME was given the 100 bp sequence centered at each peak summit. We used a fixed motif width of 14 bp, a second order background Markov model estimated from the entire peak set, and the ‘zoops’ model (zero or one site per sequence) with the restriction that at least 75% of the sequences must contain a site. The resulting PSSMs were compared by KL-divergence against the canonical monomer PSSM (four base pair unit with consensus AGAA) estimated from the previously published in vivo high-confidence HSF binding sites detected by ChIP-seq [6]. In each PSSM, one of the three monomers had on average about twice the KL-divergence as the other two. Figure S15 shows a scatter plot of the KL-divergence of the PSSMs in the ensemble
Each peak was scanned for matches to all PSSMs in the ensemble, allowing for overlapping sites. The score at each position was taken to be the maximum score across the ensemble. Peaks were split into three groups by GC% quantile, and for each group a 10 kbp sequence was simulated from a second order Markov model, which was then used to estimate the FDR associated with the score.
In our context, an appropriate FDR threshold should strike a balance between recapitulation of in vivo results and limiting the number of spurious binding sites. In vivo results are defined by high-confidence peaks, which are ChIP-seq peaks that were called by two peak calling programs and have a corresponding binding site sequence underlying the peak [6]. Whereas, spurious sites are accounted for by limiting the average number of HSE clusters per peak (set of potentially overlapping HSE no more than 10 bp apart from each other). Due to the repetitive nature of the HSE, a cluster is a better representative than a single site of a functional binding locus. We chose a 20% FDR threshold, which maximizes the fraction of peaks having a single HSE cluster while ensuring that a large fraction (97%) of the high-confidence in vivo peaks contain HSEs. This threshold resulted in 3735 clusters (71% with a single HSE, 20% with two HSEs overlapping by 10 bp, ∼5% with two HSEs overlapping by 5 bp; see Figure S17).
The final set of HSE clusters was obtained by combining data from the two experimental replicates. First, a set of genomic regions was identified by intersecting the peaks from the two experimental replicates, and retaining only those peaks for which the two replicates were in close agreement (>80% of reads fall in the overlapping region). We then identified the 2896 HSE clusters that fell in these regions (∼77% of all clusters).
The problem of measuring the intensity of each peak is complicated by the fact that some peaks contain multiple, closely spaced clusters, whose contributions are difficult to disentangle. Furthermore, peaks often include trailing edges that are dominated by the background signal. To address these concerns we experimented with various measures of intensity based on the output produced by MACS (wig files giving shifted read counts in 10 bp windows) as well as the reported ‘bandwidth’ B. We considered three measures, applied to a window of radius B centered at each cluster: maximum read count, read count sum, and an “integrated” read count based on a biweight kernel (which produces a curve at each peak that is similar to the one implied by the peak model used by MACS). We selected the biweight kernel measure, which does the best job of handling closely spaced clusters (see Figure S18).
We assume that each HSE site i is at approximately the same initial concentration in the experiment ([HSEi]initial = C). Furthermore, all sites compete to bind a shared amount of free HSF, with the remaining unbound concentration denoted by [HSF]. At the end of the experiment, a fraction of site i is bound, with concentration [HSEi : HSF], and the remainder is unbound, with concentration [HSEi]. The dissociation constant for a particular HSE site is therefore given by:
The bound HSE concentration is measured by the PB–seq experiment in terms of the number of reads at element i (Ri). This leaves two unknown quantities, [HSF] and [HSEi], in units of read counts. The first of these unknowns, [HSF], can be eliminated by considering instead the relative Kd with respect to a known reference value (for an HSE present in the experiment).
To solve for [HSEi], we express this quantity as the difference between the initial concentration C and the measured bound concentration:By substituting the expression for Kdi (above) and dividing by the Kd value for the reference HSE, Kdref, we obtain an expression with a single unknown, C:With the use of a reference dissociation value for a second HSE, we can solve for C and obtain estimates of the dissociation constants for all other HSE sites for which read counts are available. Replacing Kdi and Ri by the corresponding values for the second reference HSE and solving for C:
Our probabilistic model for HSEs was designed to capture interactions among the binding preferences of the three monomers that form the HSF homotrimer. The model consists of three PSSM-based submodels corresponding to the three 5 bp sequences (pentamers) that are bound by the HSF monomers. Each of these submodels is defined by two PSSMs, one ‘strict’ and one ‘relaxed’. These three submodels allow for eight possible combinations of strict and relaxed pentamer binding. Within each pentamer the positions are considered independent, as in standard PSSM models.
Formally, let a candidate 15 bp HSE sequence Xk be composed of random variables Xi,jk where i is the pentamer index and j is the base position within that pentamer. Additionally, let each sequence have an associated unobserved random variable Yk which indicates which of the eight combinations of strict/relaxed distributions are applied the corresponding Xi,jk (Figure S12). For simplicity, our model definition assumes that the middle monomer sequence has been reverse complemented and is therefore in the same orientation as the outer monomer binding sequences. We considered two versions of the model: a sparsely parameterized ‘constrained’ version and a more parameter-rich ‘expanded’ version, as described below.
The chromatin effect and DNase models are rule ensemble models, estimated using the RuleFit R package. This package was also used to estimate the relative importance of the model covariates. The covariates were obtained from modENCODE tracks, taking the mean value over a 200 bp window centered on the target point. Furthermore, these data were filtered to contain only points that had a value for every covariate used.
|
10.1371/journal.pcbi.1003834 | Memory Maintenance in Synapses with Calcium-Based Plasticity in the Presence of Background Activity | Most models of learning and memory assume that memories are maintained in neuronal circuits by persistent synaptic modifications induced by specific patterns of pre- and postsynaptic activity. For this scenario to be viable, synaptic modifications must survive the ubiquitous ongoing activity present in neural circuits in vivo. In this paper, we investigate the time scales of memory maintenance in a calcium-based synaptic plasticity model that has been shown recently to be able to fit different experimental data-sets from hippocampal and neocortical preparations. We find that in the presence of background activity on the order of 1 Hz parameters that fit pyramidal layer 5 neocortical data lead to a very fast decay of synaptic efficacy, with time scales of minutes. We then identify two ways in which this memory time scale can be extended: (i) the extracellular calcium concentration in the experiments used to fit the model are larger than estimated concentrations in vivo. Lowering extracellular calcium concentration to in vivo levels leads to an increase in memory time scales of several orders of magnitude; (ii) adding a bistability mechanism so that each synapse has two stable states at sufficiently low background activity leads to a further boost in memory time scale, since memory decay is no longer described by an exponential decay from an initial state, but by an escape from a potential well. We argue that both features are expected to be present in synapses in vivo. These results are obtained first in a single synapse connecting two independent Poisson neurons, and then in simulations of a large network of excitatory and inhibitory integrate-and-fire neurons. Our results emphasise the need for studying plasticity at physiological extracellular calcium concentration, and highlight the role of synaptic bi- or multistability in the stability of learned synaptic structures.
| Synaptic plasticity is widely believed to be the main mechanism underlying learning and memory. In recent years, several mathematical plasticity rules have been shown to fit satisfactorily a wide range of experimental data in hippocampal and neocortical in vitro preparations. In particular, a model in which plasticity is driven by the postsynaptic calcium concentration was shown to reproduce successfully how synaptic changes depend on spike timing, specific spike patterns, and firing rate. The advantage of calcium-based rules is the possibility of predicting how changes in extracellular concentrations will affect plasticity. This is particularly significant in the view that in vitro studies are typically done at higher concentrations than the ones measured in vivo. Using such a rule, with parameters fitting in vitro data, we explore how long the memory of a particular synaptic change can be maintained in the presence of background neuronal activity, ubiquitously observed in cortex. We find that the memory time scales increase by several orders of magnitude when calcium concentrations are lowered from typical in vitro experiments to in vivo. Furthermore, we find that synaptic bistability further extends the memory time scale, and estimate that synaptic changes in vivo could be stable on the scale of weeks to months.
| Many experiments have shown that long-lasting changes in synaptic efficacy can be induced by spiking activity of pre- and postsynaptic neurons [1], [2]. In hippocampal and neocortical in-vitro preparations, both long-term potentiation and depression can be induced by protocols in which pre- and postsynaptic neurons emit tens to hundreds of spikes in specific temporal patterns [3]–[10]. In those preparations, plasticity has been shown to depend both on relative timing of pre- and postsynaptic spikes (‘spike timing dependent plasticity’, or STDP), and the firing rates of pre- and postsynaptic neurons. These induced changes in the connectivity of a neural circuit have then been assumed to maintain or initiate a long-term memory trace of external inputs that triggered these synaptic changes [11]. However, in order for this theory to be valid, the induced synaptic changes need to survive both activity triggered by other inputs, and the ongoing background activity that is pervasive in cortex [12], [13]. How changes in synaptic connectivity survive the continuous presentation of other inputs has been the subject of several studies [14], [15]. Here, we study the decay of the synaptic memory trace due to background activity using a theoretical approach.
Synaptic plasticity has been described using a multitude of different models and approaches [6], [7], . In early plasticity models, synaptic changes were purely induced by the firing-rates of pre- and postsynaptic neurons [16], [17], [19]. At the end of the nineties, theorists introduced purely spike-timing based models [21], [23]. Finally, more recent models have been striving to capture a wide range of experimental data, and as a result capture both the spike-timing and firing rate dependence of synaptic plasticity [6], [24]–[31]. These models are natural candidates for studies of the stability of synaptic changes during ongoing activity. In this paper, we choose the model of Graupner and Brunel [28] for the following reasons: (i) the model includes the calcium concentration in the post-synaptic spine, which is known to be a crucial link between pre- and postsynaptic activity and long-term synaptic changes; (ii) the model exhibits bistability of synaptic changes accounting for experimental evidence suggesting that CA3-CA1 synapses in the hippocampus are bistable [32], [33]; (iii) the model is simple enough that the dynamics of the synaptic efficacy variable can be computed analytically.
Postsynaptic calcium entry has been identified to be a necessary [34]–[36] and sufficient [37]–[39] signal for the induction of synaptic plasticity (but see ref. [40]). However, most of the in vitro experiments evoking synaptic changes use elevated extracellular calcium concentrations, while in vivo calcium levels are estimated to be around 1.5 mM [41]. The impact of reduced calcium entry due to the lower extracellular calcium concentration in vivo on the time scale of synaptic decay has not been considered heretofore.
In the present paper, we study the persistence of synaptic changes, first in a synapse connecting a pair of independent Poisson neurons, and second in a large network of excitatory and inhibitory leaky integrate-and-fire (LIF) neurons. We show that in the absence of bistability, synaptic changes decay exponentially during ongoing activity and that the time constant exhibits a power-law like behaviour with respect to the present firing rate. We demonstrate that the reduced extracellular calcium concentration in vivo leads to several orders of magnitude longer memory time scales. The introduction of bistability in the synaptic plasticity rule further stabilises synaptic changes at low firing rates and extensively prolongs memory time scales when combined with the in vivo extracellular calcium conditions. Finally, we extend our results to a large recurrent network of LIF neurons, where we demonstrate network firing rate stability under synaptic plasticity, decay of an implanted memory for in vitro parameters and long term memory maintenance for in vivo parameters.
Memories are thought to be stored in the brain thanks to activity-dependent modifications of synaptic connectivity. According to this hypothesis, memories stored by a particular neural circuit are encoded by the state of all the modifiable synapses of the circuit. Synaptic plasticity allows particular patterns of activity to leave a trace in the connectivity matrix, but this trace is then potentially vulnerable to the ongoing activity that follows. An important question is therefore what controls the time scale of the persistence of a particular synaptic state, in the presence of such ongoing activity. To study this question, we initialize the efficacy of a synapse (either an isolated one, or part of a network) at a particular value, and study how this efficacy decays with time in the presence of background activity, using a calcium-based model of synaptic plasticity [28].
In the model, the temporal evolution of the synaptic efficacy variable, , is described by (1)where is the time constant of synaptic efficacy changes, and is the calcium concentration. The dynamics of depends on four terms:
1. The dynamics are governed by a potential for low calcium concentrations () since all other terms on the right-hand side of Eq. (1) are zero then. In the following we consider two possible scenarios for : (i) a flat potential, - in this case the synaptic efficacy variable stays constant in time in the absence of calcium transients. This means all possible values of are stable; (ii) a double well potential, (2)
In this case, evolves towards one of two possible stable fixed points (the minima of ), one at - the DOWN state -, the other at - the UP state -, depending on the initial condition.
This corresponds to a bistable synapse.
2. For intermediate calcium concentrations (), the synapse is depressed, with a depression rate .
3. For large calcium concentrations (), the synapse undergoes both potentiation, with a potentiation rate , and depression, with the same rate as in the . Since , potentiation dominates over depression in that region.
4. A noise is only active when calcium concentration is above the lowest plasticity threshold , and increases in magnitude when the upper plasticity threshold, , is also crossed. defines the amplitude of the noise, and is a Gaussian white noise process with unit variance.
Changes in are induced by increases in the postsynaptic calcium concentration, (see Eq. (11), in Methods), evoked by pre- and postsynaptic spikes. The calcium concentration increases by an amount , in response to presynaptic spikes, while it increases by an amount in response to postsynaptic spikes. It decays exponentially with a time constant in between spikes. Calcium transients induced by presynaptic activity are assumed to represent calcium influx through NMDA receptors, while calcium transients induced by postsynaptic spikes are assumed to represent activation of voltage-gated calcium channels [42] (see [28] for more details of the model).
This calcium-based model of synaptic plasticity has been used to successfully fit data from various experimental preparations [28]. Here, we use the data-set that best fits plasticity data obtained in visual cortex slices [6] - hereafter called the ‘in vitro’ parameter set. In this experiment, the extracellular calcium concentration was set to be 2.5 mM [6], which is significantly higher than the estimated in vivo concentration of about 1.5 mM [41]. Here we assume that a decrease in extracellular calcium concentration leads to a proportional decrease in the calcium influx into the post-synaptic spine. Using this assumption, we can readily predict the effects of decreasing the extracellular calcium concentration on the plasticity rule in the calcium-based model by scaling the amplitudes of the pre- and postsynaptically evoked calcium transients according to the ratio of calcium concentrations, i.e. 1.5/2.5 = 0.6. This leads to what we call the ‘in vivo’ parameter set. Values of all parameters for both conditions are indicated in Table 1.
The dynamics of the synaptic efficacy in response to calcium transients under the in vitro and the in vivo conditions are illustrated in Fig. 1. The synaptic efficacy is only modified when the calcium concentration increases above the depression threshold (Fig. 1,B–E). For the in vitro case, this happens whenever a postsynaptic spike occurs since , but for the in vivo parameter set this happens much more rarely because of the smaller calcium amplitudes ( in the in vivo case; see Table 1). In the latter case, synaptic changes are only induced whenever subsequent spikes occur in short succession such that calcium accumulates and crosses the depression and/or potentiation threshold. Such events are rare at low firing rates (Fig. 1,D–E).
We now proceed to study the time scales of synaptic decay. We start with the case of a synapse connecting two neurons firing according to uncorrelated Poisson processes, and compare the memory time constants in the flat and double-well potential cases. Simulations were performed using an event-based implementation of the synaptic plasticity model, which updates the synaptic efficacy only upon the occurrence of pre- and postsynaptic spikes (see Methods for details).
We initialise the synaptic efficacy to and investigate the time constant of decay in the presence of an ongoing constant firing rate, initially for the flat potential synapse (Eq. 1, with ). Pre- and postsynaptic neurons emit uncorrelated spikes following Poisson statistics, both with a mean rate of 1/s. Under these conditions, a fully potentiated synapse progressively decays and eventually fluctuates around a value of 0.2. On average, this decay is well described by a single exponential function (Fig. 2,A,B). The time constant of this decay is much longer in the case of the in vivo parameter set (Fig. 2,B) than in the in vitro parameter set (Fig. 2,A). The decay time constant is 2.5 minutes for the in vitro case and approximately 2 hours for in vivo in the presence of 1/s pre- and postsynaptic firing (Fig. 2,C).
The dynamics of the synaptic efficacy (Eq. 1) can be described by a truncated Ornstein-Uhlenbeck (OU) process if single calcium transients induce small changes in the synaptic efficacy and if the potential is flat (see [28] for the non-truncated case). Truncation of the process is induced by the bounds at and . In such a process, the mean synaptic efficacy decays exponentially with a time constant, , which is given by (3)to an asymptotic average efficacy, (see Eq. (22) in Methods), where and are the net potentiation and depression rates which depend on the rates and as well as on the average fractions of time spent above the potentiation and depression thresholds, and , respectively. The average fractions of time the calcium traces spend above the potentiation and depression thresholds are given by
(4)(5)where is the probability density function of the calcium variable, that can be computed analytically in the case of independent pre- and postsynaptic Poisson firing [28], [43] (see Methods for details). The theory provides an excellent match for the dynamics of the mean synaptic efficacy – compare in Fig. 2,A,B the truncated OU theory (blue and red curves), with the simulation mean (green and light blue curves).
Synaptic efficacy decay becomes faster with increasing pre- and postsynaptic firing rates since the calcium trace spends more time above depression and potentiation thresholds (Fig. 2,C). At the same time, the asymptotic value of synaptic efficacy () increases due to an increase in time spent above the potentiation threshold (Fig. 2,D). As a result of the smaller in vivo calcium amplitudes, the efficacy decay for the in vivo case is, at all firing rates, much slower than the decay in vitro (Fig. 2,C). The asymptotic efficacy value is lower, at small firing rates (/s), for the in vitro case since isolated postsynaptic spikes always cross the depression threshold () which results in a large net depression rate , compared to in vivo (Fig. 2,D).
To get a deeper understanding of the dependence of the memory time scale on the firing rates of pre- and postsynaptic neurons, we set . This simplification allows us to derive a power law relationship between the memory time scale and the firing rate , where is the number of (pre and/or post-synaptic) spikes required to clear the depression/potentiation thresholds. To compute the memory time scale, we need to compute the fraction of times spent above the depression and potentiation thresholds, and . In the case , one can show that at low rates . Consequently it is only necessary to focus our analysis on . When , the time spent above the depression threshold is (6)where is the firing rate of pre- and postsynaptic neurons (), is the decay constant for the calcium concentration and (see Eqs. (13) – (17)). This closed form solution allows us to perform an expansion for low firing rates
(7)Similarly for we have in the low rate limit, (8)where is the dilogarithm, . Thus, in both cases we find that the memory time scale depends on the firing rate as(9)where . We expect this relationship to hold in general. Intuitively, this is due to the fact that we need spikes arriving simultaneously on a time scale of order in order for the calcium concentration to cross the depression threshold , and that the probability of observing spikes in a time interval is at low rates proportional to . We also expect the result to hold in general for . In this case, we expect that .
The derived power law behaviour for is plotted in Fig. 2,C together with the full analytical solution for . We see that as expected, scales as for the in vitro parameter set, where a single spike is enough to cross , while it scales as for the in vivo parameter set, where two spikes are needed to cross the depression threshold.
The implication of this theoretical result is that, at low firing rates, there is a direct relationship between the number of spikes required to clear the lower plasticity threshold and the memory time scale. Note that the full synaptic efficacy model with is considered in the following (see Table 1)
We now turn to examine the effect of a bistability on memory time scales. The dynamics of the synapse is now described by Eq. (1), where the potential is given by Eq. (2). This double well potential leads to a bistable synapse, that can take two possible efficacy states ( and ) in the absence of activity. In the presence of background activity, transitions between these two states become possible. We investigate stability and transition times for in vitro and in vivo parameter sets as a function of pre- and postsynaptic firing rates.
The effect of background activity on the dynamics of can be explained by the fact that it modifies the potential, , leading to an effective potential (10)
(see Methods). In Eq. (10), the first term on the r.h.s. corresponds to the ‘bare’ double well potential (Eq. (2)); the second term describes the effect of depression on the potential, that tends to strengthen the stability of the lower well (DOWN state) at , at the expense of the other well that tends to disappear when increases; finally, the last term describes the effect of potentiation, that shifts the minimum of the only remaining well towards higher values of when increases.
Thus, there are two distinct regions of firing rates in the bistable case with respect to the effective potential. For sufficiently low rates, the effective potential still has two minima (see Fig. 3,A, and the effective potentials for 0.1/s and 1/s, indicated by orange and magenta curves in the inset). There is a critical value of the rates at which the high efficacy minimum disappears through a saddle-node bifurcation. Beyond this rate, the synapse is no longer bistable, and synaptic efficacy has one stable state only (Fig. 3,A), equivalent to the asymptotic efficacy value for the flat potential (Fig. 2,D). Finally, at high firing rates, the ‘bare’ potential becomes negligible, and the effective potential approaches a quadratic potential with a single stable state whose location depends on the rate (green curve in the inset in Fig. 3,A). The transition from double-well to single well regimes occurs at different firing rates for the in vitro (/s) and the in vivo (/s) parameter sets due to the larger calcium amplitudes in the former.
For the in vitro parameter set, adding bistability to the synaptic efficacy has no influence on the decay time constant for firing rates larger than approximately 0.1/s (Fig. 3,B). In contrast, for the in vivo parameter set, bistability considerably prolongs memory decay times with respect to synapses with flat potential at firing rates below <1.4/s. In the presence of two stable states, the decay of memory occurs only due to synaptic noise fluctuations that push the synaptic efficacy out of the upper well. The influence of the double well potential on the dynamics of the synaptic efficacy traps synapses in the UP state leading to long dwell times before crossing the potential barrier and converging to the low efficacy state (Fig. 3,C). The double-well has a prolongation effect on memory duration up to firing rates of about /s due to the transition between double-well and single-well regimes. At high firing rates, the potentiation and depression processes dominate and the effects of the double-well becomes negligible for both parameter sets, that is, the decay time constant is indistinguishable between flat and double-well potential synapses (see Fig. 3,B).
For low firing rates, we can accurately predict the increase in the decay time constant in the presence of bistability using Kramers escape rate for the mean first passage time across a potential barrier (Fig. 3,B; see Methods Eq. (26)). In this regime, we calculated an effective decay time constant using Kramer's escape theory given by , where is the height of the effective potential barrier and the noise term, , drives the escape of the efficacy from the upper stable state (see magenta line in Fig. 3,B for the in vivo case). Both terms and are dependent on and are detailed, along with , in Eqs. (23) and (26) (see Methods for more details). In the low rate limit, and therefore the memory time scale increases exponentially with the inverse of the rate to a power , , where is again the number of simultaneous spikes needed to cross the depression threshold. This exponential dependence extends the time scale for synaptic decay at 1/s to the order of one month for a bistable synapse with the in vivo parameter set, up from hours for a synapse with flat potential.
We next study the behaviour of the calcium-based synaptic plasticity model in a recurrent network of spiking neurons. We first examine the steady-state of synaptic efficacy and network activity. We again make use of the event-based implementation of the synaptic plasticity rule (described in Methods) allowing us to simulate much longer time scales than are normally attainable by a time stepping simulator.
The recurrent network consists of 8000 excitatory and 2000 inhibitory leaky integrate-and-fire (LIF) neurons. Each neuron receives an external input which consists of a constant (DC) term and a white noise term. External noise is independent from neuron to neuron. Each neuron also receives synaptic inputs from other neurons in the network. The connection probability between any two neurons is 0.05 and uniform in space and across neuron types. Synapses between excitatory neurons are plastic according to the calcium-based plasticity model (Eq. 1), while all synapses involving inhibitory neurons are fixed. Parameters of the network are chosen so that the network settles in a stable asynchronous irregular state [44]. Hence, correlations between neurons are weak. See Methods for more details of the network model.
The fixed point of the network can be determined analytically by solving a set of three self-consistent equations for the excitatory and inhibitory mean rates as well as for the mean excitatory-to-excitatory (EE) synaptic efficacy (see Methods). Two of these equations give the stationary firing rates of excitatory and inhibitory populations (Eqs. (32) – (33)), as a function of the mean EE synaptic efficacy [44], [45]. The third equation gives the mean EE synaptic efficacy as a function of the firing rate of the excitatory population, assuming Poisson firing statistics of neurons (Eq. (22)). Starting from the analytically determined initial conditions, the recurrent network converges to a steady-state of constant average firing rates of all neurons in the network, and constant average synaptic efficacy of the plastic connections. Figure 4,A shows how the firing rates observed in the simulations compare with the analytically predicted firing rates. It shows that at sufficiently low rates, the analytical prediction gives a very good estimate of the observed rates; however, for rates above 3Hz the observed rates are significantly lower than the analytical prediction. Likewise, the analytical prediction for the mean EE synaptic efficacy significantly overestimates the observed efficacies (green dots in 4,B).
What is the source of the difference between theory and simulations in predicting the steady network state? When synapses are fixed in the network at the efficacies predicted by the corresponding firing rate, the analytically predicted network firing rates provide a good approximation of the observed activity (blue dots in 4,A). This suggests that the underestimation of firing rates and synaptic efficacy emerges from the mapping of firing rates onto synaptic efficacy, and not due to correlations between spike trains of different neurons. We examine this effect in Fig.4,C, where we show asymptotic mean synaptic efficacy results under three different conditions. First, we simulated a population of disconnected, independent LIF neurons, receiving stochastic independent inputs with the same mean and variance as in the ‘real’ network simulation (magenta dots). By definition, this simulation led to uncorrelated spike trains. In this simulation, fake synaptic connections between neurons obeyed the plasticity rule, but had no effect on the dynamics of the neurons. Second, we simulated a connected recurrent network with constant synaptic weights. As in the first case, we simulated ‘fake’ synaptic connections that obeyed the standard plasticity rule, but these fake synapses had no effect on the dynamics (blue dots). Third, we simulated a standard recurrent network in which synaptic weights are plastic according to the plasticity rule (green dots). All three simulations show indistinguishable results, and in all three cases the average (real or fake) synaptic efficacies are consistently smaller compared to the analytical shot noise prediction (4,C, blue line). This suggests that correlations have a negligible effect on mean efficacies and firing rates, and that the differences between simulations and theory are due to differences in spiking statistics between the LIF model and a Poisson process.
To investigate further how the spiking statistics of the LIF model and in particular the interspike-interval (ISI) distribution causes the differences seen in Fig. 4, we varied the ISI distribution of the LIF neuron by changing the reset potential (, see Methods). This change had a strong effect on the average synaptic efficacy (4,D). A reset potential close to threshold ( mV, mV) yields an overrepresentation of short ISI compared to Poisson firing (4,D, inset) and in turn overestimates the average synaptic efficacy (4,D; cyan dots). Conversely, more depolarised reset potentials lead to an under-representation of short ISIs with regard to Poisson firing and consequently to an underestimation of the average synaptic efficacy (4,D; magenta, red and green dots). We use an intermediate value of mV in the following network investigations.
To conclude this section, the calcium-based synaptic plasticity rule does not affect the stability of the asynchronous irregular state in a large recurrent network of LIF neurons. Since LIF neurons in the network exhibit ISI distributions which deviate from those of Poisson neurons, the accuracy of our calculation of the average synaptic efficacy which is based on Poisson firing decreases with increasing firing rates up to a certain point. At high firing rate, calcium remains above the plasticity thresholds most of the time and the fraction of time spent above the thresholds converges to one, irrespective of the underlying neuron model.
Finally, we examine the decay of a memory trace in a network for the in vitro and the in vivo parameter set. We initialise all excitatory-to-excitatory synaptic weights at their theoretically predicted asymptotic weights, except for a randomly selected subset of 5% which are set to a weight of 1. With the in vitro parameter set, the potentiated synapses decay relatively quickly to their asymptotic value (Fig. 5,B). The time course of the average decay can be described by a single exponential function and the decay time constant is well approximated by the time constant, , of synaptic decay from the truncated OU process (see Eq. (3); Fig. 5,C). This means that the average dynamics of synaptic decay in the network is equivalent to synapses driven by independent pre- and postsynaptic Poisson neurons firing at the same rate as the excitatory neurons in the network (compare to Fig. 3,B). The addition of the double-well potential does not change the decay time constant for the in vitro parameter set, as for a single synapse driven by independent pre- and postsynaptic Poisson firing (Fig. 5,C orange stars; compare with Fig. 3,B). The lack of short ISIs in LIFs compared to independent Poisson neurons leads to a small increase in observed decay times in the network as compared with the OU theory (see Fig. 5,C).
In contrast, when using the in vivo parameter set with the double-well potential, we observe that the potentiated synapses get locked in the UP state for the duration of the network simulation with an excitatory neuron firing rate of 1/s (Fig. 6,C). None of the synapses in the potentiated subset crosses the unstable fixed point and converges to the DOWN state during a network simulation of 120 min, neither does the reverse transition occur. We expect that the escape from the UP state will be predicted by Kramers escape rate (Eq. (26)) which correctly accounted for escape dynamics of an isolated synapes driven by independent pre- and postsynaptic Poisson processes (Fig. 3B). There, the decay time constant for a firing rate of 1/s is on the order of a month, a time scale that cannot be reached by our network simulation.
Hence, as in case of independent Poisson neurons, the combination of a double-well potential with the in vivo parameter set leads to several orders of magnitude longer memory time constants, compared to the in vitro parameter set and a flat potential.
In this paper, we studied the stability of synaptic efficacy, in a plastic synapse subjected to background activity of pre- and postsynaptic neurons. We used a calcium-based plasticity model that has been shown to fit experimental data in hippocampal and neocortical preparations [28]. The model was investigated numerically, using an event-based implementation of the plasticity rule, as well as analytically, using a diffusion approximation. Thanks to this formalism, we derived scaling laws that describe how memory time scale is related to the firing rates of pre- and postsynaptic neurons. At low firing rates, we find that, when synapses are monostable, synaptic efficacies decay to an equilibrium value with a time scale that depends on the firing rates as a power law, , where k is the number of simultaneous spikes needed to cross the depression threshold. When synapses are bistable, memory decay is akin to diffusion of a particle out of a potential well, which leads to much stabler memories, with time scales that increase exponentially with the inverse of the firing rates, , at low rates. We showed that these estimates accurately reproduce the results of simulations, both of a synapse connecting two isolated independent Poisson neurons, and of a large network of LIF neurons.
We have focused here on how changes in the amplitudes of the calcium transients affect memory time scales. A change in other model parameters also affects these time scales. Changing the depression threshold, for example, has a similar pronounced effect, since the exponent in the scaling law between memory time scale and background rate depends on the ratio between this threshold and the amplitudes of the calcium transients (see (9)). On the other hand, changing other parameters of the model (such as the time constants and the potentiation and depression rates) have much milder effects, in the flat potential regime, since the time scale depends algebraically on such parameters, rather than exponentially. In the absence of further data, we have assumed a linear relationship between the external calcium concentration and the calcium influx to explore the in vivo regime. A non-linear relationship between the extracellular calcium concentration and the concentration in postsynaptic microdomains - conjectured to be relevant for synaptic plasticity - should modify our results quantitatively rather than qualitatively.
Previous studies have investigated memory maintenance in networks of neurons connected by synapses endowed with standard spike-timing dependent plasticity rules [46]. Billings and van Rossum (2009) demonstrated that the memory time scale depends dramatically on whether the rule is additive or multiplicative. In a multiplicative STDP rule, in which synaptic change depends on the current value of the weight such that synaptic changes decrease when the weights approach the bounds, distributions of weights are unimodal [46]–[48] and the memory of synaptic changes decay as , since synaptic changes occur upon coincidence of pre- and postsynaptic spikes in the characteristic time window of the STDP rule. These behaviours are very similar to the behaviour of the calcium-based rule in the flat potential case, in the parameter region in which two spikes are needed to cross the depression threshold. This is due to the fact that the calcium-based rule defined by Eq. (1) is multiplicative. In the calcium-based rule however, the exponent describing the memory decay at low rates can be set to an arbitrary integer number, through an appropriate rescaling of the ratio between the amplitude of the calcium transients and the depression threshold. In additive STDP rules, the picture changes dramatically and the synaptic weight distributions become bimodal, with weights attracted either to the lower or upper bounds through a symmetry breaking mechanism [23], [46]. In this situation, the memory time scales are much longer, and decay of synapses is similar to diffusion in a double well potential.
Several studies have shown that synaptic bi- or multi-stability can emerge from a number of mechanisms such as positive feedback loops in extensive protein signaling cascades [49], autophosphorylation of CaMKII [50]–[54], self-sustained regulation of translation [55], or modulation of receptor trafficking rates [56]. Such mechanims of bistablity are effectively implemented here in the form of the double well potential. Miller et al. (2005) studied the stability of the up state in a model of the bistable calcium/calmodulin-dependent protein kinase II system with respect to stochastic fluctuations induced by protein turnover [57]. They show that the CaMKII switch composed of a realistic number of CaMKII proteins is stable for years even in the presence of protein turnover, phosphatase as well as free calcium fluctuations. The transitions induced by background activity investigated here impose an upper limit on memory life-time which is typically lower, indicating that in vivo neuronal activity, not protein turnover, will be the limiting factor of memory life-times.
In vivo, memory in synaptic connectivity structures will be affected both by ongoing background activity, but also by changes in network activity induced by external stimuli. How ongoing presentations of external inputs affect memories of past stimuli has been the subject of several studies in recent years (e.g. [15], [20], [58]), in simpler networks of binary neurons. A detailed exploration of this issue in the model studied here is outside the scope of the present paper, but we anticipate that parameter regions that extend the robustness of synaptic memories in the face of background activity will also tend to protect the network against changes induced by external inputs.
Distributions of synaptic weights have been examined in a number of studies [4], [6], [59]–[62]. In all of these studies, distributions of synaptic weights appear unimodal and skewed, and peak at a low weight. In some cases, the distribution has been shown to be well fitted by a lognormal distribution [60], [62]. This seems at first sight at odds with the distributions of weights shown in the present paper, which are either a truncated Gaussian in the flat potential case, lacking the fatter tail of the lognormal distribution, or bimodal in the double-well case. However, the model in the flat potential case can be made consistent with the data, by choosing synaptic efficacy variables which are an exponential of the variable, rather than being linearly related to . In this case, synaptic efficacies themselves become exponentiated Ornstein-Uhlenbeck processes, consistent with [62]. The model with a double-well potential could also be made consistent with a unimodal distribution, provided the synaptic up and down states are highly heterogeneous from synapse to synapse. Finally, we should point out that the distributions we observe are asymptotic distributions under a statistically constant distribution of inputs. Synapses in vivo are typically subjected to highly non-stationary firing rates of pre and post synaptic neurons. These non-stationarities can also potentially strongly affect distributions of synaptic weights in our model.
A large number of distinct learning rules that capture quantitatively both spike-timing and firing rate effects have been proposed recently [25]–[31]. Our rule can be distinguished from most of those rules by the fact that it includes calcium concentration as its primary dynamic variable, which allows us to extrapolate parameters of the rule from in vitro to in vivo conditions, as we have explained here. Scaling laws derived here can be expected to hold also in those other models: at low rates, the time scales of memory decay are expected to be inversely proportional to the rates to a power equal to the number of spikes needed to provoke plasticity. This power should be equal to 2 for standard STDP rules, triplet rules [25], and calcium-based rules in which 2 spikes are needed to cross the depression threshold [24], [27]; 1 for spike and voltage based rules [26].
In this work, we have made the hypothesis that synaptic weights are altered during background activity, and that one can treat background activity as being essentially uncorrelated with the synaptic connectivity structure. Memory time scales could in principle be further extended by two factors. A first mechanism would be to gate plasticity by specific neuromodulator(s) that are present only during stimulus presentation. This idea is consistent with a growing body of experimental data showing how plasticity is modulated by dopamine [63], acetylcholine [64], [65], noradrenaline [66] (see also [67] and references therein). However, we note that the model we have used here is built from in vitro plasticity data where these neuromodulators were present at very low concentrations, if at all. Hence, we believe that these neuromodulators are likely to enhance plasticity during behaviourally relevant epochs, but that the memory time scales discussed here are likely not to be affected if neuromodulators are not present at high levels during background activity.
A second mechanism that would extend memory time scales would be a scenario in which background activity is in fact strongly correlated with the connectivity structure, and wanders stochastically between network states that are strongly correlated with the states of the network during stimuli presentation. This idea is consistent with a growing experimental literature [68]–[70] showing how spontaneous activity is transiently strongly correlated with sensory responses in visual and auditory cortices, and it is also consistent with the ubiquitous supra-Poissonian variability, potentially due to the doubly-stochastic process of combined rate stochasticity and individual neuronal Poisson spike processes, seen in background activity in cortex [13], [71]. Recurrence of activity states resembling the network activity during stimulus presentation could refresh existing memory traces and therefore prolong their lifetimes.
We showed here that the low extracellular calcium concentrations in vivo could have a strong impact on plasticity. A first prediction of calcium-based rules is that plasticity seen in standard protocols should be greatly reduced (and even possibly vanish altogether) at physiological calcium concentrations. While to our knowledge no study has explicitly compared plasticity results at different extracellular calcium concentration, comparisons between different studies using different extracellular concentrations seem to be consistent with this prediction. In hippocampal slices, a standard low frequency STDP protocol produces LTD for all time differences with 2 mM extracellular calcium [9], while it produces the standard STDP curve with 3 mM calcium [10]. A second prediction is that induced synaptic changes should be much more stable in the face of ongoing pre- and postsynaptic activity. These results emphasise the need for experimental studies at physiological calcium concentrations mM [72], unlike most published studies that used concentrations in the range mM. Our predictions could be easily tested in slice experiments, by providing background activity at a specified rate after the plasticity-inducing protocol. Similar experiments have been performed in the developing Xenopus retino-tectal system in vivo [73], where activity-induced modifications were shown to be erased by subsequent 10 minutes of spontaneous activity. Our model would predict that in cortical slices, at 2.5 mM calcium, induced synaptic changes should disappear on a time scale of minutes, while at 1.5 mM calcium, they should be stable on a time scale of hour.
We provided here an event-based update scheme of plastic synapses which greatly accelerates simulations and should strongly facilitate future studies of the dynamics of recurrent networks with plastic calcium-based synapses. On the theoretical front, it would be interesting to extend the theory to non-Poissonian renewal processes [74] such as for leaky integrate-and-fire neurons used here, which would give a better approximation of average synaptic efficacies, especially at higher firing rates. It would also be of great interest to examine how synaptic connectivity is modulated by non-stationary external inputs, and how such changes in connectivity affect in turn the intrinsic dynamics of the network.
Our investigations show that realistic external calcium concentration and multi-stability of synapses might stabilise memory traces against the potentially deleterious effect of ongoing background activity. These results call for studies of synaptic plasticity induction and maintenance in more realistic conditions and ideally in the intact animal. They provide a glimpse of how plasticity results obtained in vitro might translate to the living organism.
The temporal dynamics of the synaptic efficacy in the calcium-based model are given in Eq. (1) (for details see [28]).
Changes in are driven by the postsynaptic calcium concentration, c. The calcium dynamics are modelled using instantaneous increases of size and in response to pre- and postsynaptic spikes, respectively, followed by an exponential decay (11)where is the calcium decay time constant, and , the pre- and postsynaptically evoked calcium amplitudes. The sums go over all pre- and postsynaptic spikes occurring at times and , respectively. The time delay, D, between the presynaptic spike and the occurrence of the corresponding calcium transient accounts for the slow rise time of the NMDAR-mediated calcium influx.
We use two parameter sets in this paper. The in vitro parameter set is obtained by fitting the calcium-based plasticity model to plasticity data obtained in cortical slices ([6]; see [28] for details of the fitting procedure). These experiments were performed with 2.5 mM extracellular calcium concentration. The in vivo calcium amplitudes are obtained by scaling and according to the extracellular calcium concentration in vivo, estimated to be 1.5 mM [41] (see Results).
We shortly describe here how the probability density function (PDF) of the calcium concentration can be calculated if pre- and postsynaptic neurons fire as independent Poisson processes at rate (see [28], [43] for more details). In these conditions, the calcium concentration is a shot noise process, whose probability density function is given by the master equation [43], (12)
The probability density function allows us to calculate the fraction of time spent above the depression and potentiation thresholds according to and .
In the simple case , the solution to Eq. (12) is given by (13)(14)(15)(16)where is the ordinary hypergeometric function [75], (17) is Euler-Mascheroni constant, , and is the gamma function.
Fitting the calcium-based model to cortical plasticity data yields (see Table 1). In this case, the solution of Eq. (12) reads (18)(19)(20)where is a normalisation parameter which assures that . The PDF for is obtained from a numerical integration of Eq. (12).
Performing a diffusion approximation of the synaptic efficacy turns Eq. (1) into an Ornstein-Uhlenbeck process (see [28] for details). The temporal evolution of is then described by (21)for the case of a flat potential (i.e. ). and are the mean potentiation and depression rates, respectively.
The bounds at and lead to a truncated Ornstein-Uhlenbeck process, whose distribution is a truncated Gaussian, whose mean converges exponentially to (22)where , is the Gaussian with zero mean and unit variance, and is the complementary cumulative density function of G. The time constant, , of the exponential decay to is defined in Eq. (3).
In the case of a double-well potential, the diffusion approximation turns Eq. (1) into a Fokker-Planck equation (23)
This equation can be rewritten as (24)where the effective potential, , is the sum of the ‘bare’ potential U and two quadratic terms proportional to the potentiation and depression rates, respectively (see Eq. (10)), and is the amplitude of the effective noise
(25)When the effective potential is dominated by the double-well term (first term on the rhs of Eq. (10)), the ‘escape’ rate from the UP state is driven by noise and can be estimated using Kramers theory [76], [77]. The height of the potential barrier, , determines the mean dwell time in the UP state, where and are the local minima and maxima of the effective potential and are obtained solving . This allows us to calculate the expected escape time from the potential well (26)
The temporal evolution of individual synaptic weights in the calcium-based model can be calculated in an event-based manner (as opposed to a finite difference method with a fixed time step ) in an analytically exact way. This greatly accelerates network simulations since the network variables are updated at the occurrence of spikes only. In the following we describe the practical implementation of the event-based update.
For the event-based update, three variables have to be stored per synapse: the calcium concentration, c, the synaptic efficacy variable, , and the time of the last spike seen by the synapse, t. The synapse variables c and must be updated on the occurrence of three events: (1) at the presynaptic spike when the postsynaptic voltage is increased, (2) with delay D after a presynaptic spike when the presynaptically evoked calcium increase occurs (see Eq. (11)), (3) and at the postsynaptic spike when the postsynaptic calcium increase occurs.
We implemented and simulated a recurrent network of 10,000 leaky integrate-and-fire (LIF) neurons, 8,000 of which are excitatory (E) neurons and 2,000 of which are inhibitory (I). Any two neurons have a spatially uniform probability of connection of 0.05. Autapses are specifically disallowed. Synapses between E neurons are plastic and their weight dynamics are described by the calcium-based plasticity model (Eq. 1, [28]). All other synapses have fixed strength (). A presynaptic spike induces a voltage jump of size in the postsynaptic neuron.
The membrane potential of neuron i of population evolve according to (30)where(31)is a common external drive to all neurons, comprising a constant input, , and a white noise of amplitude mV. is a Gaussian white noise process with unit variance and zero mean, which is uncorrelated from neuron to neuron. In the absence of synaptic inputs each membrane potential decays exponentially to the leak potential, mV, with a time constant ms. Spiking occurs when the voltage crosses a threshold, mV, after which it is reset to the reset potential, mV. During all of our simulations, we set the refractory period, during which the membrane potential is fixed at after spiking, to zero. The three sums in the r.h.s. of Eq. (30) go over the two populations {E, I}, all presynaptic neurons j, and presynaptic spikes of neuron j in population , that occur at times . Each presynaptic spike of neuron j in population causes a jump of amplitude in the voltage of neuron i after a delay . Here, the delay is chosen to be equal to the integration time step ms (see below).
For all connections involving inhibition (i.e. all ), the connectivity matrix is set as where are independent, identically distributed (i.i.d.) Bernoulli variables, with probability 0.05, 0 with probability 0.95, and the fixed synaptic weights are mV, mV and mV. E-E synapses are given by where are again i.i.d. Bernoulli variables, with probability 0.05, 0 with probability 0.95, obeys Eq. (1), and mV. The average value of is initially, and remains throughout our simulations, much smaller than 0.5, which means that with a ratio in the E to I populations, for recurrent inhibition dominates excitation, leading to a stable asynchronous irregular state (see Fig. 8) [44].
We numerically simulated the recurrent network of LIF neurons using the forward Euler method with a time step of 0.01 ms. Synapses were updated using the event-based implementation described above. The simulations were implemented in C and OpenCL and run on general-purpose GPUs. Parallel generation of random numbers, for the Gaussian noise in the LIF equations, was implemented using the Random123 library [78].
In order to initialise the simulations close to their steady-state, with the in-vivo parameter set and the double-well potential, we first calculate the probability distribution function (PDF) for the synaptic weights assuming a 1/s pre- and post-synaptic Poisson firing process. We then use a reverse lookup of the associated cumulative distribution function (CDF) to determine the random initial values for the synaptic efficacies.
In a network of excitatory and inhibitory LIF neurons receiving white noise inputs, the mean firing rates of excitatory and inhibitory neurons are given by [44], [45] (32)(33)where is the standard LIF static transfer function [44], [45], [79], [80],(34)where is the error function, are the mean inputs to population ,
(35)(36)and is the amplitude of the fluctuations in the inputs to population ,
(37)(38)
Note that in Eqs. (35,37) is given by Eq. (22). Note also that for the parameters studied in this paper the effect of heterogeneities in numbers of inputs [81], [82] have a negligible effect on the mean firing rates of the network.
|
10.1371/journal.pntd.0003684 | A Small Antigenic Determinant of the Chikungunya Virus E2 Protein Is Sufficient to Induce Neutralizing Antibodies which Are Partially Protective in Mice | The mosquito-borne Chikungunya virus (CHIKV) causes high fever and severe joint pain in humans. It is expected to spread in the future to Europe and has recently reached the USA due to globalization, climate change and vector switch. Despite this, little is known about the virus life cycle and, so far, there is no specific treatment or vaccination against Chikungunya infections. We aimed here to identify small antigenic determinants of the CHIKV E2 protein able to induce neutralizing immune responses.
E2 enables attachment of the virus to target cells and a humoral immune response against E2 should protect from CHIKV infections. Seven recombinant proteins derived from E2 and consisting of linear and/or structural antigens were created, and were expressed in and purified from E. coli. BALB/c mice were vaccinated with these recombinant proteins and the mouse sera were screened for neutralizing antibodies. Whereas a linear N-terminally exposed peptide (L) and surface-exposed parts of the E2 domain A (sA) alone did not induce neutralizing antibodies, a construct containing domain B and a part of the β-ribbon (called B+) was sufficient to induce neutralizing antibodies. Furthermore, domain sA fused to B+ (sAB+) induced the highest amount of neutralizing antibodies. Therefore, the construct sAB+ was used to generate a recombinant modified vaccinia virus Ankara (MVA), MVA-CHIKV-sAB+. Mice were vaccinated with MVA-CHIKV-sAB+ and/or the recombinant protein sAB+ and were subsequently challenged with wild-type CHIKV. Whereas four vaccinations with MVA-CHIKV-sAB+ were not sufficient to protect mice from a CHIKV infection, protein vaccination with sAB+ markedly reduced the viral titers of vaccinated mice.
The recombinant protein sAB+ contains important structural antigens for a neutralizing antibody response in mice and its formulation with appropriate adjuvants might lead to a future CHIKV vaccine.
| Chikungunya virus (CHIKV) is transmitted by mosquitos and causes high fever and severe joint pain in humans. It is expected to spread in the future to Europe and has recently reached the USA due to vector switch and climate change. There is no specific treatment or vaccination against CHIKV infections. However, vaccination should be an efficient way to control its spread. The CHIKV envelope glycoprotein E2 enables attachment of the virus to target cells and a humoral immune response against E2 should protect from CHIKV infections. We aimed to identify small antigens of the CHIKV E2 protein that are able to induce neutralizing antibodies. These antigens should enable the production of cost effective, safe and efficient vaccines. The surface-exposed parts of the E2 domain A (sA) fused to domain B and a part of the β-ribbon that joins domain A with B (sAB+) induced most effectively neutralizing antibodies and mice vaccinated with this protein were partially protected from a CHIKV challenge infection. The protein sAB+ was identified as a useful antigen for developing a vaccine when formulated with an appropriate adjuvant.
| Chikungunya virus (CHIKV) is a mosquito-transmitted alphavirus that causes chikungunya fever in humans. Most CHIKV infections are symptomatic, with an incubation period of 2–4 days and the disease is characterized by sudden onset of fever, headache, malaise, arthralgias or arthritis, myalgias, and lower back pain. After the acute phase, polyarthritis can be recurrent and can persist for several years after infection [1]. This raises serious public health and economic problems during large outbreaks. So far, Aedes aegypti has been the most important CHIKV vector, but during a large outbreak in 2006 in La Réunion, Aedes albopictus (the Asian tiger mosquito) was the primary vector [2]. The more effective transmission via Aedes albopictus was due to only one point mutation (A226V) in the E1 envelope protein. Aedes albopictus also inhabits temperate and even cold temperate regions of the eastern and western hemispheres, including Europe and the United States of America. This trend will continue with escalating climate change and CHIKV will no longer be confined to developing nations [3]. There is no specific treatment for chikungunya fever and care is only supportive, based on the symptoms. No licensed CHIKV vaccine exists. Thus, there is an urgent demand for the development of a prophylactic vaccine.
Several vaccine approaches have been developed; however, so far without resulting in a market-approved vaccine. CHIKV vaccines have either been formulated as formalin-inactivated CHIKV [4] or live-attenuated CHIKV vaccine candidates like the CHIKV 181/25 strain [5]. CHIKV 181/25 is attenuated by only two point mutations and reversions in vaccinated mice have appeared, suggesting that genetic instability is the source of its reactogenicity [6]. Internal ribosome entry site (IRES)-based live-attenuated CHIKV vaccines (CHIK-IRES vaccines) should circumvent this problem and would additionally prevent vaccine spread by mosquitos [7]. Other approaches are chimeric vaccine strains based on the genetic backbones from Sindbis virus or the TC-83 vaccine strain of Venezuelan equine encephalitis virus [8,9], [9], a DNA vaccine based on codon-optimized consensus envelope protein (E1, E2 and E3) sequences [10], a VLP-based vaccine expressing the CHIKV envelope proteins [11], or recombinant measles [12]. As sterilizing protection can be reached with CHIKV-specific antibodies [13], protein-based vaccines might be envisioned. Recently, an E2 protein-based vaccine candidate has been described that is able to protect mice from CHIKV challenge infections [14]. In order to ease production of vaccine antigens, we were interested to test whether small linear antigens would be sufficient to elicit a protective immune response against CHIKV.
CHIKV is a (+) single-stranded (ss) RNA virus and enters cells by receptor-mediated endocytosis in a pH-dependent fusion step. CHIKV has two surface proteins: the two transmembrane glycoproteins E2 and E1. E1 is a class II viral fusion protein and E2 most likely mediates cell attachment, although the cellular receptor is still unknown [15]. E2 and E1 associate as trimers of heterodimers (E2–E1) on the particle surface. The ß-sheet-containing E2 protein interacts with E1, covers the hydrophobic E1 fusion loop, and forms the center of the trimer [16]. The E2 protein is subdivided into three immunoglobulin domains called A, B and C. Domains A and B are implicated in receptor binding [16], [15]. Domain B is located at the membrane distal part and forms the tip of E2. It is connected with domain A via a long ß-ribbon connector. Domain A is located at the center and domain C is close to the viral membrane and most likely not efficiently accessible to antibodies [15]. A linear epitope located at the N-terminus of E2 (aa 1–12, termed E2EP3) has been described to be the target of early neutralizing IgG responses of CHIKV-infected patients, mice and monkeys [17], [13]. This main epitope is located proximal to the furin cleavage site and is therefore prominently exposed on the surface of the virus, forming a stalk that points away from the virus envelope.
Here, we derived surface-exposed regions of E2, produced the recombinant proteins in E.coli and analyzed the induction of neutralizing antibodies. N-glycosylation of the E2 protein is not expected to interfere with antibody binding, since the two glycosylation sites in E2 are outside the regions used for vaccination. An antigen able to induce neutralizing antibodies was identified and partially protected vaccinated mice from challenge infection.
All cells used for this study were cultured at 37°C under 5% CO2. HEK 293T (ATCC: CRL-1573) cells were incubated in Dulbecco’s modified Eagle’s medium (DMEM; Lonza, Verviers, Belgium). BHK 21 (CCL-10) cells were grown in Roswell Park Memorial Institute medium (RPMI; Biowest, Nuaille, France). RK13 (CCL-121) cells were cultured in Eagle's Minimum Essential Medium (EMEM; Biochrom, Berlin, Germany). The used media were supplemented with 10% FBS (v/v; PAA, Pasching, Austria) and 5% L-glutamine (200 mM; Lonza, Verviers, Belgium).
The codon-optimized CHIKV E3-E1 gene (based on isolate “S27”) was synthesized by GeneArt (Life Technologies, Darmstadt) and cloned into the plasmid pIRES2-eGFP (Clontech/Takara, 78100 Saint-Germain-en-Laye, France) as described previously [18].
The CHIKV E2-derived constructs L and sAB+ were synthesized by GeneArt (Regensburg, Germany) (codon optimized, strain LR2006 derived sequences see S1 Fig) and cloned into the plasmid pET-15b (Merck Millipore [Novagen]), Darmstadt, Germany) via NdeI and BamHI (for sAB+, primers see S1 Fig). Construct L contains five repeats of the sequence encoding S1- T12 linked by glycine-serine (G-S) linkers; construct sA encodes S1-T12, I56-G82, T94-H99, G114-H123 and Q158-T164 linked by G-S linkers; B+ encodes P172-H256. Constructs sA, B+, LsA, LB+ and LsAB+ where derived from the two synthesized genes via PCR and also cloned into pET-15b (already containing the L part for LB+, LsA and LsAB+) using the primers listed in the S1 Fig Sequence identities were verified by sequencing. Addition of a secretion signal to the construct sAB+ was implemented by cloning the synthesized gene into the vector pSecTag2 B (Invitrogen, Life Technologies, Darmstadt) via SfiI and ApaI.
The wild-type CHIKV used for infections was a kind gift of Matthias Niedrig (Robert-Koch-Institut, Berlin, Germany) [19].
To clone the secretion signal containing construct sAB+ into the MVA expression plasmid pIII-pmH5, the BamHI site in the plasmid was blunted via T4-DNA polymerase. The sAB+ gene was excised from the pSecTag2 B vector by NheI and PmeI and subsequently blunted. Ligation of insert and vector resulted in the MVA vector plasmid pIII-CHIKV-sAB+. Recombinant MVA-CHIKV-sAB+ was created by parallel transfection and infection of BHK21 cells with 1 μg of plasmid DNA and MVA (wild-type) at an MOI of 0.05. Plaque selection was done on RK-13 cells as described before [20], [21]. Successful generation of the recombinant MVA-CHIKV-sAB+ was confirmed by PCR of MVA genomes derived from infected BHK 21 cells (S2 Fig). Virus amplification was performed on BHK21 cells as described by [22], [23].
Proteins were expressed in BL21-CodonPlus (DE3)-RIPL competent cells (Agilent Technologies, Böblingen, Germany) transformed with the pET-15b plasmid containing construct L, sA, B+, sAB+, LsA, LB+ or LsAB+. Bacteria were inoculated into 100 ml of LB medium containing ampicillin (0.1 mg/ml) and grown overnight (37°C, 220 rpm). After 16 hours, 2 l of LB medium were inoculated with the 100 ml overnight culture. The bacteria were grown to an OD600 of 0.5–0.7, then protein expression was induced by the addition of 1 mM IPTG. After another 2.5 hours of incubation, cells were harvested and the pellets were frozen at -20°C.
All recombinant proteins were purified from the bacteria pellets under native conditions using HisTrap FF Crude columns (GE Healthcare, Freiburg, Germany) and the ÄKTA system (GE Healthcare, Freiburg, Germany) as described by [24]. After purification, proteins were dialyzed against PBS using Slide-A-Lyzer Dialysis Cassettes 3.5K MWCO (Pierce; Thermo Scientific, Bonn, Germany) and concentrated with Ultra-4 3 kDa Centrifugal Filter Units (Merck Millipore, Schwalbach, Germany). The protein concentration was compared to marker proteins by 12.5 or 15.0% SDS-PAGE and adjusted to 1 mg/ml. Subsequently proteins were shock-frozen with liquid nitrogen and stored at -80°C. For experiments, proteins were thawed in a 37°C water bath. The identity of the purified proteins was confirmed by mass spectrometry (for example S3 Fig).
Lentiviral vector particle production was performed as described previously [18]. Briefly, HEK 293T cells were seeded in 10 cm dishes in a volume of 10 ml DMEM. Cells were cotransfected 16 hours post seeding with the plasmids pCSII-Luc, pMDLg/pRRE, pRSVrev, and pHIT-G or pIRES2-eGFP-CHIKV E3–E1 using Lipofectamine 2000 (according to the manufacturer’s protocol; Life Technologies). After 24 hours incubation, the medium was discarded and replaced by 5 ml of fresh DMEM. Another 24 hours later, the supernatant containing the vector particles was harvested, sterile filtered with 0.45 μm filters (Sartorius, Göttingen, Germany), and frozen at -80°C.
Cells were transduced with pseudotyped lentiviral vector particles in 384-well plates as described previously [18]. Briefly, 6000 HEK 293T cells per well were seeded (using a MultiFlo Microplate Dispenser; BioTek, Bad Friedrichshall, Germany) in 20 μl of DMEM in white CELLSTAR 384-well microtiter plates (Greiner Bio-One, Frickenhausen, Germany) and incubated for 16–24 h at 37°C. Sera were serially diluted in DMEM and vector particles (4 times; dilutions ranged from 1:30 to 1:2340) and mixed 1:1 with diluted vector particles (CHIKV Env or VSV-G pseudotyped particles in DMEM) in 96-U-well plates (Thermo Scientific, Rockford, IL, USA), and incubated at 4°C for 1 hour. This resulted in equal amounts of vector and serially diluted compound. The vector particle-sera mixtures were then added to the cells using a Matrix Multichannel Equalizer Electronic Pipette (Thermo Scientific, Rockford, IL, USA), transferring 20 μl each to three wells of the 384-well plate out of one well of the 96-well plate. This resulted in a final concentration of serum ranging from 1:60 to 1:4680 in the 384-well plate. Cells were incubated with the vector particle-sera mixtures for another 16–24 hours. Afterwards, 20 μl of BriteLite (PerkinElmer, Rodgau, Germany) substrate was added. After 5 minutes incubation at room temperature, the luciferase signal was detected using the PHERAstar FS (BMG LABTECH, Ortenberg, Germany).
The proteins were separated by SDS-PAGE and the gel was subsequently stained with Coomassie (Bio-Rad, München, Germany). For Western blots, a Bio-Rad semi-dry blotter was used. Proteins were transferred onto PVDF membranes with a 50 mM sodium borate pH 9.0, 20% methanol, and 0.1% SDS buffer at 100 mA per membrane for 75 min. Subsequently, membranes were blocked with Roti-Block (Carl Roth, Karlsruhe, Germany). Proteins were detected with an anti-myc antibody (BD Pharmingen, Heidelberg, Germany), or with a ß-actin antibody (Sigma, Munich, Germany). Detection was performed with the ECL system (Amersham, Freiburg).
Viral RNA was isolated from mouse sera and organs with the QIAamp Viral RNA Mini Kit and the RNeasy Lipid Tissue Mini Kit (Qiagen, Hilden, Germany), respectively, according to the manufacturer’s protocol. CHIKV RNA levels were tested with the RealStar Chikungunya RT-PCR Kit 1.0 (Altona Diagnostics, Hamburg, Germany). The kit was used according to the manufacturer’s protocol. The readout was performed using a LightCycler 480 Instrument II (Roche, Basel, Switzerland).
Mice (female, BALB/c) (Janvier, Saint-Berthevin Cedex, France) were kept at the Paul-Ehrlich-Institut in ventilated cages. They were first immunized at an age of seven weeks. Each mouse was injected subcutaneously in the neck region with 100 μg of protein in PBS mixed 1:1 with alum (Alhydrogel 2%; InvivoGen, San Diego, California, USA) per immunization. Blood was collected immediately before the immunizations and after the final immunization. Infection was carried out intranasally with 1 × 106 PFU CHIKV in 30 μl PBS. Blood was collected on day two and four post infection and mice were sacrificed on day four post infection and organs were collected. Sera were obtained from blood using Microtainer SST tubes (BD, San Diego, CA, USA). For RNA isolation, sera were directly frozen at –80°C. Organs were also frozen at -80°C in 1 ml RNAlater RNA Stabilization Reagent (Qiagen, Hilden, Germany), according to the manufacturer’s protocol. For the luciferase neutralization assay, sera were first incubated at 56°C for 30 minutes and then frozen at -80°C. All experiments were performed in accordance with German legal requirements.
Experiments were performed in accordance to legal requirements (German protection of animals act (deutsches Tierschutzgesetz) and experimental animal regulation (Tierschutz-Versuchstierverordnung)) and after approval of the regional council Darmstadt, Germany (permit number V54 19c 20/15 F107/123).
Statistical analyses were done using the GraphPad Prism 5.04 software (La Jolla, CA, USA). For the p-values of the neutralization assay, the paired two-tailed t-test was performed. For the viral titers’ p-values and ELISA analysis of anti-CHIKV antibody reactivities, the unpaired two-tailed t-test was performed.
The CHIKV E2 protein is responsible for binding to the still unidentified cellular receptor. It is organized in three domains named A, B and C, where C is unlikely to be able to interact with antibodies [25] and was therefore excluded from the analysis. Based on the crystal structure of the E2 protein [15], putatively surface-exposed, linear antigens of domain A were chosen (including one 12 amino acid subunit of the L construct) and artificially assembled and linked by G-S linkers. This resulted in a small protein named sA (surface-exposed A). The entire domain B including a part of the ß-ribbon connector implied to be surface exposed was used unaltered to produce construct B+. In addition, the linear epitope located at the N-terminus of E2 (aa 1–12), described to be the main target of early neutralizing IgG responses of CHIKV infected patients [17], was used to construct a recombinant gene containing a multimer of five of these peptide sequences joined by G-S linkers. This construct was named L. The sequences of these constructs are provided in the Supplements. Additionally, the three components were combined with each other as illustrated in Fig 1A, resulting in seven recombinant genes containing a 6-histidine tag originating from the expression vector for protein purification. The resulting proteins were expressed in E.coli and purified under native conditions by Ni2+-affinity chromatography. Fig 1B gives an overview of the purified proteins separated by SDS-PAGE and stained with Coomassie.
The recombinant proteins were subsequently tested for their ability to induce neutralizing antibodies (nAb). Female BALB/c mice (n = 3 per recombinant protein) were immunized at an age of seven weeks. Additional immunizations were performed on days 25 and 53 after the first immunization. For each immunization, 100 μg of recombinantly expressed protein in PBS were mixed 1:1 with alum and injected subcutaneously into the neck region of the mice. Blood was always collected before each immunization and nine days after the last immunization (Fig 2A). First, the sera obtained after the third immunization were evaluated by ELISA to confirm seroconversion. Except one mouse vaccinated with LsA, all mice developed IgG antibodies against the immunogen, although to varying extends (S4 Fig). The proteins LsA, sA and L developed only weak CHIKV reactivity, however LsAB+, LB+ sAB+ and B+ were highly CHIKV-positive. Subsequently, the sera were analyzed for the presence of neutralizing antibodies by a neutralization assay based on blocking the entry of CHIKV-Env-pseudotyped lentiviral vectors encoding luciferase [18]. Transduction of HEK 293T cells with VSV-G-pseudotyped vectors served as a control for CHIKV specificity. The mouse sera were diluted 1:60 to 1:4680 and the area under the curve (AUC) of luciferase units obtained with preimmune sera was divided by the AUC obtained with sera after the third immunization (Fig 2B). Three mice were analyzed per construct and the mean values are given. As expected, transduction by VSV-G-pseudotyped vector particles was not affected by the mouse sera, demonstrated by an AUC ratio of 1 (Fig 2B). Likewise the recombinant proteins sA and L did not induce nAb. However, a statistically significant induction of neutralizing antibodies compared to VSV-G-pseudotyped vectors was detected in sera of LB+-, sAB+- and B+-immunized mice. The protein sAB+ induced the highest amount of nAb, illustrated by an AUC ration of 2.7. The sA protein apparently enhanced the induction of neutralizing antibodies in combination with domain B+, although sA alone did not induce nAb. Altogether these data show that B+ was necessary and sufficient to induce nAbs in mice.
To further evaluate the protein sAB+ as a potential CHIKV vaccine, we attempted to increase the immune responses directed against sAB+ by generating a recombinant vaccine virus based on the highly attenuated strain modified vaccinia virus Ankara (MVA). MVA allows high foreign protein expression although the virus does not replicate in human cells, giving it a very good safety profile. A eukaryotic secretion signal and a myc-tag was added to the gene encoding sAB+ and the construct was subsequently cloned into the MVA expression vector pIII-mH5 [26] generating the plasmid pIII-CHIKV-sAB+, where expression is controlled by a strong early/late promoter. The recombinant virus, MVA-CHIKV-sAB+, was constructed as described previously, using K1L selection [20], [26]. Quality control was performed by PCR. Wild-type MVA and MVA-CHIKV-sAB+ had the same growth kinetics in permissive cells (S5 Fig). Western blot analysis confirmed the proper expression of the sAB+ protein to high levels at late time points in cells infected with MVA-CHIKV-sAB+ (Fig 3A). To demonstrate that the sAB+ protein is secreted, we additionally analyzed cell lysates and supernatants of MVA-infected cells 72 hours after infection. Fig 3B shows a clear band in the MVA-CHIKV-sAB+ samples at the expected size in cell lysates as well as in supernatants, which indicates that the protein is secreted by infected cells.
CHIKV challenge infections were again implemented in BALB/c mice. Although in this model the animals show no clinical symptoms, temporal CHIKV replication can be observed in target organs (muscle, brain, spleen) and in blood. It has been shown before that the intranasal CHIKV infection resulted in viremia lasting 2–3 days, while intraperitoneal infection yielded less consistent results [27], [28]. We tested whether a protective immune response could be induced with the sAB+ protein, formulated as a protein vaccine, a recombinant MVA vaccine candidate or a mixture of both. Again, female BALB/c mice (n = 5 for the CHIKV antigen groups, n = 4 for the control groups) were immunized at an age of seven weeks. Additional immunizations were then performed three, four, and six weeks after the first immunization. Each mouse was injected subcutaneously in the neck region with either 100 μg of protein in PBS mixed 1:1 with alum as before or 1 x 108 PFU MVA-CHIKV-sAB+ per immunization. The mixed immunization was performed as a MVA-CHIKV-sAB+ (1 x 108 PFU) prime, followed by three immunizations of a cocktail of protein/alum/MVA-CHIKV-sAB+ as boosts. All mice immunized with MVA-CHIKV-sAB+ and/or sAB+ protein seroconverted and were CHIKV-Env-positive when tested by ELISA with recombinant protein B+ as an immunogen (S6 Fig). The CHIKV-directed reactivity was higher in sAB+ vaccinated animals compared to MVA-CHIKV-sAB+ vaccinated animals. Although statistically significance was only detected for the combined MVA-CHIKV-sAB+/sAB+ protein vaccinated mice (S6 Fig). Two weeks after the final immunization, the mice were infected intranasally with 1x106 PFU CHIKV in 30 μl PBS. Blood was collected two and four days post infection and the CHIKV titer in the sera, lung, spleen and brain was determined by RT-PCR. Fig 4 shows the viral titers as RNA copies in serum. A statistically significant reduction of viral titers compared to animals vaccinated with MVA wt was only detected after two days of infection in the sera of mice vaccinated with the sAB+ protein alone (Fig 4, lane 3). Mice that had been infected with CHIKV seven weeks before and challenged with a CHIKV infection, however, were fully protected (Fig 4, lane 6), indicating that protein sAB+ vaccination was not fully protective in this model. MVA-CHIKV-sAB+-vaccinated animals did not show any significant alteration in viral titers compared to control animals (mock or MVA wt) and the combination of MVA-CHIKV-sAB+ with protein did not significantly enhance the immune responses of MVA-CHIKV-sAB+. However, single mice in every vaccinated group exposed to CHIKV antigen showed reduced CHIKV titers, indicating some protective activity.
The intranasal infection causes in a short time virus replication in the animals, which resulted in very low viral titers in serum four days after infection in all groups. However there was a statistically significant decrease in viral titer in MVA-CHIKV-sAB and MVA-CHIKV-sAB/protein sAB vaccinated animals although the difference between the mean values was rather low. Analysis of organs at four days after infection revealed a statistically significant decrease in viral titers in spleens of mice vaccinated either with sAB+ protein or MVA-CHIKV-sAB+/protein sAB+ and in preinfected mice (Fig 4). In lung and brain only the preinfected mice showed a drastic decrease in viral titers (Fig 4). In lungs of MVA-CHIKV-sAB+ vaccinated mice a statistically significant decrease in CHIKV titer compared to MVA vaccinated mice could be observed.
Vaccination is considered to be the most efficient way to control the spread of CHIKV. Effective experimental vaccines have been reported before, such as attenuated CHIKV strains, inactivated CHIKV or virus-like particles (VLP). These approaches are problematic because of vaccine safety or require either large scale production of CHIKV in a BSL-3 laboratory or cell-based VLP production, both economically inconvenient processes. Adoptive transfer studies indicate that neutralization of viruses by anti-CHIKV antibodies protects against CHIKV infection [29], [13]. Therefore subunit protein vaccines might constitute safe, easy to produce, and economically favorable vaccines against CHIKV. Subunit vaccines are based only on a portion of the virus and elicit humoral immune responses. Approaches using the CHIKV E1/E2 proteins expressed in E.coli have shown promising protection of mice from CHIKV infections [30], [14]. Kumar et al. showed that recombinant E2 protein adjuvanted with alum protected mice from a challenge infection to a similar extend as inactivated CHIKV [14]. Also recombinantly expressed E1 as well as E2 proteins elicited a protective immune response [30].
Here we attempted to minimize the antigens of E2 and identify concise antigens able to induce neutralizing antibodies and confer protection. The antigens were designed by taking into account the observation that mutations in the alphavirus E2 domain A and B confer escape from neutralizing antibodies, affect tissue tropism and host range [15]. In a previous study [17], mice were vaccinated with the early, immunodominant N-terminal E2EP3 peptide coupled to KLH which induced a mild neutralizing activity and partial protective responses in vivo. Based on these data, we designed a pentamer of the E2EP3 peptide linked by G-S-linkers (L). Unfortunately this polypeptide did not induce any neutralizing antibodies upon vaccination of mice, although anti-L antibodies were detectable by ELISA. The E2PE3 epitope is located proximal to the E2–E3 furin cleavage site and is therefore prominently exposed on the surface of the virus to make the side accessible to the furin protease, suggesting that it is also susceptible to antibody binding. However, in contrast to other data [17], we did not observe the induction of neutralizing antibodies, which might be attributable to differences in antigen delivery or the adjuvants used for vaccination.
The second attempt to generate a synthetic polypeptide with surface exposed antigens of the E2 domain A (sA) also failed. Although the vaccination induced sA-specific antibodies, as before with the L polypeptide, no CHIKV neutralizing activity was detected in sera of immunized mice. In spite of this, the E2 domain B+ and combinations thereof with L or sA produced CHIKV neutralizing activity. The addition of sA may have stabilized the protein and resulted in higher neutralizing activities; nevertheless the main antigenic determinant must be located in domain B+. Domain B+ contains domain B, which covers the fusion loop in the E2–E1 dimer and prevents premature membrane fusion, and also the C-terminal part of the β-ribbon connector to domain A, a part of the acid-sensitive region (ASR). During the low pH fusion step, domain B and the ASR become disordered [16]. Antibodies binding to the ASR region or domain B could prevent the dissociation of E2 from E1 following pH triggering, reducing fusion efficiency and CHIKV entry. Domain B might also be involved in viral attachment to the cell surface [16]. Therefore, antibodies directed against domain B could also prevent the binding of viral particles to the cell. Thus, the superior potential of construct B+ to induce neutralizing antibodies might be due to its ability to induce antibodies which prevent viral attachment and fusion at the same time.
In a first attempt to show proof of principle that sAB+ is a suitable antigen for vaccine development, we generated a recombinant MVA expressing sAB+ as a secreted protein to enhance antigen delivery and perhaps boost the induction of cellular and humoral immunity at the same time. MVA is a highly attenuated vaccinia virus strain with a high safety profile suitable for clinical application in immunosuppressed patients [31], [32]. MVA does not replicate in human cells but shows high protein expression and, consequently, has a very good safety profile without compromising vaccination efficiency. MVA vaccination generates antigen specific cellular and humoral immunity [23]. The protein vaccine candidate was compared to MVA-CHIKV-sAB+ and combinations of MVA-CHIKV-sAB+ and protein sAB+. Adult mouse models for CHIKV infections are still not well established and immune responses can only be analyzed in mice that do not develop disease [33]. We chose intranasal challenge for infection of BALB/c mice which allows temporal CHIKV replication to be observed in target organs and in blood [14]. Other groups use A129 mice that are deficient in α/ß interferon signaling or C57BL/6 mice and different application routes like i.m., i.p. or i.v., which makes it difficult to compare vaccine efficacy from published data. Here, in general, CHIKV-specific vaccinated mice showed a trend to lower CHIKV replication in serum at day two after infection compared to mock or MVA wt vaccinated animals. However, a significant reduction (25-fold) in mean CHIKV titers was only observed in protein sAB+ vaccinated animals. Preceding infection of mice with CHIKV conferred a sterilizing immunization to the animals. In spleen, a significant reduction of viral titers was observed in protein sAB+ vaccinated animals as well as in MVA-CHIKV-sAB+ and protein sAB+ vaccinated animals. Furthermore, MVA-CHIKV-sAB+ vaccinated animals had a slightly lower viral titer in the lungs. In summary, this indicates that MVA-CHIKV-sAB+ vaccination was also immunogenic however with much lower efficiency than protein sAB+. The induction of humoral immune responses seems to be more efficient with protein vaccination than with MVA—derived delivery. Vaccination with recombinant E1 and E2 protein also only lowered the viral titer in blood after challenge infections in another study, although brain and muscle were completely protected [30]. Recently, several recombinant MVA expressing CHIKV antigens have been described [34], [35], [36] and all shown protection of vaccinated mice from a CHIKV challenge infection. MVA-CHIKV for instance expresses the CHIKV proteins C, E3, E2, 6K and E1 [35]. E2 was membrane anchored but formed no VLPs with the capsid C. MVA-CHIKV was shown to be highly immunogenic and triggered CHIKV-specific CD8+ T cell responses and high titers of neutralizing antibodies against CHIKV. A single dose of MVA-CHIKV protected all mice from challenge with CHIKV [35]. The differences to our data might arise either from the different antigen, the experimental setting or the MVA backbone used. MVA-CHIKV is based on a MVA variant that has immunomodulatory genes (C6L, K7R and A46R) deleted. MVA-CHIKV-sAB+ produces secreted proteins and was applied subcutaneously, MVA-CHIKV was applied intraperitoneally at a 10-fold lower dose. Challenge infections were done either intranasally or subcutaneously in the dorsal side of each hind foot. CHIKV titers were analyzed by plaque assays and, in our experiments, by the more sensitive RT-PCR. The role of cytotoxic T cells in CHIKV infections is still not well understood; however, these data imply that cytotoxic T cells induced by MVA-CHIKV may contribute to viral clearance. Only the side by side testing of all recombinant MVA expressing CHIKV antigens with the same mouse model will give clear indications on the efficacy of the vaccines.
In summary, we showed that the use of the small linear epitope L or surface exposed parts of A are not sufficient to induce a protective immune response, even if the L epitopes have been shown to be a main early target of antibodies in infected individuals. In contrast, linear antigens of domain A fused to the construct B+ (sAB+) were sufficient to induce neutralizing antibodies and to partly protect immunized mice from viremia. Most likely, B+ alone plays a much more prominent role in this than sA. Considering the high CHIKV dose used for the challenge infection, it might be expected that even in the formulation as protein and alum, the vaccination might be sufficient to protect against a mosquito derived CHIKV infection. However, optimization of antigen delivery might be useful and this small polypeptide might be an appropriate antigen for the development of a CHIKV vaccine.
|
10.1371/journal.pbio.1002308 | Bidirectional Regulation of Innate and Learned Behaviors That Rely on Frequency Discrimination by Cortical Inhibitory Neurons | The ability to discriminate tones of different frequencies is fundamentally important for everyday hearing. While neurons in the primary auditory cortex (AC) respond differentially to tones of different frequencies, whether and how AC regulates auditory behaviors that rely on frequency discrimination remains poorly understood. Here, we find that the level of activity of inhibitory neurons in AC controls frequency specificity in innate and learned auditory behaviors that rely on frequency discrimination. Photoactivation of parvalbumin-positive interneurons (PVs) improved the ability of the mouse to detect a shift in tone frequency, whereas photosuppression of PVs impaired the performance. Furthermore, photosuppression of PVs during discriminative auditory fear conditioning increased generalization of conditioned response across tone frequencies, whereas PV photoactivation preserved normal specificity of learning. The observed changes in behavioral performance were correlated with bidirectional changes in the magnitude of tone-evoked responses, consistent with predictions of a model of a coupled excitatory-inhibitory cortical network. Direct photoactivation of excitatory neurons, which did not change tone-evoked response magnitude, did not affect behavioral performance in either task. Our results identify a new function for inhibition in the auditory cortex, demonstrating that it can improve or impair acuity of innate and learned auditory behaviors that rely on frequency discrimination.
| Hearing perception relies on our ability to tell apart the spectral content of different sounds, and to learn to use this difference to distinguish behaviorally relevant (such as dangerous and safe) sounds. Recently, we demonstrated that the auditory cortex regulates frequency discrimination acuity following associative learning. However, the neuronal circuits that underlie this modulation remain unknown. In the auditory cortex, excitatory neurons serve the dominant function in transmitting information about the sensory world within and across brain areas, whereas inhibitory interneurons carry a range of modulatory functions, shaping the way information is represented and processed. Our study elucidates the function of a specific inhibitory neuronal population in sound encoding and perception. We find that interneurons in the auditory cortex, belonging to a specific class (parvalbumin-positive), modulate frequency selectivity of excitatory neurons, and regulate frequency discrimination acuity and specificity of discriminative auditory associative learning. These results expand our understanding of how specific cortical circuits contribute to innate and learned auditory behavior.
| Frequency discrimination is a fundamental task in everyday hearing and can be vitally important, as spectral differences can be used to distinguish dangerous and safe sounds [1–3]. However, our knowledge of the neuronal mechanisms that support frequency discrimination remains incomplete. The auditory cortex (AC) is involved in many auditory behaviors [4–13], with some studies suggesting that it controls frequency discrimination [14–16] (but see [5,17]). It remains poorly understood which aspects of neuronal circuits in AC contribute to behavioral frequency discrimination performance.
Neurons in AC exhibit frequency selectivity in their responses to tones [18–24], and modify their tuning properties with auditory learning [25–27], providing support for the involvement of AC in frequency discrimination. Many aspects of neuronal responses to tones, including magnitude of neuronal responses and width of tuning, can in principle affect behavioral performance [28]. Furthermore, learning and experience can lead to changes in tone-evoked response patterns in AC [25,27,29–31], affecting neuronal frequency tuning and selectivity. At present, a detailed understanding of the relation between tone response properties of AC neurons and frequency discrimination behavior remains missing.
Neurons in AC form mutually coupled excitatory–inhibitory networks, which shape the responses of neurons to sounds [32,33]. Electrophysiological recordings and pharmacological studies demonstrate that GABA-ergic inhibition controls tone-evoked response amplitude, spontaneous firing rate and frequency selectivity [25,27], among other aspects of excitatory neuronal responses. The most common type of interneurons, parvalbumin-positive interneurons (PVs) [11,34–38], which target the pyramidal cell bodies, gate feed-forward thalamocortical auditory inputs [38]. We postulated that optogenetically modulating PV activity would affect tone-evoked responses in the AC, thereby enabling us to examine the effect of changing tone response properties of AC neurons on auditory behavior and learning. We focused on two behaviors, frequency discrimination—driven prepulse inhibition (PPI) of the acoustic startle response (ASR), and differential auditory fear conditioning (DAFC). Frequency discrimination-driven PPI relies on an innate behavior—startle response to loud noise, and is controlled by subcortical circuits [39]. Because PPI decreases if the startle noise is preceded by a change in an acoustic stimulus, it can be used to measure frequency discrimination acuity [3,40]. By contrast, DAFC requires both learning and memory and is controlled by interactions between the cortex and a complex circuit including the amygdala and the hippocampus [41–44]. While these two behaviors rely on different brain circuits, they can affect each other [45], with the AC facilitating this interaction [3].
We found that cortical inhibition controls frequency discrimination acuity and frequency specificity of auditory fear conditioning. These behavioral changes were correlated with changes in the magnitude of tone-evoked neuronal activity.
To manipulate the level of activity of a specific type of inhibitory interneuron, PVs, in AC, we drove them to express Channelrhodopsin (ChR2) or Archaerhodopsin (Arch), using targeted viral delivery to AC (Fig 1A and 1F) [34,38,46]. Arch is a light-driven proton pump that hyperpolarizes the neuron when activated with green light [35]. Conversely, ChR2 is a light-gated cation channel that depolarizes the neuron when activated with blue light [47]. We injected a modified adeno-associated virus (AAV), which carried the antisense code for either opsin under the FLEX cassette in AC of PV-Cre mice. Following an incubation period, PVs in AC expressed ChR2 or Arch efficiently and with high specificity (Fig 1B and 1C and Fig 1G and 1H). Analysis of light-evoked responses of putative PVs showed that PVs have a distinct waveform with relatively deep troughs (S1 Fig). We used spike waveform shape as a criterion for exclusion of putative PVs from the pool of analyzed neurons.
Throughout the study, we compared the effects of interneuron modulation with that of direct increase in the activity of excitatory neurons by photostimulation. This control allowed us to test whether a simple elevation of the activity level of excitatory neurons can account for the observed results. In order to activate excitatory neurons directly, we drove them to express ChR2 in AC using targeted viral delivery in mice that express Cre recombinase in neurons under CamKIIα promoter. This resulted in efficient and specific expression of ChR2 in putative excitatory neurons in AC (Fig 1K–1M).
To verify the effectiveness of optogenetic modulation, we measured the effect of the laser on the spontaneous firing rate of AC neurons. Spiking activity of neurons in AC of awake, head-fixed mice was recorded during acoustic presentation of a random tone sequence, a stimulus designed to measure the frequency tuning curve of neurons. Locally shining either blue (473 nm) or green light (532 nm) suppressed or activated the activity of putative excitatory neurons confined to AC, respectively (S2 Fig). Activation of PVs (473 nm, 0.2 mW/mm2 intensity at the fiber tip) significantly reduced the spontaneous firing rate (FRbase, computed during the baseline period, 0–50 ms prior to tone onset) in a large fraction of recorded neurons (Fig 1D and 1E), resulting in a reduced mean spontaneous firing rate over the recorded neuronal population. The effect scaled with increasing light intensity: the index of change in FRbase increased with increased activation of ChR2 (S3A and S3B Fig). Conversely, suppression of PVs (532 nm, 10 mW/mm2) increased mean FRbase (Fig 1I and 1J). These changes in spontaneous firing rate demonstrate that the optogenetic manipulation of PV activity efficiently altered neuronal activity in the AC.
Direct optogenetic manipulation of excitatory neurons was similarly effective: photoactivation of CamKIIα neurons by blue light increased the spontaneous activity of neurons (Fig 1N and 1O). The effect of light in the CAMKIIα-ChR2 group was either the same or larger than in the PV-Arch group (S2 Fig), allowing for comparison of effects of suppression of PVs in PV-Arch group to direct activation of excitatory neurons in CAMKIIα-ChR2 group.
We next tested the function of PVs in behavioral frequency discrimination acuity. To determine whether PV activity affects behavioral frequency discrimination acuity, we measured the change in frequency discrimination threshold (Th), while modulating PV activity. Th was determined by measuring the percent inhibition of the ASR due to a shift in frequency between a background and a pre-pulse tone for varying frequencies of the pre-pulse [3,40] (Fig 2A). Strong PPI of the ASR indicates that the mouse detected the shift in frequency between the background and prepulse tones (Fig 2B, 2D and 2F). As previously reported [3], PPI increased with larger frequency shifts between the background and prepulse tones. This method thus provides psychometric response curves for frequency discrimination over the course of a single session that lasts less than 1 hr and does not require training the subject. Th was computed as the percent difference in frequency between the background and the prepulse tone that elicited 50% of the maximum PPI (S4 Fig). To test the effect of PV activity on Th, the laser was turned on during half of the behavioral trials, overlapping with the startle and prepulse stimuli (light-On trial). On the remaining (light-Off) trials, the laser was turned on at a quasirandom time during intertrial interval. In an additional test session, the light was not used throughout (no-light).
Optogenetic modulation of PV activity significantly affected behavioral frequency discrimination acuity. Activating PVs improved frequency discrimination acuity, as evidenced by a reduction in Th for light-On trials as compared to light-Off trials and no-light session in PV-ChR2 mice (Fig 2C). Suppressing PV activity reduced frequency discrimination acuity, leading to a significant increase in Th in PV-Arch mice (Fig 2E). Combined, these results demonstrate that the level of PV activity bidirectionally controls behavioral frequency discrimination acuity.
We performed several controls to ensure that the effects of photomodulation of PV activity were specific to the shift in frequency and could not be explained by a change in the ability of the mouse to respond to and to hear the stimuli. First, we tested whether light alone affected Th. In a control group of PV-Cre mice, in which PVs were driven to express only the fluorescent marker, but not the opsin, light did not affect Th (S5 Fig). This indicates that the observed change in Th required the expression of opsins in PVs. In mice expressing ChR2 or Arch, neither activation nor suppression of PVs affected the magnitude of ASR elicited by startle stimulus alone (S6A Fig). Therefore, the observed change in Th was not simply due to a change in the magnitude of ASR. Furthermore, activating or suppressing PVs did not lead to a change in the maximum PPI elicited by the pre-pulse tone (S6B Fig). These tests indicate that photomodulation of PV activity did not affect the ability of the mouse to detect large shifts in frequencies.
To further test that photomodulation of PVs did not impair the mouse's ability to hear test tones, we measured PPI due to the prepulse tones alone, without the background, as an estimate of how strongly the mouse could detect the prepulse tone. PPI elicited by the pre-pulse tones was not significantly different on light-On and light-Off trials (S7 Fig). Furthermore, there was no difference in PPI elicited by the prepulse tone at all six frequencies tested, indicating that mice detected the different tones similarly well on both light-On and light-Off trials. Taken together, these controls demonstrate that the observed change in Th cannot be explained by changes in more basic aspects of mouse hearing or a non-specific effect of photostimulation.
We repeated the experiments, activating the excitatory neurons directly in the CamKIIα-ChR2 group. In striking contrast to the effect of PV inactivation in PV-Arch mice, direct activation of principal neurons did not affect Th (Fig 2G). This result demonstrates that the change in Th due to photosuppresson of PV activity is specific to the effect of inhibitory interneurons, and is not simply due to an increase in the mean firing rate of excitatory neurons during PV suppression.
Can the changes in neuronal activity in AC evoked by the different types of optogenetic manipulation explain the behavioral results? To answer this question, we measured how strongly photostimulation of PVs affected the responses of neurons during tone presentation in a frequency band of one octave centered at the best frequency (BF). To estimate the relative strength of population neuronal responses to tones, we computed the tone-evoked response magnitude measured as a difference between mean firing rate during tone presentation (FRtone) and FRbase (Fig 3). Photomodulation of PVs resulted in a significant change in the magnitude of normalized tone-evoked response over the population of putative excitatory neurons. Photoactivation of PVs increased the tone-evoked response magnitude (Fig 3A and 3B). This effect was due to a relatively weaker decrease in FRtone as compared to the decrease in FRbase evoked by PV photoactivation (Fig 3A and S3 Fig). By contrast, photosuppression of PVs led to a decrease in tone-evoked response magnitude (Fig 3C and 3D). This effect was due to a relatively weaker increase in FRtone as compared to FRbase (Fig 3C). These results were consistent with the mean behavioral results for changes in Th: PV photoactivation, which improved behavioral frequency discrimination acuity, also increased mean tone-evoked responses; whereas PV photosuppression, which impaired behavioral frequency discrimination acuity, also suppressed mean tone-evoked responses in AC.
The effects of PV inactivation differed between subjects. Therefore, we computed a correlation between neuronal responses and behavioral performance over subjects by comparing the mean tone-evoked response magnitude over all neurons and behavioral Th for each mouse. Changes in neuronal responses caused by photomodulation of PV activity were significantly inversely correlated with changes in Th measured behaviorally (Fig 3G). This correlation suggests that the measured change in magnitude of tone-evoked responses in AC is a good predictor for the change in behaviorally measured frequency discrimination acuity.
By contrast, direct photoactivation of excitatory neurons in the CamKIIα-ChR2 group did not affect the tone-evoked response magnitude (Fig 3E and 3F). This result is due to the strong increase in both the spontaneous and tone-evoked activity of recorded neurons by direct photoactivation of excitatory neurons (S8A and S8B Fig). These results are consistent with the lack of change in behavioral frequency discrimination acuity due to photoactivation of excitatory neurons.
Combined, our findings support the interpretation that both the bidirectional modulation of Th due to PV stimulation, and the lack of modulation due to excitatory neuronal stimulation, are due to changes in the magnitude of tone-evoked responses relative to the baseline firing rate of AC neurons.
Frequency discrimination may be controlled not only by the firing rate of neurons but also by their frequency tuning properties [48]. Therefore, we next quantified the effect of PV photo-modulation on the frequency tuning properties of putative excitatory neurons. The mean firing rate of neuronal responses to tones was used to construct a tuning curve for the frequency and intensity level of the tones, computed on light-Off and light-On trials, separately (Fig 4A–4C).
It has previously been suggested that excitatory and inhibitory inputs to the same neurons exhibit similar frequency tuning properties in AC [49]. We therefore expected that the BF (the frequency of the tone eliciting the highest firing rate) would not be affected by PV photostimulation. Indeed, the BF of recorded units was not affected by PV photoactivation and photosuppression (Fig 4D and 4E, respectively). However, PVs exhibit tuning that is similar [50] or lower [51] in selectivity to excitatory neurons. Therefore, manipulation of PV activity would likely affect the frequency selectivity of putative excitatory neurons to tones. Indeed, photostimulation modulated frequency selectivity of neuronal responses. We quantified frequency selectivity by two measures: the width of frequency tuning and the sparseness of the frequency response function. Tuning width was computed as twice the standard deviation of the Gaussian fit to the frequency response function. Tuning width decreased during activation of PVs and increased during suppression of PVs (Fig 4G and 4H, respectively). We used sparseness as an additional measure for frequency tuning selectivity, because it is less sensitive to the magnitude of the firing rate as well as spontaneous firing rate than tuning width. In addition, sparseness does not assume a specific shape of the frequency response function. A sparseness value of 1 indicates that the neuron responds to tone at only one frequency, whereas a sparseness value of 0 indicates that the neuron responds equally strongly to tones at all frequencies. Activating PVs significantly increased sparseness over the population of putative excitatory neurons (Fig 4J). The strength of the effect of photoactivation on neuronal sparseness increased with light intensity (S9A and S9B Fig) and was significantly correlated with the change in the baseline firing rate (S9C Fig). Conversely, suppressing the activity of PV interneurons significantly reduced the sparseness of neuronal responses to tones (Fig 4K). As expected, the effects of photomodulation on sparseness and tuning width were significantly correlated (Fig 4M and 4N). Combined, we found that up- or down-regulating activity of PVs did not affect the BF of neurons, but modulated the tuning selectivity of principal AC neurons, such that activating PVs increased neuronal frequency selectivity, whereas suppressing PVs reduced neuronal frequency selectivity.
On average, the mean changes in frequency selectivity were consistent with behavioral results: activation of PVs, which improved frequency discrimination acuity, increased frequency selectivity in AC neurons, whereas suppression of PVs, which impaired frequency discrimination acuity, decreased frequency selectivity in AC. However, when examined on an animal-by-animal level, there was no significant correlation between frequency sparseness and the change in behavioral threshold when tested using either parametric or nonparametric tests (Fig 4P). This result suggests that mean frequency selectivity may not be as important for behavioral frequency discrimination acuity as the response magnitude for tones of preferred frequencies.
We next tested whether photoactivation of excitatory neurons affected mean neuronal frequency tuning. Over the population of recorded neurons, the BF was not affected (Fig 4F). However, the tuning width increased significantly (Fig 4I), whereas sparseness of frequency responses decreased (Fig 4L). As in PV-Cre mice, the tuning width and sparseness significantly correlated with each other (Fig 4O). These measurements contrast with the behavioral findings that photoactivation of excitatory neurons does not affect frequency discrimination acuity, further supporting the interpretation that frequency selectivity may not be as important for behavioral frequency as changes in tone-evoked response magnitude.
Thus far in the behavioral test, we examined frequency discrimination acuity using a modified procedure that relied on measuring inhibition of the startle response by a tone preceding the startle noise—an innate behavioral response measured as PPI. We then tested whether inhibition in AC also modulated auditory associative learning [52]. In DAFC, the mouse is presented with two tones of different frequencies, one of which (CS+) is associated with an aversive stimulus (mild electric foot shock) and one that is not (CS−) (Fig 5A and S10 Fig). 24 h later, the mice typically exhibit an increase in conditioned response (freezing) during presentation of CS+ and a smaller increase in freezing during presentation of CS−. For different subjects, the freezing response may be specific to CS+ or generalize to tones at frequencies beyond CS− [3]. We hypothesized that specificity of freezing after conditioning may be controlled by PVs in AC. To test this hypothesis, we measured whether up- or down-regulating the activity of PVs in AC during conditioning affects the specificity of the freezing response.
We subjected four groups of mice to DAFC, overlapping light and tone presentation. In the PV-Arch group, suppression of PVs during conditioning led to activation of putative excitatory neurons. In the PV-ChR2 group, photoactivation of PVs during conditioning led to suppression of putative excitatory neurons. In the control group of mice, which were injected with control vector that encoded only fluorescent protein, PVs were not affected by the laser. In CamKIIα-ChR2 group, the activity of excitatory neurons was enhanced during conditioning. 24 h following DAFC, we tested the level of freezing to CS+, CS− and two additional tones during the LS test, designed to measure how specific freezing response was to conditioned tones (Fig 5A). We then assessed the level of specificity of conditioned response by measuring the relative difference in freezing response to the CS+ tone and mean freezing response to test tones (LS index, Methods).
In all groups, mice exhibited an increase in the freezing response to CS+ (Fig 5B and 5C). However, mice in which PVs were suppressed during conditioning did not exhibit differential freezing response to CS+ and CS−. By contrast, mice in both the PV-ChR2 and the control groups exhibited a significant reduction in freezing to CS− as compared to CS+. Furthermore, the specificity of learned freezing as measured by LS was significantly lower than for mice in PV-Arch group than for mice in control group (Fig 5D). Interestingly, direct activation of excitatory neurons in CamKIIα-ChR2 group did not result in significant change of LS (Fig 5C and 5D). Thus, we find that suppressing PV activity during conditioning led to a decrease in specificity of auditory fear conditioning, whereas either increasing PV activity or increasing the general level of activity of excitatory neurons did not have a significant effect on the specificity.
As expected, between subjects, the level of specificity of conditioned fear varied. If inhibition in AC controls both the frequency discrimination acuity and specificity of the conditioned response via a similar mechanism, we expected the behavioral measures for acuity and specificity to be correlated. Indeed, change in behavioral frequency discrimination acuity due to photostimulation was significantly correlated with the change in specificity of auditory fear conditioning (Fig 5E). Furthermore, there was a significant correlation between LS and the effect of photomodulation of PVs activity on neuronal tone-evoked response magnitude but not sparseness (Fig 5F). Combined, these findings demonstrate that neuronal response magnitude in AC regulates not only behavioral frequency discrimination acuity measured through a test of innate behavior, but also specificity of associative learning.
We investigated a model of excitatory–inhibitory circuit interactions to better understand why manipulation of activity of PVs, but not excitatory neurons, affects the magnitude of tone-evoked responses. We constructed a firing-rate model as an extended Wilson-Cowan model of mutually connected excitatory and-inhibitory neuronal populations [53–55]. In this model, the inhibitory neuronal population integrates depolarizing currents from tone-evoked inputs and inputs from the excitatory neurons, whereas the excitatory neurons integrate tone-evoked inputs and hyperpolarizing currents from inhibitory neurons (Fig 6A, S14A Fig and Methods). Optogenetic modulation was modeled as an additional input current delivered to either excitatory or inhibitory neuronal populations. This simple simulation provided for a biological implementation of the circuit that is consistent with our experimental findings.
Inputs from PVs to excitatory neurons have been shown to exhibit synaptic depression [56,57]. We incorporated synaptic depression at the PV to excitatory synapse in the model (Fig 6). The model here did not assume a specific form (e.g., pre- or postsynaptic) of depression. Rather, we modeled synaptic transfer function as a nonlinearity, using a closed form solution for the relation between the output of the inhibitory neuronal firing rate and the input current for the excitatory neuronal population assuming depressing synaptic dynamics (see Methods). A simulation of excitatory neuronal responses exhibited the differential effects of inhibitory and excitatory stimulation of interneurons as well as lack of effect of stimulating the excitatory neurons directly on tone-evoked responses: Activating PVs increased the tone-evoked responses, whereas suppressing PVs decreased the tone-evoked responses of the excitatory population (Fig 6B and 6C). By contrast, activating excitatory population directly did not change the tone-evoked response magnitude (Fig 6B and 6C). This simulation thus provides for one plausible biological implementation of the circuit that is consistent with our experimental findings.
To develop a more basic understanding of the circuit, we implemented an instantaneous sigmoidal input–output nonlinearity with varying coefficients after synaptic integration for either excitatory or inhibitory neurons (S14 Fig). We used three different scenarios (S14A Fig): under scenario 1, the nonlinearity operates in a linear regime for both the excitatory and the inhibitory populations; under scenario 2, the nonlinearity is saturating for the excitatory, and linear for the inhibitory, neuronal population; under scenario 3, the nonlinearity operates in a saturating regime for the inhibitory population, and a linear regime for the excitatory population. Only scenario 3 (S14B–S14E Fig right) supported our experimental findings that a) suppressing PV activity increased the magnitude of tone-evoked responses (Fig 3A); b) increasing PV activity decreased the magnitude of tone-evoked responses (Fig 3C); and c) activating excitatory neurons directly did not affect tone-evoked response magnitude (but increased both the spontaneous and the tone-evoked firing rate by the same amount) (Fig 3E). Under scenario 1, activation of excitatory neurons did not affect tone-evoked response amplitude, but neither did activation or suppression of inhibitory neurons (S14B–S14E Fig, left). Under scenario 2, activation or suppression of inhibitory neurons decreased or increased tone-evoked response magnitude, respectively (S14B–S14E Fig center); however, activation of excitatory neurons decreased tone-evoked response magnitude. Therefore, the scenario 3, under which the excitatory neurons integrate their inputs close to linear, but the inhibitory inputs are passed through a saturating nonlinearity, is consistent with our data (S14B–S14E Fig right). The synaptic depression model (Fig 6) can be viewed as a special case of scenario 3, in which the transfer function between inhibitory and excitatory neuronal population saturates. Indeed, a number of other circuits, for example activation of an additional class of interneurons, such as somatostatin-positive interneurons [51], could potentially provide for a saturating transfer function.
Our results demonstrate that auditory cortical neurons regulate auditory behaviors that rely on frequency discrimination, and that this regulation can be facilitated by the overall activity level of a specific type of inhibitory, but not excitatory neurons. Optogenetic modulation of the level of activity of PV-positive interneurons drove changes in frequency discrimination acuity and specificity of auditory conditioning (Fig 2 and Fig 5). At the neuronal level, we find that modulating the level of PV activity differentially affects the spontaneous and the tone-evoked responses of putative excitatory neurons (Fig 1 and Fig 3). The changes in tone-evoked responses magnitude were correlated with behavioral performance (Fig 3G and Fig 5F). Whereas activating PVs during fear conditioning preserved specificity of conditioned fear, consistent with a previous pharmacological study [58], suppressing PVs increased generalization of fear responses. These effects of PVs extend beyond controlling the overall firing rate of excitatory neurons as changing the gain of excitatory neuronal responses directly did not lead to similar changes in behavioral performance (Fig 2G and Fig 5C). This difference can be attributed to a nonlinear relationship between inhibitory input from PVs and output firing rate of excitatory neurons, consistent with a mechanism of synaptic depression that has been identified at the synapse from PVs to excitatory neurons (Fig 6) [57]. Combined, our results support the view that PVs regulate signal-to-noise ratio of responses of principal neurons, extending beyond the effect of a global gain control, and that this dual effect on the spontaneous and tone-evoked activity affects behavioral frequency discrimination.
Our electrophysiological results demonstrating that activating PVs leads to narrower frequency tuning of putative excitatory neurons whereas suppressing PVs leads to broader frequency tuning (Figs 3 and 4) are consistent with previous pharmacological and electrophysiological investigations of inhibitory neuronal responses [59–63]. Behaviorally, while PVs have been implicated in two separate auditory behaviors: detection of temporal gap in sound [11] and in disinhibition of responses to tones during aversive stimulus presentation in AFC [43], our results provide for the initial demonstration of the role of PVs in auditory tasks relying on frequency discrimination. Our findings are thus consistent with those in the visual system, where PVs have been found to modulate responses of principal cells to visual stimuli and affect visual discriminative behavior [64–67].
Optogenetic approaches act on different timescales than lesion studies, or pharmacological methods for neuronal activity suppression. Lesioning or pharmacologically inactivating AC previously provided mixed effects on frequency discrimination, with some studies resulting in small, if any impairments in frequency discrimination performance [3,5,17,68], whereas other studies exhibited stronger effects [69]. These results are not inconsistent with the present findings: lesioning and pharmacological studies are performed on much longer time scales (hours to days [59]), as compared to the millisecond timescale of optogenetic perturbation. Therefore, lesioning or pharmacologically suppressing AC potentially allows for other neuronal circuits to take over frequency discrimination function, similarly to brain reorganization in response to injury [70] or simply abnormal lack of activity. Our results therefore support a modulatory, but not necessary, role for AC in frequency discrimination: when AC is “online”, excitatory–inhibitory circuits control frequency discrimination behavior, and their perturbation modulates frequency discrimination behaviors. By contrast, lesioning or suppressing AC pharmacologically for extended periods of time potentially allows for other brain areas to take over control of frequency discrimination.
Behavioral frequency discrimination acuity was tested through a task that is based on an innate, rather than learned response [3,40,71]. Implementing the PPI-based behavioral task has the advantage that the animal does not need to be trained on the task, and therefore allows for dissociation of perceptual report and learning. A recent study has found that corticocollicular feedback affects learning-induced changes in auditory spatial learning [72]. Here, similarly, AC may affect PPI through corticocollicular feedback, as PPI is controlled by the inferior-colliculus to pedunculopontine nucleus connection [39,73]. Future studies, including a test of the effect of inactivation of corticocollicular feedback, are needed to determine which of the possible circuits downstream of AC drive the observed behavioral changes.
Regulation of auditory frequency discrimination by the AC is not restricted to the PPI circuit, as we find that AC also regulates how specific conditioning is to a particular frequency of the tone. A number of studies have demonstrated that the AC plays an important role in fear conditioning [30]. Our results identify that the AC shapes frequency specificity of DAFC: suppressing the activity of interneurons decreased the specificity of DAFC, as the subjects generalized the conditioned response to the full range of tones on which they were tested (Fig 5). Several circuits may underlie this effect: AC projects to the amygdala, a crucial brain area in auditory fear conditioning, via the secondary AC or via feedback through the thalamus [41,74]. Applying selective manipulation to elements in these circuits in future studies will be necessary to learn how AC controls associative learning. Interestingly, activating PVs did not increase the specificity of auditory associative learning, measured by LS, as would have been expected from frequency discrimination results. This suggests that the limits to specificity of auditory associative learning may not only be set by the AC, but may also rely on other brain regions, which would have a lower frequency resolution than the AC. Furthermore, it points to an asymmetry between the effects of activation or suppression of circuit elements: taking out a crucial element of a circuit led to a qualitatively different effect than increasing the activity of an already present element.
Our results point to remarkable robustness of frequency discrimination to the overall level of activity in the AC. Whereas direct photoactivation of excitatory neurons dramatically increased the overall firing rate in the cortex, at the behavioral level, we did not observe a change in either behavioral frequency discrimination, as measured by Th, or in specificity of DAFC (Fig 2F and 2G, Fig 5C). This robustness to the mean firing rate level may underlie important perceptual effects, such as the ability to preserve acoustic discrimination or speech comprehension in different acoustic environments.
Our results provide for a mechanism by which the AC may modulate learning-driven changes in frequency discrimination following emotional learning [3]. Previously, we found that frequency discrimination acuity and specificity of learning were correlated across subjects, pointing to a common mechanism that controls the two behaviors. We identified AC as a candidate brain area for controlling frequency discrimination acuity and DAFC, as pharmacological inactivation of AC abolished DAFC-induced change in frequency discrimination acuity [3]. Inhibitory neurons differentially process auditory information and are affected by auditory learning and experience [13,75,76]. Our present results are consistent with the possibility that the learning-driven changes in frequency discrimination may be due to inhibitory interneuron activity or plasticity in inhibitory–excitatory connections.
Combined, we find that modulating frequency response properties of neurons in AC via activity of PVs modulates frequency discrimination acuity and specificity of auditory associative learning, confirming an important role for inhibitory circuits in AC in auditory behavior. While PVs are the most common type of interneurons in AC, other interneuron types, such as somatostatin-positive and vasoactive intestinal peptide-expressing inhibitory interneurons, may play additional complementary roles in shaping frequency discrimination, through more complex circuits. It will be important to tease apart the function of different cortical circuits in the processing of spectral information.
All experiments were performed in adult male mice (supplier: Jackson Laboratories; age, 12–15 wk; weight, 22–32 g; PV-Cre mice, strain: B6; 129P2-Pvalbtm1(cre)Arbr/J; CamKIIα-Cre: B6.Cg-Tg(CamKIIα-Cre)T29-1Stl/J; wild-type control: C57BL/6J) housed at 28°C on a 12 h light–dark cycle with water and food provided ad libitum, less than five animals per cage. In PV-Cre mice Cre recombinase (Cre) was expressed in PPI, and in CamKIIα-Cre, Cre was expressed in excitatory neurons [77]. All animal work was conducted according to the guidelines of University of Pennsylvanian IACUC and the AALAC Guide on Animal Research. Anesthesia by isofluorane and euthanasia by carbon dioxide were used. All means were taken to minimize the pain or discomfort of the animals during and following the experiments. All behavioral experiments were performed during the animals' dark cycle.
At least 10 d prior to the start of experiments, mice were anesthetized with isoflurane to a surgical plane. The head was secured in a stereotactic holder. The mouse was subjected to a small craniotomy (2 x 2 mm) over AC under aseptic conditions. Viral construct was injected using syringe pump (Pump 11 Elite, Harvard Apparatus) targeted to AC (coordinates relative to bregma: −2.6 mm anterior, ±4.2 mm lateral, +1 mm ventral). Fiber-optic cannulas (Thorlabs, Ø200 μm Core, 0.22 NA) were implanted bilaterally over the injection site at depth of 0.5 mm from the scull surface. Craniotomies were covered with a removable silicon plug. A small headpost was secured to the skull with dental cement (C&B Metabond) and acrylic (Lang Dental). For postoperative analgesia, Buprenex (0.1 mg/kg) was injected intraperitonially and lidocaine was applied topically to the surgical site. An antibiotic (0.3% Gentamicin sulfate) was applied daily (for 4 d) to the surgical site during recovery. Virus spread was confirmed postmortem by visualization of the fluorescent protein expression in fixed brain tissue, and its colocalization with PV or excitatory neurons, following immuno-histochemical processing with the appropriate antibody.
Modified AAV vectors were obtained from Penn VectorCore. Vector encoding light-gated proton pump Archaerhodopsin (Arch) under FLEX promoter was used for selective suppression of PVs (Addgene plasmid 22222, AAV-FLEX-Arch-GFP [35]). Modified AAV encoding ChR2 under FLEX promoter (Addgene plasmid 18917 AAV-FLEX-ChR2- tdTomato, ChR2 [78]) was used for activation of either PVs iin PV-Cre mice and or excitatory neurons in CamKIIα-Cre mice. Modified AAV vectors encoding only GFP or tdTomato under FLEX promoter were used as a control for the specific action of Arch and ChR2 on the neuronal populations.
Brains were extracted following perfusion in 0.01 M phosphate buffer pH 7.4 (PBS) and 4% paraformaldehyde (PFA), postfixed in PFA overnight and cryoprotected in 30% sucrose. Free- floating coronal sections (40 μm) were cut using a cryostat (Leica CM1860). Sections were washed in PBS containing 0.1% Triton X-100 (PBST; 3 washes, 5 min), incubated at room temperature in blocking solution (10% normal goat serum and 5% bovine serum albumin in PBST; 3h), and then incubated in primary antibody diluted in blocking solution overnight at 4°C. The following primary antibodies were used: anti-PV (PV 25 rabbit polyclonal, 1:500, Swant), or anti-CAMKIIα (abcam5683 rabbit polyclonal, 1:500, abcam). The following day sections were washed in blocking solution (3 washes, 5 min), incubated for 1hr at room temperature with secondary antibodies (Alexa 594 or Alexa 488 goat anti-rabbit IgG; 1:1,000), and then washed in PBST (4 washes, 10 min). Sections were mounted using fluoromout-G (Southern Biotech) and confocal or fluorescent images were acquired (Leica SP5 or Olympus BX43). To quantify viral expression efficiency and specificity, cells in the proximity of injection site were identified in independent fluorescent channels and subsequently scored for colocalization using ImageJ’s cell counter plug-in.
Neurons were stimulated by application of continuous light pulse delivered from either blue (473 nm, BL473T3-150, used for ChR2 stimulation) or green DPSS laser (532 nm, GL532T3-300, Slocs lasers, used for Arch stimulation) through implanted cannulas. Timing of the light pulse was controlled with microsecond precision via a custom control shutter system, synchronized to the acoustic stimulus delivery. Prior to the start of the experiment, the intensity of blue laser was adjusted to one of three values 0.2, 0.5, or 10 mW/mm2 as measured at the tip of the optic fiber. On average, the lowest light power was sufficient to induce significant reduction in Th (paired t test, t19 = 2.68, p = 0.015). However, in a small fraction of mice (6 out of 20), higher power was needed to induce reduction in Th (0.5 mW/mm2 in 5 mice, and 10 mW/mm2 in 1 mouse). The same power was used in auditory discriminative fear conditioning for each subject. Green laser was used at intensity of 10 mW/mm2, which resulted in similar absolute magnitude of change in spontaneous firing rate over the neuronal population as the lowest level of ChR2 activation (S3 Fig). An additional experiment carried out using 6 subjects (2 PV-Arch, 2 PV-ChR2 and 2 CAMKIIα-ChR2) demonstrated that photoactivation and suppression of neurons was confined to the AC (S2 Fig). The effect of light on firing rate significantly decayed over distance, but was heterogeneous over cortical depth (S2 Fig).
During FC, the mouse was placed in a conditioning cage with a shock floor (Coulbourn) inside sound attenuation cubicle (Med Associates), housed in a single-walled acoustic chamber (Industrial acoustics). Throughout conditioning, the cage was illuminated with LED light, the color of which corresponded to the color of laser used for photoactivation of neurons (blue LED: 470 nm, 170 mW; green LED: 525 nm, 7 mW). During LS tests, a custom-made test cage of similar size but different floor and wall pattern and color was used. Auditory stimuli were provided by a free-field magnetic speaker (Tucker-Davis Technologies). Electric shock (0.5mA, 0.5 s) was delivered by precision animal shocker (Coulbourn). Freezeframe-3 software (Coulbourn) was used for stimulus control and analysis of freezing behavior.
During the PPI procedure, the mouse was placed in a custom-made tube on the sensor plate (San Diego Instruments) and head-fixed using implanted headpost. The speaker, housing, platform and webcam (Logitech) were placed in the sound attenuation cubicle (Med Associates), housed in a single-walled acoustic chamber. During tests, the housing was illuminated with LED light, the color of which corresponded to the color of the laser used for photoactivation of neurons. The speaker was positioned above the mouse. The sound delivery apparatus was calibrated using a 1/8-inch condenser microphone (Brüel&Kjær, Denmark) positioned at the expected location of the mouse's ear, to deliver each stimulus at 70 dB sound pressure level relative to 20 microPa (SPL). All pure tones presented during training and test sessions were at 70 dB SPL.
Seven to ten days after surgery, mice were subjected to at least three consequent days of habituation to experimental setups. During habituation to PPI apparatus, the duration of which gradually increased from 10 to 20 min over 3 d, mice were head fixed and optic fibers connected to cannulas. Following habituation, mice underwent daily PPI testing for frequency discrimination, which lasted for 1–3 d. Following PPI testing, a subset of mice underwent fear conditioning (FC) and one day thereafter they were tested for specificity of conditioned fear response. After termination of behavioral experiments, mice were used for electrophysiological recordings. In order to examine whether fear conditioning alters the effect of photoactivation on base firing rate of neurons and their tuning properties, we performed recordings in subgroup of PV-ChR2 mice without subjecting them to fear conditioning (“naïve” group, n = 6). Comparison of the change in spontaneous and tone-evoked firing rate and sparseness induced by photoactivation between “naïve” group and group that underwent fear conditioning did not reveal significant difference (S11 Fig). Therefore, recording data collected from these groups were pooled. All behavioral experiments were performed during animals’ dark cycle.
The measurement of frequency discrimination acuity used a modified PPI of the startle reflex protocol as previously described [3,40]. The test measured the magnitude of the ASR to the startle stimulus (SS) as a function of the difference in frequency between the background tone and the prepulse tone (PP), which immediately preceded SS. The frequency of the background tone was 15.0 kHz. The background tone (when used) was presented continuously between the end of SS and the start of PP. The transition between the background tone and PP included 1 ms ramp to avoid clicks. Five frequencies used for PP (10.2, 12.6, 13.8, 14.7, and 15.0 kHz) were presented pseudo randomly with 10–20 s ISI, which also varied randomly. Thus, PP differed from the background tone by 0, 2, 4, 8, 16 and 32%. PP was 80 ms long and was presented right before SS. SS was broadband noise, presented at 100 dB SPL for 20 ms.
The magnitude of ASR was measured using a forcesensor plate (San Diego Instruments) and defined as the maximum vertical force applied within the 500 ms window following SS minus average baseline activity during 500 ms prior to SS. In each PPI session, 50% of the strongest ASRs for each frequency were averaged and used to calculate PPI:
PPI(%)=100ASRnoPP−ASRPPASRnoPP
where ASRnoPP is the response when PP frequency is equal to the frequency of the background tone (15 kHz) and ASRPP is the response after frequency shift has occurred.
(a) To assess baseline frequency discrimination mice were subjected to the PPI procedure without photostimulation of neurons. Each test session consisted of 9 startle-only trials, followed by at least 100 pre-pulse trials, followed by one additional startle-only trial. On startle-only trials, background tone was followed directly by SS. On pre-pulse trials, each PP was presented 20 times in quasi-random order with ITI varying randomly between 10 and 20 s. Negative frequency changes were used because mice have been previously shown to be more sensitive to downward frequency shifts[3,40].
(b) To compare the effect of photoactivation or suppression of PVs on frequency discrimination, mice were subjected to a protocol similar to that described above, but including light delivery though implanted cannulas. On light ‘On’ trials, the laser was presented for 1 s, starting 0.5 s before PP onset. On light ‘Off’ trials laser was presented at quasi-random position during ITI. ‘On’ and ‘Off’ trials were shuffled randomly.
(c) To compare the subjective detectability of 5 experimental tones the background tone was omitted. The session started with 5 startle-only (no PP presentation) trials, followed by 50 pre-pulse trials, and terminated by 5 additional startle-only trials. On pre-pulse trials, each PP was presented 10 times in quasi-random order with ITI varying randomly between 10 and 20 s. The amplitude of each tone was then adjusted so that PPI induced by each tone was similar (S12 Fig).
The Th was defined as a frequency shift that caused 50% inhibition of the maximum ASR. Th is determined from a parametric fit to a generalized logistic function:
PPI=a1+exp(b+cΔf)
In a standard PPI session, 20 repetitions of each PP were presented (100 trials in total). However, if either Th was out of the range (0.5–32%) or the fit coefficient of the curve (R2) was below 0.7, the mouse underwent an additional 10 repetitions (50 trials). If Th and fit curve failed to meet the above criteria after 200 trials, the session was excluded from statistical analysis (3 out of 61 sessions).
During FC, following 5 min of silence, 10 tones (15.0 kHz, 10.5 s) co-terminated with a foot shock (CS+) were presented, at an inter-trial interval (ITI) randomly varied between 2 to 6 min. In addition, 10 tones at 11.25 kHz (10.5 s), not paired with foot-shock (CS-) were presented in random order with 2 min inter-stimulus interval (ISI). Photoactivation and suppression of neurons was performed by delivery of light through implanted cannulas. In one group of mice, photo stimulation started 0.5 s before CS+ and CS- onset and co-terminated with the tone (11 s total). In another group, photo stimulation was terminated 1 s before the tone offset to avoid overlapping with the foot-shock (10 s total).
The LS test consisted of CS+ and three test tones (3.75, 7.5, 11.25 kHz), presented 3 times at 3 min ISI. LS was assayed as the difference in freezing response to CS+ and mean freezing response to three test tones:
LS(%)=100FCS+−〈Ftest〉FCS+
Where FCS+ is freezing (%) during CS+ tone presentation and Ftest is mean freezing during test tones.
During conditioning and test sessions, freezing responses were video-recorded and analyzed offline using Freeze Frame software. Freezing responses were judged as complete immobility of the mouse for at least 1 s. Average freezing response during 20 s before the test tones was recorded as baseline, while freezing response during the test tones was recorded as the conditioned response. Subjects that exhibited either very low conditioned freezing to CS+ tone (<20%, n = 2) or very low locomotion throughout the test (>50%, n = 1) were excluded from statistical analysis.
During conditioning, photostimulation was presented during CS+ and, in most subjects, terminated 0.5 s before the onset of the footshock. However, in a subset of mice (ChR2: N = 5, Arch: N = 4), photostimulation overlapped with the footshock (S13A Fig). While overlapping the photostimulation with the footshock affected the freezing response in PV-ChR2 group (S13B Fig), as previously described [43]), it did not result in a significant difference in LS (S13C Fig). In PV-Arch group, we did not observe significant effect of the overlap of photostimulation with the footshock on either freezing response (S13D Fig) or LS (S13E Fig). Therefore, the two subsets of mice were combined for subsequent analysis within each group.
All recordings were carried out inside a double-walled acoustic isolation booth (Industrial Acoustics). Mice were placed in the recording chamber, and a headpost was secured to a custom base, immobilizing the head. Activity of neurons in the primary AC was recorded via a silicon multi-channel probe (Neuronexus), lowered in the area targeting AC via a stereotactic instrument following a durotomy. The electrode tips were arranged in a vertical fashion that permits recording the activity of neurons in different cortical laminae. Electro-physiological data from 32 channels were filtered between 600 and 6000 Hz (spike responses), digitized at 32kHz and stored for offline analysis (Neuralynx). Spikes belonging to single neurons were detected using commercial software (Plexon) [79].
Stimulus was delivered via a magnetic speaker (Tucker-David Technologies), calibrated with a Bruel and Kjaer microphone at the point of the subject's ear, to deliver tones at frequencies between 1 and 80 kHz to +- 3 dB [79]. To measure the frequency tuning curves, we presented a train of 50 pure tones of frequencies spaced logarithmically between 1 and 80 kHz, at 8 intensities spaced uniformly between 10 and 80 dB, each tone repeated twice in pseudo-random sequence, counter-balanced for laser presentation. The full stimulus was repeated 5 times. Each tone was 50 ms long, with inter-stimulus interval (ISI) of 450 ms. The light-Onset was presented during every other tone, with the onset of 100 ms prior to tone onset, and lasting for 250 ms.
The effect of the light-On FR was assessed by as an index of change in FR in light-On and light-Off trials:
ΔFR=FRON−FROFFFRON+FROFF
The change was computed separately for the spontaneous and tone-evoked firing rate. The spontaneous firing rate (FRbase) was computed by averaging FR over 50 ms before tone-Onset across light-On and light-Off trials. The tone-evoked firing rate (FRtone) was computed as the average of FR of responses to tones at 60–80 dB SPL at 0–50 ms after tone onset were averaged. To examine frequency selectivity of neurons, sparseness of frequency tuning was computed as:
Sparseness=1−(∑i=1i=nFRi/n)2∑i=1i=nFRi2/n
where FRi is tone-evoked response to tone at frequency i, and n is number of frequencies used.
To compute the width and BF of tuning, the frequency response function was fitted Gaussian function:
FR(f)=ae−(f−fb)22σ2
where fb is the BF and σ is the standard deviation of the Gaussian function. The tuning width was measured in octaves as the difference between fb + σ and fb − σ for neurons, for which the Gaussian fit had R2>0.4.
Magnitude of neuronal response to tones was defined as the difference between mean spontaneous (0–50 ms before tone onset) and tone-evoked (0–50 ms after tone onset) firing rate and, for each neuron, normalized by setting the peak response magnitude between 0 and 50 ms after tone onset on light-Off trials to 1. Only responses to tones within 0.5 octaves of BF of each neuron were included. To quantify correlation between neuronal responses and behavioral frequency discrimination, normalized tone-evoked response magnitude over all neurons recorded in each mouse was compared to changes in behavioral Th. Only mice with >5 identified single units (33 out of 36 mice) were used for statistical analysis.
First, we determined the criteria based on the spike waveform analysis. Putative PV interneurons in PV-ChR2 mice were preselected for waveform analysis if they exhibited a significant (more than 2-fold) increase in firing rate in response blue light (10 mW/mm2) and their spontaneous firing rate exceeded 3 Hz. Waveform analysis showed that spikes of these neurons have relatively low peak to trough amplitude ratio (<1.2, S1D Fig) consistently with previous reports [50]. In order to exclude PV interneurons from the pool of neurons used for the analysis of tone-evoked responses, only neurons with peak to trough ratio that exceeded 1.2 were used. In addition, putative excitatory cells were identified based on their expected response patterns to sounds and lack of significant activation of the spontaneous firing rate by the laser in PV-ChR2 mice and suppression in PV-Arch mice [50,80]. While this subpopulation may still contain inhibitory neurons, the proportion of interneurons recorded was relatively small, as we used silicon electrode probes with relatively low impedance that do not target interneurons [50]. The low impedance of the probes precluded us from conducting a more detailed analysis for fast-spiking versus regular-spiking neurons based on the spike waveform [50].
We constructed a model of the excitatory-inhibitory neuronal circuit based on a firing rate model, based on Wilson-Cowan dynamics [53–55]. The mean activity level of each population was modeled as:
dEdt=1τE[−E(t)+(k−r)S(jCamK2(t)+jETone(t)+Sinh(jIEI(t))]
dIdt=1τI[−I(t)+(k−r)S(jPV(t)+jITone(t)+jEIE(t))]
where E(t) is the firing rate of the excitatory population; I(t) is the firing rate of the inhibitory population; S(x) is the firing transfer function between the combined postsynaptic input and the neuronal firing rate; Sinh(x) is the transfer function between the inhibitory firing rate and excitatory postsynaptic current; jEI (0.2) and jIE (−0.2) are excitatory–inhibitory and inhibitory–excitatory synaptic weights; jETone(t) and jITone(t) are tone-evoked input currents to excitatory and inhibitory neurons, respectively, modeled as 50 ms long exponentially decaying inputs of maximum amplitude 3; τ E (10 ms) and τ I (10 ms) are synaptic time constants for excitatory and inhibitory neurons; k and r represent the maximum and minimum firing rates of neurons respectively (k = 15, r = 1); jCamK2(t) is the input to excitatory neurons due to ChR2-driven activation; jPV(t) is the input to inhibitory neurons due to either ChR2 (positive) or Arch (negative). The optogenetic modulation was modeled as a unitary pulse of 250 ms in duration. We simulated activation of inhibitory neurons by setting jPV(t) = 1, activation of excitatory neurons jCamK2(t) = 1.5, or suppression of inhibitory neurons by setting jPV(t) = −1.
The transfer functions is given by:
S(x)=(x−a)(b−a) for a < x < b; S(x) = 0 for x < a; S(x) = 1 for x > b; where for excitatory neurons, a = −2, b = 1.5; for inhibitory neurons, a = 0, b = 4.
For the inhibitory-to-excitatory inputs, we used a simplified saturating transfer function, Sinh(x), which is the quasistatic solution to a differential equation for the synaptic conductance g with depletion and replenishment given by:
dgdt=−gr/Td+(g0−g)/Tr
Here, r is the presynaptic firing rate, g is the synaptic conductance, g0 is the maximum conductance, and Td and Tr are the time constants for depletion and replenishment, respectively. The input to the post-synaptic neuron is given by the product gr. Then, Sinh(x)=gx1+cxwhere g = 2, c = 0.15.
For visualization, the firing rate of neurons was normalized as in Fig 3A, 3C and 3E, by subtracting the baseline firing rate, and setting the peak of the tone-evoked firing rate to 1 on light-off trials.
Because most of behavioral experiments consisted of within-subject repeated measurements, most of the data were analyzed by either two-tailed paired t test or repeated-measures ANOVA using SPSS Statistics (IBM) or Matlab (Mathworks). The effect of photoactivation and inactivation of neuronal activity on tuning properties was examined using a one-sample t test. Samples that did not pass Shapiro-Wilk test for normality were compared using Wilcoxon signed rank test. Multiple comparisons were adjusted by Bonferroni correction. Equality of variances was confirmed using Levene’s test.
|
10.1371/journal.pmed.1002716 | Health system costs for individual and comorbid noncommunicable diseases: An analysis of publicly funded health events from New Zealand | There is little systematic assessment of how total health expenditure is distributed across diseases and comorbidities. The objective of this study was to use statistical methods to disaggregate all publicly funded health expenditure by disease and comorbidities in order to answer three research questions: (1) What is health expenditure by disease phase for noncommunicable diseases (NCDs) in New Zealand? (2) Is the cost of having two NCDs more or less than that expected given the independent costs of each NCD? (3) How is total health spending disaggregated by NCDs across age and by sex?
We used linked data for all adult New Zealanders for publicly funded events, including hospitalisation, outpatient, pharmaceutical, laboratory testing, and primary care from 1 July 2007 to 30 June 2014. These data include 18.9 million person-years and $26.4 billion in spending (US$ 2016). We used case definition algorithms to identify if a person had any of six NCDs (cancer, cardiovascular disease [CVD], diabetes, musculoskeletal, neurological, and a chronic lung/liver/kidney [LLK] disease). Indicator variables were used to identify the presence of any of the 15 possible comorbidity pairings of these six NCDs. Regression was used to estimate excess annual health expenditure per person. Cause deletion methods were used to estimate total population expenditure by disease. A majority (59%) of health expenditure was attributable to NCDs. Expenditure due to diseases was generally highest in the year of diagnosis and year of death. A person having two diseases simultaneously generally had greater health expenditure than the expected sum of having the diseases separately, for all 15 comorbidity pairs except the CVD-cancer pair. For example, a 60–64-year-old female with none of the six NCDs had $633 per annum expenditure. If she had both CVD and chronic LLK, additional expenditure for CVD separately was $6,443/$839/$9,225 for the first year of diagnosis/prevalent years/last year of life if dying of CVD; additional expenditure for chronic LLK separately was $6,443/$1,291/$9,051; and the additional comorbidity expenditure of having both CVD and LLK was $2,456 (95% confidence interval [CI] $2,238–$2,674). The pattern was similar for males (e.g., additional comorbidity expenditure for a 60–64-year-old male with CVD and chronic LLK was $2,498 [95% CI $2,264–$2,632]). In addition to this, the excess comorbidity costs for a person with two diseases was greater at younger ages, e.g., excess expenditure for 45–49-year-old males with CVD and chronic LLK was 10 times higher than for 75–79-year-old males and six times higher for females. At the population level, 23.8% of total health expenditure was attributable to higher costs of having one of the 15 comorbidity pairs over and above the six NCDs separately; of the remaining expenditure, CVD accounted for 18.7%, followed by musculoskeletal (16.2%), neurological (14.4%), cancer (14.1%), chronic LLK disease (7.4%), and diabetes (5.5%). Major limitations included incomplete linkage to all costed events (although these were largely non-NCD events) and missing private expenditure.
The costs of having two NCDs simultaneously is typically superadditive, and more so for younger adults. Neurological and musculoskeletal diseases contributed the largest health system costs, in accord with burden of disease studies finding that they contribute large morbidity. Just as burden of disease methodology has advanced the understanding of disease burden, there is a need to create disease-based costing studies that facilitate the disaggregation of health budgets at a national level.
| Few studies have estimated disease-specific health system expenditure for many diseases simultaneously, keeping within the total envelope of health expenditure. Most cost of illness studies focus on a single or small set of disease(s) and overestimate costs due to attribution of comorbidities.
Few studies have estimated whether having two or more diseases increases or decreases health system expenditure, over and above that expected from having the diseases separately.
Using nationally linked health data for all New Zealanders, the additional or excess costs from having one or more chronic diseases were calculated using regression equations.
Nearly a quarter of all health expenditure (23.8%) on chronic noncommunicable disease was attributable to costs from having two or more diseases (i.e., costs due to comorbid conditions, over and above separate costs from single diseases).
Of the remaining three quarters of health expenditure, its attribution across six chronic diseases (as though they were the only disease a person had) were: heart disease and stroke, 18.7%; musculoskeletal, 16.2%; neurological, 14.4%; cancer, 14.1%; chronic lung, liver, or kidney disease, 7.4%; and diabetes, 5.5%.
Health system expenditure due to musculoskeletal and neurological diseases is sizeable, suggesting these diseases need more policy and planning consideration and research than they currently receive.
The additional costs from having two or more diseases, over and above the costs of having disease separately, reinforce the need for policy and planning to anticipate the effects of aging populations with comorbidity.
| The Global Burden of Disease (GBD) has achieved the vision of a coherent and comparable set of epidemiological estimates of disease-specific incidence, prevalence, morbidity, and mortality across all countries and since 1990 [1,2]. There are many reasons associated with planning and comparative analysis that suggest it would also be worthwhile to pursue a similar vision of national health system expenditure, disaggregating spending by disease in a manner comparable across countries. The Organization of Economically Developed Countries has promulgated such a vision under the rubric of national health system accounts that disaggregates total health expenditure by disease [3], and within the United States, there have been proposals for the creation of such accounts [4]. A comparable set of disease costs across multiple countries would facilitate comparisons both over time within countries and between countries. Such accounts would provide measures of disease burden that would generate a consistent set of costs to use in cost of illness (COI) and cost-effectiveness studies. Achieving such a vision, though, is challenging. Countries vary in the structuring of their health systems, and there is a need for standardisation of methods across countries. Using the same disease classification systems and standardising, or at least accommodating differing boundaries of total health expenditure, is critical [5].
A key limitation of disease-specific COI studies that focus on a single or small set of disease(s) is their tendency to overestimate costs due to attribution of comorbidities. If separate COI studies are naively summed, overestimation of total expenditure is likely [4]. COI studies including most/all diseases in a given country, constraining the total cost to the actual total envelope of health expenditure, are protected against this double-counting problem. Whilst such multiple or all-diseases COI studies are uncommon, the US has a reasonably long history of such studies since the 1960s (e.g., [6] through to more recent studies [7,8]), and examples are emerging in other countries (e.g., [5,9,10]). These studies are usually hybrids of top-down disaggregation and bottom-up estimation using a range of individual-level data sets. However, we are not aware of any studies in which individuals are uniquely identified across multiple data sets (e.g. inpatient, outpatient, pharmaceuticals, mortality, cancer registrations), allowing identification of disease diagnosis, progression, and death (and how costs are aligned to disease progression) and the identification of comorbidities within the same individual, linked to all event and cost data.
Rosen and Cutler (2009) [4,11] outline three approaches to attributing costs to diseases: (1) encounter based, (2) episode based, and (3) person based. Encounter based and episode based both require each encounter or episode to be coded to a primary disease, with associated or comorbid diseases as secondary codes. This coding is routinely undertaken for hospitalisation in many countries but is unlikely to be available for primary care encounters and pharmaceuticals (as many drugs are used for more than one disease). ‘Most cost of illness studies take an encounter-based approach, assigning claims to disease buckets based upon their coded diagnoses. Comorbidities are a major problem here; attributing each spending item for a patient who is both hypertensive and diabetic is not easy’ (p. S10 [4]). Dieleman et al. (2017) have proposed an extension to the encounter-based approach, merged with an attributable risk approach, to determine base costs for each disease and outflows and inflows from comorbid conditions [12]. However, this approach still requires primary and secondary diagnoses for each event. In the person-based approach, all of a person’s total expenditure is regressed on disease indicators and covariates: a model with no disease comorbidity interaction terms will estimate the independent and unconfounded costs of each modelled disease; a model with additional covariates for comorbidity interactions will estimate the greater (or lesser) costs of having two or more diseases simultaneously, over and above that expected for having each disease separately.
In this study, we capitalise on national-level data tracking all publicly funded noncommunicable disease (NCD) health events of uniquely identified individuals, over multiple years, in New Zealand to estimate costs by disease. New Zealand is a high-income country with a burden of NCDs similar to other high-income countries [2], although with less than average healthcare spending per capita than other high-income countries (US$3,648 in purchasing power parity–adjusted US dollars for 2015, compared with a population-weighted average of $5,551 across all high-income countries) [13]. We extend previous work in two ways. First, we attribute health expenditure (1) to mutually exclusive disease categories and (2) split into the single-disease costs (i.e., the cost of disease as though comorbidities did not exist), and the disease comorbidity interactions to capture the impact of people having more than one disease simultaneously.
The second extension is the estimation of costs by phase of disease progression. This includes estimating the costs in year of diagnosis and year of death if dying of that disease. These estimates are useful for subsequent simulation modelling of disease prevention and treatment in cost-effectiveness research, as the differential timing of costs can be better modelled. Our advances over previous national-level and multiple-disease COI studies may not be replicable in many other countries due to data limitations, but we aim to generate a greater depth of understanding to assist not only policy within New Zealand but also the generation of approaches for multicountry COI studies.
We address three specific research questions:
Publicly funded health events make up 82% of all health expenditure in New Zealand, and all such events are assigned a unique personal health identifier, enabling events to be linked across data sets and time by individual. These data have been assigned unit costs since 2007; in this study we used data spanning financial years 2007/2008 to 2013/2014 (i.e., 1 July 2007 to 30 June 2014) for person-year observations among ≥25-year-olds to model costs, but we also use data from previous years to determine if someone has prevalent disease(s) entering the study period. The following national data sets were included: the National Minimum Data Set (NMDS) for all inpatient events since 1990, the nonadmitted patient data set since 1998 (outpatients), cancer registrations since 1995, retail pharmaceuticals since 2005, laboratory claims since 2003, and general medical services claims since May 2000. Retail pharmaceuticals exclude spending in the inpatient facility and are based on the fee schedule, although the government may pay less due to price negotiations. For 97%–98% of the population, we assigned the primary care cost as given by the capitation bulk funding formula based on sex, age, ethnicity, deprivation, and high-user status, while the remaining 2%–3% of the population were not registered with a primary care provider and paid ‘fee for service’ transactions, which are captured in the data. For the included data sets, Ministry of Health cost weights were assigned to each event [14], adjusted for inflation to 2016 real dollars, and then converted to US dollars using the 2016 benchmark purchasing power parity of 1.450 from the Organization for Economic Cooperation and Development (http://stats.oecd.org/Index.aspx?DataSetCode=MEI_PRICES, accessed 20 November 2017).
Data cleaning undertaken by the authors included ensuring same sex, date of birth, and date of death; when an individual had multiple records and there was a disagreement, we used data from the most authoritative source (e.g., mortality record if date of death) or most commonly given (e.g., female sex if on all but one record for a given individual).
For this study, we used two distinct disease groupings. The first set, the aggregated disease set, included the first five of eight disease groupings in the 1990–2013 New Zealand Burden of Disease Study (NZBDS) [15], namely cardiovascular disease (CVD) and diabetes (which we divided into CVD and diabetes separately), cancer, chronic lung/liver/kidney (LLK) disease, neuropsychiatric conditions, and musculoskeletal conditions. The coding and capture of mental events was not yet reliable in the study window, so we excluded psychiatric conditions and had a neurological-only category. The three NZBDS groupings we did not model were as follows: other NCDs; maternal, neonatal, nutritional deficiency, and infectious disorders; and injury. The second set of diseases used in this study were 13 disaggregated diseases, including lung, breast, prostate, colorectal, and other cancers; ischaemic heart disease (IHD), stroke, and other CVD; and chronic lung, chronic liver, and chronic kidney disease.
To determine if any of these diseases were prevalent before or were incident during the 1 July 2007 to 30 June 2014 costing window, a thorough case-finding algorithm was applied, consistent with that used for the NZBDS (S1 Table). International Classification of Disease (ICD) codes (primary and other) for events and disease-specific drug and laboratory testing combinations were developed, disease by disease. For cancers, survival after diagnosis by 5 years for lung, 8 years for colorectal, 10 years for ‘other’ cancers, and 20 years for breast and prostate resulted in that person being recoded as free of that cancer, based on statistical cure times [16]. For all other disease, no remission was allowed (i.e., diagnosis was for the remainder of his or her life). Each disease was coded by phase as not present (reference category), diagnosed in that financial year, died in that financial year of that disease, and otherwise prevalent. Note, therefore, the costs for the first two categories are for people with an average of 6 months in that state (but, for the diagnosis category, still including the time and costs for events preceding the diagnosis date in the same financial year).
We did not attempt to classify what healthcare spending was or was not related to each specific disease. Rather, we used the established ‘excess’ or ‘net’ cost approach [17–20], using a statistical approach based on regression models, with total health system cost for each individual in each financial year as the dependent variable and demographics, dummy variables for calendar year, and disease and disease phase indicators as the independent variables. The coefficients for the disease indicators are therefore estimates of the excess cost of having that disease phase and are independent of other diseases (and comorbidities) in the model. The intercept term is the health system cost not attributed to the diseases in the model (e.g., due to injuries, preventive care, mental illness, maternity, etc.).
To model nonadditive costs associated with comorbidity, we interacted the disease indicators in order to identify when an individual had both diseases. It was not possible to include all possible combinations of disease interactions in regression models due to the large number of interaction terms. Thus, we retained for regression modelling the disease phase indicators for each disease as ‘main effects’ but only included all 15 comorbidity pair interactions formed by six diseases (i.e., 6!/([6 − 2]! × 2!) = 15). For example, the cancer–CVD pair was coded ‘1’ for a person-year observation for which the person had a diagnosis of both diseases, regardless of phase of diagnosis (i.e., prevalent cancer with prevalent CVD was coded the same as first year of diagnosis of cancer with prevalent CVD). In regression modelling, the disease phase indicators and disease comorbidity pairs were all interacted with age and age squared, first centring age at the midpoint of the 60–64-year-old age group, 62.5 years, and dividing by 10, so the age interaction terms are interpretable in per 10-year units. Regression equations are detailed in S1 Appendix.
To prepare the data for regression modelling, we aggregated them into unique strata formed by cross-classifying 5-year age groups by financial year, by the four levels of each disease (no diagnosis, diagnosed in that financial year, dying of that disease in that financial year, otherwise prevalent with that disease). For the disaggregated 13 disease classifications, there were 151,913 unique strata of these cross-classified categorical variables with at least one male observation, and 139,717 unique strata for females. Between-person ordinary least squares (OLS) regressions were run on these aggregated data sets, with weights equal to the number of observations in each stratum. The advantages of this approach included (1) aggregation reducing the skewness of data, (2) OLS coefficients were directly interpretable as excess costs, (3) the total predicted expenditure (when applying coefficients back onto observations and summing regression-predicted costs across all individuals) always gave the exact observed total expenditure in the data sets, and (4) model-run time was considerably reduced compared with analyses on unit-level data. We also trialled gamma regression on unit-level data, which had better residual plots than OLS but overpredicted disease and total costs, causing us to prefer the above OLS on aggregate data approach. Both because zero costs were so rare (0.09% of all person-year observations) and previous work has found that for estimating averages, a one-part model performs (nearly) as well as a two-part regression [21], we did not employ a two-step process of first estimating individuals with zero expenditure.
Four sets of models were run: disease progression models (models with all the disease-specific phases) for either the aggregated (n = 6) or disaggregated (n = 13) disease groupings, but without comorbidity interactions; and both these models with the additional 15 disease comorbidity pairs. Note that we retain the same 15 disease comorbidity pairs in the disaggregated 13-disease model, as it was impractical to model and interpret the 78 possible comorbidity pairs formed by 13 diseases.
As part of model fitting, we wanted to ensure that the intercept terms in the model and regression-predicted costs by age using the models agreed to the annual expenditure in the data for individuals with none of the disease modelled. There was good correspondence of these predictions and those observed in the data for models including the disease comorbidity pairs; however, predictions for models not including the comorbidity pairs underestimated the costs among the nondiseased at younger ages (both OLS and gamma regressions) due to strong disease–disease–age interactions. Therefore, for models without comorbidity pair interactions, we rescaled all regression coefficients. First, we used the unscaled regression equations to predict costs among those with any disease and compared this with their actual total cost in the data set. We then used age-specific scalars for the disease coefficients in the model to return the correct total cost of people with diseases. For example, if the sum of OLS regression predicted that costs of 50–54-year-olds with at least one disease was $200 million, but the actual observed total cost for these same people was $180 million, then we rescaled all disease coefficients by 180/200 = 0.9 for this age group. Second, the reciprocal procedure was performed for people without disease. Notably, no such scaling was needed for the better-fitted models with the comorbidity pairs and their interactions, with age included.
Given age interactions, we present absolute estimates of disease expenditure for 60–64-year-olds as the main results, with young (45–49 years) and old (75–79 years) estimates presented additionally in S4 Table and S5 Table. Actual regression coefficients are presented in S3 Table, allowing interested readers to calculate excess costs for alternative groupings.
We present cause-deleted estimates, the estimates of spending without the presence of comorbidities, by setting disease coefficients in the predictive equation to zero.
We undertook several robustness checks. First, for simpler models (six-disease, no disease comorbidity pairs), an OLS general estimating equation (GEE) regression model on individual-level data was undertaken to allow for intraclass correlation due to up to 7 years of observation for each individual. Second, for simpler models we also ran (non-GEE) OLS regressions on unit-level data. Third, all the above models use between-person comparisons, which may be residually confounded by unobserved individual characteristics. Therefore, we also ran fixed effects models that capitalise on within-individual changes in disease status to estimate disease costs, removing potential time-invariant confounding [22].
There were 18.9 million person-year observations for those 25 years and older between 1 July 2007 and 30 June 2014. A total of 7.1 million (37.7%) of these observations included at least one NCD (Table 1). There was a total allocated expenditure across the 7 years of US$26.4 billion, of which $20.3 billion (77.0%) occurred among observations with at least one disease. Regarding source, half or $13.1 billion arose from inpatient care and about a fifth from each of outpatients and community pharmaceuticals.
Table 2 shows the observations and total expenditure by disease phase. People with neurological conditions, musculoskeletal conditions, CVD, and diabetes had the greatest number of observations at 16.1%, 15.7%, 11.3%, and 7.5%, respectively, mostly as prevalent observations (i.e., neither diagnosed nor dying in that observation year). People with CVD had the highest total expenditure (43.1%), although this includes spending on additional diseases.
Table 3 shows the modelled health system expenditure per person-year for 60–64-year-olds, attributed to each disease phase, for both the aggregated and disaggregated-disease models, by sex, without and with disease comorbidity pairs. The first row of the table results (‘No NCD’) is the estimated expenditure for a 60–64-year-old with no NCDs; the next panel is the main effects by disease phase for each of the 6 or 13 diseases; and the last panel is the expenditure per year for the 15 comorbidity pairings, over and above that attributed to the diseases independently.
When considering the aggregated-disease model for females, without comorbidity pairs (FA model in Table 3), a 60–64-year-old without any of the six diseases was estimated to cost the health system US$626 per year. If she had a cancer diagnosed in that year, her costs were predicted to increase by $9,747 over and above the $626 base cost, by $9,815 if dying of cancer in that year, or by $1,716 if living with a prevalent cancer. If she also had prevalent CVD, her costs were predicted to increase a further $1,439.
Fig 1 shows these independent excess costs for both male and female 60–64-year-olds, for the disaggregated-disease models (i.e., estimates for the MD and FD model columns of Table 1). Excess costs in the year of diagnosis and year of death from that disease were higher than prevalent costs. However, patterns differed by disease. For example, cancer and CVD excess costs in the year of diagnosis and last year of life were roughly similar, but diabetes costs if dying of diabetes ($13,798 and $15,674, Table 1) were much higher than costs in the diagnosis year. Last year of life costs for chronic kidney disease were also notably high.
Excess costs for the disease comorbidity pairs were positive, with 95% confidence interval (CI) excluding the null for all pairs across all models, except for cancer with each of CVD, diabetes, and LLK disease (bottom panel of Table 3). That is, the costs of having diseases jointly was usually greater than the estimated summed costs for having the diseases independently (shown in the rest of Table 3). For example, using the male disaggregated-disease model (model MDC in Table 3), a 60–64-year-old with both CVD and LLK disease was predicted to cost $2,050 (95% CI $1,837–$2,263) more than that predicted based on the independent excess costs due to CVD and LLK independently. Only 1 of the 60 comorbidity pairs (15 possible pairs across four models) had a negative cost with the 95% CI excluding the null, namely the cancer-CVD pair in the male disaggregated-disease model (−$484, 95% CI −$784 to −$240). Whilst this in isolation may be a chance finding given the 60 pairs tested, this same cancer-CVD pair had a negative cost for the male aggregated-disease model and female aggregated and disaggregated models—but with 95% CI including the null.
There were significant interactions of age and age squared with most separate diseases and disease comorbidity interactions (S3 Table). Because of these age interactions, excess costs vary by age (S4 Table and S5 Table give estimates for 45–49- and 75–79-year-olds, respectively). Two patterns were evident for independent disease costs: excess costs in the last year of life were less with increasing age, and musculoskeletal and neurological excess disease costs per person in the first year of diagnosis and prevalent cases were greater for 75–79-year-olds than 45–49-year-olds. Regarding the disease comorbidity pairs, there was a strong pattern of comorbidity costs being greater for 45–49-year-olds than 75–79-year-olds—often considerably so (Fig 2). For example, excess expenditure for 45–49-year-old males with CVD and chronic LLK was 10 times higher than for 75–79-year-olds, and six times higher for females. High excess costs for comorbidity are notable for cancer and neurological, and CVD and LLK (especially at younger ages).
Total expenditure by disease costs is a function of the above per-person costs and how common the disease phases actually were (Table 2). Across the 7 years in the study, using the models without comorbidity interactions, the greatest excess expenditures were for neurological (22.3%), CVD (21.2%), and musculoskeletal (20.8%) disease (Table 4). Fig 3 shows the attribution of this disease expenditure by sex and age. A greater area under the curve for females is attributed to nonmodelled diseases at younger ages, consistent with maternity care not being included as an explicit category in the modelling. Noteworthy patterns include that neurological and musculoskeletal costs were substantial at all ages, with neurological particularly notable at younger ages for females; diabetes costs and female breast cancer costs were skewed more to younger ages; and other CVD and IHD costs were considerably greater than stroke costs at all ages.
Using the models with the 15 comorbidity pairs, the independent disease costs reduced by 23.8% due to costs now being attributed to one of the 15 possible comorbidity pairs (Table 4). Of this 23.8% comorbidity expenditure, the five comorbidity pairs including a neurological condition contributed 60.7%, and the five pairs including musculoskeletal contributed 41.4%. The comorbidity pair with the greatest contribution was neurological plus musculoskeletal (18.1%), meaning the 9 out of 15 comorbidity pairs including either/both neurological and musculoskeletal conditions explained 84.0% of additional health expenditure due to comorbidities at the population level (i.e., 60.7% + 41.4% − 18.1%; see S1 Fig for a breakdown of percentage comorbidity expenditure for each of the 15 comorbidity pairs).
Of the 76.2% of expenditure not captured by comorbidity pairs, CVD accounted for 18.7%, followed by musculoskeletal (16.2%), neurological (14.4%), cancer (14.1%), chronic LLK disease (7.4%), and diabetes (5.5%; Table 4). The expenditure attributed to these six diseases independently, when including comorbidity pairs relative to the model excluding comorbidity pairs, varied by disease, ranging from 0.60 for diabetes (i.e., 40% of apparent independent expenditure for diabetes can actually be attributed to superadditive costs arising from diabetes coexisting with other diagnoses) to 0.92 for cancer (i.e., only 8% of apparent independent expenditure on cancer can be attributed to comorbidity additive effects; last column of Table 4).
The coefficients across the preferred aggregated data OLS models and the unit-level OLS and GEE unit-level OLS regressions were identical (S6 Table). The standard errors were smallest for unit-level OLS but more similar for aggregate OLS and GEE unit level. However, given the size of the data sets and the usually sizeable health expenditures for diseases, most disease excess costs are statistically significant, regardless. Fixed effects analyses that use within-person variation (i.e., differences in expenditure between years, with and without disease, or with varying disease phase) were in reasonable agreement with our preferred OLS regressions on aggregate data (S2 Fig), given conceptual differences between the approaches and data limitations (see S1 Appendix).
This study estimates health system expenditure for many NCDs, by disease phase, for New Zealand. We capitalised on linked health data with unique identifiers for the entire population that track medical costs over multiple years. We used a regression approach to estimate the excess cost to the health system from having a given disease. At the population level, we found that 59% of all allocated health system expenditure was attributable to the included NCDs. Attributing costs to separate disease groupings and not yet allowing for comorbidities, neurological diseases accounted for 22.3% of NCD expenditure, followed by CVD (21.2%), musculoskeletal (20.8%), cancer (15.1%), chronic LLK disease (11.5%), and diabetes (9.0%). We were surprised by large costs due to neurological and musculoskeletal diseases; therefore, we plotted in Fig 4 years of life lived with disability in New Zealand by sex, age, and disease, from GBD data using the same disease categorisations for New Zealand, paralleling our Fig 3 for costs. Other studies have found a correlation of health expenditure with years of life with disability (YLDs) [9,10], although the correlation is even stronger with disability-adjusted life years. The New Zealand neurological and musculoskeletal morbidity burdens stand out even more for YLDs (Fig 4) than for costs (Fig 3), offsetting possible concerns that our costings for neurological and musculoskeletal disease might be too high.
How does this distribution of costs, ignoring comorbidities for now, compare with previous studies? There is a body of work in the US using methods that attributed events to diseases—not an excess costing approach using regression. For example, Dieleman and colleagues (2016) attributed 16 years of health expenditure using multiple data sources in the US and found that CVD contributed 11% of all health expenditure (12.5% in our New Zealand study, using the total health expenditure as the denominator in Table 4), cancers 5.5% (New Zealand = 8.9%), COPD 2.6% (4.9%), neurological 4.8% (13.2%), musculoskeletal 8.7% (12.3%), and diabetes 4.8% (5.3%) [8]. In another comparable study using a GBD grouping of diseases for Switzerland, Wieser and colleagues (2018) found that cancer contributed 6.0%, CVD 15.6%, diabetes 1.5%, neurological (including dementia) 3.7%, and musculoskeletal 13.4% [9]. Musculoskeletal expenditure in New Zealand, at 12.3%, is intermediary between these US and Swiss percentage estimates. Of note, in GBD 2016, New Zealand also had 1.31 times higher morbidity burden for back pain and 1.84 times higher ‘other’ musculoskeletal than expected based on its level of sociodemographic development [23]. This leaves the neurological category as the notable ‘stand-out’ high cost for New Zealand, perhaps due to both the range of disease and conditions included (e.g., epilepsy, dementia, migraine) and the inclusive capture of people with at least one neurological diagnosis (e.g., through the range of data sets and classifications; S1 Table).
Where our New Zealand study steps ahead of similar general COI studies is in further disaggregating disease costs by phase and comorbidities, an advance that is possible due to all health data sets being linked at the individual level. Disease costs were generally higher in the year of diagnosis and the year of death if dying of that disease (Fig 1), consistent with our previous work on multiple cancers [24]. Costs in the year of death if dying from chronic kidney disease, musculoskeletal, or diabetes were particularly prominent, ranging from $13,000 to $30,000 (Fig 1). Dying-related costs from kidney disease, if also undergoing (failed) transplant or renal dialysis, are high. Dying of diabetes per se, whilst uncommon compared with dying of diseases elevated in risk by diabetes, is also expectedly expensive.
Regarding comorbidities, we find that the majority of comorbidity pairs increase costs more than that expected based on having the same diseases separately (Table 3, Fig 1), and this superadditive comorbidity cost was far more prominent at younger ages (Fig 2). Whilst we have no direct evidence, it seems plausible that young people with comorbidities might be treated more aggressively than old people with comorbidities or perhaps that young people with comorbidities have more severe disease than their younger counterparts with just one disease. Further research to understand this this age variation in comorbidity expenditure is warranted.
We estimated that 23.8% of all NCD health expenditure is attributable to these additional increments of comorbidity, over and above the independent costs of having diseases separately. Most previous COI studies analysing the impact of comorbidities have been from the perspective of one disease only, and its increased costs due to comorbid conditions (e.g., [25–27]). An exception is Dieleman and colleagues (2017), who used encounter-level data with multiple ICD codes recorded for each event, to quantify ‘inflows’ and ‘outflows’ between all comorbidities, to reallocate health spending mutually exclusively [12]. This is a different approach to what we took using a person-based excess costing approach to directly examine disease–disease interactions based on diagnoses stated on any event linked for the same person, and quantify the independent contributions of separate diseases and comorbidity pairs to total health expenditure. We found that the majority of the 15 comorbidity pairs increased expenditure over and above having the diseases separately. There was, however, evidence that having cancer and CVD simultaneously reduced health expenditure compared to that expected from having the two diseases separately. It is possible that a diagnosis of cancer means that treatments for other diseases are down-prioritised, but this is speculation on our part.
There are limitations with our study. First, cost weights assigned to events may not have been the exact price to hospitals or other fund managers. Second, whilst New Zealand has high-quality linked heath data, it does not capture all costs. Private expenditure is estimated at 18% of all health expenditure; future research could attempt to include this but will require additional data and assumptions about how this funding is distributed compared with public funding. For example, the great majority of NCDs are treated in public hospitals, but breast cancer treatment and hip/knee replacement surgery are more commonly provided in the private sector. There are also missing costs due to support care services (e.g., rest home care of people with dementia). We were unable to capture actual primary care usage by individuals on national databases; rather, we just imputed an ‘expected’ cost based on the bulk funding formula, and we will thus have underestimated disease costs somewhat, as people with the diseases we captured will (presumably) be higher users of primary care. We also do not include prevention costs. Nevertheless, we capture the majority of costs, and certainly enough costs to confidently speak to patterns of cost by sex, age, disease, disease phase, and comorbidity pairs. Future research should aim to improve these data, however.
Third, whilst we see the excess costing approach using regression modelling as a strength overall, and it avoids the complexities of attributing each individual medical event to a disease, it has limitations. People with diseases may be higher users of health services, regardless of their disease—a form of time-invariant confounding. In fixed effects analyses exploiting changes in costs within individuals (see S1 Appendix), we found broadly similar costs, suggesting that time-invariant confounding is not too problematic. There may also be time-varying confounding, whereby people with a disease increase health services use for other reasons, including increasing surveillance and preventive activities when engaged with the health system. Fourth, addition of further comorbidity combinations (e.g., with mental illness) would likely increase the proportion of all NCD health expenditure attributable to the superadditive effect of comorbidities.
So what for researchers and policy makers? There is a surprising lack of disease-attributed costing studies involving multiple diseases at once. Governments and health systems managers and funders can improve planning and prioritisation, knowing where the money goes. Also, cost-effectiveness studies usually need costs by disease to model cost offsets from preventing disease (presentations) and, conversely, the health systems costs from living longer in evaluations using an unrelated disease costing approach [28,29]. Our analytical framework generates these costs by disease phase for such cost-effectiveness modelling (e.g., [30]). We believe the costing methods used in this paper can be applied elsewhere. For example, the OECD has as a goal that national health expenditure be disaggregated by disease; this will require standardisation as best as possible of data and methods across countries, but given data variations across countries, exact standardisation is unlikely. We therefore propose that an approach similar to that used for estimating epidemiological parameters in the GBD may be useful [2,31]. Here, one would build up from COI studies in multiple countries (e.g., [5]) and use country-level parameters such as GPD, health system configuration, and such to build a model predicting disease costs within an envelope of total health expenditure given by a system of national health accounts. Our paper and methods may help in that the regression models built within a given (or few) country predict the relative expenditure by disease—which could be merged with the population demography, disease epidemiology, and total health expenditure of another country to at least provide an initial costing.
Finally, any national costing study will be dependent on the data available and methods used. That said, we suspect that the general patterns of these results (i.e., of comorbidity impacts being superadditive) for New Zealand are likely to be generalisable to other high-income countries. The methods we applied in this paper could be applied elsewhere to test for such generalisability—or not—of comorbid costs.
|
10.1371/journal.pntd.0005832 | Larrea tridentata: A novel source for anti-parasitic agents active against Entamoeba histolytica, Giardia lamblia and Naegleria fowleri | Protozoan parasites infect and kill millions of people worldwide every year, particularly in developing countries where access to clean fresh water is limited. Among the most common are intestinal parasites, including Giardia lamblia and Entamoeba histolytica. These parasites wreak havoc on the epithelium lining the small intestines (G. lamblia) and colon (E. histolytica) causing giardiasis and amebiasis, respectively. In addition, there are less common but far more deadly pathogens such as Naegleria fowleri that thrive in warm waters and infect the central nervous systems of their victims via the nasal passages. Despite their prevalence and associated high mortality rates, there remains an unmet need to identify more effective therapeutics for people infected with these opportunistic parasites. To address this unmet need, we have surveyed plants and traditional herbal medicines known throughout the world to identify novel antiparasitic agents with activity against G. lamblia, E. histolytica, and N. fowleri. Herein, we report Larrea tridentata, known as creosote bush, as a novel source for secondary metabolites that display antiparasitic activity against all three pathogens. This report also characterizes the lignan compound classes, nordihydroguairetic acid and demethoxyisoguaiacin, as novel antiparasitic lead agents to further develop more effective drug therapy options for millions of people worldwide.
| Entamoeba histolytica, Giardia lamblia, and Naegleria fowleri pathogens are widespread throughout the world infecting and killing hundreds of thousands of people every year. They are also listed as category B bioterrorism agents by the NIH and the CDC. However, there is a serious unmet need to develop more effective therapies to treat these deadly pathogens. Herein we describe that lignans isolated from the creosote bush, common to the southwestern U.S.A. and throughout Mexico, display relatively potent antiparasitic activity against E. histolytica, G. lamblia, and N. fowleri.
| Intestinal protozoan parasite infections, through contaminated water and food supplies, are global health problems affecting hundreds of millions of people annually. The two most common intestinal parasites are Giardia lamblia and Entamoeba histolytica, which can lead to giardiasis or invasive amebiasis, respectively. G. lamblia and E. histolytica have simple infection life cycles that begin with ingesting viable cysts, excystation, followed by trophozoite multiplication in the small intestine or trophozoite migration and invasion in the colon (Fig 1A) [1–3]. Annually, giardiasis, has an estimated worldwide prevalence of 200 million cases [4], and according to the World Health Organization (WHO) giardia infections contribute substantially to the 846,000 deaths annually from diarrheal disease [5, 6]. Once G. lamblia has excysted in the small intestines, trophozoites attach to epithelial cells and elicit aberrant signaling events that disrupt organ function including the induction of programed cell death or apoptosis [3]. Although less prevalent than G. lamblia, E. histolytica infections lead to 50 million cases of invasive disease and up to 100,000 deaths, annually [7]. Invasive amebiasis is characterized by profound intestinal tissue damage and ulceration [8]. Recently, Ralston and colleagues determined trogocytosis as the mechanism by which E histolytica feeds on its host. The term ‘trogocytosis’ was taken from the Greek word trogo which means to nibble [8, 9]. The amebae damage and consume the intestinal mucosa epithelium by nibbling away at epithelial cell membranes, triggering cell death. Interestingly, Ralston et al. concluded that amebae feed on bacteria in the gut for nutrition but that host cell ingestion is done by the amebae to create a more spacious environment [8].
Free-living ameba Naegleria fowleri has been described as the cause of primary amebic meningoencephalitis (PAM) in more than 16 countries [10]. Until 2012, about 310 cases have been reported globally with a fatality rate of more than 95% [11]. According to the Centers for Disease Control and Prevention (CDC), 138 cases of PAM have been reported in the U.S.A. from 1962–2015 with a 98% mortality rate. PAM results from water containing N. fowleri entering the nasal cavity followed by migration of the amebae to the brain (Fig 1B) [12–17]. Within the brain, N. fowleri causes extensive inflammation, hemorrhage, and necrosis. The time from initial exposure to onset of illness is usually 5–7 days but may be as early as 24 h, leading to death in 3 to 7 days [18].
Treatment for giardiasis and invasive amebiasis is largely limited to the nitroimidazole drug class (e.g. metronidazole) [19]. Metronidazole, is the primary drug of choice, which requires a relatively long treatment time and high dosage to eradicate intestinal parasite infections [20]. Moreover, metronidazole is both mutagenic and carcinogenic and its use presents other significant adverse effects [21, 22]. In addition, G. lamblia and E. histolytica drug resistance and treatment failures remain an increasing problem [23–26]. Amphotericin B remains a cornerstone of therapy for PAM but is not FDA-approved for this indication and has had limited success despite its worldwide use [27]. Treatment with amphotericin B requires a high dosage and its use is frequently associated with renal toxicity and anemia, among other adverse effects [27]. Recently, an investigational drug, miltefosine, clinically used to treat leishmaniasis, has shown some promise in combination with other drugs as a treatment for PAM [28]. The CDC, through an established protocol with the FDA, is now directly providing miltefosine to the clinicians as a treatment option for PAM. However, it is still not FDA approved and has limited availability in the U.S.A. Furthermore, G. lamblia, E. histolytica and N. fowleri are listed by the United States National Institutes of Health and the Centers for Disease Control as a category B biodefense/bioterrorism pathogens due to their low infectious dose and potential for dissemination through compromised food and water supplies. Given the prevalence and mortality caused by these protozoan pathogens, compounded by their potential bioterrorism threat, more effective antiparasitic agents is a critical unmet need to treat the current pandemic and avert future outbreaks and deaths.
Natural products have played an important role throughout history in the treatment of human disease through traditional medicines and as a source for effective pharmaceutical development [29, 30]. In particular, plants have been a vast source of secondary metabolites that display potent antiparasitic activity, including protozoan parasites [30–33]. For example, G. lamblia and E. histolytica are endemic to Mexico and infections are prevalent [34, 35]. Moreover, nitroimidazole drugs display limited efficacy in the Mexican population [36]. Therefore, scientists have turned to native plants used as Mexican traditional medicines for intestinal diseases in the search for novel more effective antiparasitic agents [37, 38]. Similarly, using our established assays [39, 40], we have surveyed plants used as traditional medicines from around the world and that are common to the southwestern United States and throughout Mexico. Herein, we report the discovery of Larrea tridentata, commonly known as creosote bush or chaparral, as a novel source for antiparasitic secondary metabolites [41]. Though the extract of L. tridentata earlier showed antiparasitic activity against Trypanosoma brucei rhodesiense, T. cruzi, Leishmania donovani and Plasmodium falciparum [42], this is the first report to show their activity against a free-living amoeba N. fowleri and against diarrhea causing parasites E. histolytica and G. lamblia. We have identified seven known compounds 1–7 (Fig 2) with 1–6 displaying antiparasitic activity against E. histolytica, G. lamblia, and N. fowleri. Compounds 1 and 2 showed better activity against N. fowleri than the current drug miltefosine. In addition, we have identified two secondary metabolites, compounds 8 and 9 (Fig 2), that we isolated from the same active fractions as 1–7 that appeared to have novel structures. Compound 9 displayed modest antiparasitic activity against G. lamblia and N. fowleri. An examination of the literature indicated that 8 and 9 structures have been reported [43, 44]. Interestingly, compound 8 has not previously been isolated or structurally characterized from the creosote plant, rather, Cho and colleagues used Larreatricin 3’-hydroxylase enzyme purified from creosote and the known secondary metabolite from creosote, larreatricin, to enzymatically prepare 8, albeit in very low yield [44]. However, the structure of 9 was dubiously deduced from Graziela mollisima as an impure mixture with insufficient analytical data to accurately characterize the structure [43]. Therefore, this is the first report to unambiguously characterize the novel secondary metabolites 8 and 9 from L. tridentata. Since compounds 1 and 2 were found more active against N. fowleri than miltefosine, we selected these two compounds to investigate their ability to inhibit N. fowleri cysteine protease, an enzyme shown to play an important role in host tissue invasion by N. fowleri [45].
1H, 13C and 2D NMR spectra were recorded on a Bruker Avance III spectrometer (400 MHz for 1H NMR and 100 MHz 13C NMR). Chemical shifts are recorded in ppm (δ) using residual solvent signal as internal reference, and coupling constants (J) are reported in Hz. The following splitting abbreviations were used for NMR signals: s = singlet, d = doublet, t = triplet, q = quartet, m = multiplet, br = broad. High-resolution mass spectra (HRMS) were recorded on a Bruker Q-TOF-2 Micromass spectrometer equipped with lock spray, using ESI with methanol as the carrier solvent. Accurate mass measurements were performed using leucine enkephalin as a lock mass and the data were processed using MassLynx 4.1. Exact m/z values are reported in Daltons. Optical rotations were measured in CH3OH on a JASCO P1010 polarimeter at 589 nm (Na D-line) with a path length of 1 dm and are reported with implied units of 10−1 deg cm2 g-1. Concentrations (c) are given in g/100 mL. UV was measured in CH3OH on an Agilent 8453 UV-Visible Spectrophotometer. Analytical and preparative HPLC were performed on a Shimadzu Prominence HPLC system equipped with LC-6AD pumps, an autosampler (SIL-20AC) and manual injection port (Rheodyne, 3725i), a column oven (CTO-20A, temperature set at 27°C), a photo diode array detector (SPD-M20A, using a Deuterium lamp and a tungsten lamp as light sources) and a system controller (CBM-20A). A Phenomenex Kinetex C18 reversed phase column (5 μm, 100 Å, 250 ✕ 4.6 mm) fitted with a guard cartridge, with a flow rate of 0.7 mL/min was used for analytical chromatography, and a Phenomenex Kinetex C18 reversed phase column (5 μm, 100 Å, 250 ✕ 21.1 mm) fitted with a guard cartridge with a flow rate of 5.0 mL/min was used for preparative chromatography. The HPLC data were processed using LabSolutions Lite software (version 5.22).
The dried powdered material (11.0 g) of L. tridentata (Mountain Rose Herbs) was extracted with methanol at room temperature for 72 h. After filtration through Celite, the methanol extract was concentrated under reduced pressure to give a crude residue (2.55 g). The extract residue (2.53 g) was treated with water (150 mL) and partitioned against hexane (150 mL × 3), ethyl acetate (150 mL × 3) and n-butanol (150 mL × 2) successively to yield a hexane fraction (128.6 mg), an ethyl acetate fraction (1.5 g), a n-butanol fraction (411.7 mg), and a water fraction (504.2 mg), respectively. The parasite active ethyl acetate fraction (808.5 mg) was then chromatographed on a Sephadex LH-20 column eluted with 20% hexane in CH2Cl2 (200 mL), 60% CH2Cl2 in acetone (400 mL), 20% CH2Cl2 in acetone (200 mL), 20% CH2Cl2 in methanol (200 mL), and 100% methanol (200 mL). Ten fractions were collected: fractions A (12.9 mg) and B (13.4 mg) from 20% hexane in CH2Cl2; fractions C (28.6 mg), D (386.1 mg), E (181.9 mg), and F (81.6 mg) from 60% CH2Cl2 in acetone; fractions G (48.7 mg) and H (35.9 mg) from 20% CH2Cl2 in acetone; fraction I (52.6 mg) from 20% CH2Cl2 in methanol and fraction J (2.1 mg) from 100% methanol. Fraction E (138.9 mg) was chromatographed on reverse phase preparative HPLC and eluted with gradient 20–100% acetonitrile in water for 40 min to yield 1 (11.7 mg) and 3 (43.3 mg) as yellowish resinous solid along with sub-fraction E1 (11.0 mg). Sub-fraction E1 was re-chromatographed under similar HPLC conditions to afford 2 (4.0 mg), 7 (3.3 mg), and 8 (1.8 mg). Fraction D (386.1 mg) was chromatographed on silica gel column (13.0 g) eluted with increasing amounts of methanol in CH2Cl2 to afford seven fractions, D1 (0.6 mg), D2 (181.1 mg), D3 (73.5 mg), D4 (65.6 mg), D5 (5.6 mg), D6 (5.5 mg), D7 (7.5 mg). Fraction D2 (133.0 mg) was chromatographed on preparative HPLC and eluted with isocratic 50% acetonitrile in water to yield 2 (29.7 mg), 4 (14.6 mg), 5 (2.0 mg) and 6 (4.5 mg) as yellow resinous solids.
(7R, 7’R)-7, 7’-bis(4’, 3, 4-trihydroxyphenyl)-(8R, 8’S)-8, 8’-dimethyltetrahydrofuran (8): colorless oil; [α]D25–88.1 (c 0.16, CH3OH); UV (MeOH) λmax (log ε) 211 (3.44); 236 (2.54), 282 (1.64); 1H and 13C NMR data, see Table 1; HRESIMS m/z 301.1506 [M + H]+ (calcd for C18H21O4, 301.1439)
3-Methoxy-6, 7, 4’-trihydroxyflavonol (9): Yellow solid; UV (MeOH) λmax (log ε) 211 (5.06), 266 (4.90), 348 (4.86); 1H and 13C NMR data, see Table 2; HRESIMS m/z 301. 0690 [M + H]+ (calcd for C16H13O6, 301.0712).
Trophozoites of E. histolytica HM1: IMSS and G. lamblia WB strains were axenically maintained in TYI-S-33 medium supplemented with penicillin (100 U/ml), streptomycin (100 μg/ml) [46, 47]. Trophozoites of N. fowleri strain KUL were axenically cultured in Nelson’s medium supplemented with 10% FBS at 37°C [45]. All experiments were performed using trophozoites harvested during the logarithmic phase of growth. Four solvent partitioned fractions of an aqueous methanolic extract of L. tridentata and compounds 1–9 were screened for activity against E. histolytica, G. lamblia, and N. fowleri. For primary screening, the positive control for E. histolytica and G. lamblia was 5 μg/mL of metronidazole (Sigma-Aldrich) and 46 μg/mL of amphotericin B for N. fowleri (Sigma-Aldrich). Test samples were diluted to 10 mg/mL of extracts, HPLC fractions, and pure compounds in DMSO. Finally, 0.5 μL of diluted sample was transferred to white, solid bottom tissue culture 96-well plates (E&K Scientific) followed by addition of 99.5 μL trophozoites (5,000 E. histolytica and G. lamblia, and 10,000 N. fowleri) in TYI-S-33 medium or Nelson’s medium. The final concentration for test compounds was 50 μg/mL and 0.5% DMSO, which was the negative control and compound vehicle that we have shown has no effect the growth rate of trophozoites [39, 40, 48]. Assay plates were incubated for 48 h at 37°C. E. histolytica and G. lamblia plates were kept in the GasPak EZ Anaerobe Gas Generating Pouch System (VWR) to maintain anaerobic condition throughout the incubation period. Screening was performed in duplicate using the CellTiter-Glo assay (Promega) and luminescence was measured using an EnVision plate reader (PerkinElmer) [40, 48].
The antiparasitic activity of 1–6 and 9 were confirmed by EC50 dose response experiments, using the CellTiter-Glo assay, conducted in triplicate over a concentration range from 5-to-700 μM against trophozoites (Table 3). Miltefosine and metronidazole, current drugs for the treatment of PAM and amebiasis and giardiasis were also tested in triplicate as positive controls for EC50 determination (Table 3). Dose response curves including standard deviation (SD) calculation were processed using GraphPad Prism software 5.0. Percent inhibition relative to maximum and minimum reference signal controls was calculated using the formula: % Inhibition = [(mean of Maximum Signal Reference Control—Experimental Value)/(mean of Maximum Signal Reference Control—mean of Minimum Signal Reference Control)] × 100.
The HUVEC-TERT2 cell line was purchased from Evercyte GmbH (Vienna, Austria) and cultured and maintained in endothelial cell basal medium (Lonza) as described previously [49, 50]. Briefly, cells were seeded into a white 384-well solid bottom plate (Nunc, ThermoFisher) at a density of 1000 cells/well in 39 μL of media using a Janus liquid handler (PerkinElmer). Serial dilutions using 1 μL of compound 1 and 2 at varying concentrations were dispensed into each well in triplicate. After 48 h incubation, 40 μL of CellTiter-Glo reagent (Promega) was added into each well. The contents were mixed for 2 min on a microplate shaker to induce cell lysis and further incubated at room temperature for 10 min to stabilize the luminescent signal. Luminescence was measured with an EnVision plate reader (PerkinElmer) and %inhibition calculations were performed using the following formula for single-point normalization: %Inhibition = (1-Raw Sample Value/Mean of DMSO Signal Reference Value) × 100. Dose response curves including EC50 calculations were processed using GraphPad Prism software.
To prepare the cell lysate, N. fowleri trophozoites were removed from the culture flask surface by incubating in an ice bath for 10 min, centrifuged at 300 g for 10 min, and washed twice with PBS (pH 7.2). The cells were disrupted by four cycles of freeze thawing in PBS [51]. Protein concentration was quantified by the method of Bradford (Bio-Rad). The activity of the cysteine protease present in the crude extract after incubating in presence and absence of different concentrations of compounds 1 and 2 was assayed by the liberation of the fluorescent leaving group, 7-amino-4-methyl coumarin (AMC), from the peptide substrate Z–Phe–Arg–AMC (40 μM) (where Z is benzyloxycarbonyl, R&D Systems) [45]. The assay was performed at 25°C in an automated microtiter plate spectrofluorometer (EnVision, PerkinElmer) with excitation wavelength at 355 nm and emission wavelength at 460 nm [52]. Enzyme samples were added to the reactivation buffer (10 mM Tris, 5 mM EDTA, 50 mM NaCl, pH 7.4, 10 mM DTT), and preincubated for 20 min at 37°C prior to the hydrolysis of substrate. The rate of substrate hydrolysis at ambient temperature was determined from the rate of increase of fluorescence, monitored on a continuously recording spectrofluorometer and measured as RFU/min/μg protein.
An aqueous methanolic extract of the creosote plant was partitioned against hexane, ethyl acetate and n-butanol successively to obtain four solvent partitioned fractions. These fractions were tested for antiparasitic activity, the ethyl acetate fraction showed activity at 50 μg/mL and was selected for further study. It was fractionated on Sephadex LH-20 and the fractions were subjected to chromatographic separation by HPLC to yield 1–9 as pure compounds.
Compound 1 was obtained as a yellow resinous mass. The 1H, 13C, and HMQC NMR (acetone-d6) indicated 9 carbon resonances and corresponding proton signals, consisting of one methyl [δH 0.83 d (6.6)], four methines [δH 1.74 m], three aromatic signals displaying an ABC splitting pattern [δH 6.52 dd (7.9, 1.8); δH 6.69 d (1.8); and δH 6.73 d (7.9)], and one methylene [δH 2.21 dd (13.3, 9.2); δH 2.70 dd (13.3, 5.0)]. These data were identical with the known creosote secondary metabolite, nordihydroguairetic acid (NDGA) (Table S1 and Fig. S1-S3 in S1 Appendix) [53]. Next, we identified known compound 2 as 3’-O-methylnordihydroguairetic acid (3’-O-methyl-NDGA) [54]. Although similar in structure to 1, compound 2 is non-symmetrical, which revealed the full 19 carbon resonances and corresponding proton signals as follows: two methyls [δH 0.82 d (6.6), 0.83 d (6.6)], eight methines (δH 1.74 m, 2H), six aromatics [δH 6.58 dd (8.0, 2.0), δH 6.61 d (1.9), δH 6.64 dd (8.0, 1.9), δH 6.67 d (2.0), δH 6.77 d (8.0), and δH 6.82 d (8.0)], and two methylenes [δH 2.25 dd (13.1, 9.3), δH 2.71 dd (13.3, 4.8), δH 2.25 dd (13.1, 9.4), δH 2.68 dd (13.3, 5.0)]. In addition, DEPT-135 and HMQC supported the presence of two methyls (δc 16.6, 16.4), eight methines of which two aliphatic (δc 39.3, 39.1) and six aromatic (δc 113.2, 115.4, 115.8, 116.9, 121.2, 122.3), two methylenes (δc 40.0, 39.2) and six quaternary aromatic (δc 134.1, 134.3, 143.8, 145.4, 145.7, 48.1) (Table S2 and Fig. S4-S8 in S1 Appendix).
We identified compound 3 as Nor-3’-demethoxyisoguaiacin and 4–6 as analogs of 3 that have a tetrahydronaphthalene ring system [54, 55]. The 1H NMR (CDCl3) displayed the following signals: two methyls [δH 0.88 d (6.9), 0.89 d (6.9)], nine methines including three aliphatic [δH 3.57 d (6.2), 1.89 m, 1.99 m], two aromatic singlets (δH 6.60 s, δH 6.29 s) resulting from an A2B2 tetra-substituted phenyl ring, four signals giving an A2B2 splitting pattern [δH 6.86 (2H, d 8.5), δH 6.69 (2H, d 8.5)] due to a 1,4-disubstituted phenyl, and one methylene [δH 2.83 dd (16.4, 5.5), δH 2.41 dd (16.4, 7.2)]. The 13C NMR (acetone-d6) displayed eighteen signals and HMQC supported the presence of two methyls (δc 16.1, 16.3), three methines (δc 50.8, 41.8, 30.1), one methylene (δc 35.7), one A2B2 substituted phenyl (δc 115.9 d, 117.7 d, 128.1 s, 130.7 s, 140.0 s, 144.4 s), and one 1,4-disubstituted phenyl [δc 115.7 (2C, d), 130.8 (2C, d), 139.3 s, 156.3 s] (Table S3, Fig. S9-S13 in S1 Appendix). 4–6 were easily dereplicated due to varying methoxy and phenol substituents. Specifically, compound 4 (Nor-isoguaicin) has a methoxy in the 3’-position, which was determined by the ABC proton splitting pattern [δH 6.79 d (8.0), δH 6.52 d (1.8), δH 6.50 dd (8.0, 1.8)] from the tri-substituted phenyl ring (Table S4 and Fig. S14-S18 in Appendix). Conversely, compounds 5 (3’-Demethoxyisoguaiacin) has a methoxy group in the 7 position of the tetra-substituted ring (Table S5 and Fig. S19-S22 in S1 Appendix) and 6 (6,3'-Di-O-demethylisoguaiacin) which contains a 3’,4’-dihydroxy phenyl moiety were determined by comparison with the reported chemical shifts (Table S6 and Fig. S23-S25 in S1 Appendix) [54, 56]. Finally, 7 was purified as a colorless oil and identified as 3-hydroxy-4-epi-larreatricin with 1H and 13C NMR matching the known literature structure (Table S7 and Fig. S26-S30 in S1 Appendix) [57].
During the purification of 1–7 we identified lignan 8 and flavanol 9, however, these secondary metabolites have never been isolated from the creosote plant (8) or were not structurally well characterized (9). Therefore, we report herein the isolation and structure elucidation from the creosote plant. Compound 8, was purified as a colorless oil and the molecular formula was deduced from the HRMS and 13C NMR as C18H20O4. The 1H NMR (Table 1) displayed signals attributable to two methyl groups [δH 0.97 d (6.6), δH 0.57 d (7.1)], and eleven methines, including: two oxygenated aliphatic protons [δH 5.38 d (4.2), δH 4.54 d (9.4)], two aliphatic protons [δH 2.38–2.44, m, 2H], four aromatic protons giving an A2B2 splitting pattern [δH 7.17 d (8.1), δH 6.81 d (7.8)], and three aromatic protons giving an ABC splitting pattern [δH 6.91 br s, δH 6.81 br dd (7.2), δH 6.72 d (7.2)]. The 13C NMR revealed the occurrence of eighteen carbons resonances, DEPT-90 in conjunction with HMQC supported the presence of seven aromatic methines, including A2B2 [δc 128.0 x 2 and δc 115.5 x 2] and ABC splitting patterns (δc 118.5, δc 115.7 and δc 114.0). Further, we observed two oxygenated [δc 86.2 and δc 85.2] and two non-oxygenated (δc 48.4 and δc 44.0) methines as well as two methyl functional groups (δc 12.2 and δc 9.7). The remaining five quaternary 13C NMR signals were indicative of aromatic chemical shifts (δc 157.0, 146.0, 145.1, 136.5 and 132.8). These NMR data were identical with the previously reported enzymatically synthesized (±) 3-hydroxy-larreatricin [44]. We observed HMBC correlations from aromatic H-2 (δH 6.91) of the tri-substituted phenyl ring to C-7 (δc 86.2) of the furan ring. In addition, HMBC correlations from H-2’/H-6’ (δH 7.17) of the 1,4-di-substituted phenyl ring to C-7’ (δc 85.2) of furan ring proved the attachment of two phenyl rings at C-7 and C-7’ of furan ring, respectively (Fig 3A). These assignments were further confirmed by the HMBC correlations of H-7/C-2 and H-7’/ C-2’, C-6’. The position of two methyls of furan ring was elucidated using HMBC cross peaks between methine H-7’ (δH 5.38) and methyl C-9’ (δc 9.7) and between methine H-7 (δH 4.54) and methyl C-9 (δc 12.2). Finally, the relative stereochemistry of four stereogenic centers in furan ring was assigned by the 1D nuclear Overhauser effect (NOE) experiment (Fig 3B). Irradiation at δH 4.54 (H-7) gave enhanced signals at δH 6.92 (H-2), δH 0.97 (H-9) and δH 0.57 (H-9’), indicating the spatial proximity of H-2, H-9 and H-9’. In addition, irradiation at δH 5.38 (H-7’) gave enhanced signal exclusively at δH 7.17 (H-2’/H-6’), the absence of correlations between H-7’ and H-7 clearly indicated the trans configuration of the 2-substituted phenyl ring. Accordingly, the structure of 8 was established as (7R, 7’R)-7, 7’-bis(4’, 3, 4-trihydroxyphenyl)-(8R, 8’S)-8, 8’-dimethyltetrahydrofuran (Fig. S31-S38 in S1 Appendix), which is a stereoisomer of 7.
Compound 9 was obtained as yellow solid and its molecular formula, C16H12O6, was deduced by HRMS as well as 1H and 13C NMR analysis. In the 1H NMR (CDCl3 + CD3OD) spectrum, a methoxy functionality [δH 3.74 s] was observed as well as six aromatic methines including two singlets [δH 6.35 s, 6.20, s] and an A2B2 splitting pattern [δH 7.93 d (8.6), 2H; 6.88 d (8.6), 2H] resulting from a 1,4-disubstituted phenyl ring. The 13C NMR (Table 2) showed sixteen carbon signals and DEPT-90 in conjunction with HMQC supported the presence of one methoxy (δc 60.1) and six aromatic methines of which four [δc 130.3 x 2 and 115.6 x 2] correlated to two doublet signals giving an A2B2 pattern. In addition, we observed two signals that correlated with two aromatic proton singlets of the tetra-substituted phenyl ring (δc 98.9 and 94.1). The remaining nine quaternary 13C NMR signals include a carbonyl (δc 178.8), six aromatic and two olefinic carbons (δc 163.9, 161.5, 159.7, 157.0, 156.5, 138.4, 121.7 and 105.2). These NMR data were consistent with a flavonol ring system containing three hydroxyls and one methoxy group. The HMBC cross peaks observed between the aromatic protons in the A-ring with H-8 (δH 6.35), C-7 (δc 163.9), C-8a (δc 157.0), and C-5a (δc 105.2) (Fig 4). Cross peaks were also observed between proton H-5 (δH 6.20), C-6 (δc 161.5), and C-5a (δc 105.2) suggesting the attachment of two hydroxyl groups at C-7 (δc 163.9) and C-6 (δc 161.5). In addition, these cross peaks indicated an oxygen attachment to C-8a (δc 157.0), signifying the O-1 position of the flavonol C-ring. The flavonol B and C ring connectivity were elucidated using HMBC correlations between protons H-2’/H6’ (δH 7.93) and carbons C-2 (δc 156.5), and C-4’ (δc 159.7). The phenolic substitution on ring B was indicated through H-3’/H-5’ (δH 6.88) and carbon C-1’ (δc 121.7) correlations. Finally, the HMBC cross peak between methoxy protons (δH 3.74) and C-3 (δc 138.4) indicated that attachment at the C-3 position of the flavonol C-ring (Fig. S23-S33 in S1 Appendix) [58]. Therefore, we have precisely determined compound 9 to be 3-methoxy-6, 7, 4’-trihydroxyflavonol.
We previously developed a high-throughput screening CellTiter-Glo ATP bioluminescence-based assay to assess antiparasitic activity [48], and used this assay to test compounds 1–9 against the trophozoite stage of E. histolytica, G. lamblia, and N. fowleri. Compounds 1–6 displayed dose response antiparasitic activity against all three pathogens by reducing the culture density by 50% (EC50) compared to untreated trophozoite cultures (Table 3). Compound 1 proved to be the most potent against both G. lamblia and N. fowleri (EC50 = 36 μM) (Fig 5). However, 1 and 2 display similar EC50 values, and both exhibited only moderate activity against E. histolytica with EC50 values of 103 μM and 171 μM, respectively. Both compounds 1 and 2 were found to be about 1.5-fold more active relative to the current standard drug miltefosine (EC50 = 54.5 μM) against N. fowleri. Compound 3 was more active against G. lamblia (EC50 = 49 μM) than E. histolytica (EC50 = 94 μM) or N. fowleri (EC50 = 73 μM), whereas compound 4 had similar activity against all three pathogens with EC50 values from 74 μM to 83 μM. Compounds 5 and 6 had comparatively weak activity against the three pathogens. Similarly, 9 displayed modest antiparasitic activity against G. lamblia (EC50 = 153 μM) and N. fowleri (EC50 = 235 μM) (Table 3). Larreatricin derivatives and stereoisomers 7 and 8 displayed no antiparasitic activity.
To further assess the therapeutic potential of 1 and 2, which displayed the most potent antiparasitic activity agains N. fowleri, we conducted a cytotoxicity study with human umbilical vein endothelial cells (HUVEC), using the same CellTiter-Glo assay and time course that we used for assessing trophozoite toxicity (Fig 5B). Compounds 1 and 2 inhibit HUVEC cell viability with EC50 values of 86 μM and 59 μM, respectively. Thus, 1 and 2 are correspondingly 2.4 fold and 1.6 fold less toxic to human cells compared to N. fowleri, which is statistically significant (P<0.0001) (Fig 5C).
NDGA was previously shown to inhibit cysteine protease in cancer [59], and recent studies linked the involvement of cysteine protease in the pathogenesis of N. fowleri [45]. Thus, we investigated the effects of compounds 1 and 2 on cysteine protease activity present in total crude lysate of N. fowleri over a concentration range from 1.875-to-30 μM. The dose dependent effect varied between 1 and 2, however, both inhibited the cysteine protease activity by almost 50% at 1.875 μM (Fig 6). This data indicates that the activity of compounds 1 and 2 against whole cell N. fowleri may be due to the modulation of cysteine protease activity present in the trophozoites.
Because lignans 1–6 are from the same structural class of compounds we could assess notable structure activity relationships (SAR). For example, 1 and 2 displayed overall more potent activity compared to 3–6, which may be a result of the more flexible straight chain structure that offers more conformational flexibility compared to 3–6. In addition, introducing a methoxy group in the 3’-position of 2 appears to be negligible with regard to SAR. Conversely, 3 and 4 only differ by one methoxy group in the 3’ position (i.e. compound 4), which reduced the antiparasitic activity against G. lamblia by ~2 fold. However, this functional group was dispensable when comparing the activity between E. histolytica and N. fowleri. Similarly, introducing a phenol in the 3’ position as in 6 also results in reduced activity compared to 3. The most striking SAR is observed by introducing a methoxy group in the seven position such as in 5, which results in a substantial loss of activity compared to 3: ~3 fold (E. histolytica), 4-fold (G. lamblia), and ~ 2 fold (N. fowleri). Although 1–6 are proposed to be biosynthesized from 7 and 8 [44] and share many of the same structural features, these compounds displayed no antiparasitic activity. To better understand this SAR we compared the calculated LogP values for 1–9. Compounds 7 and 8 are 10 fold more hydrophilic (CLogP = 3.5) compared to 1–6 (CLogP = 4.5). However, the flavonol 9 (CLogP = 1.1) is 1,000 fold more hydrophilic compared to 7 and 8. Interestingly, flavonoids are known to actively diffuse through organism membranes via membrane transporters such as the ATP-binding cassette (ABC) transporters [60]. Moreover, parasitic protozoa are known to express these ABC transporters and other relevant transporters utilized by flavonoids [61], which may explain the activity of 9 compared to 7 and 8. Thus, it is plausible that the difference in hydrophilicity may be a physical property of 7 and 8 preventing diffusion into the parasite trophozoites, explaining their inactivity compared to 1–6 and 9.
Compounds 1 and 2 did not display more potent activity against E. histolytica and G. lamblia compared to metronidazole, but both compounds where 1.5 fold more potent against N. fowleri compared to miltefosine, which is used for the treatment of PAM. Therefore, we selected N. fowleri for follow-up studies with compounds 1 and 2. Interestingly, although NDGA has been shown to be cytotoxic to tumor cells by inducing apoptosis and possess antiviral activity [62, 63], it has also been shown to be a neuroprotective agent and protective of human monocytes and other human cells and tissues through its powerful antioxidant activity [62–65]. However, at high doses, NDGA has been shown to display nephrotoxicity and hepatotoxicity [62]. Importantly, our data and the collective literature reports described herein indicate that NDGA and derivatives have some therapeutic potential against N. fowleri.
Next, we investigated a potential molecular target of NDGA by review of the literature. A report by Huang et al. showed that the NDGA derivatives significantly inhibited cysteine protease activity [59]. Recent studies have also reported that N. fowleri lysate contains cysteine proteases such as cathepsin B-like protease that are important virulence factors of N. fowleri. Cysteine cathepsins are also critical to invasion, evasion, immunomodulation and are implicated in the attachment mechanism to the host tissue [45, 51, 66]. Moreover, cathepsins also potentiate N. fowleri growth [66]. Based on these studies we hypothesized that 1 and 2 may be inhibitors of cysteine protease activity present in N. fowleri. Indeed, our results validated this hypothesis and show that NDGA/derivatives inhibit 50% to 80% of N. fowleri cysteine protease activity between 1.875–7.5 μM (Fig 6), which are potencies that are consistent with our antiproliferative and antiparasitic data against N. fowleri (Table 3 and Fig 5).
In conclusion, lignans 1–8 and flavonol 9 represent two well-known classes of plant secondary metabolites [67, 68]. The well-studied flavonoid class of natural products such as 9 display a broad range of biological activity including antiparasitic activity [68, 69]. Likewise, lignan natural products have received strong interest and have been intensely studied due to their broad clinically relevant biological activity, including: antioxidant, antiviral, antibacterial, immunosuppressive, anti-inflammatory, and anticancer properties [67, 70]. Only one previous study reported in 1978 demonstrated that NDGA isolated from L. tridentata had inhibitory effect on the growth of non-pathogenic Entamoeba invadens [71]. Our report, for the first time, demonstrates that lignans isolated from L. tridentata are active against pathogenic E. histolytica and G. lamblia, which directly cause human amebiasis and giardiasis, respectively. Moreover, literature reports of natural products effective against N. fowleri growth have been limited [28, 72] and our study has identified relatively potent compounds from L. tridentata that have amebicidal activity against N. fowleri, which we show may be due to inhibiting cysteine protease activity present in the lysate of N. fowleri. Therefore, lignan secondary metabolites from the creosote bush represent a class of natural products pharmacophore that can be optimized through medicinal chemistry to translate more effective therapeutic options for amebiasis, giardiasis, and PAM.
|
10.1371/journal.pcbi.1003392 | Inferring Developmental Stage Composition from Gene Expression in Human Malaria | In the current era of malaria eradication, reducing transmission is critical. Assessment of transmissibility requires tools that can accurately identify the various developmental stages of the malaria parasite, particularly those required for transmission (sexual stages). Here, we present a method for estimating relative amounts of Plasmodium falciparum asexual and sexual stages from gene expression measurements. These are modeled using constrained linear regression to characterize stage-specific expression profiles within mixed-stage populations. The resulting profiles were analyzed functionally by gene set enrichment analysis (GSEA), confirming differentially active pathways such as increased mitochondrial activity and lipid metabolism during sexual development. We validated model predictions both from microarrays and from quantitative RT-PCR (qRT-PCR) measurements, based on the expression of a small set of key transcriptional markers. This sufficient marker set was identified by backward selection from the whole genome as available from expression arrays, targeting one sentinel marker per stage. The model as learned can be applied to any new microarray or qRT-PCR transcriptional measurement. We illustrate its use in vitro in inferring changes in stage distribution following stress and drug treatment and in vivo in identifying immature and mature sexual stage carriers within patient cohorts. We believe this approach will be a valuable resource for staging lab and field samples alike and will have wide applicability in epidemiological studies of malaria transmission.
| The human malaria parasite Plasmodium falciparum is transmitted through a mosquito vector and causes over half a million deaths per year. The microorganism cycles through asexual and sexual life cycle stages, and its successful transmission relies on cells in the sexual stage. These stages are, however, present only at low levels during infection; most infecting cells are asexually reproduced. It can be challenging to assign biomolecular activity to particular parasite life cycle stages from typical gene expression profiles, given the mixed stage composition of most samples. We developed a deconvolution model to identify components of Plasmodium transcriptional activity contributed by sexual and asexual life cycle stages, initially using samples of known composition. From these, we optimized a small set of stage-specific genes with highly informative expression patterns and trained an inference model to predict the stage composition of new samples. The model successfully inferred the parasite's transition from asexual to sexual development over time under laboratory conditions and identified a subset of patient samples harboring transmissible sexual stages. The system presented here can aid in epidemiological or laboratory perturbation in which stage composition is an important step in understanding and preventing malaria transmission.
| One of the tenets of the recently released Malaria Eradication Research Agenda (malERA) is the development of new diagnostics specifically addressing transmission reduction [1]. Individuals harboring the Plasmodium falciparum transmissible parasite stage, or gametocyte, are the primary reservoir for malaria transmission, and thus proper surveillance of gametocyte carriers is critical to transmission reduction. Surveillance is difficult, however, because gametocytes comprise only a small fraction of the total body parasite load during active infection and are only observed in the bloodstream in their mature form, while developing stages are sequestered in tissues [2]. For these reasons, quantifying gametocytes in mixed parasite populations has been an ongoing challenge ever since they were first identified more than a century ago.
Gametocytes do execute substantially different transcriptional programs from asexual parasite stages, however, as has been well-studied in vitro [3]. Like the sequential dynamics of the asexual Plasmodium life cycle [4], [5], gametocytes develop in a staged progression from immature (young and intermediate stages) to mature transmission-competent cells in preparation for meiosis and further development in the mosquito vector. The switch between asexual replication and sexual development does not occur ubiquitously in vivo or in vitro, as even the most synchronized gametocyte induction protocols result in partially asynchronous and mixed gametocyte stages [3], [6]. This problem is compounded in vivo, as blood sampled during infection is likely to contain both gametocyte and asexual parasite populations, leading to a highly convolved transcriptional mixture.
In addition to the need to dissect these signatures for analysis of microarray data, it is also of interest to develop a field-friendly approach for detecting and quantifying both immature (indication of conversion to sexual development) and mature (indication of infectiousness to mosquito vector) gametocyte stages. Transcriptional approaches such as RT-PCR, QT-NASBA and RT-LAMP have been developed [7], [8], [9] using the established mature gametocyte marker Pfs25 and the putative immature gametocyte marker Pfs16. While these approaches enable sensitive detection of these transcripts, it is unclear how the detection of these transcripts - particularly Pfs16 - relates to actual gametocyte carriage [8]. The development of a qRT-PCR-based assay has thus far been impeded primarily because this approach cannot distinguish transcript from genomic DNA when sequences are identical; the majority of P. falciparum genes lack introns and thus have identical sequences for both RNA and DNA. It is therefore worth identifying novel intron-containing markers for which exon-exon junction-spanning primers can be designed so that this approach can be used for in vivo gametocyte quantification.
Our goal was thus to develop a new transcript-based gametocyte model that addressed these challenges. Using a deconvolution approach, we quantified the stage-specificity of Plasmodium transcripts genome-wide and subsequently identified intron-containing markers from across the full range of asexual and sexual development. In order to identify expression patterns specific to different gametocyte stages, particularly the immature stages, existing in vitro asexual and sexual developmental time course samples were re-analyzed to account for their mixed stage composition. We further developed a qRT-PCR assay based on these results and, applying the model in reverse, established an algorithm to estimate the amounts of immature and mature sexual and asexual stages in a patient sample based on the expression of a small set of stage-specific markers. This was inspired by related approaches that have been used successfully in dealing with mixed cancerous/non-cancerous tissue samples [10], [11] and with mixed stages of budding yeast [12]. This framework is implemented for public use at http://huttenhower.sph.harvard.edu/malaria2013; as a transmission-focused tool, this system can be applied in epidemiological settings, and as such will ideally support efforts directed toward reducing malaria prevalence worldwide.
To gauge how this modeling process performed on patient microarray samples, we applied it to microarray data from two patient cohorts, i) a previously published cohort of severe malaria patients from Blantyre, Malawi collected in 2009 [23] and ii) a cohort of uncomplicated malaria patients from Thies, Senegal collected in 2008. While no staging information was available for the Senegal patients, a subset of Malawi patients were previously identified as gametocyte-positive by thick smears. The model inferred that the majority of patients from both cohorts have a strong ring-dominated profile, with the next largest subset being late asexual stages (trophozoites and schizonts) (Figure 4A/B). For the 10 Malawi samples in which gametocytes were observed by thick smear, our model correctly identifies 4 (40%) as such, with 0 false positive developing or mature gametocytes predicted among the 48 thick smear-negative patients (Figure 4A). Interestingly, two thick smear-negative patients are predicted to have young gametocytes, which are difficult to identify by thick smear microscopy due to their morphological similarities with asexual stages.
A subset of the uncomplicated malaria patients from Senegal were also predicted to be gametocyte carriers (6 of mature, 1 of developing and 1 of young gametocytes) (Figure 4B). As microscopy-based information was unavailable for the Senegalese cohort, we assessed how our gametocyte inferences correlated with patient parameters. Of the 6 parameters we measured for this cohort, illness duration and hematocrit differed significantly between the group of patients inferred to be gametocyte carriers and those inferred to be gametocyte-negative. The former had a longer duration of illness (6.33 days±1.02 SEM) than the latter (3.84 days±0.25 SEM, t-test p = 0.0014) as well as a lower hematocrit measured in percent cell volume (34.86%±2.17 SEM) than the latter (40.41%±1.06 SEM, t-test p = 0.031) (Supplementary Table S4). This finding agrees with published data on clinical correlates of gametocyte carriage: long illness duration (greater than 2 days) and anemia (hematocrit less than 30%) were both independently found to be risk factors of gametocytemia in uncomplicated malaria [24].
We next sought to test a variation of the microarray-based model for application to transcriptional measurements obtained by PCR, which might eventually be more appropriate for a field assay. As no MG marker that achieved our filtering criteria (see Figure 1C) for qRT-PCR also matched both the high expression levels and stage-specificity of the existing Pfs25 marker for gametocyte detection, we assessed the utility of the YG and DG markers in combination for the prediction of immature and mature gametocyte quantities when applying the model to qRT-PCR data. Specifically, PF14_0748 and PF14_0367 were likely to represent immature (IG) and mature (MG) gametocytes in combination, as PF14_0367 had a βg,s parameter similar to that of Pfs25 in mature gametocyte stages. We therefore cross-validated this 5-marker PCR set (Table 1) comparably to the 6-marker microarray set, using the in vitro microarray time courses as described above. The simplified model remained able to predict stage distribution accurately, with a root mean squared error comparable to that of the 6-marker model (Supplementary Figure S1).
In order to create a qRT-PCR assay for our sentinel transcripts, we designed exon-exon junction spanning primers (distinguishing transcripts from genomic DNA) and sequence-specific probes (distinguishing transcripts from non-specific background amplification). Following confirmation that our primer/probe sets selectively amplified cDNA and not genomic DNA or non-specific products, we validated the stage-specific expression using in vitro-derived asexual and sexual stage RNA (Table 1, and supplementary Figure S2A, and Table S5 for optimization and validation of qRT-PCR parameters). For these experiments, we used the gametocyte-producing reference line 3D7 and a gametocyte-deficient clone thereof (termed F12 [29]) to confirm the stage-specificity of each of our sentinel markers. Normalized expression data from time courses of 3D7 and F12 confirmed stage-specificity of our sentinel marker set (Supplementary Figure S2B). The asexual markers alternate with respect to time points in which there were predominately rings or trophozoites and schizonts in the culture, with similar results for both the F12 and 3D7 lines. The sexual markers demonstrate stage-specificity within the 3D7 time course and no appreciable expression in the F12 line once normalized. Specifically, PF14_0748 expression is detected in the early and mid gametocyte time points, while PF14_0367 expression is detected in both mid and late time points.
Several highly sensitive single-marker molecular assays are currently used to detect Plasmodium gametocytes. None of these existing tools have been appropriate for detection and quantification of the relevant range of parasite stages present during infection, however, due primarily to the lack of a sufficiently broad panel of stage-specific markers. Further, since malaria parasite populations exist as mixtures of the different phases of the life cycle, assays combining multiple markers require customized computational analysis methods for dealing with this complexity. We combined the development of such a bioinformatic deconvolution approach with panels of stage-specific, intron-containing markers appropriate both for microarray analysis and a newly developed qRT-PCR assay. This multi-marker platform enabled us not only to detect gametocyte carriers but primarily to infer the relative amounts of sexual and asexual stages within a sample. We provide an implementation of this platform for further development and application, particularly for refinement in field settings. This process can also be adapted bioinformatically by the exclusion or inclusion of markers to answer specific questions, such as determination of parasite sex ratios that are known to influence mosquito infectiousness [31].
Our deconvolution model provided the opportunity to define stage-specific gene sets and to characterize the biology of these stages' expression programs using tools such as GSEA, even in the absence of transcriptional data from pure stage populations. For example, our GSEA analysis confirms earlier studies that suggested increased mitochondrial and lipid metabolism during gametocyte development [3], [32]. Interestingly, the analysis also suggests significant enrichment of several markers related to endocytic trafficking in late gametocyte development but not in any other parasite stage. The biological significance of this observation remains to be determined. To put such findings into context and ultimately describe the gametocyte transcriptome at high resolution, a systematic transcriptional re-analysis of the entire P. falciparum gametocyte cycle using isolated and synchronous gametocyte stages will be required.
Transcriptional approaches have significantly increased the sensitivity of gametocyte detection in field-compatible assays [8], [9], [33], [34], [35]. However, these have been limited to either (i) qualitative assessments of multiple gametocyte markers, i.e. RT-PCR of immature and mature gametocyte markers [35], or (ii) quantitative assessments of mature gametocytes only, i.e. QT-NASBA of the gamete surface antigen Pfs25 [8]. In order to properly define the reservoir of parasite and gametocyte carriers in the field, it is imperative to determine both the absolute parasite burden and the stage composition of parasites in the blood circulation. Challenges have prevented the development of a diagnostic that can measure the latter, such as (i) the lack of transcriptional analysis methods to identify gametocytes with high specificity in a sample containing a mixture of stages, (ii) the lack of validated immature gametocyte markers, and (iii) the lack of known intron-containing qRT-PCR compatible markers for all stages. We tackled these challenges by developing a model specific to the quantification process and ensuring that it was compatible with both microarray and qRT-PCR measurements. This is distinct, of course, from models that would focus only on sensitivity and specificity of gametocyte detection from such data, which represent a potentially fruitful course of future computational investigation. Instead, by incorporating relative expression values of the markers, the model allowed us both to identify a subset of patients as gametocyte carriers and to additionally quantify sub-categories of immature and mature gametocyte fractions within the mixture of stages in the bloodstream.
Following validation of our model on samples for which stage composition was known, we applied our model to two microarray data sets in which stage composition was unknown: (i) a cohort of uncomplicated malaria patients, and (ii) two in vitro growth experiments in the presence of drug. In the former, we found that both mean illness duration and hematocrit differed between inferred gametocyte carriers and non-carriers, in agreement with published data demonstrating that long illness duration and low hematocrit is linked to gametocyte carriage [24]. In the latter, we observed an increase in the fraction of mature gametocytes as well as unexplained transcriptional signature upon the addition of drug treatment to parasites. The enrichment of mature rather than young gametocytes in response to drug treatment suggests that the drug selectively kills asexual stages, leaving gametocytes unaffected rather than inducing the development of new young gametocytes. The increase in unexplained signatures likely indicates the transition to unhealthy, dying parasite fractions. These applications demonstrate the range of potential uses for this inference tool.
As the exon-exon junction spanning primer/probe sets for 5 markers designed here represent the first attempt at a multi-marker gametocyte-staged qRT-PCR assay, further modeling of PCR-specific measurement error and careful standardization of experimental protocol for this difficult task will both improve field inferences. Like the microarray expression model, however, this model successfully recapitulated the transition from asexual to sexual development across multiple in vitro experiments even on first application. When used initially in vivo for blood samples from a cohort of children with severe malaria in Malawi, the system successfully identified a subset of patients as immature and/or mature gametocyte carriers. Because immature gametocytes in particular are present in the body several days before the more mature forms emerge, our approach for detecting them could be used in further investigations into factors that influence gametocyte conversion in vivo.
The assay and algorithm framework presented here has potential for use in epidemiological studies such as those of asymptomatic carriers, who likely represent a major reservoir for malaria transmission. Multiple such studies are already ongoing and will yield additional samples to further optimize computational models of gametocyte differentiation. This is also true of data generated from other sensitive expression platforms such as glass-slide arrays or Nanostring. The inference process may thus have applications in better understanding the natural progression of malaria in the human host, by identifying gametocytes earlier in the course of infection and determining the impact of specific drug treatments on gametocyte development. By scaling to future population-level screens, the resulting information will help better characterize the epidemiology of gametocytemia based on malaria transmission intensity, geography, climate and season.
The institutional review boards of the Harvard School of Public Health, Brigham and Women's Hospital, the University of Malawi College of Medicine, and the Ministry of Health in Senegal approved all or parts of this study. Consent was obtained from the patient or a child's guardian.
A transgenic line, 164/GFP, of a gametocyte-producing clone of the 3D7 strain of P. falciparum was used to produce the mixed stage samples for model training and validation. This transgenic line, which aided in the quantification of gametocyte stages, produces stage-specific GFP expression under the PF10_0164 gene promoter, as described previously [22]. A previously characterized non gametocyte-producing clone, F12, of the 3D7 strain was used to confirm stage-specificity of gametocyte markers [29]. Culture conditions were as described previously [38], maintaining the parasite line in O+ blood at 4% hematocrit in RPMI-1640 media supplemented with 10% human serum. Cultures were kept at 37°C in a chamber containing mixed gas (5% CO2, 5% O2, 90% N2). Prior to induction, asexual parasite cultures were synchronized for two cycles with 5% D-sorbitol [39], and subsequently induction of gametocytogenesis was performed according to the Fivelman protocol [28]. Briefly, asexual parasites were grown to a high parasitemia in the presence of partially spent (“conditioned”) medium, and then sub-cultured at the schizont stage into new dishes containing fresh media and erythrocytes. One of two methods was used to reduce the amount of asexual stages in the cultures: Treatment with D-sorbitol was applied on two days later to lyse asexual trophozoite/schizont stages and selectively enrich for unaffected early gametocytes, or N-Acetyl glucosamine was added to the medium one day later and every subsequent day to selectively kill asexual stages.
A 3D7 line was used to study the effect of drug perturbations on parasite growth. Culture conditions were performed as described above. Asexual parasite cultures were synchronized for three cycles with 5% D-sorbitol, and expanded to a parasitemia of 5–6%. Hematocrit was increased from 3 to 6% at the late schizont stage using fresh blood. Upon reinvasion drugs were added to the culture at a concentration of 5×IC50. Drug-treated and control parasites were harvested at 10, 20, 30, and 40 hours post-invasion and RNA was extracted (Qiagen).
In order to accurately quantify the stage distribution of parasites in our in vitro samples, we used a combination of standard and fluorescence microscopy. Parasite stage distribution was monitored throughout the parasite synchronization and induction protocol using Wright's Giemsa stain applied to thin blood smears. Quantification of asexual rings and trophozoite stages, as well as developing and mature sexual stages was done directly by light microscopy. In order to quantify early stages of sexual development that are morphologically similar to asexual stages, we used a combination of live imaging and immunofluorescence microscopy. Live imaging was performed using the transgenic 164/GFP line. Parasites were analyzed using the FITC channel on an inverted epifluorescence microscope (Zeiss) and quantification was done of the proportion of GFP(+) parasites out of the total number of Hoechst (+) parasites. Immunofluorescence assays were performed with cell monolayers on glass slides, prepared as described previously [40]. For labeling with the constitutive gametocyte marker Pfs16, slides were fixed in ice-cold methanol, blocked with 5% nonfat dry milk powder, incubated with polyclonal mouse antibody against Pfs16 (1∶2500) [41], washed and incubated with a secondary antibody conjugated to Alexa 488. Parasite nuclei were labeled with DAPI and quantification was done on the proportion of FITC(+) parasites out of the total number of DAPI(+) parasites. For time points in which we had data from both live and immunofluorescence experiments [41], the quantification of early gametocytes from both methods was averaged to give the final amount.
RNA (from peripheral blood of Senegalese patients and cultured in vitro drug perturbations) was assessed by Bioanalyzer (Agilent), and high quality RNA samples were labeled and hybridized to an oligonucleotide array (Affymetrix) custom-designed for the P. falciparum 3D7 genome, PlasmoFB, as published previously [5]. The raw CEL files were condensed into GCT expression files using RMA and the default parameter settings in ExpressionFileCreator in GenePattern [43].
Given the learned model parameters βg,s from stage-labeled data with known xs, the model was inverted to infer the unknown stage distributions xs in new samples. A quadratic programming approach was used to solve the system of linear equations with the constraint that the proportions of all stages must sum to 1 and that each stage contributes a non-negative fraction of expression:
We implemented this process in the R function quadprog and solved for the stage distributions using the sets of six (for microarrays) or five (for PCR data) markers ultimately selected as follows.
|
10.1371/journal.pntd.0000938 | Schistosoma mansoni Infections in Young Children: When Are Schistosome Antigens in Urine, Eggs in Stool and Antibodies to Eggs First Detectable? | In Uganda, control of intestinal schistosomiasis with preventive chemotherapy is typically focused towards treatment of school-aged children; the needs of younger children are presently being investigated as in lakeshore communities very young children can be infected. In the context of future epidemiological monitoring, we sought to compare the detection thresholds of available diagnostic tools for Schistosoma mansoni and estimate a likely age of first infection for these children.
A total of 242 infants and preschool children (134 boys and 108 girls, mean age 2.9 years, minimum 5 months and maximum 5 years) were examined from Bugoigo, a well-known disease endemic village on Lake Albert. Schistosome antigens in urine, eggs in stool and host antibodies to eggs were inspected to reveal a general prevalence of 47.5% (CI95 41.1–54.0%), as ascertained by a positive criterion from at least one diagnostic method. Although children as young as 6 months old could be found infected, the average age of infected children was between 3¼–3¾ years, when diagnostic techniques became broadly congruent.
Whilst different assays have particular (dis)advantages, direct detection of eggs in stool was least sensitive having a temporal lag behind antigen and antibody methods. Setting precisely a general age of first infection is problematic but if present Ugandan policies continue, a large proportion of infected children could wait up to 3–4 years before receiving first medication. To better tailor treatment needs for this younger ageclass, we suggest that the circulating cathodic antigen urine dipstick method to be used as an epidemiological indicator.
| In sub-Saharan Africa, intestinal schistosomiasis is a debilitating disease caused by a worm infection. To arrest disease progression, de-worming medications are given out, often en masse, to school-aged children. In Uganda, however, much younger children can be infected, and in lakeshore communities both infants and pre-school children can already show signs and symptoms of intestinal schistosomiasis. To change de-worming practices, further information on the occurrence of infections in these younger is needed for evidence-based decision making. Our study applied current methods of disease diagnosis to better define the ‘age of first infection’ and estimate general infection prevalence within a disease-endemic village. Up to 50% of young children were clearly shown to have schistosomiasis and could likely wait up to 3–4 years before obtaining first treatment if present de-worming policies are not changed. In the context of identifying future treatment needs, we propose that antigen detection methods are most suitable.
| Throughout the last decade several large-scale preventive chemotherapy campaigns, waged against neglected tropical diseases, have progressively scaled up operations to reach nationwide coverage levels in Uganda [1], [2]. For control of intestinal schistosomiasis, as caused by Schistosoma mansoni infection, an active monitoring and surveillance programme, set within the national control programme (NCP), has provided important disease-specific information, assessing the impact of treatment upon the recipient population, as well as, re-alignment of original control objectives first set forth in 2003 [3], [4].
Following WHO guidelines, mass-drug administration of praziquantel (PZQ) is typically focused towards treatment of school-aged children (≥6 years) and adults who reside within disease endemic regions [5], [6]. PZQ is provided free of charge by the NCP and analysis of school and(or) community treatment registers has shown that several million people have received at least one annual treatment of PZQ within the last five years [1], [7]. Although this represents a considerable achievement, targeted epidemiological surveys have revealed that coverage is incomplete as in certain areas, e.g. shoreline environments of Lakes Victoria and Albert, large numbers of preschool-aged children (≤5 years) and infants (≤1 years) are infected with S. mansoni and have been largely overlooked by the treatment campaign [8], [9], [10].
To ensure that this unfortunate health inequality does not persist the treatment needs of younger children are being assessed and we have recently called for formal inclusion of these young children within the Ugandan NCP [11]. It can be safely assumed, for example, that mass-treatment initiatives are vital in most in shoreline villages where infections can be common. Given the geographical focality of schistosomiasis and itinerancy of lakeshore communities, however, an important future challenge for the NCP is collection of sufficient disease-specific information to better tailor local drug needs and set parameters for subsequent programme monitoring [12], [13]. Attention will therefore focus upon those sections of villages where young children are frequently bathed in freshly drawn lake water or are within range of regular ambulation to the lake margins.
Owing to the unique natural history and developmental biology of schistosomes within the mammalian host [14], accurate identification of infected cases is challenging [15], even more so in the younger child where the founding worm population has only recently established and begun to mature. Before female worms develop their full egg-laying capacity, sporadic deposition of eggs may take place with a proportion of these being voided into the bowel lumen and ejected in faeces whilst the remainder become trapped within the host's tissues [16]. Interacting with this are also the beginnings of the child's innate and adaptive immune responses to excretory-secretory products of the worms themselves, as well as these responses being primed or modulated by maternally induced effects, for example, during pregnancy and(or) breastfeeding [17], [18], [19], [20]. It is also of particular note that the child's immune system is in a maturing flux of recognition between self- and non-self epitopes [21] and the efficacy of PZQ, which is poor against immature worms of S. mansoni [22], is only starting to be explored in this ageclass [11]. From a general diagnostic perspective as existing tools are sub-optimal, improvement of methods and techniques for detection of intestinal schistosomiasis continues [15] but in the context of the younger child, it is not yet clear which of the present methods, or combinations thereof, is either most appropriate or applicable for routine use within the NCP.
We therefore report on a field-based study which attempted to determine the age of first infection in very young children with available techniques and also estimate, as accurately as possible, the general prevalence of intestinal schistosomiasis within this ageclass from a typical lakeshore community. The performance of methods that detect schistosome - antigens in urine, antibodies to egg antigens in serum and eggs in stool - was compared. For ease of comparison, our methods are subsequently referred to as: an antigen detection method (ADM), an indirect egg detection method (IEDM) and a direct egg detection method (DEDM), respectively.
This field study was carried out in April 2009 in Bugoigo on Lake Albert (GPS co-ordinates, N 01° 54′.481″, E 31° 24′.597″), a fishing village impoverished both in terms of sanitation and hygiene that has been the location of several previous research/control studies on intestinal schistosomiasis [23], [24], [25], [26]. Prevalence of infection within local school-aged children has been continuously high (>50%) despite annual chemotherapy [27] and infections in infants and preschool children were first formally recorded in July 2007 [11].
Owing to itinerancy, the exact number of inhabitants in Bugoigo is not precisely known but is likely in the region of several thousand. The village contains up to three thousand traditional hut dwellings which stretch 3–4 km along the lakeshore and up to 1–2 km inland. Sanitation and hygiene in this village is minimal with few potable water sources and insufficient pit latrines. Household water is typically drawn directly from the lake at specific collection points and then taken back to each homestead in plastic jerry cans for subsequent domestic use. These lakeshore margins, like elsewhere on Lake Albert, provide conducive aquatic habitats for Biomphalaria spp., the intermediate snail hosts of S. mansoni, and can be found throughout the year, although infected snails vary in numbers seasonally [28], [29].
The immediate and longer-term objectives of this study were explained to the local community mobiliser who identified a total of 134 mothers that were willing to participate, bringing up to two of their infants/preschool children (≤5 years of age), and attend the two-day clinic commencing on the following day. After obtaining written informed consent from each mother on her own behalf and on behalf of her child(ren), urine, stool and fingerprick blood samples were obtained from all participants on the first day of the clinic. Mothers were then asked a suite of detailed questions recording their demography and water contact behaviours (the questionnaire is available upon request to the corresponding author).
After receipt of the second-day stool (and urine sample), all participants, regardless of their infection status, were treated for schistosomiasis and soil-transmitted helminthiasis with PZQ (40 mg/kg) (CIPLA, Mumbai, UK) and 400 mg albendazole (GSK, Uxbridge, UK) under medical supervision in conditions typical of mass-drug administration [30]. For smaller children, a chewable albendazole half-tablet (200 mg) was given and PZQ tablets were first crushed in orange juice before being administrated by spoon-feeding by their mother under supervision. The diagnostic findings for schistosomiasis here are reported for the children only.
Each child's urine sample was visually inspected for macro-haematuria/turbidity and a random sample was tested for micro-haematuria with Hemastix (Bayer, UK) to exclude the possibility of urinary schistosomiasis or other active urinary tract infections. A 50 µl aliquot was then tested for the presence of schistosome circulating cathodic antigen (CCA) using a commercially available lateral flow immuno-chromatographic urine dipstick (Rapid Medical Diagnostics, Pretoria, RSA) originally developed in Holland [31]. On a subset of 90 children, urine-CCA tests were performed in duplicate to assess variation between dipsticks.
To facilitate better recording of the visual intensity of the CCA reaction band within the test zone, results were visually graded against a reference chart for: trace, single (+), double (++) and triple (+++) positive reactions [32]. When creating binomial variables to depict infection status according to CCA, two variations were taken into account: the first considering trace results as negative infection status and the second considering trace results as positive infection status. The urine CCA reagent strip is referred to as an ADM (antigen detection method) from now on.
A commercially available ELISA kit (IVD Inc.; Carlsbad, USA) was used to test for host antibodies (IgG/M) to soluble egg antigens (SEA) according to manufacturer's instructions. Approximately 75 µl of finger-prick blood was taken from each child and serum was harvested, then diluted 1∶40 with specimen dilution buffer before loading a total of 100 µl into each ELISA microwell [11]. Positive and negative control sera were included on each batch of testing. Upon completion, each ELISA plate was placed on a white card and the colour within each microwell (ranging from colourless to yellow) was recorded by visual inspection. Positive reactions were classified either as trace (faint yellow), single (+, light yellow), double (++, yellow) or triple (+++, dark yellow) upon visual comparison with the control sera. The SEA-ELISA is referred to as an IEDM (indirect egg detection method) from now on.
Three parasitological methods Kato-Katz, percoll and FLOTAC, henceforth referred to as direct egg detection methods (DEDMs), were attempted on each stool specimen to visualise eggs. However, owing to the differing amounts of stool required for each technique, it was not always possible to assemble a complete data set for every child with each of these three methods.
Duplicate Kato-Katz (K-K) thick smears (41.7mg) were made from first and second day stool samples (N = 242 children) [33]. The four faecal smears were each examined under the microscope at x100, schistosome eggs were counted and later expressed as eggs per gram (epg) of faeces. Infection intensity was classified as light (1–100 epg), medium (101–400 epg) and heavy (>400 epg) infections according to WHO guidelines [5]. The methodology of Eberl [34] using sedimentation of schistosome eggs by centrifugation through a solution of percoll (Percoll 77237 (1.130 g/ml), Fluka, Sigma-Aldrich Chemie GmbH, Switzerland) was also implemented on-site to visualize eggs (N = 96 children on first day stool). The egg-floatation procedure known as FLOTAC [35] was performed off-site back in Kampala on a formalin-fixed stool specimen archive (N = 191 children taken from the second day stool) whereby schistosome eggs are collected by floatation centrifugation through a solution of zinc sulphate at specific gravity of 1.35.
Data were collected from each individual using pro-forma data sheets, which were then transferred into electronic format using Microsoft Excel. The data thus collated were analysed using MS Excel and R statistical package version 2.8.0 [36]. For prevalence data and diagnostic parameters, 95% confidence intervals (CI95) were estimated using the exact method [37]. Prevalence comparisons were performed using (one-tailed) Fisher's exact modification of the 2×2 chi-squared test [38]. For infection intensity values, the arithmetic mean of positive cases was chosen as the measure of central tendency. Data from the FLOTAC and percoll methods were analysed by combining with K-K results and revising the diagnostic criterion so individuals were considered positive if an egg was detected by at least one DEDM.
The diagnostic performances of the ADM (including and excluding trace reactions as a positive diagnosis) and IEDM were tested qualitatively as a rapid diagnostic for intestinal schistosomiasis, considering DEDMs as the ‘gold-standard’. Additionally, a third ‘gold standard’ was created using data from the ADM (including and excluding trace reactions as positive diagnoses) against which to test IEDM data (N = 242). Diagnostic sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated according to the different ‘gold standards’ [38]. The diagnostic powers of ADM and IEDM were calculated using all individuals, and then segregated by sex or age (≤3 years of age versus >3 years of age). P-values <0.05 were considered indicative of statistical significance [38].
Approvals for this study were granted by the Ugandan Council for Science and Technology and the London School of Hygiene and Tropical Medicine (application numbers 06.45 and 5538.09). After sensitisation of the local community to the study objectives, verbal assent was first requested from each mother which was then formalised upon written informed consent (for her and behalf of her child), as either a thumbprint or signature on data recording sheet. This was witnessed by a Vector Control Division Officer. PZQ treatment (40 mg/kg) was offered to all study participants irrespective of their infection status.
A complete data set for the ADM, IEDM and Kato-Katz examinations was obtained from a total of 242 children (134 boys: 108 girls, mean age 2.9 years, minimum 5 months and maximum 5 years). However, owing to insufficient amounts of stool available, the FLOTAC and percoll methods could only be performed on 191 and 96 children, with the former and latter finding 4 and 2 additional egg-positive cases, respectively.
The prevalence of intestinal schistosomiasis estimated by each diagnostic method, and combinations thereof, is shown in Table 1 and Fig. 1. Prevalence inferred by DEDM, ADM (including trace reactions as positive diagnoses) and IEDM (considering traces as negatives) were: 24.4%, 42.6% and 45.9%. Of the children who were egg-positive by K-K, three quarters had ‘light’ intensity infections. Girls were equally as likely as boys to be diagnosed positively for intestinal schistosomiasis by ADM (Odds Ratio (OR) = 1.06, p = 0.90) and DEDM examinations (OR = 0.72, p = 0.29). Children under the age of three, however, were less likely to be positive by ADM (OR = 0.51, p = 0.016) or by DEDM (OR = 0.26, p<0.0001) than their older counterparts. The prevalence of positives by IEDM was 45.9%. General prevalence inferred by pooling DEDM and IEDM was 47.7%, with no further change in prevalence when ADM was then added, see Fig. 1. There was no discordance between duplicate CCA testing for negative or positive classifications (data not shown).
The age of first positive (AFP) for each method is presented in Table 1. For DEDM, the youngest child with eggs in stool was 9 months old, with medium and heavy infections found at 3 and 5 years of age, respectively. For ADM, trace reactions, single, double and triple positives were found in an ascending series of 6 months, 9 months, 11 months and 2 years of age, respectively. For IEDM, trace reactions began at 5 months of age while single, double and triple positive reactions were found in children as young as 6 months, 1 year and 9 months old, respectively. All tests concur on a mean age of first infection within the third year of life. ADM detected infections slightly ahead of I/DEDMs (3.2 years v. 3.4 years v. 3.7 years, respectively). The order of this temporal series is largely concordant with an absolute minimum age of becoming first positive.
In the absence of a genuine ‘gold standard’ where the infection status of each child is precisely known, it is necessary to explore relationships between diagnostic scores and infection intensities empirically, and to cross-tabulate diagnostic permutations by investigation. There was negligible variation in diagnostic performance of all protocols tested when classifying the data according to sex and age (data not shown) and general trends were reported from now on.
Plotting the relationship between ADM and DEDM revealed some immediate trends, Fig. 2. Whilst there were children positive for ADM who were egg-negative, as the epg increases there was a corresponding increase in the proportion of positive ADM tests and once medium/heavy intensity infections were reached, all ADM tests were clearly positives, see Fig. 2A. Plotting the faecal epg of each child against the intensity of the corresponding ADM test further revealed this positive association, see Fig. 2B.
Considering the relationship between IEDM and DEDM revealed similar trends, see Fig. 3. Despite some children being positive for IEDM while being egg-negative, as the faecal epg increases there was a corresponding increase in the IEDM reaction strength, with all medium/heavy intensity infections diagnosed as clear strong positives (Fig. 3A & B).
The relationship between ADM and IDEM was less clear-cut. Children who were ADM negative or trace had a median negative (or trace) IEDM reaction, but the proportionate increase of ADM positives with rising IEDM designations of positive (+) or strong positives (++/+++) was not as great as that seen with DEDM. For example, nearly 40% of children who were IEDM strong positive elicited a negative ADM reaction, Fig. 4A. As the intensity of the ADM result stepped up towards double and triple positive reactions, this typically corresponded better with increasing IEDM classifications, Fig. 4B.
Using available data it was possible to conduct an exploration of diagnostic performances of the ADM and IEDM versus DEDM and against each other (Table 2). First, considering an ADM trace reaction to be an infection negative and comparing with positive diagnosis by at least one of the DEDMs, the ADM had a sensitivity of 59.3%, specificity of 95.6%, PPV of 81.4% and NPV of 87.9%. When considering an ADM trace reaction to be a positive infection, and comparing to diagnosis by at least one of the DEDMs, ADM had a sensitivity of 81.4%, specificity of 69.9%, PPV of 46.6% and NPV of 92.1%.
The IEDM when compared to diagnosis by DEDM, demonstrated a sensitivity of 93.2%, specificity of 69.4%, PPV of 49.5% and NPV of 96.9%. For details on the performance of the ADM or IEDM compared to diagnosis by all DEDMs (using a subset of the data), and for CI95 around each value, see Table 2.
With limited access to safe water sources, and high levels of local transmission of S. mansoni, conditions in Bugoigo are particularly conducive for young children to acquire S. mansoni infections, and from a very early age. Approximately half of our children had intestinal schistosomiasis. As might be expected, regardless of techniques used, there was an obvious positive association between increasing diagnostic patency of infection with increasing age of the child. Presumably this was resultant from a progressive temporal accumulation of antigens, eggs and antibodies. Congruence between diagnostic methods became most apparent in children between 3¼–3¾ years of age, broadly consistent with the overall mean age of infected children within our sample. Prevalence of intestinal schistosomiasis in children under 3 years of age, however, was 35.5% (CI95 27.9–43.8%) and other studies have also revealed that schistosomiasis in very young children can be common [9], [11], [39].
While egg excretions of these children were of ‘light’ intensity, such infected children will not normally receive praziquantel treatment until they have either entered primary school or if the NCP now formally extends its treatment remit to include this ageclass. Thus an infected child could therefore wait up to 3–4 years before receiving first treatment, and with this may have already entered a more ‘chronic’ stage of disease [40], [41], [42], [43]. For example, earlier clinical and ultrasound studies in Uganda in children aged 6 and above, have shown significant hepatosplenomegaly (i.e. putative morbidity from intestinal schistosomiasis), and while they have not yet developed pipe-stem liver fibrosis, up to 15% can have diffusely echogenic livers with pocketed foci, typical of image pattern B (‘the starry sky’ classification) [3], [27]. Without medication, it is likely that these preschool children will progress towards ‘moderate’ infection intensities before they become of school age. This might better explain the observations of Balen et al. that many adolescent Ugandan children have surprisingly severe intestinal schistosomiasis [44]. Although it is not yet proven that infection in very early childhood leads to heightened morbidity in later childhood and adolescence, this scenario appears plausible.
From animal models, it is known that only a fraction of penetrating cercariae successfully migrate to, and later mature in, the hepatic portal system. After adult worms reach full fecundity, schistosome eggs can be found in stool around 6 weeks after cercarial exposure and it is commonly held that females of S. mansoni produce up to 100–300 eggs per day, although many fail to be voided into the faeces [16]. Given the insensitivity of DEDMs in stool [34], it is not surprising that false negatives are inferred and the low egg-detection threshold(s) likely contribute to the longer apparent lag of 7–8 months between infection and egg-patency apparent between experimental schistosomiasis and the situation in the field. Moreover, it should be noted that the relationship between excreted eggs in stool and worm burdens is not always straightforward [45] and that infected laboratory animals are typically exposed with a single substantive dosing of cercariae. By contrast, and in this natural setting, exposure and infection is likely a more gradual process, i.e. the so-called trickle infection dynamic [46], and our children are at least two orders of magnitude greater in body size than most animal models.
While some children were patently infected during the first year of life, others were not. Thus a sub-set of children exists with increased infection risk factors which we explain by the following synopsis. As children are born throughout the year, in a largely asynchronous fashion, whilst their initial age of first exposure to unsafe water might be broadly similar (i.e. within first few months of life as mothers begin to bathe them in jerry-can collected water or in the lake directly) their accumulated infection risk will not be equivalent owing the seasonality of local transmission factors and their particular timeframe of exposure within it contingent upon their mother's infant bathing and domestic water drawing practices [11].
Estimating this accumulated risk of infection reliably over the seasonal time frame of potential exposure is problematic as day-to-day variations within water collection times, its storage and actual domestic use (within each household) introduce many stochastic processes. Estimating cumulative infection risk is therefore easily confounded but an ad hoc investigation of infection risk associated with jerry-can collected water in June 2009, however, has confirmed that sentinel laboratory-bred mice could become infected to freshly drawn water [28]. Seasonal patterns, which operate in umbrella fashion over and above these specific-exposure patterns, no doubt effect this asynchronous age of first infection. Thus there will be no ‘absolute age’ of first infection but rather a ‘range of ages’ depending upon these intricate covariates of exposure. Only after a child has passed through sufficient ‘windows of exposure’, their probability of infection rises to an eventual certainty, after which, it is incumbent on the diagnostic tools to capture their parasitological status as accurately as possible.
From first appearances the ADM looks to best capture and identify infections in early stage, especially when we consider trace results as putative infection positives. A contentious issue in the use of the CCA reagent strip has been the interpretation of the exact diagnosis of this ‘trace’ result which can be confounded by non-specific inflammatory factors or breast-feeding [32], [47]. Interpretation of ‘trace’ is more contentious when surveying children under three years of age, where worm burdens are presumably lower than what might be expected in their school-aged counterparts. Interestingly, the percentage change in prevalence estimated according to the CCA reagent strip when excluding and including trace results as a positive diagnosis is significantly larger in the very young children (≤3 years of age) –10.1% v. 36.2% (+358%) – than in those aged four and five years of age –30.1% v. 52.7% (+175%) which is fitting with our understanding of increasing worm burdens through time. Thus we postulate that using ‘trace’ as positive firmly points towards a future use of the urine CCA-dipstick as an early indicator of infections which are as yet to become egg- or antibody-patent. It is particularly notable that the prevalence based on the ADM, when considering trace as positive, is very close to that of IEDM (Fig. 1), yet the diagnostic performance with it was not particularly congruent (see Fig. 3 and Table 2) so we still have an incomplete understanding of this infection progression. The dynamics of other ADM have been explored elsewhere in the context of recently acquired infection but not in very young children [48].
The ADM showed very promising diagnostic performance and robust field performance with high sensitivity and NPV scores (83.9% and 85.3%, respectively) when we considered trace results as a positive diagnoses and very high specificity and PPV scores (95.5% and 90.5%, respectively) when we considered trace results as negative diagnoses. This bimodal use of the test criteria could be advantageous from a control perspective. For instance, if a confident estimate of the suspected occurrence of infections within a population is needed, one should consider trace results as positives. On the other hand, to monitor the prevalence of ‘actual’ infection, or rather more easily identify those who do not, one should consider trace results as negatives. The former would be important if treatments were to be given out en masse as triggered by exceeding an aggregated local prevalence threshold while the latter would be important if treatment were to be withheld in an individual patient setting on the basis of test and treat.
With the gradual rise of infection prevalence in older children (over and above our asynchronous infection hypothesis), this trend must represent the spread of several risk factors, rather incipiently, across our cohort. Aggregation of infections in schistosomiasis is well-known [49] but it would be interesting to establish why approximately half of our study cohort had no evidence of infection despite living within the same village. As we were insufficiently aware of the exact locations of sampled households within the village, this could simply represent a cryptic spatial micro-patterning (i.e. these children who live slightly further away from the lake have less contact with viable cercariae) so we are now undertaking fine scale mapping of these individual households with GPS units. If, however, other causal factors could be identified and, perhaps more importantly, were these amenable to manipulation, it could lead to future infection mitigation measures.
Presently, within the NCP there are no health education materials targeted towards these mothers and their young children. More importantly and in terms of policy realignment of the NCP, a useful formal recommendation would be to initiate cross-sectorial activities with water and sanitation NGOs to improve immediately the domestic water quality at Bugoigo and elsewhere along the Lake Albert shoreline. Rather than focusing upon expensive infrastructure development, it could be achieved by introduction of simple water storage or modification measures. For example, as schistosome cercariae are an ephemeral larval stage, freshly drawn water can be rendered harmless for schistosomiasis by simple resting for 24 hrs, by crude filtration or by introduction of mild disinfectants [16]. Thus without initiating a better dialogue with these women of children bearing age through better public health education, mothers will remain sadly ignorant of the risks that making use of this unsafe water has for themselves and that of the future health of their child [11]. In this dialogue, the NCP should be receptive to explore which infection mitigation measures are best feasible and, by this token, help to provide safe water for domestic use which is well-received, implementable and effective.
It is evident that infants and preschool children in Bugoigo, and other similar lakeshore villages of Uganda [12], are living in need of treatment. However, addressing how these children could be best identified is not yet clear, as are epidemiological parameters which should be collected for estimating treatment needs and also impact assessment. For example, should mass-treatment of all infants/preschool children take place when a sub-sample of an equivalent age range has been proven to be infected, or should treatment be allocated based at an individual level using the result of a diagnostic test in a ‘test and treat’ setting?
It is outside the remit of this present paper to make a cost-effectiveness calculation but a clear drawback of the IEDM method is that, whilst initially useful to establish if a child is infected (and there is no evidence in these data to suggest a passive maternal transfer of antibodies has been confounding), monitoring this parameter after treatment will be largely uninformative owing to residual antibody titres remaining after infection has putatively cleared. Thus IEDM is only useful at baseline but as an initial estimate of infection prevalence could be powerfully applied in identification and selection of villages, or sentinel locations, to first define the extent of the problem at intervention baseline. This of course assumes the majority of examined children can mount an antibody response and is not confounded by high levels of immune-suppression, by HIV for example, which is likely high in these fishing villages.
In contrast, both ADM and DEDMs have the potential ability to better track the dynamics of worm populations after treatment [22], [34] but the insensitivity of DEDMs is of particular concern. Put simply, numerous adult worms may reside within the host but are yet not depositing sufficient eggs to be visualised in stool on the day of sampling. Thus through lack of alternatives a pragmatic way forward would be to focus upon more widespread application of the ADM. The advantages of the ADM have been discussed elsewhere in the context of programmatic monitoring [14] but the future challenge will be for the NCP to meet the financial costs of using these rapid diagnostic tests in scale-up of operations. This is particularly true if these are to be used in a ‘test and treat’ setting when large numbers of tests would be utilized [14]. Given the low price of PZQ treatment, to maintain an affordable diagnosis versus treatment differential, a rational strategy would be to examine a sub-set of children and if local prevalence exceeded a given threshold, mass-treatment is advised. Such a strategy is presently within the resources available to the NCP but best sample sizes and prevalence thresholds remain to be determined.
|
10.1371/journal.pbio.1001276 | Role of the Gut Endoderm in Relaying Left-Right Patterning in Mice | Establishment of left-right (LR) asymmetry occurs after gastrulation commences and utilizes a conserved cascade of events. In the mouse, LR symmetry is broken at a midline structure, the node, and involves signal relay to the lateral plate, where it results in asymmetric organ morphogenesis. How information transmits from the node to the distantly situated lateral plate remains unclear. Noting that embryos lacking Sox17 exhibit defects in both gut endoderm formation and LR patterning, we investigated a potential connection between these two processes. We observed an endoderm-specific absence of the critical gap junction component, Connexin43 (Cx43), in Sox17 mutants. Iontophoretic dye injection experiments revealed planar gap junction coupling across the gut endoderm in wild-type but not Sox17 mutant embryos. They also revealed uncoupling of left and right sides of the gut endoderm in an isolated domain of gap junction intercellular communication at the midline, which in principle could function as a barrier to communication between the left and right sides of the embryo. The role for gap junction communication in LR patterning was confirmed by pharmacological inhibition, which molecularly recapitulated the mutant phenotype. Collectively, our data demonstrate that Cx43-mediated communication across gap junctions within the gut endoderm serves as a mechanism for information relay between node and lateral plate in a process that is critical for the establishment of LR asymmetry in mice.
| Superficially, humans, like other vertebrates, are bilaterally symmetrical. Nonetheless, the internal configuration of visceral organs reveals a stereotypical asymmetry. For example, human hearts are generally located on the left and the liver on the right side within the body cavity. How this left-right asymmetry is established is an area of interest, for both intrinsic biological significance and its medical application. In the mouse, the initial event that breaks left-right symmetry occurs at the node, a specialized organ located in the midline of the developing embryo. Somehow this initial asymmetry leads to a cascade of events that results in the activation of a genetic circuit on the left side of the embryo, which then leads to asymmetric organ formation. Here we show that the laterality information that is generated at the node is transferred to the lateral extremity of the embryo across the gut endoderm, which is the precursor tissue of the respiratory and digestive tracts and associated organs such as lungs, liver, and pancreas. Sox17 mutant mouse embryos exhibit defects in gut endoderm formation and fail to establish left-right asymmetry. Analysis of the mutants reveals that gap junction coupling across the gut endoderm is the mechanism of left-right information relay from the midline site of symmetry breaking to the site of asymmetric organogenesis in mice.
| The elaboration of the major axes of the embryo (anterior-posterior, dorsal-ventral, and left-right) occurs at the time of gastrulation, the morphogenetic process that forms the primary germ layers (ectoderm, mesoderm, and definitive endoderm). The molecular definition of a left-right (LR) axis precedes the establishment of an overt asymmetry in that dimension, a common feature across bilateral animals. Despite some species-specific differences, the sequence of key events and core components involved in the establishment of LR asymmetry appears conserved across vertebrates [1],[2].
An initial LR symmetry-breaking event occurs in a specialized organ located at the embryonic midline, this being the node in the mouse. Around embryonic day (E) 7.75, equivalent to the early head-fold (EHF) stage, cilia protruding from the posterior-apical surface of node cells begin rotating, thereby generating directional fluid flow within the node microenvironment [3]. Whether through interpretation of mechanical flow forces [4],[5] or asymmetric distribution of signals [3],[6], nodal flow is believed to trigger a wave of Ca2+ on the left side of the node [4], as well as left-biased asymmetric perinodal expression of Nodal [7]. By E8.5, equivalent to the ∼4 somite stage, an asymmetric readout of this symmetry-breaking event is the activation of the Nodal/Lefty2/Pitx2 genetic cascade within the left lateral plate mesoderm (LPM) [8]. Activation of these factors is required for asymmetric organogenesis [8]. The embryonic midline is thought to act as a “midline barrier” in keeping signals confined to their respective sides, both by virtue of its morphological structure and by expressing specific factors [2].
An unresolved question in the establishment of LR asymmetry in the mouse embryo is the mechanism of communication between the midline site of symmetry-breaking and the lateral plate tissues initiating asymmetric morphogenesis. By virtue of their location, cells that lie between the node and lateral plate are likely to provide the medium for signal relay. The gut endoderm and paraxial mesoderm are attractive candidate tissues for mediating signal transmission since they lie adjacent to both the node and lateral plate mesoderm. Moreover, perinodal asymmetric events, including calcium release and Nodal expression, occur in endodermal cells lining the node [4],[7],[9].
Using live imaging and genetic labeling, we previously noted that the gut endoderm of the mouse embryo forms by widespread intercalation of epiblast-derived definitive endoderm (DE) cells into the overlying visceral endoderm (VE) epithelium [10]. At gastrulation, DE progenitors intercalate into the distally positioned embryonic VE epithelium, also referred to as the emVE [11],[12], which constitutes the surface cell layer of the embryo. This multifocal intercalation of DE cells leads to the widespread dispersal and dilution of emVE cells [13]. As a result, the emergent gut endoderm tissue is comprised of cells of two distinct origins: DE and emVE. Furthermore, we noted that in addition to their scattered distribution within the gut endoderm, residual emVE cells exhibited a second stereotypic distribution in that they were absent from, but congregate around, the node and midline [10],[14],[15]. Even though the precise cellular dynamics orchestrating emVE displacement at the node and midline remain unknown, these data suggest that lateral dispersal in the future gut endoderm and midline displacement in the future notochord may be regulated independently.
The Sry-related HMG box transcription factor Sox17 is a key conserved factor involved in endoderm formation in vertebrates [16]. Mice lacking Sox17 have a depletion of DE cells, possess an abnormal gut tube, and die around E10.5 [17]. Interestingly, Sox17 mutant mouse embryos also display gross morphological features commonly observed in mutants of LR asymmetry establishment, including a failure to turn from a lordotic to a fetal position, an open body wall, and cardiac defects [17]–[19]. We therefore reasoned that a detailed analysis of the Sox17 mutant might provide further insight into the gut endoderm defects and whether gut endoderm morphogenesis and LR patterning are coupled.
Here we report that Sox17 mutant embryos exhibit a failure in emVE dispersal as well as defects in LR patterning. We noted that widespread emVE and DE cell intercalation in the prospective gut endoderm was severely affected in mutants. By contrast, emVE displacement at the midline was not. This suggested that lateral dispersal in the future gut endoderm and midline displacement in the future notochord are likely to be distinct morphogenetic processes, the former of which requires Sox17.
One mode of communication across epithelia is through gap junctions. We identified Connexin43 (Cx43) as the predominant gap junctional constituent expressed in the endoderm at a time soon after widespread emVE and DE cell intercalation is complete, when node to LPM signal relay likely occurs. In Sox17 mutants, we noted that Cx43 is absent in the gut endoderm. We demonstrated gap junctional coupling on both the left and right sides of the gut endoderm of wild-type embryos, but not in Sox17 mutants. We also observed that gap junction coupling in the mesoderm was isolated from the endoderm. Since Cx43 localization within the mesoderm was comparable in wild-type and mutant embryos, we concluded that LR signals must be propagated across the endoderm epithelium. Our studies also revealed an absence of gap junctional coupling across cells at the midline in wild-type embryos, thereby providing the first functional visualization of a midline barrier in the mouse.
Collectively our observations identify the gut endoderm as a key tissue of communication between node and LPM during the establishment of LR asymmetry in the mouse. We demonstrate that Cx43-mediated gap junction coupling across the endoderm is necessary for the correct temporal and spatial propagation of asymmetric signal(s) from the node to the LPM.
Sox17 mutant embryos exhibit a dysmorphic heart, have an open ventral body wall, and fail to turn (Figure S1A–S1D′) [17]–[19]. Since these features are characteristic of mutants with defects in LR patterning [8], they prompted us to determine whether LR asymmetry is established in Sox17 mutants. To do so, we analyzed the expression of components of the core circuitry controlling the establishment of LR asymmetry in mice: genes encoding the TGFβ family proteins Nodal and Lefty2, and the homeodomain protein Pitx2. In E8.5 wild-type embryos, Nodal was expressed around the node and along the left LPM (Figure 1A and 1B). In stage-matched Sox17 mutants, Nodal was present around the node, but absent (three out of eight embryos analyzed) or reduced and restricted posteriorly (five out of eight embryos analyzed) in the left LPM (Figure 1A′–1B″). Sox17 mutants also exhibited ectopic patchy domains of Nodal expression on both the left and right sides. Notably, these ectopic patches of gene expression were not in cells of the mesoderm layer; instead, they were located superficially and were confined to the endoderm layer of the embryo, as analyzed below.
These observations were confirmed using a NodalLacZ knock-in reporter allele [20], where lacZ activity was detected around the node and along the left LPM in NodalLacZ/+ embryos (Figure 1C and 1D). By contrast, in NodalLacZ/+;Sox17−/− embryos, the reporter was present around the node but absent from the left LPM (Figure 1C′ and 1D′). Furthermore, ectopic lacZ expressing cells were observed on both sides of the midline.
Next, we analyzed expression of Lefty1/2, which in wild-type embryos was present in the midline and left LPM (Figure 1E and 1F). In Sox17 mutants, Lefty1/2 was either absent from the left LPM (four out of nine embryos analyzed) or severely reduced and truncated anteriorly and posteriorly (five out of nine embryos analyzed) (Figure 1E′–1F″). As with Nodal, ectopic patches of Lefty1/2 expression were observed on both sides of the midline. Lefty1/2 was detectable in the midline in a subset of the Sox17 mutant embryos analyzed, but this midline domain was truncated anteriorly as compared to wild-type stage-matched embryos (Figure S1E and S1E′). In wild-type embryos Pitx2 was present in the cephalic region and in the left LPM (Figure 1G and 1H). In Sox17 mutants, Pitx2 was present in the cephalic region, but was absent from the left LPM (Figure 1G′ and 1H′). Patches of ectopic Pitx2 expression were also observed bilaterally.
Sections through mutant embryos revealed that the ectopic patches of gene expression comprised cells that were located on the embryo's surface, while in wild-type embryos gene expression was only detected in cells residing in a deeper location within the embryo, within the mesoderm (Figure S1F–S1H′). Sox17 mutant embryos at the early somite (ESom) stage (E8.25), prior to the onset of LPM expression, already exhibited ectopic expression of Nodal, Lefty1/2, and Pitx2 (Figure S1I–S1N′). Thus ectopic gene expression preceded LPM gene expression, and so could be due to a failure in the downregulation of an earlier domain of expression. Analysis of Nodal in early head-fold (EHF) stage (E7.75) Sox17 mutants revealed widespread ectopic patchy expression in the gut endoderm epithelium on the surface of the embryo (Figure S1O–S1Q′). These data demonstrate that Sox17 is required for left LPM expression of the Nodal/Lefty/Pitx2 genetic cascade operating in the establishment of LR patterning in mice, and that absence of Sox17 results in patchy ectopic expression of these genes within the gut endoderm (see summarizing cartoon in Figure 1I and Table S1).
Having confirmed that Sox17 mutants failed to establish LR asymmetry, we investigated whether this phenotype was linked to the endoderm defects that were previously reported [17]. To better understand the progression of gut endoderm morphogenesis in Sox17 mutants, we used the Afp::GFP reporter [21]. The Afp::GFP transgene permits visualization of VE cells (GFP-positive), which can be discriminated from DE cells (GFP-negative) within the gut endoderm on the embryo's surface [10].
Since the VE encapsulates the epiblast and extraembryonic ectoderm prior to gastrulation, Afp::GFP embryos exhibit homogenous widespread GFP fluorescence across the entire surface of the conceptus. By the EHF stage, the cellular movements driving gut endoderm morphogenesis during gastrulation are near complete. Extraembryonic VE (exVE) GFP-positive cells positioned proximally overlying the extraembryonic ectoderm remain homogenous in the region that will form the visceral yolk sac. By contrast, GFP-positive emVE descendents located distally overlying the epiblast become dispersed and are scattered between DE cells on the embryo's surface. We analyzed EHF stage wild-type and Sox17 heterozygous embryos that were also hemizygous for the Afp::GFP reporter and confirmed that the emVE had fully dispersed, as evident by a distally positioned scattered GFP-positive population of cells in the gut endoderm (Figure 2A–2D and unpublished data). By contrast in Sox17 mutant embryos, the distally positioned emVE appeared as a uniform GFP-positive sheet on the ventral surface of the embryo (Figure 2A′–2D′). The only regions in which GFP-negative cells could be identified were around the anterior intestinal portal, the site of foregut invagination, and around the node and midline. Overall, these observations suggested that in Sox17 mutants there was a failure to disperse the emVE within the future mid- and hindgut and that the displacement of emVE at the midline was generally unaffected.
To confirm that the Afp::GFP reporter was functioning as a short-term cell lineage label due to perdurance of GFP (as we noted previously, [10],[14],[15]) and exclude the possibility that GFP was being ectopically expressed by DE cells within the gut endoderm of Sox17 mutants, we determined the localization GFP transcripts in Sox17+/+; Afp::GFPTg/+, as well as Sox17−/−;Afp::GFPTg/+ embryos. In head-fold (HF) stage (E7.75–8.0) wild-type Afp::GFP embryos, GFP was transcribed only in the exVE (Figure S2A). In Sox17 mutants, GFP was only expressed in the exVE (Figure S2A′), indicating that no ectopic expression occurred in cells of the gut endoderm overlying the epiblast. Additionally, the pan-VE marker Apoc2 was undetectable in the gut endoderm of both wild-type and Sox17 mutants (Figure S2B–S2D′). These data suggest that in Sox17 mutants, as in wild-type embryos [10], emVE cells change their state and downregulate expression of VE markers coincident with the intercalation of epiblast-derived DE. The GFP-positive cells that colonized the gut endoderm area of Sox17 mutants were therefore emVE-derived. These cells had undergone an initial step in endoderm morphogenesis, namely cell identity reprogramming, by downregulating emVE markers, and had become poised for integration into the gut tube.
To determine if the failure to disperse the emVE persisted, we analyzed Sox17 mutants at later embryonic stages. At the ∼4 somite stage (E8.5), wild-type embryos and Sox17 mutants were indistinguishable by gross morphology (Figure 2E and 2E′). However, high-resolution examination of the localization of GFP-positive cells in embryos hemizygous for the Afp::GFP transgene revealed an aberrant distribution suggesting severe gut endoderm defects in Sox17 mutants (Figure 2F–2L′). We simultaneously analyzed the distribution of FoxA2, a marker of gut endoderm as well as node and midline cells. Anteriorly, in wild-type embryos the region in the vicinity of the anterior intestinal portal contained sparsely scattered emVE cells (Figure 2F–2H). By contrast in Sox17 mutants, a greater density of emVE cells was observed (Figure 2F′–2H′). Posteriorly, in the prospective hindgut region of wild-type embryos, emVE cells were dispersed, while in the midline emVE cells were displaced and congregated around the node and midline (Figure 2I–2K). By contrast, in Sox17 mutants the epithelium on the ventral surface of the embryo was homogenously GFP-positive and so almost entirely emVE-derived. This suggested a failure in emVE cell dispersal and DE intercalation (Figure 2I′–2K′). However, the node and midline regions appeared as largely GFP-negative areas. This suggested that the morphogenetic movements of gut endoderm morphogenesis (involving emVE dispersal), and midline formation (which we propose involves emVE displacement), can be genetically uncoupled and are likely to be distinct.
Notably, closer scrutiny of the midline structures indicated that in a subset of Sox17 mutant embryos, emergent node areas did contain some emVE-derived cells (Figure 2L–2L′). This indicates that even though gut endoderm morphogenesis and midline formation are likely to be regulated by different mechanisms, the two processes are dependent on each other for correct execution. Collectively, these observations demonstrate that defective gut endoderm morphogenesis in Sox17 mutants occurs due to a failure to disperse the emVE. Moreover, our results reveal that the morphogenetic events that drive gut endoderm and midline morphogenesis are distinct.
To better understand the role of Sox17 during gut endoderm and midline morphogenesis, we determined the localization of Sox17 at the time when these tissues form. Maximum intensity projections of confocal z-stacks of Sox17 immunofluorescence on wild-type embryos revealed that Sox17 protein was always nuclear-localized with a high signal-to-noise ratio. At the late bud (LB) stage (E7.5), when emVE cells were dispersed within the gut endoderm of wild-type embryos, and node and midline structures were emerging and displacing emVE cells [2], Sox17 was detected in both emVE and DE cells within the gut endoderm (Figure 3A–3E, and Movie S1). Anterior views of LB stage embryos revealed the region around the midline comprised mainly of emVE cells displaying low levels of Sox17 (Figure 3F and 3G, and Movie S2). Some GFP-negative patches were present, likely corresponding to first cohorts of anterior primitive streak (APS)-derived node and midline cells having reached the embryo's surface. These cell cohorts were devoid of Sox17. Posterior views of LB stage embryos revealed uniform levels of Sox17 in all gut endoderm cells, including cells overlying the primitive streak (PS) representing the posterior visceral endoderm (PVE) domain (Figure 3H and 3I).
By the EHF stage when the node and midline had emerged (E7.75), Sox17 was present in cells of the gut endoderm, but absent from cells of both the node and the midline (Figure 3J–3M, and Movie S2). At the late head-fold (LHF) stage (E8.0), when the anterior intestinal portal begins its invagination, Sox17 was present throughout the gut endoderm except in regions of the foregut where it had become downregulated, while continuing to be undetectable in cells of the node and midline (Figure 3N–3Q). These data indicate that Sox17 localized to cells of the gut endoderm irrespective of their origin and that it was absent from cells of the node and midline. This suggested that the primary site of action of Sox17 was within the gut endoderm. The observation that Sox17 was absent from node and midline in wild-type embryos suggested that Sox17 is not involved in node morphogenesis per se. Thus, the inability of emVE cells to completely clear the node area in Sox17 mutants was likely a secondary effect resulting from the failure of the emVE to disperse.
Since the node is the key symmetry-breaking organ, we further investigated the cells of the node and their behavior in Sox17 mutants. This would reveal if any emVE cells failing to clear the vicinity of the node might be responsible for the LR asymmetry phenotype. We therefore determined if nodal cells contained cilia, if these cilia were motile, and if they could generate nodal flow. Furthermore we determined if asymmetric perinodal gene expression was induced around the node of Sox17 mutants.
We analyzed the distribution of Arl13B, a small GTPase that localizes to cilia [22], in wild-type and Sox17 mutant embryos carrying the Afp::GFP VE-reporter transgene at the EHF stage (E7.75), when nodes are fully formed [22]. In wild-type embryos, robust Arl13B-positive puncta were detected in cells within the node (Figure 4A–4C). We interpreted these puncta as representing the elongated cilia present on the apical surface of cells of the node. In Sox17 mutant nodes, Arl13B-positive puncta were also present, except on emVE cells that had remained in the node region (Figure 4A′–4C′). Analysis of confocal z-stacks revealed an absence of robust Arl13B-positive puncta in cells that remained submerged and thus covered by emVE cells (unpublished data). Scanning electron microscopy (SEM) permitted the high-resolution visualization of nodal cilia in wild-type embryos (Figure 4D–4G). In the Sox17 mutant, cilia with normal morphology protruded from the posterior-apical surface of node cells, except from the larger cells that resembled emVE (Figure 4D′–4G′). Quantitation revealed that the average length of nodal cilia at the EHF stage (E7.75) was comparable in wild-type and mutant embryos (Figure 4H).
Having established that cilia were present on the nodes of mutants, we investigated their motility through high-speed, high-contrast DIC imaging. At the LHF stage (E8.0), nodal cilia of wild-type embryos and Sox17 mutants moved in a comparable clockwise motion (Movie S3). To determine if these motile cilia generated directional nodal flow, we visualized the movement of fluorescent latex beads placed on the apical surface of nodes [23]. Beads invariably migrated leftward in the nodes of both wild-type and mutant embryos (Movie S4).
Nodal flow results in left-biased perinodal gene expression asymmetries necessary to establish LR asymmetry within the LPM [7]. To determine if Sox17 mutants generated perinodal asymmetries in gene expression, we determined the expression of Nodal, Cerl2, and Gdf1, three genes asymmetrically expressed around the node at the ∼2–4 somite stage. Nodal expression [7] was slightly higher on the left side of the node, in wild-type and Sox17 mutants (Figure 4I and 4I′). In both wild-type and mutant embryos, Cerl2 expression [24] was elevated on the right side of the node (Figure 4J and 4J′), while Gdf1 expression [25] was elevated on the left side (Figure 4K and 4K′). We digitally quantified in situ hybridization signal intensities and confirmed that similar ratios of perinodal asymmetries were generated in wild-type and mutant embryos (Figure 4L–4N). These findings demonstrate that, even though in some Sox17 mutant embryos emVE cells failed to completely clear the area of the node, the node that formed was comparable to wild-type in shape, possessed nodal cilia with normal morphology and motility, generating nodal flow, and was able to induce asymmetric perinodal gene expression.
Having determined that Sox17 mutants exhibited asymmetric perinodal gene expression, we reasoned that an event downstream of the node must cause the laterality defect observed in the LPM. This implicated the gut endoderm or paraxial mesoderm, the two tissues positioned between the node and LPM, as potentially responsible for signal relay between these two distant sites. Taking into consideration that the asymmetries in the node region occurred in perinodal cells of endodermal origin [4],[7],[9], and since the Sox17 was specifically expressed by gut endoderm cells, and mutants appeared to exhibit defects specific to the gut endoderm, we favored the gut endoderm as the tissue involved. Defects within the gut endoderm might result in perturbed communication across the epithelium lying between node and lateral plate, and consequently result in a failure in the establishment of LR asymmetry.
Gap junction communication has been implicated in signal relay between the site of symmetry-breaking and tissues of asymmetric morphogenesis in several organisms, including frog, chick, and rabbit [26]. Our previous studies in the mouse demonstrated that cell-cell junctions dynamically disassemble and reassemble during emVE dispersal and concomitant DE cell intercalation takes place during gut endoderm morphogenesis [10]. We therefore investigated the presence and distribution of gap junctions within the gut endoderm of wild-type embryos and Sox17 mutants.
Since connexins are core gap junction components, we determined which connexins were expressed in the mouse embryo around the time of LR asymmetry establishment. Expression profiling of embryonic regions of EHF stage (E7.75) wild-type embryos revealed that of the 19 characterized connexin genes, all of which were present on the array, only two were expressed above background levels (Figure S3A–S3C). The most abundant was Gja1, the gene encoding Connexin43 (Cx43).
We investigated the localization of Cx43 in wild-type embryos and Sox17 mutants carrying the Afp::GFP VE-reporter transgene. Cx43 immunofluorescent localization was generally observed as puncta located at cell-cell interfaces. We interpret these puncta as representing gap junction complexes. Up until the onset of emVE dispersal, the distribution of Cx43 was comparable in wild-type and Sox17 mutant embryos. Notably, Cx43 puncta were observed in all tissue layers: throughout the visceral endoderm, in the extraembryonic ectoderm, as well as within the epiblast and mesoderm (Figure S4A–S4L′ and unpublished data). By the EHF stage (E7.75), when the emVE had dispersed, Cx43 puncta were detected within exVE and gut endoderm of wild-type embryos (Figure 5A–5H). However, in Sox17 mutants, while Cx43 puncta were detected in exVE (Figure 5A′–5D′), they were absent in the undispersed emVE (Figure 5E′–5H′). To determine if this absence of Cx43 localization persisted, we analyzed embryos at later embryonic stages. At the ESom stage (E8.25), posterior views of wild-type embryos showed Cx43 puncta in the node, midline, mesoderm, yolk sac, and all gut endoderm cells, regardless of their origin (emVE or DE) (Figure 5I–5P and Movie S5). In Sox17 mutants, Cx43 puncta were present in the node, midline, and yolk sac, but notably were absent from the gut endoderm epithelium (Figure 5I′–5P′ and Movie S5). Collectively, these observations reveal that puncta comprising Cx43, the major gap junction component expressed in embryos of these stages, were not present amongst cells of the gut endoderm in Sox17 mutants, suggesting a specific absence of gap junctions in this tissue.
These observations prompted us to examine gap junction communication between cells of the gut endoderm in wild-type and Sox17 mutant embryos. To assay for gap junctional coupling, we performed single-cell iontophoretic dye injections into living embryos (Figure 6A and 6B). Two dyes, Neurobiotin and Alexa568, were co-injected into individual endoderm cells on the embryo's surface. While the high molecular weight Alexa dye remained confined within the injected cell, the lower molecular weight tracer Neurobiotin propagated intercellularly by coupling specifically through gap junctions [27],[28].
Injections into gut endoderm cells of wild-type Afp::GFP VE-reporter expressing embryos resulted in Neurobiotin dye propagation across several cell diameters within the gut endoderm epithelium (Figure 6C–6I). Neurobiotin propagation occurred regardless of whether an emVE or DE cell was injected. Moreover, the propagation of Neurobiotin occurred irrespective of whether cells situated on the right or left side of the embryo were injected (Figure 6J–6L). Interestingly, even though we observed Cx43 puncta in the node and midline of wild-type embryos, Neurobiotin never propagated from the gut endoderm to cells of the node and midline, and so never crossed the midline between the left to the right side of gut endoderm (Figure 6J–6L and Movie S6). The Neurobiotin signal also did not couple from the injected endoderm to adjacent mesoderm cells (Figure 6H–6I). Thus, gap junction communication within the gut endoderm was uncoupled from other germ layers and was isolated between left and right sides of the gut endoderm by a midline barrier in wild-type embryos.
We then performed dye tracer experiments on Sox17 mutants. When individual gut endoderm cells in Sox17 mutants were injected, they always retained both the Neurobiotin as well as the Alexa dye (Figure 6M–6Q). We consistently failed to detect dye coupling irrespective of the side of the embryo being injected (Figure 6R–6V and Movie S7). This suggested that in Sox17 mutants, gap junction cell-cell coupling did not occur across emVE cells, likely due to the absence of Cx43 and resulting lack of functional gap junctions.
As an independent test of whether gap junction communication was required for the establishment of LR patterning in mice, we investigated whether its pharmacological inhibition might repress signal transfer from the embryonic midline (the node) where symmetry is broken, to the lateral plate, where it is affected, and in doing so affect expression of LR asymmetry markers in the LPM. Embryos from the EHF-LHF stage (E7.75–E8.0) were cultured until the ∼4 somite stage in the presence of gap junction inhibitors (Figure 7A). To ensure that inhibitors did not interfere with node morphogenesis, we selected only embryos in which the node had emerged to the surface of the embryo. To ensure that inhibitors were acting in the temporal window during which node to lateral plate signal relay must occur and were effective before LPM gene activation, we selected only embryos in which somites had not yet formed. Two inhibitors were used: Mefloquine hydrochloride, which selectively blocks Cx36 and Cx50 [29], and 18 alpha-Glycyrrhetinic acid, a general gap junction blocker [30]. After overnight culture in the presence or absence of specific inhibitors, Lefty1/2 expression was assayed.
Embryos cultured in the absence of inhibitors exhibited wild-type Lefty1/2 expression (Figure 7B and 7C). By contrast, the majority of embryos cultured in 18 alpha-Glycyrrhetinic acid exhibited no Lefty1/2 signal (10/14) (Figure 7D and 7E). Only a small subset of embryos exhibited weak Lefty1/2 signal limited to posterior regions of the left LPM (3/14) (Figure 7F and 7G). Embryos cultured in Mefloquine hydrochloride exhibited wild-type Lefty1/2 expression (Figure 7H and 7I). Embryos cultured in the presence of either inhibitor show normal cilia movement in the node and are able to create perinodal asymmetries (unpublished data). These observations indicated that blocking gap junction function at the time when LR asymmetry is established prevented correct LR patterning, and so molecularly recapitulated the mutant phenotype (Table S1).
A cascade of events establishes LR asymmetry in mice. A central unresolved question in this process is the nature of the step between symmetry-breaking at the midline and the tissues executing asymmetric morphogenesis at the lateral plate. To date, mouse mutants exhibiting LR patterning defects fall into three categories based on their expected site of gene function: the node, the LPM, and the midline [8]. No mutant affecting LR asymmetry has been reported to act within the gut endoderm, a tissue situated between the site of symmetry-breaking (the node) and the effector tissue of asymmetric morphogenesis (the lateral plate). Our studies reveal Cx43-mediated communication through gap junctions across the gut endoderm epithelium as a mechanism for information relay between node and LPM in the establishment of LR asymmetry in mice (for model, see Figure 8).
Analysis of the Nodal/Lefty/Pitx2 cascade of asymmetrically expressed genes indicated that LR asymmetry was not correctly established in Sox17 mutants. Pitx2, the gene downstream in the pathway, was never detected in the left LPM of Sox17 mutants. Interestingly, our data revealed some variability in the expression of asymmetry genes higher up in the cascade in embryos lacking Sox17. While many Sox17 mutants exhibited no expression of Nodal and Lefty1/2 in the LPM, others exhibited reduced and regionally restricted expression of these genes within the left LPM. In wild-type embryos, expression of Nodal in the LPM starts in a small region at the level of the node and subsequently expands anteriorly and posteriorly within the LPM [7]. The observation of reduced Nodal and Lefty1/2 expression in the LPM of some Sox17 mutant embryos revealed that the pathway had been activated, suggesting some signal was being relayed from the node to the LPM.
This raised the possibility that minimal emVE cell dispersal, observed in some mutants, may have been sufficient for gap junction communication. Dye coupling tracer experiments argued against the possibility of gap junction coupling in the Sox17 mutant, because Neurobiotin was never observed to propagate between cells within the gut endoderm of Sox17 mutants. Nonetheless, since it was not technically feasible to inject every cell within the gut endoderm epithelium of a single embryo, we cannot rule out the possibility that occasional gap junctional coupling may have occurred in a minor population of endoderm cells in some Sox17 mutants. Alternatively, another connexin may partially compensate for the absence of Cx43 in Sox17 mutants. Our expression profiling did identify Cx31, another connexin gene expressed at EHF stages, albeit at low levels (Figure S3). An upregulation of compensatory connexin(s) within the gut endoderm of Sox17 mutants might permit some communication between node and LPM, allowing sufficient signal transmission to occur for minimal activation of the pathway in the left LPM. The levels of signal might not be robust enough to fully activate the cascade, preventing it from spreading within the LPM. Indeed, when wild-type embryos were cultured in the presence of inhibitors of all connexins, we observed a complete failure to activate LPM gene expression.
Our analysis of LR asymmetry markers also revealed the presence of ectopic patches of expression on both sides of the midline of Sox17 mutants, with ectopically expressing cells situated on the embryo's surface. Notably, we never observed such ectopic expression in wild-type embryos that had been cultured in the presence of gap junction inhibitors, which we added at the HF stages, suggesting they originated from a defect preceding the emergence of the node and subsequent events of LR asymmetry establishment. In support of this, expression analysis in EHF stage Sox17 mutants already revealed ectopic expression of Nodal in the endoderm. In PS stage wild-type embryos, Nodal is expressed in the entire emVE and then becomes downregulated [11]. This downregulation might not occur correctly in Sox17 mutants, possibly as a consequence of the emVE not being dispersed. The ectopic Nodal patches are subsequently likely to induce Lefty2 and Pitx2.
Even though some of the features have been documented [31], the precise cellular behaviors that drive node and midline morphogenesis, and in particular regarding the emergence of cells onto the embryo's surface, remain to be elucidated. It has been proposed that groups of node precursor cells emerge from the APS and insert into the emVE epithelium at the distal tip of the conceptus [31]. Once on the embryo's surface, cohorts of node cells gradually coalesce to form a single node pit [2], thereby providing a mechanism for collectively displacing emVE cells from the midline.
The absence of Sox17 from the node and midline suggested that this transcription factor is not directly involved in node and midline formation. Accordingly, the node and midline structures emerged onto the embryo's surface in Sox17 mutants. This suggested that morphogenesis of the gut endoderm and node/midline are distinct processes and that they are uncoupled in the absence of Sox17.
We therefore interpret that failure to clear all emVE cells from the node region in a subset of mutant embryos as a secondary defect resulting from a failure in emVE dispersal. A completely emVE-derived gut endoderm might confine emVE cells to the midline, thereby hampering emergence of node progenitors onto the embryo's surface. Notably, the presence of residual emVE cells within the node of mutants did not disrupt leftward nodal flow and subsequent induction of asymmetric perinodal gene expression.
The observation that gut endoderm of Sox17 mutants lacked Cx43 puncta focused our attention on gap junctions as the potential cause for the failure in LR asymmetry establishment. We previously demonstrated that dynamic and widespread rearrangements of cell-cell junctions occur during the intercalation of DE cells into the overlying emVE layer during gut endoderm morphogenesis [10]. Prior to the onset of gastrulation, Cx43 puncta were detected at cell-cell interfaces within the emVE of Sox17 mutants. Since in Sox17 mutants the emVE did not become dispersed and would not need to be rearranged to accommodate intercalation of DE cells, we expected the Cx43 puncta to be maintained in the mutant endoderm. However, Sox17 mutants displayed no Cx43 puncta within the endoderm between the no bud (OB) and EHF stages.
LR defects have not been described in mutants lacking Cx43. The original study characterizing a Cx43 knockout mouse strain reported that mutants survive to term, dying at birth from heart defects [32]. Moreover, subsequent work revealed genetic background-related differences in the phenotypes resulting from conditional Cx43 ablations [33]. It remains to be determined whether Cx43 mutant animals exhibit features representing a failure of LR patterning, including situs inversus or heterotaxia.
Even though cells comprising the gut endoderm of Sox17 mutants were almost exclusively of emVE origin, our findings argued that they complete an initial step in gut endoderm morphogenesis by downregulating markers of VE identity, as do emVE cells in wild-type embryos prior to their dispersal [10]. Thus in the absence of DE cells within the gut endoderm epithelium of Sox17 mutants, a new wave of gap junction formation may have failed to occur. Further studies will be required to determine whether genes encoding gap junction components are direct targets of Sox17 or whether a failure in epithelial remodeling affects Cx43 localization. In either case, we reasoned that absence of Cx43 puncta within the gut endoderm caused a failure in gap junction coupling across this tissue in Sox17 mutants.
Our findings revealed that gap junction coupling occurs between gut endoderm cells and that this mode of communication propagates signals in both the left and right sides of the embryo. Gap junction coupling occurred across several cell diameters extending from a perinodal location to the lateral plate. Since coupling occurred between DE and emVE cells, cells of two distinct origins intercalate to form a congruent epithelium comprising the gut endoderm. Gap junction communication in the gut endoderm is planar and isolated from surrounding tissues. The observation that the gut endoderm was capable of gap junction communication on both the left and the right sides suggested that it acts as a passive medium for relay of asymmetric LR information. This is in accord with the fact that in certain mutants having defects in LR asymmetry, expression of genes in the LPM can be right-sided or bilateral [8]. If the node region, as a result of perturbed nodal flow, dictates to send asymmetry signals to the right or in both directions, the unbiased gut endoderm obediently propagates the signal to the right or to both sides. Furthermore, by revealing a lack of gap junctional coupling across the midline, our experiments also provide the first functional visualization of a midline barrier in the mouse. Importantly, the fact that some Sox17 mutants exhibited limited expression of Nodal and Lefty1/2 in the left LPM suggests that the LPM is responsive to LR patterning signals. Thus, we conclude that the LR asymmetry defect in Sox17 mutants lies in perturbed LR signal propagation. Notably, these observations have also been noted by Saijoh and colleagues, who have gone on to demonstrate that the left LPM of Sox17 mutants is responsive to LR patterning signals from Nodal-source cells (Y. Saijoh et al., unpublished work, personal communication).
The identity of the transmitted signal(s) is unknown: the signal that relays information must be produced in the vicinity of the node, transferred across the gut endoderm, which when reaching the interface with LPM must trigger the Nodal/Lefty/Pitx2 cascade. One molecule shown to travel through gap junctions is calcium [34]–[36]. LR asymmetric localization of calcium has been reported in fish [37], chick [38], and mice [4],[39]. The two-cilia model of left-right asymmetry establishment in the mouse proposes that non-motile mechano-sensory perinodal cilia detect flow and elicit an asymmetric influx of Ca2+ ions through activity of PKD, a calcium-permeable ion channel [4],[5],[40]. Interestingly, mutants that fail to induce asymmetric calcium levels do not express LR asymmetry genes in the LPM [4],[40]. Another candidate is serotonin [41]. Serotonin has been demonstrated to cross gap junctions [42], and disruption of serotonergic signaling leads to aberrant LR patterning in frog and chick [43],[44]. Additional candidates likely exist, and further work will be required to identify the factors that travel through gap junctions across the gut endoderm to mediate LR asymmetry establishment in mice.
Mouse strains used were: Sox17cKO/cKO [45], Sox2::Cre [46], NodalLacZ/+ [20], Afp::GFP [21], and wild-type ICR (Taconic). The Sox17cKO strain was used to generate the null allele by crossing to the Sox2::Cre strain. Work on mice was subject to approval by, and carried out in accordance with guidelines from, the MSKCC IACUC.
Embryos were dissected in DMEM-F12 (Gibco)/5% FCS (Lonza) and staged according to Downs and Davies [47]. For ex utero cultures EHF-LHF stage embryos were roller cultured in 50% rat serum/50% DMEM-F12 [48], a gaseous mixture of 5% O2; 5% CO2, and 90% N2 [49]. Inhibitors used were: 18 alpha-Glycyrrhetinic acid (1∶1,000, Sigma) and Mefloquine hydrochloride (1∶5,000, Sigma). Post-culture embryos were processed for in situ hybridization.
For in situ hybridization (ISH), embryos were fixed in 4% PFA/PBS overnight at 4°C, dehydrated, and stored at −20°C. ISH was performed using antisense riboprobes [50] and standard protocols [51]. β-galactosidase staining was performed according to standard protocols [51]. For vibrating microtome sections of stained embryos, samples were placed in 30% sucrose/PBS overnight at 4°C, transferred into 0.4% gelatin/14% BSA/18% sucrose/10% glutaraldehyde, and sectioned at 16–20 µm (VT1000S, Leica).
Immunofluorescent (IF) staining was carried out as previously described in [10]. Antibodies used were: Arl13B (1∶300, gift of K. Anderson, MSKCC), Connexin43 (1∶300, Santa Cruz), FoxA2 (1∶1,000, Abcam), and Sox17 (1∶1,000, R&D Systems). Secondary Alexa-Fluor conjugated antibodies (Invitrogen/Molecular probes) were used at 1∶1,000. DNA was visualized with Hoechst 33342 (5 µg/mL, Invitrogen). For cryosectioning, fixed embryos were taken through a sucrose gradient, embedded in O.C.T. (Tissue-Tek), and sectioned at 12 µm (CM3050S, Leica).
Widefield images were collected with Axiocam MRc, MRm, or HSm CCD cameras (Zeiss) on a Leica MZ165FC. Laser scanning confocal images were acquired on a LSM510 META (Zeiss) as previously described [10],[15]. Fluorescence was excited with: 405 nm diode laser (Hoechst), 488 nm Argon laser (GFP), 543 nm HeNe laser (Alexa-543/555/568), and 633 nm HeNe laser (Alexa-633/647). Images were acquired using Plan-Apo 20×/NA0.75 and Fluar 5×/NA0.25 objectives. Optical sections ranged between 0.2 and 2 µm. Data were processed with AIM software (Zeiss) and assembled in Photoshop CS4 (Adobe). 3-D reconstructions of confocal z-stacks are depicted as maximum intensity projections (MIPs). ISH quantitations were performed with the Gel Analyzer tool (ImageJ, NIH). For scanning electron microscopy (SEM), embryos were prepared as previously described [39] and imaged with a Field Emission Supra 25 (Zeiss).
Fluorescent latex beads (0.5 µm diameter, Sigma) were placed over the node of embryos positioned ventral side up in 1% agarose submerged in culture medium. Beads were time-lapse imaged and tracked/annotated using the Manual Tracking plug-in (ImageJ, NIH). The high speed imaging system used (Zeiss HRm CCD camera, mounted on a Leica M165FC microscope and operated with Zeiss Axiovision software) did not, however, permit analysis of the relative speeds of beads being imaged.
EHF stage embryos were bisected along the extraembryonic-embryonic junction and the embryonic portion placed in TRIzol (Invitrogen) (N = 3). Triplicates were hybridized to Mouse-6 Illumina arrays (Illumina Inc), and data were analyzed with Partek Genomics Suit (Partek Inc). These data are MIAME compliant and have been deposited in NCBI's Gene Expression Omnibus (GEO) [52], where it is publicly accessible through the accession number GSE33353.
LHF/ESom stage embryos were placed ventral side up and held in place with a metal mesh in artificial cerebral spinal fluid containing (in mM): 125 NaCl, 2.5 KCl, 1.25 KH2PO4, 1 MgCl2, 2 CaCl2, 25 NaHCO3, 1.3 ascorbate, 2.4 pyruvate, and 25 glucose (gassed with 95% O2 and 5% CO2) at room temperature. An Olympus BX51WI equipped with epifluorescence illumination, a CCD camera, and two water immersion objective lenses (UMPlanFI 10×/0.30W and LUMPlanFI/IR 60×/0.90W, Olympus) was used to visualize and target recording electrodes to cells. Glass recording electrodes (10–15 MΩ resistance) were filled with (in mM, pH 7.25): 130 potassium gluconate, 16 KCl, 2 MgCl2, 0.2 EGTA, 10 HEPES, 4 Na2ATP, 0.4 Na3GTP, 0.2% Alexa-568 (Invitrogen), and 0.5% Neurobiotin (Vector Laboratories). After obtaining a whole-cell patch recording, Alexa-568 and Neurobiotin were iontophoretically ejected through the recording electrode using anodal current repeated on (350 ms) and off (250 ms) for 420 s. Immediately after, embryos were placed in DMEM-F12 (Gibco)/5% FCS (Lonza) and live imaged.
|
10.1371/journal.pntd.0002520 | Epidemiological Trends of Dengue Disease in Brazil (2000–2010): A Systematic Literature Search and Analysis | A literature survey and analysis was conducted to describe the epidemiology of dengue disease in Brazil reported between 2000 and 2010. The protocol was registered on PROSPERO (CRD42011001826: http://www.crd.york.ac.uk/prospero/display_record.asp?ID=CRD42011001826). Between 31 July and 4 August 2011, the published literature was searched for epidemiological studies of dengue disease, using specific search strategies for each electronic database. A total of 714 relevant citations were identified, 51 of which fulfilled the inclusion criteria. The epidemiology of dengue disease in Brazil, in this period, was characterized by increases in the geographical spread and incidence of reported cases. The overall increase in dengue disease was accompanied by a rise in the proportion of severe cases. The epidemiological pattern of dengue disease in Brazil is complex and the changes observed during this review period are likely to have been influenced by multiple factors. Several gaps in epidemiological knowledge regarding dengue disease in Brazil were identified that provide avenues for future research, in particular, studies of regional differences, genotype evolution, and age-stratified seroprevalence.
PROSPERO registration number: CRD42011001826.
| Dengue disease is the most prevalent arthropod-borne viral disease in humans and is a global and national public health concern in Brazil. We conducted this review to consolidate and describe the existing evidence on the epidemiology of dengue disease in Brazil, between 2000 and 2011, to gauge the recent national and regional impact of dengue disease and provide a basis for setting research priorities and prevention efforts. We used well-defined methods to search and identify relevant research, according to predetermined inclusion criteria. Despite control measures, the increased territorial distribution of the mosquito vector and the co-circulation of multiple dengue virus serotypes have resulted in increases in the incidence and distribution of dengue disease. The number of disease-related hospitalizations and deaths has also increased. Efforts to control the increasing disease incidence have been unsuccessful. This review of dengue disease epidemiology will help enhance knowledge and future disease management. Despite the high volume of research retrieved, we have identified several avenues for future research, in particular studies of regional differences, genotype evolution and age-stratified seroprevalence that will improve our knowledge of dengue disease, contribute to a more accurate estimate of global disease incidence, and also inform evidence-based policies for dengue disease prevention.
| Dengue disease is an escalating public health problem [1]. Approximately 2·5 billion people live in over 100 endemic countries, predominantly in tropical areas where dengue viruses (DENV) can be transmitted [2]. DENV are arboviruses that are transmitted to humans by infected Aedes aegypti (Linnaeus) mosquitoes – the primary vector. Infection with any one of four DENV serotypes (DENV-1, -2, -3, or -4) can produce a spectrum of illness ranging from a mild, non-specific febrile syndrome, to classic dengue fever (DF), or severe disease forms, such as dengue haemorrhagic fever (DHF) and dengue shock syndrome (DSS), that can be fatal. The World Health Organization (WHO) estimates that >50 million dengue infections and >20,000 dengue-related deaths occur annually [1], [3], [4]. A recent disease distribution model has estimated there to be 390 (95% credible interval 284–528) million dengue infections per year, of which 96 million are apparent (i.e., cases manifest any level of clinical or sub-clinical severity) [3]. During 2001–2007, >4 million cases were notified in the Americas, and during 1995–2002, >75% of these cases were reported from Brazil [5], [6].
Ae. aegypti was eradicated from Brazil as a result of a Pan American Health Organization (PAHO) programme to control the spread of yellow fever. Additionally, DENV transmission was also suppressed in the Americas during the eradication programme. South American countries became re-infested with Ae. aegypti after the programme was discontinued and this, combined with the co-circulation of multiple DENV serotypes, led to the spread of dengue disease across the continent [5], [7]–[9].
In 1982, there was a dengue outbreak in a small city in the northern region of Brazil (Boa Vista/Roraima), which was quickly brought under control and the virus did not spread [10]. In 1986, the re-emergence of DENV-1 in Rio de Janeiro state [11] resulted in over 60,000 reported cases in 1987 and the subsequent spread of DENV increased national public health concerns [12]–[14]. Since the late 1980's the incidence of dengue disease continued to increase; 204,000 cases were reported nationally in 1999 [15], [16]. By 2000, DENV transmission was reported in 22/27 Brazilian states, and the mosquito vector was present in all states [17].
Much of Brazil is affected by a tropical wet and dry climate with high temperatures, high humidity and seasonal variations in rainfall; climate patterns that can provide appropriate conditions for breeding and survival of the Ae. aegypti mosquito. The country is divided into five regions (North, Northeast, Central-West, Southeast, and South) comprising 26 states and the federal district containing the capital city, Brasília. In 2000 there were nearly 170 million inhabitants of Brazil, increasing to more than 190 million in 2010 [18], the majority of whom live in the large cities of the Southeast and Northeast regions [19].
The National System for Surveillance and Control of Diseases (SNVS) of Brazil, operates as part of the national health system (Sistema Único de Saúde, or SUS). All reported cases from public health services or private health providers are included in the notification database (Sistema de Informacoes de Agravos de Notificacao [SINAN]), which is openly accessible via the internet [20]. Until 2011, the SNVS adopted the case definitions outlined in WHO guidelines [21], [22]. In 1997, the WHO categorized symptomatic dengue disease as: undifferentiated fever, DF and, DHF [21]. DHF was further classified into four severity grades, with grades III and IV being defined as DSS. However, difficulties in applying the criteria for DHF [23], led the WHO to suggest a new classification based on levels of severity: non-severe dengue disease with or without warning signs, and severe dengue disease [22]. During 2000–2011, both surveillance and hospitalization reporting systems in Brazil used DF and DHF; the surveillance system used an additional classification designated ‘DF with complications’ (DFC) [24]. Importantly, the articles included in this literature analysis that were based on secondary data used these surveillance sources.
Our objectives of this literature search and analysis were to describe the epidemiology of dengue disease (national and regional incidence [by age and sex], seroprevalence and serotype distribution and other relevant epidemiological data) in Brazil during 2000–2011, and to identify gaps in epidemiological knowledge requiring further research.
A literature review group, including authors of this contribution, developed a literature survey and analysis protocol based on the preferred reporting items of systematic reviews and meta-analyses (PRISMA) guidelines [25]. Our protocol prescribed well-defined methods to search, identify, and select relevant research, and set predetermined inclusion criteria. The protocol was registered on PROSPERO, an international database of prospectively registered systematic reviews in health and social care managed by the Centre for Reviews and Dissemination, University of York (CRD42011001826: http://www.crd.york.ac.uk/prospero/display_record.asp?ID=CRD42011001826; protocol: http://www.crd.york.ac.uk/PROSPEROFILES/1826_PROTOCOL_20130401.pdf) on 9 December 2011.
Between 31 July 2011 and 4 August 2011, we searched databases of published literature (Table 1) for epidemiological studies of dengue disease in Brazil. Search strategies for each database were described with reference to the expanded Medical Subject Headings (MeSH) thesaurus, encompassing the terms ‘dengue’, ‘epidemiology’, and ‘Brazil’. Google and Yahoo searches (limited to the first 50 results) were used to identify national and international reports and guidelines, congress abstracts, and grey literature (e.g., Ministry of Health data, lay publications).
To reduce selection bias, peer-reviewed contributions in English, Portuguese, or Spanish published between 1 January 2000 and 4 August 2011 were included; no limits by sex, age, ethnicity of study participants, or by study type were imposed. Single-case reports and articles only reporting data prior to 1 January 2000 were excluded. Unpublished reports were included if they were identified in one of the sources listed above. Data from grey materials supplemented that from peer-reviewed literature. Publications not identified in the target databases by the search strategy (e.g., locally published papers) and unpublished data sources meeting the inclusion criteria (e.g., theses, Ministry of Health data) were included if recommended by members of the literature review group. Editorials and data from literature reviews of previously published peer-reviewed studies were excluded.
Duplicates and articles not satisfying the inclusion criteria were removed following review of the titles and abstracts. A further selection was made based on review of the full text from the first selection of references. Included publications were summarised using a data extraction instrument developed as a series of spreadsheets. Due to the expected heterogeneity of eligible studies in terms of selection, and number and classification of cases, a meta-analysis was not conducted. For the purposes of the analysis we defined national epidemics as those years with an incidence/100,000 above the 75th percentile for the period. A trend analysis was conducted on the national incidence and case number data.
We identified 714 relevant citations, 51 of which met the inclusion criteria and were entered into the data extraction instrument (Figure 1; Table S1).
During the period 2000–2010, the incidence of dengue disease in Brazil varied substantially, reaching a peak in 2010 of >1 million cases (538/100,000 inhabitants) and the lowest value was approximately 72,000 cases in 2004 (63.2/100,000 inhabitants) (Table 2, Figure 2A–C, Table S2) [6], [15], [16], [26]–[31]. Despite the yearly variations and cyclical epidemics, trend analysis of the incidence of dengue in Brazil in the period 2000–2010 showed an overall increase in incidence over time that was not statistically significant (β = 12·9/cases per 100,000, p = 0·49). Analysis of the number of cases of dengue disease over the review period shows a growth trend that was not statistically significant (β = 47·984 cases/year, p = 0·25). Nevertheless, the trend analysis suggests a worsening of the problem over time.
There were three national epidemics (years with incidence above the 75th percentile for the period [279.95]) in 2002, 2008 and 2010. In 2002 there were 684,527 to 794,219 probable cases of DF, in 2008, 637,663 to 806,036 cases [16], [26], [27], and in 2010 there were over 1 million reported cases (Table 2; Figure 2A) [26]. A trough occurred in 2004 (71,847 to 113,000 cases) [16], [26], [27], [31], representing <10 times the number reported in the peak year, 2010 (Table 2; Figure 2A).
The number of reported severe cases also varied by year and the annual proportion of DF manifest as DHF was 0.1–0.5% over the review period. In 2000, the annual number of DHF cases was between 40 and 4502 [6], [15], [16], [26], [27]. The number of DHF cases during 2000–2010 (>18,000) is striking when compared with data from the previous decade: during the 1990s <1000 cases of DHF were reported [26]. The years in which numbers of DHF cases peaked reflected the national epidemic years for dengue disease described above, with high numbers of DHF cases in 2002 and 2008 (Figure 2B). The 2008 national epidemic of DF/DHF continued with elevated incidence into 2009/2010, with nearly 4000 cases of DHF reported in 2010 [26].
The proportion of severe cases reported is typical of countries in the Americas, but is low compared with Asia where the reported incidence of DHF is much greater [32]. In similar-sized populations, attack rates for severe dengue disease are 18 times greater in Southeast Asia than in the Americas [32]. However, differences in health surveillance system reporting guidelines and variations in case management practices may contribute to the differences in the number of cases reported, and limit the ability to make valid comparisons [33]. In Brazil, DHF cases are defined by strict application of all four criteria from the 1997 WHO guidelines, which is not the case elsewhere [1].
Similarly, hospitalizations related to dengue disease increased over the survey period to >94,000 hospitalizations in 2010 (Figure 2C) [26]. The incidence of dengue-related hospitalization was 31·6/100,000 population during the 2002 national epidemic, approximately 40·8/100,000 during the 2008 national epidemic, and 49·7/100,000 during the 2010 national epidemic [26]. These increases in hospitalization rates during epidemic years might suggest an increase in the severity of dengue disease in Brazil, although an increased awareness during epidemics and a lower threshold for hospitalization might also account for these increases.
The number of dengue-related deaths followed the same patterns as the other epidemiological indices of dengue disease. In 2010, of 13,909 cases classified as DFC and 3807 classified as DHF, there were 370 and 308 fatal cases, respectively. The overall number of DHF- or DFC-related deaths was 678 compared with only 19 in 2004 (Figure 2C) [26].
A seasonal pattern of dengue disease was observed in those studies with available seasonal case distribution data. The highest incidences occurred during January–June [34]–[38], corresponding to the period of highest rainfall and humidity, providing suitable conditions for Ae. aegypti breeding and survival. The study by Goncalves Neto et al. [35] showed that 83·3% of dengue disease cases occurred during the rainy season and demonstrated a positive Pearson correlation with the amount of rainfall (r = 0·84) and relative humidity (r = 0·76) and a negative correlation with temperature (r = −0·78).
We found published regional data for part of the study period from four of the five Brazilian regions [6], [28], [34], [35], [39]–[51]. No published data were recovered for the North region. The available data show that incidence rates varied greatly throughout the country (data not shown; Table S3). In a study of 146 Brazilian cities in October 2006, incidence rates (per 100,000 population) in the 61 cities that reported >500 dengue disease cases ranged between 24·70 (Sao Paulo) and 6222·71 (Campo Grande) [52]. By the end of 2006, 25 of the 27 states had reported local dengue epidemics [15].
The geographic distribution of the Ae. aegypti vector has widened over the 11-year review period, involving an increasing number of municipalities (Figure 2D) and this has resulted in a broader regional distribution of dengue disease. In most regions the dengue disease incidence followed national trends (Figure 2E). In the early years of the survey, the Southeast and Northeast regions were most affected by DENV infections, whereas from 2009 more cases were reported from studies within the Central-West region. Incidence rates reported in the South region were consistently lower than in other regions. The distribution of reported cases of dengue disease during the 2010 national epidemic was different from that in the 2002 and 2008 national epidemics with high attack rates observed over larger areas of Brazil [26]. These regional variations in dengue disease incidence are unsurprising given the geographically diverse nature of Brazil with its large variations in climate and population density.
A change in the age distribution of dengue disease over the survey period was evident from the available data. Young adults were most affected by DF and DHF during 2000–2007 and 2000–2005, respectively (i.e., DHF was coincident with the highest incidence of DF). However, in 2006 the incidence of DHF among children aged <5 years increased (0·47/100,000) and was higher than among those aged 10–19 years and 20–39 years (0·36/100,000 and 0·46/100,000, respectively) [9]. During 1998–2006, most DHF cases were in the 20- to 40-year age group, whereas in 2007 >53% of DHF cases occurred in children <15 years of age [53]. In 2007, a large proportion of cases of dengue-related hospitalizations (40.8%) occurred among those aged <10 years. Furthermore, children aged 5–9 years and 10–14 years showed marked increases in hospitalization rates (68·2 and 60·6/100,000 population, respectively) during the 2008 national epidemic, compared with during the 2002 national epidemic (15·9 and 23·1/100,000 population, respectively) [26]. These hospitalization data are in agreement with the distribution of hospitalizations for dengue disease according to age for 2002–2011 (Figure 3) [26], which suggests a change in age pattern in 2007–2008 (a reduction in the first quartile age) although data from 2009 suggest this change may have been transient. The median age of death from DF was approximately 38 years in 2002 and fell to 30 years between 2007 and 2009 [26].
Regional age-related data from eligible studies are sparse and inter-regional comparisons are difficult (Table 3) [35], [39]–[42], [44], . The most comprehensive data are for 2001–2008 from Ceará state, Northeast region [39]. In 2001, the highest incidence of cases occurred in those aged 20–59 years, whereas in the 2008 national epidemic, those mostly affected were aged <10 years. These data reflect the national changes in age distribution of dengue disease.
Slightly more women than men are affected by dengue disease throughout Brazil [36], which is similar to the sex distribution of reported cases in other Latin American countries [9]. During 2001–2010 the male∶female ratio of reported cases ranged from 0·75–0·82 [9], [26]. Regional data were more variable. In 2000 the ratio was 1·09 in the city of São Luís [35], and 0·5 in the City of Santos in 2010 [54]. Women with dengue disease were slightly older than men (mean age 33·7 years versus 30·2 years, respectively; p = 0.019) [37].
Despite some gaps, our literature survey and analysis provides a comprehensive overview of the evolving epidemiology of dengue disease in Brazil over the period 2000–2011. This study has several important strengths. Our survey was thorough; we screened >700 articles to identify relevant publications and we developed a comprehensive data extraction instrument to facilitate the capture of all relevant data.
Nevertheless, the lack of comprehensive and continuous data for the survey period limits our ability to make comparisons and draw firm conclusions over the years, across regions, and among different ages. For example, age-stratified data were not reported systematically and age range boundaries differed by study. Therefore, although we can suggest trends in age distribution, it is not possible to directly compare data from the selected publications.
The inclusion of publications in three languages reduced selection bias in our literature review and analysis. However, despite the inclusion of PhD dissertations and theses there is a bias towards published articles. An assessment of quality of evidence was not carried out and potential weaknesses of some studies such as inadequately described case selection, small sample sizes, and unspecified statistical methods were not reasons for exclusion. Consequently, any limitations of the original studies are carried forward into our review.
Many of the studies relied on data reported by passive surveillance systems, which can vary between regions and over time [33] and may misrepresent the number of cases due to changes in reporting behaviour and misclassifications.
Our literature survey and analysis identified several knowledge gaps, which indicate potential avenues for future study. In particular, there are gaps relating to the regional incidence of dengue disease in Brazil, national and regional age-related data, and national and regional serotype information. Further epidemiological studies may help to clarify and define regional differences.
The large increase in the number of DHF cases and the shift in age distribution of DHF towards younger age groups that occurred during the 2007–2008 national epidemic warrant explanation. One possibility is that the change in circulating DENV serotypes over time may have affected the pattern of dengue disease epidemiology in Brazil [78]. Age-stratified seroprevalence studies will improve assessment of the level of transmission and inapparent infection, as well as providing information relating to the age shift.
Further studies into the risk factors for dengue disease and its severity are also important. For example, in Southeast Asia, DENV infection has been more widespread for a longer period of time than in the Americas, creating a large group of individuals likely to experience a second or third infection [32]. These secondary infections carry an increased risk of severe dengue disease. The data in this review do not address the Southeast Asian experience and further examination as to whether this phenomenon is replicated in Brazil is required. In addition, few studies in the review specifically measured the effects of urbanization in Brazil, with effects only inferred from studies of other socio-demographic factors. The diversity of ethnic backgrounds within the population suggests that further genetic studies are warranted to determine whether ethnicity affects the clinical expression of dengue disease and the risk for severe outcomes. Studies are also required to clearly define associations with other diseases if comorbidity screening is to be used to identify patients at a greater risk of developing DHF.
We acknowledge that there are gaps in our epidemiological knowledge of dengue disease in Brazil, due, in part (as in many other countries) to the inherent weaknesses of passive surveillance systems. The majority of infections are clinically non-specific consequently dengue disease is often mis-diagnosed during inter-epidemic periods [8]. The findings presented here are in broad agreement with those of Honório et al. [79], who found only 23·3% of infections were symptomatic, and with Lima et al. [80], who showed that the number of cases reported for the Southeast region of Brazil under-represented the number of infected individuals. This was also found in studies conducted in other countries [81]. Only when an epidemic occurs is the full spectrum of the disease recognised. Consequently, the disease is likely to be under-reported during inter-epidemic periods but over-reported during epidemics [82]. Overall, we believe the national surveillance data under-estimate the true incidence of DENV infections. However, extensive representative serological surveys are required to estimate the true rate of infection and transmission and, thus, despite its drawbacks, passive reporting is important for the identification of disease trends over time.
Our review and analysis of the epidemiology of dengue disease in Brazil during the past decade suggests an overall increase in the distribution and severity of dengue disease. During the last decade (2000–2010), a total number of 8,440,253 cases were reported (the highest figure in the history of dengue disease in this region) with the highest number of severe cases (221,043; 2.6%) and fatal cases (3058; 0.036% of the total reported cases and 1.38% of the severe cases) [83]. The 1588 cases of severe dengue disease and 163 deaths reported as of epidemiological week 8 in 2011, represent 67% and 73%, respectively, of the total cases registered in the Americas [84]. The co-circulation of multiple DENV serotypes and high dengue disease endemicity may be responsible for the increased occurrence of severe forms of dengue disease and increases in the numbers of dengue disease-related hospitalizations. In addition, the increase in the number of severe cases of dengue disease and a shift in age group predominance of severe forms observed during 2007/08 confirm that dengue disease must remain a public health priority in Brazil.
Even though the studies included in this literature review have improved our understanding of the epidemiology of dengue disease in Brazil, further studies are required to clarify the epidemiological pattern and to understand regional epidemiological differences, the diversity of genotypes of circulating serotypes and the extent of herd immunity by age group. Our review has highlighted the main epidemiological characteristics of dengue in Brazil in the first decade of this century and revealed that the epidemiological pattern of dengue disease in Brazil is complex. The changes observed are likely to have been the result of multiple factors, which still require elucidation.
|
10.1371/journal.pgen.1008158 | Peptidergic signaling from clock neurons regulates reproductive dormancy in Drosophila melanogaster | With the approach of winter, many insects switch to an alternative protective developmental program called diapause. Drosophila melanogaster females overwinter as adults by inducing a reproductive arrest that is characterized by inhibition of ovarian development at previtellogenic stages. The insulin producing cells (IPCs) are key regulators of this process, since they produce and release insulin-like peptides that act as diapause-antagonizing hormones. Here we show that in D. melanogaster two neuropeptides, Pigment Dispersing Factor (PDF) and short Neuropeptide F (sNPF) inhibit reproductive arrest, likely through modulation of the IPCs. In particular, genetic manipulations of the PDF-expressing neurons, which include the sNPF-producing small ventral Lateral Neurons (s-LNvs), modulated the levels of reproductive dormancy, suggesting the involvement of both neuropeptides. We expressed a genetically encoded cAMP sensor in the IPCs and challenged brain explants with synthetic PDF and sNPF. Bath applications of both neuropeptides increased cAMP levels in the IPCs, even more so when they were applied together, suggesting a synergistic effect. Bath application of sNPF additionally increased Ca2+ levels in the IPCs. Our results indicate that PDF and sNPF inhibit reproductive dormancy by maintaining the IPCs in an active state.
| Diapause is a hormonally mediated process that allows insects to predict and respond to unfavourable conditions by altering their metabolism and behavior to resist the oncoming environmental challenges. In Drosophila melanogaster females a protective state of reproductive dormancy is induced by lower temperatures and shorter photoperiods that mimic the approach of winter. By genetically manipulating the circadian pacemaker s-LNvs cells, which express two neuropeptides, Pigment dispersing factor (PDF) and short Neuropeptide F (sNPF), we were able to modulate levels of gonadal arrest. PDF and sNPF appear to act as antagonists to dormancy, as do the Drosophila insulin-like peptides (dILPs) that are expressed in the insulin producing cells (IPCs). Indeed, we observe that the axonal projections from the s-LNvs appear to overlap with those from the IPCs implying that the clock cells signal to the IPCs. We confirm this possible communication by applying the two synthetic peptides to the IPCs and detecting a response in the IPC signal transduction pathway. We conclude that the clock neurons activate the IPCs via PDF and sNPF, which in turn release the dILPs, antagonise dormancy and lead to reproductive growth, thereby uncovering a neurogenetic circadian-overwintering axis.
| To synchronize with the Earth’s rhythmic environment, most higher organisms have evolved endogenous time-keeping systems [1,2]. While the highly conserved circadian clock is well-characterized [3,4], our knowledge of the seasonal clock that governs the overwintering response (diapause) in insects is still fragmentary [5,6]. Diapause refers to an alternative developmental program, typically induced by the shortening days and falling temperatures of the approaching winter [1,7]. Diapausing animals are characterized by low metabolic rate, drastically decreased food intake, extended lifespan, and increased stress resistance [8–15]. The fruit fly Drosophila melanogaster exhibits an adult reproductive ‘diapause’ manifested by arrested ovarian development which is stimulated by low temperatures and can be enhanced by short photoperiods [15,16]. While the Drosophila literature refers to this phenomenon as ‘diapause’ rather than ‘dormancy’ or ‘overwintering’, we recognize that it is not a classic photoperiodically-induced state because it requires cold-temperature to induce the reproductive quiescence. Nevertheless, it shows features that are commonly associated with responses that are resistant to environmental stresses [7,15,16].
An increasing body of evidence suggests that insulin-like signaling is a key regulator of diapause in numerous species [13,17–20]. In Drosophila, 4 of the 8 identified insulin-like proteins (DILP1, -2, -3, -5) are produced in 14 median neurosecretory cells (designated as Insulin Producing Cells, IPCs) of the Pars intercerebralis (PI), which are anatomically connected to the key neuroendocrine system that governs hormonal regulation of gonadal arrest [21–25]. Even though the center for dormancy control is believed to be in the PI [26–28], the neurosecretory cells in this brain area, including the IPCs, do not have a circadian clock, and would therefore have to receive any timing information (if any) from other cells [29–33]. A challenging question is how the environmental signals that trigger dormancy (i.e. decreasing photoperiod and temperature) including putative timing information, are perceived, interpreted, and converted into hormonal signals in the brain, leading to the overwintering phenotype [34].
Even though several neuropeptides, neurotransmitters, and peptide hormones have already been identified as modulators of function of the IPCs (reviewed in [35,36]), there are still gaps in our understanding of how the activity of these cells is controlled. Recent research revealed a synaptic connection between the IPCs and one group of dorsal clock neurons (DN1), raising the possibility of a direct modulatory effect exerted by the circadian system [33]. Indeed, natural variants of the timeless clock gene in D. melanogaster, namely the s-tim and ls-tim allelic variants, are known to have a dramatic effect on the inducibility of reproductive dormancy [37,38].
In our study, we primarily focused on neuronal clusters that, based on their axonal projections to the dorsal brain (dorsal protocerebrum), could play an intermediary role in conveying dormancy-inducing signals towards the IPCs. Neurites of the small ventral lateral neurons (s-LNvs) project to the dorsal protocerebrum, where they rhythmically release the circadian neuromodulator PIGMENT DISPERSING FACTOR (PDF) [39]. PDF is also expressed in the large ventral lateral neurons (l-LNvs), and in two groups of non-clock cells, a developmentally transient neuronal cluster in the Tritocerebrum (designated as PDF-Tri), and a small number of neurons in the eighth abdominal neuromere of the ventral ganglion (designated as PDFAb neurons) [40]. PDF is a key coordinator of pacemaker interactions and behavioral rhythms [41–43], of sleep and arousal [44,45] and of sexual behavior [46]. PDF may also be involved in diapause regulation in the blow fly Protophormia terraenovae [47], the mosquito Culex pipiens [48], and the bean bug Riptortus pedestris [49]. However, its effects appear to be contradictory and the mechanisms through which PDF acts on diapause are unclear.
Short Neuropeptide F (sNPF) has been implicated in the modulation of diapause in the Colorado potato beetle [50] and has been shown to stimulate ovarian development in the locust [51,52]. In Drosophila sNPF increases food intake and body size [53] and enhances growth [54]. This peptide is broadly produced in the Drosophila nervous system [53,55], including the s-LNvs [56]. A small set of bilaterally symmetric neurons in the Pars lateralis (PL), defined as dorsal-lateral peptidergic neurons (DLPs), also express sNPF. DLPs have axon terminations in the proximity of the IPCs, and co-express the multi-functional neuropeptide Corazonin (Crz) [57], which has also been proposed as a diapause regulating peptide in the hawkmoth Manduca sexta [58].
The G protein-coupled receptors for PDF, sNPF, and Crz (PDFR, sNPFR1, and CrzR, respectively) have already been characterized and extensively studied in Drosophila [57,59–67]. Interestingly, sNPFR1 and CrzR have been found to influence the activity of the IPCs [53,54,57,68]. However, to date no studies have reported PDF signaling to the IPCs.
In the present study, we demonstrate that PDF and sNPF produced by PDF-positive (PDF+) neurons, reduce the induction of dormancy in Drosophila, mainly through a direct effect on the IPCs. Conversely, the Corazonin-expressing DLP neurons do not seem to be involved. Using live imaging, we show that the IPCs respond to both PDF and sNPF peptides with increasing levels of cAMP and to sNPF additionally with increasing Ca2+ levels suggesting that PDF and sNPF positively modulate IPCs activity and thereby inhibit gonadal arrest.
To examine whether the PDF-expressing neurons have a role in the regulation of reproductive arrest, we used a Pdf-Gal4 driver to target gene expression specifically in PDF+ neurons including both s- and l-LNvs [39,41]. First, a bacterial depolarization-activated sodium channel (Na+ChBac) was expressed in these neurons (Pdf>Na+ChBac). Such manipulation will enhance the release of neurotransmitters and neuropeptides, including PDF and sNPF [69]. Pdf>Na+ChBac flies showed significantly lower levels of dormancy compared to controls (p<0.001; Fig 1A). Importantly, Na+ChBac expressing and control flies shared the same timeless (tim) background, as tim alleles (s- and ls-tim) affect the overall level of reproductive arrest (S1 Table) [37]. We also overexpressed PDF in the same subset of cells (Pdf>Pdf), which again resulted in a significant decrease in the incidence of dormancy compared to controls (p<0.001; Fig 1A).
In addition to PDF, the s-LNvs co-express the neuropeptide sNPF. Since sNPF is widely present in the nervous system [53,55], we started by overexpressing this neuropeptide with a pan-neuronal driver. This manipulation (elav>2xsNPF) produced a very significant reduction in dormancy in the experimental flies compared to the controls (p<0.001; Fig 1B). Considering that both the elav-Gal4 and the UAS-2xsNPF lines carry the ls-tim allele (S1 Table), which is known to promote ovarian quiescence [37], the antagonistic effect of sNPF is quite dramatic. We then narrowed the overexpression of the neuropeptide specifically to the PDF+ neurons (Pdf>2xsNPF) and detected a similar and highly significant reduction of ovarian arrest in the experimental flies compared to the controls (p<0.001; Fig 1C). These results suggest that the Pdf-expressing tissues (the s-LNvs, l-LNvs, the PDF-Tri and the PDF-Ab) are making the major contribution to the inhibition of dormancy. To test the importance of the s-LNvs we used the R6-Gal4 driver [70] which is active in the s-LNvs and in some other neurons but not in the remaining PDF+ cells [71]. Again, R6>2xsNPF flies showed significantly reduced dormancy when compared to controls (p<0.001; Fig 1C). Thus, we conclude that sNPF, likely released from the s-LNvs, is involved in the negative regulation of gonadal arrest.
We then considered the opposite manipulation, namely reduced neuronal excitability, which ultimately results in reduced release of neuropeptides. Neuronal overexpression of the potassium channel Ork increases potassium efflux and causes membrane hyperpolarization, thereby preventing the firing of action potentials [72]. Pdf>Ork flies showed higher levels of ovarian arrest compared to controls (p<0.001; Fig 2A). Furthermore, genetically ablating the PDF+ neurons by overexpressing the pro-apoptotic protein hid (head involiution defective), Pdf>hid), also caused a larger proportion of females to undergo dormancy compared to controls (p<0.001; Fig 2B).
As the PDF neurons co-express more than just PDF we also examined whether the Pdf01 null mutation would alter dormancy levels at two photoperiods, LD8:16 and LD16:8. The experiment was performed on the ls-tim background. Fig 2C reveals that there is a significant genotype effect (p = 1.3 x 10−4) with Pdf01 mutants showing significantly elevated levels of ovarian quiescence plus a significant photoperiodic effect (p = 6 x 10−6), with no significant interaction, so the enhancement in dormancy occurs at both photoperiods equally. We also examined the effects of overexpressing the PDF receptor (PDFR) in the IPC cells using dilp2(p)-Gal4 [21,73] in a homozygous receptor mutant han background. This was compared to parental controls (dilp2(p)-Gal4 and UAS-PDFR) that were both homozygous and heterozygous for han, as well as the dilp2(p)>+ wild-type control. All six genotypes were placed on the s-tim background. Fig 2D reveals that overexpression of PDFR causes a highly significant reduction of dormancy (p<0.001) compared to all the corresponding han/han and han/+ controls, but is not significantly different from the dilp2(p)>+ wild-type. Consequently, we appeared to have rescued the mutant phenotype in this genetic background. The heterozygous han/+ background controls also show a highly significant reduction of dormancy compared to their corresponding homozygous mutant controls (both p<0.001) and the dilp2(p)>+ wild type (p<0.01), further underscoring the dosage effect of PDFR on the phenotype. All of these experiments are consistent with the view that the s-LNvs modulate dormancy levels via the neuropeptides PDF and sNPF. However, since Pdf-Gal4 is also expressed in the non-circadian PDF-Tri and PDF-Ab neurons [39,41], an influence of the latter cannot be excluded.
The IPCs express sNPFR1 [54,57,68]; its activation by sNPF stimulates organismal growth by promoting the transcription of insulin-like peptides genes [54]. To investigate whether sNPFR1 signaling in the IPCs modulates ovarian arrest we expressed a dominant negative form of the receptor (UAS-sNPFR1-DN) under the control of two IPCs-specific drivers: dilp2(p)-Gal4 and InsP3-Gal4 [21,73]. The former drives gene expression from the 2nd larval instar, the latter becomes active mainly after larval development [73].
Inhibition of sNFR1 from early larval stages (dilp2(p)>sNPFR1-DN) increased only marginally the proportion of dormancy (Fig 2E). However, when the receptor was repressed later in development (InsP3>sNPFR1-DN), a significantly higher proportion of flies showed gonadal arrest compared to controls (p<0.001; Fig 2E). Both drivers are specific for the 14 IPCs in the brain [21,73], so we speculate that different degrees of dormancy might reflect differences in the strength of the drivers or compensatory phenomena that occur early in development. We also used the InsP3 driver to knockdown sNPFR1. We observed a significant increase in gonadal arrest in InsP3> SNPFR1 RNAi compared to the Gal4 (p = 0.017) and UAS controls (p = 0.006) (Fig 2F). These results are consistent with our previous observations regarding the antagonist nature of sNPF on this phenotype (Fig 1B and 1C), and also suggest a role for sNPF signaling in the IPCs in the regulation of reproductive arrest.
The dorsal-lateral peptidergic neurons (DLPs) are 6–7 bilaterally symmetric neurons in the Pars Lateralis (PL) whose axons ends in the proximity of the IPCs [57]. The DLPs produce the neuropeptides Corazonin (Crz) and sNPF through which they modulate the activity of the IPCs, as the latter express the relevant receptors, CrzR (Corazonin Receptor) and sNPFR [57]. The DLPs affect survival, stress resistance and levels of circulating carbohydrates and lipids [57,74]. Since diapause is associated with marked changes of these parameters, we questioned whether the DLPs are involved in the regulation of this seasonal response. We used two DLP-specific Crz-Gal4 driver lines, Crz1-Gal4 and Crz2-Gal4, to overexpress Na+ChBac and sNPF, respectively. Both Crz1>Na+ChBac and Crz2>2xsNPF flies showed a reduction in the proportion of females undergoing gonadal dormancy (p<0.001) compared to one of the parental controls but not the other (S1 Fig). Therefore, although we cannot totally exclude an effect of the DLPs on ovarian arrest we can conclude that their involvement, if any, is not as robust as that observed for the PDF+ neurons.
Next, we asked whether the dorsal terminals of the s-LNvs in the dorsal vicinity might be close enough to the IPCs to enable such sNPF and PDF signaling. By performing ICC with anti-DILP2 and anti-PDF we could not see direct contacts overlapping between the s-LNvs and the IPCs (Fig 3A). However, when we expressed GFP in the IPCs (dilp2(p)>GFP), we observed that the dorsal projections of the s-LNvs are in close proximity to fine processes originating from the IPCs (Fig 3B). To test whether these fine processes are dendrites, we expressed the dendritic marker DenMark in the IPCs (dilp2(p)>DenMark) [75]. We found prominent labeling in the IPC processes, indicating that they are of dendritic origin (Fig 3C and 3D). This explains why we could not see them by anti-DILP2 labeling.
The IPCs are neurosecretory cells that are crucial for initiating seasonal responses [13,17,19,20]. Since we have shown that PDF and sNPF have a modulatory effect on diapause (Fig 1), and that PDF+ and sNPF+ neurons appear to contact the IPCs (Fig 3B)[57], we asked whether the IPCs can respond directly to these neuropeptides. The PDF receptor signals primarily via cAMP [59,61,76], whereas the signaling cascade following activation of the receptor for sNPF uses cAMP, at least in part [77–79]. Thus, we expressed a genetically encoded fluorescence resonance energy transfer (FRET) based cAMP sensor in the IPCs (dilp2(p)>Epac1camps) to monitor real-time cAMP levels [62]. A similar experimental design had previously been adopted with success to investigate the presence of the PDF receptor in clock neurons [62].
We bath-applied 10 μM of synthetic PDF to acutely dissected fly brains. This resulted in a slow rise in the intracellular amount of cAMP, measured as the average FRET signal between 100 and 1000 s after the application of the neuropeptide (light-blue curve and bar; Fig 4A and 4B). cAMP FRET signals were 10% ca. higher (100–1000 s; p<0.001) compared to the negative (modified minimal hemolymph-like solution, HL3, light-grey curve and bar; Fig 4A and 4B) control. A similar increment was also observed after presenting PDF together with the sodium channel blocker tetrodotoxin (TTX, dark-blue curve and bar, 100–1000 s; p<0.001; Fig 4A and 4B). The latter prevents neuronal communication, suggesting that PDF activates the IPCs directly. However, the short-term (100-200s) response to PDF was unchanged compared to the negative control, either with or without TTX (Fig 4C and 4D).
Similar observations were made upon the application of 10 μM synthetic sNPF. This resulted in a slow but significant cAMP increase in the IPCs either in the absence (yellow curve and bar, 100–1000 s; p<0.001) or in the presence (orange curve and bar, 100–1000 s; p<0.001) of TTX, reflecting a direct activation of the IPCs by sNPF (Fig 4E and 4F). Moreover, as we saw for PDF, the short-term responses to sNPF did not differ from the negative control (Fig 4G and 4H).
However, when we applied sNPF and PDF together (sNPF+PDF), at a concentration of 10 μM for each peptide, we recorded an increase in cAMP FRET signal, reaching a level ~15% higher than that for the negative control (red, 100–1000 s; p<0.001; Fig 5A and 5B). Moreover, the short-term response was particularly distinctive compared to the applications of single peptides, revealing a ~8% increase in cAMP FRET signals (red curve and bar; 100–200 s; p<0.01; Fig 5C and 5D). We repeated the experiment but halving the concentration of each peptide to 5 μM (sNPF1/2+PDF1/2). This also resulted in a significant increase in cAMP (pink curve and bar, 100–1000 s; p<0.05; Fig 5A and 5B). However, we noticed that following an initial increase, after 400 s the concentration of cAMP slowly declined (pink curve and bar; Fig 5A). Focusing on the short-term response, the co-application of (sNPF1/2+PDF1/2) resulted in higher cAMP FRET signals compared to the bath treatment with single peptides; however, the difference was statistically significant only with regard to PDF (100-200s; p<0.01; Fig 5C and 5D).
We also tested (sNPF+PDF+TTX) and found an increase of ~13% in the amount of cAMP FRET signal compared to the negative control (magenta curve and bar, 100–1000 s; p<0.001; Fig 5E and 5F). This response was not statistically different from single peptide applications in the presence of TTX although in the interval 300–600 s the levels of cAMP for (sNPF+PDF+TTX) were higher (Fig 5E and 5F). Interestingly, although there were no differences in cAMP levels when comparing the application of sNPF+PDF with or without TTX in the interval 100–1000 s, the distinctive short term response disappeared in the presence of TTX (Fig 5G and 5H). This suggests that the short-term response of the IPCs to the combined application of (sNPF+PDF) is indirectly mediated by other cells.
We assessed whether the larger increase in cAMP levels induced in the IPCs by the co-application of sNPF+PDF is a specific response. We co-applied sNPF with the following neuropeptides: SDNFMRFamide (SDNFMRFa), adipokinetic hormone (AKH), Drosophila tachykinin (DTK) and allatostatin-C (Ast-C). AKH and DTK were chosen because they are known regulators of IPC activity [21,73,80,81]. When sNPF was co-applied with SDNFMRFa, AKH or DTK, the levels of cAMP were significantly reduced compared to the sNPF+PDF co-application (p<0.05, p<0.01, p<0.05, respectively) and were indistinguishable from the negative control (Fig 6). Conversely, the co-application of sNPF+Ast-C resulted in a significant increase in the amount of cAMP but the addition of Ast-C alone was sufficient to evoke similar cAMP responses in the IPCs (Fig 6). To our knowledge, no data are available on the distribution of the Ast-C receptor in Drosophila, and no link between IPCs and Ast-C has been reported previously. In summary, our data show that the strong and rapid responses of the IPCs we observed are unique to the co-application of PDF and sNPF. In particular, they are not caused either by the combination of other peptides or by receptor cross-activation due to high peptide doses applied.
To verify whether the responses to PDF were mediated by its receptor, we carried out the treatment in the PDFR-null (han) [59] background (han; dilp2(p)>Epac1camps). The application of PDF in the mutant no longer evoked an increase of cAMP (light-blue; 100–1000 s; Fig 7A–7D), and surprisingly, neither did sNPF (yellow curve and bar; 100–1000 s; Fig 7A–7D). The rapid increase in cAMP levels induced by the co-application of sNPF+PDF was also obliterated in the mutant (red curve and bar; 100–200 s; Fig 7C and 7D). To test whether sNPF acts via PDFR, we rescued PDFR expression in the IPCs in the han mutant background (han; dilp2(p)>PDFR; Epac1camps). In these flies, IPCs strongly responded to PDF, but not to sNPF. It is therefore unlikely that sNPF signals via PDFR (Fig 7E–7H). We observed a slight increase in cAMP ~400s after sNPF application, but this was not significantly different from the negative controls (yellow curve and bar; 100–1000 s; Fig 7E and 7F). The fast response to sNPF was completely absent (yellow curve and bar; 100–200 s; Fig 7G and 7H). This shows that the absence of the response to sNPF in the han mutant cannot be directly attributed to the loss of PDFR.
In several cell types PDF signals primarily through cAMP rather than calcium [59,61,76]. Here we tested whether PDF or sNPF affect the level of intracellular Ca2+ in the IPCs by expressing the genetically encoded Ca2+ sensor GCaMP3.0 (dilp2(p)>GCaMP3.0) [82] Incubating PDF with freshly dissected brains did not produce change in Ca2+ levels, which were indistinguishable from the negative control (Fig 8). However, the bath application of sNPF induced a small but significant increase in the Ca2+ signal (100-200s, 3.4 ± 2.8%, p<0.05). This was still detectable in the presence of TTX (100-200s, 5.4 ± 3.5%, p<0.05), suggesting that this response is not mediated by interneurons but is due to the direct activation of the IPCs by sNPF (Fig 8).
We have constitutively activated the PDF+ neurons with Na+ChBac [69] and observed a significant reduction in levels of gonadal arrest (Fig 1A). Since this treatment increases membrane excitability, it likely increases release of both PDF and sNPF, whose overexpression also led to a significant reduction in dormancy (Fig 1B and 1C). Manipulations with opposite effect, namely reduction of membrane excitability of the PDF+ neurons through overexpression of the K+ channel Ork [72] or induction of cell death by overexpression of the pro-apoptotic gene hid, resulted in an enhanced dormancy response (Fig 2A and 2B). Furthermore, the Pdf0 mutant showed a significantly elevated response compared to controls and overexpressing PDFR in the IPCs significantly reduced gonadal arrest on both heterozygous and homozygous mutant han backgrounds to wild-type levels. These results consistently support a model where PDF and SNF act to antagonise dormancy, possibly by enhancing dILP expression. When we expressed a dominant negative form of the sNPF receptor in the IPCs, or downregulated the receptor with RNAi, higher dormancy induction was observed especially when using a driver that is active later in development (Fig 2E and 2F). Numerous sNPF-expressing neurons can potentially target the IPCs, including the PDF+ sLNvs and the DLPs. However, the manipulations of the DLPs did not affect reproductive quiescence (S1 Fig).
One puzzling aspect of the results is that they appear, at least superficially, to contradict a recent study in which Pdf01 females showed relatively low levels of dormancy and did not reveal a photoperiodic effect [83]. The LD cycles used in these experiment were LD16:8 versus LD10:14 so not as extreme as ours and this might have had a damping effect on any photoperiodicity. Furthermore, the low dormancy levels of the mutant females suggested that the mutation might have been on the s-tim background. A congenic wild-type was not compared to the mutant, so it is difficult to predict whether such a control would have had lower levels of reproductive arrest which would be consistent with our findings. In addition, the further stress of starvation was incorporated into the experimental paradigm. It would therefore be of interest to examine whether our results, which suggest that the neuropeptides released by clock cells antagonise reproductive quiescence, can be generalized under a variety of different unfavourable environmental conditions.
Using a whole brain ex-vivo preparation, we observed that the IPCs responded to bath-applications of sNPF and PDF with increasing cAMP levels. The responses persisted when synaptic connections with the rest of the brain were inhibited by TTX, suggesting a direct effect on the IPCs of these neuropeptides (Fig 4A–4H). Expression of sNPFR1 in the IPCs had already been described [53,54,57,68]. Interestingly, sNPFR1 couple to more than one Gα-protein subtype, since both excitatory (through Gsα [77,78]) and inhibitory effects (through Goα [79]) on cAMP levels have been documented. In addition, sNPF can signal by suppressing Ca2+ in some circadian clock clusters [42] and peptidergic PTTH neurons [84]. Here, we found that sNPF increased Ca2+ levels in the IPCs showing that it can also have activating Ca2+ effects. Similar multiple G-protein coupling has also been reported for the neurokinin-1 and -2 receptors [85,86], as well as for the glucagon receptor in human atrial membranes [87].
The expression of the PDFR in the IPCs is less well-characterized. A previous study using fluorescent in situ hybridization reported prominent PDFR expression in the PI. However, it did not identify the PDFR positive cells [60]. Moreover, another study suggested that the cAMP responses evoked in the IPCs by PDF are not as robust as those registered in clock neurons [62]. Possibly, lower ligand efficacy might simply reflect less PDFR in these cells. The strong cAMP responses we observed after driving PDFR in han mutants under control of dilp2(p)-Gal4 supports this conclusion (Fig 6E–6H). On the other hand, there are 14 IPCs in the PI which show heterogeneous protein and neuropeptide composition [29,88], and rhythmic electrophysiological parameters [33]. Thus, it is possible that, like the s-LNvs [89], the responsiveness of the IPCs to PDF is also influenced by time of day. In any case, here we found significant cAMP increasing effects of PDF on the IPCs.
When we applied sNPF and PDF together, the short-term (100–200 s) response of the IPCs to the combined peptides was greater than the sum of their separate activities, pointing to a synergistic effect between the two molecules (Fig 5A–5D). Similar interactions were not observed when sNPF was co-applied with other Drosophila neuropeptides, suggesting that the interaction of sNPF and PDF is specific. Nevertheless, the addition of TTX dampened the short-term response, suggesting that additional cells participate in the synergism between sNPF and PDF (Fig 5E–5H). Interestingly, han mutants lacking PDFR did not respond to PDF, sNPF or sNPF+PDF co-application (Fig 7A–7D). For sNPF, these results are puzzling, but since the rescue of PDFR in the IPCs only restored the response to PDF but not to sNPF, we conclude that sNPF does not act via PDFR. Still, we cannot completely exclude a cross-talk between PDFR and sNPFR1. For instance, there is evidence that GPCRs can engage in homo- or hetero-oligomeric complexes, resulting in cooperativity (reviewed in [90]). Alternatively, an interaction downstream of the receptors may occur, for example at the level of the signalosomes. PDFR and sNPFR associate with specific and different signalosomes that may have enhancing or opposing effects on cAMP levels [76–79,91]. However, further studies are required to investigate these possibilities using experimental settings that are closer to physiological conditions than bath applications of peptides. Interestingly, sNPF (produced in the s-LNvs and in a subset of dorsal lateral neurons) and PDF interact on clock neurons and set the phase of their cytosolic Ca2+ rhythms according to neuron cluster [42]. PDF primarily regulates Ca2+ rhythms in the LNd and DN3 clusters, while sNPF orchestrates that of the DN1 [42]. Thus, the two signaling pathways act dynamically to facilitate the right timing of circadian neuronal activities. Similar complex regulation could also influence the seasonal clock system.
While the Pdf mRNA and protein do not cycle, it has been reported that PDF cycles at the nerve terminals with a circadian rhythm [39]. Thus, we presume that the daily rhythmic stimulation of the IPCs under normal summer conditions of warm days and long photoperiods maintains the expression of dILPs and suppresses dormancy. A simple model would have low temperatures and short photoperiods reducing the expression of PDF and sNPF from the sLNvs terminals, which might be expected to reduce IPC activation and enhance reproductive arrest levels. This could mean that the transcription/translation of these neuropeptides could be reduced under colder conditions, or alternatively, that their release was reduced from the sLNv terminal. This scenario might imply higher levels of PDF within the sLNv soma if it is sequestered there. Alternatively, the receptors for PDF and SNPF on the IPCs might be less sensitive to their ligands at lower temperatures, which would have the same effect on reproductive arrest. Future studies are required to examine the dynamic nature of PDF/SNPF and receptor expression under simulated winter conditions.
PDF and sNPF–most likely from the s-LNvs can maintain D. melanogaster, originally a tropical species, in the reproductive state. However, it is intriguing that high-latitude Drosophila species such as D. montana, D. littoralis, D. ezoana and D. virilis lack PDF in the s-LNvs [92–95]. These species have a high incidence of reproductive arrest even under long-daylengths, an adaptation to the low temperatures even under summer photoperiods at these clines. For example, D. ezoana enters diapause when day-length falls below 16 hours [96]. Although we do not know whether the s-LNvs of the high-latitude species still express sNPF, we speculate that the lack of PDF-signaling to the IPCs of these species might facilitate the termination of the reproductive state under short-day condition, and induce ovarian arrest [97]. The role of the s-LNvs as a source of PDF and sNPF may thus provide the entry point into the neuronal mechanism that allows D. melanogaster to detect the environmental conditions that predispose them to reproductive dormancy.
Flies were reared at 23°C, 70% relative humidity, under 12-hour light/12-hour dark cycles (LD 12:12) on cornmeal standard food. The following, previously described fly strains were used: Hu-S Dutch natural population [37], han5304 [59], UAS-PDFR [59], gal1118 Gal4 enhancer trap [98], R6-Gal4 [70], UAS-Epac1camps [62], UAS-hid [99], UAS-Pdf [41], and UAS-DenMark [75]. InsP3-Gal4 was a gift from Michael J. Pankratz [73], dilp2(p)-Gal4 (p, precocious) was a gift from Eric J. Rulifson [21], UAS-2xsNPF [53] and UAS-sNPFR1-DN [54] were kindly provided Kweon Yu. Out of these lines, we newly combined han5304; dilp2(p)-Gal4, han5304;;UAS-Epac1camps, and han5304;UAS-Epac1camps/CyO;UAS-PDFR. The transgenic line Pdf-LexA,LexAop-CD4::GFP11/CyO;UAS-CD4::GFP1-10/TM6b for GRASP was a gift from François Rouyer. UAS-OrkΔ-C (designated as UAS-Ork [72]) and UAS-Na+ChBac [69]) were gifts from Michael B. O’Connor. Pdf01 mutants were backcrossed for 8 generations to a natural wild-type ls-tim line from The Netherlands [37]. The following lines were ordered from Bloomington Drosophila Stock Center: Canton-S wild-type strain (#1), Crz1-Gal4 (#51976), Crz2-Gal4 (#51977), ElavC155-Gal4 (designated as elav-Gal4, #458), Pdf-Gal4 (#6900), UAS-CD8-GFP (BDSC #5137), and Oregon-R (#2376; used as control for experiments with han mutant).
In order to use a control in which the specific transgene (Gal4 or UAS) was in the same condition of heterozygosis as in the experimental line, we have crossed all the parental lines to the w1118 strain (generic genotype w1118;P-element/+). In Drosophila, two allelic variants of the timeless circadian clock gene (s-tim/ls-tim) were found to significantly influence reproductive arrest: ls-tim allele promotes reproductive quiescence at every photoperiod [37,38]. Considering this modulatory effect, controls were generated according to the tim allele of the UAS and Gal4 strains used in the experiments (see below for PCR timeless genotyping), with the help of w1118; s-tim (gift from Matthias Schlichting) and w1118; ls-tim (from our lab) lines that express either the s- or ls-tim isoform.
All fly stocks and crosses used for dormancy assays were maintained at 23°C in LD 12:12. Newly eclosed flies (within 5 h post-eclosion) were placed in plastic vials, and immediately subjected to 12°C in LD 8:16. After 11 days, flies were killed in abs EtOH, and ovaries of females were dissected in PBS. Dormancy levels were scored by considering the absence of yolk deposition in the ovarian follicles [7]. These data were presented as the proportion of females with ovarian arrest among all the dissected individuals. On average, 5 replicates of n>60 flies were dissected for each genotypes.
Flies were reared at 23°C in LD 12:12, and newly eclosed individuals were placed in short days (LD 8:16) at different temperatures (12, 18, 23°C) for 11 days. Females were collected at ZT1 (Zeitgeber Time 1, 1 h after light-on), and immediately fixed in 4% paraformaldehyde (PFA) in PBS for 100 min at room temperature (RT). After 3 washes in PBS, brains were dissected in ice-cold PBS, fixed in PFA 4% for 40 min at RT, and subsequently washed 6 times in PBS containing 0.3% Triton X-100 (PBS-T). Next, samples were permeabilized in 1% PBS-T, followed by an overnight blocking step in 1% bovine serum albumin (BSA) in 0.3% PBS-T at 4°C. Afterwards, brains were incubated in primary antibody solution (diluted in 0.1% BSA, 0.3% PBS-T) for 3 days at 4°C. After 6 washes in 1% BSA in 0.3% PBS-T at RT, another blocking step was performed in 1% BSA in 0.3% PBS-T at 4°C, followed by hybridization with the secondary antibody (diluted in 0.1% BSA, 0.3% PBS-T) overnight at 4°C. The primary antibodies used in this study: mouse anti-GFP (1:500; Thermo Fisher Scientific), rabbit-anti-GFP (1:1000; A6455, Invitrogen) and anti-PDF (1:5000; mAb C7, Developmental Study Hybridoma Bank, donated by Justin Blau), rabbit anti-DILP2, anti-sNPF recognizing the sNPF propeptide (for both 1:2000; Jan A. Veenstra [100,101]). The secondary antibodies used: Alexa Fluor 488 (goat anti-mouse, 1:250) Cy3 (goat anti-rabbit, 1:500) (Thermo Fisher Scientific), DyLight488 (goat-anti-rabbit 1:250) and DyLight649 (goat anti-mouse, 1:250) (Jackson ImmunoResearch, Dianova). Staining was visualized adopting either a semi-confocal (Nikon Eclipse 80i equipped with a QiCAM Fast Camera using the Image ProPlus software) or confocal microscope (ZEISS LSM700 running ZEN Lite software or Leica SP8).
To real-time monitor cAMP or Ca2+ concentration changes in the IPCs, live optical imaging was performed using the genetically encoded cAMP sensor, Epac1-camps [62] and the genetically encoded Ca2+ sensor GCaMP3.0 [82]. Relying on the UAS-GAL4 binary system [102], the sensors were expressed specifically in the IPCs (dilp2(p)>Epac1-camps or dilp2(p)>GCaMP3.0). Female flies, maintained at 25°C in LD 12:12, were anesthetized on ice before brain dissections in cold hemolymph-like saline (HL3 [103]) and mounted at the bottom of a cap of a plastic Petri dish (35x10 mm, Becton Dickenson Labware, New Jersey) in HL3 with the dorsal surface up. Brains were allowed to recover from dissection 15 min prior to imaging. Live imaging was conducted by using an epifluorescent imaging setup (VisiChrome High Speed Polychromator System, ZEISS Axioskop2 FS plus, Photometrics CoolSNAP HQ CCD camera, Visitron Systems GmbH) Visitron Systems GmbH) using a 40x dipping objective (ZEISS 40x/1.0 DIC VIS-IR). IPCs were brought into focus and regions of interest were defined on single cell bodies using the Visiview Software (version 2.1.1, Visitron Systems, Puchheim, Germany). Time-lapse frames were imaged with 0.2 Hz and 4x binning by exciting the CFP fluorophore of the cAMP sensor with 434/17 nm light or the GFP fluorescence of GCaMP3 with 488/10 nm light. For cAMP imaging, CFP and YFP emissions were separately detected using a Photometrics DualView2 beam splitter. After measuring baseline FRETs for ~100 s, substances were bath-applied drop-wise using a pipette, and imaging was performed for 1000 s. The neuropeptides used in this study were applied in a final concentration of 10 μM in 0.1% DMSO in HL3. The water-soluble forskolin derivate NKH477 (10 μM, Sigma Aldrich) or the cholinergic agonist carbamylcholine (1mM, Sigma Aldrich) served as positive controls in cAMP and Ca2+ imaging, respectively, while HL3 alone with 0.1% DMSO was used as negative control. In the case of tetrodotoxin (TTX) treatments, brains were incubated for 15 min in 2 μM TTX in HL3 prior to imaging and substances were co-applied together with 2 μM TTX. Inverse Fluorescence Resonance Energy Transfer (iFRET) was calculated over time according to the following equation: iFRET = CFP/(YFP-CFP*0.357) [62]. Thereby, raw CFP and YFP emission data were background corrected; in addition, YFP data were further corrected by subtracting the CFP spillover into the YFP signal, which was determined as 0.357 (35.7% of the CFP signal). Next, iFRET traces of individual neurons were normalized to baseline and were averaged for each treatment. Finally, maximum iFRET changes were calculated for individual neurons to quantify and contrast response amplitudes of the different treatments. The following synthetic neuropeptides were used in this study: pigment dispersing factor (PDF: NSELINSLLSLPKNMNDAa; Iris Biotech GmbH), short neuropeptide F-1 (sNPF-1: AQRSPSLRLRFa; Iris Biotech GmbH), adipokinetic hormone (AKH: pQLTFSPDWa, NovoPro Bioscience), allatostatin-C (Ast-C: pEVRYRQCYFNPISCF, gift from Paul H. Taghert), Tachykinin 4 (DTK-4: APVNSFVGMRa, gift from Paul H. Taghert), dFMRFamide 4 (SDNFMRFa, gift from Paul H. Taghert).
For Ca2+ imaging, brains expressed the GCamp3.0 sensor in the insulin producing cells (dilp2(p)>GCaMP3.0) [104]. The preparation of the brain samples was the same as in the case of cAMP imaging, and the same microscope was used with a modified setup, measuring GFP fluorescence without a beam splitter. The cholinergic agonist carbamylcholine (1 mM CCh) was used to generate rapid Ca2+ increases (Nakai et al. 2001). After subtraction of background fluorescence, changes in fluorescence intensity were calculated for each ROI as Δ(F/F0) = [(Fn—F0)/F0] x 100 with Fn as fluorescence intensity at time point n and F0 as the baseline fluorescence calculated prior to the application of the different substances to the brain.
To ensure genetic homogeneity for the tim locus, between the experimental flies and their corresponding controls, all the strains used in this study were genotyped in order to identify the tim allele present in their genome (summarized in S1 Table). The genomic DNA was extracted from individual adult females (10 flies per genotype) by homogenizing them in 50 μl of extraction buffer (Tris HCl pH = 8.2 10 mM, EDTA 2 mM, NaCl 25 mM); after addition of 1 μl of Proteinase K (10 mg/ml) samples were incubated at 37°C for 45 min, followed by 3 min at 100°C. The tim region containing the polymorphic site was amplified using a reverse primer (5’-AGATTCCACAAGATCGTGTT-3’) and two different forward primers (ls-tim: 5’-TGGAATAATCAGAACTTTGA-3’; s-tim: 5’-TGGAATAATCAGAACTTTAT-3’) that allow selective amplification of the different tim alleles [37].
Larvae of the following genotypes: Act>sNPFR1-RNAi, Act>+ and +>sNPFR1-RNAi, were reared under standard conditions at 23°C and LD12:12 until eclosion. Newly eclosed female flies were collected and subsequently exposed to low (12°C) or high (23°C) temperature and short photoperiod (8h:16hL:D) for 11 days. mRNA was isolated from whole bodies of 10 females. mRNA was reverse-transcribed with SuperScript II First-Strand Synthesis SuperMix (Invitrogen). PCRs were performed on a CFX96 Touch Real Time PCR Detector System (Bio Rad) with GoTaq qPCR Master Mix (Promega), using the following primers: sNPFR- F: 5′- CGACCATCAGATGCACCA -3′, R: 5′-CGTCCGTCTCGTCTGTCC -3′; rp49 F: 5′- ATCGGTTACGGATCGAACAA-3′, R: 5′- GACAATCTCCTTGCGCTTCT-3′. The results are shown as relative expression ratios obtained with the 2-ΔΔCt method ± SEM. RP49 was used as reference. Results are shown in S2 Fig.
Data were analysed with R statistical software (version 3.0.1, www.r-project.org) and plotted using GraphPad Prism 6 software. In the case of normally distributed data (Shapiro-Wilk normality test, p>0.05), statistical significance was tested by one- or two-way ANOVA with post-hoc Tukey's HSD tests, while data that were not normally distributed were analyzed by Wilcoxon or Mann-Whitney test. In the case of multiple comparisons, raw p-values were further adjusted using Bonferroni correction, and these corrected p-values served as significance levels. When analyzing dormancy assays, all data were transformed to arcsine. For simplicity, figures in the Results section show untransformed data (dormancy, %).
|
10.1371/journal.ppat.1003179 | The Abi-domain Protein Abx1 Interacts with the CovS Histidine Kinase to Control Virulence Gene Expression in Group B Streptococcus | Group B Streptococcus (GBS), a common commensal of the female genital tract, is the leading cause of invasive infections in neonates. Expression of major GBS virulence factors, such as the hemolysin operon cyl, is regulated directly at the transcriptional level by the CovSR two-component system. Using a random genetic approach, we identified a multi-spanning transmembrane protein, Abx1, essential for the production of the GBS hemolysin. Despite its similarity to eukaryotic CaaX proteases, the Abx1 function is not involved in a post-translational modification of the GBS hemolysin. Instead, we demonstrate that Abx1 regulates transcription of several virulence genes, including those comprising the hemolysin operon, by a CovSR-dependent mechanism. By combining genetic analyses, transcriptome profiling, and site-directed mutagenesis, we showed that Abx1 is a regulator of the histidine kinase CovS. Overexpression of Abx1 is sufficient to activate virulence gene expression through CovS, overcoming the need for an additional signal. Conversely, the absence of Abx1 has the opposite effect on virulence gene expression consistent with CovS locked in a kinase-competent state. Using a bacterial two-hybrid system, direct interaction between Abx1 and CovS was mapped specifically to CovS domains involved in signal processing. We demonstrate that the CovSR two-component system is the core of a signaling pathway integrating the regulation of CovS by Abx1 in addition to the regulation of CovR by the serine/threonine kinase Stk1. In conclusion, our study reports a regulatory function for Abx1, a member of a large protein family with a characteristic Abi-domain, which forms a signaling complex with the histidine kinase CovS in GBS.
| The gram-positive Streptococcus genus includes three major human pathogens that are members of the normal microflora: Streptococcus pneumoniae (also known as the pneumococcus), Streptococcus pyogenes (Group A Streptococcus), and Streptococcus agalactiae (Group B Streptococcus). Their carriage in the population is highly dynamic and mostly asymptomatic. However, each of these species can cause a wide spectrum of diseases, from local infections to systemic and fatal infections including septicemia and meningitis. Expression of streptococcal virulence-associated genes is tightly regulated at the transcriptional level. However, the signal(s) and the precise molecular events controlling the switch from commensalism to virulence are not yet understood. In this study, we identified and characterized a bacterial protein essential for virulence gene expression in Group B Streptococcus, the main pathogen of neonates. We show that this transmembrane protein, named Abx1, interacts with the histidine kinase CovS to modulate the activity of the major regulator of virulence CovR. We define how a core set of four proteins, Abx1, CovS, CovR, and the serine/threonine kinase Stk1, interact to control the expression of virulence genes in S. agalactiae. We propose that Abx1-like proteins, that are widespread in bacteria, might be part of a conserved mechanism of two-component system regulation.
| Some commensal microorganisms are also opportunistic pathogens. Harmless, and potentially beneficial, they may become causative agents of local or systemic infections [1], [2]. To date, the signals dictating the switch from commensalism to virulence are mainly unknown. The complex set of genetic and environmental factors thought to be involved would affect the equilibrium between the host and the microbes. Deciphering the molecular events that govern the transition between commensalism and virulence will contribute to understanding and controlling infections due to opportunistic pathogens.
Streptococcus agalactiae (Group B Streptococcus, GBS) is a commensal bacterium of the adult gastro-intestinal tract and is present asymptomatically in the vaginal flora of 10–30% of healthy women [3]. However, GBS is the leading cause of invasive infections in neonates (pneumonia, septicaemia, and meningitis) and a serious cause of mortality or morbidity in adults with underlying diseases [4]–[6]. As with most streptococci, the ability of GBS to cause infections is multifactorial [7]. The main virulence-associated GBS proteins identified to date are secreted or surface components, including the ß-hemolysin/cytolysin and specific adhesins [7]–[10].
Expression of several major GBS virulence genes is regulated at the transcriptional level by a two-component system (TCS) called CovSR (Control of virulence Sensor and Regulator; also known as CsrSR) [10]–[13]. TCS is the main signaling mechanism used by bacteria to respond to their changing environments [14]. CovSR controls the GBS response to acid stress in vitro and is necessary in vivo at several steps of the infectious process, such as resistance to macrophage killing and penetration of the blood-brain barrier [15]–[18]. Depending on the model of infection, inactivation of the CovSR system leads to a decrease or an increase of virulence [11]–[13], [15], suggesting that this system is tightly regulated in space and time to specifically adapt the bacterial virulence capacities to the infected compartments of the host. CovSR orthologs are present in several streptococcal species, including Streptococcus pyogenes (Group A Streptococcus, GAS), where their function as a master regulator of virulence gene expression is conserved [19]–[23].
The CovSR system belongs to the EnvZ/OmpR family of TCS, where CovS is the membrane-bound histidine protein kinase (HK) and CovR is the cytosolic response regulator (RR). Typically, HK is the sensor of an environmental stimulus that induces its autophosphorylation and, subsequently, phosphorylation of its cognate RR [14]. Phosphorylation of the RR alters its function, generally by modifying its binding affinity for target promoter regions and thereby changing expression of specific genes or operons. In contrast to the majority of TCS, the streptococcal CovSR regulatory pathway is a combination of two antagonistic negative regulators [24]. Indeed, in both GBS and GAS, the phosphorylated CovR acts mainly as a transcriptional inhibitor by direct binding to target gene promoters [11], [13], [25]–[27]. It was therefore suggested, but not formally demonstrated in vitro, that CovS acts mainly as a phosphatase on CovR to de-repress virulence gene expression during infection [13], [24], [28]. Currently, it is not known whether CovS is activated by general stresses [16], [28] and/or is a direct sensor of specific ligands [22]. The latter hypothesis is supported in GAS by the identification of extracellular Mg2+ and sub-inhibitory concentrations of LL-37, an antimicrobial peptide secreted by innate immune cells, as specific ligands that respectively activate and inhibit CovS activity [22], [29], [30].
Additional levels of TCS regulation exist that do not depend directly on environmental signal sensing by the HK sensor [31]–[33]. These cellular regulators target either the HK or the RR and were named “third components”, “auxiliary proteins”, “adaptors” or “TCS connectors”. While keeping the specific molecular links between an HK and its cognate RR, the additional cellular regulator allows to coordinate the cellular response and/or to integrate multiple signals [33]. This is the case in GBS where CovR activity is modulated by a second signaling pathway mediated by the serine/threonine kinase Stk1 [15], [25], [34]. Direct phosphorylation of the CovR T65 threonyl residue by Stk1 decreases CovR activity and interferes negatively with CovS-dependent phosphorylation of the D53 aspartyl residue [25]. These CovS- and Stk1- signaling pathways converge on CovR and are both necessary for GBS virulence [15].
In this study, we identified and characterized a transmembrane protein, Abx1, belonging to a large family of bacterial proteins of unknown function that have a conserved domain called Abi [35], [36]. We provide evidence that the putative protease activity of Abx1, inferred from its similarity to eukaryotic CaaX proteases [35], [36], is not necessary for its function. Instead, we demonstrate that Abx1 is an additional partner of the CovSR system that is necessary to regulate CovS activity by a protein-protein interaction. We show that the Abx1-CovS signaling complex and the Stk1-dependent pathway are both necessary to control CovR activity. In addition to defining the genetic network controlling virulence gene expression in GBS, we provide the first report of a cellular regulator of HK activity belonging to the large family of bacterial Abi-domain proteins.
The ß-hemolysin/cytolysin expressed by GBS is an important virulence factor encoded within a cluster of twelve genes forming the cyl operon [8], [37], [38]. Hemolytic activity of GBS is always associated with the synthesis of an orange pigment. Among the 12 genes of the cyl operon (gbs0644-gbs0655 genes), the CylE protein (gbs0651) is the critical determinant for the dual hemolytic/pigmentation phenotypes [8], [38]–[40]. Inactivation of cylE abolishes hemolytic activity and pigmentation, and the mutant is less virulent in animal models of systemic infections [39]–[41].
Screening of a collection of random Himar1 insertional mutants derived from the NEM316 wild-type (WT) strain allowed us to identify mutants that mapped in the cyl operon [40]. We extended this approach with approx. 2,500 new Himar1 mutants screened for hemolytic activity on blood agar plates and for pigmentation on Granada agar, a specific medium that stabilizes the GBS pigment [42]. We focused our analysis on four mutants displaying a strong decrease in pigmentation and hemolytic activity on Granada and Blood agar plates (see Figure 1A for two of these four phenotypically identical mutants). The four mutants have unique and independent Himar1 insertions in the uncharacterized gene gbs1532 as revealed by direct sequencing on chromosomal DNAs. This 921-bp gene encodes a 306 amino acid protein annotated as a hypothetical protein or putative protease (NCBI ref. seq. NP_735969). Transposon insertions of the four Himar1 mutants were in positions 85, 347, 625, and 632 in the DNA coding sequence, leading to truncated proteins at codons 29, 116, 209, and 211, respectively.
Blast similarity searches, domain analyses, and topology prediction (Figure S1) revealed that Gbs1532 belongs to a conserved superfamily of putative membrane-bound metalloproteases related to eukaryotic CaaX proteases [35], [36]. The characteristic domain of this superfamily of more than 5,500 proteins is the Abi-domain (Pfam02517). To our knowledge, none of the bacterial Abi-domain proteins are functionally characterized to date [36]. We therefore named the translated gbs1532 gene product Abx1, for Abi-domain protein related to CaaX protease, and renamed the corresponding gene accordingly (i.e., abx1). This abx1 gene belongs to the core genome of all sequenced GBS strains [43], [44] and 6 other genes coding for Abi-domain proteins similar to Abx1 are present in their genomes (Figure S1).
To confirm that abx1 is required for pigmentation and hemolysis, we constructed in-frame deletion mutant (Δabx1) in the NEM316 WT strain. The Δabx1 mutant displayed the same phenotypes as the Himar1 mutants (Figure 1B). Furthermore, pigment production and hemolysis were restored upon ectopic expression of abx1 transcribed from its own promoter (Figure 1A and 1B: pTCVΩabx1 complementing vector) confirming that abx1 was responsible for the observed phenotypes.
Wild-type or Δabx1 mutant strains expressing Abx1 from the complementing plasmid pTCVΩabx1 were more pigmented and more hemolytic than the control strains (Figure 1B). This suggests that the expression level of Abx1 might be limiting for pigment production and hemolysis. To test this hypothesis, we cloned abx1 downstream of two constitutive promoters with different strength (Ptet and Pcyl+) to generate the overexpressing vectors pTCVΩPtet_abx1, and pTCVΩPcyl+_abx1, respectively. Furthermore, we engineered the WT strain to replace the chromosomal Pabx1 promoter by the Pcyl+ promoter (ΔPabx1::Pcyl+). All recombinant strains carrying the different overexpression forms of abx1 displayed hyper-hemolysis on blood agar and hyper-pigmentation on Granada agar (Figure 1C). Determination of the hemolytic titer of the different mutants and quantification of the corresponding abx1 transcription levels further confirmed the link between hemolysis and abx1 expression (Table 1). Remarkably, the pigmentation level also correlated with the strength of the abx1 upstream promoter (Ptet>Pcyl+>Pabx1) on TH agar, a medium where pigmentation is usually not observed (Figure 1C). Overexpression of Gbs1037, the closest Abx1 homolog in GBS (40% identity over 222 residues, E-value = e-11; Figure S1) and of the unrelated EFGP (Enhanced Green Fluorescence Protein) did not modify pigmentation and hemolysis of NEM316, underlining the specific activity of Abx1 (Figure 1C). The activity of Abx1 is a conserved feature within the GBS species since abx1 deletion or overexpression in several clinical strains (serotypes Ia, 515; Ib, H36B; II, 18RS21; III, BM110; and V, 2603V/R; [44]) has a strong effect on their hemolytic activities and pigmentation (Figure S2).
In eukaryotes, Abi-domain-containing proteins (Pfam02517) are CaaX prenyl proteases involved in protein prenylation [35], [36], [45], [46]. These transmembrane proteases remove the carboxy-terminal aaX tripeptide of a protein after the addition of an isoprenyl group on the cysteyl residue of the CaaX motif. One of the main features of the Abi-domain is the presence of four predicted core transmembrane helical segments (labeled TMH1–4) containing conserved active-site residues (Figure S1D). The overall topology and the critical residues for protease activity are conserved in Abx1 (Figure 2A and S1). To date, prenylation was never described in prokaryotic cells but, intriguingly, the CylE hemolysin contains a cysteyl residue at the fourth position from its carboxylic end (i.e., the critical residue of the CaaX motif [46]).
We tested whether CylE could be post-translationally modified by a prenylation-like mechanism involving Abx1. Overexpression of Abx1 using pTCVΩPcyl+_abx1 in the non-hemolytic and non-pigmented ΔcylE mutant did not increase hemolysis or pigmentation (Figure 2B), confirming that Abx1 function is strictly dependent on the presence of a functional CylE. We then substituted the cysteyl residue of CylE located at the fourth position from its carboxy-terminal end by an alanyl (CylE C664A). This mutation did not alter pigmentation and hemolysis significantly (Figure 2B), which suggests that CylE is not prenylated on this cysteyl, and consequently that Abx1 might not be a CaaX protease. To further test whether Abx1 has protease activity, we replaced the conserved residues (glutamyl at positions 164 and 165 and histidyl at positions 197 and 235) described as critical for this enzymatic activity by alanyl residues (Figure 2A; [47]). All alleles obtained for Abx1 (E164A, E165A, EE164/165AA, H197A, H235A) restored hemolysis and pigment production when expressed in a Δabx1 mutant strain (Figure 2C). Hence, the putative protease activity of Abx1 is not necessary for the CylE dependent phenotypes.
Abx1 activity is dependent of CylE but appears to be independent of a CylE post-translational modification. To decipher the relationship between Abx1 and cyl-dependent hemolysis and pigment production, we took advantage of the previously described ΔCBSCyl mutant where the CovR binding sites (CBS) in the promoter of the cyl operon were deleted [40]. Since CovR acts as a transcriptional repressor of the cyl operon, deletion of CBSCyl leads to strong constitutive expression of the cyl operon. Consequently, the ΔCBSCyl mutant is hyperhemolytic and hyperpigmented in all tested conditions (Figure 3A). Whereas abx1 deletion in a WT background abolishes pigmentation and hemolysis, deletion of abx1 in a ΔCBSCyl background (Δabx1ΔCBSCyl) did not affect the hyper-hemolytic and hyper-pigmented phenotypes of the parental ΔCBSCyl strain (Figure 3A). This result suggests that Abx1 acts upstream of CovR-mediated transcriptional inhibition of the cyl operon.
To confirm this epistatic interaction, we inactivated covR (gbs1672) and the complete covSR two-component system (gbs1671–1672) in the WT and Δabx1 backgrounds. As expected, inactivation of covR or covSR in the WT strain led to hyper-pigmentation and hyper-hemolysis (Figure 3A). Similarly, inactivation of CovR or CovSR in the Δabx1 background restored pigmentation and hemolysis to levels similar to those observed in the ΔcovR and ΔcovSR mutants (Figure 3A). This suggests that Abx1 is an inhibitor of CovR activity. Consistently, Western blot analysis of the adhesin BibA and CAMP factor whose expression are negatively and positively regulated by CovR [11], [13], respectively, revealed similar expression patterns in the ΔcovR and abx1 overexpression mutants: inactivation of CovR or overexpression of abx1 dramatically increased BibA production and abolished secretion of the CAMP factor (Figure 3B). These results demonstrate that Abx1 activity is not restricted to pigmentation and hemolysin production, and suggest an antagonist function on CovR activity.
To decipher the relationship between Abx1 and the CovSR two-component system, we compared the transcriptomes of ΔcovS, ΔcovR and ΔcovSR mutants with the Δabx1 deletion mutant and two abx1 overexpression mutants (v_Oe_abx1 = vector based abx1 overexpression with the pTCVΩPcyl+_abx1 plasmid; and K_Oe_abx1 = chromosomal abx1 overexpressing strain ΔPabx1::Pcyl+). Pairwise comparisons of Log2 expression ratios (Table S1, N = 1,905 genes) revealed highly similar expression changes between the strain carrying the chromosomal overexpression of abx1 (K_Oe_abx1) and the ΔcovS and ΔcovR deletion mutants (Figure 4A: Pearson correlation = 0.704 and 0.727, respectively). Strikingly, the transcriptome profiles of the abx1 overexpression and ΔcovS deletion mutants form a distinct cluster as revealed by hierarchical clustering analysis (Figure 4B and S3A).
Detailed analyses revealed a cluster of genes containing the direct and conserved targets that are negatively regulated by CovR (Figure 4C and S3B: N = 29 and 40 genes with expression change log2>2 and >1 in at least one strain, respectively). Expression of nearly all of these genes, including the cyl operon (gbs0644–gbs0655) and the bibA gene (gbs2018), were dependent on the expression level of abx1. Moderate overexpression of abx1 in the chromosome (ΔPabx1::Pcyl+; 3.7 fold change by microarrays) induced a transcriptome response highly similar to that of the ΔcovS mutant (Figure 4C). Greater overexpression of abx1 (pTCVΩPcyl+_abx1; 9.1 fold change by microarrays) further increased expression of the direct-CovR regulated gene set to a level close to that observed in a ΔcovR mutant (Figure 4C). In contrast, downregulation below the significant threshold is observed in the Δabx1 mutant (Figure 4C). qRT-PCR analysis on a set of 6 genes regulated by the CovSR system confirmed gene expression changes identified by microarray analysis (Figure 4D). Moreover, bibA and CAMP factor transcription levels in the ΔcovR deletion and abx1 overexpression mutants are in accordance with the corresponding protein expression levels as observed by Western analysis (Figure 3B). Taken together, the similarities between abx1 overexpression mutants and cov deletion mutants suggest that Abx1 exerts a specific effect on the CovSR two-component system, most likely by acting as a CovS antagonist.
Our results suggest an inhibitory function of Abx1 on CovS activity or that Abx1 protects or sequesters CovR, impeding its phosphorylation by CovS either directly or indirectly. To discriminate between these possibilities, we used a bacterial two-hybrid system [48] to test the physical interactions of Abx1 with different putative partners. This showed that Abx1 is able to form homodimers and that it interacts indeed with CovS but not with CovR (Figure 5A and 5B). The Abx1-CovS interaction is specific, as Abx1 did not interact with two other GBS HKs that are similar to CovS (Figure 5A: Gbs2082 and Gbs0430). To define the regions involved in the Abx1-CovS interaction, we tested the interaction of Abx1 with different domains of CovS (Figure 5C). These experiments showed that Abx1 interacts specifically with the amino-terminal part of CovS containing its extracytoplasmic and transmembrane domains (Figure 5D). Deletion of the extracytoplasmic loop of CovS did not impede interaction with Abx1, suggesting that the loop is not directly required for Abx1-CovS interaction (Figure 5D: CovS form VI). No interaction was detected either with the carboxy-terminal part of CovS containing the catalytic domains or with CovS truncated forms containing only one transmembrane domain (Figure 5D). These results suggest that Abx1 interacts with the two CovS transmembrane domains involved in signal processing.
In agreement with the proposed inhibitory function of Abx1 on CovS, deletion of covS in a Δabx1 background (Δabx1ΔcovS) restores pigment production and hemolysis of the non-pigmented and non-hemolytic Δabx1 mutant to levels similar to those of the ΔcovS mutant (Figure 6A). However, as seen previously by transcriptome analysis, the ΔcovS mutant is not identical to the ΔcovR mutant. In particular, the ΔcovR mutant is hyper-pigmented and hyper-hemolytic (see Figure 3A) while the ΔcovS mutant is only slightly affected in pigmentation and hemolysis compared to the WT strain (Figure 6A).
To explain these observations, we must consider the three common activities of histidine kinases (HK): auto-phosphorylation of the kinase on a histidyl residue, phospho-transfer on an aspartyl residue of the cognate regulator, and phosphatase activity of the HK on its cognate regulator [14]. Deletion of covS abolishes the three activities. In this case, cross-phosphorylation of the regulator by small metabolites or by a non-cognate HK can be observed due to the absence of the phosphatase activity of the cognate HK [49], [50]. Thus, in the ΔcovS mutant, cross-phosphorylation of CovR might enable its partial activation. To gain experimental evidence, we constructed CovS and CovR alanyl substitution mutants in conserved residues critical for each of the three activities associated with HK [14]. First, we targeted the phospho-acceptor histidyl residue H278 to obtain an auto-phosphorylation deficient allele of CovS. Second, we replaced the aspartyl residue D53 of CovR normally phosphorylated by CovS to abolish phospho-transfer between CovS and CovR [25], [26]. Third, we replaced the CovS threonyl residue T282 that was proposed to be specifically involved in HK phosphatase activity [50], [51]. We then used pigmentation and hemolysis production as reporters of CovSR activity (Figure 6B).
The CovR D53A mutant displayed a hyper-hemolytic and hyper-pigmented phenotype like the ΔcovR mutant (Figure 6B), as also observed previously in another GBS strain [25], supporting the model of CovR as a transcriptional repressor under its D53P phosphorylated form. Analyses of the CovS H278A mutant strain showed phenotypes similar to the ΔcovS mutant (Figure 6B), suggesting that the H278 residue is essential for all three activities, as assumed by a mechanistic model of HK activity [50], [51]. We identified in CovS the conserved threonyl residues specifically required for the phosphatase reaction of several HK [50], [51]. Interestingly, the CovS T282A mutant is non-hemolytic and non-pigmented (Figure 6B) indicating that CovR is locked in an active form that fully inhibits the expression of the cyl operon. Overexpression of abx1 in the ΔcovS, CovS H278A or CovS T282A mutants did not modify their phenotypes, indicating that Abx1 activity is dependent on CovS kinase/phosphatase activities (Figure 6C). The fact that overexpression of abx1 in the WT context led to hyper-pigmented and hyper-hemolytic phenotypes, as seen for the ΔcovR and CovR D53A mutants (Figure 6), suggests that Abx1 actively stimulates CovS-dependent dephosphorylation of CovR. Expression level measurements of selected CovR-regulated genes (cylE, cylJ, bibA, gbs0791, gbs1037, and cfb coding the CAMP factor) in CovS and CovR alanyl substitution mutants confirm that the broad effect of abx1 overexpression depends upon a functional CovS histidine kinase (Figure 6D). Overall, our results support a model where Abx1 inhibits the kinase-competent form and/or stabilizes the phosphatase-competent form of CovS.
An additional regulator of the CovSR system is the eukaryotic-like serine/threonine kinase Stk1. It has been shown in vitro that Stk1 regulates CovR negatively by direct phosphorylation of the threonyl residue at position 65 (T65) [25], [34]. The CovR T65 phosphorylation by Stk1 has been proposed to be mutually exclusive of CovR D53 phosphorylation by CovS [25]. Consistently, the Δstk1 (Δgbs0307) mutant is strongly and negatively affected in pigment production and hemolytic activity (Figure 7), in agreement with an inhibitory role of Stk1 on CovR activity.
To learn whether Abx1 and Stk1 function together, we tested if Abx1 and Stk1 interact using the bacterial two-hybrid system. No interaction was observed (data not shown). Furthermore, pigmentation and hemolysis were restored in the non-pigmented and non-hemolytic Δstk1 strain upon increasing abx1 copy number with plasmid pTCVΩabx1 (Figure 7). This indicates that Abx1 is able to bypass CovR activation observed in the absence of Stk1. However, pigmentation and hemolysis levels of the Δstk1 strain carrying the pTCVΩabx1 vector are below those of the WT strain (Figure 7). This result suggests that abx1 overexpression needs a functional Stk1 to fully inhibit CovR activity.
Finally, a condition-dependent toxicity was observed when the overexpressing pTCVΩPcyl+_abx1 vector was introduced into the Δstk1 mutant. Specifically, strong overexpression of abx1 in Δstk1 is toxic on Granada agar (Figure 7). This phenotype is CovS-dependent as toxicity is alleviated in a Δstk1ΔcovS double mutant, which is unresponsive to abx1 overexpression (Figure 7). This result reinforces the link between Abx1 and a functional CovS. In addition, the Δstk1ΔcovS mutant is locked in an intermediate phenotype (Figure 7), revealing a partial inhibition of CovR in the absence of CovS. Thus, CovS appears necessary for complete activation of CovR in the absence of Stk1, most likely by an active CovR D53 phosphorylation. As shown in the model depicted Figure 8, we propose that the regulatory function of Abx1 is absolutely dependent on CovS but also depends on the serine/threonine kinase Stk1 to control CovR activity by a convergent signaling pathway.
The CovSR two-component system is a master regulator of GBS virulence gene expression. In order to assess the role of Abx1 in GBS virulence, neonatal rat pups were infected by intraperitoneal injection of 5×106 bacteria of either the Δabx1 deletion or abx1 overexpressing mutants. In this model, both mutants exhibited reduced virulence compared to the parental WT strain (Figure 9). It is noteworthy that the abx1 overexpression mutant is totally avirulent (100% survival) and that neonates did not display any sign of illness during the course of two independent experiments (data not shown). These results show that abx1 expression at an appropriate level is required to develop GBS infection.
The CovSR two-component system (TCS) is the major regulator of virulence gene expression in GBS. In this study, we identified the Abx1 transmembrane protein as essential and limiting for CovSR activity due to its interaction with the CovS histidine kinase (HK). Regulation of HK activity by an interacting protein is increasingly recognized when studying TCS signaling in a cellular context [31]–[33]. Instead of being a linear signaling pathway, our data show that the CovSR TCS is embedded in a regulatory network involving several regulatory proteins, including the Abi-domain protein Abx1.
We first observed that Abx1 is required for hemolysin and pigment production in GBS and that abx1 overexpression in a WT background increases hemolysin and pigment production in a dose-dependent manner. These abx1 effects depend upon a functional CovSR pathway, which directly controls expression of the cyl operon. In agreement with a regulatory activity of Abx1 on the CovSR system, modulation of abx1 expression affected not only hemolysin/pigment production but also production of the adhesin BibA and the CAMP factor, two direct targets of CovR [11], [13]. Indeed, genome-wide transcriptome profiling suggested a specific and antagonistic activity of Abx1 on the CovSR system.
By combining genetic analysis and protein-protein interactions, we show that Abx1 acts through the HK CovS to control the activity of CovR. The transmembrane protein CovS is the cognate HK of CovR. However, in contrast to the majority of TCS, CovR is mainly a transcriptional inhibitor when phosphorylated on the conserved D53 residue and CovS is assumed to be primarily a phosphatase necessary to relieve repression of virulence genes during infection [11]–[13], [24]. Our data support this regulatory logic (i.e., the inactivation by CovS of the CovR inhibitor) but show in addition that Abx1 exerts a cellular control on the dual kinase/phosphatase activity of CovS (Figure 8). By using the cyl operon as reporter of CovR activity, we observed constitutive CovR activation with a CovS allele (CovS T282A) mutated in the conserved threonyl residue specifically involved in the phosphatase activity of several other HKs [50], [51] or by deletion of Abx1 (Figure 8B). Conversely, CovR can be inactivated by a D53A substitution, mimicking a non-phosphorylated allele, or by overexpression of Abx1 (Figure 8C). Furthermore, all of these Abx1 effects are dependent on a functional CovS. Taken together, these results suggest that the level of Abx1 is critical to control the equilibrium between the kinase and phosphatase activities of CovS.
It is noteworthy that inactivation of both CovS kinase and phosphatase activities, either with a CovS H278A allele or by deleting covS, led to intermediate phenotypes compared to those obtained by CovR inactivation. This pattern has been observed previously in GBS [13] and GAS for the CovSR ortholog [24], [52], [53]. In this condition, it is likely that CovR is partially activated by cross-phosphorylation, either by small metabolites like acetyl-phosphate or by a non-cognate HK. The specific phosphatase activity of CovS avoids inappropriate activation of CovR by other pathways, a general mechanism to insulate TCS signaling pathways [49].
Importantly, HK phosphatase activity is not simply the reverse reaction of the kinase activity [50], [51]. The two activities involve different, but potentially overlapping, catalytic residues [50] and are mutually exclusive since they depend on different conformations of HK homodimers [14]. Both Abx1 and CovS are transmembrane proteins. We found that Abx1 interacts directly with CovS input domains (i.e., the extracellular and transmembrane domain), and ompacts CovS output domains (i.e., the catalytic cytoplasmic domain). Current models of HK activity regulation are based on different conformational states of HK homodimers [14]. It is therefore most likely that Abx1 interferes with the dynamic process of CovS conformational changes propagated along the whole proteins to control the kinase/phosphatase switch [14]. In the absence of Abx1, CovS appears to be locked into a kinase-competent conformation while overexpression of Abx1 will result in stabilization of the phosphatase-competent conformation (or destabilization of the kinase-competent form). Of note, the CovS extracellular loop is not necessary for the physical interaction with Abx1 and it is therefore likely that the main domains of CovS interacting with Abx1 are the transmembrane domains flanking this loop. These HK transmembrane domains control the conformational rotation of the cytoplasmic subunits of HK homodimer by a rotation or a piston mechanism [54], [55]. Thus, intra- and inter-molecular interactions interfering with this dynamic process will impact the catalytic activity by blocking HKs in a kinase or a phosphatase form [56], [57].
Positive and negative regulation of HK activity by physical interactions with transmembrane regulators has been described in both gram-negative and gram-positive bacteria [31], [32]. Members of this category are mainly small hydrophobic peptides, like the PhoQ-B1500/SafA [58], PhoQ-MgrB [59] and EnvZ-MzrA [60] pairs in enteric bacteria, or proteins with a single transmembrane domain, like the YycH and YycI proteins regulating the essential HK YyfG in Bacillus subtilis [61], [62]. In particular, structural modeling suggests that the YyfG HK is inactivated by its interactions with YyfH and YyfI in the membrane environment [62]. In analogy, this suggests a mechanism of CovS inactivation by its interaction with the eight predicted transmembrane spanning domains of Abx1 (Figure S1).
Abx1 controls GBS virulence gene expression by regulating CovS activity, via a mechanism independent of the direct sensing of an environmental signal by CovS. This cellular regulation of TCS by a “third” component raises the question of how this additional regulator is itself regulated [32], [33]. The HK regulator may set up a feedback loop when located in the same operon as the HK, as for B. subtilis YycHI [61], [62], or it may act as a TCS connector when its transcription is controlled directly by another TCS [31] as for E. coli B1500/SafA, MgrB, and MzrA peptides mentioned above [58]–[60]. We did not find evidence that abx1 transcription is directly controlled by CovR (data not shown) but cannot exclude transcriptional regulation by an as yet unidentified RR.
An alternative hypothesis would be an allosteric regulation and/or a scaffolding function of Abx1 [33]. Regulation of HK activity by signal(s) not directly recognized by the HK but sensed and/or transmitted by its interactor(s) might be a common mechanism, as already described for antimicrobial cationic peptide resistance modules in Firmicutes [63], [64] and for HK-HK complex formation in Pseudomonas aeruginosa [65]. Until now, it is not known how CovS activity is regulated in GBS. In GAS, the CovS ortholog (displaying 50.3% identity with the GBS CovS) is thought to be a direct sensor of magnesium and of the human antimicrobial peptide LL-37 [22], [29], [30]. However, only two Abi-domain proteins are present in GAS genomes and neither is an Abx1 ortholog (Figure S1). Although the biological function of the CovS orthologs in virulence is conserved [19], [20], GAS and GBS inhabit different ecological niches (nasopharynx and skin for GAS; gut and vagina for GBS), suggesting a species-specific response by CovS. The identification of conditions bypassing the Abx1 effects in GBS will be valuable for characterizing the putative signal(s) directly sensed by CovS.
We found that activity of the Abx1-CovS module depends upon a second signaling pathway mediated by the serine/threonine kinase Stk1 [25], [34]. The two pathways converge on CovR to regulate virulence gene expression (Figure 8). In vitro, direct phosphorylation of the CovR T65 threonyl residue by Stk1 interferes negatively with phosphorylation of the conserved D53 aspartyl residue and decreases CovR affinity for its target promoters [25], [34]. By genetic analysis, we found that the Abx1-CovS and the Stk1 signaling pathways are dependent on each other to fulfill complete activation or inhibition of CovR. CovR phosphorylation on T65 and D53 are mutually exclusive [25], strongly suggesting that CovR has three possible states (Figure 8D: D53P T65 = active; D53 T65 = inactive I dephosphorylated; D53 T65P = inactive II). It should be noted that Stk1 has additional direct targets beyond CovR and that its activity depends upon its associated phosphatase Stp1 [66]–[69]. Interestingly, a Δstk1 mutant is affected in cell cycle progression ([66] and data not shown), as observed in S. pneumoniae and other bacteria [70], [71], and Stk1 might respond to the release of cell wall components into the extracellular environment by growing bacteria [72]. Therefore, it appears that CovR integrates at least two types of signals to coordinate virulence gene expression (via Abx1-CovS) with the control of bacterial division (through Stk1).
This integrated system might be necessary to accurately control GBS lifestyle from commensalism to invasive infection [15]. The attenuated virulence of both deletion and overexpression abx1 mutants supports this dynamic view of the infectious process. The survival of animals challenged intraperitoneally with the overexpressing abx1 mutant is reminiscent of the decreased virulence of ΔcovR mutants previously described [11]–[13]. Several virulence genes controlled by CovR are involved at several steps of the infectious process, ranging from stress resistance to adherence to host cell [10], [11], [13], [16], [18], [73]. Among the genes directly regulated by CovR, the hemolysin/cytotoxin CylE has key functions [39]. At sub-lytic concentrations, CylE induces an anti-inflammatory response [74], while high CylE expression induces a pro-inflammatory response [41]. The inappropriate activation of a strong pro-inflammatory response during initial steps of the infectious process might explain, at least in part, the loss of virulence of mutants overexpressing abx1 or inactivated for covR. In contrario, at a latter stage of infection, these mutants can be more virulent due to the cytotoxic effect of CylE on blood cells, as seen with a ΔcovR mutant injected intravenously in mice [15]. Overall, these results indicate that the Abx1-CovS-CovR-Stk1 regulatory network controls the fine-tuning of virulence gene expression which is critical for GBS disease progression.
The demonstration that Abx1 is a cellular regulator of CovS represents a new function for a bacterial Abi-domain protein. The Abi-domain (Pfam02517) of Abx1 is the characteristic domain of a large transmembrane protein family (>5,000 sequences) with up to 15 members per bacterial species [35], [36]. This family is mostly uncharacterized to date in prokaryotes but includes the eukaryotic type II CaaX proteases involved in protein prenylation [36], [45]. However, prenylation appears restricted to eukaryotes and the two examples of prenylated bacterial proteins, SifA of Salmonella typhimurium and AnkB of Legionella pneumophila, are bacterial effectors modified by the eukaryotic enzymes after their injection into the host cell [75]–[77]. Our characterization of targeted mutations (Figure 2) did not sustain our initial hypothesis of a prenylation-like modification of the CylE hemolysin. Moreover, the putative catalytic residues inferred from similarities with CaaX proteases are not necessary for Abx1 activity. Given the high conservation of these residues (Figure S1), we cannot rule out that Abx proteins have protease activity. However, the main function of Abx1 as a regulator of the CovS histidine kinase does not require this putative protease activity.. Strikingly, the absence of a protease activity was already suggested for a Staphylococcus aureus Abi-protein involved in lysostaphin resistance [78]. Interestingly, three of the four S. aureus Abi-domain genes were identified recently in a genome-wide screen for altered targeting of cell-surface proteins [79]. These defects are mainly due to decreased transcription of the corresponding genes in the absence of the Abi-domain proteins [79], suggesting a conserved mechanism in gene transcriptional regulation.
Finally, it should be mentioned that about one-fifth (>1,000) of bacterial Abi-domain proteins are annotated as “abortive infection protein” due to a historical miss-annotation [36] leading to confusion between Abi phenotypes [80] and Abi-domain proteins [36]. It was also suggested that Abi proteins are involved in self-immunity against bacteriocins [81]. However, only a small subset of Abi-proteins (less than 1%) is localized in putative bacteriocin-encoding loci [81] or in operons encoding small toxins [82], [83]. In particular, the gene encoding the Abi-domain protein, SagE is located in the operon directing synthesis of the S. pyogenes ß-hemolysin known as streptolysin S or SLS [8], [84]. However, genetic analysis of this locus did not provide any conclusive clues about the SagE function [84]–[86]. Abi-domain proteins localized in toxin operons remain to be experimentally characterized for their involvement in toxin production and/or regulation.
Studies of TCS signaling in cellular context have identified a great variety of additional TCS partners, targeting either the HK or the RR [31], [32]. In this study, we demonstrated that the Abi-domain protein Abx1 is a regulator of the HK CovS in GBS. The CovSR system was presumed to be a linear signaling pathway. We show here that the CovSR TCS is the core of a signaling network and propose a new regulatory model, depicted in Figure 8, which includes Abx1-dependent CovS regulation and Stk1-dependent CovR regulation [25], [34]. Deciphering the additional CovSR input/output associated with these cellular regulators might prove useful for understanding the GBS transition from commensalism to virulence. Strikingly, Abx1 belongs to a very large family of multi-spanning transmembrane proteins that remain mostly uncharacterized to date [35], [36] and are highly conserved at the species level. It is therefore tempting to extend our findings to other members of this family. The characterization of other Abx-like proteins in GBS (Figure S1) and in other bacteria will reveal whether they belong to a functionally conserved family of HK regulators.
Animal experiments were performed at the Institut Pasteur (Paris, France) animal husbandries in accordance with the policies of the European Union guidelines for the handling of laboratory animals (http://ec.europa.eu/environment/chemicals/lab_animals/home_en.htm) and were approved by the Institut Pasteur animal care and use committee (N°04.118).
The reference WT GBS strain used in this study is NEM316, a ST-23 and serotype III clinical isolate responsible for a fatal case of septicaemia, whose sequenced genome [43] is accessible (NCBI RefSeq NC_004368.1). The relevant characteristics of the bacterial strains and plasmids used in this study are summarized in Table S2. GBS was cultured in Todd Hewitt (TH) broth (Difco Laboratories) at 37°C without agitation. Escherichia coli DH5α (Invitrogen) and XL1 Blue (Stratagene) were grown in Luriani Broth (LB) medium. When specified, antibiotics were used at the following concentrations: for E. coli: ticarcillin, 100 µg/ml; erythromycin, 150 µg/ml; kanamycin, 25 µg/ml; for GBS: erythromycin, 10 µg/ml; kanamycin, 1,000 µg/ml.
Pigmentation and ß-hemolytic activity were detected on Granada agar and Columbia agar supplemented with 5% horse blood, respectively (BioMérieux, France). Overnight GBS cultures in TH were washed, adjusted to 108 CFU/ml, serially diluted (dilution factor 10) in microplates and spotted on appropriate agar plates. Horse blood plates were incubated at 37°C for 16–24 h, followed by an additional 16–24 h incubation at room temperature or 4°C depending on the hyper- or hypo-hemolytic mutants being tested. To highlight the halo of lysis due to the ß-hemolysin, photographic acquisitions of blood agar plates were taken with light from below. Whole images were further converted to gray scale and processed (Photoshop CS4, Adobe, US) to adjust contrast and brightness. Granada plates were incubated 20–36 h at 37°C in anaerobic condition (AnaeroGen, Oxoid, UK). Photographic acquisitions of TH and Granada plates were done with side lighting from above and whole images were processed for contrast and brightness.
GBS hemolytic titers were determined by a semi-quantitative method as described previously, with slight modifications [11]. Briefly, culture adjusted to 2.108 CFU/ml in Phosphate Buffer Saline (PBS) solution supplemented with 0.2% glucose were serially diluted (dilution factor = 2) in microplates. One volume (100 µl) of 1% defibrinated horse blood (Oxoid, UK) in 0.2% glucose PBS was added in each well and plates were incubated 1 h at 37°C. After gentle centrifugation (5 min. at 1000 rpm) to pellet unlysed cells, the amount of hemoglobin in 100 µl of supernatants was quantified by optical absorbance at 420 nm. Hemolytic activity of each strain was defined as the minimum dilution that lysed at least 10% of red blood cells. Hemolytic titers were defined by the ratio between hemolytic activities of each strain against the hemolytic activity of the WT strain (titer X = activity X/activity WT). Negative and positive controls were PBS ( = 0% lysed cells) and 0.1% SDS ( = 100% lysed cells) instead of bacterial cells, respectively. All assays were performed in triplicates with independent cultures.
Purification of GBS genomic DNA and E. coli plasmids DNA was done on columns following manufacturer instructions (DNeasy Blood and Tissue kit and Quiaprep Spin Minipreps kit, respectively, Qiagen). Oligonucleotides (MWG and Sigma) used in this study are listed in Table S3. Analytical PCR used standard Taq polymerase (Invitrogen, Life technologies) and preparative PCR for cloning and sequencing was carried out with a high fidelity polymerase (Phusion, Finnzymes). Sanger sequencing was done at the Institut Pasteur sequencing core facility (Paris, France) using the ABI PRISM 3.1 dye terminator cycle sequencing kit (Applied Biosystems) or outsourced (Beckman Coulter Genomic, UK).
Plasmids for overexpression and complementation (pTCV backbone, Table S2), deletion (pG+host5 backbone, Table S2) and for bacterial double hybrid (pUT18 and pKNT25 backbones, Table S2) were constructed by standard cloning procedures and all inserts were fully sequenced. Primer pairs, DNA matrix and restriction enzymes used for each vector are detailed in Table S4. For several constructs, we used a splicing by overlap-extension method [87] to generate in-frame deletion cassettes, site-directed mutagenesis inserts or translational fusions (Table S4). For instance, deletion cassettes for chromosomal in-frame deletions were generated by combining two 500 bp PCR products, corresponding to genomic sequences flanking the region to be deleted, that were designed to have 25–50 bp of homology with each other at one end [87]. The 1 kb amplification products obtained with external primers were subsequently cloned into the thermosensitive shuttle plasmid pG+host5 backbone and sequenced (Table S4).
Plasmids were introduced in GBS by electroporation. For chromosomal gene inactivation/modification, allelic exchanges were selected as described [88], [89]. Deletions and site-directed substitutions were confirmed by sequencing PCR products obtained with primers designed outside the genomic region used for the construction of the corresponding cassette (Table S3 and S4).
Secreted proteins were purified from 40 ml TH broth cultures collected at mid-exponential phase (DO600 nm = 0.5). Supernatants were filter sterilized and concentrated 50-fold using Vivaspin 20 columns (Sartorius). Cell-wall proteins were prepared from overnight cultures at 37°C in TH by mutanolysin (Sigma-Aldrich) digestion in osmo-protective buffer as described [73], [89]. Following SDS-PAGE electrophoresis, proteins were transferred onto a nitrocellulose membrane (GE Healthcare) and detected using rabbit specific polyclonal antibodies [73] and horseradish peroxidase (HRP)-coupled anti-rabbit secondary antibodies. Signals were detected by chemiluminescence (ECL, GE Healthcare).
Total RNAs were extracted from exponentially growing cells (OD600 = 0.4–0.5) in TH at 37°C with a phenol/Trizol-based purification method as previously described [11]. Reverse transcription was done with Superscript indirect cDNA kit (Invitrogen, Life technologies) and qPCR was carried out with SYBR Green PCR kits (Applied Biosystems, Life technologies). Relative quantification of specific gene expression was calculated with the 2−ΔΔCt method, with gyrA as the housekeeping reference, and normalized against the NEM316 wild-type. Each assay was performed at least in triplicate on three independent cultures.
For microarray analysis, identical culture conditions and RNA purification procedures were used. The three independent cDNA preparations of each strain were labeled with Cy5 or Cy3 (Amersham Biosciences) for dye swap hybridizations against NEM316 wild-type strain RNA prepared in parallel. The 15K custom microarray (Agilent Technologies) contains 8,691 60-mer oligonucleotides specific for the 2134 predicted genes and long intergenic region of the NEM316 strain [90]. Arrays were scanned in an Axon 4000B dual laser scanner and treated as described [90]. Raw data have been submitted to the ArrayExpress database under the Accession number E-MEXP-3703 (http://www.ebi.ac.uk/arrayexpress/). Hierarchical clustering were done with Cluster 3.0 and visualized with Java TreeView.
For BACTH assays [48], chimeric GBS proteins fused to either T18 or T25 fragments of adenylate cyclase (CyaA) were constructed by conventional cloning in pUT18 and pKNT25 vectors for T18/25 C-terminal tagging or in pUT18C and pKT25 for T18/25 N-terminal tagging (Table S2 and S4). Physical interactions were assayed in E. coli DHT1 cells (Table S2) co-transformed with recombinant pKNT25/pKT25 and pUT18/pUT18C vectors. Interaction efficiencies between hybrid proteins were quantified by measuring ß-galactosidase activity in 96-well plates assays. Bacteria were grown overnight at 30°C in 1 ml LB broth in the presence of 0.5 mM IPTG and appropriate antibiotics in 2.2 ml 96-deepwell plates (Thermo Scientific). To permeabilize cells, 100 µl of bacterial cultures in 96 deepwell polypropylene plates were treated with 500 µl of buffer containing ß-mercaptoethanol (50 mM), SDS (0.2% v/v) and chloroform (3% v/v) and vortex two times 1 min (Mixmate, Eppendorf). For the enzymatic reaction, 40 µl of ONPG (4 mg/ml) was added to 120 µl of the permeabilized cells solution in a new microplate. Reaction kinetics at 28°C were followed by recording the OD420 every 4 min for 60–90 min in a microplate reader (Synergy, BioTek). Slopes (min−1) with a correlation coefficient r2≥0.98 were used to calculated enzymatic activities relative to internal negative (pUT18 and pKNT25 empty vectors; activity = O arbitrary units) and positive controls (pUT18-zip and pKNT25-zip vectors, [48]; activity = 1,000 arbitrary units) incorporated in each microplate. Slopes420 was divided by the OD600 of each culture to normalize for initial cell density. At least four independent cultures were done for each plasmid combination. All recombinant pKNT25/pKT25 and pUT18/pUT18C vectors were tested against the corresponding empty vector and gave background level of ß-galactosidase activity (not depicted in Figure 5).
Two-days old neonatal Sprague-Dawley rat pups (Janvier, Le Genest Saint Isle, France) were used for mortality curve experiments. Randomized groups of 10 neonatal rat pups were infected by intraperitoneal (i.p.) injection with 5×106 bacteria in 100 µl PBS. Survivals were monitored for five days after injection and two independent experiments were carried out.
|
10.1371/journal.pntd.0002289 | Sensitivity and Specificity of a New Vertical Flow Rapid Diagnostic Test for the Serodiagnosis of Human Leptospirosis | Background: Leptospirosis is a growing public health concern in many tropical and subtropical countries. However, its diagnosis is difficult because of non-specific symptoms and concurrent other endemic febrile diseases. In many regions, the laboratory diagnosis is not available due to a lack of preparedness and simple diagnostic assay or difficult access to reference laboratories. Yet, an early antibiotic treatment is decisive to the outcome. The need for Rapid Diagnostic Tests (RDTs) for bedside diagnosis of leptospirosis has been recognized. We developed a vertical flow immunochromatography strip RDT detecting anti-Leptospira human IgM and evaluated it in patients from New Caledonia, France, and French West Indies.
Methodology/Principal Findings: Whole killed Leptospira fainei cells were used as antigen for the test line and purified human IgM as the control line. The mobile phase was made of gold particles conjugated with goat anti-human IgM. Standards for Reporting of Diagnostic Accuracy criteria were used to assess the performance of this RDT. The Microscopic Agglutination Test (MAT) was used as the gold standard with a cut-off titer of ≥400. The sensitivity was 89.8% and the specificity 93.7%. Positive and negative Likelihood Ratios of 14.18 and 0.108 respectively, and a Diagnostic Odds Ratio of 130.737 confirmed its usefulness. This RDT had satisfactory reproducibility, repeatability, thermal tolerance and shelf-life. The comparison with MAT evidenced the earliness of the RDT to detect seroconversion. When compared with other RDT, the Vertical Flow RDT developed displayed good diagnostic performances.
This RDT might be used as a point of care diagnostic tool in limited resources countries. An evaluation in field conditions and in other epidemiological contexts should be considered to assess its validity over a wider range of serogroups or when facing different endemic pathogens. It might prove useful in endemic contexts or outbreak situations.
| The major burden of leptospirosis happens in low-income populations from tropical or subtropical regions. Because of nonspecific symptoms in human leptospirosis, the biological confirmation is needed to ascertain the disease. However, this biological diagnosis relies on sophisticated and time-consuming techniques that are most often hardly (if ever) available to clinicians in peripheral health centers. Yet, the outcome of leptospirosis in humans largely depends on an early antibiotic treatment. Taken together, these factors highlight the need of rapid simple diagnostic tests for leptospirosis that could be used directly on the bedside even in remote health centers. In this study, we developed and evaluated a prototype point of care strip test for the serological diagnosis of human leptospirosis in New Caledonia, mainland France, and the French West Indies. The sensitivity was 89.8% [95% CI, 84.7–93.4] and the specificity 93.7% [95% CI, 89.65–96.2]. This easy, early and portable diagnostic test will be evaluated in other epidemiological conditions and under field conditions.
| Leptospirosis is a bacterial disease of high incidence in many tropical and sub-tropical areas [1], [2]. Its clinical presentation is both highly variable and most often characterized by non-specific signs and symptoms; the complete triad of Weil's disease (hepatic failure, renal failure and hemorrhage) is recognized to account for less than one third of human cases [3], [4]. Most of the early signs and symptoms point to the so-called “acute fever of unknown origin” (FUO), a major diagnostic challenge in tropical and subtropical areas. Beside the epidemiological context and patient exposure history, to quickly diagnose and implement an appropriate treatment, an etiological investigation is necessary, especially in malaria, hantavirus, or viral hepatitis endemic regions or during influenza, chikungunya or dengue outbreaks. Leptospirosis is also reported to be an emerging or re-emerging disease in industrialized countries, with probable increasing impacts due to global warming and increasing travel-related cases [5], [6]. Unfortunately, the biological confirmation of leptospirosis is both tedious and rarely available in a timely manner. It notably requires sophisticated techniques that are most frequently available only in central reference laboratories [1]. These techniques are of prime importance in disease surveillance and epidemiological investigations but are inappropriate for early clinical care in peripheral health centers that support a major part of the leptospirosis burden. Additionally, an early and proper antibiotic treatment is known to be a key to rapid recovery and a major determinant of outcome in leptospirosis [3], [7], [8]. The need for portable rapid diagnostic tests (RDT) is striking and largely recognized to improve clinical management of leptospirosis patients, notably in remote dispensaries of tropical and subtropical regions. As an example, the major part of the leptospirosis burden in New Caledonia occurs away from the city and the central hospital of Noumea [9], [10].
In this study, we developed a vertical flow immunochromatographic RDT for the early diagnosis of leptospirosis to detect Anti-Leptospira human IgM and evaluated its sensitivity, specificity, reproducibility, repeatability, temperature stability and simulated shelf-life in the context of leptospirosis endemic (New-Caledonia and French West Indies) or non-endemic (mainland France) countries using clearly defined case definitions and the reference Microscopic Agglutination Test (MAT) results as the gold standard [4].
The antigen was prepared at the Institut Pasteur, National Reference Center for Leptospirosis (Paris, France) as follow: a culture of L. fainei serovar Hurstbridge strain BUT 6T at OD = 0.5 at 420 nm was killed by 0.2% formalin for 3 hours at ambient temperature, boiled for 45 minutes and its pH adjusted to 9.6. This preparation was kept at 4°C until use as an antigen for the Vertical Flow RDT and was also used for an in-house IgM ELISA at the National Reference Center [11](Bourhy et al., manuscript in revision).
The positive control line was made of purified human IgM (MP Biomedicals) at 2 mg/mL. Both control and test lines were sprayed as lines onto nitrocellulose membranes. Gold particles labelled with goat anti-human IgM (BBI International BA.GAHM40/X) adjusted to the concentration of OD520 nm = 3 were used as the capture mobile phase to construct our one-step vertical flow immunochromatography RDT, as previously described [12].
Two batches of RDT were produced at the Institut Pasteur in Paris (platform 5) and used in this study: batch numbers 110,211 and 120,511.
Leptospirosis cases were defined as confirmed when a clinical and epidemiological suspicion was complemented by either a positive specific PCR evidencing genomes of pathogenic Leptospira spp. in the blood or urine of the patient, or when the MAT on acute and convalescent sera showed a seroconversion (from nil to a titer ≥400) or a significant seroascencion (at least a fourfold raise in titers) [13]. Because most of the serum specimens originated from leptospirosis endemic regions where a high MAT titer threshold is usually used, we adopted this ≥400 threshold throughout the study. Probable cases were defined as clinical and epidemiological suspicion together with a unique serum with a MAT titer ≥400. The panels of strains used for MAT were adapted to the local epidemiology of New Caledonia, mainland France and French West Indies [9], [14] and are provided in details in Table S1. High rates of agglutination of the serum with one particular strain are used to identify the presumptive serogroup of the infecting bacterium as described elsewhere [4].
All sera used in this study were addressed to the Institut Pasteur in Nouméa or Paris for diagnostic purpose and originated from patients from New Caledonia, mainland France or the French West Indies (Martinique and Guadeloupe). The workflow is summarized as a flowchart in Figure S1. These laboratories are territorial (New Caledonia) and national (France and West Indies) reference centers for leptospirosis and receive all (New Caledonia) or more than 75% (France and French West Indies) of leptospirosis diagnosis requests. Sera were stored at −20°C, selected according to case definitions, and then tested blindly with RDTs. To assess the sensitivity of the RDT, only MAT-positive sera from confirmed cases were used: we tested 187 confirmed leptospirosis cases sera with a MAT titer ≥400 (120 from New-Caledonia, 38 from mainland France, 29 from French West Indies). The specificity was assessed using 221 sera (142 from New-Caledonia, 79 from mainland France): 12 anti-Chikungunya virus IgM positive sera, 58 anti-dengue virus IgM positive sera from all 4 serotypes, 6 anti-hepatitis A virus total Ig positive sera, 7 rheumatoid factor positive sera, 25 syphilis (TPHA and VDRL) positive sera, one acute malaria serum and 112 sera from healthy blood donors (Platform ICAReB, Institut Pasteur). These 221 sera were then confirmed to be MAT-negative (titer<100) within the same week.
Preliminary experiments determined dilution of sera between 1/100 and 1/800 in Phosphate Buffer Saline (PBS, pH 7.4) as suitable for testing with RDT. Because rapid tests are mostly to be used in endemic regions and similarly to the high MAT titer threshold, a 1/400 dilution of sera was used throughout the study. Briefly, Vertical Flow RDT strips were introduced into 200 µL diluted serum in 5 mL polystyrene tubes, for 15 minutes. The strips were then removed and placed on absorbent paper and read within 5 minutes. All results were recorded using a grading scale from 0 (no visible trace on test band) to 3+ (intensity of the test band equal to the intensity of the control band). The grading included a “weak” value for low but visible traces on the test band. Weak, 1+, 2+ and 3+ were then considered positive for further analysis and 0 was considered as negative.
All analyses were run blindly: any person involved in one particular analysis had no access to the results of the other tests results from the same serum.
To assess the sensitivity of the RDT, only the first MAT-positive serum from each confirmed case was used. For specificity evaluation, all 221 negative control sera have been tested using MAT and were all negative (titer<100) (see Table 1).
Possible false negative results due to high levels of anti-Leptospira IgM (prozone phenomenon) was controlled using two positive sera with 25,600 and 51,200 MAT titers: serial two-fold dilutions of the sera (1/400 to 1/6,400) were tested and test band intensity of the Vertical Flow RDT were compared.
The strips were stored at 4°C in sealed aluminium foils with a silica gel bag to avoid exposition to humid conditions, and the test was performed at laboratory room temperature (20°–25°C in New Caledonia and Paris).
The predictive shelf life of the RDT was assessed by testing serial dilutions of a MAT-positive serum (MAT titer = 800, pointing to serogroup Icterohaemorrhagiae) twice per week over a period of 3 weeks exposure of the strips at 60°C. During this period, the positive control serum was kept at 4°C to avoid repeated freeze-thaw cycles. This accelerated stability method is equivalent to two years of actual storage time at 25°C [15], [16].
Several experiments were performed in New Caledonia to evaluate the the Vertical Flow RDT. For reproducibility and repeatability, four sera (3 positives and one negative) were tested blindly by three different operators on three different days; one serum was tested blindly 14 times by integrating 14 aliquots of the same specimen within 4 independent experiments and using two different batches of RDTs by the same operator; 177 sera from both confirmed and probable cases (28 negative and 149 positive samples) were read independently by two technicians, including 157 sera (16 negative and 141 positive samples) by three technicians. We additionally simulated tropical field conditions by performing the tests in parallel at 37°C in an incubator (simulating tropical conditions) and at laboratory temperature (standard condition)
The earliness of IgM seroconversion using MAT or RDT was assessed on serial sera (day 2 to day 18 after the onset of symptoms) from 17 confirmed cases, based on the date of onset as declared by the patients.
The RDT was also tested on early sera from 99 patients who were tested positive by PCR but negative by MAT (titer = 0, 100 or 200).
One hundred and fifty MAT-positive sera from probable cases, including 124 sera from the IPNC collection and 26 from the French National Reference Centre, were tested using RDT.
To compare the newly developed RDT with currently available techniques, we compared its performance on identical sera from New Caledonia. To assess the sensitivity, 72 MAT-positive sera from confirmed cases were randomly selected from the 118 New Caledonian control sera. For the specificity, 72 negative controls were randomly-selected, corresponding to 10 anti-Chikungunya virus IgM positive sera, 30 healthy blood donors, 11 anti-dengue virus IgM positive sera from all 4 serotypes, 6 anti-hepatitis A virus total Ig positive sera, 7 rheumatoid factor positive sera, 7 syphilis (TPHA and VDRL) positive sera, one acute malaria serum. The results using our RDT were compared with those obtained using two Elisa assays (Leptospira IgM ELISA, Panbio, Inverness Medical, QLD Australia, and SERION ELISA classic Leptospira IgG/IgM, Institut Virion/Serion GmbH, Germany) and one lateral flow IgM immunochromatography assay (Leptocheck, Zephyr Biomedicals, India). The Serion ELISA test was used together with the Rheumatoid Factor Absorbent as recommended by the manufacturer. All tests were made within a 5 day period. For calculations, the “uncertain” results of ELISA were considered as positive.
The evaluation of our RDT for the serodiagnosis of leptospirosis was performed according to the WHO Tropical Diseases Research Diagnostics Evaluation Expert Panel for the evaluation of diagnostic tests for infectious diseases [17].
Data were captured into Excel 2007 (Microsoft Corporation, Redmond, United States of America).
We calculated sensitivity (Se), specificity (Sp), positive and negative predictive values (PPV and NPV, respectively) of the RDT, using the reference MAT serology as the gold standard. The 95% confidence intervals (95% CI) were calculated using the Wilson's method. The variations of the PPV and NPV according to the prevalence of the disease were also plotted.
We also calculated likelihood ratios (LR). The positive LR (LR+ = Se/[1 - Sp]) indicates how many times a positive result is more likely to be observed in specimens with the target disorder than in those without the target disorder. The negative LR (LR− = [1 - Se]/Sp) indicates how many times a negative result is more likely to be observed in specimens with the target disorder than in those without the target disorder. The more accurate the test is, the more LR differs from 1. LR+ above 10 and LR− below 0.1 are considered convincing diagnostic evidence [18]. The 95% CIs were calculated for LR+ and LR− [19].
The diagnostic odds ratio (DOR) measures the test performance by combining the strengths of sensitivity and specificity, with the advantage of representing a single indicator of accuracy. These characteristics make the DOR particularly useful for comparing tests whenever the balance between false negative and false positive rates is not of immediate importance [20]. The DOR is defined as the ratio of the odds of positive test results in specimens with the target disorder relative to the odds of positive test results in specimens without the target disorder. It was calculated as follows:The DOR does not depend on prevalence and its value ranges from 0 to infinity, with higher values indicating better discriminatory test performance. The 95% CIs for DOR values were also calculated [21].
The IPNC and the French NRC are reference diagnostic laboratories for leptospirosis. In New Caledonia, leptospirosis is a notifiable disease. The serum samples used in this study were selected from the 2008–2001 IPNC and 2009–2011 French NRC collections of sera issued from routine diagnostic activities and as part of public health surveillance. This biobank of sera was declared to the French Ministry of Research (DC-2010-1222, Collections number 1 and 2). This study was part of a protocol approved by the Institut Pasteur (protocol # RBM2008-16) and the French Ministry for Education & Research (protocol # AC-2007-44). All sera were tested as anonymous samples. Negative sera from mainland France were provided by Platform ICAReB (Investigation Clinique et Acces aux Ressources Biologiques). The STARDT checklist is provided as Table S2.
The sera used in this study were from patients or donors of both sex and all age classes, being selected among leptospirosis suspicions (positive sera), other pathologies or blood donors (negative sera). All along the study and whatever the batch used, we observed no invalid test: all RDT displayed an intensely marked control line and very little to no background coloration. Out of the 187 gold standard positive sera tested, 168 were RDT positive, including 15 RDT with a test line intensity graded as “weak” (8.9% of positives). The putative serogroups of the 19 RDT negative sera were: Icterohaemorrhagiae (n = 12), Pyrogenes (n = 3), Australis (n = 2), Panama (n = 1) and one could not be determined due to co-agglutination of multiple serogroups.
Out of the 221 MAT-negative sera tested, 207 were RDT negative. All 14 RDT positive sera were graded “weak” and originated from 9 healthy blood donors and five patients positive for anti-dengue virus IgM.
The sensitivity and specificity of the RDT were therefore, respectively, Se = 89.8% [95% CI, 84.7–93.4] and Sp = 93.7% [95% CI, 89.65–96.2]. The Likelihood Ratios (LR) were therefore LR+ = 14.18 [95% CI, 8.52–23.56] and LR− = 0.11 [95%; 0.01–0.17]; and the Diagnostic Odds Ratio DOR of 130.74 [95% CI, 63,65–268,52].
The results are summarized in Table 1, positive and negative predictive values of our RDT according to prevalence are presented in Figure 1.
The absence of false negative due to prozone phenomenon was demonstrated using dilutions of sera with highest MAT titers: serial two-fold dilutions actually yielded test lines of decreasing intensities..
To simulate tropical conditions, the RDT results of 10 MAT-positive sera run at 37°C were compared and proved identical to those run at 25°C. For accelerated shelf-life evaluation, serial two-fold dilutions (from 1/400 to 1/12,800) of one MAT positive serum (titer 800) were tested twice a week for three weeks with RDT stored at 60°C. At day 1, the RDT was positive at a 6,400 dilution, and remained the same until day 17. At day 21, the 3,200 dilution was the last giving a positive test line (a one dilution decrease of the serum).
One serum tested 14 times with strips from the two different batches gave 14 similar results, including the grade of the test line.
Inter-readers variability was assessed by two independent operators on 177 sera (28 negative and 149 positive) of which 157 sera were read by three independent operators. These readings provided an excellent inter-operator agreement (>99%) in all cases but one: one weakly positive RDT from a probable case was rated “weak” by two operators but negative by the third one.
Inter-operator variability was also assessed using 4 sera (RDT graded from negative to 3+) blindly and independently tested on three different days by three different operators. Two operators provided perfectly concordant grading results on all three tests, the third one graded “weak” a negative serum once out of the three tests.
Of 17 confirmed cases analysed (see Table 2), one patient (number 1) seroconverted for MAT at day 6 (pointing to Icterohaemorrhagiae) but remained negative for RDT; oppositely, 5 PCR-confirmed patients (numbers 2–6) were MAT negative whereas they were RDT-positive. For one of these patients (number 6), PCR and RDT tests were both positive at day 4 after onset of symptoms. Five patients (numbers 7–11) were positive for MAT and RDT on the same day (days 5–11 after the onset of symptoms); lastly, for 6 patients (numbers 12–17), the RDT was positive earlier than the MAT (day 3 to day 7). Out of these 6, four (numbers 12, 15, 16 and 17) had a positive blood PCR and RDT results on the same day (on days 5, 4, 7 and 3 respectively).
Similarly, in 16 out of 99 early sera from confirmed patients from New Caledonia, the RDT was positive whereas the MAT was still negative (6 out of 62 MAT negative) or displayed low titers (4 out of 21 with a MAT titer of 100; and 6 out of 16 with a MAT titer of 200).
Of 150 sera from probable cases of leptospirosis (unique sera with a MAT≥400), 109 gave a positive result using the RDT, corresponding to a concordance of 72.7% [65–79.1].
Out of these, 108 had a MAT>400, from which 81 (75% [66.1–82.2]) were RDT-positive, while 63 had a MAT titer >800, from which 53 (84.1% [73.2–91.1]) were RDT- positive.
The use of 72 gold standard positive (MAT≥400 from confirmed cases) and 72 negative (MAT<100) serum specimens selected randomly allowed a comparison of our RDT with three commercially available tests: two ELISA tests (from Serion (using the recommended RF absorbant) and from Panbio) and one IgM lateral flow immunochromatographic assay (Leptocheck). The results of these tests are detailed in Table 3. The IgM ELISA from Panbio had 100% specificity on these particular specimens together with the lowest sensitivity (75%). This 100% specificity does not allow the calculation of a Diagnostic Odds Ratio (DOR) that would however be very high. The ELISA test from Serion had both a good sensitivity (91.7%) and a good specificity (81.9%), therefore showing a good DOR of 49.9. Another rapid diagnostic test, namely Leptocheck (from Zephyr) had a very good sensitivity (91.2%) but a quite low specificity (52.8%), giving a DOR of 39.1. The Vertical Flow RDT we developed displayed a very good specificity (95.8%) and a good sensitivity (81.9%) and had therefore a very good DOR of 104.4. The corresponding curves of predictive values according to the prevalence of the two IgM rapid tests on these specimens are compared in the figure 2.
Serum samples from leptospirosis patients contain antibodies that become detectable approximately one week after the onset of symptoms [4]. MAT is a long used gold-standard method for the serological diagnosis of leptospirosis. This method relies on the detection of agglutinating antibodies (both IgM and IgG) against antigens that are live Leptospira strains corresponding to representative serogroups of epidemiological significance. However, many laboratories and hospitals do not have the facilities and expertise required to perform the MAT. Therefore, there is a growing use of qPCR for early diagnosis [22], providing the opportunity to rapidly confirm leptospirosis suspicions in acute phase sera. More simple and rapid serological diagnostic tests including ELISA-based assays detecting antibodies are also used. However, none of these techniques are prone to be implemented in health centers of endemic regions where the highest burden of leptospirosis occurs.
Pathogenic Leptospira strains are classified into 9 species and more than 200 serovars reflecting the structural heterogeneity in the carbohydrate component of the lipopolysaccharide. ELISA-based assays using crude whole-cell lysates of Leptospira strain (usually the saprophyte L. biflexa serovar Patoc strain Patoc 1) as antigens may not recognize the diversity of circulating strains and the sensitivity of these tests are generally poor [3], [4]. A major challenge is to discover antigens that are conserved across the major leptospiral strains. In this study, we tested a one-step vertical flow immunochromatography RDT coated with heat-killed L. fainei serovar Hurstbridge as an antigen. L. fainei belongs to the intermediate group of Leptospira [23] and, as such, may share common antigenic features with saprophytes and pathogens which constitute the two other phylogenetic groups in the genus Leptospira. In addition, a previous study has suggested that L. fainei serovar Hurstbridge may cross-react with different pathogenic serovars [24].
Rapid diagnostic tests should ideally be accurate, simple to use, relatively inexpensive, easy to interpret, stable under extreme conditions, with little or no processing, and give the results within less than 2 hours.
A proper evaluation of a diagnostic test has to face two major challenges: first, the samples used for validation must have a very well-defined status with regard to the diagnostic target of interest; second, the results of the test under evaluation must be compared with the results of the same samples characterized using a validated reference test defined as the “gold standard”. Our study evaluated the sensitivity and specificity of a new Vertical Flow RDT for the serological diagnosis of leptospirosis in endemic (New Caledonia and French West Indies) and non-endemic (mainland France) countries. We only used sera from confirmed leptospirosis cases [13] for this evaluation. Therefore, the positive samples for the evaluation of sensitivity were both gold standard positive (a MAT titer of at least 400) and from confirmed leptospirosis cases (either a positive PCR or a seroconversion from nil to ≥400 or a ≥4-fold rise in MAT titer in paired sera). This high MAT threshold was chosen because most of the specimens (149 out of 187) originated from endemic regions, where a similar MAT threshold is used for diagnosis and surveillance [4], [9]. Additionally, negative sera for specificity evaluation were tested blindly using the reference MAT and were only considered as true negatives if the MAT titer was below 100. These latter originated from both healthy volunteers and a selection of patients with pathologic conditions of relevance in endemic countries. Using this clearly defined case definition, sensitivity and specificity were assessed using a high dilution of sera (1/400) reflecting the high MAT threshold titer. The sensitivity and specificity of the new Vertical Flow RDT in these conditions were 89.8% and 93.7% respectively. These results compare and are slightly better than the ones reported by Smits et al. who reported a 85.8% sensitivity and a 93.6% specificity with another Vertical Flow RDT [25].
The performance of ELISA tests vary widely in terms of sensitivity and specificity. For example, a commercial IgM ELISA (Panbio) gave a sensitivity and specificity of 76% and 82% in northeast Thailand [26], 35% and 98% in Hawai [27], and 61% and 66% in Laos [28], respectively. Reported variations in diagnostic assay performance may reflect population-related differences such as past exposure to leptospirosis or environmental leptospires. This can also be attributable to differences in the choice of the case definition; MAT is usually used as the reference test [29].
To increase the statistical power of our evaluation, we included as many serum samples as possible, only including the earliest positive serum when serial samples were available. Because the patient population recruited through our laboratories represents all (New Caledonia) or around 80% (France and West Indies) of leptospirosis suspicions, our collection can be regarded as representative of the total patient population in these regions. These included sera from New Caledonia older than 3 years, stored frozen at −20°C. It is well recognized that the long term storage of serum specimens at −20°C and their freeze/thawings may result in a drop of IgM titers. Actually, the sensitivity was higher in sera stored for less than two years than in sera stored for more than two years (90.6% versus 81.5%). This may have resulted in a slight under-estimation of the sensitivity of this Vertical Flow RDT.
The RDT we developed reacts with IgM to at least serogroups Australis, Autumnalis, Ballum, Bataviae, Canicola, Cynopteri, Grippotyphosa, Hebdomadis Icterohaemorrhagiae, Panama, Pomona, Pyrogenes, Sejroe and Tarassovi, indicating that the assay reacts broadly with antibodies mounted against Leptospira strains circulating worldwide.
Probable leptospirosis cases were defined as cases with a leptospirosis-compatible clinical presentation but a unique serology with a MAT titer ≥400. We tested our Vertical Flow RDT using these unique sera from such cases. Interestingly, the proportion of Vertical Flow RDT -positive sera was significantly lower than the sensitivity of the test as determined using confirmed cases (69.4% versus 89.8%, χ2 = 9.08, p<0.01), suggesting a poorer positive predictive value of the RDT for patients classified as probable cases. Two main reasons might contribute in explaining this difference. First, our Vertical Flow RDT only detects IgMs whereas the MAT is known to detect both IgMs and IgGs. Because IgM titers are known to decline faster than IgG, some positive MAT results may reveal IgGs remaining from previous exposure to leptospires. MAT could therefore be less specific than IgM-specific assays to detect acute and recent leptospirosis. It is also well known that direct visual methods such as MAT (agglutination of bacteria using microscopy) are less sensitive than indirect amplified techniques such as ELISA or colloidal gold particles immunochromatography assays. The possibility of false positive MAT results was already observed in other contexts [30]. Actually, though MAT is the recognized reference technique for the serological diagnosis of leptospirosis, it also suffers some drawbacks and weaknesses. Some concerns about both its sensitivity and its specificity have been raised and discussed [30]–[36]. More recently, a mathematical modeling study again demonstrated the limitations of MAT as a gold standard [29]. Since no diagnostic assay is adequately sensitive and specific enough to diagnose all acute cases of leptospirosis, results should be confirmed by another method. Regarding sensitivity, the most widely recognized weakness of MAT is its low sensitivity in early acute phase sera. The ability to detect anti-Leptospira antibodies earlier in the course of the disease with specific IgM detection tests than with MAT has already been largely recognized [30], [33], [34], [37]–[41]. We also observed an earlier positivity of our IgM Vertical Flow RDT when compared to MAT in serial sera. In our study, 11 out of 17 leptospirosis seroconversions could be diagnosed earlier with the Vertical Flow RDT than with the MAT (Table 2). Similarly, in 16 out of 99 early sera from confirmed cases, the Vertical Flow RDT was positive before the seroascension (10/37) or seroconversion (6/62) could be evidenced with the MAT. However, for strictness reasons, our strategy was to only use gold standard-positive specimens (MAT titers ≥400) for sensitivity assessment. Therefore, the sensitivity evaluated here might not reflect the conditions of rapid tests use in routine medical conditions, where patients may be seen and tested before seroconversion.
Previous studies have generally found that ELISA-based assays detect anti-Leptospira antibodies earlier in the course of the disease than with MAT [4], [30], [33], [37], [38]. Anti-leptospiral IgM cannot be detectable before 4–5 days after onset of symptoms, before the appearance of IgG and agglutinating antibodies [42].
Because an early diagnosis is of prime importance in the clinical management of leptospirosis [43], the possibility to ascertain the disease earlier in the course of the infection should be regarded as a real asset.
When considering the need of RDT for bedside diagnosis, the comparison of our Vertical Flow RDT with a RDT that is commercially available shows that our test has a lower sensitivity (81.9% versus 97.2%) but a much higher specificity (95.8% versus 52.8%) and therefore a better Diagnostic Odds Ratio (104.4 versus 39.1). This better performance is also shown by the comparison of the curves of their predictive values according to prevalence (Figure 2). There may be various reasons for these differences, including a different antigen used for IgM detection. However, the most probable cause is a much lower dilution of serum specimens in the Leptocheck lateral flow IgM assay (ca. 1/20) when compared with the dilution used for our Vertical Flow RDT (1/400). Because the MAT positive threshold titers may vary depending on the region [44], endemic countries usually using a higher threshold, it might be worth also considering using a different serum dilution for RDT according to the local epidemiology of leptospirosis. Because leptospirosis is endemic in New Caledonia and French West Indies, we decided to consider both a MAT titer of 400 as a gold standard and a 1/400 dilution of the serum for the Vertical Flow RDT evaluation. It is highly probable that the use of a lower dilution (1/200 or 1/100) would result in both an increased sensitivity and a decreased specificity.
Our results demonstrate that this new rapid diagnostic test could prove useful in endemic contexts, especially in low and middle-income countries. Actually, most of the leptospirosis burden occurs in the back-country with delayed access to the reference laboratories. In epidemics situations, especially during post-disaster periods like in the Philippines in 2009 [45], reference diagnostic tests are seldom if ever available. Therefore, a RDT with good diagnostic performances would also be particularly useful [1]. For easier use, further development of the technique could allow its use with capillary blood.
However, because an early initiation of antibiotherapy is a major contributor to a rapid recovery, the recommendation of treating the patient on the sole basis of a clinical and epidemiological suspicion should be maintained. The use of this Vertical Flow RDT as an initial screen for leptospiral infections would still allow facilitating the difficult differential diagnosis of leptospirosis [46]. Lastly, because the MAT provides important epidemiological information at the population level, it should still be recommended that sera be sent to the reference laboratory for subsequent confirmation by MAT, as suggested by other authors [41].
|
10.1371/journal.pntd.0001812 | Field Evaluation of a Coproantigen Detection Test for Fascioliasis Diagnosis and Surveillance in Human Hyperendemic Areas of Andean Countries | Emergence of human fascioliasis prompted a worldwide control initiative including a pilot study in a few countries. Two hyperendemic areas were chosen: Huacullani, Northern Altiplano, Bolivia, representing the Altiplanic transmission pattern with high prevalences and intensities; Cajamarca valley, Peru, representing the valley pattern with high prevalences but low intensities. Coprological sample collection, transport and study procedures were analyzed to improve individual diagnosis and subsequent treatments and surveillance activities. Therefore, a coproantigen-detection technique (MM3-COPRO ELISA) was evaluated, using classical techniques for egg detection for comparison.
A total of 436 and 362 stool samples from schoolchildren of Huacullani and Cajamarca, respectively, were used. Positive samples from Huacullani were 24.77% using the MM3-COPRO technique, and 21.56% using Kato-Katz. Positive samples from Cajamarca were 11.05% using MM3-COPRO, and 5.24% using rapid sedimentation and Kato-Katz. In Huacullani, using Kato-Katz as gold standard, sensitivity and specificity were 94.68% and 98.48%, respectively, and using Kato-Katz and COPRO-ELISA test together, they were 95.68% and 100%. In Cajamarca, using rapid sedimentation and Kato-Katz together, results were 94.73% and 93.58%, and using rapid sedimentation, Kato-Katz and copro-ELISA together, they were 97.56% and 100%, respectively. There was no correlation between coproantigen detection by optical density (OD) and infection intensity by eggs per gram of feces (epg) in Cajamarca low burden cases (<400 epg), nor in Huacullani high burden cases (≥400 epg), although there was in Huacullani low burden cases (<400 epg). Six cases of egg emission appeared negative by MM3-COPRO, including one with a high egg count (1248 epg).
The coproantigen-detection test allows for high sensitivity and specificity, fast large mass screening capacity, detection in the chronic phase, early detection of treatment failure or reinfection in post-treated subjects, and usefulness in surveillance programs. However, this technique falls short when evaluating the fluke burden on its own.
| A coproantigen-detection technique (MM3-COPRO ELISA) was evaluated in 436 and 362 schoolchildren of Huacullani, Bolivia, and Cajamarca valley, Peru, respectively. Classical techniques for egg detection were used for comparison. In Huacullani, using Kato-Katz as gold standard, sensitivity and specificity were 94.68% and 98.48%, respectively, and using Kato-Katz and COPRO-ELISA test together, they were 95.68% and 100%, respectively. In Cajamarca, using rapid sedimentation and Kato-Katz together, these results were 94.73% and 93.58%, respectively, and using rapid sedimentation, Kato-Katz and copro-ELISA together, results were 97.56% and 100%, respectively. There was no correlation between coproantigen detection by optical density (OD) and infection intensity by eggs per gram of feces (epg) in Cajamarca low burden cases (<400 epg), nor in Huacullani high burden cases (≥400 epg), although there was in Huacullani low burden cases (<400 epg). Six cases of egg emission appeared negative by MM3-COPRO, including one with a high egg count (1248 epg). The coproantigen-detection test allows for high sensitivity and specificity, fast large mass screening capacity, detection in the chronic phase, early detection of treatment failure or reinfection in post-treated subjects, and usefulness for surveillance programs. However, this technique falls short when evaluating the fluke burden on its own.
| Fascioliasis is an important human and animal disease caused by the trematode species Fasciola hepatica and F. gigantica. At present, fascioliasis is emerging or re-emerging in numerous regions of Latin-America, Africa, Europe and Asia, both in humans and animals, a phenomenon which has partly been related to climate change [1]. Major human health problems are encountered in Andean countries (Bolivia, Peru, Chile and Ecuador), the Caribbean (Cuba), northern Africa (Egypt), western Europe (Portugal, France and Spain) and the Caspian area (Iran and neighbouring countries) [1]. Emergence, long-term pathogenicity [2]–[4] and immunological interactions [5], [6] prompted the WHO to include this disease among the so-called neglected tropical diseases (NTDs), which are chronic, debilitating, poverty-promoting and among the most common causes of illness in developing countries. Their control and elimination is now recognized as a priority to achieve the United Nations Millennium Development Goals and targets for sustainable poverty reduction [7], [8].
Among Andean countries, the highest human fascioliasis prevalences and intensities are encountered in the Northern Altiplano of both Bolivia [9], [10] and Peru [11], as well as in the Cajamarca valley (Peru) [12], where F. hepatica is the only fasciolid species present [13] and children and females are the subjects most affected [1]. Within the human fascioliasis high altitude transmission pattern related to F. hepatica transmitted by lymnaeid vectors of the Galba/Fossaria group in Andean countries, two different subpatterns have been distinguished according to physiographic and seasonal characteristics [1], [13]: a) the Altiplanic pattern, with endemicity distributed throughout an area of homogeneous altitude and transmission throughout the whole year caused by high evapotranspiration rates leading lymnaeid vectors to concentrate in permanent water bodies, e.g. the Northern Bolivian Altiplano [14]; b) the valley pattern, with endemicity distributed throughout an area of heterogeneous altitude and seasonal transmission related to climate, e.g. the Cajamarca valley in Peru [12], [15], [16].
In recent years, the availability of a very effective drug against fascioliasis, namely triclabendazole [17], prompted the WHO to launch a worldwide initiative against human fascioliasis [18], [19]. This initiative includes interventions in human fascioliasis endemic areas presenting different epidemiological situations and transmission patterns [1]. Bolivia and Peru are two of the countries selected for priority intervention due to the very large public health problem posed by this disease. Different pilot schemes were designed to assess the best control strategies according to the different epidemiological situations and transmission patterns. The Northern Bolivian Altiplano was chosen as an example of the Altiplanic pattern, while the Cajamarca valley was chosen as an example of the valley pattern.
An alternative to coprological egg detection is the use of immunodiagnostic tests based on the detection of anti-Fasciola antibodies and/or coproantigens released by the parasite. In the last decades, several ELISA methods based on the use of polyclonal and monoclonal antibodies have been reported to be useful for detection of F. hepatica and F. gigantica in the feces of sheep and cattle [20]–[23] and also rat feces [24]. Nevertheless, surveys on human fascioliasis have usually been made through various coprological techniques verifying the presence of eggs in stools [25] and antibody detection tests to confirm the diagnosis of human fascioliasis [26]. Among these techniques, classical coprological egg detection methods are the most frequently used [27]. However, so far, the use of coproantigen detection was applied to diagnose F. hepatica infection in patients in Cuba only [28], [29].
Enzyme-linked immunosorbent assay (ELISA) methods developed for determination of Fasciola coproantigens in stool samples from animals and humans provide an alternative to coprological examination [30], [31]. One of these methods is the MM3 capture ELISA (MM3-COPRO) test for fascioliasis diagnosis detection of fecal excretory-secretory antigens (ESAs) using a monoclonal antibody (mAb), whose usefulness for detection of F. hepatica and F. gigantica coproantigens in experimental and natural Fasciola infections of sheep and cattle has already been demonstrated [22], [32]. This test proved to be highly sensitive (confirmed by necropsy) and specific (no cross reaction was observed with antigens from other helminths), and allowed for the detection of Fasciola infections 1–5 weeks before patency in cattle. Furthermore, other researchers recently tested a commercial version of the test, and its appropriateness for the detection of F. hepatica infections in cattle was confirmed under field conditions [33]. The suitability of the MM3-COPRO method for detection of Fasciola coproantigens in both fresh and preserved stools from hospital patients has been demonstrated [34], but its applicability for detection of F. hepatica infections in humans under field conditions has not been proved.
An efficient coproantigen test for human fascioliasis diagnosis represents a valuable tool to facilitate population screening and post-treatment surveillance in control campaigns, above all in communities where people are reluctant to furnish blood samples due to ethnic/religious aspects. The aim of the present study is to evaluate the coproantigen technique MM3-COPRO ELISA under field conditions for human fascioliasis diagnosis in human hyperendemic areas of Andean countries, using classical coprological techniques for egg detection for comparison purposes (rapid sedimentation and Kato-Katz). Thus, two endemic areas were chosen: Huacullani (Bolivia) representing the Altiplanic pattern with high prevalences and intensities, and the rural areas of Cajamarca (Peru), representing the valley pattern with high prevalences but with low intensities. Results of the pilot intervention implemented in Huacullani to assess the feasibility of a strategy of large-scale administration of triclabendazole, with a focus on safety and efficacy, are included in another article [35].
In Bolivia, the study was approved by the Comisión de Etica de la Investigación of the Comité Nacional de Bioética. In Peru, it was approved by the Comité Institucional de Etica of the Universidad Peruana Cayetano Heredia and the Comité de Ética of the Instituto Nacional de Salud.
All subjects involved provided written informed consent. Samples from children were obtained after consent from the children's parents, following the principles expressed in the Declaration of Helsinki. Consent was also obtained from the local authorities of the communities and heads and teachers of the school.
In Huacullani, activities were performed in collaboration with the Servicio Departamental de Salud La Paz and the Unidad de Epidemiología of the Bolivian Ministry of Health and Sports (MSyD). In Cajamarca, the study was done in cooperation with the Dirección Regional de Salud of Cajamarca, and the Estrategia Nacional de Zoonosis, Dirección General de Salud de las Personas, Ministerio de Salud (MINSA), Lima.
Coprological studies were carried out in the locality of Huacullani, which belongs to the municipality of Tiwanaku, third section of the province of Ingavi of the Departamento de La Paz, Bolivia. This locality is situated 85 km from the capital La Paz, at the western end of the so-called Tambillo-Aygachi corridor of the Northern Bolivian Altiplano. Huacullani has 2525 inhabitants, according to the last 2005 census of the Bolivian Instituto Nacional de Estadística. Stool collection was performed in the school of the locality and samples were obtained from a total of 436 children. Previous surveys in that locality showed very high prevalence rates of 38.2% in the year 1992, 31.2% in 1993, and 34.8% in 1996 in children, and 18.4% in 1996 and 11.8% in 1997 in total community surveys (children plus adults) [10].
Stool samples were also obtained in the Departmento de Cajamarca, Peru, which covers an area of around 35,400 km2 in the northern Andean part of Peru and is inhabited by 1,416,000 people. This Department comprises 13 provinces and the province of Cajamarca in turn includes 12 districts [12]. A total of 362 fecal samples were obtained from children of the schools of Escuela de Varones del Distrito (Jesus district), Santa Rosa de Chaquil (La Encañada district), and Andres Avelino Caceres (Baños del Inca district). Previous surveys showed very high prevalences in that endemic area, with a mean of 24.4% and the maximum prevalence of 47.7% in Santa Rosa de Chaquil, the hitherto highest local prevalence detected in Peru [12].
Classical coprological techniques for egg detection were used for qualitative (rapid sedimentation and Kato-Katz) and quantitative (Kato-Katz) diagnosis. The combined use of highly specific techniques has been reported as a means of compensating the low sensitivity of the Kato-Katz technique [36]. Thus, identification of true positive and true negative cases was carried out by using two criteria: i) finding of F. hepatica eggs in feces; ii) egg finding plus COPRO ELISA test results. Applying the Kato-Katz technique, eggs were detected in fresh stools after analysis of three Kato-Katz slides (Helm-Test, AK test, AK Industria e Comércio Ltda, Belo Horizonte, Brasil) per sample, depending on the concentration of Fasciola eggs following WHO recommendations, using a template delivering about 41.7 mg of feces [37]. The average egg output was calculated as eggs per grams of feces (epg). Parasitological analysis was done microscopically by a trained parasitologist. Intensity of infection, measured as eggs per gram (epg), was used as an indicator of F. hepatica burden in infected subjects. Kato-Katz was used in both study areas and rapid sedimentation was an additional test done in Cajamarca. In the case of the Huacullani samples, a single Kato-Katz slide was used for each sample. In the case of the Cajamarca samples, the rapid sedimentation procedure was applied and those fecal samples positive for the MM3-COPRO ELISA were also quantitatively analyzed by three Kato-Katz slides.
Children were not included in the study if they presented any chronic or acute hepatic disease, pregnancy, breast-feeding, any acute infection within a week of enrolment, or receiving treatment for any other disease or condition. In Huacullani, at the time of the baseline survey (April 2008), the school population consisted of 459 children aged 5 to 14 years, who were all considered eligible for an interventional treatment study. A total of 447 children returned the plastic container. From these, 437 fecal samples from an equivalent number of children were examined (four children returned an empty plastic container, and six other children provided insufficient stool quantities to apply both Kato-Katz slide and COPRO ELISA). Thus, fecal samples were obtained from 437 children (220 males and 217 females) of 5–16 years of age (mean ± SD = 8.8±2.3). A clean, plastic, wide-mouthed, numbered container with a snap-on lid was given to every participant. All subjects were then asked to try to fill the container with their own feces and to return it immediately. One stool sample per subject was collected and personal data (name, sex, and age) were noted on delivery of the container. Samples were transported to the parasitological laboratory of the Faculty of Pharmacy, Universidad Mayor de San Andrés (UMSA), La Paz, within 1–3 h after collection. One aliquot was used to carry out the MM3-COPRO ELISA and another was preserved at 4°C to make the Kato-Katz slides. All Kato-Katz slides were made at the Laboratory of Parasitology of the Faculty of Medicine, UMSA, and were initially examined within 1 h of preparation to avoid overclarification of some helminth eggs.
In Cajamarca, at the time of the baseline survey (December 2007), the target population was 616 school children (age range 6–15 years old), with a coverage of 4.25% of the school children population and 0.86% of the overall population from the three aforementioned districts. Thus, in the present study, fecal samples were obtained from 362 children (264 males and 98 females), 7–15 years of age (mean ± S.D. = 9.9±2.2), by similar procedures. Samples were transported to Cajamarca city within 1–3 h after collection and stored at 4°C until being sent to the Laboratory of Parasitology at the Instituto de Medicina Tropical Alexander von Humboldt, Lima, where coproparasitological analyses were carried out. Both ELISA and Kato-Katz slides were applied to two aliquots of the material preserved at 4°C. A third aliquot was preserved in 10% formalin solution (1∶3) for subsequent egg detection by means of the rapid sedimentation technique [38].
To assure quality standards and possible handling differences, procedures in the two laboratories were implemented by the same personnel of the Valencia team in addition to the respective local personnel of each laboratory. In Cajamarca, stool samples were distributed into two groups according to the 400-epg threshold used for identifying high intensity infections [18]: a high burden group (≥400 epg) and a low burden group (<400 epg). However, in Huacullani, as precautionary measure, a lower threshold (300 epg) was requested to be applied by Bolivian health responsibles to distinguish between samples whose respective infected children were in need to be hospitalized for prevention follow up of potential post-treatment colics, and samples whose respective infected children were not hospitalized and were treated on an out-patient basis [35].
Statistical analyses were done using PASW 17 software. For the evaluation of categorical variables, the chi-square test or Fisher's exact test was used. Bivariant correlations (Pearson's correlation) were calculated to assess the relationship between optical density (OD) and epg of F. hepatica. A P value below 0.05 was considered significant.
Theoretical positive predictive values (PPV) and negative predictive values (NPV) were calculated from sensitivity and specificity values obtained using only classical coprological tests for the identification of F. hepatica eggs in feces as “gold standard”. The following formulas were used for their calculation:
The MM3-COPRO ELISA kits were prepared and tests performed as previously described [22], [32], [34]. Kits were provided by Dr. F.M. Ubeira (University of Santiago de Compostela, Spain). Briefly, polystyrene microtiter 1×8 F strip plates (Greiner Bio-One GmbH, Frickenhausen, Germany) were coated overnight with 100 µL/well of a solution containing 10 µg/mL of rabbit anti-Fasciola polyclonal IgG antibody in phosphate buffered saline (PBS) (wells from odd-numbered rows), or with 100 µL/well of a solution containing 10 µg/mL of IgG polyclonal antibodies from non-immunized rabbits (wells from even-numbered rows). Uncoated sites were blocked with 1.5% of sodium caseinate in PBS for 1 h at RT, and each fecal supernatant (100 µL) was then added in quadruplicate (2 odd-numbered wells plus 2 even-numbered wells), and incubated overnight at 4°C. After washing 6 times with PBS containing 0.2% Tween-20 (PBS-T), 100 µL of a solution containing 0.3 µg of biotinylated MM3 antibodies in PBS-T plus 1% bovine serum albumin (PBS-T-BSA) was added to each well and incubated for 1 hr at 37°C. After washing as above, bound MM3 antibody was detected by incubation, first with peroxidase-conjugated neutravidin (Pierce, Rockford, Illinois; dilution 1∶2000 in PBS-T-BSA) for 1 hr at 37°C, and then with the substrate (buffered H2O2 and o–phenylenediamine [OPD], Sigma-Aldrich, Madrid, Spain). After incubation for 20 min at RT, the reaction was stopped by addition of 3N H2SO4. Finally, OD was measured at 492 nm. The OD value for each sample was calculated as OD1–OD2, where OD1 is the mean for the 2 even-numbered wells (coated with anti-Fasciola polyclonal antibodies), and OD2 is the mean for the 2 odd-numbered wells (coated with irrelevant polyclonal antibodies). The OD value for each sample was calculated by subtracting the OD of the blank well from the OD of the test well using the cut-off point 0.097 [34].
Diagnostic parameters of the MM3-COPRO ELISA were estimated by (i) only choosing coprology as the “gold standard” assay to detect F. hepatica infection in humans, and also by (ii) considering results of both coprology and COPRO ELISA together. Positive cases of the MM3-COPRO ELISA and egg detection techniques of F. hepatica infection and performance characteristics of the MM3-COPRO ELISA according to study site are summarized in Table 1.
Huacullani positive cases were globally 24.77% using MM3-COPRO ELISA and 21.56% applying an egg detection technique (Kato-Katz). No significant differences were encountered between either % (P = 0.093).
In this Bolivian locality, using Kato-Katz as gold standard, sensitivity and specificity were 94.68% and 98.48%, respectively, and using Kato-Katz and COPRO-ELISA test together as gold standard, sensitivity and specificity were 95.68% and 100%, respectively.
Of 436 samples assayed, 94 showed the presence of eggs through the Kato-Katz technique (21.56%). MM3-COPRO ELISA was positive in 108 samples (24.77%), which included samples with Fasciola eggs (89) and without Fasciola eggs (19), i.e. 82.40% of the children who were positive for the MM3-COPRO ELISA were also positive through the Kato-Katz procedure. It should be emphasized that there were five children shedding eggs with emissions of 48, 72, 96, 120 and 1248 epg, whose MM3-COPRO ELISA results were negative (1.14%). The stool sample showing 1248 epg was repeatedly re-analyzed and a negative result was always obtained with the MM3-COPRO ELISA test.
The geometric mean egg content in F. hepatica positive samples was 142.17 epg, and the arithmetic mean was 334.98 (with SD of ±92.56), with a range of 24 to 8088 epg (Table 1). In these samples from Huacullani, epg data were distributed into two groups: a high burden group (≥400 epg) and a low burden group (<400 epg) of samples.
The OD values obtained for individual F. hepatica positive and negative fecal samples from Huacullani are shown in Figure 1. Positive samples with F. hepatica eggs showed OD values above the cut-off value except in five cases (determined by the Kato-Katz technique).
In children who were positive in egg emission, the bivariant correlation between OD and epg data from low and high burden groups was carried out separately. A significant positive correlation was detected only between OD and low burden (r2 = 0.20) (Figure 2), but no significant positive correlation was detected when considering OD and high burden (r2 = 0.01) (Figure 3).
Theoretical PPVs and NPVs vs fascioliasis prevalence are represented in Figure 4A, showing the expected PPVs and NPVs depending on whether the test was used in low, medium or high prevalence scenarios in this Altiplanic highly endemic locality.
Cajamarca positive cases were globally 11.05% using MM3-COPRO ELISA and 5.60% employing egg detection techniques (rapid sedimentation and Kato-Katz). Significant differences were detected between both % (P = 0.007).
Differences between the two local patterns were detected, i.e. significant differences were found when comparing MM3-COPRO ELISA positive cases % from Huacullani and Cajamarca (P = 0.0025), and also when comparing egg detection positive cases % from Huacullani and Cajamarca (P = 0.001).
In Cajamarca, using rapid sedimentation and Kato-Katz together as gold standard, sensitivity and specificity were 94.73% and 93.58%, respectively, and using rapid sedimentation, Kato-Katz and copro-ELISA together as gold standard, results were 97.56% and 100%, respectively.
In this Peruvian locality, of 362 samples assayed, 19 showed the presence of eggs through the rapid sedimentation and Kato-Katz techniques (5.24%), whereas MM3-COPRO ELISA was positive in 40 samples, which included the samples with Fasciola eggs (18) and without Fasciola eggs (22), i.e. 45.0% of the children who were positive by MM3-COPRO ELISA were also positive through coprological egg detection procedures. Interestingly, one child shed eggs (by the rapid sedimentation technique) but was negative by MM3-COPRO ELISA. The remaining 321 MM3-COPRO ELISA negative samples, however, included 237 negative samples, and 84 positive samples for parasitic infections other than Fasciola. They involved one or more parasitic protozoans (Blastocystis hominis, Chilomastix mensnilii, Giardia intestinalis, Entamoeba histolytica/E. dispar/E. moshkovskii, E. coli, Endolimax nana, Iodamoeba buetschlii) and helminth species (Strongyloides stercoralis, Ascaris lumbricoides, Trichuris trichiura, Enterobius vermicularis and Hymenolepis nana).
The geometric mean egg content in F. hepatica positive samples from Cajamarca was 89.80 epg, and the arithmetic mean was 116.47 (with SD of ±84.80), with a range of 16 to 376 epg (Table 1), i.e. samples were all considered as belonging to the low burden group as their epg counts were <400.
The OD values obtained for individual F. hepatica positive and negative fecal samples from Cajamarca are shown in Figure 5. Positive samples with F. hepatica eggs showed OD values above the cut-off value except in one case (determined by the rapid sedimentation technique).
In children who were positive in egg emission, the bivariant correlation between OD and epg data (low burden) was carried out. No significant positive correlation between OD and low burden (r2 = 0.05) was detected (Figure 6).
Theoretical PPVs and NPVs vs fascioliasis prevalence are represented in Figure 4B, showing the expected PPVs and NPVs depending on whether the test was used in low, medium or high prevalence scenarios in this Peruvian highly endemic locality.
Sensitivity is defined as the proportion of people with the disease who have a positive test for the disease. A sensitive test will rarely miss people with the disease. Specificity is the proportion of people without the disease who have a negative test. A specific test will rarely misclassify people as having the disease when they do not [39]. Knowing true positive and true negative cases is essential when calculating sensitivity and specificity, respectively. The identification of true positive and true negative cases was carried out using classical coprological tests for the identification of F. hepatica eggs in feces. Nevertheless, in the case of human fascioliasis, the application of the rapid sedimentation or Kato-Katz techniques may result in false negative cases. The ethiological diagnosis based on egg detection in stools is complicated because parasite eggs are not found during the prepatent period [27], [40], when juvenile worms migrate through the intestinal wall to the peritoneal cavity (at one week), penetrate the liver parenchyma (at five to seven weeks), and pass into the biliary tract where they ultimately reach maturity (at two months or more). Previous studies have even estimated a period of at least three to four months to be necessary for F. hepatica flukes to attain sexual maturity in humans [27], [41]. Once the worms have matured, diagnosis still remains difficult because commonly employed microscopic techniques for quantitative diagnosis of Fasciola eggs are very specific but rather insensitive. In addition, in some cases diagnosis is also difficult during the biliary stage, due to the intermittent excretion of parasite eggs. Fecal egg counts are known to follow inter- and intraindividual variations in fascioliasis [42], [43]. In our case, we used the Kato-Katz technique as a “gold standard” because it is considered the best available for quantitative analysis, although taking into account that it is admittedly rather imperfect. Therefore, results from the rapid sedimentation were also considered to improve the first gold standard in Cajamarca, and the combination of results from both coprological methods and COPRO ELISA were used for both study areas.
When liver-flukes are located in the bile ducts, excretory-secretory (ES) products are released, being eliminated via feces. The detection of these products by means of a sandwich-ELISA reflects the installation of flukes in the bile ducts and the presence of the biliary stage of the disease [30]. No statistically-significant differences were detected between prevalence results obtained using egg detection techniques and the MM3-COPRO ELISA in Huacullani, where egg intensities are higher according to the typical feature of the Altiplanic pattern. On the contrary, such differences were detected in Cajamarca, probably as a consequence of the low egg intensities characteristic of the valley pattern, i.e. in Cajamarca low burdens are common and therefore the higher probability of infected subjects intermittently shedding very few eggs is higher, with the consequence that such cases go unnoticed.
Five and one cases of egg emission were negative using the MM3-COPRO ELISA in Huacullani and Cajamarca, respectively, including one case of very high egg count (1248 epg) in the Bolivian locality. Such cases pose a question mark. This result raises the question as to whether these false negative cases may be interpreted as being inherent to the kit design, or due to external factors not attributable to the ELISA kit. However, considering that there is broad experience in detecting Fasciola coproantigens in ovine and bovine samples, and that animals with egg emission were not found to be negative with the MM3 ELISA, it is unlikely that these false negative results were due to kit construction. Furthermore, a recent study has shown that the monoclonal antibody used in the MM3 assay recognizes L1 and L2 cathepsins [44], as does the ES78 in the FasciDig test [20], and there were no observations of false negative results with this test either. Three possible explanations include (i) an intermittent release of the ES products from the liver to the intestine through the bile ducts, (ii) spurious infections, and (iii) the existence of food remains in the intestine masking or interfering with the detection of the fluke ES antigens. The first does not seem to be the case, as previous studies using this and other kits do not indicate the emission of cathepsins L1 and L2 in either human or animal species to be intermittent (unlike with eggs). A spurious infection was excluded after meticuluous study of the aspect of eggs in each one of these children. A potential negative influence of high temperatures during the transport or/and an inadequate handling of the samples at a given moment throughout the whole procedure cannot be ruled out, although the relatively short storage period does not suggest a considerable denaturation of the L-cathepsins to have occurred.
In both Huacullani (representing the fascioliasis Altiplanic pattern) and Cajamarca (representing the fascioliasis valley pattern), cases of coproantigens present in the feces of humans without F. hepatica eggs in stools were detected. Previous time-course studies in animals on the detection of F. hepatica coproantigens by ELISA indicated that coproantigens were detectable prior to patency [45]. Furthermore, a marked increase in the levels of these coproantigens at the beginning of fecal egg output was observed [32]. Considering the positivity/negativity of MM3-COPRO ELISA and the presence/absence of eggs in feces, two situations were established in the current study: the Altiplanic pattern with a correlation between positivity of MM3-COPRO ELISA and the presence of eggs, and the valley pattern with a larger number of positive cases applying MM3-COPRO ELISA but without presence of eggs in feces.
The Altiplanic pattern, characterized by higher prevalences and intensities, showed no statistical differences between the percentage of children who were positive in coproantigens and eggs in feces. Thus, it may be concluded that the majority of children with liver-flukes in the biliary ducts shed eggs. Nevertheless, in the valley pattern, characterized by high prevalences but low intensities, differences were detected between the percentage of children who were positive to coproantigens and the reduced number of these children that shed eggs in feces. This suggests that children did not shed eggs or only a much reduced, undetectable number even despite presenting parasites in their biliary canals.
In Europe, for instance, the diagnosis of human fascioliasis is frequently established using serological tests, because the detection of F. hepatica eggs in stools is not always possible. Thus, in an epidemiological survey from 1970 to 1999 to record cases of human fascioliasis detected in the Limousin region (central France), egg detection in stools was positive in only 27.6% of a total of 711 persons with fascioliasis [46]. Future studies are needed in Cajamarca (and other endemic areas in valleys of the Andean countries) to verify whether in these cases the non-detection of eggs implies that the parasite (i) has not reached the biliary ducts or is located in the bile-ducts but oviposition has not yet started (suggesting a more or less recent infection) or (ii) oviposition is taking already place but with only very low egg numbers or with intermittent shedding (indicating that subjects present only one or a very few flukes in the chronic stage).
Negative results by the MM3-COPRO ELISA after treatment, which occurs approximately one to three weeks in animals, is usually accepted when determining the efficacy of anthelminthic treatment of biliary fascioliasis [22]. Contrarily, serological methods have limitations when determining the efficacy of anthelminthic treatment because the presence of antibodies indicates previous exposure to the parasite rather than the existence of a current infection. Additionally, after successful anthelminthic treatment, several months have to pass for serological antibody-detection tests to become negative. Hence, the detection of specific antigens in feces allows for the confirmation of a current infection, whereas antibody detection tests need to be complemented by another technique to confirm the results obtained in treated subjects. Future studies should be carried out to determine the time required for negativization of MM3-COPRO ELISA results after effective treatment in humans.
The MM3-COPRO ELISA is also a reliable method for detecting F. gigantica coproantigens in fecal samples from experimentally infected sheep [32]. Although most reported cases of human fascioliasis are caused by F. hepatica, infections by F. gigantica have also been reported [25]. The fact that the MM3-COPRO ELISA can detect infections by both species may be of great value to ensure diagnosis of human and animal fascioliasis in countries where F. gigantica predominates, or where both species of Fasciola are present [25], [34].
Determining a patient's parasitic burden is crucial given the necessity to monitor drug treatment in order to prevent a hepatic colic as the consequence of the massive expulsion of liver-flukes [18], similar to other helminth diseases [47]. The Kato-Katz is usually employed as a coprological quantitative technique. Nevertheless, this technique has a low sensitivity, and the elaboration of several slides from the same individual stool sample is indispensable. The application of the Kato-Katz technique in community surveys becomes problematic because (i) it is pronouncedly time consuming when the number of samples is high, (ii) microscopic egg count is also time consuming in cases of heavy egg burdens, and (iii) it requires an additional technique to increase the sensitivity in areas where subjects shed a very low number of eggs in an intermittent way.
Results obtained in the samples from Huacullani showed that the concentration of coproantigens in feces is correlated with epg in the low burden group (<300 epg). This result agrees with a previous study using the MM3-COPRO ELISA in cattle, which showed that the concentration of coproantigens in feces is also correlated with the number of flukes found in the livers of animals collected after slaughter [22], as well as with the results of positive correlation found with another coproantigen test in fascioliasis infected patients in Cuba [30]. Nevertheless, our findings in the high burden group (≥300 epg) showed that the concentration of coproantigens in feces is not correlated with epg. This result agrees with the absence of any correlation between egg shedding in human samples from Hospital patients, measured by the Kato-Katz technique, and coproantigen concentration, measured by the MM3-COPRO ELISA [34]. One possible explanation for this discrepancy may be that the positive cases analyzed in Cuba [30] probably corresponded to recent infections with less than a year of age (early chronic stage), whereas our samples were from patients with chronic infections, in which egg excretion is probably more erratic. It must be kept in mind that fasciolid flukes may survive for up to 13.5 years in humans, and the pattern of egg shedding is not linear but fluctuates between maximum and minimum values [43]. By comparison, the kinetics of coproantigen release versus the kinetics of egg shedding showed a similar pattern but with a two-week time lag in epg [32].
In Cajamarca, chronic fascioliasis in valley samples, coproantigen levels did not show a good correlation with epg. Therefore, the use of only one coproproantigen technique appears to be insufficient to evaluate the fluke burden.
In these hyperendemic areas, the number of subjects who participate in surveys of this kind is very large, which implies the problem of transporting and preserving the fecal samples, as the coproantigen degrades at ambient temperature within a few days and the fecal material cannot be treated with any classical fixative. The monoclonal antibody MM3 recognizes a single conformational epitope, located in Fasciola cathepsins L1 and L2, which are the main cysteine proteases produced by adult flukes gut [44]. The stability of the antigen was observed during a period of 5 weeks, except for samples preserved in CoproGuard, which were observed for 17 weeks. Comparison of the different preservation conditions revealed that even when maintained at 37°C, only the antigenicity of coproantigens in the samples diluted with CoproGuard did not vary throughout the observation period. In contrast, biocides such as sodium azide and thimerosal did not preserve the antigenicity, as the start signal decreased to approximately 30% by the end of the observation period. When the samples were maintained at 4°C, the F. hepatica coproantigens retained about 70% of their initial antigenicity after 5 weeks. However, the antigens are relatively stable in some stools. This suggests that degradation of MM3-recognized Fasciola coproantigens depends on the presence of particular protease species, or other factors, which differ for each patient [34].
Other studies have also referred to the stability of Fasciola coproantigens. The monoclonal antibody F10 recognizes a 26–28 kDa antigen which is a monomeric proteoglycan secreted and excreted from the tegument and the gut of the flukes. The antigenicity of that coproantigen was noted to be stable or even enhanced by the action of proteolytic enzymes found in the digestive tract and under a variety of standard laboratory storage conditions. Storage at various temperatures resulted in some break down of the protein. The storage of the purified protein at room temperature overnight gave rise to several new bands ranging from 8 kDa to 20 kDa. Incubation of the purified coproantigen at 4°C for two months resulted in a major band at 8 kDa and a minor band at 20 kDa which decreased in size with longer incubation. Storage of ES for more than three years resulted in a major band at 8 kDa not seen in fresh ES. All these bands were recognized by the monoclonal antibody. The 26–28 kDa band was always detectable and the smaller bands are lower in intensity, suggesting that the coproantigen is relatively stable during storage. Thus, that degradation probably represents a loss of carbohydrate, since antigenicity is maintained [48], [49].
One possible alternative would be freezing the samples at −20°C, but this poses the additional problems of (i) difficult and expensive transport of frozen samples to the laboratory where determination is to take place, and (ii) the non-appropriateness of frozen samples for the diagnosis of other parasite species present in coinfections. Another solution is the use of Coproguard, which has been demonstrated to be convenient for sample preservation in this kind of surveys including the application of a coproantigen-detection test [34].
In Huacullani and Cajamarca, the PPVs calculated for diverse epidemiological situations are very different. PPVs in hyperendemic situations were very high, making this test recommendable for such situations. Contrarily, NPVs calculated for diverse epidemiological situations are similar.
Current efforts for the control of human fascioliasis need diagnostic techniques which allow for high sensitivity and specificity, large mass screening, detection in the chronic phase, early detection of treatment failure or reinfection in post-treated subjects, and usefulness in surveillance programs. Our results indicate that a coproantigen-detection test such as MM3-COPRO ELISA fulfils all these aspects. It provides a good tool to detect biliary fascioliasis in humans under field conditions in Andean hyperendemic countries, including a higher sensitivity than egg detection techniques, especially in areas where burdens are usually low, such as in areas of the valley transmission pattern. Hence, the MM3-COPRO ELISA appears to be not only useful for individual diagnosis in hospitals, but also in human surveys in fascioliasis endemic areas characterized by low to high parasitic burdens.
The present MM3-COPRO ELISA validation is expected to facilitate the improvement of human fascioliasis diagnosis in endemic areas (a commercial version of the MM3-COPRO ELISA is today available). The practical application of this sensitive and convenient method for large scale surveillance in the control programs in the Northern Bolivian Altiplano and Cajamarca could improve screening of human fascioliasis in these endemic areas by detecting infected humans in the biliary stage of the disease, as a large number of samples can easily be processed. Keeping in mind that most affected subjects are usually children, the attainment of fecal samples is easier and faster than taking blood samples, which is considered invasive. The former does not pose difficulties for community elders, school head teachers and parents who usually give their consent. Moreover, to many of these indigenous communities, blood extraction is culturally not acceptable.
Furthermore, our experience in Huacullani and Cajamarca indicates that MM3-COPRO ELISA offers the easiest and fastest way to adequately face large mass screenings, by initially applying the coproantigen technique to all the coprological samples obtained in the community survey and thereafter applying the Kato-Katz technique only or first in coproantigen-positive samples. It is recommended to treat subjects with coproantigen-positive samples but with negative egg detection. This allows for a quick selected treatment action, lending the positive effects of (i) fast response in the communities surveyed that verify that infected subjects are treated within a few days after the survey, and (ii) reducing the probability of drug resistance appearance. The remaining coproantigen-negative samples may finally be analyzed for the eventual detection and subsequent selected treatment of very few subjects shedding eggs, although this last step will unavoidably be time-consuming.
Given the aforementioned advantages a coproantigen-detection test offers, one wonders why there are only relatively few of such tests for parasitic diseases affecting the digestive system available: amebiasis [50], giardiasis [51], opisthorchiasis [52], taeniasis [53], trichinelliasis [54], strongyloidiasis [55], hookworm infection [56]. Developing and/or improving highly specific coproantigen-detection test for diseases in which coprological diagnosis requires specialized personnel and time-consuming microscope work would evidently be welcome.
|
10.1371/journal.ppat.1002300 | A Viral Nuclear Noncoding RNA Binds Re-localized Poly(A) Binding Protein and Is Required for Late KSHV Gene Expression | During the lytic phase of infection, the gamma herpesvirus Kaposi's Sarcoma-Associated Herpesvirus (KSHV) expresses a highly abundant, 1.1 kb nuclear noncoding RNA of unknown function. We observe that this polyadenylated nuclear (PAN) RNA avidly binds host poly(A)-binding protein C1 (PABPC1), which normally functions in the cytoplasm to bind the poly(A) tails of mRNAs, regulating mRNA stability and translation efficiency. During the lytic phase of KSHV infection, PABPC1 is re-localized to the nucleus as a consequence of expression of the viral shutoff exonuclease (SOX) protein; SOX also mediates the host shutoff effect in which host mRNAs are downregulated while viral mRNAs are selectively expressed. We show that whereas PAN RNA is not required for the host shutoff effect or for PABPC1 re-localization, SOX strongly upregulates the levels of PAN RNA in transient transfection experiments. This upregulation is destroyed by the same SOX mutation that ablates the host shutoff effect and PABPC1 nuclear re-localization or by removal of the poly(A) tail of PAN. In cells induced into the KSHV lytic phase, depletion of PAN RNA using RNase H-targeting antisense oligonucleotides reveals that it is necessary for the production of late viral proteins from mRNAs that are themselves polyadenylated. Our results add to the repertoire of functions ascribed to long noncoding RNAs and suggest a mechanism of action for nuclear noncoding RNAs in gamma herpesvirus infection.
| Almost all eukaryotic messenger RNAs (mRNAs) have a string of 150–200 adenylates at the 3′ end. This poly(A) tail has been implicated as important for regulating mRNA translation, stability and export. During the lytic phase of infection of Kaposi's Sarcoma-Associated Herpesvirus (KSHV), a noncoding viral RNA is synthesized that resembles an mRNA in that it is transcribed by RNA polymerase II, is methyl-G capped at the 5′ end, and is polyadenylated at the 3′ end; yet this RNA is never exported to the cytoplasm for translation. Rather, it builds up in the nucleus to exceedingly high levels. We present evidence that the function of this abundant, polyadenylated nuclear (PAN) RNA is to bind poly(A) binding protein, which normally binds poly(A) tails of mRNAs in the cytoplasm but is re-localized into the nucleus during lytic KSHV infection. The interaction between PAN RNA and re-localized poly(A) binding protein is important for formation of new virus, in particular for the synthesis of proteins made late in infection. Our study provides new insight into the function of this noncoding RNA during KSHV infection and expands recent discoveries regarding re-localization of poly(A) binding protein during many viral infections.
| Kaposi's Sarcoma-Associated Herpesvirus (KSHV) is the causative agent of several human cancers and immunoproliferative disorders, including Kaposi's Sarcoma, Multicentric Castleman's Disease and Primary Effusion Lymphoma [1], [2]. Like other herpesviruses, KSHV infection is characterized by two states: viral latency and lytic growth. During latency, very few viral genes are expressed, reducing the number of viral epitopes available to trigger a host immune response. Given appropriate but incompletely understood stimuli, the virus activates the lytic program of infection. This is characterized by three ordered waves of viral gene expression producing “immediate early,” “delayed early” and “late” proteins, as well as replication of the viral genome. Ultimately, the new genomes are packaged into virions, which are released from the cell for expansive host infection.
Upon KSHV entry into the lytic phase, an intronless viral noncoding (nc)RNA called polyadenylated nuclear (PAN) RNA, also known as T1.1 or nut-1, begins to be synthesized at unusually high levels [3], [4]. Although the 1.1 kb PAN RNA resembles an mRNA in being transcribed by RNA polymerase II, methyl-G capped at its 5′ end, and polyadenylated at its 3′ end, it is not exported to the cytoplasm for translation as are other viral transcripts. Instead, PAN RNA accumulates to astonishingly high levels, reaching ∼500,000 copies per nucleus and ultimately accounting for up to 80% of the total polyadenylated RNA in the cell [3]. Much has been learned regarding the mechanism that enables PAN RNA to accumulate to such high levels. Specifically, a 79-nucleotide element located near the 3′ end of the RNA, termed the expression and nuclear retention element (ENE), serves to stabilize the RNA in the nucleus [5], [6], [7]. Deletion of the ENE dramatically reduces the levels of transfected PAN RNA in HEK 293 cells, while insertion of the ENE into an intronless β-globin transcript significantly increases its nuclear levels. Insertion of the ENE has also been shown to enhance the abundance of nuclear pri-miRNAs [8]. It was hypothesized that a U-rich internal loop within the ENE engages the poly(A) tail, thereby sequestering it from deadenylases that initiate RNA decay [6], [7]. A recent x-ray crystal structure of the ENE complexed with oligo(A) reveals the formation of a triple helix that clamps the oligo(A) [9].
To address how PAN RNA contributes to lytic infection of KSHV, we began by investigating protein components of the PAN RNP and identified poly(A)-binding protein C1 (PABPC1). PABPC1 normally functions in the cytoplasm where it binds the poly(A) tails of mRNAs, regulating their stability by either antagonizing or enhancing the activity of cytoplasmic deadenylases [10], [11], [12], [13], [14]. PABPC1 also mediates circularization and enhances translation of mRNA via physical interactions with the translation initiation factor eIF4G, and assists in the export of mRNAs from the nucleus to the cytoplasm [14], [15], [16], [17]. However, since PAN RNA resides exclusively in the nucleus of KSHV-infected cells and does not shuttle (Conrad and Steitz, unpublished observations), re-localization of PABPC1 to the nucleus is a prerequisite for significant binding of PABPC1 to PAN RNA. Indeed, several groups have reported that re-localization of PABPC1 from the cytoplasm into the nucleus occurs during lytic KSHV infection of TIME endothelial cells and BC3 and BCBL1 TReX-RTA lymphoid cells [18], [19], [20]. The phenomenon is driven by the shutoff exonuclease (SOX) protein, as transient transfection of SOX into uninfected cells causes nuclear accumulation of PABPC1 in the absence of any other viral gene product [19]. SOX is also responsible for the host shutoff effect of the virus, which selectively downregulates host mRNAs while viral mRNAs persist [21], [22], [23]. SOX-mediated PABPC1 re-localization to the nucleus is critical for the host shutoff effect since knockdown of PABPC1 diminishes the ability of SOX to effect shutoff. GFP mRNA levels were downregulated in cells transfected with SOX, but less so in cells pre-treated with anti-PABPC1 siRNAs [19]. Furthermore, overexpression of PABPC1 targeted into the nucleus, using a nuclear retention signal from hnRNPC1, recapitulates aspects of host shutoff in the absence of any viral gene product [24]. Finally, a point mutation that disrupts the ability of SOX to re-localize PABPC1, but does not disrupt SOX's unrelated function as an alkaline DNase, ablates its ability to mediate host shutoff [19].
Here, we demonstrate that PAN RNA interacts with PABPC1 after it has re-localized to the nucleus during the lytic phase of KSHV infection. In transient transfection experiments, SOX strongly upregulates the levels of PAN RNA by a mechanism that is dependent on the ability of SOX to mediate the host shutoff effect, on the re-localization of PABPC1 into the nucleus and on the existence of a poly(A) tail on PAN RNA. In infected cultured cells activated into lytic phase, PAN RNA expression is coincident with the host shutoff effect, and correlates with PABPC1 re-localization from the cytoplasm into the nucleus. Yet, knockdown of PAN RNA using RNase H-targeting oligonucleotides shows that it is not required for the shutoff effect or the nuclear re-localization of PABPC1. Instead, the amount of virus released into the supernatant of cultured cells is severely reduced upon knockdown. This striking effect on viral titer is explained by our observation that knockdown of PAN RNA adversely affects the expression of a subset of viral genes during the late stage of lytic infection.
BCBL1 TReX-vector and TReX-RTA cells (gift from Jae Jung, USC) were maintained in RPMI supplemented with pen/strep, L-glutamine and 20% tetracycline-compatible FBS (Clonetech). iSLK.219, 293 DC-SIGN (gift from Robert Means and Sabine Lang, Yale University) and 293T cells were maintained in DMEM supplemented with pen/strep, L-glutamine and 10% FBS. 293 tet-on cells were maintained in DMEM supplemented with pen/strep, L-glutamine and 10% tetracycline-compatible FBS. Anti-orf6/SSB was a gift from G. Hayward (Johns Hopkins University), anti-LANA1 and anti-orf4 were gifts from Y. Chang (University of Pittsburg), anti-K8.1 antibody was from Advanced Biochemicals Incorporated, anti-orf65 antibody was a gift from G. Miller (Yale University), anti-myc antibody was from Santa Cruz Biotechnology. Electroporation was conducted in 0.4 cm electroporation cuvettes at 975 µF/220 mV for BCBL1 cells, and 975 µF/210 mV for iSLK.219 cells. 293T cells were transfected with MirusTransIT reagent according to the manufacturer's protocol. For some oligonucleotide knockdown experiments in BCBL1 TReX-RTA cells, cultures were first synchronized into S phase by serum starvation for 24 hours, followed by the re-addition of serum and incubation for 16 hours [25], [26]. Cells were then electroporated with RNase H-targeting modified oligonucleotides and allowed to recuperate overnight in RPMI supplemented with 20% tetracycline-compatible FBS (Clonetech), followed by induction the next day with 0.2 µg/mL doxycycline. Other transfection methods were tested, such as nucleofection with both Amaxa and Mirus reagents. Nucleofection with Amaxa reagent, as per the manufacturer's suggestions for BCBL1 cells, gave results comparable in terms of knockdown efficiency and cell death to the electroporation method used here.
KSHV-infected BCBL1 TReX-vector or TReX-RTA cultures were grown to a density of 0.8 million cells/mL, and then treated with 2 ug/mL doxycycline for 24 hours. 100 million cells were pelleted, washed with PBS, resuspended in 500 uL 50 mM Tris-HCl, pH 7, 100 mM KCl, 10 mM MgCl2, 1 mM DTT, 10% glycerol, 1 mM ATP, protease inhibitor cocktail (CalbioChem) and RNase Inhibitor (Roche), and syringe lysed by 10 passages through a 20G needle, 15 passages through a 25G7/8 needle and 20 passages through a 27G needle. Lysate was clarified by spinning at 14,000 x g for 10 minutes at 4°C, before nutating with 500 µL bed volume DEAE-sepharose resin that had been pre-equilibriated with lysis buffer. After 4 hours at 4°C, the resin was pelleted at 3000 rpm for 1 minute at 4°C, washed twice with lysis buffer, and subjected to successive batch elutions with increasing concentrations of KCl in lysis buffer. Maximal yield of the PAN RNP was observed in the 0.3 M KCl fraction. 1 mM MgCl2, 1.5 mM ATP, 5 mM creatine phosphate and 2–5 pmol/µL of biotinylated 2′-O-methylated anti-PAN oligonucleotides, complementary to nucleotides 70 to 89 and 993 to 1012 of PAN RNA, was added to 35 µL (175 µg) of the 0.3 M KCl fraction in a 100 µL total volume. 150 µL bed volume of streptavidin beads was pre-blocked at 4°C for 15 minutes in 100 µg/mL E. coli tRNA, 100 µg/mL glycogen and 1 mg/mL BSA, washed three times in 20 mM Tris pH 7.6, 50 mM NaCl and 0.01% NP-40. Extract was nutated with streptavidin beads at 4°C for 1 hour, before pelleting, removing the supernatant, washing beads twice with lysis buffer and eluting bound proteins by treatment with micrococcal nuclease (Worthington Biochemical Corporation) supplemented with 2 mM CaCl2 at 37°C for 30 minutes. Proteins were fractionated by 10% SDS PAGE; the PAN RNP-specific bands were excised and identified by mass spectrometry (Columbia University). Further details on all peptides identified in each band can be found at links provided in Fig. S2C.
BCBL1 TReX-vector or TReX-RTA cells were fixed onto glass slides pre-treated with poly-L-lysine, fixed with 4% paraformaldehyde and permeabilized with 0.5% Triton-X. Anti-PAN RNA oligonucleotides (SB2: ACAAATGCCACCTCACTTTGTCGC; SB85: CGCTGCTTTCCTTTCACATT; SB88: GTGAAGCGGCAGCCAAGGTGACTGG), which were labeled with digoxigenin-dUTP using the DIG oligonucleotide Tailing Kit (Roche), were hybridized with the samples in 50% formamide, 10% dextran sulfate, 2X SSC, 0.1% RNase-free BSA, 500 µg/mL salmon sperm DNA, 125 µg/mL E. coli tRNA and 1 mM vanadyl ribonucleoside complexes and detected using FITC-conjugated anti-DIG antibody (Jackson Lab Immunologicals). PABPC1 was visualized with anti-PABPC1 mouse monoclonal or rabbit polyclonal antibodies (gift of G. Dreyfuss, UPenn, and Abcam), and anti-mouse or anti-rabbit Alexa Fluor 594 antibody (Invitrogen). FLAG-PABPC1-NRS, a gift from B. Glaunsinger (UC Berkeley), was visualized with rabbit anti-FLAG antibody (Sigma) and anti-rabbit Alexa Fluor 594 (Invitrogen) [24]. rRNA was visualized with Y10B mouse monoclonal antibody [27] and anti-mouse Alexa Fluor 488 antibody (Invitrogen). Images were collected on a Leica TCS SP5 confocal microscope.
RNase H-targeting oligonucleotide [28], [29] sequences used were as follows. SB215 anti-PAN RNA oligo 1 (complementary to nucleotides 70 to 89 of PAN RNA): mC*mC*mA*mA*mU*G*A*A*A*A*C*C*A*G*A*mA*mG*mC*mG*mG; SB216 anti-PAN RNA oligo 2 (complementary to nucleotides 993 to 1012 of PAN RNA): mU*mG*mA*mG*mC*T*C*T*A*G*G*C*A*C*G*mU*mU*mA*mA*mA; SB230 anti-GFP oligo: mC*mU*mG*mC*mC*A*T*C*C*A*G*A*T*C*G*mU*mU*mA*mU*mC; SB232anti-K7 oligo (complementary to nucleotides -305 to -285 of K7 mRNA, with respect to the PAN RNA transcriptional start site): mA*mA*mU*mC*mG*A*G*C*A*G*A*G*T*A*G*mC*mC*mA*fmA*mG, where m represents 2′-O-methyl substitutions of the ribose ring and * represents phosphorothioate substitutions.
BCBL1 TReX-RTA cells were induced with doxycycline for 8 days. Supernatant was collected, passed through a 0.45 micron filter, and incubated with 6.25 µg/mL RNase A (Sigma) and 20 units/mL DNase One (New England Biolabs) for 1 hour at 37°C to degrade extracellular RNA and any viral DNA that had not been properly packaged into viral capsid protein and released into the supernatant. The supernatant was then spun at 15,000 rpm using a Beckman TLA 100.2 rotor for 2 hours at 4°C, and the resulting pellet was resuspended in 600 µL of lysis buffer (20 mM Tris-HCl pH 8.0, 10 mM EDTA, 100 mM NaCl and 0.5% SDS) and incubated at room temperature for 10 minutes to inactivate the DNase. Following proteinase K treatment (0.1 mg/mL final concentration and 2 hours incubation at 37°C), 3 ng of a control plasmid DNA was added to each sample as a normalization control for loss of DNA during subsequent steps of the purification process. After extraction with phenol:chloroform:isoamyl alcohol (25∶24∶1), 600 µL of the aqueous phase was precipitated with isopropanol, sodium acetate and GlycoBlue (Ambion), and viral DNA levels were quantified using qRT-PCR. The viral DNA signal was normalized to signal from the control plasmid DNA (J. Ziegelbauer, personal communication).
For measuring virus production from iSLK.219 cells, cells were induced with doxycycline for 2-3 days, and supernatant was collected and passed through a 0.45 micron filter. 50 µL of supernatant was added to 300 µL of DMEM +10% FBS and placed on 293 DC-SIGN cells that had been grown in 12-well plates. Cells were then spun at 2000 rpm for 1 hour at room temperature in a Beckman Coulter J6-MI centrifuge fitted with a JS-4.2A rotor [30]. 48 hours later cells were harvested by trypsin, washed with PBS and fixed with 4% formaldehyde in PBS for 30 minutes at room temperature. GFP levels were quantified by flow cytometry using BD Biosciences' FACSCalibur platform.
BCBL1 TReX-RTA cells were electroporated and induced as above for 3 days. 5 million live cells were washed with PBS, resuspended in 250 mM Tris pH 8.5, 125 mM EDTA, 1 mg/mL protease K and 1% SDS, and incubated at 60°C for 2 hours. 1/5th volume of 5 M potassium acetate pH 5.2 was added, lysate was incubated on ice for 20 minutes, and clarified by centrifugation at 12,000 rpm for 15 minutes at 4°C. Supernatant was transferred to a fresh tube, and incubated with 20 µg/mL RNase A at 37°C for 15 minutes. Then, 2.5-fold volume of ice-cold 100% ethanol was added, and DNA was precipitated overnight at −20°C. Pellets were resuspended in 100 µL water, and viral DNA levels were measured by qRT-PCR as above, normalizing to human DNA using GAPDH-specific primers.
To determine what proteins bind to PAN RNA in the nucleus of lytically infected cells, we isolated the PAN RNP from a KSHV-infected cell line containing a doxycycline-inducible version of RTA (BCBL1 TReX-RTA), the viral transcription factor that is necessary and sufficient to promote entry into the lytic phase of viral infection [31], [32], [33]. After anion exchange and affinity chromatography using a selection oligonucleotide complementary to PAN RNA (Fig. 1), we identified both hnRNP C1/C2 and PABPC1 as specifically co-purifying with PAN RNA (see Fig. S1 and S2 for fractionation and mass spectrometry data). These proteins were not seen when extracts from the control doxycycline-treated BCBL1 TReX-vector cell line were used. hnRNP C1 has been previously observed to bind to PAN RNA, although the significance of this interaction is not understood [34].
Since PAN RNA is exclusively nuclear, proteins that interact with PAN RNA must reside in the nucleus. While PABPC1 is essentially a cytoplasmic protein, nuclear re-localization of PABPC1 during KSHV lytic phase had been reported in the literature [18], [19], [20]. We therefore performed immunofluorescence experiments to confirm that PABPC1 re-localizes into the nucleus of the BCBL1 TReX-RTA cells [20] and to establish whether this re-localization correlates with the expression of PAN RNA. Cells were fixed and stained with primary and secondary antibodies to visualize PABPC1, and in situ hybridization probes and detection antibodies to visualize PAN RNA. As seen in Fig. 1C, PABPC1 indeed re-localizes into the nucleus during lytic infection, and the subnuclear distribution of PAN RNA and of re-localized PABPC1 appear strikingly similar. rRNA, however, remains predominantly cytoplasmic during both latent and lytic stages of infection (Fig. S3). We manually scored the immunofluorescence pattern of several hundred cells from a number of microscopic fields, and found that ∼80% of the cells in which PABPC1 had re-localized into the nucleus also expressed PAN RNA (representative images from the manual scoring experiment are provided in Fig. S4). The coincidence of PABPC1 re-localization and PAN RNA expression was further verified in the original BCBL1 cell line (data not shown). High throughput analysis, using multispectral imaging flow cytometry (Amnis Corporation), of PABPC1 nuclear re-localization and PAN RNA expression in BCBL1 cells has extended this correlation (S. Borah, L. Hassman, D.H. Kedes and J.A. Steitz, manuscript in preparation).
Additionally, we noted that unique peptides in the mass spectrometry data also identified PABPC4 as a protein that co-purifies with PAN RNA (Fig. S2A). Recently, nuclear re-localization of this protein has also been reported to result from KSHV lytic infection; knockdown of both PABPC1 and PABPC4 in transient SOX-transfection experiments revealed that these proteins might have redundant functions with respect to SOX-mediated nuclear retention of polyadenylated RNA in 293T cells [24].
Since SOX induces both host shutoff and the nuclear accumulation of PABPC1 [19], [22], and since PAN RNA binds re-localized PABPC1 in the nucleus (see Fig. 1), we hypothesized that PAN RNA might be involved in the host shutoff effect mediated by SOX. To elucidate the relationship between PAN RNA, the SOX protein and the host shutoff effect, we co-expressed PAN RNA and SOX protein in transient transfection experiments, in the absence of other viral genes. As seen in Fig. 2A, transfection of SOX protein into 293 tet-on cells had no effect on 18S rRNA or the RNase P RNA, which are normally unaffected during host shutoff, but resulted in a modest 2- to 4-fold reduction in the level of endogenous GAPDH mRNA and a 2- to 3-fold reduction in the level of co-transfected GFP mRNA. In contrast, transfection of SOX very strongly stimulated the level of PAN RNA (20- to 30-fold). This stimulation was independent of the tetracycline-regulated promoter used to express PAN RNA, as upregulation was observed in experiments using PAN's endogenous RTA-dependent promoter (Fig. 2B and 2C) or a CMV promoter (data not shown).
The effect of SOX on PAN levels was also seen by immunofluorescence (Fig. 2B). When a PAN RNA expression vector that includes 1 kb of PAN's endogenous, RTA-responsive promoter was co-transfected into 293T cells with an RTA expression plasmid, only an occasional in situ hybridization signal for PAN RNA was observed (see Fig. 2B, middle right panel). However, when the SOX gene was co-transfected, the PAN RNA in situ hybridization signal was robust (bottom right panel). Furthermore, those cells in which a robust PAN RNA in situ signal was observed were the same cells in which PABPC1 had re-localized to the nucleus (Fig. 2B, middle column). This suggested that high levels of PAN RNA expression are not only dependent on SOX expression, but might also involve PABPC1 re-localization. Accordingly, when a mutant version of PAN RNA in which the poly(A) tail was replaced by a 3′ stem-loop structure derived from a histone mRNA [6] was transfected into these cells, SOX-mediated enhancement of PAN RNA levels was reduced by about 6-fold (see Fig. 2A, PAN RNA Δ-poly(A) tail, and Fig. S5).
The SOX protein is a dual-function protein, acting both as the mediator of the host shutoff effect and as an alkaline exonuclease that is important for processing the viral DNA [35]. These two functions are separable; point mutants that are unable to carry out the shutoff function are wild-type for exonuclease function, and vice versa. Moreover, mutations that inactivate only the shutoff function of the protein also ablate its ability to re-localize PABPC1 into the nucleus, arguing for a link between these phenomena. We therefore tested each of these mutants to determine which function of SOX is responsible for the strong stimulation of PAN RNA levels. As seen in Fig. 2C, the P176S SOX mutant, which retains exonuclease activity but does not mediate host shutoff, failed to enhance PAN RNA expression. Conversely, the Q129H mutant, which retains host shutoff but does not exhibit exonuclease activity, stimulated PAN RNA expression to the same extent as wild-type SOX. These results indicate that SOX's exonuclease activity is not required for its stimulatory effect on PAN RNA. They are also consistent with the possibility that upregulation of PAN is mediated by SOX's host shutoff function and ability to re-localize PABPC1.
To address whether nuclear localized PABPC1 alone can upregulate levels of PAN RNA, in the absence of SOX, we co-transfected genes for PAN RNA and a mutant PABPC1-NRS that is retained in the nucleus [24]. As seen in Fig. 2D, co-expression of PABPC1-NRS resulted in upregulation of PAN RNA. As with SOX expression, upregulation of PAN RNA levels were specific to those cells in which nuclear retention of PABPC1 had occurred (Fig. 2E). While this result clearly demonstrates that nuclear accumulation of PABPC1 results in increased PAN RNA levels, the levels of PAN RNA did not reach levels seen upon co-transfection with SOX (see Fig. 2D). This may be due to a different efficiency of PABPC1 nuclear re-localization in SOX-transfected versus PABPC1-NRS-transfected cells, or due to some other aspect of SOX expression that further augments PAN RNA expression, in addition to the nuclear localization of PABPC1.
A previous study showed that although SOX and its murine herpesvirus-68 homolog reside in both the nucleus and cytoplasm, it is the cytoplasmic fraction of these proteins that is responsible for the host shutoff effect and for PABPC1 re-localization [36]. A more recent study demonstrated that SOX promotes PABPC1 re-localization by depleting cytoplasmic polyadenylated RNA, which in turn allowed direct intereaction between PABPC1 and importin α [37]. The fact that upregulation of PAN by SOX (Fig. 2A and B) depends on the nuclear re-localization of PABPC1, as well as on the presence of PAN's poly(A) tail, suggests how cytoplasmic SOX protein strongly upregulates the levels of a nuclear RNA: SOX indirectly enhances PAN abundance through re-localizing PABPC1, which in turn stabilizes PAN RNA by binding to its poly(A) tail. This scenario is consistent with the role of poly(A) tails as stabilizing constituents of mRNAs by antagonizing deadenylases [14] and with the presence of nuclear degradation factors that destabilize PAN in the absence of its poly(A) tail [6].
If increased levels of PAN RNA in the nuclei of lytically reactivated cells are due to PABPC1 re-localization, as observed in transiently transfected cells (Fig. 2), then PAN RNA should accumulate at the same time during the course of KSHV infection that host mRNA levels decrease because of the shutoff effect. To test this, BCBL1 TReX-vector or TReX-RTA cells were induced into lytic phase, and total RNA was extracted from culture samples after different times. Northern blot and qRT-PCR analysis (Fig. 3A and B) showed that increases in SOX and PAN RNAs coincide with the decrease in β-actin and GAPDH mRNA (∼12–24 hours post-induction). Thus, PAN RNA expression correlates with the host shutoff effect in the context of actual viral infection. Furthermore, at approximately the same time during the course of lytic infection, PABPC1 re-localizes into the nucleus (data not shown and [19]).
To ask whether lytic induction results in the accumulation of other abundant, polyadenylated noncoding RNAs coincident with PABPC1 re-localization, we measured the levels of host MALAT1 and NEAT1 RNAs by northern blot and qRT-PCR. MALAT1 RNA exists either as a long polyadenylated transcript that localizes to nuclear speckles, or a shorter transcript that localizes to the cytoplasm [38]. Similarly, NEAT1 RNA exists either as a shorter polyadenylated transcript or a longer transcript whose 3′ end is generated by endonucleolytic cleavage by RNase P [39]. We observed that the levels of the nuclear, polyadenylated forms of MALAT1 or NEAT1 RNA did not increase dramatically (Fig. 3A and C) but rather declined in abundance during lytic infection, after normalization to 18S rRNA levels. Thus, PAN RNA selectively accumulates in the nucleus during the KSHV lytic phase, while two other similar noncoding polyadenylated RNAs, MALAT1 RNA and NEAT1 RNA, do not.
To investigate whether PAN RNA expression directly contributes to the host shutoff effect and/or PABPC1 re-localization during the KSHV lytic phase, we knocked down the levels of PAN RNA using 2′-O-methylated and phosphorothioate-substituted antisense oligonucleotides that target endogenous RNase H to cleave PAN RNA (α-PAN oligos 1 and 2 in Fig. 4A) [28]. Interpretation of PAN knockdown experiments is complicated by the fact that the region of the KSHV genome from which PAN RNA is transcribed is included in the overlapping K7/survivin transcript (Fig. 4A) [40]. To differentiate between the involvement of PAN RNA and the K7/survivin protein, we designed an additional antisense oligonucleotide that targets the K7 transcript in a region not included in the sequence of PAN RNA (α-K7 oligo in Fig. 4A). We reasoned that any results of transfection of oligonucleotides targeting both PAN RNA and K7 mRNA (Fig. 4A, α-PAN oligos 1 and 2) that are not seen upon transfection of the K7-specific oligonucleotide (α-K7 oligo) would be specifically due to PAN RNA knockdown. First, we showed by qRT-PCR that the K7-specific oligonucleotide selectively lowered the level of the K7 mRNA ∼20-fold (Fig. 4B). However, the K7-specific oligonucleotide did not downregulate levels of PAN RNA, as seen in the northern blot of Fig. 4C. The combination of the two anti-PAN RNA oligonucleotides decreased not only PAN RNA to 5%–10% of wild-type levels in BCBL1 TReX-RTA cells (see Fig. 4C), but also the K7 mRNA, as expected (see Fig. 4B). Thus, differential knockdown of the overlapping K7 and PAN transcripts can be achieved by use of modified oligonucleotides.
Revealingly, the anti-PAN oligonucleotides had no effect on the host shutoff effect, as indicated by GAPDH and β-actin mRNA levels, despite a knockdown of PAN RNA levels by ∼90%–95% (Fig. 5). Transfection with a GFP- or K7-specific control oligonucleotide also did not alter the magnitude of host shutoff in BCBL1 TReX-RTA cells (Fig. S6). Confocal analysis indicated that PABPC1 was re-localized into the nucleus of BCBL1 TReX-RTA cells upon lytic induction despite PAN RNA knockdown (data not shown). We conclude that the expression of PAN RNA may be related to but is not required for SOX-dependent host shutoff during the KSHV lytic phase.
Since knockdown of PAN RNA does not appear to affect the SOX-induced shutoff of host protein synthesis, we asked whether PAN RNA might alter the expression of viral proteins instead. We compared the effects of knockdown of PAN RNA on viral protein expression in two different cells lines: BCBL1 TReX-RTA cells, which harbor the authentic KSHV virus, and iSLK.219 cells, in which the recombinant rKSHV.219 virus has been re-introduced into a KS cell line that has lost the original KSHV genome [41], [42]. In both these cell lines, the virus is induced into lytic phase by overexpression of RTA controlled by a tetracycline-responsive promoter.
Western blot analyses of lysates from BCBL1 TReX-RTA that had been transfected with no (mock), anti-GFP, anti-K7 or anti-PAN RNA oligonucleotides revealed that knockdown of PAN RNA affects late rather than early KSHV proteins (Fig. 6). Several early viral proteins, such as Orf50/RTA, vIL-6 and Orf6/ssDNA-binding protein (Fig. 6A), as well as the latency-associated nuclear antigen Orf73/LANA1 (data not shown), accumulate normally. In contrast, the levels of the late viral proteins K8.1 and Orf65/small viral capsid antigen (Fig. 6A), as well as Orf4/complement-binding protein (data not shown) were significantly lowered upon knockdown of PAN RNA, compared to the control oligonucleotides against GFP or K7 mRNAs. Densitometric quantifications by two different methods (see Fig. S7 legend for description of methods) of the immunoblot signals for the immediate early protein RTA, early protein vIL-6 and late protein K8.1 from 9 independent experiments are averaged in Fig. 6B (7 independent experiments for signal from cells transfected with anti-K7 oligonucleotide). The efficiency of PAN RNA knockdown ranged from 75% to 95% (data not shown) and the immunoblots from all 9 experiments are presented in Fig. S7. Together these data demonstrate the selective downregulation of the late protein K8.1 upon knockdown of PAN RNA.
Although non-specific toxic effects were experienced upon electroporation followed by lytic induction, particularly in the BCBL1 TReX-RTA cells (see Materials and Methods), similar levels of cell death were seen regardless of the oligonucleotide transfected (Fig. 6C). Cultures were routinely stained by trypan blue and scored for percent viability prior to harvest in order to analyze lysate from the same number of living cells for each condition. Thus, the loss of late gene expression in PAN knockdown cells is not due simply to increased levels of cell death.
Likewise, several lines of evidence indicate that the decreased levels of late protein expression observed upon knockdown of PAN RNA are not simply due to a non-specific decrease in RTA expression. Although some of the 9 immunoblot experiments revealed a decrease in RTA expression upon transfection with the K7 and PAN RNA oligonucleotides (Fig. 6B), in others a significant decrease in K8.1 expression was observed despite robust expression of RTA (see Fig. S7, panels A, D, E, G and H). Moreover, an independent assay was devised to test whether the effect of PAN RNA knockdown on K8.1 was due to decreased RTA expression. The expression levels of both K8.1 and myc-tagged RTA were measured by dual immunofluorescence staining and analysis by confocal microscopy. Manual scoring of RTA-positive BCBL1 TReX-RTA cells revealed that fewer cells expressed K8.1 upon knockdown of PAN RNA (Fig. S8). Importantly, since K8.1 expression was scored only in cells that also expressed RTA, the observed effect could not be due to decreased RTA expression. We conclude that the effect of PAN RNA knockdown on K8.1 expression is independent of any effect on RTA.
The results of PAN RNA knockdown in iSLK.219 cells were also assessed (Fig. 7A). Knockdown efficiency in this cell line was 90%–95% (data not shown). In iSLK.219 cells, knockdown of PAN RNA resulted in some decrease in all viral proteins tested, including the early proteins Orf6/ssDNA binding protein and vIL-6 (Fig. 7A). However, the effect on the late K8.1 protein was much more pronounced: K8.1 protein levels dropped 12.5-fold in cells transfected with anti-PAN oligonucleotides compared to cells transfected with anti-K7 oligonucleotide (Fig. 7B), whereas vIL-6 protein levels decreased only 2.5-fold. We conclude that PAN RNA preferentially enhances the expression of late viral proteins in KSHV-infected cells.
Since knockdown of PAN RNA adversely affects expression of important late viral genes, we predicted that its knockdown would affect overall viral yield. To assess the role of PAN RNA in the production of new virus, we harvested virus from the supernatant of induced BCBL1 TReX-RTA cells 8 days post-induction. As seen in Fig. 8A, knockdown of PAN RNA significantly reduced viral production, as assayed by qPCR measurement of DNase resistant, encapsulated viral DNA released into the supernatant. The effect was comparable to treatment with ganciclovir, an inhibitor of viral DNA replication [43], [44].
The effect of PAN RNA knockdown on production of infectious virus from iSLK.219 cells was also assayed by harvesting the supernatant from induced iSLK.219 cells and infecting target 293 cells that stably express the DC-SIGN receptor. Infection was assessed by visualization (Fig. 7C) and by FACS analysis of GFP expression arising from the recombinant KSHV genome in target cells 48 hours later. Fig. 7C reveals that treatment of iSLK.219 cells with anti-PAN RNA oligonucleotides resulted in virtually no GFP-positive 293 DC-SIGN cells upon infection with virus harvested from cells treated with anti-PAN oligonucleotides. However, GFP-positive cells were detected among 293 DC-SIGN cells incubated with virus harvested from cells treated with the anti-K7 oligonucleotide. Thus, release of infectious virus as well as encapsulated viral DNA levels were both diminished by knocking down PAN RNA.
Since viral DNA replication is required for the expression of herpesvirus late proteins [45], [46], [47], the effect of PAN RNA knockdown on the accumulation of intracellular viral DNA during the lytic phase was also measured by qPCR. Increases in intracellular viral DNA levels 4 days post-induction in BCBL1 TReX-RTA cells were very modest (∼5–7 fold), in agreement with other reports [48], [49]; this level was further decreased when cells were electroporated prior to induction. The low levels of intracellular viral DNA make it difficult to conclude definitively whether the knockdown of PAN RNA specifically inhibits viral DNA replication. However, no dramatic differences in intracellular viral DNA levels were detectable in cells transfected with control versus anti-PAN RNA oligonucleotides (Fig. 8B), suggesting that PAN RNA is not directly involved in viral DNA replication.
The data presented here indicate that the highly abundant PAN RNA binds to the normally cytoplasmic poly(A) binding protein PABPC1 once it has been re-localized to the nucleus of KSHV-infected cells (Fig. 1). This is supported by both the composition of the PAN RNP and the observation that the detailed patterns of PAN RNA and PABPC1 concentration within the nucleus coincide. The abundance of PAN RNA (upwards of 0.5×106 transcripts per nucleus) corresponds well to the abundance of PABPC1 (estimated at 7×106 copies in the average HeLa cell), given that the length of PAN RNA's poly(A) tail is estimated to be similar to that of the average host mRNA (Conrad and Steitz, data not shown) and that each PAN RNA thus likely binds 8–10 PABPC1 molecules [3], [50]. Our results extend prior reports of PABPC1 re-localization by SOX protein alone and the concurrent host shutoff effect [19] by showing that that the level of PAN RNA is likewise dependent on expression of the SOX protein in transient transfection assays (Fig. 2). In infected cells, PAN RNA is not required for the re-localization of PABPC1 nor for the host shutoff effect (Fig. 5), which promotes this re-localization.
Since PAN RNA expression is concurrent with the host shutoff effect in the context of viral lytic infection (Fig. 3), PAN RNA likely functions downstream of SOX action. An alternative possibility-that RTA levels (and therefore all late functions) are compromised by treatment with anti-PAN oligonucleotides-is not consistent with several observations. Because lytic reactivation is dependent on RTA expression, an across-the-board reduction in all lytic-related events would be expected but was not observed. Specifically, PAN RNA knockdown did not affect: 1) the host shutoff effect and PABPC1 nuclear re-localization (Figs. 5 and S6 and data not shown), 2) accumulation of intracellular viral DNA (Fig. 8B), and 3) expression of the viral lytic marker vIL-6 (Fig. 6A, B). Instead, knockdown of PAN RNA adversely affected the expression of late viral genes in cells of both lymphoid (BCBL1, Figs. 6, S7 and S8) and endothelial (iSLK.219, Fig. 7) origin, which are two major KSHV targets in vivo.
Ultimately, knockdown of PAN RNA adversely affects release of new infectious virus into the supernatant from BCBL1 TReX-RTA cells (Fig. 8A), as would be expected from the downregulation of late viral gene expression. This effect appears to be independent of viral DNA replication (Fig. 8B). We conclude that PAN RNA plays an important role in the expression of a subset of viral genes, perhaps related to the nuclear re-localization of PABPC1. Further work is needed to establish whether this effect is at the level of transcription, translation or mRNA stability, although selective downregulation of mRNAs for the late genes K8.1, Orf18 and Orf29 was observed using qRT-PCR analysis (data not shown). It is also possible that the effects seen on virus production and late viral protein expression result from dual knockdown of both PAN RNA and K7. However, the fact that the knockdown of PAN RNA is required to observe these effects, as K7 knockdown alone is insufficient, underscores the importance of PAN RNA.
KSHV targets host gene expression at several levels during the lytic phase. First, host mRNA levels are specifically downregulated by the SOX protein, a phenomenon that is itself related to nuclear re-localization of PABPC1, a highly abundant translation factor that displays nanomolar affinity for the polyadenylate tails of mRNAs [24], [50]. Rowe and colleagues have found that the Epstein Barr virus SOX homolog BGLF5 mediates host shutoff in EBV-infected cells [51], and Glaunsinger and colleagues have extended their findings of SOX function to the homologous murine herpesvirus-68 (MHV-68) protein [36]. Both EBV and MHV-68 SOX homologs drive PABPC1 re-localization into the nucleus, like their KSHV-counterpart [19], and we have extended the observation of PABPC1 nuclear re-localization to EBV-infected HH514-16 cells upon lytic activation (S. Borah, R. Park, G. Miller and J.A. Steitz, unpublished observations). Second, in addition to the downregulation of host mRNA levels, other changes in host translation during the KSHV lytic phase [20] have been reported to include increased levels of 4E-BP1 and eIF4E phosphorylation, which are expected to enhance rates of eIF4F assembly onto and of translation of viral mRNAs. Third, KSHV mRNAs, many of which are unspliced, are preferentially exported by the viral export factor Orf57, which tethers the nuclear export factor TAP protein to viral mRNAs and enhances their translation [52], [53], [54].
Our data support a model in which the highly abundant PAN RNA contributes importantly to the viral manipulation of gene expression. Aspects of viral infection that downregulate host gene expression, such as host shutoff and PABPC1 re-localization, do not require PAN RNA, but instead affect the expression level of PAN RNA itself both in transiently transfected cells and in bona fide infected cells. PAN RNA then impacts the expression of at least a subset of viral genes; possible molecular mechanisms for this regulation are presented below.
Poly(A) binding protein is a central player in cellular gene expression and is targeted by a number of viruses, usually as a means to achieve shutdown of host gene expression. Some, such as picornaviruses and caliciviruses, have evolved specific proteases that cleave PABPC1 [55], [56], [57]. Rotavirus instead expresses the NSP3 protein, which displaces PABPC1 from binding to eIF4G. This reduces the translational efficiency of host mRNAs, whereas viral mRNAs, which lack a poly(A) tail, are efficiently translated [58]. Expression of NSP3 also leads to nuclear accumulation of PABPC1, although the significance of this re-localization is not fully understood [59], [60]. Nuclear re-localization of PABPC1 is also observed during infection with Bunyamwera virus. Furthermore, although siRNA-mediated knockdown of PABPC1 decreased translation of a polyadenylated luciferase reporter, translation of a reporter whose 3′ end was derived from Bunyamwera virus mRNA, and thus lacked a poly(A) tail, was unaffected by knockdown [61].
PABPN1, a distinct nuclear poly(A) binding protein [14], is also targeted by viruses. Influenza NS1 protein binds both PABPN1 and the cleavage and polyadenylation stimulation factor (CPSF) within the assembled 3′-end processing machinery [62]. Thus, poly(A) polymerase does not processively extend the poly(A) tail of nascent mRNAs past ∼12 residues, and multiple PABPN1 molecules do not assemble onto the mRNA. This failure to properly form a poly(A) tail [63] and interact with PABPN1 are thought to underlie the lack of host mRNA export during influenza virus infection [64]. NS1 has other effects on PABPN1 as well, including inhibition of PABPN1 shuttling and re-distribution of PABPN1 from nuclear speckles to a uniform pattern throughout the nucleoplasm [62]. However, to date, no role for a noncoding RNA has been identified in the mechanisms by which these viruses target PABPC1 for host shutoff.
Given the striking abundance of PAN RNA and the role that it plays in KSHV gene expression within the context of PABPC1 re-localization, it might be expected that homologs of this RNA would be found in related herpesviruses in which re-localization of PABPC1 occurs. Viruses whose mRNAs lack poly(A) tails, such as rotavirus and Bunyamwera virus, would be exceptions. Indeed, the EBV homolog of the SOX protein, BGLF5, has been shown to mediate host shutoff in cells infected with EBV and to drive PABPC1 re-localization [36], [51]. Although no PAN-like RNA has yet been reported in EBV or other herpesviruses, we have recently discovered putative homologs in several members of the gamma herpesvirus family. Indeed, expression of a PAN homolog has been verified in the closely related Rhesus Rhadinovirus (RRV), even though the overlapping K7 open reading frame does not appear to be conserved (K. Tycowski, S. Borah, M. Shu and J.A. Steitz, manuscript in preparation) [65], [66]. Additional studies are required to explore the possibility that PAN RNA-like RNAs also exist in more distantly-related viruses such as EBV and MHV-68. Interestingly, the sequence of PAN-like RNAs may not be conserved, whereas their abundance, nuclear localization and the presence of a poly(A) tail may be critical for function. Furthermore, there is no reason to expect that the role of PAN RNA be fulfilled by a single RNA transcript. Perhaps in other viruses, PAN RNA is functionally replaced by two or three different polyadenylated, nuclear RNAs, which together sequester PABPC1. Immunoprecipitation of PABPC1 from the nuclei of other herpesviruses during the lytic phase and sequencing the RNAs that co-precipitate might identify such PAN-like RNAs.
Interestingly, the best-characterized viral noncoding RNAs, the VA RNAs of adenovirus, which facilitate viral translation by suppressing protein kinase R and competitively inhibiting miRNA export and processing [67], [68], are not found outside of adenoviruses. Perhaps their biological functions are fulfilled by alternative mechanisms in other viruses.
By what molecular mechanism might PAN RNA enhance the expression of late viral genes, and could this role be related to its ability to bind re-localized PABPC1 in the nucleus of KSHV-infected cells? It is first critical to understand why PABPC1 re-localization might be harmful to the cell's normal functioning. PABPC1 is proposed to have three major functions: 1) to synergistically enhance mRNA translation by interaction with eIF4G, 2) to regulate mRNA fate by inhibiting deadenylation of the poly(A) tail or promoting interaction with deadenylating factors, and 3) to facilitate nuclear export of mRNA [14]. It seems likely that loss of PABPC1 from the cytoplasm would be detrimental to any or all of these functions in uninfected cells. In the context of viral infection, re-localization of PABPC1 appears to be important for the shutoff effect since siRNA-mediated knockdown of PABPC1 diminishes the ability of SOX to target GFP mRNA for degradation, and since point mutations in SOX that ablate its effect on PABPC1 re-localization also abolish its downregulation of host mRNAs [19]. Importantly, aspects of host shutoff have been recapitulated by transient transfection of a nuclear targeted PABPC1 protein [24]. These results argue that it is not merely the absence of PABPC1 from the cytoplasm that is detrimental for host mRNA stability, but that its presence in the nucleus is pivotal.
How might nuclear re-localized PABPC1 contribute to host mRNA degradation? Under normal conditions, PABPN1 is required for the synthesis of and co-transciptionally binds to the newly-extended poly(A) tails of mRNAs. It has been suggested that PABPN1 is further required for proper export of mRNA via its interaction with nuclear export factors [63]. PABPC1, which shuttles in and out of the nucleus, is also thought to aid in mRNA export by interacting with polyadenylated mRNA at an early stage in maturation [69], [70], [71], [72], [73]. The dramatic influx of PABPC1 into the nucleus during KSHV lytic infection might therefore perturb the process of mRNA nuclear export. Given the extreme abundance of PABPC1 and its higher affinity for poly(A) tails, nuclear PABPC1 (Kd ∼7 nM) could displace PABPN1 (Kd ∼555 nM) from polyadenylated mRNAs and interfere with PABPN1's role in export [50], [74]. As mRNAs that are not properly processed and exported from the nucleus become hyperadenylated and degraded [19], [75], [76], [77], the re-localization of PABPC1 to the nucleus could be the initiating step in their degradation during shutoff (Fig. 9A and B). Similar ideas are discussed in two recent studies, by Glaunsinger and colleagues, on the inhibition of nuclear PABPC1 on mRNA export [24], [37].
If a highly abundant, inert, polyadenylated RNA were expressed in the nucleus, then the re-localized PABPC1 would be stoichiometrically bound, or nearly so. Thus, PAN RNA could serve as a “buffer” against changes in nuclear PABPC1 concentration (Fig. 9C). The fact that host-cell encoded abundant, nuclear polyadenylated RNAs, such as MALAT1 and NEAT1 RNA, do not accumulate to exceptionally high levels in response to PABPC1 re-localization (Fig. 3) suggests that there is something unique about PAN RNA. It should be noted that PABPN1 knockdown also diminishes SOX-mediated mRNA degradation, so it may be the balance of the two proteins on the poly(A) tail that is critical [19]. A comparison of intranuclear PABPC1 and PABPN1 immunofluorescence signals relative to PAN RNA in situ hybridization signal should be performed.
A challenge for the model (Fig. 9) is explaining how PAN RNA specifically protects viral mRNAs. Should not host mRNAs be equally protected from the negative effects of PABPC1 re-localization by the abundant expression of PAN RNA? Some possible explanations derive from fundamental differences between viral and host mRNAs in the nucleus. First, the viral Orf57 protein selectively binds intronless viral mRNAs, mediating their export by anchoring the human transcription and export complex (hTREX) [52]. Since most KSHV genes lack introns, Orf57 is not only important, but in fact essential for virus replication [53]. Second, transcription rates of viral mRNAs appear to greatly outpace those of host mRNAs during the lytic phase [23]. Thus, viral mRNAs may be preferentially exported and translated during the late lytic phase simply because they outcompete their host counterparts. PAN RNA would therefore function in concert with several other proteins (PABPN1 and PABPC1, SOX and Orf57) to create an environment that favors the export and expression of viral but not host mRNAs (Fig. 9C).
|
10.1371/journal.pgen.1005894 | Novel NEK8 Mutations Cause Severe Syndromic Renal Cystic Dysplasia through YAP Dysregulation | Ciliopathies are a group of genetic multi-systemic disorders related to dysfunction of the primary cilium, a sensory organelle present at the cell surface that regulates key signaling pathways during development and tissue homeostasis. In order to identify novel genes whose mutations would cause severe developmental ciliopathies, >500 patients/fetuses were analyzed by a targeted high throughput sequencing approach allowing exome sequencing of >1200 ciliary genes. NEK8/NPHP9 mutations were identified in five cases with severe overlapping phenotypes including renal cystic dysplasia/hypodysplasia, situs inversus, cardiopathy with hypertrophic septum and bile duct paucity. These cases highlight a genotype-phenotype correlation, with missense and nonsense mutations associated with hypodysplasia and enlarged cystic organs, respectively. Functional analyses of NEK8 mutations in patient fibroblasts and mIMCD3 cells showed that these mutations differentially affect ciliogenesis, proliferation/apoptosis/DNA damage response, as well as epithelial morphogenesis. Notably, missense mutations exacerbated some of the defects due to NEK8 loss of function, highlighting their likely gain-of-function effect. We also showed that NEK8 missense and loss-of-function mutations differentially affect the regulation of the main Hippo signaling effector, YAP, as well as the expression of its target genes in patient fibroblasts and renal cells. YAP imbalance was also observed in enlarged spheroids of Nek8-invalidated renal epithelial cells grown in 3D culture, as well as in cystic kidneys of Jck mice. Moreover, co-injection of nek8 MO with WT or mutated NEK8-GFP RNA in zebrafish embryos led to shortened dorsally curved body axis, similar to embryos injected with human YAP RNA. Finally, treatment with Verteporfin, an inhibitor of YAP transcriptional activity, partially rescued the 3D spheroid defects of Nek8-invalidated cells and the abnormalities of NEK8-overexpressing zebrafish embryos. Altogether, our study demonstrates that NEK8 human mutations cause major organ developmental defects due to altered ciliogenesis and cell differentiation/proliferation through deregulation of the Hippo pathway.
| Genes mutated in ciliopathies encode proteins with various localizations and functions at the primary cilium. Here we report novel NEK8 mutations in patients with renal cystic hypodysplasia and associated ciliopathy defects. NEK8 belongs to a protein complex defining the Inversin compartment of the cilium. It is also a negative regulator of the Hippo signaling pathway that controls organ growth. We report genotype-phenotype correlation in the patients. We functionally demonstrate that the two types of mutations (missense versus nonsense) differentially affect ciliogenesis, cell apoptosis and epithelialisation. We also show that all the mutations lead to dysregulation of the Hippo pathway through nuclear YAP imbalance but that the nature of this imbalance is different according to the type of mutation. We confirm alteration of the Hippo pathway associated with Nek8 mutation in vivo in Jck mice. Remarkably, we show that morphogenesis defects observed in Nek8 knockdown epithelial cells or zebrafish embryos are rescued by Verteporfin, a specific inhibitor of YAP transcriptional activity, demonstrating the causative role of YAP dysregulation in the occurrence of these defects. Altogether, this study links NEK8 mutations to dysregulation of the Hippo pathway and provide molecular clues to understand the variability of the multiorgan defects in the patients.
| Ciliopathies are a group of autosomal recessive disorders caused by a dysfunction of the primary cilium. These conditions are multisystemic disorders, affecting left-right symmetry (situs inversus) and various organs such as retina (retinitis pigmentosa, Senior-Løken syndrome), brain (cerebellar vermis aplasia, Joubert syndrome), liver (cysts, intrahepatic biliary fibroadenomatosis), pancreas (cysts) as well as skeleton (cone shape epiphysis, narrow thorax, polydactyly), and/or kidney (renal cystic dysplasia (RCD), nephronophthisis (NPH)) [1, 2]. RCD and NPH are major genetic causes of end stage renal failure in children and perinatal death for RCD. RCD is a kidney developmental defect whose antenatal diagnosis by ultrasound examination reveals hyperechogenic kidneys. Phenotypes range from enlarged cystic dysplastic kidneys to undersized, hypodysplastic kidneys. RCD is usually classified among the spectrum of CAKUT (congenital anomalies of the kidney and urinary tract). NPH is characterized by atrophic kidney tubules with thickened basal membrane, interstitial fibrosis and, at a later stage, the development of cysts at the cortico-medullary junctions. Kidney size can be normal or reduced.
The primary cilium is a microtubule-based antenna-like structure present at the cell surface of almost all vertebrate cells, which controls signaling pathways (Hedgehog, canonical Wnt and planar cell polarity (Wnt/PCP)) with a major role during development and homeostasis of the kidney and other organs. In renal tubular cells, the primary cilium functions as a mechano/chemo-sensor regulating cell cycle and PCP in response to urine flow, in order to control the orientation of the mitotic spindles along the axis of the elongating tubules and the organization of the epithelial cells with respect to their neighbors in the tissue. Defects in these processes result in cyst formation [3]. Ciliopathies are genetically heterogeneous diseases and mutations in >100 genes encoding ciliary proteins have been identified in affected patients [4]. Genotype-phenotype correlation analyses revealed that different mutations in the same gene could result in phenotypes with varying severity. Among these genes, missense mutations in NEK8/NPHP9 have been reported to lead to early onset isolated NPH [5]. However, a homozygous nonsense NEK8 mutation leading to absence of the protein was also identified in a family with three fetuses presenting with a more severe phenotype similar to Ivemark I and II syndromes, characterized by enlarged cystic dysplastic kidneys, pancreas and liver, associated with skeletal abnormalities, asplenia and congenital heart defects [6]. NEK8 is a serine/threonine kinase composed of an N-terminal kinase domain and five C-terminal Regulator of Chromosome Condensation 1 (RCC1) repeat domains that belongs to the family of Never in Mitosis gene A (NIMA) proteins involved in the control of cell cycle progression [7]. In the cilium, NEK8 is located at the “Inversin (INVS) compartment”, a specific subcompartment of the proximal part of the axoneme, distal to the transition zone [8]. The function of this compartment is poorly understood, but human or mouse mutations in genes encoding components of the INVS compartment, INVS/NPHP2, NPHP3 and ANKS6/NPHP16, are known to lead to infantile nephronophthisis with cystic kidneys, congenital heart defects and laterality defects [9–13]. Additionally, NEK8 is also present in the nucleus where it controls the replication fork progression during S-phase and regulates DNA damage response [14].
Recently, NEK8 has been proposed as a regulator of the Hippo signaling pathway [15]. The Hippo pathway regulates organ size by controlling the balance between cell proliferation and cell cycle arrest through the phosphorylation state and nuclear shuttling of transcriptional co-factors YAP/TAZ [16, 17]. Phosphorylation and nuclear shuttling of YAP/TAZ are strictly correlated to cell density, cell polarity and cellular actin cytoskeleton organization [18]. In low cell density conditions, YAP/TAZ are mainly unphosphorylated and able to translocate into the nucleus, resulting in cell proliferation. Conversely, at high cell density, YAP/TAZ are mainly phosphorylated and retained in the cytoplasm, leading to proliferation arrest. NEK8 has been reported to favor TAZ nuclear translocation, a process that is enhanced by NPHP4, encoded by another gene causing NPH, highlighting these two proteins as inhibitors of the Hippo pathway [15, 19].
Here, we report novel NEK8/NPHP9 mutations in five unrelated cases with severe multisystemic phenotypes. This study highlights the dual phenotype associated with the nature of the mutations and the key functions of NEK8 in ciliogenesis and cell proliferation/differentiation through regulation of YAP.
To identify novel mutations responsible for renal ciliopathies, we performed exon-enriched NGS targeting 1,222 genes associated with cilia structure/function, including all genes already known to be associated with ciliopathies (“ciliome sequencing”) [20–22] in two distinct cohorts of affected individuals: 342 patients with isolated or syndromic NPH and 200 fetuses or neonatal death cases with syndromic cystic dysplasia, including Meckel and Ivemark syndromes. Eight novel recessive mutations were identified in NEK8/NPHP9 in five unrelated families with severe overlapping phenotypes (Fig 1A, Table 1).
All five cases presented with kidney involvement associated with extra-renal defects including situs inversus (4 cases), cardiomegaly (3 cases), paucity of bile ducts (3 cases), pancreas defects (3 cases), narrow thorax and short bowed femurs (2 cases), and brain defects such as corpus callosum or vermis agenesis (2 cases) (Table 1). Patients/fetus from families 1, 2 and 3 shared developmental abnormalities including asymmetric renal hypodysplasia with one absent or extremely reduced in size kidney and the other one with major dysplasia, diffuse interstitial fibrosis and incomplete cortico-medullary differentiation, cardiac septal hyperplasia and liver alterations including paucity of bile ducts (Fig 1B and Table 1). The fetus from family 3, diagnosed with Ivemark I syndrome, presented with short pancreas and asplenia in addition to the defects listed above. The patient from family 2 and his affected sibling have already been reported [23]. In contrast, the fetuses from families 4 and 5 exhibited enlarged multicystic kidneys, cystic pancreas and liver, and agenesis of the vermis (fetus 4). The patient from family 1 carried compound heterozygous missense mutations: a paternally-inherited c.259A>G variation in exon 3 leading to a missense alteration in the kinase domain (p.Thr87Ala) and a maternally-inherited c.1804C>T variation in exon 13 resulting in an amino acid change in the RCC1 domain (p.Arg602Trp) (Figs 1A, 1C and S1A, Table 1). The patient from family 2 carried a homozygous missense mutation in exon 13 (c.1738 G>A) leading to an amino acid substitution (p.Gly580Ser) in the RCC1 domain (Figs 1A, 1C and S1A, Table 1). The fetus from family 3 carried a maternally-inherited heterozygous variant of the last base pair of exon 4 (c.618G>A) resulting in the in-frame skipping of exon 4 and loss of 44 amino acids in the kinase domain (p.Val163-Ala206del), associated with a paternally-inherited heterozygous variation (c.1246G>A) leading to a missense mutation (p.Gly416Ser) in the RCC1 domain (Figs 1A, 1C and S1A, Table 1). All four missense mutations were predicted as damaging by PolyPhen-2 and SIFT. The fetus from consanguineous family 4 carried a homozygous mutation affecting the first base of the 5' essential splice site in intron 1 (c.47+1 G>A). This mutation, which is predicted to totally abolish intron 1 splicing, is thus expected to lead to an absence of the protein. Finally, the fetus from family 5 carried two compound heterozygous nonsense mutations (paternally-inherited p.Arg127* and maternally-inherited p.Arg462*) that may result in truncated proteins and/or RNA decay as indicated by NEK8 mRNA quantification (S1B Fig).
We first examined the impact of patient mutations on the targeting of NEK8 to the cilium as well as on ciliogenesis, using patient fibroblasts and mIMCD3 kidney cells.
We obtained primary fibroblasts from skin biopsies of family1 and 5 affected cases F1 II.1 and F5 II.1 (subsequently referred as PT1 and PT5) harboring compound heterozygous mutations p.T87A/p.R602W or p.Arg127*/p.Arg462*, respectively. While NEK8 was located in the proximal part of the ciliary axoneme in control ciliated fibroblasts, it was absent from cilia in PT1 fibroblasts carrying missense mutations (Fig 2A and 2A’), and was instead detected onto a cytoplasmic juxtanuclear vesicular compartment that we identified as the Golgi apparatus (S2A Fig). As an aside, we noted that endogeneous NEK8 or transfected NEK8-GFP proteins were also present at the Golgi in control fibroblasts (S2A and S2B Fig). However, in these cells, NEK8 Golgi localization seemed to be transient as it was clearly observed at low confluence (S2C Fig) and decreased in ciliated cells, suggesting a dynamic localization of NEK8 during ciliogenesis. In PT5 fibroblasts carrying nonsense mutations, no specific NEK8 staining was detected, indicating that the truncating mutations result in the loss of protein expression. Subsequent evaluation of the impact of NEK8 mutations on ciliogenesis showed that the percentage of ciliated cells was significantly decreased in PT1 fibroblasts compared to control cells (50% vs 80%; Fig 2B). In contrast, ciliogenesis was not affected in PT5 fibroblasts (Fig 2B), in agreement with previously reported data showing that renal and MEF cells from Nek8 knockout mice do not show ciliogenesis defects (10,25). In addition, cilia length was more reduced in cells with missense mutations than in those with loss-of-function mutations (Fig 2C).
In order to characterize the specific effect of each NEK8 missense mutation, we generated murine inner medullary collecting duct (mIMCD3) cells depleted for Nek8 expression using a lentiviral vector for shRNA expression. A 80% reduction of the mRNA level was obtained in the Nek8 knock-down cell line (shNEK8) compared to the control cell line transduced with the empty vector (pLKO) leading to a non detectable expression of the protein by immunofluorescence and western blot analyses (S3A, S3B and S3D Fig). The shNEK8 cells were then stably transfected with human NEK8-GFP constructs encoding either the wild-type protein (WT) or the missense variants identified in the patients. In addition, the p.H425Y NEK8 variant, located in the RCC1 domain and previously associated with isolated infantile NPH, was used as a control as it has been reported to alter NEK8 localization at the cilium [5, 24]. The level of expression of human NEK8-GFP fusions was in the same range in all the cell lines (S3C and S3D Fig). The WT NEK8-GFP protein localized to the ciliary axoneme at the “inversin (INVS) compartment” (Fig 2D and 2D'). The proteins with mutations in the RCC1 domains, p.G580S and p.R602W, were no longer present at the cilium and accumulated into the cytosol, similar to the p.H425Y protein [5]. This is in agreement with the major role of the RCC1 domains in cilia localization [24]. In contrast, an axonemal staining, weaker than in control cells, was detected in half of the ciliated cells expressing the NEK8 protein mutated in the kinase domain (p.T87A) (Fig 2D and 2D'). This result is coherent with a previous study showing that some NEK8 mutations in the kinase domain and affecting the kinase activity do not prevent its localization to the cilia in cell culture and in mice [10]. However, the decreased localization of NEK8 T87A-GFP protein points to a role of the kinase domain, possibly independently of the kinase activity, in ciliary NEK8 targeting process.
Then, we evaluated ciliogenesis in the various mIMCD3 cell lines. As observed in PT5 fibroblasts, ciliogenesis was not significantly affected in shNEK8 cells (Fig 2E). However, expression of all but WT and p.H425Y NEK8-GFP proteins led to a reduced percentage of ciliated cells, similar to what we observed in PT1 fibroblasts (p.T87A/p.R602W), highlighting a “gain of function” effect of the missense mutations on this process (Fig 2E).
We next addressed the effect of NEK8 mutations on the ciliary localization of its partner ANKS6 [11]. While ANKS6 was located at the INVS compartment in 58% of control ciliated fibroblasts, its staining at the cilium was completely lost in PT1 and PT5 fibroblasts (Fig 3A and 3A’). As previously reported [10], a dramatic reduction of ANKS6 positive cilia was also observed in shNEK8 mIMCD3 cells (40% compared to 90% in control cells) (Fig 3B and 3B'). This phenotype was rescued by the re-expression of WT NEK8-GFP (80% of ANKS6-positive cilia), whereas none of the mutant forms were able to fully restore ANKS6 localization at the cilium (Fig 3B'). Moreover, in ANKS6 positive cilia of NEK8 mutant expressing cells, ANKS6 localization was restricted to the base of the cilium and did not extend along the INVS compartment as in control or NEK8-WT expressing cells (Fig 3B”). We next investigated the impact of NEK8 mutations on the interaction with ANKS6. Co-immunoprecipitation experiments demonstrated that only the p.T87A mutation led to a significant decrease of ANKS6 binding (Fig 3C and 3C’). This observation is in accordance with the recent report showing that NEK8 interacts directly with ANKS6 via its kinase domain [10, 11]. It is also in agreement with the observed lack of ciliary ANKS6 in shNEK8 cells expressing the p.T87A mutation. Altogether, these data demonstrate that all the NEK8 missense mutations affect both ciliogenesis and biogenesis of the INVS compartment in renal tubular cells and fibroblasts, by preventing correct targeting of NEK8 to cilia and/or binding to or recruitment of ANKS6.
As NEK proteins are involved in cell cycle regulation [7], we sought to examine how NEK8 mutations affect the cell cycle and proliferation. Counting the cell number over 7 days of culture revealed that patient fibroblasts behave differently depending on the type of mutation. While the population of PT1 cells with the missense mutations failed to expand as fast as control cells, PT5 fibroblasts with loss-of-function mutations expand much faster (Fig 4A). In order to characterize this difference, we analysed cell cycle progression of control and patient fibroblasts. At low cell confluence, Ki-67 staining, a marker of G1 to M phase, did not show any difference (Fig 4B). However, analysis of cell cycle by flow cytometry revealed a higher proportion of cells in S-phase in fibroblasts from patients compared to control, at the expense of cells in G0/G1 (S4 Fig). We also analysed Ki-67 staining at high cell confluence, when cells are ciliated. As expected, in this condition, most if not all control cells were Ki-67-negative, likely in quiescent G0 status. In contrast, ~20% of PT1 and PT5 cells remained Ki-67-positive, including ciliated cells (Fig 4B and 4B’; insets), indicating that patient cells fail to enter into G0. This was confirmed by flow cytometry analysis, showing an increased proportion of cells in G2/M for both patient cell lines (S4 Fig).
Altogether, these results indicate a dysregulation of the cell cycle in patient cells, likely resulting in increased cell proliferation. This is in agreement with the data obtained by counting PT5 cell number that increases over time; however, for PT1 fibroblasts, we observed by Annexin-V staining that they underwent more apoptosis than control and PT5 cells after 48 hours of culture (33% vs 15% of control and PT5 cells; Fig 4C and 4C'), explaining why the population PT1 do not expand as well as PT5 fibroblast cells.
NEK8, like other proteins involved in ciliopathies, has recently been described as a regulator of DNA damage response (DDR), with loss of NEK8 dramatically affecting S-phase progression upon DNA stress conditions [14]. Unlike the mentioned report, in non stress conditions we detected a significant increase in the proportion of nuclei positive for γH2AX (phosphorylated form of H2AX), a marker of DNA double-strand breaks [25], in PT1 fibroblasts compared to control and PT5 cells (40% vs 20% positive cells, respectively) (Fig 4D and 4D’). The increase of γH2AX staining was also detected in kidney sections from the same patient, compared to control kidney (Fig 4E). Hence, NEK8 missense mutations cause defects in DNA repair, which may lead to apoptosis during cell proliferation.
In order to study the impact of NEK8 mutations on epithelial organization, we performed 3D matrigel culture on mutated NEK8 re-expressing mIMCD3 cells [26]. As previously reported [14], 20% of the structures formed by shNEK8 cells were abnormal, without a clearly formed lumen (multi-cellular aggregates, tubular structures with excessive branching, malformed spheroids) after 5 days of culture (Fig 5A and 5A’). However, the most striking and previously unreported phenotype consisted of shNEK8 enlarged single lumen spheroids (Fig 5A, 5A’ and 5A”). These enlarged structures correlated with a sustained Ki-67 staining (Fig 5B and 5B’) during spheroid formation, as observed in patient fibroblasts in high cell density. Indeed, while at 2 days of culture the majority of control and shNEK8 spheroids were Ki-67-positive, at 5 days of culture only 10% of the structures remained Ki-67-positive in the control compared to 58% in shNEK8 spheroids. Expression of NEK8-WT protein, as well as p.H425Y form, partially rescued the proportion of normally-sized spheroids (Fig 5A’ and 5A”). In contrast, expression of the three other missense mutated proteins led to an increased proportion of malformed or tubular-shaped structures (p.G580S, p.R602W) (Fig 5A’) or failed to restore the correct size of the spheroids (p.T87A) (Fig 5A”). These results indicate that in vitro the loss of function of NEK8 leads to an overgrowth of the spheroids (enlarged spheroid structures), whereas the RCC1 missense mutations identified in the severe cases (p.G580S and p.R602W) affect epithelial morphogenesis, which may reflect the pathogenic effect of the different types of NEK8 mutations during kidney development.
Since NEK8 has been reported as a regulator of the Hippo pathway, a critical pathway controlling growth and organ size [15], we investigated the impact of NEK8 mutations on the regulation of this pathway. For this, we analyzed YAP localization and phosphorylation in low and high cell density, reflecting proliferating (YAP active) and quiescent (YAP inactive) states. As expected [18, 27], in control fibroblasts, YAP strongly accumulated into the nucleus in the low cell density condition (non ciliated cells), whereas it was hardly detectable in the nucleus when cells reached high cell density and became ciliated (Fig 6A, insets and Fig 6A’). In contrast, both patient cell lines present altered nuclear YAP staining. PT1 fibroblasts showed a weak nuclear YAP staining compared to control cells at low cell density and this expression level was not modulated by confluence (Fig 6A and 6A’). In contrast, a strong nuclear YAP staining was observed in PT5 fibroblasts at low cell density, that decreased at high cell density but remained higher than in control cells (Fig 6A and 6A'). Consequently, whereas the vast majority of control ciliated cells did not show any nuclear YAP staining (Fig 6A), maintenance of nuclear YAP staining occurred in about 40% of ciliated patient fibroblasts (Fig 6A, insets and Fig 6A”). These results indicate that NEK8 mutations lead to an improper regulation of nuclear YAP shuttling from proliferation to quiescent state, which may affect cell growth and differentiation.
Upon cell confluence, inactivation of YAP is mediated by its phosphorylation on Ser127 [17]. As expected, immunofluorescence analysis revealed that cytosolic phospho-YAP increased with cell density in control fibroblasts (S5 Fig). A similar increase was observed in patient fibroblasts, indicating that NEK8 mutations do not affect YAP phosphorylation (S5 Fig). YAP phosphorylation is regulated by the MTS1/2-SAV1 complex, which has also been recently reported to promote ciliogenesis and to localize at the basal body in ciliated cells [28]. We thus carefully examined phospho-YAP localization and detected the presence of phospho-YAP along the axoneme in 80% of cilia in control and PT5 fibroblasts (Fig 6B and 6B’). However, in PT1 fibroblasts, phospho-YAP was absent from cilia and the staining was restricted to the cilium base in half of the ciliated cells (Fig 6B and 6B’). Altogether, these data underline that NEK8 mutations differently impair the cell density regulated nucleocytoplasmic shuttling of YAP, whereas only the missense mutations alter the localization of phospho-YAP at the cilium.
In order to better understand how NEK8 mutations affect YAP nucleocytoplasmic shuttling, we analysed the ability of WT and mutated NEK8-GFP to promote nuclear YAP-myc localisation in co-transfected HEK393 cells. The co-transfection of each NEK8 mutant form with YAP-myc decreased the nuclear translocation of YAP-myc compared to WT NEK8-GFP, confirming the results observed in PT1 fibroblasts (S6A Fig). Moreover, proximity ligation assay performed on cells co-expressing YAP-myc and NEK8-GFP-WT revealed that the two proteins are in close vicinity and are likely to interact at the perinuclear region (S6B Fig), supporting a direct role of NEK8 in YAP regulation.
Then, in order to examine if nuclear YAP imbalance had an impact on target gene regulation, we analysed the expression of YAP target genes as well as the transcriptional YAP co-regulator TEAD4 in control and patient fibroblasts. In control cells, the expression of CYR61, CTGF and TEAD4 was decreased in confluent cells versus non-confluent cells (Fig 6C), thus following the amount of nuclear YAP in these cells, as previously described [29]. In contrast, in PT1cells the expression of these genes was maintained at a similar level in high versus low confluent cells, consistent with maintenance of nuclear YAP localization in confluent cells. In PT5 cells, the expression of YAP target genes also reflected the nuclear YAP localization, with a high level of expression in non-confluent cells that decreased when cells reached confluence, although remaining at a higher level than in control cells.
Among the signalling pathways reported to be downstream of YAP, we examined the Notch pathway, crucial for kidney and liver development [30, 31]. In agreement with the upregulation of this pathway upon cell-cell contact, we observed an overexpression of JAG1 as well as the downstream target HES1 at high cell confluence in both control and PT1 fibroblasts. However, this increase was much higher in patient cells (Fig 6D), indicating that dysregulation of nuclear YAP may also affect Notch signaling.
In order to investigate if NEK8 mutations also led to YAP activation in vivo, we studied Yap and target gene expression in mutant juvenile cystic kidney (Jck) mice which bear a missense p.G448V mutation in the highly conserved RCC1 domain of Nek8 [32, 33]. Immunohistochemistry assay showed that Yap expression was predominantly cytoplasmic in the kidneys of wild-type mice, whereas it was markedly increased in nuclei of kidneys of 5-week old Jck mice, an age at which these mice exhibited a polycystic kidney disease (Fig 7A). Yap staining was particularly intense in the nuclei of tubular epithelial cells lining the cysts (Fig 7A, insets). Western blot analysis of whole-kidney extracts (Fig 7B and 7B') confirmed that the expression of both Yap and phospho-Yap (S127) was increased in mutant mice. Nevertheless, the phospho-Yap (S127) / Yap ratio was decreased, pointing again to an upregulation of Yap activity. In line with these observations, quantitative RT-PCR confirmed that Yap target genes, i.e. Ctgf, Cyr61, Ankrd1 and Birc5 were upregulated in Jck mice (Fig 7C).
To determine if the epithelial morphogenesis abnormalities observed in shNEK8 and mutated NEK8 re-expressing cells were caused by deregulation of the Hippo pathway, we quantified nuclear YAP staining during spheroid formation. Control and shNEK8 cells grown in 3D in matrigel were fixed after 2, 3 and 5 days of culture and stained for YAP (Fig 8A). After 2 days, the majority of control and shNEK8 spheroids were positive for YAP. In control cells, the proportion of YAP-positive spheroids dramatically decreased to 15% at day 5 (Fig 8A and 8A’). However, nuclear staining of YAP was still present in 75% of shNEK8 spheroids after 5 days of culture. The continuous activation of YAP in the nucleus in shNEK8 mIMCD3 cells could thus promote cell growth in forming structures, causing abnormal enlarged spheroids as recently described in MDCK cells [34]. To confirm this hypothesis, we performed size rescue experiments using Verteporfin, an inhibitor of YAP-TEAD4 interaction [35]. Indeed, we observed that 1 μM of Verteporfin caused a reduction in size of the spheroids formed by shNEK8 cells after 5 days (Fig 8B). In parallel, we confirmed that the increased expression of YAP targets observed in patient fibroblasts and shNEK8 IMCD3 cells was reduced with Verteporfin treatment (S7 Fig).
We also investigated if persistence of nuclear YAP in confluent patient fibroblasts was involved in the abnormal activation of the Notch pathway. As shown in Fig 8C, Verteporfin treatment performed in the high cell density condition dramatically reduced JAG1 expression in patient fibroblasts, further demonstrating the link between YAP activation and Notch dysregulation in NEK8 mutant cells.
Finally, we examined the impact of NEK8 mutations identified in patients in zebrafish, an in vivo model relevant for ciliopathies [36]. Embryos injected with nek8 morpholino (MO) displayed the classical ciliopathy-related phenotype including curved body axis, laterality defects and pronephric cysts (Fig 9A and 9A’ and Fig 8A and 8A’), as previously described [37]. Body curvature was partially rescued by co-injection of human WT NEK8-GFP RNA (43% of normal embryos compared to 20% in nek8 morphants; Fig 9A’) but not by mutated NEK8-GFP RNA (p.T87A and p.R602W), thus confirming the pathogenicity of the human missense mutations. Of note, we observed that co-injection of nek8 MO with WT NEK8-GFP RNA led to shortened dorsally curved embryos with occasionally a unique centered eye (Fig 9A and 9A’), a phenotype that was exacerbated by NEK8 missense mutations (60% vs 30%), further indicating their gain of function effect. Overexpression of human NEK8 accounts for the shortened dorsally curved phenotype since it was observed in 40% of embryos injected with WT RNA only (Fig 9A”). We also observed laterality defects (70% of embryos) and pronephros abnormalities (cysts or developmental defects in 50% of embryos) upon human NEK8 overexpression (S8A and S8A’ Fig). As a similar dorsal curvature phenotype has been reported for embryos injected with human YAP RNA [38], we performed rescue experiments using Verteporfin treatment (Fig 9A”, S8 Fig). WT NEK8-GFP RNA-injected embryos were treated with 20 μM Verteporfin from 90% epiboly stage to 34 hours post fertilisation (hpf). Analysis of Yap target gene expression by qPCR revealed that human NEK8 overexpression does induce an upregulation of the target genes, which is blocked by Verteporfin treatment (S8B Fig). Contrary to laterality defects which remained unchanged, the proportion of stunted dorsally curved embryos and pronephros abnormalities decreased by 50% and 25% respectively upon treatment (Fig 9A”, S8A” Fig). These data indicate that the NEK8 overexpression-related phenotype partially results from an upregulation of YAP activity in zebrafish.
Altogether, these data demonstrate that abnormal YAP activation accounts for the epithelialisation, signaling and morphogenesis defects linked to NEK8 mutations.
To date, only two recessive human NEK8 mutations had been reported, one missense mutation in the RCC1 domain in a patient with early onset NPH and one nonsense mutation in the same domain in three fetuses from a consanguineous family with Ivemark I/II syndromes including cystic dysplastic lesions occurring in kidneys, liver and pancreas, and heart and skeletal defects [5, 6]. Here, we describe 8 novel NEK8 mutations in five cases with severe multi-organ developmental defects, and the first association of NEK8 mutations with renal hypodysplasia and agenesis, situs inversus, agenesis of the vermis and bile duct paucity. Based on our results, NEK8 seems to be a major gene for renal dysplasia, since mutations were identified in 5 out of 200 analyzed families with dysplastic kidneys. Conversely, the previously identified mutation in a patient with infantile NPH seems to be a rare event, as we did not identify any other NEK8 mutation among the 342 analyzed NPH families. We also report the first two human mutations in the serine/threonine kinase domain of the protein.
We observed a strong genotype-phenotype correlation. Fetuses with total NEK8 loss-of-function mutations (c.47+1G>A, family 3; p.R127*/p.R462*, family 5) presented enlarged cystic kidneys and pancreas associated with proliferative cystic biliary ducts, characteristics of Renal-Hepatic-Pancreatic Dysplasia syndrome (OMIM #208540), as described for the previously reported fetuses with a nonsense mutation [6]. In contrast, the three patients carrying missense mutations (p.T87A, p.R602W, p.G580S and p.G416S) or in-frame deletion due to a splicing defect (p.V163-A206del) presented with asymmetric dysplasic/hypodysplasic kidneys (agenesis in one case) with loss of differentiation, cortical interstitial fibrosis, dilated tubules and cartilage nodules, associated with paucity of bile ducts (Table 1).
The functional analyses of the NEK8 mutations indicate that loss-of-function and missense mutations differentially alter ciliogenesis, proliferation/apoptosis and epithelial morphogenesis. Indeed missense mutations exacerbate some of the defects due to NEK8 loss of function both in vitro and in vivo (zebrafish), highlighting their likely gain of function effect. In particular, only missense mutations lead to prominent ciliogenesis defects with reduction of percentage of ciliated cells and cilia length in fibroblasts and mIMCD3 cells. While both types of mutations affect cell cycle regulation, missense mutations also alter the function of NEK8 as a regulator of DNA damage response [14], resulting in increased cell apoptosis in fibroblasts and kidney tissue. 3D culture assays showed that shNEK8 mIMCD3 cells (loss of function) form large spheroids (cystic phenotype) compared to NEK8 mutant form re-expressing mIMCD3 cells that mostly fail to get organized into spheroids (dysplastic phenotype). Finally, co-injection of human mutant NEK8 RNA in nek8 zebrafish morphants further enhances their severe morphological alterations, resulting in a shortened dorsally curved body axis. It is noteworthy that the missense mutation previously reported in a patient with infantile NPH (p.H425Y, [5]) does not have the same gain of function effect, thus explaining the less severe phenotype. Therefore, this genotype/phenotype correlation points out the dual function of NEK8 for which a loss of function (nonsense mutations) leads to proliferative/cystic phenotypes and a gain of function (missense mutations) to hypodysplastic phenotypes with loss of differentiation (S9 Fig).
Although Nek8 mouse model phenotypes are different from those of human cases, a genotype-phenotype correlation seems to also exist in rodents in term of severity of the renal phenotypes. The Nek8jck/jck mice, carrying a missense homozygous mutation in the RCC1 domain (p.G448V), which has been demonstrated to be a gain of function, develop enlarged cystic kidneys [32]. In contrast, Nek8 knockout mice (Nek8tm1Bei) present a mild renal phenotype, with dilated proximal tubules and glomerular cysts [39] and the mouse model with a missense mutation (p.I124Y) in the kinase domain (Nek8roc) exhibits hydroureter, cystic tubular dilations and small glomerular cysts [10]. However, both Nek8tm1Bei and Nek8roc mice also present situs inversus and heart defects that lead to death at birth and which may prevent renal cyst formation during the final, post-birth steps of murine nephrogenesis. Association of kidney defects with situs inversus and heart defects was also observed in the five human cases. This phenotype is consistent with a general alteration of the ciliary function of NEK8 and the integrity of the INVS compartment (absence of NEK8 or ANKS6 proteins in the cilia). Indeed, mutations in genes encoding other components of the INVS compartment (INVS/NPHP2, NPHP3 and ANKS6/NPHP16) are known to lead to infantile NPH associated with enlarged cystic kidneys or to kidney cystic dysplasia associated with congenital heart defects and situs inversus [9, 11–13]. Interestingly, we also identified a homozygous frameshift mutation (c.1010_1011del, p.G337Afs*16, family 6) in ANKS6 (S10 Fig) in a fetus whose phenotype was similar to that of NEK8 loss of function cases, i.e. enlarged cystic kidneys associated with enlarged fibrotic pancreas, situs inversus and cardiopathy. This is in agreement with the phenotype of Anks6Streaker mice whose Anks6 mutation (p.M187K) decreases the binding to and activation of Nek8 and leads to cystic kidneys, situs inversus and congenital heart defects, thus mimicking the phenotype of patients with NEK8 mutations [10]. Identification of this ANKS6 mutation, together with our functional data, strengthens the close relationship between NEK8 and ANKS6, i.e. NEK8 recruits its target ANKS6 to the cilium, which in return enhances NEK8 kinase activity [10].
Besides its function at cilia, NEK8 is critical for cell cycle regulation. We demonstrate that NEK8 mutations lead to defective Hippo pathway regulation, with a decreased amount of nuclear YAP in proliferating cells (missense mutations) as well as maintenance of a pool of nuclear YAP in confluent ciliated cells (missense and loss-of-function mutations) in vitro in patient fibroblasts. Such defects were also observed in shNEK8 mIMCD3 spheroids and in vivo in Jck cystic tubular cells. Maintenance of nuclear YAP in patient confluent ciliated cells was accompanied by higher expression of YAP targets compared to controls, sustained Ki-67 staining and abnormal cell cycle with an increased proportion of cells in G2/M. Therefore, NEK8 mutant cells still undergo proliferation and fail to differentiate when reaching confluence. Moreover, we show that cells harboring NEK8 missense mutations are more subject to DNA damage than NEK8 defective cells and undergo apoptosis, which can contribute to the difference in cell growth resulting in hypodysplastic versus enlarged multicystic kidneys respectively (S9 Fig).
Several mechanisms could account for nuclear YAP misregulation. Association of NEK8 with TAZ, the other major Hippo pathway effector, favors TAZ translocation into the nucleus [6, 15]. We show that NEK8 interacts with YAP in a perinuclear region, suggesting a similar regulation process for YAP and TAZ. Moreover, we show that NEK8 missense mutations alter YAP nuclear translocation. As all mutations in both kinase and RCC1 domains affect NEK8 localization into the nucleus in mIMCD3 cells (S3E and S3E’ Fig), NEK8 mutant forms might thus affect YAP shuttling into the nucleus, through their own defective nuclear translocation. NPHP4 has previously been reported to favor NEK8 and TAZ translocation into the nucleus [15]. We can hypothesize that NEK8 mutations affect binding to NPHP4, resulting in a less efficient translocation of NEK8, and consequently YAP into the nucleus. During normal cell differentiation, ciliogenesis and quiescence are accompanied by activation of the Hippo pathway (i.e. YAP inactivation) and proteasome-mediated degradation of cytoplasmic NEK8 [24]. However, the constant level of NEK8 protein expression detected in low and high confluent cells of PT1 with missense mutations suggests that these mutations preserve NEK8 from degradation, resulting in maintenance of nuclear YAP and consequently the lack of proper cell differentiation. This result is in agreement with our in vivo observations showing that NEK8 overexpression in zebrafish embryos mimics the YAP overexpression phenotype [38]. Moreover, the preserved YAP nuclear localization in cells with NEK8 loss-of-function mutations indicates that other proteins help YAP to translocate into the nucleus in proliferative condition, but also control its partial downregulation in confluent cells. This may explain why loss-of-function NEK8 mutations partially preserve nephrogenic differentiation, evidenced by the presence of some mature glomeruli in fetal renal biopsies.
The maintenance of YAP in the nucleus in confluent patient fibroblasts and renal cells in 3D culture may also be associated with defective Hippo pathway activation at the cilium in NEK8 mutant conditions. Recent studies on MST1/2, two major activators of the Hippo pathway, showed that they localize at the basal body and promote ciliogenesis [28]. In this study, we report the presence of phospho-YAP at the cilium and that this localization is partially affected in the presence of NEK8 missense mutations (S9 Fig). Phospho-YAP at the cilium may thus be a key component of activation of Hippo pathway under the control of NEK8. Finally, soft matrix, such as matrigel in 3D culture assays, is known to promote cytoplasmic retention of YAP and TAZ resulting in limited cell growth, via LATS independent non-canonical Hippo pathway activation, involving cytoskeleton, cell junctions, RhoGTPases or GPCR signaling [40]. Maintenance of nuclear YAP in enlarged spheroids in the absence of NEK8 suggests that NEK8 could also regulate YAP through non-canonical mechanisms.
The Hippo pathway is a highly integrative pathway whose regulation is connected to that of many other signalings crucial for organogenesis, including Wnt/β-catenin, TGF-β, BMP and Notch. YAP dysregulation due to NEK8 mutations is thus expected to play a major role in the development of the multisystemic defects presented by the patients/fetuses. Specific inactivation of Yap in the nephrogenic lineage (YapCM-/-) leads to a reduced number of nephrons [41], as seen in the patients with NEK8 missense mutations. We can hypothesize that an abnormal amount of nuclear YAP during kidney development, as shown in patient fibroblasts and Nek8jck/jck mice, would affect the expression of downstream targets that might participate to the pathophysiological processes. CYR61, a gene expressed in tubules and glomeruli in fetal kidney, encodes a CCN protein that interacts with integrins to mediate cell adhesion, migration and differentiation during nephrogenesis. Upregulation of the Notch1 ligand JAG1 may induce a dysregulation of Notch pathway, required during nephron tubular development [42, 43], thus contributing to renal hypodysplasia. Finally, CTGF, encoding a regulator of cartilage morphogenesis and mediator of fibrosis [44, 45] could be involved in the formation of cartilage islets and fibrosis observed on most of the kidney biopsies of NEK8 mutated patients/fetuses. In the liver, YAP is highly expressed in bile ducts and regulates the Notch pathway for ductal specification during development [31]. In the pancreas, YAP-TEAD regulates the transcriptional network controlling pancreatic cell proliferation and differentiation [46]. As for kidney defects, liver and pancreas defects vary according of the type of NEK8 mutation, suggesting that NEK8 loss of function and gain of function differentially affect the hepatic/pancreatic transcriptional program, leading to either proliferation or paucity of bile ducts in the liver, and cysts or fibrosis in the pancreas, respectively. Finally, Hippo pathway dysregulation is associated with heart overgrowth in mice [47], thus highlighting the involvement of this pathway in the pathophysiological mechanisms leading to cardiac defects seen in the NEK8 patients/fetuses.
In conclusion, we demonstrate that NEK8 is a multifunctional protein whose alterations lead to severe developmental abnormalities due to the synergic effect of dysfunction of key processes and signaling pathways. The demonstration of the central role of YAP dysregulation in NEK8 mutant conditions highlights potential therapeutic targets for the patients.
This study was conducted with the approval of the « Comité de Protection des Personnes pour la Recherche Biomédicale Ile de France II ». Approval was obtained under numbers 2007-02-09/DC-2008-229 and 2009-164/DC-2011-1449 (fetuses) and 2008-A01039-46/DC-2008-229 (nephronophthisis patients). For each patient/fetus, written informed consent was obtained from the parents. For studies using animal data: housing and handling of mice were performed in accordance with the guidelines established by the French Council on animal care "Guide for the Care and Use of Laboratory Animals": EEC86/609 Council Directive—Decree 2001–131. The project was approved by the departmental director of "Services Vétérinaires de la Préfecture de Police de Paris" and by the ethical committee of the Paris Descartes University (approval number: A75-15-34).
342 patients with isolated or syndromic NPH and 200 fetuses or early neonatal death cases with syndromic cystic dysplasia, including Meckel and Ivemark syndromes, were studied. Genomic DNA was isolated from peripheral blood or frozen tissues using standard procedures.
Ciliary exome targeted sequencing and bioinformatic filtering was conducted in affected individuals using a custom SureSelect capture kit (Agilent Technologies) targeting 4.5 Mb of 20,168 exons (1221 ciliary candidate genes), including NEK/NPHP9. Briefly, Agilent SureSelect capture libraries were prepared from 3 μg of genomic DNA samples sheared with a Covaris S2 Ultrasonicator according to manufacturer’s instructions. The SOLiD molecular barcodes for traceable ID of samples were added at the end of the capture step. The Ovation Ultralow System (NuGEN Technologies) was used to prepare HiSeq2500 pre-capture barcoded libraries. The ciliome capture by hybridization was performed on a pool of 10 to 16 barcoded precapture libraries. Sequencing performed on SOLiD5500XL (Life Technologies) and HiSeq2500 (Illumina) was done on pools of barcoded ciliome librairies (64 barcoded ciliome libraries per SOLiD FlowChip and 16 ciliome libraries per lane of HiSeq FlowCell). Paired-end reads were generated (75 + 35 base reads for SOLiD, 100 + 100 base reads for HiSeq) and mapped on human genome reference (NCBI build37/hg19 version) using Burrows-Wheeler Aligner (Illumina) or mapread (SoliD). Downstream processing was carried out with the Genome Analysis Toolkit (GATK), SAMtools, and Picard Tools, following documented best practices (http://www.broadinstitute.org/gatk/guide/topic?name=best-practices). All variants were annotated using a software system developed by the Paris Descartes University Bioinformatics platform. The mean depth of coverage obtained was greater than 90x, and more than 89% of the exome was covered at least 15x. Different filters were applied to exclude all variants located in non-exonic regions, pseudogenes, UTRs or known polymorphic variants with a frequency above 1%, i.e. present in databases such as dbSNP, 1000 genome projects and all variants identified by in-house exome sequencing (5150 exomes and 1020 ciliomes). The functional consequence of missense variants was predicted using SIFT (http://sift.jcvi.org/www/SIFT_enst_submit.html) and PolyPhen2 (http://genetics.bwh.harvard.edu/pph2/) softwares.
Control and affected individual fibroblasts were cultured in Opti-MEM supplemented with 10% fetal bovine serum, penicillin, streptomycin, uridine, sodium pyruvate and Ultroser G G serum substitute (Pall Corporation). Control and patient fibroblasts (1.5 × 104 or 2.5× 104 cells respectively) were plated on coverslips and grown for 2 days (low confluence) or 6 days followed by 48-hour serum deprivation (high confluence) before fixation. Murine inner medullary collecting duct (mIMCD3) cells were cultured in DMEM F-12 and HEK293T in DMEM both supplemented with 10% fetal bovine serum, penicillin, streptomycin and L-Glutamine (all from Life Technologies). For immunofluorescence, 2.5 x 104 cells were plated on coverslips and grown for 5 days before fixation. In all the experiments, the level of confluence was visually checked and counted to ensure similarity between control and samples.
mIMCD3 and fibroblasts were fixed in 4% PFA in PBS 1X for 15 min followed by treatment with 50 mM NH4Cl for 15 min. Antibodies used for immunofluorescence were: NEK8 (kind gift of D. Beier [32]), ANKS6 (1:50, Sigma-Aldrich HPA008355), GFP rabbit (1:500, Life Technology A11122), GFP chicken (2B Scientific, 1020), GM130 (1:50, BD 558712), YAP (1:50, Cell Signaling #4911), phospho-YAP (Ser127) (1:50, Cell Signaling #4911), γH2AX (1:500, Millipore 05–636) and anti-acetylated α-tubulin (1:10000, Sigma-Aldrich). Cells were permeabilized with Triton 0.5% for 10 min at room temperature and treated with blocking solution constituted of PBS 1X, 0.1% Tween 20, 3% (for fibroblasts) or 1% (for mIMCD3) BSA before incubating with primary antibodies overnight. Then, cells were washed 3 times with PBS 1X for 10 min and stained with appropriate Alexa Fluor-conjugated secondary antibodies (1:200, Molecular Probes). Nuclei were stained with Hoechst. For Annexin-V assays, cells were first incubated with a cold solution constituted by 10 mM HEPES, 140 mM NaCl and 25 mM CaCl2. Annexin V (Life Technologies) was secondly incubated in the same solution for 30 minutes at room temperature. Fixation was performed with PFA (4%) for 20 minutes and nuclei were stained with Hoechst.
Tissue biopsies embedded in paraffin blocks were sectioned (8 μm section thickness) using a Leica microtome. Next, sections were immersed in xylene baths (5 minutes in the first bath, 5 minutes in the second bath), then rehydrated for 5 minutes in ethanol baths of decreasing concentrations (100%, 95%, 70%, and 40%) and finally immersed in MilliQ water for 5 minutes. Dako target retrieval solution (Dako ref. S1699) was used according to the manufacturer's instructions. The slides were blocked for 45 minutes at 4°C by 10% NDS (Normal Donkey Serum) diluted in PBT (DPBS with 0.1% Triton X100). Fluorescein label Peanut Agglutin (PNA) (1:200, Vector Fl-1071) was used to detect collecting tubules. Other primary and secondary antibodies were used as described above. Slides were mounted in adapted medium, and analysed under an inverted confocal microscope Zeiss LSM 700.
Nek8-knockdown (KD) was performed in mIMCD3 cells by lentiviral infection of a shRNA expressing construct in pLKO puromycin vector (Sigma-Aldrich sh1570), as previously described [39]. Puromycin-resistant Nek8-KD cells were then transfected using Lipofectamine2000 with wild-type and mutated human NEK8-GFP constructs [24] and stable cell lines re-expressing NEK8-GFP were selected with double selection with geneticin and puromycin. NEK8-GFP variants were obtained through site-directed mutagenesis using Pfu turbo kit (Invitrogen). YAP-myc construct has been described in [18].
HEK293 cells were transiently transfected using the calcium phosphate method. After 48 hours, cells were harvested with ice-cold PBS 1X. A small aliquot of this cell suspension was immediately removed and lysed directly in SDS-PAGE sample buffer as a whole cell lysate. The remaining harvested cells were lysed and treated in accordance with the Miltenyi-Biotec beads protocol. Protein dosage was performed using the BCA protein assay kit (Thermo Scientific). Fifty micrograms of proteins were loaded on a 8% acrylamide gel (Bio-rad), and Western blot was conducted using the indicated anti-FLAG M2 (1:1000, Sigma-Aldrich F1804), anti-GFP (1:1000, Roche #1814460001), anti-tubulin (1:10000, Sigma-Aldrich T5168).
For in situ Proximity Ligation Assay (PLA) (OLINK Biosciences, Uppsala Sweden), HEK293 cells were fixed 48h after transfection in 4% PFA for 15 min, permeabilized 10 min with PBS-0.1% Triton before treated with blocking solution, labeled with anti-rabbit GFP and anti-mouse Myc (Thermo Fischer Scientific, #MS139P1) antibodies and then incubated with a pair of nucleotide-labeled secondary antibodies (rabbit PLA probe MINUS and mouse PLA probe PLUS) in hybridization solution. Interactions between the PLA probes, possible when within a distance less than 40 nm, were revealed by adding a ligase and by amplification of a rolling-circle product using labeled oligonucleotides and a polymerase, according to the manufacturer's instructions. Signals indicative of interactions were detected by confocal microscopy as fluorescent dots in visible red.
Total cellular mRNA was isolated using Qiagen Extraction Kit and then treated with DNase I. 1.5 μg of total RNA was reverse-transcribed using Superscript II (Life Technologies). Relative expression levels of genes of interest were determined by real-time RT-PCR using the Absolute SYBR Green ROX Mix (ABgene) and specific primers as follows: human NEK8 forward 5’-GCCTCAAGAGGGCTTTCGA-3’ and reverse 5’-AAGGTGCCACTCATGATCTTCAG-3’; mouse Nek8 forward 5'-GCACCTTGGCCGAGTTCAT-3' and reverse 5'-GCCAGCAGGATCTGCACAA-3'; human CTGF forward 5'-CGAAGCTGACCTGGAAGAGAA-3' and reverse 5'- GTACTCCCAAAATCTCCAAGCCT-3'; human CYR61 forward 5'-GAGTGGGTCTGTGACGAGGAT-3' and reverse 5'-GGTTGTATAGGATGCGAGGCT -3'; human TEAD4 forward 5'-GGACACTACTCTTACCGCATCC-3' and reverse 5'- TCAAAGACATAGGCAATGCACA-3; human JAG1 forward 5'-GCCGAGGTCCTATACGTTGC-3' and reverse 5'-CCGAGTGAGAAGCCTTTTCAA-3'; human HES1 forward 5'-TCAACACGACACCGGATAAAC-3' and reverse 5'-GCCGCGAGCTATCTTTCTTCA-3'. Experiments were repeated at least three times and gene expression levels were normalized to GAPDH.
For qPCR analyses in IMCD3 cells, we used mouse primers described below.
Experiments were performed on 5-week-old female mutant juvenile cystic kidney (Jck) mice bearing a Nek8 mutation (The Jackson Laboratory) and compared to wild-type littermates. Animals were fed ad libitum and housed at constant ambient temperature in a 12/12-hour light/dark cycle.
For mouse samples, 4 μm sections of paraffin-embedded kidneys were submitted to heat-mediated antigen retrieval and incubated with antibody to Yap (Cell Signaling Technology, 4912, 1:100), followed by a donkey anti-rabbit biotinylated antibody (GE Healthcare) at 1:200. Biotinylated antibodies were detected using HRP-labeled streptavidin (Dako) at 1:2000 and 3–3′-diamino-benzidine-tetrahydrochloride (DAB) revelation.
Western blot analyses were performed as previously described [48]. Briefly, protein extracts from kidneys were resolved by SDS-PAGE before being transferred onto the appropriate membrane and incubated with antibodies to phospho-YAP (Ser127) (Cell Signaling Technology, 4911, 1:1000), and YAP (Santa Cruz, sc-101199, 1:1000), Gapdh (Millipore, 1:5000) followed by the appropriate Alexa-conjugated secondary antibody (Life Technologies). Fluorescence was acquired using a ChemiDoc MP Imaging System (Bio-Rad), and densitometry was performed using Image Lab software 5.0.
For real-time RT-PCR, mRNA were extracted from whole kidney samples and Ctgf, Cyr61, Birc5 and Ankrd1 expression were analysed by real-time RT-PCR using CFX96 Touch Real-Time PCR Detection System (Bio-Rad). Primers (Eurogentec) were as follows: Ctgf forward 5’-GCTGACCTGGAGGAAAACATTAA-3’ and reverse 5’-TGACAGGCTTGGCGATTTTAG-3’; Cyr61 forward 5’-CCTTCTCCACTTGACCAGAC-3’ and reverse 5’-ATATTCACAGGGTCTGCCTTCT-3’; Birc5 forward 5’-CCCGATGACAACCCGATAGAG-3’ and reverse 5’-TGACGGGTAGTCTTTGCAGTC-3’; Ankrd1 forward 5’-CTGTGAGGCTGAACCGCTAT-3’ and reverse 5’-CCAGTGCAACACCAGATCCA-3’. Rpl13 was used as the normalization control.
A total of 7.5 x 104 cells/well were plated in triplicate in 6 well plates and grown for 1–7 days. Cells were incubated with complete medium as previously described. The number of cells was counted at the indicated time-points in triplicate.
For flow cytometry analysis, cells were plated at a density of 1 x 105 cells/ml. The cells were pulse-labeled with BrdU for 30 min, washed with PBS, and treated with trypsin. Cells were fixed with ethanol and stained with anti-BrdU-FITC antibody (BD Biosciences) and propidium iodide, according to the manufacturer’s instructions. Flow cytometry analysis was carried out with the BD LSRII flow cytometry system and BD FACSDiva software.
96 well plates were coated with a thin layer of collagen (collagen I, Rat Tail, Corning #354236) that was allowed to polymerize at 37° C for 30 minutes. 4 x 104 cells per well in the appropriate medium (with antibiotics) was mixed with Matrigel (BD) and allowed to polymerize at 37°C for 30 minutes. Subsequently, the appropriate medium was added and changed every 2 days. Samples were fixed after 2, 3 or 5 days of culture. After two washes with PBS 1X, PFA 4% was added for 30 minutes. Antibody stainings were done as previously described; only incubation time with blocking solution was prolonged to 1 hour at room temperature.
Zebrafish were maintained at 28.5°C under standard protocols. Tg(cmlc2:GFP) and Tg(wt1b:GFP) transgenic lines were used to assess heart looping and pronephros morphology, respectively. Control and nek8 (ATG) morpholinos [37] were injected into one-cell stage embryos at 0.4 pmol per embryo. Human full length NEK8-GFP RNA was obtained by in vitro transcription with mMESSAGE mMACHINE kit (Ambion) and injected into one-cell stage embryos at 100 pg per embryo. For Verteporfin treatment, embryos were injected with 100 pg of RNA and GFP-positive animals were selected at shield stage. Embryos were then treated with either DMSO or 20 μM Verteporfin from 90% epiboly stage to 34 or 52 hours post fertilization (hpf), time points at which body curvature and laterality/pronephros phenotypes were measured, respectively. Phenotypes were analysed using a Leica M165FC stereoscope.
For real-time RT-PCR, mRNA was extracted from whole embryos at 34 hpf by TRIZOL and ctgfa, reported to be specific to Yap unlike ctgfb [49], cyr61 and tead4 expression were analysed by real-time RT-PCR using Absolute SYBR Green ROX Mix (ABgene) and specific primers as follows: ctgfa forward 5’-TCCTCACAGAACCGCCACCTTGCCCAT-3’ and reverse 5’-TCACGCCATGTCGCCAACCATCTTCTTGT-3’; cyr61 forward 5’-CCGTGTCCACATGTACATGGG-3’ and reverse 5’-GGTGCATGAAAGAAGCTCGTC-3’; tead4 forward 5’-AAGGAGGACTGAAGGAGCTGTTCGAGAAGG-3’ and reverse 5’-GCCGAATGAGCAGACTTTAGTGGAGGAGGT-3’. gapdh was used as the normalization control.
For treatment of human fibroblasts, Verteporfin (Sigma-Aldrich, SML0534) was added after 5 days of culture when cells had achieved confluence. Several drug concentrations were tested and 0.5–0.075 μM were chosen as optimal non toxical conditions. For treatment of shNEK8 re-expressing NEK8-GFP mIMCD3 cells, we used drug concentration ranging from 0.5 to 4 μM. Verteporfin treatment was also used on control pLKO1 and shNEK8 mIMCD3 cells for rescue experiments in the matrigel 3D assay. In this case, the drug (1 and 2 μM) was added after 2 days of culture and maintained until fixation at 3 or 5 days.
|
10.1371/journal.pmed.1002777 | Octreotide-LAR in later-stage autosomal dominant polycystic kidney disease (ALADIN 2): A randomized, double-blind, placebo-controlled, multicenter trial | Autosomal dominant polycystic kidney disease (ADPKD) is the most frequent genetically determined renal disease. In affected patients, renal function may progressively decline up to end-stage renal disease (ESRD), and approximately 10% of those with ESRD are affected by ADPKD. The somatostatin analog octreotide long-acting release (octreotide-LAR) slows renal function deterioration in patients in early stages of the disease. We evaluated the renoprotective effect of octreotide-LAR in ADPKD patients at high risk of ESRD because of later-stage ADPKD.
We did an internally funded, parallel-group, double-blind, placebo-controlled phase III trial to assess octreotide-LAR in adults with ADPKD with glomerular filtration rate (GFR) 15–40 ml/min/1.73 m2. Participants were randomized to receive 2 intramuscular injections of 20 mg octreotide-LAR (n = 51) or 0.9% sodium chloride solution (placebo; n = 49) every 28 days for 3 years. Central randomization was 1:1 using a computerized list stratified by center and presence or absence of diabetes or proteinuria. Co-primary short- and long-term outcomes were 1-year total kidney volume (TKV) (computed tomography scan) growth and 3-year GFR (iohexol plasma clearance) decline. Analyses were by modified intention-to-treat. Patients were recruited from 4 Italian nephrology units between October 11, 2011, and March 20, 2014, and followed up to April 14, 2017. Baseline characteristics were similar between groups. Compared to placebo, octreotide-LAR reduced median (95% CI) TKV growth from baseline by 96.8 (10.8 to 182.7) ml at 1 year (p = 0.027) and 422.6 (150.3 to 695.0) ml at 3 years (p = 0.002). Reduction in the median (95% CI) rate of GFR decline (0.56 [−0.63 to 1.75] ml/min/1.73 m2 per year) was not significant (p = 0.295). TKV analyses were adjusted for age, sex, and baseline TKV. Over a median (IQR) 36 (24 to 37) months of follow-up, 9 patients on octreotide-LAR and 21 patients on placebo progressed to a doubling of serum creatinine or ESRD (composite endpoint) (hazard ratio [HR] [95% CI] adjusted for age, sex, baseline serum creatinine, and baseline TKV: 0.307 [0.127 to 0.742], p = 0.009). One composite endpoint was prevented for every 4 treated patients. Among 63 patients with chronic kidney disease (CKD) stage 4, 3 on octreotide-LAR and 8 on placebo progressed to ESRD (adjusted HR [95% CI]: 0.121 [0.017 to 0.866], p = 0.036). Three patients on placebo had a serious renal cyst rupture/infection and 1 patient had a serious urinary tract infection/obstruction, versus 1 patient on octreotide-LAR with a serious renal cyst infection. The main study limitation was the small sample size.
In this study we observed that in later-stage ADPKD, octreotide-LAR slowed kidney growth and delayed progression to ESRD, in particular in CKD stage 4.
ClinicalTrials.gov NCT01377246; EudraCT: 2011-000138-12.
| Autosomal dominant polycystic kidney disease is the most frequent genetically determined renal disease and affects approximately 10% of patients on chronic dialysis therapy because of end-stage kidney disease.
The disease is characterized by relentless growth of renal cysts, with consequent disruption of normal parenchyma and progressive reduction of kidney function up to terminal kidney failure.
Previous studies found that octreotide long-acting release (octreotide-LAR), an analog of somatostatin, slowed cyst growth and progressive renal function loss in patients with normal or slightly impaired kidney function.
In this randomized and placebo-controlled trial, we evaluated whether the renoprotective effect of octreotide-LAR could be extended to patients with severe renal disease, that is, with chronic kidney disease stage 3b or 4.
One-hundred patients with estimated glomerular filtration rate 15–40 ml/min/1.73 m2 were randomized to receive two 20-mg intramuscular injections of octreotide-LAR (n = 51) or sodium chloride (placebo; n = 49) every 28 days for 3 years.
We found that 3-year treatment with octreotide-LAR did not appreciably affect glomerular filtration rate decline compared to placebo, but significantly slowed cyst growth and progression to end-stage kidney failure, in particular in patients with more severe renal insufficiency (stage 4) to start with, and was safe and tolerated well.
Our present data—combined with previous evidence of a protective effect against kidney and liver volume growth and renal function loss in patients with normal or moderately reduced kidney function—indicate that octreotide-LAR could be a novel disease-modifying therapy that benefits patients with autosomal dominant polycystic kidney disease, including those with later-stage disease.
Octreotide-LAR is an expensive medication. The identification of a subgroup of patients with CKD stage 4, accounting for approximately 10% to 15% of patients with autosomal dominant polycystic kidney disease, who are at high risk of kidney failure and at the same time may benefit the most from treatment may help increase the cost-effectiveness of octreotide-LAR for the prevention of end-stage renal disease (and related costs of renal replacement therapy and its complications) in this population.
| Every year worldwide, 4.8 to 15.3 per million persons with autosomal dominant polycystic kidney disease (ADPKD) progress to end-stage renal disease (ESRD) [1]. In Europe approximately 10% of all patients undergoing renal replacement therapy have ADPKD [2]. In ADPKD patients, mutations in the genes encoding for either polycystin 1 or polycystin 2 result in polycystin complex dysfunction. This dysfunction results in reduced intracellular calcium concentration, leading to high activity of adenylyl cyclase enzyme and up-regulation of 3′,5′-cyclic adenosine monophosphate (cAMP) levels [3]. In the kidneys, the sustained high intracellular cAMP levels in the proximal and distal nephrons as well as collecting ducts lead to aberrant tubular epithelial cell proliferation and chloride-driven fluid secretion, the 2 key components of the process of cyst formation and growth in ADPKD [4]. Uncontrolled cyst growth results in crowding of adjacent nephrons, destruction of normal renal parenchyma, and, eventually, substantial enlargement of the kidneys and progressive renal failure [5].
Somatostatin, an endogenous cyclic peptide with pleiotropic endocrine, paracrine, and autocrine actions [6], inhibits in vitro adenylyl cyclase and post-cAMP events in shark rectal gland [7]. High-affinity, specific SST2 receptors for somatostatin are expressed in human kidney [8] and co-localize with adenylyl cyclase in the basolateral membrane of renal tubular epithelial cells [9]. This evidence suggested the possibility of targeting SST2 receptors with a somatostatin analog in order to limit cell proliferation and fluid secretion by inhibiting cAMP production in renal cells. In a pilot study, we found that octreotide long-acting release (octreotide-LAR), a synthetic somatostatin analog with longer half-life and higher SST2 affinity than the naïve polypeptide [10], slowed the increase in total kidney, and even liver, volume compared with placebo in 12 patients with ADPKD [11,12]. A subsequent study in a rodent polycystic kidney disease model found that the protective effect of somatostatin analogs against hepatorenal cystogenesis was associated with decreased cAMP production [13].
In the “A Long-Acting somatostatin on DIsease progression in Nephropathy due to autosomal dominant polycystic kidney disease” (ALADIN) trial, we compared octreotide-LAR for 3 years versus placebo in adults with ADPKD with normal kidney function or mild-to-moderate renal insufficiency (estimated glomerular filtration rate [eGFR] ≥ 40 ml/min/1.73 m2). Results from ALADIN showed a significant reduction in kidney growth and cyst growth, and stabilization of glomerular filtration rate (GFR) at 1 year compared with progressive decline in GFR in the placebo group [14]. The ALADIN 2 trial was designed to assess the effect of octreotide-LAR on kidney growth at 1 year and GFR decline at 3 years in patients with ADPKD with more severe renal insufficiency (chronic kidney disease [CKD] stage 3b to 4).
This was an internally funded, prospective, parallel-group, double-blind, placebo-controlled phase III trial aimed at assessing the renal effects of the somatostatin analog octreotide-LAR in adults with later-stage ADPKD. GFR, estimated by the Modification of Diet in Renal Disease Study 4-variable equation (eGFR), at inclusion was 15 to 40 ml/min/1.73 m2. Participants were randomized to receive 2 intramuscular injections of 20 mg octreotide-LAR (n = 51) or 0.9% sodium chloride solution (placebo; n = 49) every 28 days for 3 years. Central randomization was 1:1 using a computerized list stratified by center and presence or absence of diabetes or proteinuria. Co-primary short- and long-term outcomes were 1-year total kidney volume (TKV) growth assessed by computed tomography (CT) scans and 3-year decline of GFR directly measured with the iohexol plasma clearance technique (for further details see S2 Text).
Participants were identified among patients with ADPKD referred to the outpatient clinics of 4 hospitals in Italy coordinated by the Istituto di Ricerche Farmacologiche Mario Negri IRCCS (see S1 Appendix).
Adult (>18 years) men and women with ADPKD according to Ravine criteria [15] and eGFR between 15 and 40 ml/min/1.73 m2 were eligible. We excluded patients with confounding factors that could affect renal function loss independently of kidney growth and treatment allocation (HbA1c > 8%, systolic/diastolic blood pressure > 180/110 mm Hg, urinary protein excretion > 3 g/24 h); patients with abnormal urinalysis suggestive of concomitant, clinically significant glomerular disease; patients with urinary tract lithiasis or infection; patients with symptomatic gallstones, cancer, or major systemic disease; those who were unable to provide informed consent; and pregnant, lactating, or potentially childbearing women without adequate contraception (for further details, please see S1 Protocol and https://clinicaltrials.gov/ct2/show/NCT01377246).
The study protocol was approved by each site’s institutional review board: the Comitato di Bioetica of the Local Health Authority of the Province of Bergamo, the Comitato Bioetico of the Local Health Authority of the Province of Agrigento, the Comitato Etico of the University of Naples Federico II, and the Comitato Etico per la Sperimentazione of the Province of Treviso. The Comitato Etico of the Local Health Authority of Lecce and the Comitato Etico of the Fondazione IRCCS Cà Granda Ospedale Maggiore di Milano also approved the protocol, but the centers of Lecce and Milan did not include patients. Written informed consent was obtained from all participants in compliance with the Declaration of Helsinki. Data were locally recorded in dedicated electronic case report forms and centralized into the database at the coordinating center. This study is reported as per the Consolidated Standards of Reporting Trials (CONSORT) guideline (S1 Checklist).
Eligible participants were stratified for the presence or absence of risk factors that might affect renal function loss (diabetes mellitus and/or 24-hour proteinuria > 1 g). Participants were randomly assigned to treatment groups 1:1 by an independent investigator (G. A. Giuliano see: ALADIN 2 Study Organization in S1 Appendix), using a web-based, computer-generated randomization list created using SAS (version 9.2), stratified by center and the presence or absence of risk factors with a random block size of 4 or 8.
At the baseline evaluation, blood pressure was measured in the dominant arm after a 10-minute rest in the sitting position. The mean of 3 measurements, taken 2 minutes apart, was recorded for statistical analyses. Blood samples were collected in the morning after overnight fasting for routine blood tests including renal and liver function tests, and peripheral blood cell counts. Twenty-four-hour urine collections were sampled for protein, albumin, sodium, creatinine, urea, glucose, phosphorus excretion, and osmolality assessment. Additionally, albumin-to-creatinine ratio was assessed in spot morning urine samples. GFR was centrally measured by iohexol plasma clearance technique [16]. TKV was quantified on CT scans. After baseline evaluation and every 28 days thereafter for 3 years, participants allocated to active treatment received two 20-mg intramuscular injections of octreotide-LAR, whereas those assigned to placebo were given 2 intramuscular injections of 0.9% sodium chloride solution [14]. All injections were administered at the clinic. Any drug administration was registered in patient case record forms for treatment adherence recording.
Vital sign, physical examination, and laboratory variables were assessed every 3 months, together with gallbladder biliary tract and kidney ultrasound assessment. GFR was measured by the iohexol plasma clearance technique every 6 months during the 3-year follow-up. Blood samples for the measurement of iohexol plasma concentration were collected at 120, 180, 240, 300, 360, 420, and 480 minutes after the injection.
CT images were obtained as previously described [17,18] at baseline and the 1-year and 3-year visits the day after GFR measurement, and collected in Digital Imaging and Communications in Medicine (DICOM) format by the coordinating center for central quality control and subsequent analysis. Kidneys were manually outlined by trained operators using the ImageJ polyline method [19], and double-checked by a single operator (AC). All of the operators were blinded to patient treatment allocation (see S1 Text). TKV was obtained as volume of kidney outlines and finally corrected for height (height-adjusted TKV [htTKV], ml/m) [20] to adjust for sex-related volume differences. The use and choice of a contrast agent was left to the radiologist performing the test according to the center’s procedures.
All participants were encouraged to comply with dietary recommendations as per center practice. Adjustments of existing antihypertensive therapy were allowed to optimize blood pressure to a target of ≤130/80 mm Hg throughout the study. Concomitant changes in blood glucose and HbA1c levels and need for hypoglycemic therapy were carefully monitored during the follow-up. Appropriate treatments were allowed to maintain markers of mineral-bone metabolism and acid/base balance in recommended targets. If gallbladder sand or stones were documented during the scheduled serial ultrasound evaluations, treatment with ursodeoxycholic acid was prescribed. No patient received antidiuretic hormone antagonists.
The primary short-term outcome was absolute change in TKV, as measured by CT scan, from baseline to 1-year follow-up. The primary long-term outcome was the chronic rate of GFR decline from 6 months to study end as assessed by serial measurements of iohexol plasma clearance.
Secondary endpoints were the measurement of total liver and liver cyst volumes and a composite endpoint of progression to doubling of serum creatinine (versus baseline) or ESRD at 3-year follow-up. Sensitivity analyses considered 1- and 3-year changes in htTKV. Safety variables included vital signs, clinical laboratory tests, and adverse events.
Statistical analyses of endothelin and MCP-1 urinary excretion mentioned in the protocol were explorative in nature and were not performed because of fund constraints. Also, exploratory analyses of quality of life and societal costs were performed in a subgroup of consenting patients by using the Quality of Life Questionnaire–Version 1 of SF-36, validated in Italy, and the Short-Form Health and Labour Questionnaire. We preferred not to report the above results since in our opinion they were poorly informative. Consistent with the strategy previously adopted for the ALADIN trial [14,21], we decided to report data on liver volumes separately from those on renal outcomes.
Sample size was estimated for the main prespecified outcome, absolute TKV change at 1 year, assuming use of a 2-group t test (2-sided) of the difference between octreotide-LAR and placebo. On the basis of data from the interim results of the ALADIN trial [14], a mean increase of 103.4 ml (SD 149.5) was expected in the placebo group at 1-year follow-up, and octreotide-LAR treatment was predicted to reduce such an increase from 103 ml to 0 ml. Based on these assumptions, and assuming 30% dropout, a sample size of 49 patients per group (total sample size 98) would give the trial 80% power to detect as statistically significant (α = 0.05, 2-tailed test) the expected difference in TKV change between the 2 treatment groups over 1 year.
As for the long-term primary endpoint at 3-year follow-up, assuming a yearly GFR decline (mean ± SD) of 6.31 ± 4.47 ml/min/1.73 m2 in the placebo group (data from ADPKD patients with severe renal insufficiency included in the REIN study [22]), the sample size of 49 patients per group was expected to provide the trial 81% power to detect as statistically significant (α = 0.05, 2-tailed test) a 50% (or larger) reduction in the rate of GFR decline (as observed for chronic GFR slopes in ALADIN [14]) in the octreotide-LAR treatment arm (i.e., from 6.31 to 3.16 ml/min/1.73 m2) compared to the placebo group.
All statistical analyses were done by modified intention-to-treat, using SAS (version 9.4) and Stata (version 12). All adjusted models included age and sex as covariates and 1, or maximum 2, additional baseline covariates [23]. Changes in TKV and htTKV at 1 and 3 years and all other between-group effects were assessed by nonparametric (because of non-normal data distribution) ANCOVA also adjusted for age, sex, and baseline TKV (or htTKV) using the SAS/IML “NParCov3” Macro [24]. GFR decline was assessed with a linear regression analysis and compared between groups with the Wilcoxon rank-sum test. Exploratory linear mixed models using SAS PROC MIXED were also used for TKV, htTKV, and GFR repeated measures, with age, sex, and baseline value as covariates. For the composite endpoint doubling of serum creatinine or dialysis, a Cox regression model was used, also adjusted for age, sex, and baseline serum creatinine and TKV. Between-group differences and their 95% confidence intervals (CIs) for TKV and htTKV were calculated using the SAS/IML “NParCov3” Macro [24]. The between-group difference in median GFR slope and its 95% CI were determined by means of Hodges–Lehmann estimation using the SAS PROC NPAR1WAY.
To assess whether and to what extent the treatment effect was affected by the severity of renal insufficiency, we evaluated in a post hoc, unplanned analysis, not mentioned in the protocol, all study outcomes in the 2 subgroups of patients with eGFR > 29 (range 30–44) or ≤29 (range 15–29) ml/min/1.73 m2 who, according to KDIGO recommendation statements [25], could be classified as patients with moderately to severely decreased eGFR (CKD stage 3b) or severely decreased eGFR (CKD stage 4), respectively.
Data are expressed as mean (SD), median (IQR), or number (%) unless otherwise specified. Percent changes were determined for each participant before calculating descriptive statistics. The normality assumption was assessed by means of the Shapiro–Wilk test. Adjustment for multiplicity was considered for the interim analysis (see S1 Protocol): no further adjustment for multiple testing was performed. All p-values are 2-sided. As planned in the study protocol, an interim analysis was performed by the independent data and safety monitoring board on March 3, 2014, and the board decided to continue the study as per protocol guidelines. This trial is registered with ClinicalTrials.gov (NCT01377246) and EudraCT (2011-000138-12).
Of 104 assessed patients, 3 withdrew consent and 1 had eGFR < 15 ml/min/1.73 m2. Thus, 100 patients were randomized from October 11, 2011, to March 20, 2014 (51 to octreotide-LAR and 49 to placebo), and followed for a median (IQR) of 36 (24 to 37) months (Fig 1). Forty-eight patients allocated to octreotide-LAR and 47 allocated to placebo had evaluable TKV at baseline. After randomization, 4 patients on octreotide-LAR and 2 on placebo withdrew consent, 2 on octreotide-LAR left the study because of adverse events, and 1 on placebo progressed to ESRD. At 1 year, 45 patients on octreotide-LAR and 46 on placebo were available for GFR slope analyses; 37 on octreotide-LAR and 39 on placebo also had CT scan data evaluable for TKV analyses. After the first year, 3 patients on octreotide-LAR and 7 on placebo progressed to ESRD. All patients in the study at 1 year also had GFR slope data evaluable for analyses at the 3-year evaluation. In each group, 35 patients also had CT scan data for TKV analyses (Fig 1). All patients received all planned doses of octreotide-LAR or placebo from randomization to final visit. Thus, compliance to treatment was 100%.
Baseline characteristics—including distribution of imaging classes [26]—of patients randomized to the 2 treatment arms were similar in the study group considered as a whole and in the 37 patients with CKD stage 3b and 63 patients with CKD stage 4 considered separately (Tables 1 and S1; S1 Fig). Independent of treatment allocation, median (IQR) urinary protein excretion was significantly higher in the CKD stage 4 group than in the CKD stage 3b group (355.0 [170.0 to 714.5] versus 180.0 [110.0 to 320.0] mg/24 h, p = 0.008; S2 Table). Concomitant medications were distributed similarly between treatment groups (S3 Table).
Median (IQR) TKV increased less with octreotide-LAR than with placebo at 1 year (135.5 [40.4 to 453.1] versus 257.7 [112.6 to 497.7] ml) and 3 years (604.2 [339.1 to 1,145.1] versus 939.1 [515.5 to 1,318.0] ml). Compared to placebo, octreotide-LAR reduced median (95% CI) TKV growth from baseline by 96.8 (10.8 to 182.7) ml at 1 year (p = 0.027) and 422.6 (150.3 to 695.0) ml at 3 years (p = 0.002) (Fig 2; Table 2). Similar results were obtained for absolute increases in htTKV (Table 2). Median (IQR) percentage increase in TKV was significantly less with octreotide-LAR than placebo at 1 year (5.2% [1.6% to 10.2%] versus 8.8% [5.2% to 13.7%], p = 0.036) and numerically lower at 3 years (29.9% [13.0% to 41.8%] versus 37.1% [23.2% to 54.6%], p = 0.091).
Compared with baseline, measured GFR [16] decreased by 11.3% in the octreotide-LAR group and by 7.0% in the placebo group after 6 months of treatment (Table 3). Thereafter, the reduction in the median (95% CI) rate of GFR decline (0.56 [−0.63 to 1.75] ml/min/1.73 m2 per year) with octreotide-LAR compared to placebo was not significant (p = 0.295) (Table 3 and S2 Fig show individual values for GFR decline over 3 years, GFR reduction from baseline to 6 months, and chronic GFR decline from 6 months to study end).
Sensitivity analyses restricted to the 70 patients without concomitant diabetes and without proteinuria showed that, compared to placebo, octreotide-LAR reduced median (95% CI) TKV growth from baseline by 90.8 (−4.7 to 186.4) ml at 1 year (p = 0.062) and 410.1 (105.8 to 714.3) ml at 3 years (p = 0.008). The difference in the median (95% CI) GFR slope (0.88 [−0.52 to 2.27] ml/min/1.73 m2 per year) between the octreotide-LAR and placebo groups was not significant (p = 0.181) (S4 Table).
During the study, 9 of 51 patients (17.6%) on octreotide-LAR progressed to the composite endpoint of doubling of serum creatinine or ESRD compared to 21 of 49 (42.9%) on placebo (crude hazard ratio [HR] 0.412 [95% CI 0.188 to 0.899], p = 0.026) (Fig 3). Treatment effect was significant even when the analyses were adjusted for age, sex, and baseline serum creatinine and TKV (adjusted HR 0.307 [95% CI 0.127 to 0.742], p = 0.009). Three patients of 51 (5.9%) on octreotide-LAR progressed to ESRD considered as a single endpoint compared to 8 of 49 patients (16.3%) on placebo (crude HR 0.376 [95% CI 0.100 to 1.418], p = 0.149). Four patients (95% CI 1 to 7) needed to be treated to prevent 1 composite endpoint, and 10 (95% CI −2 to 21) to prevent 1 ESRD event considered as single endpoint, within the 3-year analysis period. Analyses that were not prespecified showed that all ESRD events were observed in the subgroup of 63 patients with CKD stage 4. In this subgroup, 6 of the 31 patients on octreotide-LAR (19.4%) progressed to the combined endpoint, compared to 18 of the 32 on placebo (56.3%). The difference was significant (crude HR [95% CI]: 0.341 [0.135 to 0.860], p = 0.023) even after adjusting for age, sex, and baseline serum creatinine and TKV (adjusted HR [95% CI]: 0.199 [0.065 to 0.606], p = 0.005) (Fig 4A). Three patients of 31 on octreotide-LAR (9.7%) progressed to ESRD compared to 8 of 32 (25.0%) on placebo, an effect that was significant after adjusting for age, sex, and baseline serum creatinine and TKV (adjusted HR [95% CI]: 0.121 [0.017 to 0.866], p = 0.036) (Fig 4B).
At baseline and at each study visit up to study end, systolic and diastolic blood pressure (S3 Fig) and HbA1c serum level were similar between groups. The median (IQR) urinary protein excretion rate increased significantly, from 260 (130 to 460) mg/24 h at baseline to 420 (160 to 710) mg/24 h over the whole follow-up period (p < 0.001), in the placebo group, but did not change appreciably in the octreotide-LAR group (S2 Table). Data on eGFR and other clinical and laboratory variables at baseline and follow-up are shown in S5 and S6 Tables. Notably, throughout the whole observation period, urinary osmolality was slightly lower in patients randomized to octreotide-LAR compared to those allocated to placebo (S6 Table). The distribution of blood-pressure-lowering medications and all other considered treatments was similar between groups at baseline and follow-up, with the exception of calcitriol treatment and sodium bicarbonate supplementation, which over the whole follow-up period were more frequent in patients on octreotide-LAR than in those on placebo (S3 Table).
In patients with CKD stage 4, at 3 years of treatment, both TKV and htTKV absolute change from baseline were numerically, though not significantly, lower in the octreotide-LAR than the placebo group. Between baseline and 6 months, median (IQR) measured GFR decreased more with octreotide-LAR than with placebo (−5.6 [−7.7 to −1.7] versus −1.2 [−4.3 to +1.7] ml/min/1.73 m2, p = 0.03), whereas between 6 months and study end, median (IQR) chronic GFR decline tended to be slower with octreotide-LAR than with placebo (Table 3). Individual data are shown in S2 Fig. In the same cohort, urinary protein excretion increased significantly, from 320 (180 to 570) mg/24 h at baseline to 508 (325 to 750) mg/24 h over the whole follow-up period (p = 0.002), in the placebo group, but it did not change appreciably in the octreotide-LAR group (S2 Table).
Overall, 6 of the 37 patients with CKD stage 3b progressed to doubling of serum creatinine: 3 on octreotide-LAR and 3 on placebo. Data on TKV changes at 1 and 3 years and GFR slopes in this subgroup are shown in Tables 2 and 3, respectively.
Twelve of 51 (23.5%) participants in the octreotide-LAR group and 11 of 49 (22.4%) in the placebo group had at least 1 serious adverse event (p = 0.898). Overall, distribution of serious (Table 4) and non-serious (S7 Table) adverse events was similar between groups. However, 2 of 51 patients (3.9%) on octreotide-LAR compared to 9 of 49 (18.4%) on placebo (p = 0.021) had a serious (1 versus 3; Table 4) or non-serious (1 versus 6; S7 Table) renal cyst rupture or infection. These events were considered serious or non-serious according to standard criteria detailed in the study protocol.
Diarrhea, biliary sand, and cholelithiasis were more frequent in the octreotide-LAR group. In this group, diarrhea and other gastrointestinal symptoms recovered spontaneously within the first month of treatment. Biliary sand and cholelithiasis recovered with ursodeoxycholic acid treatment. In addition to renal cyst rupture or infection, other possibly disease-related events including back pain and hepatic cyst rupture appeared to be more frequent in the placebo group (S7 Table).
At 1 and 3 years, body weight and all blood variables were comparable between groups (S6 Table), with the exception of blood glucose concentration, which was higher in the octreotide-LAR group than in the placebo group at both time points. However, new-onset diabetes was not reported in a single patient. Twenty-four-hour urine output and urea, phosphate, and sodium excretion were similar between treatment groups (S6 Table).
No participant required treatment interruption or dose down-titration during the study.
In this study we found that 3-year treatment with octreotide-LAR did not appreciably affect GFR decline compared to placebo in 100 patients with later-stage (CKD stage 3b or 4) ADPKD. Active treatment, however, slowed kidney volume growth and progression to the combined endpoint of doubling of serum creatinine or ESRD, and prevented the urinary protein increase observed in controls randomized to placebo. Octreotide-LAR was well tolerated, and no patient required treatment interruption or even transient dose down-titration during the study. The overall incidence of serious and non-serious adverse events was similar between groups. Our present findings confirm and extend evidence from the ALADIN trial [14] that octreotide-LAR may slow kidney volume growth and renal function loss in ADPKD patients with normal or moderately reduced kidney function. Moreover, our study provides the novel information that a somatostatin analog may slow the progression to a hard clinical endpoint such as ESRD in patients affected by ADPKD. Only one-sixth of patients on octreotide-LAR progressed to the combined endpoint of ESRD or doubling of serum creatinine compared to two-fifths of those on placebo. This finding may have implications for healthcare providers since postponing or even preventing ESRD, in addition to preserving patient quality of life and physical function, also reduces the direct and indirect costs for chronic renal replacement therapy. Notably, only 4 patients needed to be treated to prevent 1 composite endpoint, and 10 to prevent 1 ESRD event considered as a single endpoint, during the 3-year follow-up.
Notably, all ESRD events were observed in patients with CKD stage 4, and the protective effect of octreotide-LAR against progression to the combined endpoint, or to ESRD considered as a single endpoint, was fully driven by treatment effect in this subgroup. In these patients, the reduction in event rates was associated with an acute GFR reduction at 6 months that conceivably reflected amelioration of compensatory glomerular hyperfiltration [11,27–29], a tendency (admittedly non-significant) toward slower chronic GFR decline, and a protective effect against the increase in proteinuria observed on placebo. Thus, in ALADIN 2 patients with CKD stage 4, octreotide-LAR reduced the incidence of ESRD with only marginal effects on chronic GFR decline, an effect that conceivably could be explained by the extremely high number of ESRD events, which increased the power of event-based analyses compared to the power of slope-based analyses. Altogether, these data converge to indicate that even in later pre-terminal stages of ADPKD, when kidney architecture is largely disrupted, octreotide-LAR may still exert a specific and clinically relevant protective effect against progression of the disease.
Another finding that merits further investigation is that nephroprotection appeared to be partially explained by mechanisms—additional to those related to slowed kidney volume growth—similar to those of renin angiotensin system inhibitors, such as amelioration of hyperfiltration [14] and reduction of proteinuria, effects that in this specific context could be mediated by inhibited growth hormone secretion and action [30] and, notably, are not associated with hyperkalemia. As observed in other proteinuric chronic nephropathies [31], these effects may protect residual functioning units from accelerated dysfunction and sclerosis. Thus, based on the above considerations, it is conceivable that proteinuria might be an additional risk factor for disease progression and a specific treatment target for octreotide-LAR in patients with ADPKD and CKD stage 4 [32].
Throughout the whole observation period, urinary osmolality was slightly lower in patients randomized to octreotide-LAR compared to those randomized to placebo. This finding is of potential interest because a retrospective analysis of the Modification of Diet in Renal Disease Study, including 139 patients with ADPKD and chronic kidney disease [33], and of the TEMPO 3:4 trial [34] identified low urinary osmolality as a risk factor for faster renal function loss independent of treatment allocation. Consistently, defective urinary concentration is more evident in patients with larger kidneys [34,35] and appears to worsen in parallel with the progression of cystic lesions and consequent reduction in the interstitial osmotic gradient. This process is associated with peripheral resistance to vasopressin and decreased V2R expression/function in the distal nephron [34,36]. Thus, evidence that ADPKD patients randomized to octreotide-LAR experienced slower kidney growth and delayed progression to ESRD compared to controls, in spite of lower urinary osmolality, further corroborates the working hypothesis that octreotide-LAR may have a renoprotective effect in patients with ADPKD.
Our present findings differ from those of the DIPAK 1 study [37], an open-label randomized clinical trial with blinded endpoint assessment that tested the renal effects of 2.5-year treatment with lanreotide, another somatostatin analog, in 309 patients with ADPKD who had an eGFR of 30 to 60 ml/min/1.73 m2. Unlike ALADIN 2, DIPAK 1 failed to detect any treatment effect on worsening of kidney function, defined as a 30% decrease of eGFR compared to baseline or start of dialysis. However, in ALADIN 2 all ESRD events were observed in patients with CKD stage 4, and the protective effect of octreotide-LAR against progression to ESRD considered as a single endpoint or in combination with doubling of serum creatinine from baseline was fully driven by the treatment effect in this subgroup. Exclusion of patients with CKD stage 4 may explain why only 5 (3 on lanreotide) of the 309 randomized patients (1.6%) progressed to ESRD during the DIPAK 1 study, compared to 11 of 63 patients with CKD stage 4 (17.5%) progressing to ESRD during the ALADIN 2 trial [37]. Thus, unlike ALADIN 2, DIPAK 1 was underpowered to detect a treatment effect on ESRD because of a markedly lower incidence of events in the study population. An additional and plausible, but fully speculative, explanation for the differing results of these two studies could be that, because of amelioration of glomerular hyperfiltration and proteinuria, somatostatin analogs are more renoprotective in patients with later-stage ADPKD than in those with less severe renal dysfunction, who may have less or no hyperfiltration or proteinuria. Alternatively, lanreotide could be just less effective than octreotide-LAR in preventing ADPKD progression to ESRD.
The hypothesis of different drug-specific effects is corroborated by the fact that we did not observe any episodes of hepatic cyst infection in our participants given octreotide-LAR. This is at variance with the increased risk for hepatic cyst infection reported during treatment with lanreotide in the DIPAK 1 study, especially in those with a previous history of hepatic cyst infection [38]. Notably, despite the more advanced stage of disease in ALADIN 2 compared to the ALADIN trial, the safety profile of octreotide-LAR did not differ [14]. Morning fasting blood glucose was significantly higher in the octreotide-LAR than placebo group, but serum HbA1c values were similar between groups throughout the whole study period. Thus, it is conceivable that treatment impaired fasting blood glucose without appreciably affecting average blood glucose levels throughout the day. Consistently, no case of new-onset diabetes was observed in the octreotide-LAR group. As expected, diarrhea was more frequent in the octreotide-LAR group. However, in the affected 15 patients, it recovered spontaneously within 1 month from randomization. Biliary sand or stones were detected by routine ultrasound evaluation in 8 otherwise asymptomatic patients on octreotide-LAR, versus none on placebo, and dissolved in all cases with ursodeoxycholic acid supplementation. Adverse events that were most likely related to the disease, including renal cyst rupture or infection (which was serious in 4 cases), were more frequent in the placebo arm.
Our present data confirm the good safety profile of octreotide-LAR reported in the ALADIN trial [14], in a pilot safety study [11], and in a small pilot trial [39]. However, these findings must be taken with caution since they were obtained by relatively small studies that, combined with the ALADIN 2 trial, included a total of only 131 patients with ADPKD who were exposed to octreotide-LAR for a relative short period, ranging from a minimum of 6 to a maximum of 36 months.
On the other hand, octreotide-LAR has been used for years in thousands of patients for the treatment of acromegaly [40] and neuroendocrine tumors [41], and no major worrisome signal has emerged so far. Independent of the above considerations, however, data from larger series of patients with longer exposure to treatment are needed to better establish the risk/benefit profile of octreotide-LAR in the specific context of ADPKD.
Major strengths of this technically challenging study were the measurement of TKV and GFR by gold standard techniques and the centralized assessment of data by investigators with specific expertise. In particular, the use of CT scans with manual contouring to evaluate TKV has been validated in several studies [11,17,42], and comparative analyses between CT and magnetic resonance (MR) images in ADPKD patients [19] found that kidney volume reproducibility was higher for CT scans than for MR images for all considered methods, likely due to lower image quality on MR images, making kidney identification more operator-dependent. Study findings are unlikely to be explained by unbalanced distribution of risk factors for more severe outcome since baseline characteristics were much the same between groups. Moreover, sensitivity analyses restricted to the 70 patients without potentially confounding factors such as diabetes or proteinuria found that the treatment effect on the primary outcomes was very much the same in this subgroup as in the study group considered as a whole. Consistently, the treatment effect on kidney volume was significant even after adjusting for age, sex, and baseline kidney volume, and that on renal events after adjusting for age, sex, and baseline kidney volume and serum creatinine. Similar findings were observed when kidney volume data were corrected by patient height in order to adjust for the potential confounding effect of sex-related differences in kidney volume. Throughout the whole study period, blood pressure control and the distribution of antihypertensive drugs, including renin angiotensin system inhibitors, diuretics, and lipid-lowering agents, were similar between groups. Moreover, evidence that urinary output and 24-hour urea, phosphate, and sodium urinary excretion were almost the same in the 2 treatment groups reasonably excluded any appreciable role of potential confounding factors such as water, protein, and salt intake. Parametric multiple imputations by chain equations confirmed that study results were robust to missing data. The double-blind design was an additional strength. In particular, the decision to initiate chronic renal replacement therapy was made on the basis of standard clinical criteria by physicians who were blinded to both treatment assignment and GFR measurements [43], which enhanced the robustness of the results and their generalizability to everyday clinical practice. Finally, despite the highly labor-intensive design, and the relatively invasive treatment that required 2 intramuscular injections every 28 days, the study had a high retention rate of enrolled participants and full (100%) adherence to the study interventions.
Our study has a number of limitations. At randomization, GFR, eGFR, osmolality, and urinary protein excretion were slightly different between treatment groups. However, randomization (and even stratification in our study) in a clinical trial does not guarantee that patients allocated to the different treatment groups will be similar with respect to all characteristics evaluated at baseline, with potential differences among groups being attributable to chance [44]. Data on progression to doubling of serum creatinine or ESRD were obtained by analyses of a secondary efficacy outcome and need to be confirmed in larger trials. Furthermore, the data from patients with CKD stage 4 must be interpreted with caution, since they were generated by analyses in 63 patients that were not prespecified. Indeed, the sample size of our study was relatively small, and the possibility of a type I error cannot be definitely excluded. As prespecified in the study protocol, the use of contrast agents during CT scan acquisition to discriminate cyst volumes from intermediate or parenchyma volumes could be avoided when radiologists were concerned by the risk of nephrotoxicity in patients with renal insufficiency. Because of this cautious approach, data were obtained from a too-small number of patients, which did not allow us to perform informative analyses of treatment effect on different kidney compartments. Explorative analyses of endothelin and MCP-1 urinary excretion were not performed because of fund restriction.
Study findings may have implications for healthcare providers. Indeed, octreotide-LAR is an expensive medication. The identification of a subgroup—accounting for approximately 10% to 15% of patients with ADPKD [45]—who are at high risk of ESRD and at the same time may benefit the most from treatment may help increase the cost-effectiveness of octreotide-LAR for the prevention of ESRD (and related treatment costs) in this population. Our data may pave the way to large-scale randomized trials with progression to ESRD as the primary outcome, to definitively demonstrate the nephroprotective effect of octreotide-LAR even in patients with less advanced (CKD stage 3a and 3b) disease. This trial could also secondarily test the treatment effect on concomitant polycystic liver disease [21], cardiac function and morphology [46], and fatal and nonfatal major cardiovascular events.
In conclusion, in this internally funded, parallel-group, double-blind, placebo-controlled phase III trial, we assessed whether the renoprotective effect of the somatostatin analog octreotide-LAR shown in patients with early-stage ADPKD could be extended to patients with CKD stage 3b or 4. We found that 3-year treatment with octreotide-LAR did not appreciably affect GFR decline compared to placebo, although secondary analyses suggest that octreotide-LAR may help to postpone ESRD, particularly in patients with CKD stage 4. Further research, involving larger series of patients with longer exposure to treatment, is needed to investigate these signals further. The results of ALADIN 2 confirm and extend previous evidence showing that for adults with ADPKD octreotide-LAR is safe and may have a protective effect against kidney growth and GFR decline, and could be a novel disease-modifying therapy for patients with later-stage disease.
|
10.1371/journal.ppat.1002242 | Protease-Sensitive Conformers in Broad Spectrum of Distinct PrPSc Structures in Sporadic Creutzfeldt-Jakob Disease Are Indicator of Progression Rate | The origin, range, and structure of prions causing the most common human prion disease, sporadic Creutzfeldt-Jakob disease (sCJD), are largely unknown. To investigate the molecular mechanism responsible for the broad phenotypic variability of sCJD, we analyzed the conformational characteristics of protease-sensitive and protease-resistant fractions of the pathogenic prion protein (PrPSc) using novel conformational methods derived from a conformation-dependent immunoassay (CDI). In 46 brains of patients homozygous for polymorphisms in the PRNP gene and exhibiting either Type 1 or Type 2 western blot pattern of the PrPSc, we identified an extensive array of PrPSc structures that differ in protease sensitivity, display of critical domains, and conformational stability. Surprisingly, in sCJD cases homozygous for methionine or valine at codon 129 of the PRNP gene, the concentration and stability of protease-sensitive conformers of PrPSc correlated with progression rate of the disease. These data indicate that sCJD brains exhibit a wide spectrum of PrPSc structural states, and accordingly argue for a broad spectrum of prion strains coding for different phenotypes. The link between disease duration, levels, and stability of protease-sensitive conformers of PrPSc suggests that these conformers play an important role in the pathogenesis of sCJD.
| Sporadic Creutzfeldt-Jakob disease (sCJD) is the most common human prion disease worldwide. This neurodegenerative disease, which is transmissible and invariably fatal, is characterized by the accumulation of an abnormally folded isoform (PrPSc) of a host-encoded protein (PrPC), predominantly in the brain. Most researchers believe that PrPSc is the infectious agent and five or six subtypes of sCJD have been identified. Whether or not these subtypes represent distinct strains of sCJD prions is debated in the context of the extraordinary variability of sCJD phenotypes, frequent co-occurrence of different PrPSc fragments in the same brain, and the fact that up to 90% of protease-sensitive PrPSc eludes the conventional analysis because it is destroyed by protease treatment. Using novel conformational methods, we identified within each clinical and pathological category an array of PrPSc structures that differ in protease-sensitivity, display of critical domains, and conformational stability. Each of these features offers evidence of a distinct conformation. The link between the rate at which the disease progresses, on the one hand, and the concentration and stability of protease-sensitive conformers of PrPSc on the other, suggests that these conformers play an important role in how the disease originates and progresses.
| Prions cause a group of fatal and rapidly progressing neurodegenerative diseases, originally described as transmissible spongiform encephalopathies (TSEs) [1], [2]. The most common of these diseases is sporadic Creutzfeldt-Jakob disease (sCJD), which accounts for ∼85% of all CJD cases worldwide [3]. Although 40 years ago sCJD was shown to be transmissible to nonhuman primates [4], its pathogenesis remains enigmatic.
Most researchers today believe that all prion diseases are caused by the accumulation of an aberrantly folded isoform, termed PrPSc, of the prion protein PrP [5]. Having a basic amino acid composition and an unstructured N-terminus, PrP can assume at least two conformations: (1) native, α-helix–rich PrPC and (2) disease-causing, β-sheet–rich PrPSc [6]–[8]. The latter represents a misfolded isoform of the normal cellular prion protein PrPC, which is host-encoded by the chromosomal gene PRNP and expressed at different levels in mammalian cells [9]. Yet despite the impressive progress that has been made in understanding the molecular basis of prion diseases, the molecular mechanism of initial misfolding and the high-fidelity replication of the pathogenic conformation of PrPSc in vivo both remain elusive [2], [10]–[12].
Many lines of evidence from experiments with laboratory prion strains support the view that the phenotype of the disease—its distinctive incubation time, clinical features, and brain pathology—is enciphered in the strain-specific conformation of PrPSc [13]–[17]. Although remarkable progress has been made in understanding the structure of laboratory strains of rodent prions [2], [10], [18]–[20], knowledge of the molecular basis of human prion diseases has lagged behind. Researchers generally agree that the genotype at codon 129 of the chromosomal gene PRNP underlies susceptibility to these diseases and, to some degree, their phenotype [21]. However, in contrast to the experiments with laboratory rodent prion strains, in which the digestion of brain PrPSc with proteolytic enzyme proteinase K (PK) consistently results in a single protease-resistant domain with mass ∼19 kDa, the outcome in sCJD is more complex. Distinctive glycosylation patterns and up to four PK-resistant fragments of the pathogenic prion protein (rPrPSc) found in sCJD brains are easily distinguishable on western blot (WB) [14], [21]–[25]. The WB findings together with PRNP gene polymorphism led Parchi, Gambetti, and colleagues to posit a clinicopathological classification of sCJD into five or six subtypes; notably, the WB characteristics of PrPSc breed true upon transmission to susceptible transgenic mice [14], [21], [22]. An alternative classification of the PrPSc types and their pairing with CJD phenotypes has been proposed by Collinge and collaborators [23], [24], [26], [27]. This classification differs from the previous one in two major aspects: First, it recognizes three (not two) PrPSc electrophoretic mobilities; and second, it identifies also PrPSc isoforms with different ratios of the three PrP glycoforms [26]. Although the disease phenotypes of patients with sCJD are remarkably heterogeneous, 21 kDa fragments of unglycosylated PrPSc (Type 1) frequently differ from the phenotypes associated with the 19 kDa fragments of unglycosylated PrPSc (Type 2) [14], [21], [22], [28].
Cumulatively these findings argue that the PrPSc type represents yet an additional major modifier in human prion diseases; accordingly, WB-based clinicopathologic classifications became an important tool in studies of prion pathogenesis in human brains and in transgenic mice models [14], [26]. Now, inasmuch as two distinct PK cleavage sites in PrPSc Types 1 and 2 most likely stem from distinct conformations, some investigators contend that PrPSc Types 1 and 2 code distinct prion strains [14], [23], [28], [29]. However, the heterogeneity of sCJD, along with a growing number of studies including bioassays, all suggest that the range of prions causing sCJD exceeds the number of categories recognized within the current WB-based clinicopathologic schemes [30]–[32]. Additionally, recent findings revealed the co-occurrence of PrPSc Types 1 and 2 in up to 44% of sCJD cases and thus created a conundrum [33]–[38]. Finally, up to 90% of brain PrPSc in sCJD eludes WB analysis because it is destroyed by proteinase-K treatment, which is necessary to eliminate PrPC. Consequently, the conformation or role of this major protease-sensitive (s) fraction of PrPSc in the pathogenesis of the disease is a subject of speculation [30], [39], [40].
Aiming to advance our understanding of the molecular pathogenesis of human prion diseases, we used the conformation-dependent immunoassay (CDI) [15], [30], [41] to determine the conformational range and strain-dependent molecular features of sCJD PrPSc in patients who were homozygous for codon 129 of the PRNP gene. Even relatively minute variations in a soluble protein structure can be determined by measuring conformational stability in a denaturant such as Gdn HCl [42]. Utilizing this concept, we designed a procedure in which PrPSc is first exposed to denaturant Gdn HCl and then exposed to europium-labeled mAb against the epitopes hidden in the native conformation [15]. As the concentration of Gdn HCl increases, PrPSc dissociates and unfolds from native β-sheet-structured aggregates; and more epitopes become available to antibody binding. These experiments involve insoluble oligomeric forms of PrPSc, and denaturation of this protein is irreversible in vitro; consequently the Gibbs free energy change (ΔG) of PrPSc cannot be calculated [43]. Therefore we chose instead to use the Gdn HCl value found at the half-maximal denaturation ([GdnHCl]1/2) as a measure of the relative conformational stability of PrPSc. The differences in stability reveal evidence of distinct conformations of PrPSc [15], [42], [43]. Because CDI is not dependent on protease treatment, it allowed us to address fundamental questions concerning the concentration and conformation of different isoforms of sCJD PrPSc, including protease-sensitive (s) and protease-resistant (r) PrPSc. We found a broad spectrum of structures that are likely responsible for the phenotypic heterogeneity of sCJD and we identified the structural characteristics of PrPSc that are linked to the duration of the disease.
From 340 patients with an unequivocally definite diagnosis of Type 1 or Type 2 sCJD and who were homozygous for codon 129 polymorphism in the PRNP gene, we selected samples from 46 patients. The descriptive statistics and Kaplan-Meier survival curves indicate that these cases are representative of the whole group collected at NPDPSC and are similar to those previously reported by us and others (Compare Figure 1 and Figure S1, Table 1) [21], [38], [44]. As expected, we did not observe statistically significant differences in sex ratio or age at onset of the disease [21], [44]. Kaplan-Meier analyses of survival (Figure 1) demonstrated that patients with PrPSc Type 1 had a significantly shorter disease duration than patients with PrPSc Type 2 (P = 0.002) despite identical codon 129 MM polymorphism, age, and sex distribution (Table 1). Moreover, there is an apparent tendency toward longer survival of patients with Type 2 rPrPSc(129 V) than patients with Type 1 rPrPSc(129 M) (P = 0.017). The difference in survival between patients with Type 2 rPrPSc(129 V) and Type 2 rPrPSc(129 M) was also significant (P = 0.008) with shorter survival of those homozygous for valine (Figure 1).
To ensure that the brain homogenate analyzed by CDI contained only Type 1 or 2 rPrPSc, each brain homogenate underwent a second WB (Figure S2). The results confirmed the original diagnostic classification but we found two atypical patterns: Case #833 (Type 2 PrPSc(129 M) and Case #162 (Type 2 PrPSc(129 V) revealed, in addition to a band of unglycosylated rPrPSc with apparent molecular mass ∼19 kDa, a second band with electrophoretic mobility corresponding to mass ∼17 kDa. The observation of different glycoform patterns of PrPSc in different sCJD cases before protease K treatment and distinct resistance to proteolytic degradation of different glycoforms of PrPSc is interesting and deserves further investigation.
To measure the concentration of different forms of PrPSc in the frontal cortex, we used europium-labeled mAb 3F4 [45] for detection and 8H4 mAb (epitope residues 175–185) [46] to capture human PrPSc in a sandwich CDI format (Figure S4) [30], [47]. The analytical sensitivity and specificity of the optimized CDI for detection of both protease-sensitive (s) and protease-resistant (r) conformers of PrPSc was previously reported by us and others in numerous publications [15], [30], [41], [48]–[50] and has been shown to be as low as ∼500 fg (∼20 attomoles) of PrPSc. This sensitivity of CDI is similar to the sensitivity of human prion bioassay in Tg(MHu2M)5378/Prnp0/0 mice [30].
First, we determined the concentration of disease-causing PrPSc in subpopulations of sporadic sCJD patients (Table 1 and Figure 2). We observed wide interindividual variations, and approximately sixfold more accumulated PrPSc in the frontal cortex of patients with Type 2 PrPSc(129 M) than those with Type 1 PrPSc(129 M) or Type 2 PrPSc(129 V). A large portion of PrPSc in all groups is protease-sensitive, constituting a pool of sPrPSc conformers (Table 1 and Figure 3a). The digestion with proteinase K (PK) was performed with 3 IU/ml (100 µg/ml) of 10% brain homogenate containing 1% sarkosyl for one hour at 37°C. The protocol for PrPSc digestion, validated in previously published experiments, was selected according to the following criteria: 1) complete digestion of PrPC determined with CDI in control samples; 2) complete shift of the bands of PrPSc to PrP 27–30 on WBs; 3) unequivocal WB differentiation of Type 1 and Type 2 rPrPSc in all tested samples [15], [30], [38], [40], [47], [51]. Additionally, the complete digestion of the PrPSc N-terminus with PK was monitored on WBs in all samples (Figure S2).
In patients with Type 2 PrPSc(129 M), significantly higher concentrations of total PrPSc and sPrPSc protein (Table 1) are associated with extended duration of disease. However, the concentration of sPrPSc vary greatly between individual patients, with numerous overlapping values between each classification group (Figure 3a). Thus, when the concentration of sPrPSc is expressed as a percentage of total PrPSc, no significant difference between groups appears, and the proportion of sPrPSc varies from 5% to 90% in individual patients (Figure 3b). We concluded from these observations that a major portion of pathogenic sCJD PrPSc is protease-sensitive and that the highest levels of sPrPSc are present in Type 2 PrPSc(129 M). The observed large interindividual differences in PK sensitivity likely indicate a broad range of PrPSc conformers within each PRNP genotype and WB pattern [15], [39]. Since the proteolytic sensitivity of PrPSc is considered a reliable and constant marker of a distinct prion strain, the data support the conclusion that distinct prion structures are present within each classification group.
The partial exposure of epitopes 108–112 and 175–185 in native pathogenic PrPSc reflects differences in the conformation of native PrPSc [15], [52]. When we adopted this approach previously, we found considerable differences among eight laboratory prion strains passaged in Syrian hamsters [15]. The denatured state is a reference corresponding to the concentration of PrPSc; the ratio between the fluorescence signal of europium-labeled mAb 3F4 reacting with PrPSc in the native (N) or completely denatured (D) state represents a relative measure of the degree of exposure of these epitopes.
The highest D/N PrPSc ratio was found in patients with Type 2 PrPSc(129 M); and despite a large spread of values, the difference is statistically significant (P = 0.002) (Figure 4). PK treatment eliminated most of the exposed 108–112 and 175–185 epitopes in patients with Type 1 PrPSc(129 M) and in patients with Type 2 PrPSc(129 V), resulting in the increased D/N ratios (Figure 4). The opposite trend was observed in patients with Type 2 PrPSc(129 M). After PK treatment the PK-induced differences among the three cohorts proved statistically significant to a remarkable degree (P<0.001). Large variations in D/N values exceed what we expect from our experiments with laboratory prion strains [15] and suggest that a high degree of conformational heterogeneity exists in PrPSc aggregates. Protease treatment change the ratio in all groups and reduced the heterogeneity in MM2 sCJD, and as a result, each group could be reliably differentiated. The increased frequency of exposed epitopes in codon 129 MM samples with Type 2 rPrPSc after PK treatment is unexpected and may indicate one of three possibilities: that the ligand protecting the 3F4 epitope was removed by PK treatment; that epitope 108–112 was protected by the N-terminus of PrPSc; or that conformational transition resulted in more exposed 108–112 epitopes. Whether the epitopes hindrance in undigested PrPSc is the result of lipid, glycosaminoglycan, nucleic acid, or protein binding to the conformers unique to the MM2 sCJDF PrPSc remains to be established.
First, we asked whether the PTA precipitation has an impact on the stability of PrPSc. This step in the protocol was important for eliminating high concentrations of PrPC and for concentrating PrPSc in brain samples with relatively low levels of PrPSc. (Figure S5). The denaturation curves performed on 5% brain homogenate before PTA precipitation, on PTA pellet and on PTA pellet washed with an excess of H2O, were superimposable, an effect which indicated that PTA quantitatively concentrated all PrPSc conformers and did not influence the stability in CDI. This conclusion accords with numerous previously published data, including bioassays, which indicate that PTA dose not precipitate PrPC and recovers specifically ≥95% of infectious PrPSc in the pellet, regardless of protease sensitivity or prion strain [15], [30], [53]–[55]. The error of the method does not exceed 5% in monitoring [Gdn HCl]1/2 values in the same repeatedly measured brain samples (Figure S5 and Figure S6).
Since the dissociation and unfolding of oligomeric PrPSc may be dependent on protein concentration [42], we first followed the process with CDI at different dilutions of PrPSc (Figure S5). The resulting overlapping dissociation/unfolding curves of PrPSc with variation in Gdn HCl1/2 values <3% indicate that in the 10–250 ng range, the dissociation/unfolding is independent of concentration and is highly reproducible. Furthermore, to ensure the same conditions in all dissociation/unfolding experiments, the PrPSc content in all samples was maintained at a constant 50 ng/ml concentration. As we observed previously with the western blot technique, the Gdn HCl1/2 values obtained with frontal, temporal, parietal, and occipital cortex, thalamus, and cerebellum in three typical sCJD cases were superimposable, indicating that the same conformers of PrPSc are present in different anatomical areas (data not shown) [38].
Next we examined the frontal cortex of individual sCJD patients homozygous for methionine or valine at codon 129 of the PRNP gene. Typical examples of dissociation/unfolding curves are shown in Figures 5a, 5b, and 5c. Comparing all sCJD cases, we found a broad range of Gdn HCl1/2 values ranging from 1.3 to 3.5 M (Figure 6a). Because of the wide spread of values, the difference between the cases with Type 1 and 2 PrPSc(129 M) is only marginally significant (P = 0.040) and there is no statistically significant difference among other groups. The possible cluster of Gdn HCl1/2 values at ∼3.0 M is discernible in cases with Type 1 PrPSc(129 M) (Figure 6a). We concluded from these experiments that PrPSc proteins in different brains of sCJD patients display a vast range of unique conformations within each classification group.
We next investigated the conformational impact of the proteolytic digestion of sPrPSc conformers and the loss of N-terminal residues in rPrPSc. The proteolysis of PrPSc with PK resulted in increased conformational stability in Type 1 rPrPSc(129 M) and Type 2 rPrPSc(129 V) but did not significantly reduce the range of values (Figure 6a). In contrast, PK treatment uniformly decreased Gdn HCl1/2 values in Type 2 rPrPSc(129 M) (Figure 6a). The marked drop in this group's stability is statistically significant to a high degree (P<0.001). Additionally, there is a discernible cluster of Type 2 PrPSc(129 M) cases at ∼2.6 M (Figure 6a). We interpret the data as providing evidence of a wide range of unique conformations in each subgroup. Proteolytic treatment selects the conformers having a more stable core in Type 1 rPrPSc(129 M) and Type 2PrPSc(129 V). The opposite effect of PK, as well as decreased stability, was observed in samples with Type 2 PrPSc(129 M). These data suggest that PK treatment generates a unique set of conformers in Type 2PrPSc(129 M), characterized by increased exposure of 108–112 and 175–185 epitopes (Figure 5) and, upon PK treatment, decreased stability of the core rPrPSc(129 M).
To investigate the conformational stability of sPrPSc separately from rPrPSc, we subtracted the relative fractional change in stability of rPrPSc after PK treatment from the PrPSc values obtained before PK (Figures 5a, 5b, and 5c). The resulting differential curves exhibit Gaussian distribution with the peak at the median stability of sPrPSc; the height and integrated peak area is proportional to the relative fraction of PK-digested conformers. Overall stability of Type 1 sPrPSc is, as expected, lower than that of rPrPSc and we estimate, from these data alone, that sPrPSc conformers constitute 13–72% of the PrPSc (Figure 6a). A larger spread of positive values obtained with Type 2 sPrPSc(129 V) coincides with a generally larger spread of Gdn HCl1/2 values in this group. In contrast, the negative differential curves for Type 2 sPrPSc(129 M) indicate that sPrPSc is more stable than rPrPSc in this patient group (Figure 6b). Notably, the only positive value in this group came from a sample having an atypical 19 and 17 kDa doublet of unglycosylated rPrPSc on WBs (Figure 6b and Figure S2). Since the stability of sPrPSc and of rPrPSc reflect different initial conformation, the observed spread of values suggests a broad range of unique PrPSc conformers within each PRNP genotype and WB pattern [15], [39], [56].
To determine whether unifying trends exist, we examined which PrPSc characteristics have an impact on duration of the disease in individual patients in all groups using regression analysis. In contrast to analysis of variance (Anova) used to compare MM1, MM2, and VV2 groups (Table 1), the regression analysis is testing the relationship between a dependent variable (duration of the disease) and independent variables (e.g., sPrPSc levels) in individual patients. From concentrations of PrPSc, sPrPSc, and rPrPSc, only the levels of sPrPSc (Figure S7a) correlated significantly with longer duration of the disease. The overall dependency is driven mainly by the higher levels of sPrPSc in Type 2 sPrPSc(129 M) and longer duration of the disease in this subgroup (Table 1). Additionally, the measurement of absolute concentration of sPrPSc is clearly a better indicator of this relationship than the estimate of the relative fraction (percentage) of sPrPSc (Figure 3b). Despite a wide spread of values, this observation corroborates the conclusion, drawn from previous experiments with eight laboratory strains of prion, that incubation time, and by extension duration of the disease, is linked to the higher levels of sPrPSc [15].
We then analyzed the conformational characteristics of PrPSc. The stability of rPrPSc clearly did not correlate with duration of the disease in individual cases (Figure 7a). In contrast, the change in the stability of PrPSc upon PK treatment (Figure S7b) or relative levels of sPrPSc conformers eliminated by PK (Figure 7b) expressed as a fraction of all conformers, both demonstrated better correlation with duration of the disease than did any other parameter in both Type 1 and Type 2 cases. In contrast to simple measurement of sPrPSc concentration, the stability assay performed before and after PK treatment cumulatively determines the shift in the stability of PrPSc, change in the slope of the denaturation curve (dissociation/unfolding rate), and relative levels of the sPrPSc conformers in the total PrPSc pool. This effect leads to the clear separation of Type 1 from Type 2 sPrPSc(129 M) cases (Figure 7b). We interpret these findings as evidence of the differential impact of protease treatment on different conformers, resulting in either increased or decreased stability of the remaining rPrPSc core (PrP 27–30). Taken together, higher levels of more stable sPrPSc conformers are associated with extended duration of the disease. Conversely, lower concentrations of unstable sPrPSc correlate with faster progression of the disease.
The discovery of heritable polymorphic PK cleavage sites and glycosylation patterns in PrPSc have been used for the initial diagnostic classifications of sCJD cases. In concert with the codon 129 PRNP haplotype, the different rPrPSc types broadly correlate with distinct disease phenotypes [14], [21], [27]–[29], [57]. The majority of sCJD patients are homozygous for methionine at codon 129 of the PRNP gene; they also accumulate Type 1 rPrPSc and present with so-called classic sCJD, characterized by rapidly progressive dementia, early myoclonus, visual disturbances including cortical blindness, disease duration of approximately 4 months, and fine punctate (synaptic) deposits of PrPSc [21], [30]. In contrast, patients with the second most frequent phenotype are homozygous for valine at codon 129 of the PRNP gene, accumulate Type 2 PrPSc and manifest a different disease course, with early ataxia, predominant extra-pyramidal symptoms, relatively late-onset dementia in the extended course of the disease, and large plaque-like deposits of PrPSc [21].
In the increasing number of subsequent sCJD cases which were examined with more sensitive and specific techniques, investigators began to recognize the extensive variability of the sCJD phenotypes, as well as the extreme complexity of brain immunohistochemistry and western blot patterns of PrPSc [25], [32], [37], [38], [57]–[59]. Although the western blot systems provided early evidence that molecular characteristics of PrPSc are transmissible, evidence regarding the original conformation of PrPSc remains indirect and limited to the most protease-resistant fractions. Because variable fractions of PrPSc are protease-sensitive, we decided to determine the conformational characteristics directly, by using CDI. This method allowed us to compare the conformational features of human PrPSc independently of proteolytic treatment and in addition provided quantitative data on levels of PrPSc, sPrPSc, and rPrPSc [15], [30]. The CDI techniques represent a major improvement over previously used semi-quantitative WB-based methods, the finding that has been independently confirmed by another group [60], [61]. The dissociation and unfolding of PrPSc in a presence of increasing concentration of Gdn HCl can be described as follows: [PrPSc]n→[sPrPSc]n→iPrP→uPrP, where [PrPSc]n are native aggregates of PrPSc, [sPrPSc]n are soluble protease-sensitive oligomers of PrPSc, iPrP is an intermedite, and uPrP is completely unfolded (denatured) PrP [7], [43], [62]. The CDI monitors the global transition from native aggregates to fully denatured monomers of PrPSc. In contrast, the WB based techniques monitor either the partial solubilization of PrPSc [63] or conversion of rPrPSc to protease-sensitive conformers [16] after exposure to denaturant. As a result, the stability data on soluble protease sensitive oligomers and intermediates of PrPs cannot be obtained with WB techniques and lead to the markedly underestimated values [60].
The sixfold difference in concentrations of PrPSc between Type 1 and Type 2 PrPSc(129 M) (Figure 2) revealed in the frontal cortex by means of CDI was surprising, even though some variability was to be expected due to differences in the predominantly affected areas in distinct sCJD phenotypes [30]. The average levels of PrPSc are up to 100-fold lower than those in standard laboratory prion models such as Syrian hamsters infected with Sc237 prions [15]; and together with the up to 100-fold variability within each phenotypic group, these lower levels of PrPSc may partially explain why some sCJD cases are difficult to transmit, and why lower endpoint titers are obtained with human prions in transgenic mice expressing human PrPC [14], [26], [30], [64].
As we observed previously, up to 90% of the pathogenic prion protein was protease-sensitive [30]. In this study, we found the highest concentrations in Type 2 PrPSc(129 M). The broad range of absolute and relative levels of rPrPSc and sPrPSc offers evidence of a broad spectrum of PrPSc molecules differing in protease sensitivity in each group with an identical polymorphism at codon 129 of the PRNP gene and an identical WB pattern (Figure 3). Moreover, these findings signal the existence of a variety of sCJD PrPSc conformers; and since protease sensitivity is one of the characteristics of prion strains, they also suggest that distinct sCJD prion strains exist [15], [30], [31], [58], [62], [65].
The CJD cases studied in this paper represent 75–90% of all clinical and pathologic diagnostic categories of sCJD [21]. In order to allow unequivocal interpretation of the CDI data, we had to exclude sCJD patients heterozygous for codon 129 polymorphism in the PRNP gene, even though they represent ∼15–20% of sCJD cases. The CDI cannot differentiate PrPSc with codon 129 M from V in a mixture which is present in sCJD heterozygots, and therefore we were unable to differentiate the conformational impact of codon 129 polymorphism. We also excluded the VV1 type of sCJD because of its rarity. This rare form of sCJD constitutes ∼1% of all sCJD cases and we did not collect enough cases to allow statistical comparison with the other groups [21].
The heterogeneity of PrPSc conformations found with CDI within sCJD patients homozygous for codon 129 plymorphism of the PRNP gene is remarkable (Table 1 and Figure 6), with a range corresponding to that of stabilities found in more than ∼30 distinct strains of de-novo and natural laboratory rodent prions studied up to now [15], [16], [66]. The high sensitivity and reproducibility of CDI, together with broad inter-individual variability detected with techniques based on three different principles—PK sensitivity, epitope exposure, and conformational stability—all indicate that the intragroup variations did not originate in the CDI technique but rather reflect differences in the structure of PrPSc in different patients. The intriguing effect of PK treatment on the stability of Type 2 PrPSc(129 M) suggests that the protease-resistant core of Type 2 was profoundly destabilized. Since sCJD cases with Type 2 PrPSc(129 M) have remarkably extended disease durations, the molecular mechanism underlying this effect calls for detailed investigation.
Several theories have been proposed to explain the origin of sCJD. One argues for spontaneous somatic mutations in PRNP; another, for rare stochastic conformational changes in PrPC [26], [67]. Yet a third hypothesis holds that low levels of PrPSc are normally present and cleared, but rise to pathogenic levels when the clearance mechanism fails [40]. Cumulatively, our findings indicate that sCJD PrPSc exhibit extensive conformational heterogeneity. Whether this heterogeneity originates in a stochastic misfolding process that generates many distinct self-replicating conformations [26], [67] or in a complex process of evolutionary selection during development of the disease [17] remains to be established.
We discovered this fraction of PrPSc while developing a conformation-dependent immunoassay (CDI), which does not require proteolytic degradation of ubiquitous PrPC [15]. Although the original definition of sPrPSc was only operational, considerable additional data demonstrate that (1) sPrPSc replicates in vivo and in vitro as an invariant and major fraction of PrPSc; (2) sPrPSc separates from rPrPSc in high speed centrifugation; and (3) the proteolytic sensitivity of PrPSc can reliably differentiate various prion strains [15], [30], [31], [58], [62], [65]. Accumulation of sPrPSc precedes protease-resistant product (rPrPSc) in prion infection [40], [68]; and up to 90% of PrPSc accumulating in CJD brains consists of sPrPSc [30]. Thus, the detection by CDI of sPrPSc as a disease-specific marker is widely regarded as a more reliable basis for diagnosing prion diseases. This improved detection led to the discovery of a new human prion disorder, variably protease-sensitive prionopathy (VPSPr) [15], [30], [39], [69], [70]. It is noteworthy that protease-sensitive synthetic prions generated in vitro during polymerization of recombinant mouse PrP into amyloid fibers produced upon inoculation into wild mice prions composed exclusively of sPrPSc [66].
In laboratory rodent prion models, we found that levels of sPrPSc varied with the incubation time of the disease [15] but the molecular mechanism of this link was unknown [15], [30], [40]. Subsequent experiments with yeast prions indicated that replication rate may be an inverse function of the stability of misfolded protein [71]. The hypothesis based on these experiments posits that the less stable prions replicate faster by exposing more available sites for growth of the aggregates. Additionally, experiments with laboratory and synthetic prions in mouse suggested that the yeast prion principle may apply to mammalian prions as well. However, these experiments were based entirely on the correlation of the shorter incubation time of mouse inoculated with PrPSc that on WBs converted to protease-sensitive isoforms at a lower denaturant concentration, whereas the replication rates of mammalian prions were never determined [72].
In this paper we determined the conformational features and stability of human sPrPSc in sCJD. The data indicate that the levels as well as stability are linked to the progression rate of the disease. Despite the inevitable influence of variable genetic background and the potential difficulties in evaluating initial symptoms, the disease progression rate and incubation time jointly represent an important parameter, which is influenced by replication rate, propagation, and clearance of prions from the brain [2], [40]. The correlations among the levels of sPrPSc, the stability of sPrPSc, and the duration of the disease found in this study all indicate that sPrPSc conformers play an important role in the pathogenesis. When sPrPSc is less stable than rPrPSc, the difference in stability correlates with less accumulated sPrPSc and shorter duration of the disease. Conversely, when sPrP conformers are more stable than rPrPSc, we observe the opposite effect—more accumulated sPrPSc and extended disease duration. It remains to be determined if these effects represent an outcome of different replication rates and clearance, or whether they stem from as yet unknown aspects of the pathogenesis of sCJD.
All procedures were performed under protocols approved by the Institutional Review Board at Case Western Reserve University. In all cases, written informed consent for research was obtained from patient or legal guardian and the material used had appropriate ethical approval for use in this project. All patient's data and samples were coded and handled according to NIH guidelines to protect patients' identities.
We selected 46 representative subjects from a group of 340 patients with definitive diagnosis of sCJD. The criteria for inclusion were (1) availability of clinical diagnosis of CJD according to WHO criteria [73]–[75] and clearly determined and dated initial symptoms upon neurological examination to ascertain the disease duration; (2) methionine or valine homozygous at codon 129 of the human prion protein (PrP) gene (PRNP); (3) unequivocal classification as pure Type 1 or Type 2 sCJD according to WB pattern; (4) unequivocal classification of pathology as definite Type 1 or 2 at the National Prion Disease Pathology Surveillance Center (NPDPSC) in Cleveland, OH; (5) demographic data distribution within 95% confidence interval of the whole group resulting in no difference between selected cases and the whole group in any of the statistically followed parameters.
Retrospective charts review was carried out for all subjects, with particular attention to the documented initial cardinal clinical signs of sCJD such as cognitive impairment, ataxia, and myoclonus [73]–[75]. We also reviewed the findings on electroencephalography, brain magnetic resonance imaging, and CSF markers when available.
All Type 1–2 patients or uncertain cases were excluded from this study. DNA was extracted from frozen brain tissues in all cases, and genotypic analysis of the PRNP coding region was performed as described [29], [30], [76]. On the basis of diagnostic pathology, immunohistochemisty, and western blot (WB) examination of 2 or 3 brain regions (including frontal, occipital and cerebellum cortices) with mAb 3F4, the pathogenic PrPSc was classified as (1) Type 1 PrPSc(129 M) (n = 16); (2) Type 2 PrPSc (129 M, n = 16); or (3) Type 2 PrPSc (129 V, n = 14). Patients lacked pathogenic mutations in the PRNP and had no history of familial diseases or known exposure to prion agents. These cases underwent additional detailed WB analyses of the PrPSc so that we could ascertain the accuracy of their original classification and confirm that the same brain homogenate analyzed by CDI contained pure Type 1 PrPSc(129 M), Type 2 PrPSc(129 M), and Type 2 PrPSc(129 V).
Coronal sections of human brain tissues were obtained at autopsy and stored at 80°C. Three 200–350 mg cuts of frontal (superior and more posterior middle gyri) cortex were taken from each brain and used for molecular analyses. The other symmetric cerebral hemisphere was fixed in formalin and used for histologic and immunohistochemical purposes.
Slices of tissues weighing 200–350 mg were first homogenized to a final 15% (w/v) concentration in calcium- and magnesium-free PBS, pH 7.4, by 3 75 s cycles with Mini-beadbeater 16 Cell Disrupter (Biospec, Bartlesville, OK). The homogenates were then diluted to a final 5% (w/v) in 1% (v/v) sarkosyl in PBS, pH 7.4 and rehomogenized. After clarification at 500× g for 5 min., one aliquot of the supernatant was treated with protease inhibitors (0.5 mM PMSF and aprotinin and leupeptin at 5 ug/ml, respectively). The second aliquot was treated with 50 µg/ml of proteinase K (Amresco, Solon, OH) for 1 h at 37°C shaking 600 rpm on Eppendorf Thermomixer (Eppendorf, Hauppauge, NY) and PK was blocked with PMSF and aprotinin-leupeptin cocktail. Both aliquots were precipitated with final 0.32% (v/v) NaPTA after 1 h incubation at 37°C as described [15]. The samples were spun 30 min at 14,000× g in Allegra X-22R tabletop centrifuge (Beckman Coulter, Brea, CA) and the pellets were resuspended in 250 ul of deionized water containing protease inhibitors (0.05 mM PMSF, aprotinin and leupeptin at 1 ug/ml each, respectively, and stored for analysis at −80°C.
Both PK-treated and untreated samples were diluted 9-fold in 1× Laemmli Buffer (Bio-Rad, Hercules, CA) containing 5% (v/v) beta-mercaptoethanol (ME) and final 115 mM Tris-HCl, pH 6.8. Samples were heated for 5 min at 100°C and ∼2 ng of PrP per lane was loaded onto 1 mm 15% Polyacrylamide Tris-HCl, SDS-PAGE gels (Bio-Rad) mounted in Bio-Rad Western Blot apparatus. After electro-transfer to Immobilon-P Transfer Membranes (Millipore, Bedford, MA), the membranes were blocked with 2% (w/v) BSA in TBS containing 0.1% of Tween 20 (v/v) and 0.05% (v/v) Kathon CG/ICP (Sigma, St. Louis, MO). The PVDF membranes were developed with 0.05 ug/ml of biotinylated mAb 3F4 (Covance, Princeton, NJ) followed by 0.0175 ug/ml Streptavidin-Peroxidase conjugate (Fisher Scientific, Pittsburg, PA) or with ascitic fluid containing mAb 3F4 (kindly supplied by Richard Kascsak) diluted 1∶20,000 followed by Peroxidase-labeled sheep anti-mouse IgG Ab (Amersham, Piscataway, NJ) and diluted 1∶3000. The membranes were developed with the ECL Plus detection system (Amersham) and exposed to Kodak BioMax MR Films (Fisher Scientific) or Kodak BioMax XAR Films (Fisher Scientific).
The CDI for human PrP was performed as described previously [30], [47], with several modifications. First, we used white Lumitrac 600 High Binding Plates (E&K Scientific, Santa Clara, CA) coated with mAb 8H4 (epitope 175–185) [46] in 200 mM NaH2PO4 containing 0.03% (w/v) NaN3, pH 7.5. Second, aliquots of 20 µl from each fraction containing 0.007% (v/v) of Patent Blue V (Sigma) were directly loaded into wells of white strip plates prefilled with 200 µl of Assay Buffer (Perkin Elmer, Waltham, MA). Finally, the captured PrP was detected by a europium-conjugated [15] anti-PrP mAb 3F4 (epitope 108–112) [45] and the time-resolved fluorescence (TRF) signals were measured by the multi-mode microplate reader PHERAstar Plus (BMG LabTech, Durham, NC). The recHuPrP(90–231,129 M) and PrP(23–231,129 M) used as a calibrant in the CDI was a gift from Witold Surewicz, and preparation and purification have been described previously [77]. The initial concentration of recombinant human PrP(23–231) and PreP(90–231) was calculated from absorbance at 280 nm and molar extinction coefficient 56650 M−1 cm−1 and 21640 M−1 cm−1, respectively. The purified recombinant proteins were dissolved in 4 M GdnHCl and 50% Stabilcoat (SurModics, Eden Prairie, MN), and stored at −80°C. The concentration of PrP was calculated from the CDI signal of denatured samples using calibration cure prepared with either recPrP(23–231) for samples containing full length PrPSc or recPrP(90–231) for samples containing truncated rPrPSc (PrP 27–30) after proteinase-K treatment. This separate calibration was necessary due to the ∼3.5-fold lower affinity of mAb 3F4 with full length hurman PrP(23–231,129 M) compared to PrP(90–231,129 M) (Figure S3).
The denaturation of human PrPSc was performed as described previously [15], with several modifications. Frozen aliquots of PrPSc were thawed, sonicated 3×5 s at 60% power with Sonicator 4000 (Qsonica, Newtown, CT), and the concentration was adjusted to constant ∼50 ng/ml of PrPSc. The 15 µl aliquots in 15 tubes were treated with increasing concentrations of 8 M GdnHCl containing 0.007% (v/v) Patent Blue V (Sigma, St. Louis, MO) in 0.25 M or 0.5 M increments. After 30 min incubation at room temperature, individual samples were rapidly diluted with Assay Buffer (Perkin Elmer, Waltham, MA) containing diminishing concentrations of 8 M GdnHCl, so that the final concentration in all samples was 0.411 M. Each individual aliquot was immediately loaded in triplicate to dry white Lumitrac 600, High Binding Plates (E&K Scientific, Santa Clara, CA), coated with mAb 8H4, and developed in accordance with CDI protocol using europium-labeled mAb 3F4 for detection [15], [30], [41], [78].
The raw TRF signal was converted into the apparent fractional change of unfolding (Fapp) as follows: F = (TRFOBS−TRFN)/(TRFU−TRFN) where TRFOBS is the observed TRF value, and TRFN and TRFU are the TRF values for native and unfolded forms, respectively, at the given Gdn HCl concentration [7]. To determine the concentration of Gdn HCl where 50% of PrPSc is unfolded ([Gdn HCl]1/2), the data were fitted by least square method with a sigmoideal transition model (Equation 1):
The apparent fractional change (F) in the TRF signal is the function of Gdn HCl concentration(c); c1/2 is the concentration of Gdn HCl at which 50% of PrPSc is dissociated/unfolded and r is the slope constant. To determine the impact of protease treatment on the conformational stability of PrPSc, the values of fractional change after PK were subtracted from Fapp values obtained before PK (ΔFapp = F0−FPK) and then fitted with a Gaussian model to estimate the proportion and average stability of sPrPSc conformers (Equation 2):In this model, the Pk-induced fractional change is ΔF, F0 is fractional change at 0 concentration of Gdn HCl, and c0 is the Gdn HCl concentration at the maximum height A of the peak.
We investigated the effect of the following demographic and laboratory variables on survival: sex; age at onset; duration of the disease; electrophoretic Type of PrP 27–30; and the concentration and stability of PrPSc in Gdn HCl before and after PK treatment [15]. Cumulative survival curves were constructed by the Kaplan–Meier method, both overall and by stratifying for each of the above variables. For each type of PrPSc and PRNP gene polymorphism, we report descriptive statistics and the overall survival times stratified for each variable. In the comparison of different patient groups, P values were calculated using Anova. Comparisons of survival curves among groups were carried out by the log rank (Mantel-Cox) and generalized Wilcoxon test. To evaluate the dependency of disease duration upon the concentration and stability of PrPSc in individual CJD cases, the data were analyzed by non-linear regression using the logistic function or the nonlinear models with the best fit. To obtain significance and to compare the relative importance of each characteristic of PrPSc, we used ANOVA and F statistics with regression mean square (MSR) divided by the residual mean square (MSE). All the statistical analyses were performed using SPSS 17 software (SPSS Inc., Chicago, IL).
|
10.1371/journal.pntd.0006988 | A randomized trial of AmBisome monotherapy and AmBisome and miltefosine combination to treat visceral leishmaniasis in HIV co-infected patients in Ethiopia | Visceral leishmaniasis (VL) in human immunodeficiency virus (HIV) co-infected patients requires special case management. AmBisome monotherapy at 40 mg/kg is recommended by the World Health Organization. The objective of the study was to assess if a combination of a lower dose of AmBisome with miltefosine would show acceptable efficacy at the end of treatment.
An open-label, non-comparative randomized trial of AmBisome (30 mg/kg) with miltefosine (100 mg/day for 28 days), and AmBisome monotherapy (40 mg/kg) was conducted in Ethiopian VL patients co-infected with HIV (NCT02011958). A sequential design was used with a triangular continuation region. The primary outcome was parasite clearance at day 29, after the first round of treatment. Patients with clinical improvement but without parasite clearance at day 29 received a second round of the allocated treatment. Efficacy was evaluated again at day 58, after completion of treatment.
Recruitment was stopped after inclusion of 19 and 39 patients in monotherapy and combination arms respectively, as per pre-specified stopping rules. At D29, intention-to-treat efficacy in the AmBisome arm was 70% (95% CI 45–87%) in the unadjusted analysis, and 50% (95% CI 27–73%) in the adjusted analysis, while in the combination arm, it was 81% (95% CI 67–90%) and 67% (95% CI 48–82%) respectively. At D58, the adjusted efficacy was 55% (95% CI 32–78%) in the monotherapy arm, and 88% (95% CI 79–98%) in the combination arm.
No major safety concerns related to the study medication were identified. Ten SAEs were observed within the treatment period, and 4 deaths unrelated to the study medication.
The extended treatment strategy with the combination regimen showed the highest documented efficacy in HIV-VL patients; these results support a recommendation of this regimen as first-line treatment strategy for HIV-VL patients in eastern Africa.
www.clinicaltrials.gov NCT02011958.
| Visceral Leishmaniasis is a complex parasitological disease and is particularly challenging to treat in patients coinfected with human immunodeficiency virus (HIV). Antimonial drugs used in first-line treatments for immunocompetent patients in eastern Africa are more toxic in immunocompromised patients. In 2010, a WHO expert committee recommended a lipid formulation of amphotericin B as first line treatment for HIV/VL co-infected patients, based on a single clinical trial conducted in Spain and empirical information obtained from scattered case reports using AmBisome (liposomal amphotericin B). In addition, Médecins Sans Frontières began a compassionate use regimen combining AmBisome and miltefosine a in a treatment centre in Northwest Ethiopia with encouraging results. Here, we report the results of a trial to assess the efficacy and safety of both the currently internationally recommended treatment of AmBisome monotherapy and the new AmBisome-miltefosine combination regimen, in Ethiopian patients. The results of this trial show that one course of treatment with either regimen could be insufficient to clear parasites in a high proportion of patients and that an extended treatment strategy, of administrating a second course of treatment, could lead to a high parasite clearance rate in patients treated with the combination regimen.
| Human immunodeficiency virus (HIV) affects visceral leishmaniasis (VL) by increasing its incidence, altering its clinical manifestation and severity, and, more importantly, by worsening treatment outcomes and relapse rates [1,2]. VL is the second most deadly protozoan infection after malaria. HIV-VL co-infection has been observed in at least 35 countries on four continents [3–6]. Following the introduction of highly active anti-retroviral therapy (HAART), the incidence of VL in HIV patients has decreased in most settings [7]. Northwest Ethiopia has the highest burden globally, with HIV rates among VL patients ranging between 20–40% [2,3]. Typically, young male seasonal workers migrate to the lowlands to harvest crops, sleep in improvised shelters, and are exposed to the bites of sand flies. In addition, migrant workers are at high risk of HIV infection [2,8,9].
The current WHO recommended regimen is infusion of amphotericin B lipid formulations 3–5 mg/kg daily or intermittently up to a total dose of 40 mg/kg [10], despite the absence of proper evaluation in most endemic areas. The Ethiopian National Guidelines (2013) [11] recommend liposomal amphotericin B and sodium stibogluconate (SSG) as the first and second line treatments for HIV-VL patients. While effectiveness studies at 40 mg/kg AmBisome are lacking, effectiveness at 30 mg/kg was less than 60% among HIV co-infected patients [12]. SSG has poor effectiveness (43%–70%) and considerable toxicity, with an increased risk of death in HIV co-infected patients [13,14]. Thus, there is an urgent need for a better treatment approach to VL in HIV co-infected patients.
Preventing recurrence of disease is another important factor, as relapse cases are even more difficult to cure, consequently becoming reservoirs of the parasite and playing a role in transmission [15]. Negative aspirate after treatment is the best predictor of absent/delayed relapse. Surviving HIV patients develop some tolerance to the parasite. The annual VL relapse rate among HAART-taking HIV patients in Northwest Ethiopia was more than 60% if their CD4 count was less than 100 cells/μl [16]. A recent cohort study in Northwest Ethiopia demonstrated 71% disease-free survival one year after VL treatment using pentamidine as secondary prophylaxis [17].
This clinical trial was conducted primarily to evaluate the day 29 efficacy of a combination of AmBisome (30 mg/kg) with miltefosine (100 mg/day for 28 days) and AmBisome monotherapy (40 mg/kg) for VL in HIV co-infected patients in Northwest Ethiopia. Secondary objectives were to assess efficacy at day 58 and the safety of the regimens. A long-term evaluation of relapse-free survival up to one-year follow-up (Day 390), including an estimate of the relapse rate in those receiving pentamidine as a prophylactic treatment, will be reported in another publication.
This study was conducted according to principles of the Helsinki declaration, Good Clinical Practice (GCP) rules, and local regulations. The Protocol and the Informed Consent Sheet were revised and approved by the University of Gondar Institutional Review Board, the Ethiopian National Research Ethics Review Committee, the Médecins Sans Frontières Ethics Review Board, the London School of Hygiene and Tropical Medicine Research Ethics Committee, the Antwerp University Hospital Ethics Committee, and the Prince Leopold Institute of Tropical Medicine Institutional Review Board. Approval from the Food, Medicine and Healthcare Administration and the Control Authority of Ethiopia was also obtained before inclusion of the patients. Patients were included after completion of a written informed consent process.
This study, registered on www.clinicaltrials.gov (NCT02011958), included two components. The first component was a non-comparative randomized open-label clinical trial to evaluate the end of treatment efficacy and safety of two treatment regimens for VL in HIV co-infected patients (Fig 1). This component used a sequential design, in which the sample size is not fixed in advance but depends on the accruing endpoint data. More specifically, we used a triangular continuation region which is a graphical representation of the stopping rules of the trial [18,19] (Fig 2). The results cannot easily be interpreted without reference to the graphical representation. The trial design allows for regular interim analyses, and hence possible stopping, after every 10 patients, based on acceptable or unacceptable efficacy values set in the protocol. This design allows recruitment to be stopped as patients are being evaluated if the observed efficacy is too low (lack of promise), i.e. crossing the lower boundary of the triangular region (pink area), or sufficiently high (promise), i.e. crossing the upper boundary (blue area). Otherwise recruitment continues, corresponding to the interior of the triangular continuation region in the graphical representation (green area). In more technical terms, the null hypothesis was that the proportion of patients reaching negative parasitology at day 29 (p) is less than or equal to a value p0, which we set to 75%. A test statistic falling outside the continuation region in the graphical representation would lead to a decision to stop recruitment. If the upper boundary is crossed during an interim analysis, then the null hypothesis is rejected and we conclude that p>75%. Crossing the lower boundary implies that the null hypothesis (proportion reaching negative parasitology ≤75%) is not rejected and there is a specified power to exclude a proportion reaching negative parasitology of pa, for which we chose a value of 90%. Less formally, on crossing the upper boundary, the trial should be stopped for promise (efficacy above 75%), and on crossing the lower boundary it should be stopped for lack of promise (efficacy below 90%). The type I error rate and power of the study were pre-specified as 5% and 95%, respectively (α = β = 0·05). Interim analyses were specified after every 10 patients in each arm had reached the Day 29 endpoint. An independent Data Safety Monitoring Board (DSMB) evaluated the results and confirmed decisions to continue or stop with recruitment in each arm. Based on the sequential trial design and the parameters described above, the maximum sample size per arm was 66. The second component included a follow-up for 12 months with secondary prophylaxis for severely immunocompromised patients. Results will be described in a separate publication.
Patients were recruited at two facilities in Ethiopia, the Leishmaniasis Research and Treatment Centre, located within the teaching hospital of the University of Gondar, and the Abdurafi Health Centre, located in a rural setting on the border with Sudan and supported by Médecins Sans Frontières. These two sites are the main VL treatment centers in Ethiopia. On average 400 and 800 VL patients are managed every year respectively at these two centers.
Patients were diagnosed for VL and for HIV as per the Ethiopian national guidelines [11]. Upon presentation of clinical signs and symptoms of VL (e.g. irregular fever for more than 2 weeks, hepatomegaly and/or splenomegaly, weight loss), the patients were diagnosed by identification of the Leishmania parasite by microscopy in tissue aspirate (spleen aspirate was the preferred methodology, or bone marrow aspirate in case of contra-indication e.g. bleeding tendency, platelets below 40,000/mm3, haemoglobin below 5 g/dL). Patients were eligible regardless of whether this was the first episode of VL (primary case) or whether it was a relapse case with single or multiple relapses. The parasite strains were not identified, but according to epidemiological reports [20,21], the circulating parasite causing visceral leishmaniasis in Ethiopia is Leishmania donovani. HIV status is routinely determined by two rapid tests followed by a third confirmatory test in case of discrepancy. Within the trial, it was reconfirmed using an enzyme immunoassay (ImmunoComb II HIV 1&2 BiSpot, Orgenics Ltd.).
Subjects were allocated to treatment using random block sizes, stratified by site (Gondar & Abdurafi) and by patient type (whether the VL episode at screening was a primary or relapse case).
The randomization list was prepared by the data management team. Site investigators were blinded to block sizes. Randomization codes were prepared in sealed, sequentially numbered, opaque envelopes and were under the control of the site investigator.
Patients and treating physicians were not masked to study treatment due to the considerable differences in the administration of the treatment arms (different dosing schedule of an infused treatment plus oral administration).
Patients received in-patient treatment. Liposomal amphotericin B (AmBisome, Gilead Inc.) was stored and reconstituted as per the manufacturer’s recommendations. In the monotherapy arm, a dose of 40 mg/kg total dose of AmBisome (adapted from WHO recommendations) was administered by intra-venous (IV) slow infusion of 5 mg/kg on days 1 to 5, 10, 17, and 24.
In the combination arm, a 30 mg/kg total dose of AmBisome was administered according to the following schedule: IV slow infusion of 5 mg/kg on days 1, 3, 5, 7, 9, and 11; with a miltefosine (Paladin Therapeutics, Inc.) 50 mg capsule orally twice a day for 28 days (all patients included were above 25 kg).
For both arms, a pre-testing dose of 1 mg of AmBisome was administered to patients before treatment to rule out allergic reactions, as recommended in the AmBisome Summary of Product Characteristics.
An initial assessment of cure was conducted at D29, through clinical and parasitological examination (spleen or bone marrow aspiration). Patients with negative parasitology were considered cured of VL and started the follow-up period (Fig 1). Patients who had clinically improved but still had detectable parasites at D29 were given extended treatment using the same regimen to which they had been randomized. Patients with detectable parasites at D29, and who were clinically unwell, and patients with parasites at D58 were given a rescue medication of the clinician’s choice (Fig 1).
Once the patients had a negative parasitology result, they started a follow-up period of one year (up to D390) to assess long-term relapse-free survival and safety (not reported here).
The primary endpoint was parasitological clearance at day 29 (D29), which was defined as absence of parasites in tissue aspirate at D29 (bone marrow or spleen, with spleen aspirate as the preferred option). Treatment failure at D29 was defined as presence of parasites at the D29 assessment, or death prior to the D29 assessment, or no clinical response to treatment requiring rescue medication on or before D29.
A secondary endpoint to assess treatment outcome after extended treatment was defined as efficacy at day 58 (D58). D58 treatment success was defined as: (i) being parasite free at D29 and no recurrence of symptoms by D58 or (ii) being parasite free at D58 after extended treatment. Thus, D58 failures were patients who (i) received rescue treatment prior to, or at, the D58 visit, or (ii) were confirmed to be parasite positive at D58 or (iii) died up to D58. A patient with detectable parasites at D29 who then received extended treatment would be a treatment failure at D29 but a success at D58 if no parasites were detected at D58.
All adverse events (AEs) and serious adverse events (SAEs) were captured up to one month after the last dose of study medication (D58 if patients received one round of treatment or D86 if patients received a second round). Grading of the severity of the events was based on Common Terminology Criteria for Adverse Events (CTCAE), Version 4.0 [22].
Safety laboratory assessments were performed at baseline, D3, D10, D29, and D58, and during follow-up as needed. CD4 count was measured at baseline, on D29, and within one month of reaching negative parasitology. HIV viral load was measured at baseline and every 6 months as per routine practice, samples were sent to central regional or national reference laboratories. Results of the viral load were often not available in time to support HIV case management.
Data capture and management utilized OpenClinica and Stata software [23]. Data analysis was performed using Stata, version 14.
The intention to treat (ITT) population was pre-specified as primary analysis population for the sequential interim analyses of treatment success at D29. Both ITT and per-protocol (PP) populations for treatment success at D29 and D58 were used for final analyses of treatment success at D29 and D58. ITT was considered primary analysis population. Interim analyses took place each time 10 patients per arm had reached the D29 primary endpoint assessment. Decision making was based on the position of the test statistic relative to the triangular continuation region, as described above [18]. Recruitment and randomization of new patients were not stopped during interim analysis, but were stopped on the recommendation of the DSMB when the interim analysis showed that the boundary of the continuation region had been crossed.
In a sequential trial, recruitment is stopped when a pre-specified difference is reached in the accruing data. In the current trial, this difference is between the observed and expected numbers of treatment successes. Stopping when a difference of interest has been achieved implies a risk that the final results are a “random high” [24]. In other words, the results will tend to be more extreme than if the sample size had been fixed in advance. To allow for this, specific analysis methods are required. In particular, simple maximum likelihood estimates—which in the current trial would be simple proportions—in general suffer from ‘substantial bias’ [25], in sequential trials. Here, the pre-specified approach, taking into account the sequential design, was to use point and interval estimates of efficacy obtained at D29, as per Whitehead [26]. Because this approach assumes that the final data lie close to the boundary of the continuation region (Fig 2), an additional post-hoc analysis, as per Liu et al. [27], was done of D29 efficacy to allow for this and for ‘over-run’, i.e. inclusion, in the final analysis, of people who had been randomized to the trial before the final interim analysis, but whose D29 data were not then available for inclusion.
The sequential stopping rule applied only as far as the primary outcome at D29. Hence, the D58 outcome is subject to a mixture of sequential and classical follow-up. To take this into account, the probability of treatment success at D58 was estimated by a probability tree method previously developed for this purpose [28].
All adverse events were coded according to Medical Dictionary for Regulatory Activities (MedDRA, Version 14.0). Safety outcomes were the number (%) of patients experiencing a serious adverse event (SAE) at any time during the trial or follow-up, the number (%) of patients experiencing an adverse event (AE) between Day 1 of initial treatment and up to one month after completing treatment, and the incidence of adverse drug reactions (ADRs) by preferred MedDRA term estimated as the number (%) of patients experiencing at least one ADR for each MedDRA lower level preferred term. We also report the rate of ADR accounting for time at risk based on the number of rounds of treatment.
CD4 counts were compared between time points by paired t test.
Patients were recruited between 14 August 2014 and 18 August 2015. Out of the 536 parasitologically confirmed VL patients (Fig 3), 81 were HIV positive according to the Ethiopian national HIV testing algorithm and confirmatory test and 59 patients were enrolled.
Both arms had a similar distribution of baseline characteristics, including body mass index, VL status, CD4 count, and prior antiretroviral treatment (Tables 1 and 2).
Prior to D29 there were no protocol deviations, thus the ITT populations were identical to the PP. At D58, the PP population excluded 5 patients with major protocol deviations (Fig 3). There were no missing outcome data. One patient died after randomization before receiving any treatment and was excluded from all analyses.
In interim analyses, both arms crossed the lower side of the boundary (Fig 2), indicating treatment success of less than 90% at D29. The AmBisome monotherapy crossed first, based on data from 10 patients, and the combination arm crossed the boundary based on data from 20 patients.
In the AmBisome arm, there were only 7 treatment successes out of 19 subjects in the final data. Stopping in this way, for a difference of a given magnitude, is characteristic of a sequential trial and required particular statistical methods, as explained above. This pre-specified analysis gives an estimated efficacy of 70% (95% CI 45–87%) at D29 in the AmBisome arm. However, this does not take account of the substantial over-run to the lower side of the boundary, visible in Fig 2 (see the Fig 2 legend for an explanation of over-run). After accounting for this, the estimated efficacy was lower, at 50% (27–73%). In the combination arm, the estimated success rate is 81% from the pre-specified efficacy analysis (67–90%), and with an adjusted efficacy of 67% (48–82%), taking account of over-run (Table 3).
Efficacy was also estimated at Day 58 (Table 3) to evaluate the success of the extended treatment strategy. Using the over-run adjusted D29 analysis as inputs, the D58 ITT efficacy was 55% (95% CI 32–78%) in the monotherapy arm, and the PP efficacy was 59% (35–83%). The same analysis in the combination arm gave 88% (79–98%) and 91% (81–100%) for ITT and PP respectively. These D58 efficacy estimates are higher than those at D29 because improvement in status between these two time points was more likely than deterioration, thanks to prolongation of the treatment. If the over-run phenomenon is not considered, D58 efficacy estimates are slightly higher in both arms, as was the case for D29 (Table 3).
All patients experienced at least one AE (Table 4). A higher percentage of patients experienced at least one ADR in the combination arm than in the monotherapy arm. However, the rate of ADR was similar between arms, after accounting for time-at-risk for those receiving one or two rounds of treatment (Table 5). AEs and ADRs were mostly mild or moderate. ADRs occurring in more than 10% of patients were dyspepsia, gastritis, vomiting mainly related to miltefosine, increased blood creatinine, and hypokalaemia related to AmBisome. Two hypokalaemia events (one per arm) were severe in intensity as per CTCAE [22] criteria (between 2·5 and 3 mmol/L) and required supplementation. All ADRs were considered expected as per the reference documentation for each drug [29,30]. Ten SAEs were reported during the treatment phase (Table 6), including four deaths, mostly due to infectious events (sepsis, strongyloidiasis hyperinfection syndrome, meningitis/encephalitis due either to toxoplasma or tuberculosis). One patient died from pancreatitis/renal failure related to rescue treatment with sodium stibogluconate and paromomycin, along with ARV drugs, a combination known to be toxic in HIV-VL co-infected patients [14].
At admission, approximately 70% of patients were on antiretroviral treatment; overall and by arm. The tenofovir-lamivudine-efavirenz combination, the first line treatment according to Ethiopian guidelines, was the most common ARV drug combination used. Only two patients were receiving a protease inhibitor based regimen. All newly diagnosed HIV patients started the ARV treatment after completion of the VL treatment, except for one refusal. Three patients changed their ARV regimen during the VL treatment. CD4 recovery was substantial at the end of VL treatment compared to baseline, 23 patients presented with CD4 count above 200 cells/μl compared to only one at baseline (Fig 4A and 4B). In the monotherapy arm, the average increase in CD4 from baseline to day 29 was 52 cells/μl (95% CI 24–79, paired t-test p = 0·001) and in the combination arm 111 cells/μl (95% CI 67–155 paired t-test p<0·001). HIV viral load was available for 55 patients at baseline (Table 2). Only 16 patients had undetectable viral load among the 41 already receiving ART treatment at inclusion.
The burden of VL in HIV patients and the associated poor treatment outcome with all currently available medicines underline the urgent need for better treatment approaches. This trial was therefore designed to evaluate a combination treatment (AmBisome plus miltefosine) and the current WHO recommended regimen (AmBisome monotherapy), with the aim of providing better quality evidence for future guidelines. The combination arm showed an efficacy rate of 67% on day 29 and 88% on day 58 with prolonged treatment (ITT). In the PP, efficacy at D58 reached 91%.
In 2008, Ter Horst and colleagues [16] made 5 recommendations for the management of VL and HIV co-infected patients in Ethiopia: (i) ART should be provided to all HIV positive individuals; (ii) VL should be an AIDS-defining illness and a valid entry point to ART, irrespective of CD4 count, to reduce the chance of relapse; (iii) secondary prophylaxis is necessary when the risk of relapse is high; (iv) parasitological clearance is a crucial end point for VL treatment; and (v) combination therapy can minimize the risk of developing resistance. Our study focuses on an initial effective treatment defined by parasite clearance at the end of VL therapy (responding to points iv and v), supplemented by secondary prophylaxis and ART (responding to points ii and iii; manuscript in preparation).
The limited number of study reports available on HIV-VL treatment in the region generally show poor treatment outcomes with all the available anti-leishmanial drugs. In a previous study that compared SSG and miltefosine for the treatment of VL, a subgroup analysis showed cure rates of 90% and 78%, and death rates of 7% and 2% for SSG and miltefosine respectively [31]. This study included mainly primary VL, and a death rate of 19% was also reported among patients with unknown HIV status and treated with SSG (patients suspected of advanced AIDS disease). Subsequent studies evaluating SSG in treatment of HIV-VL co-infected patients showed a cure rate that ranged from 43% to 70%, and death rates during treatment that ranged from 14–17% (or ~15%) [2]. AmBisome at a 30 mg/kg dose showed a cure rate of 60% and failure rate of 32%, with a worse outcome in relapsed cases as compared to primary HIV-VL co-infected patients [12].
A randomized trial that enrolled 57 HIV-VL patients from Spain, infected with a different species (Leishmania infantum), compared two regimens of amphotericin B lipid complex (ABLC 15 mg/kg and 30 mg/kg total dose), with meglumine antimoniate [32]. The initial cure rates were 33% (95% CI 13%-59%), 42% (95% CI 16%-62%), and 37% (95% CI 16%-62%), respectively. Treatment with the amphotericin B lipid formulation had fewer safety concerns and treatment discontinuations. Despite the limited number of patients per arm in this randomized study (18, 20, and 19 respectively), this evidence, together with a number of reported case series using AmBisome, was used by the WHO Expert Committee to recommend the use of amphotericin B lipid formulations up to 40mg/kg for the treatment of HIV-VL co-infected patients with an ‘A’ grade of evidence [10].
The clinical trial described here is the first to evaluate the efficacy of the combined regimen of AmBisome and 28-day miltefosine. It has demonstrated much higher efficacy in African HIV-VL co-infected patients than the previous trial in Spain using ABLC monotherapy [32]. This result was obtained when the combination was used within a strategy of extended treatment based on the patient’s initial clinical and parasitological response to treatment. Indeed, this strategy achieved a parasite clearance rate of 91% at the end of treatment for patients with high compliance (i.e. per protocol analysis population). Notably, the combination treatment offers encouraging efficacy in relapse patients with previous history of VL treatment, as they represent half of the patient population included in the trial.
Importantly, in the context of this trial, we observed the value of parasitological assessment at the end of treatment, especially for HIV-VL patients, clinical evaluation not being sufficiently sensitive to detect patients who failed to clear parasites after the initial treatment. This allowed us to demonstrate that prolongation of same treatment can be effective.
In terms of safety, although all patients experienced at least one adverse event, which is to be expected in this seriously ill population, no AE led to VL treatment discontinuation. Adverse drug reactions were reported in a significant number of patients, but most events were of mild intensity, and mostly corresponded to a single episode of vomiting, known to be associated with miltefosine. Sporadic hypokalaemia remains a concern with AmBisome and requires close monitoring, as previously reported in India [33]. These data therefore suggest a satisfactory safety profile in a population with a high burden of concomitant illness and medication. Prolonging treatment using the extended treatment strategy does not seem to create any safety concern impacting compliance to treatment, although the number of patients in the trial was not conducive to detecting rare events.
Although the combination regimen and treatment strategy identified by this randomized trial shows promising results for the HIV-VL co-infected population, the medicines used remain very expensive, and quality of HIV care remains a challenge to ensuring long-term patient survival. VL control programs in the region are mostly dependent on the support of international organizations that should consider HIV-VL patients as fully part of their main target, despite the financial burden that they represent. Indeed, since HIV-VL co-infected patients are chronically infected, frequently suffer relapse episodes (with high parasitaemia), and can infect sand flies [15], they act as a reservoir in the population. Implementation of a regimen that enhances parasite clearance is thus of major importance both for the individual patient and at the public health level to reduce circulation of the parasite in the community. Ensuring access and optimal care for these patients could impact the efficacy of elimination programmes such as the Indian Kala-azar Elimination Program.
Considering both individual and public health benefits, there is a strong case for the prompt adoption of this strategy into national and regional African guidelines. The Ethiopian authorities have committed to rapidly endorsing the AmBisome and miltefosine combination as the first line regimen in their national treatment guidelines for HIV-VL co-infected patients.
Because both arms crossed the triangular boundary on the lower side, corresponding to a lack of promise, defined as an efficacy of less than 90%, this can be considered as a limitation to the study. This 90% parameter was arguably set too high for this patient population, because efficacy values, for example, in the range 80–90% are still higher than observed in previous studies in HIV-VL patients. Moreover, the data over ran the triangular boundary by a large amount, while the primary analysis assumes this to be negligible. It was therefore felt necessary to take this into account in an additional analysis which is given prominence in the current report. Because of the scarcity of suitable clinical trial sites accessible to patients from remote areas and providing sufficient level of security and stability, the trial was not designed or powered for a comparison of efficacy between arms. However, there is no overlap between 95% CIs around the ITT efficacy at D58 accounting for over-run, tending to suggest higher D58 efficacy in the combination arm.
The numbers of HIV-VL co-infected patients seems to be decreasing in Ethiopia, probably due to the implementation of prevention strategies and large scale anti-retroviral treatment programs. However, some gaps remain in identifying patients with low compliance to HIV treatment, leading to ART failure and emergence of viral resistance, with a consequently late switch to second line ARV treatment. HIV viral load services are limited and results are reported only after long delays. Suboptimal HIV treatment might result in re-emergence of VL cases. This study was not designed to evaluate the importance of ART regimen on the efficacy and sustainability of the response to VL treatment. Closer monitoring of the evolution of CD4, cytokine profile, and HIV viral load would have allowed for a better understanding of the pathophysiology of co-infection, but this was not possible in the clinical settings where the trial was conducted.
In terms of generalizability, geographic variation in efficacy is observed with the commonly used anti-leishmanial medicines, in particular between Africa and India. This study was conducted in Northwest Ethiopia where, as in Sudan, L. donovani has high strain diversity and is often difficult to treat. These results, even if encouraging, cannot be extrapolated to other settings without reservation. Experience shows that in the Indian subcontinent patients usually respond to lower doses of treatment compared to in Africa. A retrospective study conducted in India with a similar combination regimen (same dose of AmBisome (30 mg/kg) but 14 instead of 28 days of miltefosine) suggested this regimen would be safe, effective, and tolerable [34]. There is now an ongoing randomized trial in India to evaluate this regimen (CTRI/2015/05/005807). In Europe and Brazil, which share the same parasite L. infantum, differences in the circulating parasites would probably justify bridging studies.
In conclusion, the results of this randomized trial strongly support a change in the treatment recommendations for HIV-VL co-infected patients, from AmBisome monotherapy to combination therapy as the first line treatment. A new case management strategy where duration of treatment is dependent on reaching a negative parasitology, by using one or two rounds of treatment, should be adopted. AmBisome-miltefosine combination therapy has a satisfactory safety profile and is highly efficacious.
|
10.1371/journal.pntd.0005872 | Spatial distribution and risk factors of Schistosoma haematobium and hookworm infections among schoolchildren in Kwale, Kenya | Large-scale schistosomiasis control programs are implemented in regions with diverse social and economic environments. A key epidemiological feature of schistosomiasis is its small-scale heterogeneity. Locally profiling disease dynamics including risk factors associated with its transmission is essential for designing appropriate control programs. To determine spatial distribution of schistosomiasis and its drivers, we examined schoolchildren in Kwale, Kenya.
We conducted a cross-sectional study of 368 schoolchildren from six primary schools. Soil-transmitted helminths and Schistosoma mansoni eggs in stool were evaluated by the Kato-Katz method. We measured the intensity of Schistosoma haematobium infection by urine filtration. The geometrical mean intensity of S. haematobium was 3.1 eggs/10 ml urine (school range, 1.4–9.2). The hookworm geometric mean intensity was 3.2 eggs/g feces (school range, 0–17.4). Heterogeneity in the intensity of S. haematobium and hookworm infections was evident in the study area. To identify factors associated with the intensity of helminth infections, we utilized negative binomial generalized linear mixed models. The intensity of S. haematobium infection was associated with religion and socioeconomic status (SES), while that of hookworm infection was related to SES, sex, distance to river and history of anthelmintic treatment.
Both S. haematobium and hookworm infections showed micro-geographical heterogeneities in this Kwale community. To confirm and explain our observation of high S. haematobium risk among Muslims, further extensive investigations are necessary. The observed small scale clustering of the S. haematobium and hookworm infections might imply less uniform strategies even at finer scale for efficient utilization of limited resources.
| The World Health Organization is spearheading the war on neglected tropical diseases, including helminth infections, by encouraging its member states to intensify control efforts. This call has recently been answered in most endemic regions of helminthiasis and governments are scaling up chemotherapy-based control programs in collaboration with private and public partners. However, it is necessary to clearly understand factors driving local transmission dynamics of helminth infections to design effective control programs. Here, we conducted a cross-sectional survey of 368 primary schoolchildren in Kwale, Kenya, and identified factors associated with the intensity of Schistosoma haematobium and hookworm infections. The negative binomial generalized linear mixed model showed the intensity of S. haematobium infection was much higher among Muslims and schoolchildren from low socioeconomic status households. High intensity of hookworm infection was associated with sex, SES, distance to river and history of anthelmintic treatment. Our findings demonstrate considering social and cultural drivers of NTDs could be beneficial in designing of efficient control programs and expediting NTDs control.
| Schistosomiasis and soil-transmitted helminthiases are among neglected tropical diseases targeted for control by the World Health Organization (WHO) [1]. Globally, soil-transmitted helminths (STHs), such as hookworms (Ancylostoma duodenale and Necator americanus), Ascaris lumbricoides and Trichuris trichiura, infect 1.5 billion people [2]. By 2014, 258 million individuals were estimated to be suffering from schistosomiasis, which is endemic in 78 countries worldwide. In Kenya, approximately 17.4 million people are at risk of schistosomiasis [3] and approximately 9.1 million Kenyans are in danger of soil-transmitted helminthiases [4].
Two schistosome pathogens dominant in Kenya are Schistosoma haematobium, which causes urogenital schistosomiasis, and Schistosoma mansoni, which is responsible for intestinal schistosomiasis [4]. Disease distribution chiefly depends on the presence of Bulinus spp. and Biomphalaria spp. as intermediate host snails for S. haematobium and S. mansoni, respectively [5]. Along the Kenyan coast, schistosomiasis is almost entirely caused by S. haematobium. The constant high temperature along the coast restricts the proliferation of Biomphalaria spp. host snails in the area [6]. Streams, seasonal pools, quarry pits and drainage canals are primary habitats for Bulinus spp. along the Kenyan coast [7].
Apart from distribution of intermediate host snails, sanitation and human contact with infested water play a significant role in schistosomiasis transmission [8,9]. The intensity of infection is influenced by water contact frequency and duration in infested water [8]. Factors such as age, gender, occupation, female household head’s education level, religion, SES and house location can influence a person’s contact with infested water [10–13]. Therefore, dynamics of helminth infection can, to some extent, be viewed as a result of the behavior and livelihoods of individuals in the context of their physical, economic, social and cultural environments.
Small-scale spatial heterogeneity is one of the most striking features of schistosomiasis from an epidemiological point of view and is chiefly due to locally determined factors [13–17]. Past studies have also linked persistence of schistosomiasis and soil-transmitted helminthiasis in endemic regions to low socioeconomic status (SES). Low SES can result in lack of access to safe water and improved sanitation in addition to poor hygiene practices [18,19].
Identifying local epidemiological drivers of helminthiasis in endemic areas is necessary to generate vital data for improving current control programs toward achieving maximum benefits. This study, therefore, determined factors associated with the intensity of S. haematobium and hookworm infections among schoolchildren in Kwale, Kenya.
The study was carried out in Kwale, a rural setting located on the south coast of Kenya (Fig 1). There is an established Health and Demographic Surveillance System (HDSS) in Kwale which covers an area of 384.9 km2 with 7,617 households and 42,585 inhabitants. HDSS Kwale lies between latitudes 4°17′S and 4°5′S and longitudes 39°15′E and 39°29′E [20]. Compared to other counties in Kenya, Kwale is among the poorest. More than half of the population does not have access to improved sanitation [21]. Residents of Kwale engage in farming as their primary economic activity for subsistence. The two largest religions are Islam and Christianity.
The net primary school enrolment rate is about 80% which is lower than the national average [22]. Kwale benefited from deworming exercise of the Kenya National Program for Elimination of Lymphatic Filariasis. Individuals aged 2 years and over received a single-dose of albendazole and diethylcarbamazine citrate in 2003, 2005 and 2008. The respective treatment coverage was 77%, 76% and 62.8% [23,24]. The Kenya National School-Based Deworming Programme (NSBDP) to control schistosome and STH infections was launched in 2009 where 3.6 million school aged children were dewormed with albendazole in endemic regions. In 2012, the NSBDP was scaled up and albendazole and praziquantel were co-administered to school children in Kwale County in 2013 and 2014. During the year 2015 only albendazole was administered in Kwale County due to logistical challenges experienced by NSBDP in the country [25–27].
A cross-sectional study was conducted from January to March 2012. Our study targeted schoolchildren in class 4. Only full grade primary schools were included in this study. There were 40 primary schools in HDSS Kwale, of which 10 were private schools as of January 2012 (Data manager HDSS Kwale, self-report). Twenty-three schools with 1,502 children in class 4 met the inclusion criteria. Since the prevalence of helminthiasis in the study area was unknown, we set it at 50%. Precision and design effect (α = 0.1) were set at 5% and 3%, respectively. Based on these parameters, a sample of 270 children was deemed sufficient for this study. The average size of class 4 in the eligible schools was 47 pupils; with the assumption of a 95% response rate, six schools were adequate for this study. Random cluster sampling of six schools (Fig 1) was performed using R statistical software version 2.13.1 [28]. Ninety-two percent of parents/guardians consented and consequently, 427 children were recruited in the study.
Trained interviewers gathered demographic and socioeconomic data from parents/guardians in home settings using a pretested questionnaire. A SES index was constructed based on main floor, wall and roof material of the house. Additionally, we included number of household members sharing a sleeping room and land size. Other components considered for SES were possession of: solar panel, bicycle, radio and mobile phone. Principal component analysis of wealth related variables was conducted in SPSS version 17 [29]. We created a wealth quintile and categorized participants into “Most poor” “very poor,” “poor,” “less poor” and “least poor” groups. Generation of wealth index by PCA is detailed in S1 File. We also gathered data on female household head’s education level. The question on education level had five categories: “none,” “incomplete primary,” “complete primary,” “secondary level” and “at least college level.” This was later categorized as “none/incomplete primary” and “above primary” since majority of participants did not complete primary school. Household religious affiliation was categorized as “Christian,” “Muslim” or “Atheist”. The interviewers verified the age of the children during home visits by cross-checking official birth certificates or baptism cards. Household geo-coordinates were recorded using a handheld global positioning system unit (Garmin eTrex H, Deutschland, Garmin International, Germany). The water contact behavior and shoe wearing practices of the children were assessed through a questionnaire administered to the children at school. Shoe wearing habit was recorded as “always” or “never.” For water contact frequency, the children were asked how often they bathed or washed in the river: “daily”, “3–6 times per week”, “1–2 times per week” or “never”.
A day before parasitological screening commenced, participating children were issued stool containers. The research team instructed them on how to collect a portion of their stool in the morning on the next day. The study group distributed urine sample containers to participants on the actual screening day between 09:00 and 13:00 hours. We collected both fecal and urine samples for 3 consecutive days. The laboratory staff labeled specimen containers with a unique code assigned to each child. S. mansoni and STH infections were examined by the Kato-Katz fecal thick smear technique for stool [30]. Briefly, thick fecal smears prepared using 41.7 mg plastic templates were observed within 1 hour using a light microscope to detect and quantify hookworm eggs. The slides were left to clear within 24 hours for identification of S. mansoni, T. trichiura and A. lumbricoides eggs. We multiplied the number of eggs observed by 24 to express infection intensity as the number of eggs per gram of feces (EPG).
For S. haematobium assessment, 10 ml of urine was aliquoted using disposable syringes and filtered through a polycarbonate filter membrane. The filters measured 25 mm in diameter with a pore size of 12 μm (Whatman, Kent, UK). The urine was filtered at the sample collection sites. The filtrates were placed on labeled microscope glass slides and stored in slide boxes for subsequent analysis under a light microscope. We expressed the intensity of S. haematobium infection as the number of eggs detected per 10 ml of urine. Our team examined urine on 3 succeeding days to prevent misdiagnosis due to day-to-day variation in egg excretion. Arithmetic mean for the three slides, both for stool and urine, was used to express infection status of each child. An individual was deemed to be S. haematobium positive if at least one egg was observed on microscopic examination of urine on either day. For school infection intensity, geometric mean was obtained using the n+1 transformation for a series of egg outputs including zero. We categorized the extent of S. haematobium infection intensity as light (1–49 eggs/10 ml urine) or heavy (≥50 eggs/10 ml urine). Hookworm infection was categorized as light (1–1,999 EPG), moderate (2,000–3,999 EPG) and heavy (≥4,000 EPG). T. trichiura infection was categorized as light (1–999 EPG), moderate (1,000–9,999 EPG) and heavy (≥10,000 EPG) based on WHO guidelines [31].
Data were entered into Microsoft Excel 2007 spreadsheets (Microsoft Corp., Redmond, WA, USA) and exported to the statistical package R version 3.2.4 where all statistical analyses were performed [28]. For the final analyses, we included 368 children with complete parasitological and questionnaire data. Since intensity yields morbidity details, unlike prevalence, we tested the association between the intensity of helminth infections and fixed factors. The intensity of helminth infections was expressed as the arithmetic mean of number of slides examined per child. For school mean intensity, we showed the intensity of helminth infections as the geometric mean. Infection intensity is a statistic that measures the estimation of worm numbers per person. Due to the over dispersion of egg counts, such data are well described using the negative binomial probability model [32].
To identify factors associated with the intensity of S. haematobium and hookworm infections, we fitted negative binomial generalized linear models (NB-GLM) using MASS library for bivariate analysis. For multivariate analysis, we employed the glmmADMB library of the R statistical package [28] to fit negative binomial generalized linear mixed models (NB-GLMM). School variable was included in the NB-GLMM as a random factor to control for the influence of living conditions. For factors associated with the intensity of S. haematobium infection, the response (i.e., dependent variable) was the average count of S. haematobium eggs/10 ml urine for each child. Fixed factors (covariates) included age, sex, SES, female household head’s education level, last deworming, bathing/washing in river, house distance from the river, religion and school as a random factor (to control for variability in living conditions). To identify factors associated with the intensity of hookworm infection, the mean hookworm EPG for each child was the response variable. Covariates included age, sex, SES, female household head’s education level, last deworming, religion, shoe wearing habit and school as a random factor. A map of the spatial distribution of the intensity of S. haematobium infection was developed using QGIS version 2.12.3 [33]. The shortest distance from a participant’s house to the river was measured using the QGIS software. Spatial clustering of the intensity of S. haematobium and hookworm infections was determined by SaTScan software version 9.4 [34]. To detect high and low clusters we applied normal model to the log (N+1) transformed egg count.
The scientific steering committee and the ethical review board of Kenya Medical Research Institute (SSC No. 2084) authorized this study. The ethical review board of Nagasaki University, Institute of Tropical Medicine, Japan (No. 140829127) also approved this study. Before the commencement of field activities, meetings were held with parents/guardians, school administrators and teachers to discuss the purpose and procedures of the study. We also informed relevant district education and health officers of the research. Parents/guardians consented in writing while children assented to the study before enrollment. On completion of sample analysis, a clinician treated all children infected with schistosomes with 40 mg/kg of praziquantel and those infected with STHs with 400 mg of albendazole in the six schools according to WHO guidelines [31].
The age range of the children was 8–18 years with a median of 12 years. There were 186 girls (50.5%) and 182 boys (49.5%). Over half (59.8%) of the children were from poor or worse off SES households. A majority (86.1%) of female household heads had not completed primary school education. Islam was the most frequently reported religion (78.5%) in the study area. Table 1 shows characteristics of the study participants. The median shortest distance from participants’ houses to the river was 1,295 meters (range, 33–6,249 meters). Approximately one-third (32.1%) of the children had a daily river water contact. Over half (52.2%) of the children did not wear shoes while outdoors.
The overall prevalence of at least one helminth infection was 46.2% (95% CI: 41.1–51.3), ranging from 14.7% to 80.0% in the six schools. As indicated in Table 2, the prevalence of S. haematobium, hookworm and T. trichiura infection was 33.2% (95% CI: 28.3–38.0), 26.1% (95% CI: 21.6–30.6) and 1.6% (95% CI: 0.3–2.9), respectively. We did not observe any cases of S. mansoni and A. lumbricoides infections.
The geometric mean egg count for S. haematobium was 2.0 eggs/10 ml urine (range, 0.4–8.0 eggs/10 ml urine) in the six schools. Among children infected with S. haematobium, 13.3% (95% CI: 9.8–16.8) showed heavy egg numbers in their urine. The geometric mean of hookworm eggs was 2.2 EPG (range, 0–16.5 EPG). The majority of hookworm cases were light infections. All five positive cases for T. trichiura were light infections.
In bivariate analysis without controlling for school effects, the NB-GLM revealed the intensity of S. haematobium infection to be associated with SES, religion and last deworming. In reference to the least poor, the infection intensity was high in the very poor and less poor categories. Participants affiliated to Islam were more intensely infected than Christians. At school level, Islam was associated with high infection risk except in Vyogato. The children who were dewormed more than one year prior to the study had lower infection intensity compared to those dewormed within a year’s time. Details of bivariate analysis results are shown in Table 3.
On inclusion of school in the NB-GLMM as a random factor, the effects of last deworming became nonsignificant. High intensity of S. haematobium infection was associated with very poor P = 0.0208 and Muslim P = 0.0011 (Table 4).
The spatial distribution of the intensity of S. haematobium infection in the study area is illustrated in Fig 2. Generally, the geometric mean intensity (indicated in parentheses) was high among Muslims compared to Christians in all schools except Vyogato. The percentage of S. haematobium positive cases among Muslims (34.6%) was higher than positive cases among Christians (27.8%) but not significant (χ2 = 0.99044, P = 0.3196).
Spatial analysis revealed clustering of S. haematobium infection. A high risk cluster including 14 children with a radius of 620 meters was identified in Burani/Bahakanda. The mean of log (N+1) transformed egg count was 4.46 and 0.96 inside and outside the cluster respectively P = 0.001. A low infection risk cluster of radius 3050 meters including 102 children in Yapha and Dumbule was found. The mean inside 0.24 while the outside mean was 1.41, P = 0.011 (Fig 3).
The outcome of bivariate analysis on the association between the intensity of hookworm infection and the potential risk factors with the exclusion of school effects (NB-GLM) is displayed in Table 5. Participants who were dewormed more than one year prior to the study were more intensely infected than those dewormed within a year’s time P = 0.04951. Other factors which showed a significant association with high infection risk were; SES, latrine availability, religion and main source of drinking water. There was a significant low infection risk among participants who were residing far away from the river.
In the NB-GLMM on inclusion of school random factor, the effects of latrine, religion and source of drinking water became non-significant. SES, sex and distance to the river were significantly associated with the intensity of hookworm infection. Table 6 details the final predictors of hookworm infection risk.
In Fig 3, a significant high risk cluster for hookworm infection was singled out in Burani and Bahakanda (radius 2470 meters and included 103 children). The respective mean of log (N+1) transformed egg counts inside and outside was 3.07 and 0.42, P = 0.001. A low cluster of the intensity of hookworm infection was identified around Amani, Yapha and Dumbule schools. All children in the three schools except 14 children in Amani were included. The mean of log (N+1) transformed egg counts inside the low cluster was 0.095 while outside mean was 2.23, P = 0.001.
The current study demonstrates both S. haematobium and hookworm infections are significant public health problems in Kwale. Identifying local risk factors is essential for expediting disease control by targeting high-risk groups or by informing possible intervention strategies to stakeholders involved in helminthiasis control. The proportion of schoolchildren infected with S. haematobium was 33.2%. This result is consistent with recent reports preceding this study [35,36]. However, this frequency is lower than that in the 1980s [37,38], when the prevalence was >70% among school-aged children in Kwale. Among STHs, hookworm is the most common in Kwale, and this finding corroborates past studies [36,39].
Clustering of the density of S. haematobium infection was evident in the micro-geographical study. The proportion of children with heavy infection intensity was lower in Dumbule and Yapha compared to other schools in the study area. Focality of schistosomiasis even in small-scale geographical settings is a well-known phenomenon [13–17]. Notably, the two schools with lower infection density were located in a dry area compared to the other schools. Such environments are not suitable for the propagation of intermediate host snails of schistosomiasis.
In NB-GLMM analysis, children from Muslim households excreted large numbers of S. haematobium eggs. The school was included in the model as a random factor to adjust for environmental effects, with the assumption that children attending a given school were clustered around the school. On stratification of the study population by the school, the intensity of S. haematobium was consistently high among Muslims in all schools except Vyogato, where the infection intensity appeared to be similar both in Muslims and Christians. The high intensity of S. haematobium among Muslims compared to Christians is of great interest. A further investigation of the underlying religion determinants of our observed difference in the intensity of S. haematobium in this population based on religious affiliation is necessary. A past study in Kwale indicated Muslims had lower participation in control and related operational research for urogenital schistosomiasis and soil-transmitted helminthiasis by 50% compared to Christians [40]. We could not clarify the reason why Muslims only showed higher intensity of S. haematobium infection than Christians but not both intensity and prevalence. Health seeking behavior can be influenced by religious or cultural beliefs [41]. A few heavily infected individuals can maintain the transmission of schistosomiasis. To effectively control schistosomiasis, identifying and targeting such heavily infected individuals is critical. Furthermore, profiling such heavily infected individuals can help us understand the disease epidemiology in endemic regions. The density S. haematobium infection was high among children from households with low SES. This finding is in agreement with a former study [42]. The relationship between low SES and infection intensity could be attributed to the correlation between low SES and poor sanitation and inaccessibility to safe water. We did not observe a relationship between the intensity S. haematobium infection and sex, contrary to past studies [39,43,44] but in agreement with an earlier study in coastal Kenya [45]. Past studies also found association of Schistosoma spp. infection with proximity to open water sources [13–17], there was no relationship between the intensity of S. haematobium infection and house distance to the river in Kwale setting.
Cluster analysis revealed a low hookworm infection cluster around Amani, Dumbule and Yapha schools. Notably, these schools are located in a drier area compared to the rest of schools in the study site. Transmission of hookworm is least likely to be supported in such dry conditions.
The final predictors of hookworm infection risk were: SES, sex, and distance to the river. Higher intensity of hookworm infection was observed among the poorer categories i.e., the less and most poor compared to the least poor group. Poverty is associated with multiple factors such as the absence of concrete floors in home dwellings, inadequate sanitation and lack of access to ant-helminths. Such factors can promote hookworm infection risk [46]. The hookworm infection risk among boys was lower compared to girls. Our findings are contradictory to the past studies where higher infection has been observed in boys [39,47,48]. The risk of hookworm infection risk declined with increased residence distance from the river. This can be explained by the survival rate of the infective larval stage that depends on the presence of optimal soil humidity and temperature conditions [49]. The three schools with low hookworm risk were in a drier area with high mean distance of residence from the river. History of anthelmintic treatment was marginally associated with the intensity of hookworm infection in Kwale. Individuals are prone to reinfection especially when chemoprophylaxis is the only strategy for hookworm control. Hookworm infection density was lower among children who received anthelmintic medication within 1 year before our study. This is in agreement with a study in Uganda where hookworm infection intensity was lower among participants who reported anthelmintic treatment within the last 6 months [48]. To expedite the control of hookworm infection in our study setting, intensified preventive chemotherapy strategies, i.e., increased frequency and coverage, are necessary. Past studies observed an association between hookworm infection and age [50–53]. However, in this study, hookworm infection intensity was not associated with age.
We acknowledge some limitations of our study. First, we only inquired about river water contact not considering other potential sources of schistosomes. Second, we did not investigate the duration of contact with water infested with Schistosoma larvae. Quantifying contact activities with infested water is necessary to assess the contribution of water contact behavior to schistosomiasis in endemic regions [54]. Third, use of a questionnaire to gather information on past deworming history is subject to recall bias. Finally, the study participants could easily confuse other medications taken in the past to be anthelmintic treatment.
Both S. haematobium and hookworm infections showed micro-geographical heterogeneities in this Kwale community. To confirm and explain our observation of high S. haematobium risk among Muslims, further extensive investigations are necessary. The observed small-scale clustering of the S. haematobium and hookworm infections might imply less uniform strategies even at finer scale for efficient utilization of limited resources.
|
10.1371/journal.pgen.1006404 | Multiple Origins of the Pathogenic Yeast Candida orthopsilosis by Separate Hybridizations between Two Parental Species | Mating between different species produces hybrids that are usually asexual and stuck as diploids, but can also lead to the formation of new species. Here, we report the genome sequences of 27 isolates of the pathogenic yeast Candida orthopsilosis. We find that most isolates are diploid hybrids, products of mating between two unknown parental species (A and B) that are 5% divergent in sequence. Isolates vary greatly in the extent of homogenization between A and B, making their genomes a mosaic of highly heterozygous regions interspersed with homozygous regions. Separate phylogenetic analyses of SNPs in the A- and B-derived portions of the genome produces almost identical trees of the isolates with four major clades. However, the presence of two mutually exclusive genotype combinations at the mating type locus, and recombinant mitochondrial genomes diagnostic of inter-clade mating, shows that the species C. orthopsilosis does not have a single evolutionary origin but was created at least four times by separate interspecies hybridizations between parents A and B. Older hybrids have lost more heterozygosity. We also identify two isolates with homozygous genomes derived exclusively from parent A, which are pure non-hybrid strains. The parallel emergence of the same hybrid species from multiple independent hybridization events is common in plant evolution, but is much less documented in pathogenic fungi.
| The genus Candida is one of the leading causes of fungal morbidity in humans. Many pathogenic Candida species are diploid, and do not have have a full sexual cycle. The evolutionary origin of Candida orthopsilosis is unclear. Here, we use whole genome sequencing of 27 C. orthopsilosis isolates from around the world to show that C. orthopsilosis arose from hybridization (or mating) of two distinct parental species. Unusually, the hybridization event did not occur only once; we identify at least four events, and we suggest that hybridization is ongoing. The “species” C. orthopsilosis therefore does not have one single origin. We have identified one of the parental lineages involved, but the other remains elusive. Our results suggest that inter-species hybridization has an evolutionary advantage. However, unlike in plant pathogens, it does not appear to result in increased virulence of C. orthopsilosis.
| Hybridization or mating between different species can promote the emergence of new species by creating extreme (transgressive) phenotypes allowing adaptation to new ecological niches [1]. In the human fungal pathogen Cryptococcus neoformans, hybridization has been associated with phenotypic evolution and increased virulence [2, 3], and in plant fungal pathogens hybridization is associated with increased host range and the emergence of new species [4–6]. Hybridization is particularly common in yeast species used in the preparation of food and drink, such as Zygosaccharomyces and Saccharomyces [7–9]. Natural hybrids between many of the members of the Saccharomyces species complex have been identified [10, 11]. For example, Saccharomyces pastorianus formed at least twice from recent hybridizations between Saccharomyces cerevisiae and Saccharomyces eubayanus, and this event has been associated with the acquisition of cold tolerance in the lager yeast [12–14].
Homoploid hybrid speciation (without an increase in chromosome number) can lead to the formation of new species, for example in natural populations of Saccharomyces paradoxus [15]. Polyploidization was probably important for speciation of up to 1/3 of plants, and has been reported in both plants and animals [16]. The increased use of whole genome sequencing has made it relatively easy to identify hybrids and to study their genome evolution at high resolution [9], and indeed recent evidence suggests that the whole-genome duplication in the S. cerevisiae lineage arose from an ancient hybridization between two closely related species [17].
Here, we investigate hybridization in members of the yeast CTG-Ser clade (species that translate the codon CTG as serine and not leucine [18]). Several of these species are human fungal pathogens, including Candida parapsilosis, which is particularly associated with infections of neonates [19–21]. The C. parapsilosis sensu lato species complex consists of three defined species: C. parapsilosis sensu stricto, C. orthopsilosis and C. metapsilosis [22]. C. parapsilosis sensu stricto is the most frequently isolated from human infections, followed by C. orthopsilosis (up to 26% of C. parapsilosis sensu lato isolates) and C. metapsilosis (up to 11% of C. parapsilosis sensu lato isolates) [23, 24]. There is however a large variation in the frequency of isolation of the individual species, which may be related to geographic region. Several studies fail to identify any C. metapsilosis isolates [23, 24], whereas in a 12-year study in Taiwan, approximately equal numbers (10%) of C. parapsilosis sensu lato isolates were identified as C. orthopsilosis and C. metapsilosis [25]. A recent study in Chinese hospitals identified more C. metapsilosis than C. orthopsilosis isolates [26]. The C. parapsilosis sensu lato species vary significantly in virulence and drug susceptibility, with C. parapsilosis being the most virulent, followed by C. orthopsilosis and C. metapsilosis [25, 27, 28].
C. parapsilosis sensu lato species are obligate diploids, and mating and meiosis have never been observed [29–31]. The level of heterozygosity in C. parapsilosis sensu stricto isolates is much lower than in other CTG clade species [29, 32–34]. For example, SNP frequency in one sequenced C. parapsilosis isolate is approximately 1 SNP per 15 kb, which is 70 times lower than in the related species Lodderomyces elongisporus [29]. Low levels of heterozygosity were confirmed by sequencing three additional genomes, though some copy number variations were identified [34]. In addition, all C. parapsilosis sensu stricto isolates characterized to date contain only one mating idiomorph (MTLa) at the Mating-Type Like locus, and MTLa1 is a pseudogene [31]. Genome structure in C. metapsilosis however suggests a different evolutionary history in that species. Sequencing genomes of 11 clinical C. metapsilosis isolates showed that they were all highly heterozygous, and most likely resulted from hybridization between two parental species that differed by approximately 4.5% at the genome level [35]. Although earlier analysis suggested that C. metapsilosis isolates contained only MTLα idiomorphs [31] genome sequencing revealed that a second idiomorph was formed by introgression at MTLa generating a chimeric locus, containing the MTLa regulatory genes a1 and a2, and MTLα2 [35]. The authors suggested that a single ancient interspecies hybridization event was followed by global expansion of C. metapsilosis and loss of heterozygosity [35]. In C. orthopsilosis, AFLP (amplification fragment length polymorphism) analysis and sequencing of ITS sequences identified some heterogeneity among isolates, suggesting the presence of at least two sub-groups [31, 36–38]. This was supported by our identification of two MTLa and two MTLα idiomorphs in 16 C. orthopsilosis isolates which differed by approximately 5% [31]. Some isolates were heterozygous at MTL, and we suggested that the two different MTLa/α combinations represented two distinct subspecies, named Type 1 and Type 2.
Sequencing of a putative C. orthopsilosis Type 2 genome (isolate 90–125) showed that it is highly homozygous, similar to C. parapsilosis [29, 39]. However, further studies identified two highly heterozygous isolates, which were suggested to result from the same hybridization event, possibly between Type 1 and Type 2 parents [40]. Here, we carried out a population genomics analysis of 27 worldwide C. orthopsilosis isolates. We report that most C. orthopsilosis isolates are hybrids most likely formed by mating between two parental species that are about 5% different in sequence, followed by loss of heterozygosity (LOH) to form mosaic genomes.
Although some aspects of C. orthopsilosis evolution are remarkably similar to the recently described structure of C. metapsilosis populations [35], we show that C. orthopsilosis has arisen from at least 4 distinct hybridizations between the same two parental species, whereas all known C. metapsilosis strains derive from a single ancestral hybridization. We propose a model for C. orthopsilosis hybrid origins that places Type 1 and Type 2 strains in the hybrid context, and shows that Type 1 and Type 2 are not useful descriptions of the parental species, but are reciprocal combinations of mating partners. The existence of recombinant mitochondrial genomes and a non-hybrid “Parent A” lineage indicates that the formation of C. orthopsilosis by hybridization is recent and probably ongoing. Recurrent hybridization has been shown to lead to increased virulence of some plant and animal fungal pathogens [2, 4, 41–44], but our study is the first to show that it is also occurring in the Candida clade.
We sequenced the genomes of 27 C. orthopsilosis clinical isolates from around the globe, including Europe, US and Asia (Table 1). Four isolates were sequenced at >400X coverage with the reminder at >70X with Illumina technology. One isolate (Sample 427) was also sequenced using PacBio technology, which was used to characterize genome structure. Previously sequenced isolates 90–125 and MCO456 [39, 40] were included in the subsequent analysis. We identified an assembly error in the 90–125 reference genome, an artefactual translocation between chromosomes 2 and chromosome 6 (S1 Fig). The corrected assembly has been submitted to GenBank (BioProject number PRJEA83665). A description of the main findings from the genome data, including an analysis of copy number variation, is provided in S1 File and is summarised in Table 1.
Homozygous and heterozygous single nucleotide polymorphisms and insertions and deletions relative to the reference genome 90–125 were identified as described in Methods (Fig 1). Only one strain has a highly homozygous genome similar to 90–125 (Sample 428, <4,000 heterozygous SNPs, mostly derived from incorrectly assembled regions in the reference strain). In contrast, the heterozygosity levels of the remaining isolates varied from approximately 100,000 (Sample 1799) to 400,000 (Sample 498) heterozygous sites, and are much more similar in number to isolate MCO456 described by Pryszcz et al [40].
Pryszcz et al [40] suggested that high levels of heterozygosity in C. orthopsilosis MCO456 and in a related isolate AY2 reflects their origin from a hybridization between two related, but different, species or sub-species, where one parent is highly similar to 90–125. This proposal was supported by their observation that there is a bimodal distribution of differences between homozygous regions of MCO456 and 90–125. Approximately 41% of the MCO456 genome is similar to 90–125 (<1.8% divergence), and approximately 32% is different (>1.8% divergence). We carried out a similar analysis of our data, and we found that all the heterozygous isolates exhibited a similarly bimodal pattern of differences to 90–125 (Fig 1B, S2 Fig). This result indicates that the majority of C. orthopsilosis isolates arose from hybridization between two parents. The average nucleotide difference is 5.1%. One of the parents is very similar to 90–125.
To further characterize the parental lineages, we identified the regions of each genome derived from each parent, and used these in a phylogenetic reconstruction. We first identified highly heterozygous regions (defined as 1 kb regions that were heterozygous in at least 20 isolates, see Methods). We then identified the homozygous and heterozygous SNPs from these regions and assigned them to haplotypes. Those that were identical to isolate 90–125 were assigned to haplotype A, and those that were different were assigned to haplotype B. Maximum likelihood trees were generated separately from each haplotype using RAxML [45] (Fig 2). When all SNPs are considered together, the isolates fall into only two groups (Fig 2A), suggesting that they all arose from the same two parents, where haplotype A is derived from one parental species, and haplotype B from the other.
The phylogenetic trees derived independently from SNPs in each of the two haplotypes are very similar (Fig 2). The isolates fall into 4 major clades, with the two homozygous isolates forming outgroups to Clades 1 and 3 (Sample 428) and Clade 4 (90–125). Clade 4 is somewhat divergent, and for discussion we divide it into sub-clades 4.1 and 4.2, with a single isolate (Sample 498) in subclade 4.3. The overall sequence divergence in the trees for haplotype B is almost twice as large as for haplotype A (indicated by scale in Fig 2).
We next assigned all sections of each genome (in 1 kb windows) into one of four categories: homozygous haplotype A, homozygous haplotype B, heterozygous (AB), and undefined, based on the distribution of homozygous and heterozygous SNPs relative to 90–125 (Fig 3, see Methods). Loss of heterozygosity (LOH) events (homozygous A or homozygous B) are shared by members of the same clade, though some events are specific to individual isolates within each clade. Clade 1 isolates have the highest amount of LOH, followed by isolates in Clade 2, 3 and 4. The majority of large LOH events occurred towards telomeric regions, particularly in the most heterozygous isolates in Clade 4.
To test whether A and B are the only two haplotypes present in the sequenced C. orthopsilosis isolates, we investigated whether there is any evidence for genomic regions that originate from a third source. To do this, we used the 90–125 strain as a reference for haplotype A, and inferred an almost-complete (84.6%) reference for haplotype B from sample 424, which has the lowest amount of homozygous haplotype A regions. We then repeated the SNP analysis looking for genomic regions that are divergent from both the A and B references (S3 Fig). The majority of the genomes are similar to either haplotype A or haplotype B, except for regions where haplotype B cannot be inferred (S3 Fig). Some short regions on chromosomes 7 and 8 in some isolates differ from both haplotypes A and B, but this is an artifact of short LOH events (< 500 bp) in sample 424 (S3 Fig). We therefore conclude that all heterozygous isolates descended from the same two parental species, A and B.
For the heterozygous isolates we determined the contribution from each parent by calculating the percentage of each strain’s genome that is homozygous AA, homozygous BB, heterozygous AB or undefined (Fig 4, see Methods). Most isolates have approximately equal contributions from haplotype A and haplotype B genome-wide. Isolates with the biggest difference include Sample 498 (12% more haplotype A than B) and Sample 436 (11% more haplotype B than A). Overall however, LOH events appear mostly random in C. orthopsilosis isolates with little preference towards one or the other parental species.
To determine if haplotypes A and B underwent recombination we took advantage of the PacBio data from Sample 427. We restricted the analysis to five regions where two haplotypes were assembled, ranging in size from 20 kb to 64 kb. In all five cases one contig from the PacBio assembly matched 90–125 (the A genome), and the other represented the B parent, indicating that no recombination occurred, at least at these regions in Sample 427 relative to 90–125 (S1 Table, S4 Fig).
We previously characterized the Mating-Type-Like (MTL) locus of 16 isolates of C. orthopsilosis and showed that both the MTLa and MTLα idiomorphs occurred in two types [31]. The types diverged in sequence by approximately 5%, for both MTLa and MTLα. The majority of isolates in that study were homozygous for either MTLa or MTLα and only two were MTLa/α heterozygotes. In that study, we assumed that these MTL heterozygotes resulted from mating between isolates in a single type or group; the MTLa and MTLα idiomorphs from isolate J981224 were designated as Type 1, and those from isolate CP125 (Sample 425) as Type 2. The 5% sequence divergence between the Type 1 and Type 2 MTL idiomorphs is similar to what we now observe between haplotypes A and B at other loci.
We analyzed the MTL idiomorphs in our genome sequences (which included 14 of the strains studied by Sai et al [31]). Six isolates are MTLa/α heterozygotes (S5 Fig). Analysis of adjacent SNPs shows that MTLα Type 1 and MTLa Type 2 are both in physical linkage with haplotype A in different strains (S5 Fig). Similarly, MTLα Type 2 and MTLa Type 1 are both linked to haplotype B. Isolates that are homozygous at MTL (either MTLa/a or MTLα/α) have undergone LOH, and SNP analysis shows that these tracts of LOH extend into the regions flanking the MTL locus itself so that both chromosomes are derived from the same parental haplotype (e.g. Samples 423 and 426 are MTLα/α homozygotes derived from haplotypes A and B respectively by LOH; S5 Fig). Some of the isolates heterozygous at MTL show a small amount of LOH in a region to the right of the MTL (Sample 425; S5 Fig).
We can now see that the MTLa/α heterozygotes designated “Type 1” by Sai et al [31] (such as Sample 498) are heterozygotes containing MTLα from haplotype A and MTLa from haplotype B. The heterozygotes that were designated as “Type 2” (such as Sample 425) have the reciprocal combination of MTLa from haplotype A and MTLα from haplotype B. Therefore, the “Type 1” and “Type 2” labels for MTL heterozygotes represent the two complementary ways that cells from two putative parental species corresponding to haplotypes A and B could combine by mating. We refer to these parental species as ‘Parental Species A’ and ‘Parental Species B’. Although many Candida species are asexual, a parasexual cycle has been described in some diploid species [46]. Cells of opposite mating type hybridize to form a tetraploid, followed by concerted chromosome loss to regenerate a diploid [47, 48]. We propose that similar events occurred during hybridization of Parental Species A and Parental Species B during the evolution of C. orthopsilosis, though we cannot rule out the possibility that the ancestral species were fully sexual.
For isolates in Clades 1, 3 and 4, Parental Species A contributed MTLα, and Parental Species B contributed MTLa. Conversely, for Clade 2 isolates, Parental Species A contributed MTLa, and Parental Species B contributed MTLα (S5 Fig; Fig 5). This discrepancy in MTL genotypes indicates that Clade 2 and Clades 1/3/4 cannot be descendants of a single ancestral mating event between two cells from the two parental species, so at least two separate hybridization events occurred. By analysis of other datasets described below, we inferred a model for C. orthopsilosis evolution that postulates at least 4, and possibly 5 separate hybridizations between the parental species (Fig 5).
We assembled the mitochondrial genome sequences from the 27 sequenced isolates and from strain 90–125 [39], and compared them to the three previously published C. orthopsilosis mtDNAs [49, 50]. Twenty-eight isolates have linear mitochondrial genomes, and three have circular genomes (S2 Table). In the linear genomes, all the genes are located in a central 24 kb region, which is flanked on each side by a large inverted terminal repeat (Fig 6A). This repeat consists of a subterminal region (sub) followed by multiple tandem copies of a telomeric repeat (tel). The PacBio assembly of mtDNA from Sample 427 has nine complete copies of tel on the left arm, and six on the right, followed by an incomplete copy on both arms (100 to 103 bp). C. parapsilosis mtDNA molecules, which have a similar organization, have been reported to contain up to eight telomeric repeats on each arm [51].
The circular genomes in Samples 436 and 90–125 are caused by recombination at microhomologies in the subterminal repeats, similar to those previously described [49]. Although the Illumina assemblies of the linear mtDNAs are incomplete in the telomeric regions, there are at least two different types of subterminal region (451 and 396 bp) and two types of telomeric repeat (565 and 777 bp) in C. orthopsilosis (S2 Table).
The phylogenetic analysis showed that the C. orthopsilosis mitochondrial genomes belong to four mitotypes (Fig 6B). Isolates from three mitotypes (mt1, mt2 and mt3) correspond to the nuclear Clades 1, 2 and 3. We designated strain 90–125 as mitotype mt4 because this strain is closely related to nuclear Clade 4 and its mtDNA appears not to be recombinant (see below). There is almost no variation within the mitotypes (0–6 SNPs in the whole genome) whereas divergence between the mitotypes is significant (up to 222 nucleotide substitutions or 1.1% between mt2 and mt3).
In contrast to Clades 1–3, nuclear Clade 4 isolates have mtDNAs that fall at two distinct positions on the tree. Most belong to a single clade designated as clade mtR1, but Sample 498 forms a distinct lineage (mtR2) that is more closely related to mt4 (90–125) and to the mt1 and mt3 clades (Fig 6B).
The predicted phylogenetic relationship of mt1, mt2, mt3 and mt4 isolates is the same, irrespective of which region of the mtDNA is used for the phylogenetic analysis. However, the placement of the mtR1 and mtR2 isolates varies when different mtDNA regions are used to construct trees (Fig 6C–6F). This suggests that the mtDNA in mtR1 and mtR2 resulted from an inter-lineage recombination, as was previously proposed by Valach et al [50] for the strain MCO471 (mtR1). The mtR1 and mtR2 mtDNAs are both derived from recombination between the mt2 and mt4 mitotypes, but they were formed by two separate events because the recombinations occurred at different sites in the genome (Fig 6G and 6H).
Analysis of diagnostic SNP sites in mtDNA (Fig 6G) shows that in mtR1 the center of the mitochondrial genome (between rrnL and rrnS) comes from mt2 whereas both arms come from mt4. In contrast, in mtR2 only the right arm (from nad5 to the telomere) comes from mt2. The left recombination event in mtR1 can be mapped to within the rrnL gene for the large subunit rRNA, because mtR1 shares a 44 bp insert at the 5’ end of rrnL with mt2, but lacks the rI1 and rI2 introns that are present at the 3’ end of rrnL only in mt2. A further polymorphism in C. orthopsilosis mtDNA concerns the cox1 intron ai3, which is present only in the mt2, mt3 and mtR1 clades (S2 Table).
Analysis of the nuclear polymorphisms, MTL loci and mitochondrial genomes together shows that there were at least four independent hybridizations between Parental Species A and B (Fig 5). We infer that Parental Species A and B were both quite diverse, with multiple populations having distinct nuclear lineages (lineages B1 to B4 within Parental Species B; and lineages A1 to A4, 90–125 and 428 within Parental Species A) and distinct mitochondrial lineages (mt1-mt4). Each hybridization was a mating event that gave rise to one of the four nuclear clades. After each clade was formed by hybridization, it diversified (probably clonally, undergoing LOH), resulting in congruent SNP trees for its A and B subgenomes.
Because hybridization has been associated with increased virulence of the human fungal pathogen Cryptococcus neoformans [2] and has been postulated as a virulence mechanism in C. metapsilosis [35], we measured the virulence of C. orthopsilosis isolates using the model host Galleria mellonella (Fig 7). We identified substantial variation, ranging from avirulent to highly virulent isolates (Fig 7A). Isolates in Clade 2 have significantly reduced virulence compared to isolates in Clade 3 and Clade 4 (Fig 7B). However, we did not identify any correlation between levels of heterozygosity and virulence, or between heterozygosity and doubling time in rich media, either by comparing survival endpoints or by using Kaplan-Meier analysis (Fig 7, S6 Fig). Notably, one homozygous C. orthopsilosis isolate (Sample 428) is virulent (survival rate <25%), whereas the other (90–125) is not (Fig 7, S6 Fig).
All the isolates studied here would be classified as C. orthopsilosis based on their ribosomal DNA sequences, but our population genomics data shows that the name ‘C. orthopsilosis’ has been applied to two quite different types of isolates. A minority (2 of the 29 studied here) have ‘pure’ genomes that are simply Parental Species A. The first C. orthopsilosis genome sequenced, strain 90–125 [39] was fortuitously a strain of this type. In contrast, the majority of C. orthopsilosis isolates (93%; 27 of 29) are hybrids formed by mating between this species and a second species (Parental Species B) that is 5% different in genome sequence and that has not yet been isolated in a ‘pure’ (non-hybrid) form. This relationship is reminiscent of the beer yeasts, where a minority of strains are pure S. cerevisiae, the majority are interspecies hybrids (S. cerevisiae x S. eubayanus), and for a long time there was no known example of a ‘pure’ lineage of the S. eubayanus parent [12].
Our data show unequivocally that hybridization between the A and B parents of C. orthopsilosis occurred by mating, and that it occurred on at least four separate occasions. Since the pure A lineage was found in only 7% of isolates, and the pure B lineage was not found at all, this observation suggests that the interspecies hybrids are significantly more successful (i.e., more viable or more virulent) than their parents. Hybridization has been proposed to increase the virulence capacity of human fungal pathogens [2, 35] and the host range of plant fungal pathogens [4, 42]. However, our assays show no consistent correlation between virulence and heterozygosity in C. orthopsilosis, and a large difference in virulence between the two homozygous representatives of lineage A. Our results also indicate that C. orthopsilosis must be capable of mating, even though it has never been seen to mate in the laboratory.
The inference that multiple hybridizations occurred means that the species C. orthopsilosis does not have a single clonal origin. Although hybridization is common in yeasts, to our knowledge the only other known example of an ascomycete yeast species with multiple origins by parallel hybridizations is the beer yeast S. pastorianus. Multiple interspecies hybridizations have been reported in other pathogenic fungi (such as hybridizations between Cryptococcus neoformans and C. deneoformans, or between C. neoformans and C. gattii), but in these examples the hybrid lineages are a minority and the parental species are abundant and readily identifiable [2, 41]. We suggest that, for both C. orthopsilosis and C. metapsilosis, the parental species of these hybrids may be yeasts that are pathogens of other mammalian species and are not normally (or frequently) associated with humans. Formation of the hybrid may have facilitated a change in host range and pathogenicity to humans [2, 4, 35, 42]. Further investigation into geographical and ecological variation in the C. parapsilosis sensu stricto clade will be needed to understand the circumstances in which the parental species encounter one another and these pathogenic hybrids can emerge.
For Illumina sequencing, the 27 C. orthopsilosis isolates were cultured overnight in a shaking incubator at 30°C in 5 ml YPD medium (1% yeast extract, 2% peptone, 2% glucose). Genomic DNA was extracted using the Qiagen Genomic Tip-kit (20/G, product code 10223).
For PacBio sequencing, Sample 427 was grown in 100 ml synthetic complete media (2% glucose, 6.7% yeast nitrogen base, 2% Bacto agar, 0.2% dropout mix) overnight to reduce the carbohydrate concentrations. DNA was extracted using the Qiagen Genomic-tip kit (500/G, product code 10262) using a modified protocol for yeast (available at PacBio SampleNet, www.pacbiosamplenet.com).
C. orthopsilosis strains (Table 1) and C. parapsilosis CLIB214 were grown overnight in a shaking incubator at 30°C in 5 ml YPD medium (1% yeast extract, 2% peptone, 2% glucose). Yeast strains from overnight cultures were centrifuged and washed in Phosphate Buffered Saline (PBS, Oxoid) and diluted to 108 cells/ml for C. parapsilosis, and 5 × 107 cells/ml for C. orthopsilosis strains in PBS. G. mellonella in their final larval stage were obtained from Lifefoods Direct Ltd, Sheffield, UK, and stored at 15°C in the dark for use within 7 days from shipment. Twenty larvae, similar in size and weight, were used to analyze the virulence of each fungal strain, in two separate experiments. Larvae were injected with 10 μl of the diluted strains through the last left proleg, using insulin syringes. Untreated larvae and larvae injected with PBS were used as negative controls to assess the general viability and the effect of injection, respectively. After inoculation, larvae were placed into two petri dishes with filter paper (10 larvae per dish) and incubated at 30°C in the dark. Viability of larvae was monitored every 24 hours, for four days. To determine whether the larvae are alive or dead, they were gently touched with tweezers. If no movement was observed, larvae were considered to be dead [52]. Virulence was analysed by comparing the endpoint survival means (using one-way ANOVA with Bonferroni correction) and comparing the survival curves over time (by Kaplan-Meier estimate with log-rank test). The statistical analyses were performed using the SPSS Statistics software package and survival package implemented in R.
To compare doubling times, 28 C. orthopsilosis isolates were grown overnight in YPD broth in a shaking incubator at 30°C. Cultures were diluted to an A600 of 0.1 and incubated in a 96 well round bottom plate. A600 measurements were taken every 15 min over 48 h using a shaking plate reader (Biotek Synergy HT: Multi-Detection Microplate Reader), to generate growth curves. The doubling times of the isolates were calculated from three biological replicates, each with three technical replicates, using the software tool GATHODE [53].
Library preparation and Illumina sequencing of 27 C. orthopsilosis isolates in two Illumina HiSeq 2500 lanes (150 bp paired-end) was carried out by the DNA Core Facility, University of Missouri, USA. Five isolates (Sample 427, Sample 831, Sample 422, Sample 282 and Sample 320) were sequenced on one lane, and the remaining samples were multiplexed in the second lane. Raw read numbers ranged from 40.59 to 69.22 million reads for the first five genomes and from 6.19 to 13.17 million reads for the remainder (Table 1). Raw reads were downloaded for previously published C. orthopsilosis genomes 90–125 [39] (Illumina GAIIX, 75 bp single end) and MCO456 [40] (Illumina HiSeq 2000, 100 bp paired-end). For each isolate, raw reads were trimmed using Skewer v0.1.117 [54] with parameters—quiet (without progress updates) -m pe (paired-end mode) -l 36 (minimum read length allowed after trimming) -q 15 (trim 3’ end of read until a quality of 15 or higher is reached) -Q 15 (lowest mean quality for a read allowed before trimming) and -t 4 (number of threads).
Reads were mapped to the reference genome (90–125) using the mem algorithm from bwa (with -t 8 threads and default options), generating BAM files for each isolate. AddOrReplaceReadGroups from Picard Tools v1.82 from the Broad Institute [55] was used to add read groups to BAM files in order to pass requirements of Genome Analysis Toolkit (GATK) for BAM files. BAM files were indexed using Samtools [56].
The GATK HaploTypeCaller [57] (with -nct 42 threads and default options) was used to obtain a high quality set of single nucleotide variants (SNVs). Identified SNVs from all samples were merged using GATK CombineVariants. For the SNP analysis, insertions and deletions were removed using a custom script. Each genome was divided into 1 kb windows and assigned to haplotype A, haplotype B or haplotype A/B depending on the number of homozygous and heterozygous SNPs compared to the reference 90–125, using a custom R script. Regions were defined as homozygous A if they had <10 homozygous SNPs and <10 heterozygous SNPs, homozygous B if >10 homozygous SNPs and <10 heterozygous SNPs, heterozygous A/B if <10 homozygous SNPs and >10 heterozygous SNPs, and undefined (>10 homozygous SNPs, >10 heterozygous SNPs). Undefined regions ranged from 0.93% (Sample 424) to 4.81% (Sample 434), which correlates with the overall heterozygosity levels (Pearson correlation coefficient of 0.86). Undefined regions probably arise because the 1 kb windows used sometimes span the start or the end of a LOH event. These 1 kb regions then correspond partly to a heterozygous region and partly to a homozygous region.
To infer haplotype B, all 1 kb regions in Sample 424 defined as either haplotype B or haplotype A/B were extracted. For heterozygous SNPs, the base that differed from 90–125 was used. In total, SNPs from 10.71 Mb were extracted. For each extracted SNP the base in the 90-125-reference genome was substituted with the base from haplotype B. GATK HaplotypeCaller was used to call SNPs for all isolates against the inferred parent B. SNPs in 1 kb regions were binned as described above.
Mitochondrial genomes were identified as contigs in genome assemblies made using Platanus v1.2.1 [58] and annotated by reference to Kosa et al [49]. Linear mtDNAs assembled as contigs ending in telomeric (tel) repeats (Fig 6). The circular mtDNAs present in two strains assembled as contigs containing junctions between the left and right subtelomeric (sub) regions and lacking telomeric repeats. All 28 mitochondrial genomes that we sequenced have been submitted to the EMBL database (accession numbers LT594353-LT594380).
BAM files generated in the variant analysis step were used to characterize depth of coverage using the DepthOfCoverage tool in GATK with default parameters. Expected genome-wide coverage was calculated using the total number of mapped reads multiplied by the read length and divided by the size of the reference genome (90–125). Log2 ratios for copy number analysis for each isolate were calculated in 1 kb windows using the formula: log2 (observed coverage in 1 kb window / expected coverage) + log2 (total expected coverage / total observed coverage). Log ratios were then smoothed using the smooth.CNA() function of the DNAcopy package in Bioconductor (R package version 1.42.0). We used circular binary segmentation as implemented in the DNAcopy package to extract regions of equal copy number. All identified CNVs were manually verified.
The genomes were divided into 1 kb windows and, for each isolate, all SNPs (heterozygous and homozygous SNPs) in a specific 1 kb region were extracted if that region contained more than ten heterozygous SNPs in 20 or more of the analyzed isolates (excluding the two homozygous strains 90–125 and Sample 428). We identified 57530 SNPs from 1195 kb (9.43% of the genome). The SNPs were converted to a FASTA file and split into parental alleles (named A and B) using a custom script. In brief, if a base was identical to 90–125 it was assigned to A and an N was inserted for B; a homozygous SNP was assigned to B and an N was inserted for A; a heterozygous SNP was split into A (equal to 90–125) and B. If neither base of a heterozygous SNP matched 90–125 they were alphabetically ordered (A to T) and the first base was assigned to A and the second to B. RAxML v8.1.21 (raxmlHPC-PTHREADS) [45] was used to generate 20 maximum likelihood trees (options -m GTRCAT -p 12345) and 1000 bootstraps (-m GTRCAT -p 12345 -b 12345). Bipartitions were calculated by drawing all bootstraps onto the best maximum likelihood tree using RAxML (-m GTRCAT -p 12345 -fb). Trees were visualized using FigTree v1.4.2 (http://tree.bio.ed.ac.uk/software/figtree/)
To identify structural variations, including insertions, deletions and intra- and inter-chromosomal rearrangements, we utilized BreakDancer [59] with the BAM files generated in the variant analysis step. The script bam2cfg.pl from BreakDancer was used with 10000 random paired-end reads for each isolate to generate configuration files that list read length as well as lower, upper and mean insert size of the paired-end read fragments and its standard deviation. The insert size ranged from 412.01 to 442.29 bases, with a standard deviation between 95.09 and 109.56 bases. BreakDancer was then executed with default parameters and the results were analyzed manually.
For PacBio sequencing of Sample 427, SMRT-bells generation, quality control and sequencing on two SMRT Cells using P6-C4 chemistry was outsourced to GATC Biotech Ltd., Constance, Germany. PacBio SMRT Portal version 2.3.0.140936.p2.144836 was used for quality assessment of reads, generation of subreads, genome assembly using the Hierarchical Genome Assembly Pipeline (HGAP3) algorithm and AHA (A Hybrid Approach) scaffolding. The SMRT portal was locally modified to allow execution of commands on a single 48 core Linux server with 256 GB of memory. A total of 300584 polymerase reads were generated from two SMRT Cells, with a total read base count of 1.09 billion and a N50 of 18093 bases. After filtering (with parameters minimum subread length of 500, minimum polymerase read quality of 0.8 and a minimum polymerase read length of 100), 79850 polymerase reads with a total read base count of 894.95 million and a N50 value of 19514 bases were used to extract 165541 subreads with a total read base count of 875.25 million and a N50 of 6992 bases.
The subreads were used as input for the RS_HGAP_Assembly.3 protocol in the SMRT Portal (with parameters minimum seed read length of 6000, number of seed read chunks of six, alignment candidates per chunk of 10, total alignment candidates of 24 and minimum coverage for correction of six). The draft assembly contained 263 polished contigs with a N50 of 292.25 kb, a length of 17.14 Mb and a mean coverage of 46.62x. Scaffolding with the RS_AHA_Scaffolding.1 protocol using five iterations resulted in 241 scaffolds, a N50 of 322.40 kb, a length of 17.16 Mb and 22 gaps with a total length of 19.04 kb. The 34 longest scaffolds (ranging from 988.53 kb to 91.93 kb) had a total sum of 12.65 Mb, compared to a total sum of 12.66 Mb for the Co_90–125 assembly. The mitochondrial genome was represented in a single 33.78 kb contig.
|
10.1371/journal.pntd.0000501 | Co-authorship Network Analysis: A Powerful Tool for Strategic Planning of Research, Development and Capacity Building Programs on Neglected Diseases | New approaches and tools were needed to support the strategic planning, implementation and management of a Program launched by the Brazilian Government to fund research, development and capacity building on neglected tropical diseases with strong focus on the North, Northeast and Center-West regions of the country where these diseases are prevalent.
Based on demographic, epidemiological and burden of disease data, seven diseases were selected by the Ministry of Health as targets of the initiative. Publications on these diseases by Brazilian researchers were retrieved from international databases, analyzed and processed with text-mining tools in order to standardize author- and institution's names and addresses. Co-authorship networks based on these publications were assembled, visualized and analyzed with social network analysis software packages. Network visualization and analysis generated new information, allowing better design and strategic planning of the Program, enabling decision makers to characterize network components by area of work, identify institutions as well as authors playing major roles as central hubs or located at critical network cut-points and readily detect authors or institutions participating in large international scientific collaborating networks.
Traditional criteria used to monitor and evaluate research proposals or R&D Programs, such as researchers' productivity and impact factor of scientific publications, are of limited value when addressing research areas of low productivity or involving institutions from endemic regions where human resources are limited. Network analysis was found to generate new and valuable information relevant to the strategic planning, implementation and monitoring of the Program. It afforded a more proactive role of the funding agencies in relation to public health and equity goals, to scientific capacity building objectives and a more consistent engagement of institutions and authors from endemic regions based on innovative criteria and parameters anchored on objective scientific data.
| The selection and prioritization of research proposals is always a challenge, particularly when addressing neglected tropical diseases, as the scientific communities are relatively small, funding is usually limited and the disparity between the science and technology capacity of different countries and regions is enormous. When the Ministry of Health and the Ministry of Science and Technology of Brazil decided to launch an R&D program on neglected diseases for which at least 30% of the Program's resources were supposed to be invested in institutions and authors from the poorest regions of Brazil, it became clear to us that new strategies and approaches would be required. Social network analysis of co-authorship networks is one of the new approaches we are exploring to develop new tools to help policy-/decision-makers and academia jointly plan, implement, monitor and evaluate investments in this area. Publications retrieved from international databases provide the starting material. After standardization of names and addresses of authors and institutions with text mining tools, networks are assembled and visualized using social network analysis software. This study enabled the development of innovative criteria and parameters, allowing better strategic planning, smooth implementation and strong support and endorsement of the Program by key stakeholders.
| The World Health Organization (WHO) classifies diseases as Type I, Type II and Type III, which largely corresponds to Global, Neglected and Most Neglected diseases in the vocabulary of the international organization Medécins Sans Frontières (MSF) [1],[2]. Type I/Global diseases know no geographic boundaries while Type II–III/Neglected-Most Neglected are predominantly or exclusively prevalent among populations of developing countries. Types II and III diseases (from now on “neglected diseases”), being prevalent in poor regions, are not prioritized by pharmaceutical and biotechnological industries responsible for the manufacture of goods such as vaccines, drugs and diagnostic kits. This generates what is known as ‘market failures’ - the inefficient allocation of products and services through usual free market mechanisms.
Several procedures have been suggested to cope with the three types of “health failures”: (i) Science failures (insufficient knowledge prevents the development of health products such as malaria and HIV vaccines): Stimulate basic, fundamental research and technological development, (ii) Market failures (high prices prevent access of drugs by needy populations): Price reduction policies (resulting e.g. from negotiations between governments and industry) or creating subsidizing mechanisms leading to lower prices and (iii) Health service failures (inexpensive drugs do not reach the patients): Fighting corruption, reducing inequalities and coping with cultural, religious or infrastructure barriers, etc. that prevent access to cheap or free drugs by poor countries [3],[4].
Several initiatives have recently been proposed to stimulate research, technological development and production of vaccines, drugs and diagnostics for neglected diseases by both Big Pharma and Small Biotech of developed countries such as “Push” mechanisms, like Public Private Partnerships (PPPs) or Partnerships for the Development of Products (PDPs), funded in general by philanthropies or governments [5],[6] and “Pull” mechanisms, like Advance Market Commitments (AMCs), Orphan drug legislation (e.g. the US Orphan Drug Act of 1983) and Priority Review Vouchers issued under the Food and Drug Administration Amendments Act of 2007 (FDAAA).
These mechanisms have in general been conceptualized and implemented by the developed world and either international or philanthropic organizations. They do not take full advantage of the brainpower and infrastructure existing in middle-income developing countries or in some innovative developing countries (IDCs) [7] such as Brazil, where considerable progress has recently been made in defining and implementing a national policy for science, technology and innovation in health [8],[9],[10].
Research and development on neglected diseases is one of the key strategic areas of Brazil's priority agenda for health research [8],[11]. In 2005 the Ministry of Health together with the Ministry of Science and Technology, through their funding agencies DECIT (Department of Science and Technology, http://dtr2001.saude.gov.br/sctie/decit/index.htm) and CNPq (National Council for Scientific and Technological Development, http://www.cnpq.br/english/cnpq/index.htm), launched a joint Program to support research, technological development and innovation on six diseases that disproportionately hit poor and marginalized populations in Brazil: dengue, Chagas disease, leishmaniases, leprosy, malaria and tuberculosis. In 2008 schistosomiasis was added to the list and a 2nd call for applications instituted (http://www.cnpq.br/editais/ct/2008/034.htm). For additional detais on this DECIT/CNPq Program see Serruya et al [11],[12].
As equity and capacity building were considered critical components of the Program, it was decided to invest at least 30% of the financial resources in the three Brazilian geographic Regions where these diseases are still prevalent (North, Northeast and Center-West). Since the scientific productivity related to neglected diseases is less than in other areas of health sciences and several institutions located in these Regions are still maturing, traditional indicators such as number of scientific articles and impact factor of the journals where they were published would be of only limited value. We therefore decided to develop new approaches and criteria based on social network analysis [13],[14],[15],[16], to allow for a fair and efficacious allocation of resources without losing sight of scientific standards.
Publications by Brazilian authors on the seven diseases were retrieved as raw data files from the ‘Web of Knowledge’ database of the Institute for Scientific Information (ISI), a database that lists the full addresses of all authors of every paper. Queries were made in ‘advanced search’ mode directed simultaneously at the country name and at words in the titles of the papers, e.g. [CU = Brazil AND TI = (Chagas OR cruzi)] to retrieve papers with at least one researcher from Brazil among the authors and having “Chagas disease” or “Trypanosoma cruzi” in the title.
The ISI raw data files were imported into the text-mining software VantagePoint (http://www.thevantagepoint.com) with the appropriate ISI filters. A process of standardization was carried out to bring together the various different names of a particular author or institution [17] and VantagePoint thesauri for names and addresses were created in order to process additional name and address lists.
Co-occurrence matrices of authorship data were built into VantagePoint and exported to UCINET software for social network analysis [18]. A co-occurrence matrix shows the number of records in the dataset containing two given list items. Symmetrical, co-occurence matrices (also called ‘adjacency matrices’) were created using the same set of authorship data in rows and columns in order to map co-authorships between authors (authors×authors matrices) or institutions (institutions×institutions matrices). For additional details on the use of matrices in social network analysis see for instance Chapter 3 of Scott [19], “Handling Relational Data”. Networks were assembled, visualized and analyzed for several parameters such as network components and cut-points with the softwares NetDraw or Pajek [20] which are embedded in the UCINET package.
The scientific environment where the Program is based and operates can be assessed analyzing the scientific productivity of Brazilian authors and institutions in peer-reviewed international journals. Table 1 and Figure 1 display that it varies widely among the diseases covered by the Program, being for instance 5-fold greater for Chagas disease and leishmaniases as compared with dengue and leprosy.
Co-authorship network analyses were carried out at several stages of the two phases of the Programme: Phase I included six diseases, funding projects during the biennium 2007–2008 and the ongoing Phase II addresses seven diseases during the 2009–2010 biennium. We decided to focus our attention on network components and network cut-points, basic elements of social network analysis [19],[21] that generate visual information readily useful for Program managers and decision-makers. In this way we emphasized the generation of graphical displays over a purely quantitative, numerical analysis.
A component of a network is a portion of the network in which all actors are connected, directly or indirectly, by at least one tie (one co-authorship in the present work) [21]. Fig. 2 shows the component analysis of the 2001–2008 dengue co-authorship network, where 172 authors are distributed among 9 components, each one addressing in isolation its own set of specific, complementary or overlapping research topics and subjects.
A cut-point of a network is an actor (author or institution in our case) whose removal would increase the number of components by dividing the sub-graph into two or more separate subsets between which there are no connections. Cut-points are therefore pivotal points of articulation between the elements that make up a component [19]. The role of cut-points is exemplified in Fig. 3, which shows the 2006–2007 tuberculosis institutional co-authorship network with the cut-point nodes labeled and identified as red squares. In this network, for instance, the removal of the cut-point “Inst. Trop. Med. Prince Leopold” would disconnect FURG and IVIC from the network and the removal of the cut-point “UNICAMP” would do the same for the University of Illinois. The visualization of this network also demonstrates the power of graphic displays to rapidly detect and emphasize unique features of a given network. In this figure, the large agglomerate of nodes at the upper left immediately stands out, drawing one's attention to the presence of a publication involving a large number of coauthors and their institutions, an indicator of projects involving global networks.
Traditional scientific production indicators routinely adopted as criteria for evaluating scientific proposals and research funding programs, such as the number of publications in a given period of time and impact factor or H-index [22], have intrinsic shortcomings [23],[24] and are of limited value beyond ‘Mode 1’ of knowledge production (disciplinary, primarily cognitive, context) [25] or when the publication output of the work field or the scientific community under consideration is of small size. In fact a ‘Catch-22’ type challenge (a no-win situation or a double bind dilemma, see http://en.wikipedia.org/wiki/Catch-22) arises when considering these indicators to select candidates eligible for capacity building purposes, as the researchers and institutions most in need of support are exactly those who have a modest scientific curriculum or performance. Traditional evaluations therefore become a real barrier to career progress or towards institutional development.
The management of the DECIT/CNPq Program, having received the double mandate to adhere to high scientific standards and strengthen capacity in the less developed Regions of the country, as two pillars of the initiative, realized that new strategies and indicators would be needed. The 2001–2008 survey of publications shown in Tables 1 and 2 well illustrates some of the challenges the Program would face, for instance: (i) three out of the four most active research communities (Chagas disease, schistosomiasis and tuberculosis) are located in the developed South and Southeast of Brazil, far from the target Regions for capacity building and (ii) dengue, a disease that has caused serious problems for public health in recent years, has been addressed by one of the smallest scientific communities and needs ‘fast-track’ capacity building actions.
Two social network analysis tools proved to generate particularly valuable information for the strategic management of the Program, the identification of components and cut-points of the co-authorship networks:
Component analysis generates a picture of the overall network structure, revealing how fragmented it is and therefore providing valuable information on its status and opportunities for strategic management. As shown in Fig. 2 for the dengue researcher's network, the analysis of the work areas of the nine individual components, based on article keywords, suggested for instance, a collaboration between component III and VIII, as their researchers were all working on dengue vector control but did not engage in formal collaborations.
The identification of network cut-points became a very important analytical tool for the management of the Program, particularly in relation to its capacity building/strengthening mandate. As the majority of institutions in the less developed Regions still need to mature, a selection based exclusively on scientific productivity would place them at a clear disadvantage in comparison with sister institutions from the developed Southeast and South. We realized that institutions acting as network cut-points were critical key players as they were responsible for keeping several institutions from these Regions in the loop and should therefore be considered as fundamental partners for training, capacity building and institutional strengthening. This reasoning is supported by work in other fields that made evident the importance of scientists who play roles as brokers for communications among others [14], the function of nodes critically involved in connecting or bridging modular subregions of a network [26] or the cruciality of ‘creative elements’ in cells, social networks and ecosystems [27].
Table 2 shows that by adopting this cut-point criterion to help the selection of institutions worth strengthening, nine ‘cut-point institutions’ could be added to the eleven ‘top-10 institutions’ identified by classical high-productivity criteria. The Program could therefore double the number of potential investment targets in the North, Northeast and Center-West Regions with objective science-based parameters: the traditional, productivity-based indicators together with the new ones derived from the network analysis proposed in this article.
The Program was shaped to operate in ‘Mode 2’ of knowledge production (broader, transdisciplinary social and economic contexts) [25] as its mission goes beyond academic goals to also address capacity building, institution strengthening, product development, disease control and public health. In Brazil's national health system (SUS - Sistema Único de Saúde) the participation of civil society and communities is assured at all levels of government - federal, state and municipal [28]. The process leading to the prioritization of R&D on neglected diseases, which made possible the launching of the DECIT/CNPq Program and set its main objectives and goals, involved strong participation of these key stakeholders e.g. at the National Health Council (http://conselho.saude.gov.br/apresentacao/index.htm) and at the 2nd National Conference on Science, Technology, and Innovation in Health held in 2004 which involved 15,000 participants [8].
Mobilizing the scientific community, disease control managers and policy/decison-makers to collaborate under the umbrella of this initiative required a sort of ‘cultural change’ from everyone involved. For this purpose the process adopted by the Program included: (i) Organizing priority setting workshops with equal representation by researchers, policy/decision-makers and managers interested in the seven diseases of the Program; (ii) Adopting guiding principles such as burden of disease and classical/network-based science indicators as the basis for workshop agendas and discussions; (iii) Structuring these workshops on disease-specific working groups with equal representation of policy/decision-makers, managers and scientists of high productivity and/or affiliated to network cut-point institutions; (iv) Mobilizing the participation of scientific communities through ‘Call for Applications’ based on the recommendations of the working groups and published in the websites of the funding agencies; (v) Peer reviewing of the proposals taking into account the need to allocate a minimum of 30% of the funds to projects submitted by principal investigators affiliated to institutions in the North, Northeast and Center-West Regions.
Fig. 1 suggests that the DECIT/CNPq Program has been successful in stimulating scientific productivity on the six diseases in its first phase which did not include schistosomiasis as one of the targets. The future assessment of the full impact of the two phases, however, will need a thorough in-depth evaluation exercise based on input, output, outcome and impact indicators addressing scientific, technological and public health goals. Co-authorship network analysis has been employed to evaluate scientific journals [29],[30], institutions [31] and collaboration patterns in specific scientific fields [17]. The innovative contribution brought by this analytical approach during the shaping and implementation of the Program will be expanded and become critical when assessing the evolution, performance and robustness of the networks involved.
Our results also suggest that co-authorship network analysis could become an important tool for international organizations or partnerships targeting the elimination or eradication of diseases, providing science-based information relevant to strategic analysis and planning. Lessons from past eradication campaigns demonstrated the importance of maximizing the utilization of scarce human and financial resources, functioning within existing health service structures and encouraging research at all levels [32]. Applied to today's planned efforts towards the elimination/eradication of malaria [33],[34] or neglected tropical diseases [35], these lessons would mean identifying and engaging health services, researchers and institutions from developed and endemic countries, an immense challenge that co-authorship network analysis could help address, providing a substantial contribution to global health.
|
10.1371/journal.pntd.0006396 | Gene target selection for loop-mediated isothermal amplification for rapid discrimination of Treponema pallidum subspecies | We show proof of concept for gene targets (polA, tprL, and TP_0619) that can be used in loop-mediated isothermal amplification (LAMP) assays to rapidly differentiate infection with any of the three Treponema pallidum subspecies (pallidum (TPA), pertenue (TPE), and endemicum (TEN)) and which are known to infect humans and nonhuman primates (NHPs). Four TPA, six human, and two NHP TPE strains, as well as two human TEN strains were used to establish and validate the LAMP assays. All three LAMP assays were highly specific for the target DNA. Amplification was rapid (5–15 min) and within a range of 10E+6 to 10E+2 of target DNA molecules. Performance in NHP clinical samples was similar to the one seen in human TPE strains. The newly designed LAMP assays provide proof of concept for a diagnostic tool that enhances yaws clinical diagnosis. It is highly specific for the target DNA and does not require expensive laboratory equipment. Test results can potentially be interpreted with the naked eye, which makes it suitable for the use in remote clinical settings.
| Sustainable eradication of human yaws benefits from applicable and reliable assays to detect possible reemerging yaws cases. Our study provides proof of concept for LAMP assays that are capable of rapid diagnoses and discrimination of active Treponema pallidum infection. While current clinical diagnosis is based on the clinical manifestations in combination with serology, the selected targets and LAMP assays allow for DNA based differentiation between skin ulcers caused by the subsp. pallidum (syphilis), subsp. pertenue (yaws), and the subsp. endemicum (bejel). The presented LAMP assays require limited expensive technical equipment and can be run in virtually any clinical setting. The method is thus capable of enhancing yaws diagnosis in particular in a low capacity environment.
| Human yaws is a tropical skin disease of children caused by the bacterium Treponema pallidum subsp. pertenue (TPE) [1]. Skin ulcers are the most characteristic clinical manifestations associated with infection in all three active disease stages (primary, secondary, and tertiary yaws) [2]. The disease is currently subject to global eradication efforts [3], which face challenges that arise from the biology and distribution of the yaws bacterium as well as its diagnosis and treatment [4]. It is largely believed that the first yaws eradication campaigns conducted in the mid-1950s to late 1960s were successful in terms of reducing the prevalence by 95% but failed to eradicate the disease when local efforts to prevent new cases proved insufficient [5]. The majority of affected populations belong to poor and marginalized societies, with only rudimentary access to health care systems (‘Where the road ends, yaws begins’) [6]. Until today, standard diagnosis of yaws in clinical settings is based on clinical manifestations in combination with serology [1]. T. pallidum (TP) elicits a strong antibody response [7, 8]. Although it is possible to distinguish current infection (active or latent) from past infection when non-treponemal and treponemal tests are used in combination [9], it remains impossible based on serology and in some instances clinical manifestations, to differentiate yaws infection (TPE) from syphilis (caused by subsp. pallidum (TPA)) or bejel (caused by the subsp. endemicum (TEN)). Moreover, it has been shown that other diseases are capable of mimicking yaws infection. In particular, Haemophilus ducreyi has been reported to cause yaws-like skin ulcers [10]. Lastly, a larger number of skin ulcers in rural Africa remains etiologically undiagnosed [11], which increases the chance of overlooked infection with TPE. Other diseases which are capable of mimicking yaws infection are cutaneous leishmaniasis, scabies, or fungal infections [1]. Eradication of yaws is further challenged by the finding that nonhuman primates (NHPs) are infected with TP [12, 13]. Notably, all whole genome sequenced simian strains must be considered TPE strains [14, 15]. NHPs therefore must be considered as a possible natural reservoir for human infection [13]. The West African simian TPE strain Fribourg-Blanc, which was isolated from a Guinea baboon (Papio papio) in the 1960s [16], caused sustainable infection when inoculated into humans [17].
Post-eradication surveillance following the currently ongoing mass-azithromycin treatment phase [4] would benefit from rapid and cost-effective molecular tests that are able to distinguish TPE infection [18] from infections with all other TP subspecies (TPA and TEN) and bacteria that are involved in tropical skin ulcers and which may fall together with TP seropositivity. Potentially a single overlooked yaws case would result in a failure of global yaws eradication. Loop-mediated isothermal amplification (LAMP) was first described by Notomi et al. in 2000 [19] and since then has been extensively used to improve infectious disease diagnostics [20]. The highly specific method recognizes the DNA target using six distinct sequences initially and four distinct sequences subsequently [19]. LAMP uses a DNA polymerase with high strand displacement activity to perform a fast running auto-cycling strand displacement synthesis. Reactions run at constant temperature (isothermal) and therefore do not require expensive technical equipment such as PCR cycling machines. Our objective was to identify suitable gene targets that can be used for LAMP assay design to rapidly distinguish between yaws infection, including simian strains, and syphilis or bejel.
DNA samples of human TPA laboratory strain Mexico A, Nichols, Seattle 81–4, SS14, TPE strain Gauthier, CDC-1, CDC-2, Samoa D, Sei Geringging K403, Kampung Dalan K363, as well as the simian TPE strain Fribourg-Blanc were obtained from rabbit-in vivo inoculation experiments (S.A. Lukehart and DS). These experiments were not directly associated with this study. DNA extracts from human TEN strain Bosnia A and Iraq B originate from whole genome amplified clinical samples (DS) that were not directly associated with this study. DNA from a TP-infected olive baboon (Papio anubis; 6RUM2090716) originates from a clinical sample collected for a different study at Ruaha National Park (RNP) in Tanzania in 2015 (DFG KN1097/3-1 (SK)). Details and further reference for each strain included into the study can be found in the Supplementary S1 Table. ‘Good Veterinary Practice’ rules were applied to all procedures where animals were handled.
Three different LAMP assays were designed. First, we generated a LAMP assay that is able to detect DNA of all three TP subspecies (TPA, TPE, and TEN). This assay served as an initial control and was designed for the use in NHPs where little is known about the TP subspecies that circulate in wild NHP populations. Second, a LAMP assay was designed to distinguish TPE strain infection from infection with TPA or TEN strains. Third, a LAMP assay that differentiates between infection with TPE or TEN and infection with TPA strains has been established. All LAMP reactions were run with four human TPA, six human TPE and two simian TPE strains, as well as two human TEN strains of known copy number (S1 Table). All tests were run as triplicates and included a DNA-free negative control.
Dilution series of target DNA were used to identify the analytic limits of detection for each of the specific LAMP reactions using appropriate strain material. 10-fold serial dilutions of the target DNA were applied to cover a range of at least five decimal powers, from the maximum of TP copy numbers (strain Nichols 10E+5, all other strains 10E+6) until 10E+0. Negative controls that contained no DNA and dilution steps that contained ≤10E+2 TP copies were run as at least six replicates. A StepOnePlus Real-Time PCR System (ThermoFisher Scientific) was used to run the reactions and to collect the data. Due to software restrictions, it was necessary to introduce a (neglectable) thermal cycling step into the protocol. Each LAMP run therefore encompassed continuous 40 cycling steps each consistent of 63°C for five seconds followed by 64°C for one minute and data collection.
LAMP reactions were performed in a volume of 25.0 μl using the Mast Isoplex DNA Kit (#REF67dnamp, Mast Diagnostica GmbH). According to the manufacture’s guidance, each reaction consisted of 12.5 μl of the kit’s 2x reaction mix, 1.0 μl Bst polymerase enzyme, 1.0 μl fluorochrome dye, and 2.0 μl of the primer mix. One microliter target DNA was included and distilled water (molecular grade) was used to top up the reaction volume until 25.0 μl were reached. All primers were heat pre-treated at 95°C for 5 min and immediately cooled on ice prior to adding them to the master mix. The primer mix contained 1.6 μM each FIP and BIP, 0.2 μM each F3 and B3, as well as 0.8 μM each LF and LB primer. All reactions were run on a MircoAmp Fast Optical 96-well reaction plate (#4346907, ThermoFisher).
Oligonucleotide primers were designed using the PrimerExplorer v5 Software (http://primerexplorer.jp/e/). Each LAMP primer set consisted of six oligonucleotide primers (Table 1). The design followed the description given by Yoshida et al. 2005 [21]. Briefly, a set of four primers (F3, B3, the forward inner primer [FIP], and backward inner primer [BIP]), which bind six loci of the target gene (F1, F2, F3, B1, B2, and B3) are necessary. The two inner primers (FIP and BIP) are a sequence combination of sense and antisense sequences of the DNA. This is essential for the priming in the first stage and the self-priming in the later stages. Therefore, FIP primers consist of the combination of sequences defined as F1c (c = complementary) and F2. Likewise, BIP primers are composed of primer sequences B1c and B2. To enhance amplification efficacy, two loop primers LF and LB were added to each of the LAMP primer sets. To confirm the specificity of the newly designed primers, we performed a search for orthologous sequences using BLASTn at the NCBI homepage (http://blast.ncbi.nlm.nih.gov/Blast.cgi).
The LAMP primer set ‘TP’ targets the polymerase I (polA) gene (TP_0105) of TP. The locus is highly specific for all TP subspecies [22] and has only one orthologue in the lagomorph infecting Treponema paraluisleporidarum ecovar Cuniculus. The latter is not capable of infecting humans [23, 24]. This locus therefore allows the reaction to become positive for DNA of any known TPA, TPE, or TEN strain (Fig 1A).
LAMP primer set ‘TPA/TEN’ targets the tprL gene (TP_1031) of TP. At this locus, a 278-bp long deletion exists that distinguishes known human TPA and TEN strains from human and simian TPE strains [25, S1 Fig]. This primer set was specifically designed to bind within the deletion part, which creates the specificity for TPA and TEN strains (Fig 1B).
LAMP primer set ‘TPE/TEN’ targets the T. pallidum TP_0619 gene, which has recently been investigated in a multilocus-typing study on TPE strains [26]. This locus has a 179-bp long sequence part that distinguishes known human and NHP TPE as well as TEN strains from TPA human strains (Fig 1C, S2 Fig). Primer sequence data of all three LAMP primer sets are listed in Table 1.
All TP strains used in this study were quantified using an established [27] but slightly modified TaqMan PCR (qPCR) targeting the polA gene. A dilution series of a plasmid containing the target amplicon was used as a standard curve from 10E+7 to 10E+0 copy numbers. Briefly, each reaction volumes contained 10.0 μl TaqMan Universal Mastermix II (no Uracil N-glycosylase, Applied Biosystems), 1.8 μl each 10 μM primer and probe, 3.6 μl molecular grade water (RNase-free; Qiagen), and one microliter of the target DNA. Samples were quantified using a StepOne Plus Real-time system with the following temperature steps: 50°C for two minutes, 95°C for ten minutes, followed by 50 cycles of 95°C for 15 seconds and 60°C for one minute. At the end of each cycle, fluorescence was measured. All samples and standards were run as triplicates.
LAMP performance as well as qPCR data were retrieved from the StepOnePlus Real-Time PCR System and extracted as RAW data into Excel sheets utilizing the StepOne Software v2.3 (Life Technologies). Statistical analyses were performed with Prism 7.0 (GraphPad Software). In LAMP dilution series with low copy numbers (≤10E+2) and in qPCR data, single replicate outliers were excluded.
The LAMP assay targeting the polA gene was positive for all tested TP strain samples including the four TPA, six human TPE, two simian TPE, and the two human TEN strains (Fig 1A, S1 Table). The tprL targeting LAMP was positive for all tested TPA and TEN strains, while human and NHP TPE strains did not amplify (Fig 1B). The LAMP assay that uses a part of the TP_0619 gene generated positive results for all TPE strains including simian TPE strains as well as the two human TEN strains (Fig 1C). The onset of exponential fluorescence increase (ΔRn) started reproducibly between 5 min and 15 min incubation time (Fig 1A–1C). Melting curves for each LAMP assay are shown in S3 Fig. All curves were of appropriate shape and without any additional peaks indicative for unwanted side products of primer dimers.
Analytic limits of detection were assessed as demonstrated in several published studies [28–30]. The LAMP assay that targets the polA locus amplifies all TP strains but differed slightly in its detection limit across the different TP subspecies. While the TPA strain Nichols failed to amplify between 10E+3 and 10E+2 copies (Fig 2A), the TPE strain Gauthier showed a non-exponential increase in fluorescence at 10E+2 copies (Fig 2B). TEN strain Bosnia A failed to exponentially amplify at 10E+1 copies (Fig 2C; Table 2). The LAMP targeting the tprL locus had a detection limit of 10E+2 copies for Nichols (Fig 2D) and 10E+3 for TEN strain Bosnia A (Fig 2E; Table 2).
The LAMP assay that utilizes the TP_0619 locus amplified TPE (strain Gauthier) and TEN (strain Bosnia A) DNA until a total copy number of 10E+2 copies was reached (Fig 2F and 2G; Table 2).
In many areas where endemic treponematoses occur, syphilis can also be found at meaningful prevalence rates (e.g., Ghana 3.7% [31], Papua New Guinea 7.9% (men)-12.9% (women) [32]). While this is a problem for the serological based diagnosis of yaws in the presence of etiologically unrelated skin ulcers, it is not an issue for LAMP assays, which specifically target the DNA of the pathogen. The TPE/TEN LAMP was able to reliably discriminate yaws and simian TPE infection from infection with syphilis causing strains. It will, however, not discriminate yaws-causing strains from those known to cause bejel (TEN strains). While in theory this could be a problem, bejel is a disease found in the dry areas of Sahelian Africa and Saudia Arabia and thus its spatial distribution does not overlap with yaws reporting countries in Western and Central Africa, Southeast Asia, and the Pacific Islands [33]. In cases where a clear differentiation between yaws and bejel infection is important, the combination of the LAMP targeting tprL1 and TP_0619 will enable the distinction of both subspecies since only TEN strains will amplify in both assays. In the future, either TEN specific assays or a LAMP multiplex assay can be designed [34]. The latter, however, would require a fluorescence measuring device and thus may restricts the use in remote tropical health care facilities. Our study used the StepOnePlus Real-Time PCR System, but all LAMP assays described in this study can be run equally well on a portable system (e.g., ESEQuant TS2, Qiagen) that allows easy transportation and use under field conditions. In low-income clinical settings, it would even be possible to detect amplification by the naked eye through the detection of turbidity generated by the precipitation of magnesium pyrophosphate or through the addition of calcein, a fluorescent metal indicator [35]. Lyophilization allows for ambient storage of formulated LAMP reagents [36].
As indicated in the methods, all three gene targets that were selected for the LAMP assays are highly specific for the human and NHPs specific pathogenic TP, but also the lagomorph infecting T. paraluisleporidarum ecovar Cuniculus and Lepus, respectively. However, lagomorph infecting treponemes are not capable of infecting humans [23, 24] and most probably also NHPs. False positive test results due to infection with non-TP bacteria are therefore unlikely. In light of a recently published report on failure of qPCR due to variations in primer binding sites [37], it should be noted that the number of published genomes, in particular non-draft genomes, in any of the TP subspecies is low. At this stage, a general statement on genome variability at the selected gene target sites is therefore not possible. However, based on our research, which included representatives of the full range of published TP genomes (Table 1, S1 and S2 Figs), the relevant primer binding sites are conserved across the different subspecies and strains.
It has been proposed that yaws eradication in humans is possible through total community treatment in combination with subsequent total target treatment [38]. Rapid and reliable identification of yaws infection is important because successful global eradication requires an infinite zero-case scenario. In the first years after eradication has been declared in humans, it might well happen that few cases reemerge from yet unidentified relapsing latent yaws cases as well as there is a theoretical change that sporadic spillover from infected wild NHPs in Africa occurs. Either way, an available molecular test such as a LAMP assay could effectively and timely identify new cases from etiologically unrelated skin ulcers at the very beginning and could help to prevent yaws from re-emerging in areas where PCR machines and expensive laboratory equipment are not available. The analytic limits of detection for all three LAMP assays were around 10E+2 copies per reaction (Table 2), which is sufficient for clinical samples from human primary and secondary syphilis infection [39]. The same numbers can be expected for human yaws samples. Furthermore, the amount of TP in chronically infected monkeys also falls within the detection range of the TPE/TEN LAMP [27].
NHP TPE strains have been discussed as a possible source for human yaws infection in Africa [13]. The identification of NHP populations that harbor the pathogen, not only in Africa but also Asia [12], must be considered an important research priority [4]. Post-treatment surveillance needs to focus in particular on areas where NHPs and humans are in close contact. The TPE/TEN LAMP performance of the NHP samples (strain Fribourg-Blanc and DNA extracted from a clinical sample of a baboon at Ruaha National Park in Tanzania (RNP)) that were included into this study were similar to the results obtained for the human yaws-causing strains (Fig 1C). This is not surprising, given the fact that NHP TPE strains are genetically and functionally highly similar to human yaws causing strains [14, 15]. However, the full diversity of NHP infecting TP is unknown and it is possible that monkeys from Sahelian Africa and Saudia Arabia may carry TEN strains. In this case, the TPE/TEN LAMP assay would become positive. Due to the fact that currently all naturally occurring NHP infections with TP should be whole genome sequenced to fuel our understanding on yaws epidemiology and evolution, the TPE/TEN LAMP assay result would be more of academic than practical interest. The whole genome data derived from simian isolates would reveal the subspecies status of the isolate.
In humans, infections with all TP subspecies have reported potential to cause atypical clinical manifestations. A striking example is the frequent syphilis-like manifestations associated with TEN strains [40,41]. A rapid, highly sensitive and specific LAMP assay would therefore contribute to the identification of atypical clinical manifestations caused by TP. It would further help to identify possible NHP-to-human infection in countries like Tanzania, where human yaws has not been reported since decades. Syphilis screening programs in Tanzania would currently not detect possible NHP-to-human transmission events, since serological tests cannot discriminate between the TPA and TPE infection. Our target selection for LAMP assays that discriminate infection with TP from other causes of skin ulcers, represents a basis for the implementation of a One Health approach in yaws eradication and its post-eradication surveillance. Fig 3 illustrates the proposed new way of diagnosing TPE infection in humans. The new LAMP assays would simplify and accelerate yaws diagnosis. We note here that we have reached proof of concept for the suitability of the described gene targets, but further validation in a statistically adequate number of clinical samples is necessary to achieve confidence of the LAMP assays to be used in a non-research environment.
The selected gene targets are suitable for the diagnosis and discrimination of all three TP subspecies, which is currently not possible using clinical signs of infection in combination with serology. The next step must be to conduct tests to evaluate the sensitivity and specificity of the newly created assays in various clinical samples that originate from humans and NHPs. The designed LAMP assays do not require expensive laboratory equipment and can be run in virtually any clinical setting. Results are available within a few minutes and thus outrun the existing methods, which require reasonable laboratory infrastructure [4].
|
10.1371/journal.pcbi.1002016 | Exploring Off-Targets and Off-Systems for Adverse Drug Reactions via Chemical-Protein Interactome — Clozapine-Induced Agranulocytosis as a Case Study | In the era of personalized medical practice, understanding the genetic basis of patient-specific adverse drug reaction (ADR) is a major challenge. Clozapine provides effective treatments for schizophrenia but its usage is limited because of life-threatening agranulocytosis. A recent high impact study showed the necessity of moving clozapine to a first line drug, thus identifying the biomarkers for drug-induced agranulocytosis has become important. Here we report a methodology termed as antithesis chemical-protein interactome (CPI), which utilizes the docking method to mimic the differences in the drug-protein interactions across a panel of human proteins. Using this method, we identified HSPA1A, a known susceptibility gene for CIA, to be the off-target of clozapine. Furthermore, the mRNA expression of HSPA1A-related genes (off-target associated systems) was also found to be differentially expressed in clozapine treated leukemia cell line. Apart from identifying the CIA causal genes we identified several novel candidate genes which could be responsible for agranulocytosis. Proteins related to reactive oxygen clearance system, such as oxidoreductases and glutathione metabolite enzymes, were significantly enriched in the antithesis CPI. This methodology conducted a multi-dimensional analysis of drugs' perturbation to the biological system, investigating both the off-targets and the associated off-systems to explore the molecular basis of an adverse event or the new uses for old drugs.
| Idiosyncratic drug reactions (IDR) generally cannot be identified until after a drug is taken by a large population, but usually result in restricted use or withdrawal. Clozapine provides the most effective treatment for schizophrenia but its use is limited because of a life-threatening IDR, i.e., the agranulocytosis. A high impact clinical study demonstrated the necessity of moving clozapine from 3rd line to 1st line drug; therefore, intensive research has aimed at identifying genes responsible for clozapine-induced agranulocytosis (CIA). Olanzapine, an analog of clozapine, has much lower incidence of agranulocytosis. Based on this phenomenon, we proposed an in silico methodology termed as antithesis chemical-protein interactome (CPI), which mimics the differences in the drug-protein interactions of the two drugs across a panel of human proteins. e.g., HSPA1A was identified to be targeted by clozapine not olanzapine. Furthermore, the gene expression of the HSPA1A-related gene system was also found up-regulated after clozapine treatment. This approach can examine the system's perturbation in terms of both the off-target and the off-system's interaction with the drug, providing theoretical basis for decoding the adverse drug reactions or the new uses for old drugs.
| Clozapine (CLZ) provides one of the most effective therapeutic treatments for schizophrenia [1]. It is classified as an atypical antipsychotic drug because of its binding to serotonergic and dopamine receptors. However, its usage is limited due to potential life-threatening adverse drug reaction, mainly agranulocytosis [2], [3], [4]. FDA therefore requires blood testing for patients taking CLZ, complicating the clinical use of the drug. A recent high impact clinical study demonstrated the necessity of moving CLZ from a 3rd line drug to a 1st line drug based on its overall benefit/risk ratio [1]. Thus the identification of the biomarkers for clozapine induced agranulocytosis (CIA) could greatly broaden the usage of this drug. Organizations such as the severe adverse event consortium (SAEC) and Duke University are collaborating on identifying genetic risk factors for CIA via genetic association studies (http://www.genomeweb.com/dxpgx/saec-duke-collaborate-rare-variants-adverse-events-research). However, due to the rarity of suitable patients, such an approach requires global collaboration. Even if some statistically significant SNPs are identified by using genome wide association studies [5], [6], identifying the causal mechanism of such SNPs and using them in prediction models still presents a challenge. Instead of the traditional association study, we proposed an alternative computational methodology to identify the genetic risk factors for CIA, by identifying the known risk genes, explaining the relevant mechanism by observing chemical-protein interactions and providing a “most likely” candidate list [7] for pharmacogenetic and pharmacogenomic studies [8].
Drug-induced agranulocytosis is a form of idiosyncratic drug reaction (IDR). It is dose independent and is a form of serious adverse drug reaction [9], [10], [11]. One of the major causes of IDR is unexpected drug-protein interactions in human proteins [12], [13], [14], [15], [16], [17], [18]. Olanzapine (OLZ) is a CLZ analog, but has inferior efficacy in treating schizophrenia. It is reported to cause much less agranulocytosis compared with CLZ [19], [20], [21], a fact that is also confirmed in our statistical test (Fisher's exact test p = 8.2E-21, Table 1). Differences in their interaction profile towards human proteins (off-targets) might explain the etiology of CIA. Hence we hypothesized that if a human protein tends to be targeted by CLZ but not OLZ, the protein should be regarded as the candidate mediator of CIA, and the genes sharing a biological function with the off-targets (off-system, short for ‘off-target associated system’) should also display expression perturbation in cell lines treated by the drug. For example, we identified from a 410 protein target set retrospectively that Hsp70 protein as the off-target of CLZ but not OLZ, and that genes sharing the biological function with HSPA1A (Hsp70's gene) or acting as neighbors in Human Protein Reference Database (HPRD), a protein-protein interaction (PPI) database, with HSPA1A were found up-regulated in cell lines treated by CLZ. Another hypothesis is that if a protein target is preferably targeted by all drugs causing agranulocytosis (case) but not targeted by the agranulocytosis- drugs (control), the protein is a candidate mediator of the agranulocytosis. Using this hypothesis, we identified NQO2 gene as the candidate gene of agranulocytosis.
To identify unexpected drug-protein interactions, we utilized chemical-protein interactome (CPI) [13], [22], [23], which gives a score array generated by docking a panel of drug molecules across a set of human proteins. A CPI delivers two types of information, the binding conformation and the binding strength (Fig. 1a). It can be constructed via wet lab techniques [24], [25], [26], [27], but the most convenient way is to generate an in silico CPI. We used the DOCK [28] program to evaluate the chemical-protein interaction strength because it is an open-source software and had been widely used along with its success in identifying the unexpected chemical-protein interactions.
To prepare an unbiased protein set, we utilized a pocket set comprising 410 human protein pockets (381 unique proteins, Table S1), representing all the available human protein structure models from third-party target structural databases. The ligand binding pockets on each protein were then processed manually for docking preparation (see Methods).
We then mined from literature and the FDA adverse event reporting system (AERS) the drugs that were reported to cause agranulocytosis (case) or not cause agranulocytosis (control, Fig. S1a), aiming at identifying proteins tend to be targeted by case but not control drugs (red dashed rectangle in Fig. S1b). According to our criteria (Methods), there were 39 case and 15 control drug molecules selected for agranulocytosis, including the parent drug and their major metabolites and isomers. The control drugs did not share significant 2D structure similarity (Fig. S2), their indications covering a broad therapeutic categories (covering nine 1st level of ATC codes). To generate a comprehensive distribution of docking scores for each protein across many drug molecules, we also incorporated other drug molecules. Although for effective performance and classification, a larger data set should be used [22], e.g., all the FDA approved drugs), we restricted our analysis to drug molecules from our former studies because of the CPU time for array docking. Thus, a total of 255 drug molecules, including the CLZ and OLZ, were selected for docking (Table S2).
Here 255 chemicals were docked into the 410 human proteins using DOCK, generating a docking score matrix of 255×410 elements. A 2-directional Z-transformation (2DIZ) [23] was then applied to transform the raw docking score into a Z′-score, extending the multiple active site corrections concept [29]. The docking scores were normalized by each drug and then by each protein (Fig. 1b), thus the “endogenous” variance among proteins, such as the free energy variation across the binding pockets, has been normalized and contribute almost zero to the variance of the Z′-scores (Table S3). The major contributions of the variance are from the chemical effects and the chemical-protein interactive effects after the 2DIZ, which means that each chemical can ‘fish’ its targets only based on Z′-score without noises from the “endogenous” variance among proteins.
A basic assumption in using antithesis binding profile from CPI between CLZ and OLZ is that, 1) the two drugs are broadly similar in their effects, except for some side-effects, such as agranulocytosis, and that therefore, apart from some minor differences, their overall protein binding profile should be similar; 2) these minor differences in protein binding profile are highly likely to be associated with CIA. To verify the comparability between CLZ and OLZ, we calculated the Pearson's correlation coefficients (PCC) between Z′-score vectors of CLZ and OLZ across all 410 human proteins (with missing values removed). All four CLZ-OLZ pairs (2 CLZ ionization states×2 OLZ ionization states) obtained high positive PCC values (Fig. S3a). Their mean PCC value was distinctly higher (p = 0.0009 for permutation test in Fig. S3b). The high correlated protein binding profiles of CLZ and OLZ underlined their structural and pharmacological similarity, which also indicated the structural variability of all 255 drug molecules in the construction of the CPI. We therefore hypothesized that the proteins exhibiting different binding affinity against CLZ and OLZ might account for the agranulocytosis risk of these two analogs.
In order to identify the minor distinctions, we defined the antithesis score (A-score) for protein i as the Z′-score difference between CLZ and OLZ towards protein i,
We also calculated the probability of an A-score less than between two randomly selected drug molecules among 255 molecules at protein i (Fig. 1c), which could be expressed as,
We performed permutations for each target by randomly selecting drug-pairs and calculating their A-scores 10,000 times. Here the p value was the one-tailed probability when the A-score of the drug-pair was less than that of the CLZ-OLZ pair. Targets with p value less than the 0.05 cutoff are shown in Table 2. For the four CLZ-OLZ pairs, we chose only the pair that recalled most known CIA related genes reported in the genetic association studies.
A chemical-protein interaction with a Z′-score less or greater than −0.48 was defined as interactive or not interactive, respectively. As indicated in our previous training set [22], Z′-scores above such cutoff captured 70% of the true bindings and were enriched more than three-fold as compared with the false binding. For protein i, ai, bi, ci, and di, denoting the number of interactive (ai or bi) and not interactive (ci or di) by case or control drug molecules, respectively, were counted and the relative ratio (RR) was calculated as follows,
To identify proteins preferentially interacting with the case drugs, we performed Fisher's exact tests for each protein. The significance (one-sided) for each of the protein pockets with RR value exceeding one were computed and were used as a measure to prioritize the potential protein mediating agranulocytosis. Table 3 shows protein targets with p values less than 0.05.
Besides human leucocytes antigen (HLA) markers, three CIA susceptible genes have been identified in genetic association studies [30], namely HSPA1A [31], TNF [32] and NQO2 [33]. None of the HLA proteins were included in our pocket set since they did not meet our criteria of choosing protein pockets. Proteins coded by these three susceptibility genes all happen to be included in our pocket set comprising third party targetable protein databases (Table S1).
HSPA1A codes the heat shock 70 kD protein 1 (Hsp70 protein, PDB ID: 2E8A) and has been reported in a high profile journal to be associated with CIA with its causality in CIA discussed [31]. It is also well known for its druggability in antitumor drugs [34], which in general, cause the death of the cell. The gene was prioritized in our binomial antithesis CPI (Table 2). Significant binding strength differences between CLZ and OLZ towards Hsp70 were identified with the binding conformations visualized in Fig. 2. The CLZ molecule fits deeply into the Hsp70 pocket (Fig. 2b). By contrast, the methyl group of OLZ was difficult to accommodate in the narrow pocket using the similar binding pose as CLZ (Fig. 2c).
We further performed the site-moiety map analysis [35] of the Hsp70 pocket by examining the moiety preferences of the docked ligands and the physicochemical properties of the pocket. One van der Waals-interacting anchor site was identified with three essential residues (R272, R342 and G339, Fig. 3a). Among the docked drug molecules, most used the aromatic moiety or conjugated bonds to interact with this center (Fig. 3b). Theoretically, both CLZ (Fig. 3c) and OLZ (Fig. 3d) should have been capable of insertion into this pocket, however, the methyl on the OLZ molecule made it difficult to hold the same binding direction as that of the CLZ (see molecule structures in Fig. 3c, d). The CLZ molecule was inserted deep into the pocket and used most of its conjugated ring system to interact with the R272 and R342 via π-π interaction. Compared with CLZ, OLZ could not use the majority of its conjugated system due to steric hindrance caused by his methyl group. The above findings add evidence to the hypothesis that the Hsp70 protein was the off-target of CLZ but not of OLZ.
Ribosyldihydronicotinamide quinone dehydrogenase (coded by NQO2; PDB ID: 1SG0), the known risk gene for CIA, was prioritized from the multiple antitheses CPI (Table 3), together with other 44 proteins with p value less than 0.05. The protein was preferably targeted by the case but not the control drugs. The Kolmogorov-Smirnov test of the Z′-scores between cases and controls showed significant differences on two pockets (p = 0.002 and p = 0.004 for pocket 1 and 2, respectively). As for the binomial antithesis CPI, NQO2 protein ranked 37th among the 410 proteins (top 9%) when ordered by p value. Although the p value did not exceed the 0.05 threshold, the A-score was −1.18, indicating that there were still differences between the interaction strength of CLZ and OLZ towards this protein.
Myeloperoxidase and NADPH-oxidase are functionally involved in the pathogenesis of the drug-induced agranulocytosis [36], [37]. Myeloperoxidase (PDB ID: 1D2V) was found in Table 2 whereas two oxidoreductases using NADPH as the co-enzyme, namely Carbonyl reductase NADPH 3 (2HRB) and NAD(P)H dehydrogenase quinone 1 (1KBQ) were found in Table 3.
We also investigated the genetic polymorphisms of genes coding Hsp70, NQO2 protein, Myeloperoxidase and NADPH-oxidase. Some nonsynonymous single nucleotide polymorphisms (SNPs) were identified but none of these was found to affect the ligand binding pockets.
Besides bindings between chemicals and proteins, the drug-target relationship may also be reflected in the expression changes of genes related to the off-target associated system [38] after chemical treatment. If the mRNA expression of a set of genes related to off-target X is significantly changed after drug treatment, both target X and the associated system X could corroborate each other for their roles in the adverse reaction. Since Hsp70 was identified as the putative off-target of CLZ, we sought to investigate whether the CLZ treatment resulted in perturbation of Hsp70 and the related gene system. We analyzed the data from Connectivity Map (cMAP) [39], a collection of gene expression data from drug-treated human cell lines on Affymetrix U133A microarrays. Cells were treated by particular drug and vehicle respectively to measure the change of gene expression. One such drug-vehicle pair was defined as an instance. For all 6,100 instances, 22,283 probes were ranked by fold-change values with higher fold-change ranked at the top (close to rank 1), forming a 22283×6100 matrix. We recruited all four instances (instance 1170, 1289, 2689 and 6188) performed on the human promyelocytic leukemia (HL60) cell line to specifically address the drug effect of CLZ on the leukocytes. Instances performed on other cell lines were also investigated.
We then manually extracted genes related to HSPA1A in Gene Ontology (GO) (Fig. 1d) [40]. HSPA1A was associated with 7 GO terms in the biological process. As agranulocytosis is basically the death of neutrophil and is known to be correlated to apoptosis pathways [41], we choose the term “anti-apoptosis” (GO:0006916) to characterize the role of HSPA1A in CIA. We selected all human genes linked to this term that collectively represented the Hsp70 off-system. These genes were mirrored to probes on microarray (439 probes corresponding to 235 genes). For each probe, we calculated the average rank of the probe across four CLZ instances (R′ rank), with higher R′ (closer to rank 1) indicating generally up regulated status and lower R′ down regulated status. We compared the R′ of the Hsp70 system and other genes on the U133A probe set. The anti-apoptosis system exhibits an R′ distribution quite distinct from that of the genome background (Fig. 4a), with significantly higher mean R′ than the random 235 gene set (258 out of 10000 sets showed higher R′, p = 0.0258 for permutation test, Fig. 4b). The general up regulation of Hsp70 related genes indicates that CLZ treatment clearly changes the bioactivity of the Hsp70 system in human HL60 promyelocytic leukemia cells. The Hsp70 off-system's perturbation was further confirmed using HSP1A1's ‘neighbor’ in HPRD [42] network) following the same procedure as for investigating the anti-apoptosis system (Fig. 4c,d). Both GO term-based off-system and the PPI-based off-system corroborate the important role of Hsp70 in CIA. The cMAP also contains breast cancer cell line MCF7 and human prostate cancer cell line PC3, however, none of the perturbation of the Hsp70 system could be detected in these two cell lines. The significant perturbation could not be detected on other six GO terms of HSP1A1.
The drug-(off) targets interaction and the gene expression change are the molecular events at two different dimensions after drug treatment. To get an overview of the systems perturbation of the off-targets prioritized in Table 2, we investigated the PPI-based off-systems for them. We did not choose the GO term-based off-systems because each gene was related to multiple GO terms, and it was difficult to objectively choose the appropriate GO terms related to agranulocytosis. Furthermore, using PPI-based off-systems to study the drug's perturbation on the biosystems has been proved to be applicable [43]. Among 17 off-systems, three were found to be significant perturbed with a permutation p value less than 0.05 (Table 2), including Hsp70 off-system.
The PPI-based off-systems were then visualized in Fig. 5, where the gene expression perturbation ‘landscape’ of the off-systems was shown. These off-systems were found to be connected by several hub nodes, such as apoptosis associated gene (TP53), the gene coding Bcl-2-binding protein (BAG1) and the transcriptional regulator of vitamin D3 receptor (TRIM24) et al. Interestingly, NQO2 was also found to be involved in HSPA1A off-system and significantly up-regulated after CLZ treatment. Besides preferably inhibited by CLZ, most of the oxidoreductases were found down-regulated or remain unchanged after CLZ treatment. The whole picture demonstrated that the impact of CLZ on the HL60 cell line is reflected on the up-regulation of the anti-apoptosis systems and the inhibition or the down-regulation of the oxidoreductases.
Interestingly, oxidoreductases were found to be significantly enriched in prioritized proteins. For example, quinone oxidoreductase (PDB ID: 1YB5), an isozyme of the NQO2 protein, also appears in Table 2. Seventy out of 410 protein pockets (17%) were oxidoreductases (Table S1). However, as Table 2 shows, oxidoreductases were significantly enriched (10 out of 19, 53%, Fisher's exact test p = 6.6E-4). Among targets prioritized by multiple antitheses CPI (Table 3), 15 out of 44 pockets (34%) belonged to oxidoreductases (p = 7.9E-3). In addition, only 12 out of 410 protein pockets (3%) were related to glutathione metabolite, which plays key role in antioxidation. However, as Table 3 shows, 7 out of 45 (16%) were significantly enriched (p = 1.2E-3).
Identification of off-targets has potential application in drug repurposing [44], [45] and personalized medicine [13], [46]. Compared with the similarity ensemble approach [47] and the naive Bayesian classifiers approach [48] to off-target identification, both of which build new drug-protein connections within the space of the known therapeutic target, the chemical-protein interactome approach is a step towards analyzing the entire human proteome, although the available human protein structrome is limited. Several of the pocket comparison algorithms have also tried to explore the off-target spaces facing the entire human proteome [15], [17], or tried to map the off-targets onto the pathways [49] or the metabolic network [50], but our study is the first one examining the system's perturbation in terms of both the off-target identification and the off-system's gene expression change, providing candidates for pharmacogenetic and pharmacogenomic studies, respectively. Further work may combine the off-target and the off-system in elucidating and predicting adverse drug reactions.
In the retrospective studies, the antitheses CPI recalled the accredited susceptible genes for CIA. As a complement to genetic association studies [6], the CPI reveals the possible mechanism of the CIA based on the drug-protein interaction, the primary step in drug reaction. The difference between the interaction conformation and the interaction strength of CLZ and OLZ towards the off-targets could account for the difference in patients' susceptibility to agranulocytosis. Since none of the nonsynonymous SNPs was found around the ligand binding pocket of the four proteins reported to be involved in CIA, we deduced that individual differences in CIA susceptibility could be explained by a variation in the expression level of the protein. In fact, NQO2 was found to have lower expression levels in CIA susceptible patients [33]. The lower expression level in this detoxification enzyme could make the patient more sensitive to the drug. It is also reasonable to expect subsequent discoveries (e.g. some genotypes correlated to Hsp70 or NQO2 expression level) supporting the CLZ off-target hypothesis, which could lead to biomarker development at genotype and gene expression level [51] in CLZ therapy.
The reactive oxygen hypothesis is one of the major hypotheses of agranulocytosis etiology [37]. In our results, CLZ and other drugs causing agranulocytosis tended to affect the oxidoreductases, which play an important role in reactive oxygen clearance. For example, NQO2 protein and myeloperoxidase are key enzymes in the detoxification of active radicals thus protecting the cells from drug-induced oxidative and electrophilic stress [52]. Furthermore, alpha-tocopherol transfer protein is a prioritized target of clozapine (Table 2). Blocking the transferring of tocopherol, which is a strong endogenous antioxidant [53], may also explain clozapine's impact on the detoxification system. Clozapine can be oxidized to reactive nitrenium ions [54], which preferably reacts with sulfhydryl and is detoxified by glutathione. In our results, glutathione related enzymes were significantly enriched in the CPI, implying that the drug causing agranulocytosis not only affected the detoxification system of oxidoreductases, but might also interfered in the glutathione system, which is essential to the detoxification of the major metabolites of CLZ.
Besides the unexpected drug-protein interactions, the expression change of the off-system may explain CIA etiology. The perturbation of anti-apoptosis genes by CLZ treatment reflects the fact that CLZ disturbs cell death pathways by binding with Hsp70, and the general up regulation of anti-apoptosis genes can be explained as a feedback towards elevated apoptotic stress mediated by Hsp70 and the anti-oxidation system, since the inhibition of oxidoreductases and the perturbation of oxidoreductase system is a well known mediator of apoptosis [55]. By breaking the balance of oxidation and reduction, CLZ can stimulate apoptosis via Hsp70 inhibition and enhanced oxidative stress. Along with the CPI results, biological effects of CLZ further support the hypothesis that Hsp70 and oxidoreductases together with their respective system serve as the off-targets(-systems) of CLZ and potentially mediate CIA. Since HL60 is derived from peripheral blood leukocytes, which is a representative cell model for the immune system, the finding of the systems perturbation in HL60 cells but not in MCF7 (breast cancer) and PC3 (prostate cancer) cell lines strengthens the antiapoptosis and the oxidoreductases systems' function in immune related events. In summary, 53% and 34% of prioritized proteins from the CPI are oxidoreductases, and 16% of the proteins are related to gluthathione metabolism. These findings suggest a much higher participation of the detoxification/antioxidant systems in drug-induced agranulocytosis than previously thought and the off-targets/-systems identified in this study can represent candidates for biomarker development in wet-lab experiments and pharmacogenetic/pharmacogenomic screening in the future.
However, the 410 binding pocket set is a limited representation of the entire human proteome. For instance, it does not include any HLA proteins according to our target preparation criteria, which may be involved in agranulocytosis as a mediator of the immune etiology. Drug-HLA interaction was reported to be an important step determining the drug-HLA specificity in IDR [56]. In our previous study, we have built the abacavir-HLA-B*5701 interaction models for abacavir-induced hypersensitivity [13]. The identification of the drug-HLA interaction at the F-pocket of HLA molecules has been cited by several immunologists [57], [58]. Since HLAs have been identified as the key factors in IDRs [5], [6], [59], [60], the drug-HLA interactome will be systematically studied in future.
Identification of the related genes and the systems is the first step towards understanding and more importantly, predicting the IDR. The IDRs were regarded as unpredictable in response to compounds [61]. In this study, we argue that the IDRs are predictable, and the challenge of personalized medicine is not to predict adverse reaction for a compound but for a patient. The biomarkers could be either the genetic variations causing a binding affinity change of the drug towards the off-targets [62], [63], the expression level alteration of one gene [33], or the off-systems' perturbation. Our study demonstrates that beside polymorphisms around the binding pocket that alter the drug efficacy via a change in the binding affinity [64], [65], the off-system expression change could also determine individual variability towards the same drug, suggesting a new way of identifying biomarkers or constructing a prediction model for personalized medicine. Such an approach could also be applied to personalized drug repurposing [66], [67], [68], where the off-targets and the off-systems accounting for the new therapeutic area could also be patient specific.
Adverse drug reaction and the new indication are two ‘off-effects’ of the drug towards human being. So this study will also illuminate the drug repositioning by, 1) helping explain the mode-of-action of the serendipitous repositioned drugs via identifying their off-targets/-systems; 2) predicting the new use for existing drugs based on their interaction profiles with the off-targets and their perturbations on the off-systems. For example, one can recruit the case and the control molecular set for a particular indication. After identifying the off-targets/-systems using the methodology in this study, one can predict the indication of a new compound based on its impact on these newly identified off-targets/-systems.
The reports were downloaded from the FDA's AERS (http://www.fda.gov/Drugs/GuidanceComplianceRegulatoryInformation/Surveillance/AdverseDrugEffects/default.htm). This system tracks adverse events that are voluntarily reported but only the records from 2004 were freely available. All reports bearing CLZ and OLZ as the primary or secondary suspected drug were counted. The numbers of agranulocytosis cases were then counted for each drug. We performed Fisher's exact test to examine the frequency difference.
Protein targets were obtained from third-party protein structure databases, including a drug adverse reaction target database [69], a drug-induced toxicity related protein database [70], a therapeutic target database [71] and a protein database for drug target identification [72]. Every pocket was examined manually when constructing the target set for DOCK according to the following criteria. First, the species should be confined to Homo sapiens; secondly, a co-crystallized ligand must be contained to indicate the targetable state of the protein; thirdly, the pocket should not contain missing residues. Spheres whose radii ranged from 1.1–1.4 Å were generated to fill in the pocket. A grid box was constructed 3–5 Å from the spheres. EC classifications of the enzymes were taken from the annotations of UniProt [73]. Finally, we achieved 410 protein pockets from 384 PDB entries, 74% of which have the resolution less than 2.5 Å.
Drugs reported in the PubMed literature (up to September, 2009) as being associated with agranulocytosis were chosen as candidates and further examined in the AERS administered by the FDA/Center for Drug Evaluation and Research (http://www.fda.gov/Drugs/GuidanceComplianceRegulatoryInformation/Surveillance/AdverseDrugEffects/default.htm). All AERS raw data were downloaded from the FDA website and then placed in a relational database (MySQL 5.1). Accessible data were limited to the period from Jan 2004 to March 2009. In any adverse event report, only the primary or the secondary suspected drugs were regarded as linked to agranulocytosis. The candidates were only included if the number of reports exceeded 3. The candidates for control drugs were collected from AERS data, on condition that there were no reports of agranulocytosis. Candidates were then confirmed as control drugs only if they had never been co-cited with agranulocytosis in PubMed literature and the first 10 results of a Google search (up to September, 2009; with drug name AND “agranulocytosis” as query term). The major metabolites and the isomers of the drugs were also included. In the end, 39 case and 15 control drug molecules were selected for agranulocytosis endpoint. These 15 controls do not share significant 2D structure similarities. The SMILES code of the drugs and their derivatives was retrieved from PubChem. The 3D conformations of chemicals were simulated using CORINA. Charges and hydrogens of proteins and chemicals were added using Chimera [74].
The background drugs were chosen from the molecules prepared in our previous studies, including anti-Alzheimer drugs [22], drugs referred to in the study by Lamb et al [39] on using the cMap, case and control drugs for rhabdomyolysis, cholestasis, deafness and Stevens-Johnson syndrome and QT prolongation [13]. A total of 255 drug molecules, including case and control drugs for agranulocytosis, were involved in constructing the CPI.
A CPI comprising 255 drugs towards 410 protein pockets was constructed using the DOCK [28] program controlled by Bash shell scripts. The parameters for docking corresponded to the default settings. The 2DIZ transformation [23] was performed where the docking score matrix was normalized first by one drug towards the 410 proteins then by one protein pocket towards the 255 drugs. The empirical threshold −0.48 of the Z′-score was set to distinguish binding and non-binding, based on the findings of the previous studies [22], [23].
To determine the significance level of similarity between four CLZ-OLZ pairs (2 CLZ ionization states×2 OLZ ionization states) across their protein binding profile, we randomly recruited 10,000 sets with four drug pairs from all 255 drugs in the CPI, and identified 9 pairs with mean PCC not lower than the mean PCC of the four CLZ-OLZ pairs.
Suppose there are n genes sharing a specific GO term or linked to the same hub in the HPRD network. Each probe was independently ranked according to expression change for each instance in cMAP, with most up-regulated being at the top. For the cMAP instance # 1170, 1289, 2689 and 6188, which were the CLZ-treated instances, we calculated the mean rank R′ of each probe aswhere R1170, R1289, R2689 and R6188 indicate the rank in instance 1170, 1289, 2689 and 6188, respectively. For evaluation on the perturbation status of a system, we randomly recruited 10,000 sets with n genes, obtaining m sets with mean rank higher than the object system. The p value was calculated as m/10000.
Polymorphism information for the genes was retrieved from dbSNP [75] and UniProt [73]. The ‘coordinations’ of the amino acid sequence in the PDB files were adjusted to match the ‘coordination’ of dbSNP. The distance between the polymorphism site and the ligand binding pocket of the protein was visualized on PyMOL.
|
10.1371/journal.pntd.0006132 | 4-aminopyridyl-based lead compounds targeting CYP51 prevent spontaneous parasite relapse in a chronic model and improve cardiac pathology in an acute model of Trypanosoma cruzi infection | Chagas disease, caused by the protozoan Trypanosoma cruzi, is the leading cause of heart failure in Latin America. The clinical treatment of Chagas disease is limited to two 60 year-old drugs, nifurtimox and benznidazole, that have variable efficacy against different strains of the parasite and may lead to severe side effects. CYP51 is an enzyme in the sterol biosynthesis pathway that has been exploited for the development of therapeutics for fungal and parasitic infections. In a target-based drug discovery program guided by x-ray crystallography, we identified the 4-aminopyridyl-based series of CYP51 inhibitors as being efficacious versus T.cruzi in vitro; two of the most potent leads, 9 and 12, have now been evaluated for toxicity and efficacy in mice.
Both acute and chronic animal models infected with wild type or transgenic T. cruzi strains were evaluated. There was no evidence of toxicity in the 28-day dosing study of uninfected animals, as judged by the monitoring of multiple serum and histological parameters. In two acute models of Chagas disease, 9 and 12 drastically reduced parasitemia, increased survival of mice, and prevented liver and heart injury. None of the compounds produced long term sterile cure. In the less severe acute model using the transgenic CL-Brenner strain of T.cruzi, parasitemia relapsed upon drug withdrawal. In the chronic model, parasitemia fell to a background level and, as evidenced by the bioluminescence detection of T. cruzi expressing the red-shifted luciferase marker, mice remained negative for 4 weeks after drug withdrawal. Two immunosuppression cycles with cyclophosphamide were required to re-activate the parasites. Although no sterile cure was achieved, the suppression of parasitemia in acutely infected mice resulted in drastically reduced inflammation in the heart.
The positive outcomes achieved in the absence of sterile cure suggest that the target product profile in anti-Chagasic drug discovery should be revised in favor of safe re-administration of the medication during the lifespan of a Chagas disease patient. A medication that reduces parasite burden may halt or slow progression of cardiomyopathy and therefore improve both life expectancy and quality of life.
| Chagas disease is a parasitic disease caused by the Trypanosoma cruzi. The infection may result in gastrointestinal manifestations and cardiomyopathy. Benznidazole, the current treatment, has limited efficacy and often leads to serious side effects. Aiming to develop new treatments, our group has identified new inhibitors that block the synthesis of parasitic lipids, resulting in parasite death. In this work, we evaluated the safety and efficacy of two of these compounds, 9 and 12, in mouse models of T. cruzi infection. Both compounds were well-tolerated by animals throughout the 28-day administration. In acutely infected mice, the compounds drastically reduced bloodstream parasites and increased survival. When treatment was initiated during the chronic phase, parasitemia dropped to background levels and remained undetectable for 4 weeks after drug withdrawal; parasites were re-activated by chemically-induced immunosuppression. Thus, the experimental compounds tested in these studies had an acceptable safety profile, achieved a marked reduction in parasite load and prevented heart injury due to inflammation, even in the absence of sterile cure. We conclude that the development of non-toxic medications capable of slowing the progression of cardiomyopathy is a valuable treatment option for Chagas disease patients because it could enhance the quality of life.
| Chagas disease afflicts about 7 million people in South and Central America [1], where it is the leading cause of heart failure. More than 10,000 deaths are estimated to occur annually due to this disease. Despite joint efforts in Latin America to eradicate the transmission of Trypanosoma cruzi through screening of blood banks and control of triatomine vectors, Chagas disease presents a risk to 70 million people living in endemic countries [1,2]. International travel, infected blood transfusions, co-infection with HIV, mother to fetus transmission, and northward migration of the “kissing bug” insect vector [3], all help to drive up the number of cases and push the incidence outside its historic range. Chagas disease is now seen in Europe, North America and Asia and seems set to become an urgent public health issues in countries far beyond its focal source in South America [4,5]. An annual economic burden due to Chagas disease, calculated by simulation models as overall cost, reaches 7.19 billion US dollars, largely from the loss of productivity and premature mortality caused by cardiomyopathy [6,7].
Human infections by T. cruzi result in a significant mortality rate in children in the acute phase, or may lead to cardiomyopathy in chronically infected adults [8,9]. About 40% of infected individuals develop chronic manifestations of the disease: ten percent of patients develop gastrointestinal symptoms (e.g., mega colon and mega esophagus); and 30% of patients develop cardiac disease characterized by cardiomyopathy, arrhythmias and interstitial fibrosis accompanied by cardiac inflammation[9]. The clinical treatment of Chagas disease is limited to two drugs: nifurtimox and benznidazole, developed about 60 years ago. Nifurtimox is now discontinued in several countries [10,11], while benznidazole has been recently FDA-approved only for use in children of 2 to 12 years old [12,13].
Both benznidazole and nifurtimox are about 80% effective in the acute stage of Chagas disease [14]. Limitations of current therapy include variable efficacy against T. cruzi of different genetic backgrounds and elevated toxicity with severe side effects, including widespread dermatitis, digestive intolerance, polyneuritis and bone marrow depression, leading to poor patient compliance [10,11]. Both drugs are used in cases of new infections, congenital infections, reactivation and/or re-aggravation associated with immunosuppression and as a preventive measure against laboratory accidents [8]. The efficacy of benznidazole against the more prevalent chronic stage of Chagas dissease was investigated in the BENEFIT clinical trial [15,16]—the first randomized, placebo controlled, clinical study on the effects of benznidazole on the clinical progression of chronic Chagas disease patients with compromised cardiac function. Drug treatment led to a marked reduction of the circulating parasite load in patients from Brazil (strain TcII) and Argentina and Bolivia (strains TcV and TcVI), but not in patients from Colombia or El Salvador (strain TcI). In all cases, benznidazole failed to reduce cardiac function deterioration when evaluated at the 5–7 year follow-up. [15]. These results suggest that benznidazole has limited clinical utility in patients with moderate to advanced cardiac compromise (class I or II heart failure, New York Heart Association terminology). However, an important qualification is that previous observational, not randomized, studies [17] suggest that the drug is effective in patients in the asymptomatic (indeterminate) stage or those with incipient cardiac compromise. A combination of benznidazole with posaconazole in the treatment of asymptomatic patients (the STOP-Chagas clinical trial) also showed no advantage over benznidazole monotherapy, as judged by the PCR test alone. Clinical disease as evidenced by decreased cardiac function or other cardiomyopathy signs were not assessed in this study [18]. In either case, there is an urgent need for safer and more efficacious drugs and drug combinations to meet the etiological challenges of this complex disease.
As an alternative to the use of benznidazole in patients with chronic Chagas disease [19], significant efforts have been made to repurpose antifungal azole drugs targeting sterol biosynthesis. Among validated sterol biosynthetic targets, CYP51 is one of the most extensively exploited for the development of new therapeutics for fungal and parasitic infections [20,21]. The CYP51 inhibitors posaconazole (Noxafil, Merck) and ravuconazole (E1224, Eisai, Tokyo), which have undergone extensive pharmacological and toxicological optimization in antifungal programs, have demonstrated efficacy and curative activity in animal models of Chagas disease [22], and alleviated chronic Chagas disease in a patient with systemic lupus erythematosus [23,24]. Both drugs have been tested in controlled clinical trials for Chagas disease [18,25,26]. The perceived inferiority of both drugs to the current standard-of-care drug, benznidazole, [25,27] was due to their failure to produce sterile cure (PCR negative), and triggered discussions in the Chagas research community about the validity of CYP51 as a target [28,29]. Two concerns have been expressed: (i) differential activity of CYP51 inhibitors against different strains of T. cruzi or between the replicative (amastigote) and non-replicative (trypomastigote) stages of the parasite and (ii) the slow-acting mechanism of CYP51 versus fast-acting benznidazole [28–30]. A third factor that may have affected the outcomes of the clinical trials is that the repurposed antifungal drugs, including posaconazole and ravuconazole, were not optimized to target T. cruzi CYP51.
In parallel with the clinical trials, a number of laboratories pursued novel chemical scaffolds specifically targeting T. cruzi CYP51 (reviewed in [21]). Using a target-based structure-aided drug discovery approach, a 4-aminopyridyl-based scaffold was identified as efficacious and further developed into a series of lead compounds active against T. cruzi both in vitro and in vivo [31–37] (Fig 1). Two optimized leads of the series, 9 [37] and 12 [35] (Fig 1, compound numbers correspond to those in the cited references), have now been evaluated for both toxicity and parasitological cure in the acute and chronic animal models of T. cruzi infection. Although a sterile cure was not achieved, 9 and 12 were proven safe for long term administration in mice and suppressed parasitemia in both the acute and chronic phases. In the acute model, these lead compounds improved survival, protected mice from hepatic injury and drastically reduced cardiac inflammation. In the chronic phase, these lead compounds prevented spontaneous T. cruzi relapse for up to 4 weeks post-treatment.
The No Observed Adverse Effect Level (NOAEL), of 9 and 12 was evaluated according to the Organization for Economic Cooperation and Development (OECD) guidelines. Escalating doses of 12 were administered orally to male and female Swiss mice every hour; adverse effects were observed only at concentrations higher than 300 mg/kg for male and 250 mg/kg for female mice. Cumulative in vivo effects were analyzed using uninfected BALB/c male and female mice treated with 9 or 12 at 25 mg/kg orally for up to 28 days, b.i.d. No adverse clinical signs (such as ruffled fur, hunched posture, reduced mobility, or tremor) or alteration in general health were observed in any of the mice. Blood was collected after the end of treatment and serum was evaluated in a chemistry panel that included liver enzymes and markers of renal function. No alteration of blood levels for alanine aminotransferase (ALT), aspartate aminotransferase (AST), bilirubin (BIL), albumin (ALB), blood urea nitrogen (BUN) or creatinine (CRE) was detected after the course of treatment (Fig 2). Histological analysis of brain, heart, liver, kidney, GI tract and lungs did not show any alteration of tissue morphology. The weights of the animals remained steady throughout the treatment.
Since no toxicity was detected in uninfected mice, we performed a 28-day oral treatment at 25 mg/kg of compounds in Swiss male mice infected with T. cruzi Y strain (104 inoculum), an established model of acute infection recommended for drug screening and development by the Fiocruz Program for Research and Technological Development on Chagas Disease (PIDC/Fiocruz) and the Drugs for Neglected Diseases Initiative (DNDi) [38]. Treatment with 9 or 12 at 25 mg/kg significantly reduced parasitemia, reaching the minimum limit of detection by the Pizzi-Brener method at 9 days post infection (dpi), with inhibition levels of 99.9% for 9 and 99.3% for 12. Parasitemia remained undetectable in the treated mice, while untreated and vehicle-treated mice showed high parasitemia and all of these mice died by 18 dpi (Fig 3A). While parasites were not detected, only 20% of the infected mice treated with 9 and 80% of the infected mice treated with 12 survived the entire 30 day study (Fig 3B). 100% of benznidazole-treated mice survived. All mice were euthanized at the end of treatment. Compared to the negative controls, treatment with the test compounds did provide partial protection and delayed the death of the mice. Death of the treated animals in this model of acute disease with Y strain parasites may be due to yet unknown T. cruzi Y strain-specific factor(s) that could interfere with the compound effect in the infected mice. That effect with Y strain parasites was not seen in uninfected animals or animals infected with T. cruzi CL-luc strain.
Upon study completion, mice were euthanized and serum was assessed for hepatic enzymes and renal function markers (Fig 3C–3F). Data from the blood chemistry analysis showed that T. cruzi Y infection induced elevated serum levels of alanine aminotransferase (ALT) and aspartate aminotransferase (AST), indicating infection-induced liver injury. The animals treated with 9 or 12 had reduced levels of these enzymes when compared to untreated controls. Urea and creatinine (CRE) levels, markers of renal function, were also analyzed; treated animals show normal levels similar to that of untreated controls, suggesting no renal toxicity was caused by 9 or 12.
Given that neither 9 nor 12 produced sterile cure in the T. cruzi Y-infected animals, we next evaluated their effect in a less severe acute model using the transgenic CL-Brenner T. cruzi strain expressing “red-shifted” luciferase (CL-luc, a gift from Dr. John Kelly, UK). This strain carries a stable bioluminescent marker, which allows one to detect live parasites in tissues of a live mouse with a sensitivity limit exceeding that of the RT-PCR method for up to a year after infection [39,40]. Since mice gender influences the level of infection [41,42], we used both male and female mice for the 28 day b.i.d. treatment beginning 14 dpi. Males not treated with experimental or reference compounds showed levels of parasitemia higher than females (Figs 4, 5 and 6).
As judged by the bioluminescence (Fig 4), the level of CL-luc T. cruzi infection markedly decreased at 18 dpi, which corresponds to 4 days of treatment with the experimental or reference compounds, benznidazole or posaconazole. By 32 dpi (18 days of treatment), the parasite bioluminescence in all treated mice was reduced to the background level, with the photon count lower than that of the uninfected mice injected with luciferin. Both groups of negative control mice, including a group of infected but untreated mice, as well as a group of infected and vehicle-treated mice, had bioluminescence levels two orders of magnitude higher than the uninfected control mice. This trend was maintained up to 49 dpi (7 days after the end of treatment). However, by 53 dpi (11 days after the end of treatment) animals from the benznidazole (male and female), 12 (female), and 9 (male and female) groups all showed resurgence of parasites (Figs 4, 5 and 6). Posaconazole treated mice showed resurgence of parasites at 67 dpi.
By 81 dpi, the majority of the animals in the 9- and 12-treated groups were T. cruzi-positive as indicated by bioluminescence. All animals in the male group treated with 9 showed parasitemia as high as untreated controls at 81 days post infection. Both compounds reduced the parasite load in female mice, where bioluminescence levels were 10× lower than untreated controls in 4 out of 5 mice for both compounds at 81 dpi, with one 9-treated female negative (Table 1). Posaconazole reduced T. cruzi bioluminescence to undetectable levels in 1 out of 5 males and 2 out of 4 females, with the remaining mice showing reactivation of parasites. Posaconazole markedly reduced parasite load compared to the untreated controls. Finally, benznidazole was not able to suppress parasites in all the mice, with one female showing circulating parasites after treatment withdrawal (Figs 4, 5 and 6, Table 1).
Ex vivo bioluminescence imaging of internal organs was performed at 81 dpi. As previously described for this model [39,43], parasites were detected consistently in the gastro-intestinal (GI) tract in all T. cruzi-positive mice (Figs 7 and 8). Moreover, bioluminescence above the background level was also observed randomly in heart, skeletal muscles, liver and mesenteric fat in T. cruzi-positive mice throughout the groups, suggesting a dynamic infection process (Fig 7). Ex vivo quantification of the parasite burden in the organs analyzed (heart, liver, kidneys, lungs, spleen, gastro-intestinal tract, skeletal muscle and mesenteric fat) was consistent with the whole-mouse imaging. One female mouse treated with benznidazole showed traces of infection in the GI tract, while posaconazole-treated animals revealed T. cruzi in the GI tract and lung. For the experimental inhibitors, no parasites were detected in two females treated with 12 and one female treated with 9. All other animals in the 9- or 12-treated groups showed T. cruzi bioluminescence in GI tract, liver, lung and mesenteric fat (Figs 7 and 8).
Tissue architecture and inflammation in the heart was evaluated through conventional histology and H&E staining from both lethal acute and bioluminescent models, with the levels of inflammation being quantified using FIJI software [44]. Uninfected mice had normal cardiac tissue, as expected (Fig 9D). T. cruzi infection in untreated mice resulted in marked inflammation in heart tissue with inflammatory infiltrates and interstitial fibrosis in both acute models (Fig 9A and 9F). BALB/c mice infected with CL-luc showed mild cardiac inflammation when compared to Swiss mice infected with T. cruzi Y strain; the latter showed levels of inflammation at least 2× higher than the former and included the presence of amastigote nests, which were not observed in BALB/c mice infected with CL-luc (Fig 9A, 9F, 9E and 9J). In both models, mice treated with benznidazole (Fig 9B and 9G), 9 (Fig 9C and 9H), or 12 (Fig 9I) had normal heart tissue, with a significant reduction of inflammatory cells (Fig 9E and 9J) compared to vehicle treated controls, and no signs of interstitial fibrosis or amastigote nests, suggesting that the reduction of parasite load induced by treatment with the CYP51 inhibitors improved cardiac pathology, even without sterile cure.
Most patients in need of treatment are in the chronic phase of Chagas disease. In this later stage, parasite load is low enough to require sensitive techniques for parasite detection [45]. To recapitulate these conditions, we evaluated performance of 9 in a chronic mouse model. Because males are more susceptible to infection than females, BALB/c males were infected with T. cruzi CL-luc strain allowing highly sensitive bioluminescence detection. Following the treatment scheme reported [46], the compounds were administered at 25 mg/kg for 28 days starting at 126 dpi, when chronic infection was established and the parasite signal was consistently detected. Similar to the acute model described above, T. cruzi bioluminescence levels dropped soon after the start of treatment. After 28 days, all treated groups, including 9, showed only background luminescence (Fig 10A–10C). Mice were then followed for 4 weeks after compound administration had ceased. For up to 27 days post-treatment, groups treated with posaconazole or 9 had bioluminescence levels slightly above the background defined by the uninfected mice and 10–100× lower than the untreated controls (Fig 10D and 10E). Since the parasite load was below the detection level 4 weeks post-treatment, the animals were immunossupressed with cyclophosphamide. After 2 rounds of immunossupression, parasites relapsed as evidenced by the bioluminescence levels similar to those of the untreated controls.
A similarity in sterol biosynthesis pathways between T. cruzi and fungi is that both produce ergosterol and ergosterol-like sterols as membrane building blocks [47]. This similarity encouraged the application of antifungal drugs for the treatment of Chagas disease. However, in human clinical trials for Chagas disease, both posaconazole and ravuconazole failed to demonstrate superiority to the current standard-of-care drug, benznidazole, using PCR as a marker of continued or reactivated T.cruzi infection [25,27]. The failure of posaconazole and ravuconazole to attain sterile cure in humans raised concerns about the CYP51 target. Differential activity of CYP51 inhibitors against the replicative (amastigote) and non-replicative (trypomastigote) stages of T. cruzi, a slow-acting mechanism of action, and the stochastic nature of T. cruzi infection with the non-replicating or rarely-replicating cryptic amastigotes ‘hidden’ inside the tissues [39], were listed as potential drawbacks of the CYP51 target [28,48]. On the other hand, it has been argued that both the dose and the duration of anti-fungal agents used in the clinical trials to treat human T. cruzi infection have been suboptimal. Urbina et al. noted that the plasma exposure in patients for the dose used in clinical trials corresponds to 10–20% of the curative dose in mice [29,49]. The post-clinical trial tendency to balance risks in the Chagas drug discovery portfolio, and to identify drug candidates aimed at other molecular targets is logical. At the same time, it is critical to not reject promising targets based on clinical studies with drugs not properly optimized, dosed, or clinically evaluated.
An important, but overlooked factor, that may have affected the performance of the anti-fungal drugs in Chagasic patients, is the loose drug-target fit demonstrated by posaconazole in co-crystal structures with Trypanosome CYP51. The electron density of the bound drug is poorly defined and its pendant phenyl-2-hydroxy-pentantriazolone group adopts alternative conformations to make multiple interactions outside of the active site [50,51]. Several novel CYP51 inhibitors developed in this collaboration [33–37] and elsewhere [51–53] demonstrated drug-target fits superior to posaconazole. The most potent 4-aminopyridyl-based inhibitors of the 4-aminopyridyl-based series bind entirely in the CYP51 target interior, making tight interactions with hydrophobic residues constituting the CYP51 active site [35,37]. Improved drug-target interactions may be responsible, at least in part, for the superior potency of the experimental inhibitors in the acute and chronic mouse models of infection [54].
Neither of the two lead compounds of the 4-aminopyridyl-based series evaluated in these studies attained a sterile cure. However, both leads were proven safe for long term administration in mice and, efficiently suppressed parasitemia in both acute and chronic models of infection. In the acute phase, compounds improved survival in highly stringent acute mouse models, protected mice from hepatic injury, and drastically reduced acute cardiac inflammation. In a model of chronic Chagas disease, 9 prevented spontaneous T. cruzi relapse for up to 4 weeks post-treatment. Similar results—supression of parasitemia, no spontaneous relapse after treatment withdrawal and parasite reactivation after immunossupression—were also achieved by other research groups using CYP51 inhibitors based on different molecular scaffolds [41,46,55].
Collectively, 9 is more efficacious in the treatment of the chronic phase of the disease with low parasite load. Although 9 did not eradicate cryptic reservoirs of parasites in vivo after 28 days of treatment, it successfully kept parasites under control and prevented the inflammation responsible, in part, for cardiac tissue damage. This outcome is not unique for inhibitors that target the ergosterol synthesis pathway. Treatment of chronically infected mice with N,N-dimethylsphingosine, an inhibitor of sphingosine kinase, also failed to produce a sterile cure, but reduced parasite load leading to a marked decrease in inflammation and fibrosis. Furthermore, there was a reduction of inflammatory mediators and an improvement of heart function measured as exercise capacity [56]. In addition, mice infected with resistant strains of T. cruzi showed decreased tissue parasitemia, reduced myocarditis and less electrocardiographical alterations after treatment with benznidazole, even though the drug failed to completely eliminate parasites in this model [57].
Several mechanisms have been proposed to explain the pathogenesis of Chagas’ cardiomyopathy, including parasite-dependent inflammation, autoimmunity, autonomic neuronal degeneration and damage of microvasculature [58,59]. Although more than one mechanism may be involved in Chagas disease pathogenesis, a consensus is that tissue damage is related to parasite persistence [58–60]. At the same time, 60–70% of infected individuals are asymptomatic. A balance between host and parasite in asymptomatic cases may be maintained by expression of the anti-inflammatory cytokine IL-10, while cardiomyopathy is associated with inflammation triggered by IFN-gamma and TNF-alpha [61]. Reducing the parasite burden diminishes inflammation even without complete elimination of the parasite [60]. In this regard, several non-randomized clinical trials have shown that etiological treatment of chronic patients with benznidazole resulted in slower progression to advanced stages of cardiomyopathy evaluated by electrocardiography and echocardiography [17,62].
Sterile cure is a highly desirable treatment outcome, however, it may not always be achieved, and drug discovery efforts are often hampered by deficiencies in understanding the nuances of disease pathogenesis. There is currently no sterile cure of HIV infection. The desirable outcome of the antiretroviral treatment is a long term plasma HIV-RNA count below 50 copies/ml [63]. The WHO recommends antiretrovirals in people of all ages, including pregnant women as soon as the diagnosis is made; once treatment is begun, it is recommended to continue throughout the entire life span without interruptions [63]. Benefits of treatment include a decreased risk of progression to AIDS and a decreased risk of death [64]. Highly active antiviral therapy options are available as drug ‘cocktails’ consisting of at least three medications belonging to at least two different classes of antiviral agents [65]. As of 2017, 19.5 million people are accessing antiretroviral therapy and more than half of all people living with HIV are on treatment [66].
By analogy with HIV/AIDS, the treatment option of non-toxic medications should be developed for Chagas patients to slow down progression to cardiomyopathy and to improve life expectancy and quality of life. With the scarce arsenal of anti-T. cruzi agents, the drug discovery community cannot afford to be prejudiced against CYP51, or any other target, if the inhibitors have acceptable safety profiles and achieve a marked reduction in parasite load, even in the absence of sterile cure. Regardless of the molecular target affected by the drug, development of an efficacious and safe treatment for Chagas disease would be a breakthrough for society, medicine and science.
Research performed at UC San Diego was conducted in compliance with the Animal Welfare Act and other federal statutes and regulations relating to animals and experiments involving animals and adheres to the principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council, 2011. The facility where this research was conducted is fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. Animal research was conducted under approved protocol S14187 from the Institutional Animal Care and Use Committee, University of California, San Diego. Research performed at Oswaldo Cruz Foundation—FIOCRUZ, Rio de Janeiro, Brazil, was approved by the Committee for Ethics in the Use of Animals of FIOCRUZ, under protocol number LW-37/13 and is in compliance with Brazilian Federal Law number 11794/08, Federal Brazilian Decree number 6899/09 and Brazilian Normative Resolution number 1 (July 9th, 2010) of the National Council for the Control of Animal Experimentation. Euthanasia was accomplished by CO2 inhalation or by sodium pentobarbital overdose (60 mg/kg), followed by cervical dislocation. These methods of euthanasia have been selected because they cause minimal pain and distress to animals, are relatively quick, and do not adversely impact interpretation of the results of studies. All methods are in accord with the recommendations of the Panel on Euthanasia of the American Veterinary Medical Association.
Compounds 9 and 12 were synthesized by following the procedures previously reported [35,37].
In vivo experiments were performed at the University of California San Diego (UCSD), La Jolla, California, USA and Oswaldo Cruz Foundation—FIOCRUZ, Rio de Janeiro, Brazil. At FIOCRUZ, Swiss Webster male and female mice weighting 18–20 g were obtained from CEMIB (Centro Multidisciplinar para Investigação Biológica), UNICAMP (Campinas, SP, Brazil). At UCSD, male and female 6 weeks old BALB/c mice, in the same weight range, were purchased from Jackson Laboratories (Farmington, CT, USA). Mice were housed in a maximum number of 5 animals per cage and kept in a conventional room at 20 to 24°C under a 12 h/12 h light/dark cycle. The animals were provided with sterilized water and chow ad libitum.
Acute toxicity was evaluated by administration of escalating doses of the compounds to male and female Swiss Webster mice (n = 2/group) orally by gavage, at 100 ul/hour 50 mg/kg dose formulated in 20% solutol (also known as Kolliphor HS15) (Sigma #42966). The general health of the animals was closely monitored for up to 48 h and the last dose before the onset of toxic symptoms were observed was defined as NOAEL according to the OECD guidelines. Cumulative toxicity after prolonged treatment was evaluated using BALB/c females (n = 5/group) treated orally by gavage with the experimental CYP51 inhibitors, 9 (25 mg/kg) or 12 (25 mg/kg), dissolved in 10% solutol, at 100 ul/dose, b.i.d, for 28 consecutive days. Mice were weighed once a week and their general health was assessed daily. After treatment, mice were euthanized, blood was collected for analysis of several biochemical markers of general health. Brain, heart, liver, kidney, gastro-intestinal (GI) tract and lungs were collected, briefly rinsed in PBS and fixed in buffered formalin solution including 10% formaldehyde, 33 mM NaH2PO4, 45 mM Na2HPO4 for histological evaluation.
The T. cruzi Y parasites were obtained from the bloodstream of infected Swiss Webster mice at the peak of parasitemia, as previously described [67]. Transgenic T. cruzi CL Brener parasites expressing a red-shifted luciferase that emits light in the tissue-penetrating orange-red region of the spectrum (a gift from Dr. John Kelly, London School of Hygiene and Tropical Medicine, London, United Kingdom), were obtained as described previously [39]. Epimastigote forms were maintained at 28°C in LIT media supplemented with 10% FBS and 100 μg/ml of antibiotic G418 to keep selective pressure in favor of the luciferase marker [68]. Epimastigotes were induced to differentiate to trypomastigotes through metacyclogenesis as previously described [69]. Metacyclic trypomastigotes were used to infect C2C12 myoblasts monolayers. After 5–7 days, trypomastigotes released in supernatant were collected by centrifugation for 15 min at 3300 rpm, re-suspended in DMEM and used to infect mice.
Swiss Webster male mice weighting 18–20 g were infected intraperitoneally with 104 bloodstream trypomastigote form of T. cruzi Y parasites. For bioluminescence imaging, six week old BALB/c male and female mice were infected by intraperitoneal injection with 103 T. cruzi CL-luc trypomastigotes derived from cell culture supernatant.
All drugs were solubilized in 10% solutol and administered orally, b.i.d, at previously optimized doses: 25 mg/kg for 9 and 12, 50 mg/kg for benznidazole, and 20 mg/kg for posaconazole. The treatment of Swiss mice acutely infected with T. cruzi Y strain was started with parasitemia onset at 5 days post-infection (dpi). The treatment of BALB/c mice infected with CL-luc parasites started at 14 dpi (acute phase), when parasitemia reached a peak as detected by bioluminescence, or at 126 dpi (chronic phase), when a chronic state of infection was established [46]. In all models, only parasite-positive mice (5 mice/group) were used in the treatment course lasting for 28 days. To assess if sterile cure was achieved, immunossupression was performed in the chronic model of infection 4 weeks after the end of treatment, with two doses of cyclophosphamide (200 mg/kg) by intraperitoneal (i.p.) injection at 3-day intervals.
In Swiss mice acutely infected with T. cruzi Y strain, parasites in the blood of each animal were quantified by using the Pizzi-Brener method. The total number of parasites are counted in 50 fields under 400X magnification of freshly prepared blood samples (5 μl drops) obtained from the tail veins of mice, collected 3 times a week, starting at 5 dpi and continued until the end of treatment [70]. Mortality was monitored daily and % survival was calculated using GraphPad prism software.
BALB/c mice infected with parasites carrying a bioluminescent marker were imaged at 13 dpi, before treatment was initiated, and then once a week, both during the 28-day treatment period and 39 days post-treatment, as previously described [35]. Briefly, mice were injected i.p. with 150 mg/kg D-luciferin potassium salt in PBS (Gold Biotechnology, St. Louis, MO), and 5 minutes later, anesthetized by isofluorane inhalation (3–5%) and imaged using IVIS Lumina in vivo imaging system (PerkinElmer, Waltham, MA) with 180s exposure time. Data acquisition and analysis were performed with the LivingImage V4.1 software (PerkinElmer, Waltham, MA). Uninfected controls were imaged in parallel to establish a negative threshold.
To evaluate sites of parasite persistence in BALB/c mice infected with T. cruzi expressing luciferase, we performed ex vivo imaging of selected internal organs according to the protocol adapted from Lewis et al., 2014 [39]. Briefly, the animals were injected i.p. with 150 mg/kg of D-luciferin, euthanized in a CO2 chamber and perfused with 10 ml of D-luciferin. Then, heart, liver, kidneys, lungs, spleen, mesenteric fat, skeletal muscle (excised from left thigh) and the whole gastro-intestinal (GI) tract were removed, placed in a petri dish with PBS containing 300 μg/ml of D-luciferin, and imaged using the IVIS Lumina system.
Brain, heart, liver, kidney, GI tract and lungs from uninfected mice for toxicity analysis, and heart and GI tract from infected animals were removed and fixed as described above. Samples were processed for routine histologic examination in the Histology Core of Moore Cancer Center (UCSD), embedded in paraffin, sectioned and stained with hematoxylin and eosin. The slides were scanned using Nanozoomer Slide Scanner (Hamamatsu Photonics, NJ, USA) and images were obtained through NDP viewer software (Hamamatsu Photonics, NJ, USA).
To quantify levels of inflammation, 5 random images of mouse hearts (10× magnification) were obtained from each animal, 5 animals/group. At this magnification, 5 fields comprise the majority of the area of the heart section. Image processing was performed using Fiji software [44], where cell nuclei was segmented through the Particle Analyzer plugin and the fraction of total area of the image occupied by all cell nuclei was then measured. Even though cardiomyocytes and cardiac fibroblasts cell nuclei are being counted together with inflammatory cells, uninfected heart sections were used as controls and provide a baseline number.
Terminal blood collection was performed via cardiac cavity exsanguination in uninfected and T. cruzi-infected mice. Blood was collected in serum separator tubes (Microtainer, BD Biosciences), allowed to clot for 0.5–2.0 h and then centrifuged for 90 s at 10000 g. Serum was removed and analyzed at the Central Animal Facilities of the Oswaldo Cruz Foundation (Rio de Janeiro, Brazil, CECAL/Fiocruz platform) using Vitros 250 (Ortho Clinical-Johnson & Johnson), or at the UC Davis Comparative Pathology Laboratory (Davis, CA, USA), where samples were analyzed using Roche Cobas Integra 400 Plus clinical chemistry analyzer. In both facilities, tests were performed for electrolytes and enzyme metabolites indicative of liver, kidney and cardiac functions, including alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin (BIL), albumin (ALB), alkaline phosphatase (ALP), blood urea nitrogen (BUN), creatinine (CRE), urea, calcium, phosphorus, glucose, total protein.
Student’s t-test was used for evaluation of differences in experimental data between groups. Values were considered statistically significant when p≤ 0.05. Statistics were analyzed by GraphPad Prism Software.
|
10.1371/journal.pgen.1007208 | A molecular basis behind heterophylly in an amphibious plant, Ranunculus trichophyllus | Ranunculus trichophyllus is an amphibious plant that produces thin and cylindrical leaves if grown under water but thick and broad leaves if grown on land. We found that such heterophylly is widely controlled by two plant hormones, abscisic acid (ABA) and ethylene, which control terrestrial and aquatic leaf development respectively. Aquatic leaves produced higher levels of ethylene but lower levels of ABA than terrestrial leaves. In aquatic leaves, their distinct traits with narrow shape, lack of stomata, and reduced vessel development were caused by EIN3-mediated overactivation of abaxial genes, RtKANADIs, and accompanying with reductions of STOMAGEN and VASCULAR-RELATED NAC-DOMAIN7 (VDN7). In contrast, in terrestrial leaves, ABI3-mediated activation of the adaxial genes, RtHD-ZIPIIIs, and STOMAGEN and VDN7 established leaf polarity, and stomata and vessel developments. Heterophylly of R.trichophyllus could be also induced by external cues such as cold and hypoxia, which is accompanied with the changes in the expression of leaf polarity genes similar to aquatic response. A closely-related land plant R. sceleratus did not show such heterophyllic responses, suggesting that the changes in the ABA/ethylene signaling and leaf polarity are one of key evolutionary steps for aquatic adaptation.
| Evolutionary adaptation into aquatic environment is widely observed in diverse clades of land plants. To understand the molecular basis behind such adaptation, we analyzed Ranunculus trichophyllus, an amphibious plant producing different leaf shape depending on the growth conditions. Aquatic leaves of this plant produce higher levels of ethylene, which causes overactivation of genetic circuits composed of EIN3, an ethylene signaling transducer, and abaxial genes that suppress genes regulating stomata and xylem development. In contrast, terrestrial leaves produce higher levels of ABA, which activates adaxial genes and causes activation of stomata and xylem developments. Such changes in the ABA/ethylene signaling and leaf polarity after submergence were not observed in the closely-related land plant R. sceleratus, indicating that they are key evolutionary steps towards aquatic adaptation.
| Since plants are sessile organisms, specific adaptations to their given environments are critical for their survival. Thus, compared to animals, plants show higher levels of phenotypic plasticity, differential phenotypic alterations exhibited in the same species depending on their surrounding environments [1, 2]. One of the most dramatic plant plasticity is heterophylly, an ability to produce morphologically different types of leaves depending on the growth environments [3]. Amphibious plants produce different shapes of leaves when grown under water compared to terrestrial growth; they usually produce thin and slender leaves in aquatic conditions but produce thick and stout leaves in terrestrial conditions [3]. Currently, the molecular mechanisms behind such heterophylly of amphibious plants are not well known.
Plant hormones participate in various plant developments so that plant architecture is shaped by the accurate regulation of the hormones [4, 5]. Plasticity by water adaptation is also regulated by plant hormones. Abscisic acid (ABA), auxin, ethylene, and gibberellin (GA) were proposed to mediate perception and responses to submergence into water [6, 7]. For example, auxin influences hyponastic growth and development of adventitious roots in submerged condition [6, 8]. Ethylene also regulates adventitious roots and rapid shoot growth when submerged, especially in deep water rice [7, 9]. Exogenous ABA treatment induces terrestrial leaf development in many aquatic plants whereas exogenous GA induces aquatic leaf development in some aquatic plants [3].
Leaf, as a photosynthetic organ, is a major plant organ showing plastic development depending on the environments [10, 11]. Leaves are developed from the shoot apical meristem as lateral organs and the leaf development is coordinated through three axes, a proximo-distal axis, an adaxial-abaxial (dorso-ventral) axis and medio-lateral axis. Adaxial-abaxial polarity has been well studied at the molecular level because establishment of this axis is critical for leaf morphogenesis [12]. Recent studies have identified several families of transcription factor genes determining adaxial and abaxial cell fate in the leaf [12, 13]. For example, KANADI (KAN) and YABBY (YAB) families and two AUXIN RESPONSE FACTOR genes (ARF3/ETTIN and ARF4) play critical roles in determination of abaxial cell fate whereas class III homeodomain-leucine zipper (HD-ZIPIII) genes, ARP (ASYMMETRIC LEAVES1, ROUGH SHEATH2, and PHANTASTICA) class Myb genes, and a LOB domain transcription factor, ASYMMETRIC LEAVES2 (AS2), determine adaxial cell fate [14–18]. In addition, the expression domains of leaf polarity genes are finely delimited by small RNA; i.e., miR165/166 degrades HD-ZIPIII transcripts in abaxial side and tasiRNA erases the transcripts of ARF3 and ARF4 in adaxial side of the leaves [14, 19]. Likewise, abaxial fate-determining genes act antagonistically with adaxial fate-determining genes. For example, the HD-ZIPIII genes are ectopically expressed in abaxial side of the leaves in kan1 kan2 double mutant, thus causing adaxialization [15]. In addition, overexpression of KAN2 causes reduced expression of PHB, a HD-ZIPIII gene [15], indicating that KAN genes suppress the expression of HD-ZIPIII. In contrast, gain-of-function of HD-ZIPIII causes adaxialization whereas loss-of-function of HD-ZIPIII genes like phb phv rev triple mutant causes abaxialization, indicating that HD-ZIPIII genes antagonistically suppress KANs [13, 20]. Interestingly, both abaxialization and adaxialization cause partial radialization of the leaves.
Land plants have evolved from aquatic plants, algae, in Silurian period, ca. 400 million years ago [21]. Afterwards, they have developed various traits for land adaptation such as vascular structure, stomata and seed development [22, 23]. In addition, they have evolved a plant hormone, abscisic acid (ABA), and ABA signaling to endure dehydrated environments [24]. During the evolutionary process, diverse families of plants re-colonized water and turned into aquatic plants [25, 26]. Although derived from diverse clades of land plants, many submerged plants share common phenotypes such as thin and cylindrical leaves [3]. As submerged plants are subjected to the same selection pressure, there may be a common evolutionary developmental (evo-devo) mechanism modified from the genetic circuits present in the terrestrial plants. Such evo-devo mechanisms have yet to be disclosed.
In this study, we delved into the potential evo-devo adaptive mechanism of an amphibious plant, Ranunculus trichophyllus var. kadzusensis, which is an endangered species in Korea that lives in rice pad. We hypothesized that amphibious plants are an evolutionary bridge between land and aquatic plants, thus, the elucidation of adaptive molecular mechanism in R. trichophyllus would provide an insight how land plants re-adapted to aquatic environments. Ranunculus is a widespread genus containing hundreds of species adapted to various habitats in the northern hemisphere [27]. Many Ranunculus species live near the water, and some species adapted to aquatic environments [27, 28]. Therefore, Ranunculus genus is a good model system to investigate how land plants recolonized aquatic environments.
Here we show that the heterophylly of an amphibious plant, R. trichophyllus, is widely controlled by two plant hormones, ethylene and ABA. The protoplast transfection assays, at least in cellular level, demonstrated our hypothesis indicating that ethylene, increased at aquatic conditions, induces ETHYLENE INSENSITIVE3 (EIN3)-mediated overactivation of abaxial genes, KANs, and suppression of STOMAGEN (STO) and VASCULAR-RELATED NAC-DOMAIN7 (VDN7), which cause cylindrical leaf morphology, lack of stomata, and reduced xylem development, three hallmarks of aquatic plants. In contrast, ABA, increased at terrestrial conditions, establishes leaf polarity through ABSCISIC ACID INSENSITIVE3 (ABI3)-mediated activation of adaxial genes, HD-ZIPIIIs, in adaxial side of the leaves, and induces STO and VDN7 for the development of stomata and vessel elements. In addition, we show that molecular changes have occurred in the expressions of ABA biosynthetic gene and leaf polarity genes in aquatic R. trichophyllus compared to a land plant relative, R. sceleratus.
Ranunculus trichophyllus is an amphibious plant that grows both on land and under water. Depending on their growth environments, they develop morphologically different types of leaves (Fig 1). Under terrestrial conditions, R. trichophyllus develops thick and broad leaves whereas under aquatic conditions it produces thin and cylindrical leaves (Fig 1A). The leaf index, a leaf length-to-width ratio, in aquatic leaves was approximately 10-fold higher than that in terrestrial leaves (Fig 1Ai), which is an indicative of slender appearance of aquatic leaves. Microscopic analyses showed that terrestrial leaves have well-developed stomata, particularly on adaxial surfaces, whereas aquatic leaves completely lack stomata (Fig 1Ab, c, f and g). When comparing cell structure, terrestrial leaves showed stout and irregular-shaped epidermal cells, whereas aquatic leaves showed slender and rectangular epidermal cells. We also observed that terrestrial leaves have a higher number of developed vessel elements than aquatic leaves (Fig 1Ad and h). In contrast, a close sister species R. sceleratus, which lives near the waterside, does not show such heterophylly (Fig 1B). The leaf index and leaf morphology of R. sceleratus, were not affected by 1 week submergence. In addition, stomata density and the number of vessel elements were little changed by submergence (Fig 1B). Moreover, R. sceleratus showed severe growth retardation by 3 weeks of long-term submergence, which is similar to Arabidopsis thaliana (S1 Fig). It suggests that the heterophyllic response of R. trichophyllus is evolutionarily adaptive trait for long-term submergence.
To understand the molecular basis of heterophylly in R. trichophyllus, we performed quantitative whole gene expression analysis using terrestrial leaves vs aquatic leaves by RNA sequencing. A total of 77,459 transcripts were analyzed, and ca. 15.8% of transcripts were up- or down-regulated in aquatic leaves compared to terrestrial leaves (Fig 2A). In general, the genes involved in the response to internal and external stimuli and stress-response genes showed significant up-regulation in aquatic plants (Fig 2B). Among the Gene Ontology (GO) terms related to stress response, ‘response to hypoxia’ (GO:0001666), ‘defense response’ (GO:0006952), ‘response to osmotic stress’ (GO:0006970) were prominent for up-regulation in aquatic plants, which may have evolved to protect plants from environmental stimuli and hypoxia stress in aquatic environments. We also found that genes involved in stomata and vascular developments were considerably down-regulated in aquatic leaves, which reflect lack of stomata and underdeveloped vessel elements. Moreover, well-known pathogen-resistance genes were up-regulated and genes for wax biosynthesis were down-regulated (Fig 2C). More importantly, the transcriptome analysis clearly pointed out that the two plant hormones, ethylene and ABA, are related to heterophyllic leaf development (Fig 2D). Such transcriptional changes may be required for evolutionary adaptation into aquatic environments. To test this hypothesis, we analyzed the effects of submergence on the expressions of the orthologous genes from R. sceleratus. In contrast to R. trichophyllus, the orthologous genes from R. sceleratus showed no significant differential expression in response to submergence (S2 Fig).
In addition to our transcriptome analysis, there are studies showing that ethylene, and GA can cause land-grown amphibious plants to develop an aquatic leaf-like morphologies [29, 30]. Thus, we wondered if any of the plant hormones effect on the heterophyllic development of R. trichophyllus seedlings (Fig 3). We found that exogenous ethylene treatment of terrestrial plants caused an increase of the leaf index, reduced number of stomata and vessel elements, whereas treatment of the aquatic leaves with silver nitrate (AgNO3), an inhibitor of ethylene biosynthesis, caused the opposite effects such that decreased leaf index and increased the number of stomata and vessel elements (Fig 3A). In contrast, when aquatic plants were treated with ABA, the leaf index was dramatically reduced whereas the numbers of stomata and vessel elements were increased (Fig 3B). GA treatment on the terrestrial plants did not reduce the number of stomata and vessel elements (Fig 3D and 3E). Likewise, paclobutrazol (PBZ), an inhibitor of GA biosynthesis, did not affect to stomata and vasculature development even though the leaf index was decreased (Fig 3C–3E). In addition, auxin and brassinosteroid (BR) treatments caused almost no effect (S3 Fig). The results suggest that aquatic leaf morphologies of R. trichophyllus are dependent on ethylene whereas terrestrial ones are dependent on ABA. GA, auxin, and BR do not appear to be involved in the heterophylly of R. trichophyllus. In contrast to R. trichophyllus, R. sceleratus did not show any morphological changes in response to ethylene and ABA (S4 Fig), indicating that ethylene and ABA signaling could control leaf development in R. trichophyllus but not in R. sceleratus.
Since the treatments of plant hormones indicated that ABA and ethylene mediates heterophyllic leaf development of R. tricophyllus, we analyzed the contents of ABA and ethylene in terrestrial and aquatic leaves (Fig 4A and 4B). As expected, terrestrial leaves contained 3 times higher level of ABA than aquatic leaves whilst aquatic leaves contained 4 times higher level of ethylene than terrestrial leaves. Then, we checked if expressions of any specific genes encoding the enzymes involved in the critical steps of ABA and ethylene biosynthetic pathways are differentially regulated according to environments. The gene encoding enzyme for critical step of ABA biosynthesis is ABA Aldehyde Oxidase (AAO) and that for ethylene biosynthesis is ACC Oxidase (ACO) [31, 32]. Thus, we cloned the orthologs of AAO and ACO from R. trichophyllus and compared the expression levels depending on the growth condition (Fig 4C and 4D). Consistent with the hormonal contents, terrestrial plants showed higher expression of RtAAO than aquatic plants whereas aquatic plants showed higher expression of RtACO than terrestrial plants in general. For ethylene biosynthesis, RtACO4B and RtACO4C showed remarkable increase in aquatic plants compared to terrestrial plants (Fig 4C). To address if submergence of land plants into water causes rapid changes in the expression of ABA/ethylene biosynthesis genes, we checked dynamic expressions of the genes at 7 hours, 1 day, and 2 days after submergence. The results showed that ethylene-biosynthesis genes, RtACO4B and RtACO4C, showed rapid increase within 7 hours, then slow increase until 2 days after submergence (Fig 4C). In case of ABA-biosynthesis genes, RtAAO1 and RtAAO3, the transcript levels were highly decreased within 7 hours of submergence but slowly increased afterwards (Fig 4D). We also checked if genes responsive to ABA and ethylene are increased in terrestrial and aquatic leaves of R. trichophyllus respectively. As expected, aquatic leaves showed higher expression of ethylene-responsive genes whereas terrestrial leaves showed higher expression of ABA-responsive genes (Fig 4C and 4D). Moreover, dynamic expression patterns of ABA/ethylene responsive genes after submergence showed rapid changes within 7 hours after submergence. Such results suggest that aquatic condition triggers in vivo ethylene signaling cascades and suppresses ABA signaling pathway.
In addition, we found that although submergence of R. trichophyllus into water rapidly downregulates expression of the ABA biosynthesis gene, RtAAO3 (ABA-aldehyde oxidase), expression of AtAAO3, an Arabidopsis ortholog, is not reduced, instead increased by submergence, perhaps due to hypoxic stress (Fig 4E). In R. sceleratus, a waterside plant, the expression of ortholog, RsAAO3, was reduced relatively weakly by submergence (Fig 4E). This result suggests that the suppression of ABA biosynthesis in aquatic environments is an evolutionary adaptation developed in amphibious R. trichophyllus.
To address the molecular mechanism behind heterophyllic leaf development, we explored the roles of several leaf development genes in the differential leaf morphologies in aquatic and terrestrial environments. Since the leaf structure and morphology is mainly governed by leaf polarity genes, we cloned three KAN and three HD-ZIPIII homologs, which determine abaxial and adaxial identity, respectively [14–16]. We named these genes KANa, -b, and -c, and HD-ZIPIIIa, -b, and -c (S5 Fig). Overexpression of RtKANa and RtHD-ZIPIIIa in Arabidopsis caused narrow or curling leaf morphology, which phenocopied the transgenic lines overexpressing Arabidopsis homologs (S5C Fig) [14, 16, 33]. The expression of the abaxial genes, RtKANs, was much higher in aquatic than in terrestrial leaves, suggesting that RtKANs are overexpressed in aquatic environments. In contrast, expression of adaxial genes, RtHD-ZIPIIIs, was significantly reduced in aquatic leaves (Fig 5A and 5B). In situ hybridization showed that RtKANa expression is mainly detectable around the midvein and abaxial side, but is not detectable in the adaxial side of terrestrial leaves (Fig 5F and S6 Fig). However, strong expression of RtKANa throughout aquatic leaves was observed (Fig 5H). In contrast, the expression domain of RtHD-ZIPIIIa was confined to the adaxial side of terrestrial leaves and was barely detectable in aquatic leaves (Fig 5J and 5L). These findings strongly support the hypothesis that the axial expressions of RtKANs and RtHD-ZIPIIIs are perturbed in aquatic leaves. In contrast to R. trichophyllus, Arabidopsis and R. sceleratus did not show any such alteration of the polarity gene expression following submergence (S7 Fig), indicating that these traits are acquired during re-adaptation to water.
Next, we investigated whether the expressions of leaf polarity genes are affected by ethylene and ABA. Under the terrestrial condition, treatment of plants with ethylene resulted in increased expression of the abaxial genes, RtKANs, thus, phenocopying aquatic leaves. However, the expression of adaxial genes, RtHD-ZIPIIIs, were not significantly affected (Fig 5C). Under aquatic condition, treatment of plants with ABA and AgNO3 led to increased expression of RtHD-ZIPIII genes, although no remarkable decrease in RtKAN expression was observed (Fig 5D). These results support the hypothesis that ethylene activates RtKANs, whereas ABA activates RtHD-ZIPIIIs as shown in the model (Fig 5E). The RNA expression pattern observed by in situ hybridization also supported these results: ethylene treatment increased the expression domain of RtKAN in terrestrial leaves whereas ABA treatment increased that of RtHD-ZIPIIIa in aquatic leaves (Fig 5I and 5M).
To address if ABA and ethylene directly regulate leaf polarity genes, RtHD-ZIPIIIs and RtKANs, we developed a Ranunculus protoplast transient expression assay using seedlings grown on solid MS media. The promoters of RtKANa and RtHD-ZIPIIIa were fused to the luciferase reporter gene (LUC) and tested for their response to ethylene and ABA in transiently transfected protoplasts. As expected, RtKANa promoter was rapidly induced by ethylene but was not affected much by ABA whereas the RtHD-ZIPIIIa promoter was strongly induced by ABA but was not significantly affected by ethylene (Fig 6A and 6D). This indicates that ethylene and ABA directly regulate the promoters of RtKANa and RtHD-ZIPIIIa respectively. Then, we searched for candidate transcription factors that might mediate the ethylene and ABA signaling by directly acting on the promoters of RtKANs and RtHD-ZIPIIIs. Interestingly, when RtEIN3 is cotransfected with proKANa-LUC into protoplasts, it caused strong activation of luciferase activity. Similarly, when RtABI3 is cotransfected with proHD-ZIPIIIa-LUC, it caused strong activation of luciferase activity (Fig 6B and 6E). This indicates that RtEIN3 and RtABI3 directly activate RtKANs and RtHD-ZIPIIIs respectively. Consistently, transfected RtEIN3 increased expressions of all three endogenous RtKAN genes and transfected RtABI3 increased all of RtHD-ZIPIIIs (Fig 6C and 6F), which supports the model shown in Fig 4E.
Finally, we wondered if ethylene and ABA signalings directly control heterophyllic leaf development through transcriptional cascades. Thus, we transfected upstream transcription factors, EIN3, ABI3, and HD-ZIPIII into protoplasts, then checked the expression of two key regulators controlling stomata and vessel developments; STO, encoding a peptide protein turning on stomatal development[34], and VDN7, encoding a NAC domain transcription factor controlling vascular development.[35] RtSTO and RtVND7 were down-regulated in aquatic leaves, which are consistent with the lack of stomata and reduced number of vessel elements in aquatic leaves (Fig 6G). Protoplast transfection assays showed that transient overexpression of RtABI3 or RtHD-ZIPIIIa increases transcript levels of RtSTO and RtVDN7, suggesting that stomata and vessel developments in terrestrial leaves are controlled by an ABI3-RtHD-ZIPIIIa regulatory module (Fig 6H). In contrast, overexpression of RtEIN3 decreases transcript levels of RtSTO and RtVDN7. Taken together, ethylene and ABA signaling control leaf polarity, stomata development and vascular development, the three hallmarks of heterophyllic development in R. trichophyllus.
It has been reported in some species of amphibious plants that certain environmental conditions such as cold cause aquatic leaf development mimicking aquatic condition [36]. Thus, we checked if any environmental conditions cause aquatic leaf development in R. trichophyllus (Fig 7). We found that 4°C cold temperature and hypoxia (less than 1% O2 concentration) caused significant increase of leaf index. In addition, the plants grown at cold temperature showed lack of stomata and decrease of vessel numbers, indicating that cold temperature mimics the aquatic condition well. However, hypoxia caused reduced number of stomata and vessel elements, suggesting that hypoxia mimics aquatic condition partially (Fig 7A). Then, we checked if cold and hypoxia effect on the expressions of leaf polarity genes similar to aquatic condition (Fig 7B and 7C). As expected, expression of KAN genes was higher whereas that of HD-ZIPIII genes was lower in the plants grown under cold temperature compared to room temperature. Consistent with the phenotypic effect, hypoxia caused less effects on the expression of both polarity genes than cold temperature. This result suggests that environmental cues inducing aquatic leaf development also cause similar molecular changes in R. trichophyllus.
Although embryophytes, land plants, have evolved from water to land by acquiring land adaptation such as vascular development and broad-leaf morphology, diverse plant species from different phylogenetic clades have returned to aquatic environments, indicating that water re-adaptation is quite common [21, 37]. R. trichophyllus is an amphibious plant, an evolutionary bridge between land and aquatic plants, and produces typical aquatic leaves if grown under water. In this report, we show that the heterophyllic leaf development of this plant is mainly determined by ABA and ethylene signalings which regulate leaf polarity genes. In aquatic environments, ethylene level is increased and the ethylene signaling overactivates the expression of abaxial genes, RtKANs, which antagonistically suppresses adaxial genes, RtHD-ZIPIIIs. Such overexpression of abaxial genes is most likely the molecular mechanism behind the cylindrical shape of aquatic leaves. In contrast, in terrestrial environments, ABA level is increased and ABA signaling activates the expression of adaxial genes, RtHD-ZIPIIIs, which establishes adaxial-abaxial polarity and causes broad leaf development (Fig 4E).
ABA is a well-known stress hormone in plants; it is accumulated by various abiotic and biotic stresses and confers resistance against them [38, 39]. Since flooding is also a stress to land plants, it is plausible that myriad of land plants show increased levels of ABA after submergence [40, 41]. Such ABA accumulation seems to be evolutionarily adaptive to land plants because it renders the plants to adopt ‘stunt strategy’; enduring the flooding period by inducing growth retardation which restricts energy consumption [42]. However, submergence-tolerant species such as deepwater rice and Rumex palustris have evolved differently. They show the opposite response to submergence in which ABA contents decreased [6, 43]. It indicates that reducing ABA level is adaptive to aquatic environments in some plant species. Consistent with this, ABA contents and the expression levels of RtAAO, a gene involved in a critical step of ABA biosynthesis, are decreased in R. trichophyllus in aquatic condition. It is likely that the suppression of ABA biosynthesis under water is widely occurred among submergence-tolerant plants and R. trichophyllus has also adopted similar evo-devo mechanism during evolution for aquatic adaptation.
Increase of ethylene level by flooding is also observed among various plant taxa, which is achieved by the enhancement of biosynthesis or local entrapment by submergence [7]. In addition, in wide range of plants from moss to Arabidopsis, the treatment of exogenous ethylene mimics submerged growth [44, 45]. Therefore, it is likely that the activation of ethylene signaling is a widely conserved response to submergence in plants. Consistently, ethylene biosynthesis and signaling are increased in R. trichophyllus by submergence. Interestingly, the antagonistic interaction between ABA and ethylene found in heterophylly of R. trichophyllus is also observed in many developmental processes in plants. For example, hyponastic growth of leaf in submerged R. palustris is regulated by the antagonistic interaction of ABA and ethylene [8]. Therefore, heterophylly of R. trichophyllus seems to be evolved from the common mechanism of ABA/ethylene interaction observed in other land plants.
Based on the roles of leaf polarity genes known in A. thaliana, differential expressions of leaf polarity genes in R. trichophyllus seem to lead to the three developmental changes which are required for adaptation to aquatic environments. That is, cylindrical shape leaves and reduced numbers of stomata seem to be caused by overactivation of the abaxial genes, RtKANs, as ectopic expression of KAN1 or KAN2 in Arabidopsis throughout the leaf primordia results in abaxialized radial organs, with a concomitant loss of HD-ZIPIII expression [14, 16]. Subsequently, the loss of RtHD-ZIPIIIs seems to cause reduced number of vessel elements, similar to loss-of-function of HD-ZIPIII genes in Arabidopsis showing reduced xylem [14]. Recent reports showed that a KNOX-GA module is critical for the heterophyllic development of Rorippa aquatica [36], which is a different mechanism than that we have found in R. trichophyllus. KNOX-GA module seems not to be a main mechanism for the heterophylly in R. trichophyllus. First, morphological pattern of heterophylly is not similar in the two species (S8A Fig). Heterophylly of R. aquatica is achieved by deep serration of leaves, thus it changes simple leaves to dissected compound leaves. In contrast, in case of R. trichophyllus, leaf complexity is increased in both terrestrial and aquatic leaves during growth and leaf branching pattern is not significantly different between terrestrial and aquatic leaves (S8A Fig). Instead, heterophylly of R. trichophyllus is achieved by leaf elongation and radialization rather than leaf serration. Second, GA has little effect on the heterophylly of R. trichophyllus although it is a main participant of the heterophylly in R. aquatica (Fig 3D and 3E). Thus, the heterophyllic developments in R. tricophyllus and R. aquatica have adopted different mechanisms, indicating that convergent evolution has occurred.
In spite of such differences in overall architecture, there is some convergent point between R. trichophyllus and R. aquatica. Like R. aquatica, aquatic leaves were generated by cold environment in R. trichophyllus (Fig 7). It is still unclear why amphibious plants induce aquatic leaf development in response to cold stress. Wells and Pigliucci (2000) proposed an “anticipatory plasticity hypothesis” in which plants can show similar phenotypic plasticity in response to different external cues coming together in nature. For example, submergence into water in nature causes combination of changes in diverse environmental cues, e.g. humidity, temperature, changes in light quality, etc. Thus, cold and hypoxia-induced molecular changes in the expression of leaf polarity genes in R. trichophyllus seem to be supportive to our hypothesis that changes in leaf polarity drive the evolution of amphibious adaptation in R. trichophyllus. The more direct evidence will be obtained through the analyses of mutants and transgenic plants which show defects in the heterophyllic development. In that sense, the recent report suggesting Hygrophilia difformis as a model plant to study heterophylly of amphibious plant is interesting [46]. H. difformis is easy to grow and propagated vegetatively well and above all, it can be easily transformed by Agrobacterium tumefaciens. However, the molecular basis of heterophylly of H. difformis is similar to that of R. aquatica but is different with that of R. trichophyllus. For example, in H. difformis, GA is a major regulator determining heterophylly and aquatic leaf development is achieved by deep serration of leaves, thus changing simple leaves to dissected compound leaves. Such characteristics are very similar with those of R. aquatic but dissimilar to those of R. trichophyllus. Therefore, H. difformis as a model plant for amphibious plants is limiting.
From comparative studies using two land plants, A. thaliana and R. sceleratus, and one amphibious plant, R. trichophyllus, we found that at least two molecular changes have occurred in R. trichophyllus during evolutionary adaptation to aquatic environments; the suppression of ABA biosynthesis and abaxialization of leaf development. Since ABA signaling components and the regulatory mechanism of HD-ZIPIII transcription factors seem to have evolved during land colonization by plants [24, 47], such molecular changes observed in R. trichophyllus are suggestive of evolutionary trend rearranging pre-existing gene networks instead of generating novel one [48]. Since many aquatic plants share similar morphological traits observed in submerged R. trichophyllus, further analysis of this plant will provide deep insight into the understanding of convergent evolution occurred in aquatic plants.
Seeds of Ranunculus trichophyllus var. kadzusensis were collected from its native habitat at Ganghwa Island, South Korea. Seeds of Ranunculus sceleratus, collected from Namyangju City, were donated from the Korea National Arboretum. Seeds were sterilized with 70% ethanol and with 1% NaOCl and 0.5% Tween-20 solution. Seeds were sown on half-strength Murashige-Skoog (MS) medium containing 50 μM carbenicilin, 75 μM cefotaxim, and 1% agar. Seeds of R. trichophyllus and R. sceleratus were germinated on solid MS media for 1 week, at that time root radicles were just protruded. Then germinated seedlings are transferred to aerial or aquatic/submerged environments. The true leaves produced at 7 days after transference were used for morphological analysis and RNA expression analysis. For transcriptome analysis, the plants 10 days after transference were used for RNA expression. In case of Arabidopsis, 4 days-old seedlings after germination were transplanted, then submerged into water for 2 weeks. For Arabidopsis thaliana, Col-0 seeds were used. The growth room was maintained at 22°C, 60 ± 10% relative humidity in long day photoperiod (16h light/8h dark).
cDNA libraries were obtained using 1 μg of total RNA extracted from whole plant tissues of R. trichophyllus. 100 base pair paired-end libraries were sequenced by Illumina HiSeqTM 2000. The libraries were quantified according to the qPCR Quantification Protocol Guide and qualified using an Agilent Technologies 2100 Bioanalyzer. RNA-seq reads were de novo assembled and mapped using Trinity and TopHat programs and the relative transcript levels were calculated by FPKM (Fragments Per Kilobase of exon per Million fragments mapped) using Cufflinks software. Excluded transcripts were filtered with 1 FPKM value and transformed to logarithm scale. They are normalized by quantile normalization method. Transcripts were assigned a putative function, then gene ontology analysis was performed by using DAVID tool (http://david.abcc.ncifcrf.gov/).
One-week-old seedlings of R. trichophyllus were used. The concentrations of hormones used were 10 μM NAA (1-naphthaleneacetic acid, Duchefa, N0903), 50 μM ACC (1-aminocyclopropane-1-carboxylic acid, Sigma Aldrich, A3903), 1 μM EBL (epi-brassinolide, Sigma Aldrich, E1641), 50 μM bikinin (Sigma Aldrich, SML0094) and 10 μM gibberellin (GA, bioWorld, 714248), respectively. For submerged treatment, abscisic acid (ABA, Sigma Aldrich, A0149), silver nitrate (AgNO3, Sigma Aldrich, S8157), and paclobutrazol (PBZ, Sigma Aldrich, 46046) were added into the aquatic media. The concentrations used were 1 μM ABA, 10 μM AgNO3, and 10 μM PBZ, respectively. After 10 days of growth, the first true leaves from the seedlings were analyzed.
For whole mount clearing, first true leaves were soaked in clearing solution (2.5 g chloral hydrate; 0.3 ml 100% glycerol; 0.7 ml distilled water), then incubated for 3 h at 55°C. The epidermis and xylem elements were observed using an Axio Imager A1 microscope (Carl Zeiss) under DIC optics. Images were captured using an AxioCam HRc camera (Carl Zeiss).
Seedlings (using about 20 individual seedlings) were grown in MS with or without 150 ml distilled water. Using 3 ml disposable syringe, we harvested capped air in headspace containing ethylene, then sealed by parafilm. Using Hamilton syringe, 100 μl gas was extracted from sealed air, then feeding to gas chromatography with flame ionization detector (Agilent 7890B GC). We used HP-5 column (#19091J-413, Agilent). Ethylene production was normalized by seedling weight. For measuring ABA contents, we used ABA ELISA kit (CSB-E09159Pl). Intensity of 450 nm fluorescence was determined by using Plate reader-Powerwave X (Bio-Tek). ABA production was normalized by sample weight. The measurements were performed from three biological replicates and two technical replicates each.
Candidate genes were selected using information of A. thaliana and the TAIR database (www.arabidopsis.org). The Arabidopsis sequences were used to search for orthologous genes from Aquilegia coerulea, for which the genome database (http://www.phytozome.net/search.php?method=Org_Acoerulea) is available. Aquilegia coerulea is the closest relative to the Ranunculus genus among plants that have an available sequence database.
For real time-qPCR, total RNA was isolated from leaves using TRI reagent (Sigma Aldrich, T9424) and RNeasy Plant Mini Kit (Qiagen, 74904). cDNA was generated using 4 μg of total RNA, 5 unit of reverse transcriptase (Fermentas, EP0442), 4 μl of 2.5 mM dNTP, 2 μl of 50 mM oligo(dT), and ddH2O to 40 μl. For real time qPCR, 0.3 μl of synthesized cDNA was mixed with 2 μl of 5 μM primers and 10 μl of SYBR Green qPCR Master Mix (Bio-Rad), and ddH2O to 20 μl. Real-time-qPCR analysis was performed by CFX96 Real-Time PCR system (Bio-Rad). The relative transcript levels were calculated according to the ΔΔCt method. [49]
Leaflets of R. trichophyllus which was grown on short days (8 h light/16 h dark, 22°C) were used for the isolation and transfection of protoplasts. The method of transfection was based on previously described.[50] For transfection, we used 10% PEG final concentration. After 1 day of incubation, the protoplasts were harvested for real time-qPCR and luciferase activity assays. For determining promoter activity, we used luciferase assay system (Promega, E1500) and microplate luminometer (Berthold).
Multiple alignment of amino acid sequences was performed using the ClustalX2.1 program (http://www.clustal.org/download/current/), which generates aligned phy format files. These aligned files were passed through the PHYLIP program (version 3.69) for phylogenetic analyses (http://evolution.genetics.washington.edu/phylip.html). In the PHYLIP software, SEQBOOT, PROTDIST, NEIGHBOR, and CONSENSE programs were run sequentially to generate draft unrooted phylogenetic trees and to obtain bootstrap values. The phylogenetic tree was drawn using the TreeView program (http://taxonomy.zoology.gla.ac.uk/rod/treeview.html).
All in situ hybridization experiments were performed as described previously.[51] For signal detection using NBT/BCIP (Roche, 11681451001), 100 ng of DIG-labelled RNA probes per mL of ALP buffer was used for hybridization. The images were obtained by light microscopy.
For fluorescence detection using HNPP (2-hydroxy-3-naphtoic acid-2'-phenylanilide phosphate, Roche, 1758888001), leaves were hybridized with DIG-labelled RNA probes, then stained with a mixture of 10 μl HNPP and 0.25 mg Fast Red TR solution per mL in ALP buffer containing 2 mM levamisole for 30 min at room temperature. Leaves were washed in distilled water for 10 min and incubated with 0.2 μg per mL DAPI (4,6-diamidino-2-phenylindole) for 10 min at RT for nuclear counter-staining. Fluorescence was detected by confocal laser scanning microscopy (LSM700, Carl Zeiss).
Statistical analyses were performed using an unpaired Student’s t-test. For multiple comparisons, we used a one-way ANOVA and post-hoc test. We considered P < 0.05 as statistically significant. All statistical analyses were performed using the statistical package R.[52]
|
10.1371/journal.pcbi.1006642 | Molecular basis for the increased affinity of an RNA recognition motif with re-engineered specificity: A molecular dynamics and enhanced sampling simulations study | The RNA recognition motif (RRM) is the most common RNA binding domain across eukaryotic proteins. It is therefore of great value to engineer its specificity to target RNAs of arbitrary sequence. This was recently achieved for the RRM in Rbfox protein, where four mutations R118D, E147R, N151S, and E152T were designed to target the precursor to the oncogenic miRNA 21. Here, we used a variety of molecular dynamics-based approaches to predict specific interactions at the binding interface. Overall, we have run approximately 50 microseconds of enhanced sampling and plain molecular dynamics simulations on the engineered complex as well as on the wild-type Rbfox·pre-miRNA 20b from which the mutated systems were designed. Comparison with the available NMR data on the wild type molecules (protein, RNA, and their complex) served to establish the accuracy of the calculations.
Free energy calculations suggest that further improvements in affinity and selectivity are achieved by the S151T replacement.
| RNA is an outstanding target for oncological intervention. Engineering the most common RNA binding motif in human proteins (called RRM) so as to bind to a specific RNA has an enormous pharmacological potential. Yet, it is highly non trivial to design RRM-bearing protein variants with RNA selectivity and affinity sufficiently high for clinical applications. Here we present an extensive molecular simulation study which shed light on the exquisite molecular recognition of the empirically-engineered complex between the RRM-bearing protein Rbfox and its RNA target pre-miR21. The simulations allow predicting a variant, the S151T, which may lead to further enhancement of selectivity and affinity for pre-miR21.
| The RNA recognition motif (RRM) is the largest family of eukaryotic RNA-binding proteins [1], involved in virtually all post-transcriptional regulatory events [2]. RRMs bind a wide-range of single-stranded RNAs [3], stem-loops and other RNA structures [2, 4–6]. Therefore, engineering RRM binding interfaces to target specific RNAs may create widely applicable tools for regulating gene expression [7, 8]. Yet, a variety of factors have hampered such efforts, including the complexities of the protein-RNA interactions, a poor understanding of the structural and biophysical basis for specificity, and the idiosyncratic way in which various RRM domains bind to RNA [9, 10].
Recently, some of us were able to engineer the conserved RRM domain of the human Rbfox protein by modulating its specificity for a target RNA [8]. The protein is part of a small family of tissue-specific alternative splicing regulators. It was chosen for its ability to bind with high sequence specificity and affinity -in the low nM range- to the r-GCAUG sequence in specific RNAs. These are the single-stranded RNAs and the hairpin microRNA precursors that code for miR107 and miR20b (referred to as pre-miR20b, hereafter, see Fig 1)[3, 6]. The r-G29AAUC33 sequence in the terminal loop of the chosen RNA target, the oncogenic precursor miRNA 21 (pre-miR21) [11], bears two nucleotide changes (at positions 30 and 33) from the r-GCAUG sequence. These mutations are sufficient to nearly abolish Rbfox binding [8]. The successfully engineered R118D-E147R-N151S-E152T quadruple mutant (Rbfox* hereafter, Fig 1) binds tightly to the pre-miR21 terminal loop sequence (Kd ~ 13 nM) [8], but also to pre-miR20b, with a dissociation constant only ~10 fold higher (Kd ~ 150 nM) [8]. Further improvements in binding specificity could be facilitated by understanding of the structural dynamics of key interactions at the protein-RNA interface at atomic level of description.
Molecular Dynamics (MD) simulations in explicit solvent are a useful tool to dissect the nature of interactions and specificity in biomolecular complexes [12, 13], providing information beyond what can be obtained experimentally. In particular, MD nicely complements NMR experiments on RNA interactions with RRM class of binding domains [14–16] by providing insights into specific interactions that are not revealed by experiments.
In this manuscript, we report the use of molecular simulation approaches to predict the structural determinants of the Rbfox*•pre-miR21 complex. After performing standard simulations, we use free-energy calculations to investigate a new mutant (S151T Rbfox*) that is predicted to improve selectivity towards the pre-miR21 target RNA relative to Rbfox*. The accuracy of our simulations is established by a comparison with the available NMR data and structure of the Rbfox•pre-miR20b complex [6, 17].
As a first task (i), we tested our computational setup by performing extensive MD simulations, extended to the microsecond timescale, on the Rbfox•pre-miR20b complex in explicit solvent. Note that the RNA hairpin loop is remodeled by the protein compared to the free RNA; in the complex the hairpin segment is larger to accommodate the protein [6]. The RNA/protein interface is stabilized by a number of intermolecular stacking interactions and hydrogen bonds, which provide tight sequence specificity for the nucleotide sequence–G29C30A31U32G33-. The MD results reproduce the available NMR structural data well and describe accurately the interactions at the binding interface. Comparison with simulations of the two isolated molecules (protein and RNA) suggests significant changes of the protein flexibility upon complex formation. Next (ii), we constructed a model of the binding interface of the Rbfox*•pre-miR21 complex by replacing G28, C30 and G33 with U, A and C, respectively, and by substituting R118, E147, N151 and E152 residues with D, R, S, and T, respectively. We performed a series of explicit-solvent simulations on the resulting model. To ensure adequate sampling of the conformational space of the mutated complex, we used an enhanced sampling method (Replica Exchange with solute scaling -REST2- [17] simulations). The results were consistent with affinity data.
(iii) To further cross-validate the simulations predictions on the Rbfox*•pre-miR21 binding interface, we performed two additional simulations on the Rbfox•pre-miR21 and Rbfox*•pre-miR20b systems. The molecular description of the interactions at the two binding interfaces is in qualitative agreement with the experimental binding affinity data.
Finally (iv), we used the simulated model of the Rbfox*•pre-miR21 complex to design a mutant with predicted higher affinity and selectivity.
The studied protein-RNA complexes are characterized by a complex interplay between the sequence and structural dynamics. Therefore, quantitative analysis of the simulation trajectories is not trivial. To do so, we have employed a wide range of different descriptors to characterize the protein conformational dynamics and plasticity (RMSD and PAD [18]), RNA structural variation (εRMSD [19] and structural parameters: torsion angles, base-pair and base-pair steps parameters), and to quantify the change in protein stability upon mutations and RNA binding (conformational entropy).
Recent MD studies on diverse protein-RNA complexes provided indication that the standard equilibration protocols, usually sufficient to equilibrate isolated medium-size RNA or protein molecules, might be inadequate for simulations of protein-RNA complexes [20]. Therefore, we performed multiple microsecond-long simulations, and exploited the existing NMR information as restraints at the early stages of most simulations (details provided in Materials and Method section).
The properties described below are calculated on the unrestrained parts of the initially restrained simulations of the system (Table 1, sim. 3–7 and 9–13), and on the fully unrestrained runs (Table 1, sim. 2 and 8) for comparisons. The individual trajectories sampled a similar conformational space (S1 Fig). Consequently, the average structural and dynamic properties calculated over the entire MD ensemble (all trajectories merged) do not significantly differ from those determined over the individual trajectories.
The root mean square deviation (RMSD) of the RRM domain (residues 117–193) in the Rbfox•pre-miR20b complex and in the free state in aqueous solution fluctuate around an average of 0.17 ± 0.02 nm, and 0.30 ± 0.02 nm, respectively, after only 200 ns (S2 Fig). This suggests that the systems are well equilibrated for most of the dynamic runs. These RMSD values are within the uncertainty of the NMR ensemble.
As expected [6], the protein RRM domain (residues 117–193) becomes generally more rigid upon RNA binding. Indeed, we calculate a substantial decrease in the per residue conformational entropy upon binding (S3 Fig).
Finally, the backbone flexibility, described here in terms of the so-called Protein Angular dispersion for the Ramachandran angles (PAD) [18], is larger in the free state than in the Rbfox•pre-miR20b complex. The larger the PAD value, the more flexible the protein backbone. The same analysis also allows identification of conformational transitions of the backbone during simulations [18]. These involve residues belonging to the β2 and β3 strands and to the β2β3 loop. This region is inserted into pre-miR20b terminal loop and anchors the RNA to the protein surface (Fig 2) in a manner reminiscent of the structure of the U1A complex (PDB 1URN, [23]). However, unlike U1A, the Rbfox RRM binds much more strongly to a single stranded RNA compared to a stem-loop with the same binding sequence[6]. The pronounced flexibility of the β2β3 loop might not be optimal for binding to structured RNAs [24, 25].
The RNA conformational ensemble in the Rbfox•pre-miR20b complex simulations is compatible with that of the NMR ensemble (S5 Fig). In particular, the conformational flexibility of nucleotide U27, not bound to the protein, is relatively large both in the NMR ensemble and in simulations. It dominantly contributes to the observed relatively large εRMSD values (S4 Fig). If calculated on the loop nucleotides directly bound to the protein (G28GCAUG33), the average εRMSD value is only 0.76 ± 0.14.
The backbone torsion angles values of the loop region (nucleotides 28–33) and the stem base pair parameters remain in agreement with those calculated for the structures of the NMR ensemble (S6 Fig).
In contrast, the simulations show immediate and large-scale conformational changes within the apical loop of the free pre-miR20b RNA (we reiterate that the RNA structure in isolation differs significantly from the complex). Unsatisfactory behavior of simulations of RNA hairpin loops has been widely analysed in literature [13, 26]. It is ascribed to accumulation of various inaccuracies in the force fields, such as an overstabilization of non-native base-phosphate and/or sugar-phosphate interactions, underestimated stability of the hydrogen bonding interaction in base pairing and various difficulties in describing the sugar-phosphate backbone substates [13]. Therefore, the description of the apical loop of the free pre-miR20b RNA can be expected to be less accurate than that of the RNA in complex with the protein. Indeed, in all simulations of the free pre-miR20b RNA (Table 1, sim. 2–7), performed with the standard AMBER force field (χOL3; described in Methods), the U27GGCAU32 loop is rearranged, and the original NMR conformation is never recovered afterwards (Fig 3, S7 Fig). Note that in the NMR structure, the loop is rigidly ordered and characterized by a U27-G28 stacking interaction and a G28-U32 type 3 base-phosphate (3BPh) interaction [27] (Fig 3).
To illustrate the changes in the loop conformations, we show the overlap of frames from each simulation and the NMR starting structure in S7 Fig. We observe high mobility of the G29 base, which flips around its χ torsion from anti to syn to stack with the G28 base (Fig 3), followed by the loss of the G28-U32 3BPh interaction. The C30 base is either bulged out or forms stacking interactions with A31. The high mobility of C30 and U32 results in a significant distortion of the backbone torsion angles (Fig 3). Due to our inability to reproduce the NMR structure of the isolated RNA hairpin, we did not attempt any simulations of the isolated RNA starting from its conformation seen in the RNA/protein complex.
The pronounced flexibility of the terminal loop, however, does not affect the pre-miR20b stem dynamics, as shown by a comparison between the base pair and base–pair steps parameters of the simulated and NMR-solved ensembles of structures (S8 Fig).
Likewise, the inability to reproduce the experimental conformation of the apical loop in simulations of the free pre-miR20b RNA should not affect our investigations of the protein-RNA complex where the loop is sufficiently stabilized by the protein, in addition, the splayed RNA conformation is likely less strained than in the isolated hairpin loop as suggested by the experimental data [6].
It has been demonstrated that using the revised van der Waals phosphate’s oxygen parameters reported in ref. [21], along with the 3-charge, 4-point OPC water model [22] partially improves simulations of RNA tetranucleotides by stabilizing the native A-form like conformations [28, 29]; this protocol is referred as χOL3-CP-OPC force-field combination, hereafter. However, for the pre-miR20b loop, the χOL3-CP-OPC force field-based simulations did not achieve better accuracy than those based on the parent χOL3 force field (Table 1, sim. 1–3 (χOL3-CP-OPC)); a similar unsatisfactory outcome was reported also for other hairpin-loop systems investigated recently [26]. Indeed, the pre-miR20b χOL3-CP-OPC simulations showed syn/anti nucleobase flips and alternative stacking conformations for all the terminal loop nucleotides (S6 Fig). Since no improvement was detected with this protocol, we did not attempt χOL3-CP-OPC simulations of the protein-RNA complexes. The use of the modified phosphate parameters, while improving the A-form single-strand simulations, might destabilize for example some native BPh interactions in folded RNAs. Note that simulations of free RNA hairpin loops remain a fundamental challenge for all currently available RNA force fields [13][30].
Analysis of average ion occupancies [31] revealed an average local concentration of approximately 1 M sodium near the U27/G33 base pair in simulations of the free RNA. This is a significantly elevated ion concentration, well above the bulk value. Here, the Na+ ions interact with the (U27) O4 atom with a residency time of few ns, at a distance of ~ 0.25 nm (S9 Fig). This ion-binding site is completely absent in the complex structure since the U27/G33 base pair is disrupted by protein binding. Instead, in the complex, the Na+ concentration is strongly localized at the dinucleotide step 34–35 (S9 Fig).
The interface structure sampled in MD simulations is similar to that of the NMR-resolved ensemble (Fig 4). In particular, G29 stacks with F126 and R184. The G29/R184 stacking interaction is always observed in the simulations (S10 Fig) even though it is absent in some frames of the NMR ensemble. The network of interactions is further stabilized by a bifurcated hydrogen bond involving the I124 backbone and the G29 base (Table 2). As in the NMR ensemble, G29 and A31 form a trans Watson Crick/Shallow groove (tWS) base-pair [32] in the simulations. The G29/A31 interaction is further stabilized by a water molecule, coordinated by the A31 N7 atom and C30 phosphate group (Fig 4). The adenine base forms water-mediated hydrogen bonds with the S122 and K156 side chains. The residence time of water molecules in these interactions are of a few tens of ns, sensibly longer than the common time-scale (50–500 ps) of short-residency hydration sites around RNA molecules [33–35]. These results are fully consistent with the earlier report of structured hydration sites in the simulations of the Rbfox RRM in complex with single-stranded RNA [15].
In our simulations, the U32 base forms stacking interactions with H120 (S10 Fig) and hydrogen bonds with the backbone of N190 and T192 (Table 2, Fig 4). G33 base is in syn conformation and forms stacking interactions with F160 (Fig 4, S10 Fig) and hydrogen bonds with T192 backbone and R118 side chain (Table 2, Fig 4). The latter interaction, always observed with good hydrogen bond geometry (Table 2) in the simulations, is present in eight conformers of the NMR ensemble, while in the other twelve the residue is solvent-exposed, perhaps due to insufficiently clear experimental information. Notably, the R118 side chain forms a very similar hydrogen-bonding interaction in the NMR structure of the Rbfox-RRM bound to the single-stranded RNA r-U1G2C3A4U5G6U7 [3].
Next, we present several comparisons with the primary NMR data, which is a very stringent way to judge the accuracy of simulations [14].
On average, the back-calculated Chemical Shifts (CSs) for 13C and 1H atoms of free and bound pre-miR20b, along with the 13C’, 13Ca, 13Cb, and 15N atoms of Rbfox in its complex with pre-miR20b are, within the accuracy and the limits of the empirical methods (LARMORd [36] and SHIFTX+ [37]) used for the predictions (S11 Fig), in fair agreement with experimental observations.
The agreement between observed and calculated chemical shifts is only fair because of a variety of reasons. These include the fact that SHIFTX+ is expected not to be able to accurately predict shifts of residues in close proximity to RNA. Indeed, the characteristic ring current and charge for non-protein like molecules are not included in the SHIFTX2 parameterization [38]. LarmorD suffers from the same drawback as it was parameterized by excluding RNA’ structures in complex with proteins or other ligands in its training data set [36]. Moreover, the apparent agreement (S11 Fig, Pearson correlation coefficients R = 0.99 for 13C and R = 0.97 for 1H) with the measured chemical shifts for the free pre-miR20b RNA (S11 Fig), which shows large conformational changes in simulations, suggest that LarmorD sensitivity to structural changes might be limited.
However, the SHIFTX+ predictions were still sensitive to spurious motions of F150 and F158 as shown by the calculated distributions of the N nuclei, which are characterized by multiple peaks (S13 Fig). This might be related to temporary flips of the χ1 dihedral angles from gauche(+) to gauche(-) during the simulations (S13 Fig). A similar pronounced flexibility for phenylalanine and tyrosine side chains was observed also in our previous simulations using the ff14SB force field [15, 16]. This potentially erroneous behavior can be related to the energy barrier for side chain rotation around the χ2 angle [15], which might be too low in the employed force field.
The chemical shift predictions for the Ca nuclei of residues V146, E147, and V151, located on β2 strand, deviate from the experimental values beyond two standard deviations (S12 Fig). In this case, the divergence reflects the dependence of the carbon shifts on backbone φ and ψ angles, for which we already described enhanced fluctuations during the simulations (Fig 2).
None of the back-calculated CSs of the bound RNA show significant differences compared to the experimental values (S15 Fig).
On average, ~84% of the NOE upper bounds are satisfied for the isolated protein and RNA and only ~3% of the violated NOE have violations greater than 0.05 nm (S1 Table). Not unexpectedly, in free RNA, the larger NOE violations are typically localized to the G28, G29 and U32 nucleotides of the apical loop, which exhibit major conformational changes in the simulations (see above).
For the complex: on average, the percentage of satisfied NOE upper bounds is about 76%. Interestingly, the single trajectory generated without the initial application of the NMR restraints (13), shows the largest number of violations (Table 3). This indicates that the application of the experimental restraints in the early part of production trajectory leads to visibly better agreement with the inter-molecular NOEs in the subsequent unrestrained trajectories. However, this assessment should be taken with some caution, because it is made based on only a single unrestrained trajectory. The greatest distance violations (>0.3 nm) are observed for the inter-molecular NOEs involving amino acids F150 and F158 and nucleotides U25, and U32, possibly because of the relatively high flexibility of the two phenylalanine side chains noted above.
Altogether, these analyses demonstrate that the MD-derived conformational ensemble of structures reproduces fairly well the experimentally sampled conformations for the Rbfox•pre-miR20b complex. These and previous [14] results suggest that our protocol can be employed to study the dynamic properties of the engineered Rbfox*•pre-miR20b* complex and to compare it with the wild type complex from which it was designed.
One of our main goals was to perform atomistic simulations of the Rbfox*•pre-miR20b* complex, for which the experimental structure is not available, and characterize the molecular interactions at its binding interface. The complex features R118D, E147R, N151S and E152T mutations on Rbfox, as well as G28U, C30A and G33C mutations in the pre-miR20b RNA. To achieve wide exploration of the engineered binding interface conformational space, we used enhanced sampling methods, and specifically Hamiltonian replica exchange (HREX) MD simulations. These have become a common way to elucidate conformational ensembles of proteins [39–41] and nucleic acids [42, 43]. HREX simulations should eliminate any bias caused by the initial building up of the mutated structure. A wealth of recipes for HREX has been proposed in the last years (among others, [39, 44–50]). One of the most successful approaches is the so-called replica exchange solute tempering in its REST2 variant [17], in which only the solute Hamiltonian is scaled. Being mainly interested in the properties of the Rbfox*•pre-miR20b* binding interface, we have used a promising cost-saving variant of REST2, where only this part of the solute is scaled (Methods, REST2 PS, Table 1, sim. 15). However, a standard REST2 simulation was also performed (Methods) for comparison (Table 1, sim. 16). A discussion of convergence issues of these types of simulations is offered in the SI.
Overall, the structure of the mutated complex remains very similar to that of the wild type, while the flexibility is reduced (Fig 5 and Table 4). Unlike in the wild-type, no backbone conformational transitions are observed for the β2β3 loop in the MD simulations of either the free (Table 1, sim. 14) or the bound (Table 1, sim. 17–18) Rbfox* (Fig 5).
The results of this analysis are consistent with a considerable loss of per-residue conformational entropy of the Rbfox* β2β3 loop residues (Table 4 and S4 Fig). Upon RNA binding, the protein and the β2β3 loop become even less mobile. This is shown by a calculation of the so-called PAD values, which provide a measure of proteins’ backbone conformational flexibility and of the conformational entropy differences (Fig 5 and Table 4).
This increase in rigidity of the protein structure, and in particular of the loop—whose stiffening might provide a better steric fit for RNA binding [23]—might contribute to the 2-fold increase of binding affinity for E152T relative to the wild type protein [8].
At the RNA/protein interface of the mutant, the hydrogen bonds, and stacking interactions of the G29, A31 and U32 bases (parts of the RNA/protein interface which are not mutated) observed in the wild-type simulations (Fig 4) are also preserved in the mutant complex (Fig 6, Table 5, S16 Fig). Interestingly, the water molecules coordinated by A31, S122 and K156 side chains as in the wild-type complex exhibit slow exchange with bulk solvent, with residence time of tens of ns (Figs 4 and 6). This leads us to suggest that these hydration sites could be indeed important in stabilizing the binding interface [15].
Overall, the simulations convincingly suggest that the system is able to structurally tolerate the mutations without altering the overall Rbfox*•pre-miR20b* binding mode.
A30 forms, about 30% of the time, a hydrogen bond with the S151 backbone (Table 5). This observation is consistent with the experimental data which shows only slight improvement of the mutant binding affinity relative to the wild type protein•pre-miR21 complex upon the N151S mutation [8]. Note that the S151 side chain mostly interacts with the N6 atom of A30 in simulations, instead of the N1 atom, as was originally suggested (8). When not present, the S151/A30 hydrogen bond is most often replaced by an intramolecular S151/G154 interaction (Table 5), which contributes to stabilizing the β2β3 loop.
C33, in the anti conformation (G33 in Rbfox•pre-miR20b complex was in syn), establishes either direct or water-mediated hydrogen bonds with the R147 guanidinium group (Fig 6). There are also hydrogen bonds with D118 and T192 side chains in most of the simulations (Table 5, Fig 6). This is consistent with the larger binding affinity gain of the R118D-E147R mutant for pre-miR21 relative to that of the wild type protein [8]. Indeed, these two mutations most significantly, increased the binding affinity of the mutant protein to pre-miR21 (by ~102 fold) compared to the wild type protein•pre-miR21 complex [8]. Lastly, we note that the A31 forms an intermolecular stacking interaction with residue R153 within the β2β3 loop. This interaction is absent in the wild type complex. We suggest that the network of intermolecular interactions shown by our simulation is qualitatively consistent with the experimentally measured affinities and the rationale behind the design of the mutations [8].
To further investigate the accuracy of our predicted interactions at the engineered complex binding interface, we performed two 1-microsecond long simulations (Table 1, sim. 23 and 24). The first focuses on the Rbfox•pre-miR20b* complex and shows that the interface is solely maintained by the stacking interactions of G29 with F126 and R184 along with a hydrogen bond between I124 backbone and the G29 base (Table 6 and S18 Fig). This finding is qualitatively consistent with electrophoretic mobility shift assay experiments, which indicate an extremely weak binding for these substitutions [8]. The second simulation is carried out on the Rbfox*•pre-miR20b complex. The binding interface resulting from our simulation features equivalent hydrogen bonds and stacking interactions (Table 6) as observed in the wild type (Table 2 and Fig 4) and the Rbfox*•pre-miR20b* (Table 5 and Fig 6) complexes. These interactions involve the G29, A31 and U32 bases. The RNA does not interact with the protein’s β2β3 loop, and, in particular, with the mutated S151 and T152. Most notably, G33 maintains its syn conformation (as in the Rbfox•pre-miR20b complex) and forms a hydrogen bond with the mutated R147 side chain. The latter in turn forms a hydrogen bond with the mutated D118 (S19 Fig). Hence, the simulation suggests a strong compensatory effect upon amino acids substitution at this site as new interactions are formed to maintain complex stability. This may be consistent with the good binding affinity of the mutated Rbfox* protein features for the pre-miR20b terminal loop [8] (see Introduction).
These results, although based on single trajectories, further establish the simulations predictive power through their qualitative agreement with the experimental binding assays.
Based on overall consistency between predictions and available experimental data, we sought to identify a mutation, which would further improve affinity and selectivity for the target pre-miR21b RNA. Specifically, our simulations show that the N151S substitution as suggested by the experiments [8] does not lead to significant interactions with the RNA, possibly because of the intrinsic flexibility of the protein β2β3 loop. We therefore reasoned that placing a bulkier group, such as Thr, in position 151 would be advantageous. Both S151 and T151 are capable of forming the same H-bonds with the N1 and/or the N6 atoms of A30. However, the bulkier side chain of T151 might influence the dynamics of the β2β3 loop. We therefore investigated the structure of S151T Rbfox*•pre-miR20b* by MD simulations (Table 1, sim. 19–20) and the change in affinity upon the S151T mutation by alchemical calculations using non-equilibrium approach (see Methods) [51]. The method has been successfully applied to a variety of protein mutants[51], and more recently, to protein–DNA–mutant complexes [52], providing accurate free energy estimates [52]. We refer to other works for a detailed comparison between the alchemical non-equilibrium and the equilibrium free energy calculations [53, 54].
While the basic structure of the complex is overall unaffected, the A30(N6) forms hydrogen bond with the S155 backbone oxygen, while the A30(N1) forms a hydrogen bond with the T151 hydroxyl group (Fig 7). These interactions might decrease the flexibility of the β2β3 loop compared to the previous mutant (Fig 7), and lead to an indirect stabilization of the position of the U28 base, which is able to form a stable H-bond with the S155 backbone oxygen, and a stacking interaction with F126 (Fig 7). None of these interactions are observed in the Rbfox*•pre-miR20b* complex, where the U28 base is always solvent-exposed. Note that identical binding pattern for the U28 was also observed in earlier MD simulation studies of the wild type Rbfox complexed with a single-stranded RNA [14, 15]. The T151 thus might be better in overall accommodation of the pre-miR20b* RNA than the S151.
The free energy change associated with the S151T mutation, calculated using computational alchemy over two different simulation time windows (see Materials and Methods), is either -1.2 ± 0.3 kcal mol-1 or -1.3 ± 0.1 kcal mol-1. Hence, this estimation appears to be well converged and suggests that the mutation increases to a small, yet significant extent the affinity of the complex.
The presence of the U28 /protein interactions might also improve the selectivity of this mutant for the r-U28GAAUC33 sequence in pre-miR20b* RNA over r-G28GCAUG33 found in pre-miR20b. Indeed, in the wild type complex, G28 (equivalent to U28) exhibits pronounced flexibility in simulations and does not form any contact with the protein, ([6] and this work). Hence, our simulations suggest that the proposed mutation would alter the preference of the binding interface for the pre-miR20b* sequence over the pre-miR20b RNA, improving both the affinity and the selectivity of the engineered protein for the target pre-miR21 RNA.
MD simulations of protein-RNA complexes remain somewhat limited by practical considerations of sampling (i.e. simulation time-scale) and inaccuracies resulting from force-field limitations [12, 20], yet they can supply important insight that often cannot be obtained by experiments, specifically on free-energy contributions and persistence of intermolecular contacts. The MD simulations in explicit solvent conducted here, covering overall about 50 microseconds of simulation data, including several state-of-the art simulation techniques and validated by their full consistency with experimental data, provide a detailed atomistic picture of the effect of mutations in the Rbfox*•pre-mir20b* interface. The simulations also suggest a new mutant, S151T, which is predicted to be more selective and have higher affinity for the pre-miR-21 sequence than the S151 suggested in the original design.
We used the lowest energy structures of the NMR ensembles 2cq3 [55], 2n7x [6] and 2n82 [6] as starting structures for the simulations of the free Rbfox protein, pre-miR20b RNA and Rbfox-pre-miR20b complex, respectively.
The starting structure of Rbfox* was prepared by introducing R118D, E147R, N151S and E152T mutations into Rbfox (in both free and bound states) using the Swiss MODEL software (available at https://spdbv.vital-it.ch/) [56–59]. The starting model of pre-miR20b* was obtained by replacing the G28, C30 and G33 in the pre-miR20b structure with U, A and C, respectively, using the tleap module of Amber 16 (available as AmberTools16 at http://ambermd.org/AmberTools16-get.html) [60]. The pre-miR20b* sequence (r-G19GUAGUUUUU28GAAUC33ACUCUACC41) is equivalent to that of the pre-miR21 only in the terminal loop (nucleotides 28–33: UGAAUC). This is part of the protein-RNA interface. The remainder of the sequence does not interact with the protein and was therefore left unchanged and identical to the pre-miR20b (r-G19GUAGUUUUG28GCAUG33ACUCUACC41).
The molecules were solvated in truncated octahedral water boxes with minimal distance of 0.10 nm between solutes and the box border. The solutes were neutralized with sodium ions followed by addition of a sufficient number of Na+/Cl- ion pairs to reach the excess salt concentration of 80 mM. Similar solvent conditions were shown to work well for other protein-RNA systems [14, 15, 20]. The topology and coordinate files for the simulations were prepared using the tleap module of Amber 16 [60].
TIP3P [61], Joung and Cheatham [62], and the amber ff14SB [63], and χOL3 [64] force fields were used for water, ions, proteins, and RNA respectively. This combination has shown satisfactory behavior with other protein-RNA complexes [14].
We performed also a second set of MD simulations of the free pre-miR20b RNA. These were carried out in exactly the same way as the first set except that they included the recently suggested modification of van der Waals oxygen radii for organic phosphates (atom types O2, OH, and OS)[21], along with the OPC water model [22].
All systems were subjected to energy minimization, and equilibration using a standard equilibration protocol [20]. In order to reduce the likelihood of instabilities in the production runs [14], NMR restraints, when available, were applied in the early stages of the majority of the simulations of the pre-miR20b RNA (Table 1, sim. 3–7) and of the Rbfox•pre-miR20b complex simulations (Table 1, sim. 9–13). Specifically, after the initial standard equilibration, the systems were simulated in the following way: 0–100 ns—all available NMR hydrogen restraints (both inter- and intra-molecular NOE interactions) were utilized, 100–120 ns—only protein–RNA (intermolecular NOE) restraints were utilized, and after 120 ns—entirely unrestrained simulations were conducted. The aim of the procedure is to guarantee a sufficient equilibration of the systems before data is gathered. Since the restraints are lifted in the later stages of the simulations, they do not bias the results. Only the primary NMR data (NOE distance restraints) were used, and were introduced with a flat-well potential [14]. Earlier, this approach was shown to be able to prevent the abrupt structural disruptions which can otherwise occur in beginning of MD simulations of protein-RNA complexes. By giving the structures more time to relax without immediate deviations from the NMR ensemble, it is possible to achieve more stable simulations of protein-RNA complexes [14]. Some simulations were also performed without the initial use of NMR restraints (Table 1, sim. 2 and 8). For detailed discussion of this protocol see [14].
Covalent bonds involving hydrogen atoms were constrained using the SHAKE algorithm [65]. Periodic boundary conditions and a 2 fs integration step were employed. The particle mesh Ewald (PME) approach[66] was used for handling electrostatic interactions. The cut-off distance of the non-bonded Lennard-Jones interactions was 0.9 nm. We used the Nose−Hoover thermostat [67] and Andersen−Parrinello−Rahman barostat [68] to maintain the systems at a temperature of 298 K and pressure of 1 bar, respectively. The completely unrestrained simulations were performed using GROMACS 5.1 (http://www.gromacs.org) [69]. Simulation runs initially using the NMR restraints were performed with the pmemd module of AMBER 14 (http://ambermd.org) [70].
In order to provide proper sampling of the Rbfox*•pre-miR20b* binding interface conformational space, we performed two distinct Replica Exchange with Solute Scaling (REST2)[17] simulations. The method is based on a modification of the potential energy, so that the interactions between solute atoms are scaled by a factor λ, solvent–solvent interactions remain unscaled, and solute–solvent interactions are scaled by λ1/2. Scaling the energy by a factor λ is equivalent to scaling of the temperature by 1/λ. Thus, in the case of REST2, only the solute atoms are effectively heated up in REST2. Solvent–solvent interactions that typically contribute the most to the energy differences between replicas, do not contribute to exchanges, allowing to effectively reduce the number of replicas and the computational cost [17].
In a first simulation run (Table 1, sim. 15 (REST2 PS)), we explored the possibility to enhance sampling of the mutated binding interface only, by rescaling the force field parameters of the nucleotides A30 and G33 along with their flanking phosphates and the protein residues within 0.5 nm of those nucleotides (the complete list of included atoms is reported in S1 Table). Eight replicas were simulated with scaling factor λ ranging from 1 (reference replica) to 0.6, according to a geometric distribution, and leading to an average acceptance rate of 22%. Each replica was simulated for 2 μs, giving a cumulative time of 8 x 2 μs = 16 μs. For this simulation, an in-house modified Amber 16 version was used and the same simulation setup described above was adopted.
The above-proposed simulation protocol requires decoupling the degrees of freedom of the binding interface from rest of the system, but this procedure might affect fundamental molecular properties such as electrostatics and hydrophobicity [71]. Therefore, to test the accuracy of the calculations, a second REST2 simulation was conducted using a standard protocol, namely rescaling the force-field parameters of the entire solute. In this case, the Hamiltonian Replica Exchange (H-REX) code [71] as implemented in the Plumed-HREX patch of Gromacs 5.1 (https://plumed.github.io/doc-v2.3/user-doc/html/hrex.html)[71] was used. Sixteen replicas of the system were simulated, with the setup described above. A geometrical distribution of sixteen λ values ranging from 1 to 0.7 was used, which resulted in an average acceptance rate of ~20%. Each replica was simulated for 1 μs (Table 1, sim. 16 (REST 2)).
A cluster analysis was performed to identify the most populated conformers in the total simulated ensemble. In order to ensure that the clusters found would be consistent across both REST2 runs, clustering was performed on the combined trajectory obtained from the two reference (unbiased) trajectories. The k-means clustering algorithm implemented in cpptraj module [72] of Amber 16 [60] was used based on the Root Mean Square Deviation (RMSD) of the interface of the protein-RNA complex (nucleotides 28–33 and the amino-acids residues within 0.45 nm from those nucleotides). The combined clustering results were also parsed to obtain results for each individual simulation [73, 74].
A representative structure for each cluster was identified as the centroid conformer of the cluster (i.e., the trajectory frame with the lowest cumulative RMSD distance to every other point in the cluster). Subsequently, two additional unbiased MD simulations (Table 1, sim. 17–18) were started from the representative structures of the two most populated clusters (accounting for ~44% of all structures). Here we used Gromacs 5.1 [69] and the same protocol described above.
To compare the conformational space sampled by the two REST2 simulations and their efficiency with respect to conventional MD, we estimated the probability density ρ(x) of observing the system in a state x using a Gaussian kernel density estimate [75] (Gaussian KDE) along two collective variables (CV) [76].
Overall changes are described by the differences in the distribution of reciprocal interatomic distances (DRID)[77] with respect to the representative structure of the most populated cluster. The distribution is evaluated from the inverse intra-molecular distances between all the Ca and the P atoms of the protein and RNA. For each Ca and P inverse distance distribution, three features are extracted: the mean, the square root of the central moment, and the cube root of the third central moment. This assigns a feature matrix vn∈R3×N to each configuration n. The difference between configuration n and the reference structure is then
DRID=13N∑i=1N‖vn(∙,i)−v0(∙,i)‖
(1)
where N is the number of residues, vn(∙,i) denotes the feature vector of the ith Ca or P atom in vn, and v0 is the feature matrix of the reference configuration.
Local changes were captured from the fraction of conserved contacts Q between the protein and the RNA at the binding interface. Q is defined via a list of contact pairs between the heavy atoms i of residues 28–33 of the RNA loop and the heavy atoms j of the protein residues:
Q(x)=1N∑(i,j)11+exp[β(rij−λrij0)]
(2)
where the sum runs over N pairs of contacts (i,j), rij(x) is the distance between i and j in configuration x, rij0 is the distance between i and j in the reference conformation, β is a smoothing parameter taken to be 0.5 nm and the factor λ, taken to be 1.8 as default [78], accounts for fluctuations when the contact is formed.
The DRID feature vector and the fraction of native contact were obtained using the MDtraj code (http://mdtraj.org/) [79].
A model of the S151T Rbfox*•pre-miR20b* complex was prepared from the representative structure of the most populated cluster of the Rbfox*•pre-miR20b* complex simulations (see above for details). A threonine residue at position 151 was introduced using the Swiss MODEL software [56–59] and two standard independent MD simulations (Table 1, sim. 19–20) were conducted using the same protocol as described above.
The free energy difference associated with the S151T mutation (ΔΔG) was computed according to the thermodynamics cycle equation: ΔΔG = ΔGco−ΔGs = ΔGS151–ΔGS151T. The ΔGco and ΔGs represent the results of the non-equilibrium alchemical calculations[52] of the S151T protein-RNA complex and of the free protein state, respectively. The ΔGS151 and ΔGS151T are the dissociation energy of the Rbfox*•pre-miR20b* and of the S151T Rbfox*•pre-miR20b* complex, respectively.
The free energy calculations were conducted in a following fashion: From the equilibrium production simulations of the Rbfox*•pre-miR20b* complex (Table 1, sim. 17) and of the Rbfox* protein (Table 1, sim. 14), 10 conformations were extracted equidistantly in time, and, for every configuration, a hybrid structure/topology for the S151T mutation was generated using the pmx utilities (http://pmx.mpibpc.mpg.de/) [51, 80]. Subsequently, a 1 ns MD simulation for every configuration was performed to equilibrate the velocities on the introduced dummy atoms.
From each equilibrium simulation, 100 snapshots were extracted equidistantly in time, and finally, a 200 ps (Table 1, sim. 21 (FE)) or 1 ns (Table 1, sim. 22 (FE)) alchemical transition was initiated to morph the system from one physical state to the other. The alchemical transformations were performed in both directions: S151 to S151T and vice versa. A soft-core function with the default parameters (α = 0.3, σ = 0.25, p = 1)[51, 81] was used for the non-bonded interactions during the non-equilibrium transitions. The work values from the non-equilibrium transitions were used to calculate free energy differences based on the Crooks theorem [82] utilizing the maximum likelihood estimator [83]. The protocol described above was applied to all the alchemical simulations.
Table 1 reports the complete list of the simulations performed in this work (overall more than 50 μs of molecular simulation).
Hydrogen bonds were analyzed using the cpptraj [72] module of AMBER 16 (available as AmberTools16 at http://ambermd.org/AmberTools16-get.html) [60]. We used a distance cut-off of 0.35 nm between the relevant heavy atoms and an angle cut-off of 135° for the intervening hydrogen atom. These interactions are characterized by the percentage of the trajectory during which they are observed (i.e. occupancy). Aromatic amino acids and nucleobases were considered to form stacking interactions if the distance between their centers of mass was less than 0.5 nm and the angle between the two planes was less than 30°.
RNAs base pair, base-pair steps and the ion distributions around the RNA helical axes in the simulated systems were analyzed with the Curves+ program [31] and the Canal and Canion utilities (available at https://bisi.ibcp.fr/tools/curves_plus/). Average ion molarities were calculated by setting the groove limit at a radius of 0.11 nm from the RNA helical axis, while the angular limits were determined by the average position of the sugar C1’ atoms.
Deviations relative to the initial RNA structure were calculated using the εRMSD metric [19], a recently suggested RNA-specific structural metrics that is considered more robust than the notoriously insensitive and ambiguous RMSD [84, 85]. Two structures with εRMSD of 0.7 or lower are considered to be significantly similar [19].The εRMSD was calculated using the baRNAba package (available at https://github.com/srnas/barnaba).
The protein’s deviations from the initial structure were analyzed in terms of the RMSD, calculated using the cpptraj [72] module of Amber 16 [60]. The protein backbone conformational plasticity was calculated in terms of PADω angle from the T-PAD analysis (freely available upon request) [18]. The latter provides a quantitative analysis of local plasticity of individual residues in terms of the angular dispersion ω, which is the sum of the Ramachandran angles Φ and ψ. Moreover, it allows distinction between backbone local fluctuations and conformational transitions (from one region of the Ramachandran plot to another) even when they occur with the same amplitude [18]: the tag “F” is assigned to fluctuations, “T” to long transitions (i.e., contributing more than 30% of the simulation time) and “t” to short transitions (i.e., contributing less than 30% of the simulation time). This analysis has been successfully used in the past to evaluate proteins backbone fluctuations from MD simulation trajectories and NMR structures [86].
The conformational entropy has been estimated by calculating the chain’s conformational entropy from the distribution of the backbone (φ, ψ) and side-chains rotameric angles, [χn] following ref. [87]. The calculation has been performed on the trajectories of Rbfox and Rbfox* free and in complex with RNA.
Protein‘ chemical shifts (CS) were predicted using SHIFTX2 v. 1.07 (http://www.shiftx2.ca/) [37, 38]. LARMORD software (https://brooks.chem.lsa.umich.edu/) [36] was used for RNA. In the SHIFTX2 program the sequence information is not used in the prediction, so that the predictions are identical to those of the SHIFTX+ program (http://www.shiftx2.ca/performance.html). We run SHIFTX2 and LARMORD on each frame extracted from the un-restrained simulations every 10 ps of the free pre-miR20b and of the Rbfox•pre-miR20b complex, for which experimental CS are available. The chemical shifts predictions for these 48,000 pre-miR20b and 48,000 Rbfox•pre-miR20b conformers were then linearly averaged to make a final prediction for the protein’ 13Ca, 13C’, 13Cb, 15N and for the RNA’s 13C and 1H CS.
For the set of experimental upper bound distance restraint rNOE, the simulated NOE’s 〈ri,j〉 were calculated according to:
〈ri,j〉=(1Nf∑(ri,j)−6)−16
(3)
where ri,j is the interatomic distance between atoms i and i, and the sum runs over the Nf trajectories frames. The average distance violation was defined as:
1NNOE∑(rNOE−〈ri,j〉)ifrNOE<〈ri,j〉
(4)
where the sum runs over all reported intermolecular NOE-based distance restraints. The conformers with best match with the NOEs upper bounds were then selected to produce an “MD-adapted structure ensemble” using the same protocol as in [14]. In particular, we used the combined simulation trajectories of the Rbfox•pre-miR20b complex and from each we selected 10% of frames with fewest NOE violations. K-means clustering algorithm was used to cluster this group of frames based on the RMSD of the complex. The representative structures of the 20 clusters obtained constitute the “MD-adapted structure ensemble”: sets of atomic coordinates (deposited as PDB files at https://doi.org/10.5281/zenodo.1297931) that capture the flexibility and the conformers suggested by MD simulations while still retaining the highest possible level of agreement with the primary NMR data.
|
10.1371/journal.pntd.0006788 | Leprosy in children under 15 years of age in Brazil: A systematic review of the literature | Leprosy is a chronic infectious disease neglected, caused by Mycobacterium leprae, considered a public health problem because may cause permanent physical disabilities and deformities, leading to severe limitations. This review presents an overview of the results of epidemiological studies on leprosy occurrence in childhood in Brazil, aiming to alert health planners and managers to the actual need to institute special control strategies.
Data collection consisted of an electronic search for publications in eight databases: Literatura Latino-Americana e do Caribe em Ciências da Saúde (LILACS), Scientific Electronic Library Online (SciELO), PuBMed, Biblioteca Virtual em Saúde (BVS), SciVerse Scopus (Scopus), CAPES theses database, CAPES journals database and Web of Science of papers published up to 2016. After apply selection criteria, twenty-two papers of studies conducted in four different regions of Brazil and published between 2001 and 2016 were included in the review. The leprosy detection rate ranged from 10.9 to 78.4 per 100,000 inhabitants. Despite affecting both sexes, leprosy was more common in boys and in 10-14-year-olds. Although the authors reported a high cure proportion (82–90%), between 1.7% and 5.5% of the individuals developed a disability resulting from the disease.
The findings of this review shows that leprosy situation in Brazilian children under 15 years is extremely adverse in that the leprosy detection rate remains high in the majority of studies. The proportion of cases involving disability is also high and reflects the difficulties and the poor effectiveness of actions aimed at controlling the disease. The authors suggest the development of studies in spatial clusters of leprosy, where beyond the routine actions established, are included news strategies of active search and campaigns and actions of educations inside the clusters of this disease. The new agenda needs to involve the precepts of ethical, humane and supportive care, in order to achieve a new level of leprosy control in Brazil.
| Leprosy remains as a severe health problem in Brazil and its transmission in children under 15 years of age occurs mainly through intradomiciliary contacts. The number of leprosy cases in this age group is considered an important indicator for the surveillance of this disease. To understand how the epidemiological studies in Brazil have shown the situation of young people affected by leprosy, we performed a systematic review of the literature, searching for published articles about the situation of leprosy in this age group. We reviewed 22 studies published during 2001 to 2016 and concluded that the negative effects of leprosy still remain high in most of studied places in Brazil. This disease was more common in boys, aged between 10 to 14 years old, with a remarkable proportion of disabilities due to leprosy. These disabilities can limit their routine activities and reflect failure in the public medical care. We hope that our review should contribute to arguments in order to improve the control of this disease in children.
| Leprosy is a chronic infectious disease caused by Mycobacterium leprae. Since the disease may cause permanent physical disabilities and deformities, leading to severe limitations in individual’s ability to perform daily activities, this disease is considered a public health problem worldwide [1, 2]. The incubation period of M. leprae is very long, in some cases up to ten years, and for this reason the majority of cases only become clinically detectable in adulthood. The occurrence of leprosy in children under 15 years of age suggests early exposure and persistent transmission of the agent [3].
Leprosy control has improved markedly around the world over the past thirty years. During this period the leprosy prevalence fell from 21.1 per 10,000 inhabitants in 1983 to 0.24 per 10,000 inhabitants in 2000. This decline occurred due to the generalized use of multidrug therapy (MDT), in addition to nationwide campaigns and an improvement in the quality of health services directed to leprosy treatment in endemic countries [1]. In 2016, the World Health Organization (WHO) launched a new global strategy entitled “The Global Leprosy Strategy 2016–2020: Accelerating towards a leprosy-free world” [4].
The total number leprosy new cases registered by the WHO in 2016 was 214,783 (2.9 per 100,000 inhabitants), 95% of which are concentrated in only 14 countries of high endemicity where over 1,000 new leprosy cases are notified each year [1]. These countries are geographically situated within the tropics, with India in first place and Brazil coming second in terms of the number of cases detected annually.
In 2017, the total number of leprosy new cases in Brazil was 22,940, of which 1,718 were in children under 15 years of age, corresponding to 7.5% and detection rate of 3.72 cases per 100,000 inhabitants [5]. One of the goals of the Global Leprosy Strategy 2016–2020 is to further reduce global and local leprosy burden, aiming to reduce to zero the number of children with disabilities due to this disease. However, in the last five years in Brazil, the percentage of children <15 years old that has grade 2 disability (G2D) ranged from 2.9% (2013) to 4.1% (2017), with an average proportion of 3.7%, reflecting long delay in diagnosis [5].
One of characteristics of the distribution of leprosy is the occurrence in clusters, and in Brazil the leprosy detection rate in the all population and in the under 15 years of age, varies greatly among regions and cities. A region is considered hyperendemic when the leprosy detection rate in under 15 years of age is above 10 per 100,000 inhabitants [6], in many areas, the values of this indicator reach higher levels those considered hyperendemic [5].
In 2016, the detection rate of new cases in this country reached 12.2 per 100,000 inhabitants, considered “high” according to the reference parameters established by the Ministry of Health [5]. In the northern region, this indicator in the general population was 28.7 per 100,000 inhabitants, and of 8.92 per 100,000 inhabitants under 15 years of age. In the Midwest, 30.01 new cases per 100,000 inhabitants were registered for the general population and 6.42 per 100,000 inhabitants under 15 years of age. In the Northeast of the country, the detection rate in the general population was 19.29 per 100,000 inhabitants, while in those under 15 years of age was 5.78 per 100,000 inhabitants [7, 8].
Based on fact that leprosy in children under 15 years of age is hiperendemic in Brazil and that produce important negative effect on life of the affected children and their families, the research question that guided this systematic review was: Does the epidemiological studies on leprosy in childhood, in Brazil, have warning to the policy makers and health managers that the severity of negative effects of this disease is demanding immediate attention? This review presents an overview of the results of epidemiological studies on leprosy occurrence in children under 15 years, in Brazil, aiming to alert health planners and managers to the need to institute special control strategies for this disease, which is one of the most neglected problems of public health.
This study was conducted according to the guidelines established in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [9], with the study protocol registered in PROSPERO with reference code CRD42016033006PROSPERO.
The review was conducted between May 17 and August 10, 2016 using the databases listed in Table 1. Equivalent keywords or subject terms were identified using health sciences descriptors (DeCS), a trilingual structured vocabulary created by the Latin American and Caribbean Center on Health Sciences Information (LILACS/BIREME) [10], in the three languages (Portuguese, English and Spanish) included in this review. A simulated search by language was then conducted to verify which terms would achieve the optimal results in each database. The appropriateness of the language used in the search was assured by preserving the three languages as filters in the databases.
Although the term “children under 15 years of age” was used as a keyword in various studies conducted in Brazil and internationally, no such equivalent was found in the DeCS, with “minors” being the closest term found. Therefore, in addition to the terms child and adolescent, the term “children under 15 years of age” was also used. Likewise, after a preliminary reading of the international papers, it was found that two forms were used in international papers when referring to “children under 15 years of age”: “under 15 years” and “younger than 15 years”. Therefore, the following search terms were used and restricted to the fields “title”, “abstract” and “keywords”:
Brazilian databases: BVS, CAPES theses database and Scielo: Hanseníase AND “Epidemiologia” AND “Criança”; Hanseníase AND “Epidemiologia” AND “Adolescente”; “Hanseníase AND “Epidemiologia” AND “Menores de 15 anos”.
International databases: LILACS, CAPES journals, PubMed, Scopus, Web of Science: "Leprosy" AND "Epidemiology" AND “Child”, "Leprosy" AND "Epidemiology" AND “Adolescent”, "Leprosy" AND "Epidemiology" AND “Under 15 years” OR “Younger than 15 years”, “Lepra” AND “Epidemiología” AND “Niño”, “Lepra” AND “Epidemiología” AND “Adolescente”, “Lepra” AND “Epidemiología” AND “menor de 15 años”.
The protocol defined the following inclusion criteria: complete papers and theses available in the selected databases in English, Portuguese or Spanish. Since there were few scientific papers specifically on leprosy in children under 15 years of age, a decision was made not to restrict the beginning of the search period to any specific year, thereby including all the articles, dissertations and theses found up to the cut-off date of August 2016, in which the type of epidemiological study was described in the Methods section.
The exclusion criteria consisted of papers involving animals; articles published as a short communication or poster; papers in which the methodology used was not described; any other type of documents; and duplicates.
The data extracted from the selected papers were transferred to an Excel 6.0 version spreadsheet using double data entry and evaluated independently by two reviewers (MCAV and KVFA). All disagreements were settled by a third reviewer (MGT).
The relevant data that were collected in the publications selected for the systematic review were: authors, year of publication, journal, study site, study period, study design, population, age group, sex, education, type of housing, leprosy detection rate, epidemiological characterization, operational classification, clinical classification, vaccination with Calmette-Guérin Bacillus (BCG) scar, smear test result, disability degree at diagnosis, disability degree at cure, Hansen's reaction, close contact with leprosy patient, contact examination, 2nd BCG in contacts, detection mode, reason for discharge, complete treatment, relapse and sequelae.
A scale was created to evaluate the quality of the articles selected for this systematic review (S1 Table). This scale was based on and adjusted according to the proposals made by Downs and Black [11] and Boas and Neto [12]. Scores were awarded for the selected items on a 17-point scale and analyzed according to the mean score awarded for the quality of the articles.
A total of 2,075 papers were identified in the databases following a search based on the selected keywords. Subsequently, 1,880 articles were excluded because: they did not meet the study objectives, they dealt with subject matter other than that proposed, duplicates, the papers were not available in their entirety, the age group included in the study was older than that established for this review, or the objective of the study was not included in the keywords, title and/or abstract. Fig 1 consists of a flowchart depicting the selection of the articles in accordance with the PRISMA guidelines.
The 22 studies selected for inclusion [3, 13–33] were published between 2001 and 2016, and the most productive year was 2010 (04 papers). The predominant database was the BVS from which eight papers were retrieved and included in the review. Geographically, the papers identified came from four regions of the country: the Northeast, Southeast, Midwest and North, as described in Table 2.
All studies included in this systematic review were based on scientific research (n = 22). Most of the studies were linked to universities (n = 20) and only two were from the Brazilian Ministry of Health. The studies were financed by the authors' own resources (n = 14), by the National Scientific and Technological Development—CNPq (n = 06) and Ministry of Health (n = 02).
S2 Table 2 describes the papers in greater detail: author, year of publication, type of study, study population, study period, objective, main findings and score according to the methodology validation scale. The studies were ecological, cross-sectional or case series descriptive. Sample sizes ranged from 24 to 2,455 individuals of both sexes. Most of the studies provided descriptions of the participants according to age (0–4 years, 5–9 years and 10–14 years), sex, detection rate, epidemiological classification, operational classification, clinical classification and degree of disability. Of the 14 publications in which leprosy was presented proportionally by sex, males were more affected in 8. With respect to the distribution of cases according to age group, the greatest number of cases occurred in 10-14-year-olds (n = 10). Only 13 papers provided the clinical classification of the disease and in 6 of these articles the most common was tuberculoid leprosy. Operational classification was described in 16 studies, with the paucibacillary form being predominant in 14. The classification of the epidemiological situation of the studies place was described in 17 articles, with 11 reporting hyperendemic levels in children under 15 years of age, with detection rates that ranged from 10.9 to 78.4 per 100,000 inhabitants. Of these 19 papers, 5 reported prevalence in the states of Alagoas, Minas Gerais, Mato Grosso, Pará and Rio de Janeiro, with the level being highest in Mato Grosso at 25.9 per 100,000 inhabitants in 2014.
The proportion of cases diagnosed with grade 2 disability (n = 13) ranged from 1.7% to 5.5%. However, the percentage of individuals in this group who had not undergone evaluation (n = 5) ranged from 2.4% to 19.20%. Only five articles reported the proportion of grade 2 disabilities following cure (ranging from 0.5% to 11.1%) and only three reported on the proportion of the sample that was not evaluated, citing percentages that ranged from 18% to 61.5%.
Household contact, with the family being the main source of transmission, was reported in three publications. According to the reports, 40% of the children had parents with leprosy, 20–36% a grandparent, 18% an uncle or aunt, and 4% had siblings with the disease.
The examination of contacts was reported in four publications, with 24.2% to 90% of contacts having been examined. The principal form of detection reported (n = 8) was through spontaneous demand at healthcare facilities (13.8% to 79.6%), followed by referral from another healthcare service (12.4% to 75.6%). Other forms of detection such as collective examinations (n = 6) and the examination of contacts (n = 7) were also reported, with percentages ranging from 1% to 6.6% and from 8.9% to 20.4%, respectively.
The presence of a BCG vaccination scar and of a second dose of this vaccine for the contacts was described in only two studies, with results showing that the presence of a scar from the first dose of this vaccine was high; however, revaccination of contacts was very low, between 13.8% and 15.5%.
Five studies reported on acid-fast microscopy results, with this test having been performed in 20.6% to 94.6% of cases and with positivity ranging from 4.3% to 15%. In 4.4% to 79.3% of cases, this test was not performed.
Only two papers described the occurrence of a leprosy reaction. In one of these articles, reaction occurred in 33.4% of patients: immediately following diagnosis in 24.4% and following cure in 9%. The other study simply mentioned that 44.1% of the patients did not develop a leprosy reaction.
Complete treatment involving six doses of multidrug therapy (MDT) was administered in 41.7% to 66.7% of cases, with 12 doses of MDT being administered in 11.8% to 45.8% of cases. In another 2.14% to 21.5% of cases, treatment regimens were longer or unknown. Recurrences were recorded in 3.4% of children. Complete cure ranged from 81.9% to 90% of cases. Few studies emphasized the importance of drawing the attention of health planners and managers to the need to develop special actions for childhood.
This systematic review included 22 articles and showed that the detection rate of leprosy in children under 15 years of age, in Brazil, comes decreasing, but it remains very high in the most of sites investigated reaching levels compatible with hyperendemicity, or at least high or moderate endemicity. It was shown strong association between poverty and leprosy and that its spatial distribution is in clusters. The boys were the most affected by this disease; the proportion of multibacillary cases was high (23.4% to 75%) considering that they were children; the proportion of cases with grade 2 disability varied in the different studies analyzed, from moderate (1.7% to 5.5%) to high at the time of bacteriological cure (18% to 61.5%)and; the proportion of reaction leprosy was also high (33.4% to 55.1%). The coverage of intradomiciliary contacts exams did not reach satisfactory levels, as well as was very low the coverage of the second dose BCG.
These findings clearly indicate that the socio environmental conditions for determination of leprosy occurrence in Brazil are maintained, despite efforts, which have been undertaken to eliminate it, in the last three decades, when there was prevalence reduction near 94%, in total population. These efforts has made possible the continuous reduction in the prevalence and leprosy detection rate in adults and in children in this country as a whole [34], but has not been able to avoid the continued transmission and expansion of this disease that persist in the clusters, particularly in the Midwest, North and Northeast regions [35]. Such situation shows the early contamination of young people, consequent to the failure to detect and treat timely adults bacilliferous cases who maintains the chain of transmission in communities [20]. Luna et al. [16] and Santos et al [20] demonstrate that the distribution of new cases of leprosy differs from area to area, and highlight the need to sensitize managers to perform actual diagnosis by region, with actions being prioritized according to the epidemiological profile. In this perspective, the implementation of health education measures would theoretically bring benefits, mainly in the major cities, where leprosy has been found concentrated in bounded clusters.
The fact that children with leprosy were mostly from families living in great social vulnerability related to low socioeconomic conditions [24, 25, 32] is not a new or unknown fact. It is not by chance, that some of the studies included here reveal that these children have more difficulties in progressing in schooling, that will result in the future in perpetuating poverty, similar to their parents. This is not determined by the physical damages that the disease causes, but by the disease stigma and insufficient social protection. Unfortunately, this situation has been maintained until the present day as more recent studies show. For example, it was found a statistically significant association between greater coverage by the conditional cash transfer program Bolsa Família and reduction in the detection rates of leprosy. The authors argue that, in view of its characteristics, this social protection strategy is capable of acting on different social determinants of leprosy such as poverty and economic inequality [36, 37].
As Brazil´s Leprosy Control Programme is integrated in the primary health care while in other countries leprosy interventions are developed in vertical programmes, it is difficult to establish comparisons among endemic countries. Although in Brazil the actions of this Programme are implemented in the whole network of public health services one of difficulties faced is that the majority of new cases reported are known from passive detection by these services, in which the demand is mostly constituted by poor population. There is evidence that when active case search is implemented, unexpectedly higher new case detection rates have been found, even in cities with the highest human development index (HDI), as in São Paulo [38] and Brasília-District Federal [39].
As regards to the higher frequency of boys with leprosy observed in the analyzed studies, further studies need to be conducted to clarify whether boys are indeed at a greater risk of acquiring the disease. If confirmed, a hypothesis to explain this fact could be their greater exposure to social interactions more frequent and intense when compared to girls, increasing the possibility of contact with people with leprosy outside the dwelling [25].
As we know, the basic strategy for the control of leprosy focuses on eliminating the sources of infection using MDT. For this strategy to be effective it has to be adopted timely, i.e. as early as possible to ensure that patients are protected from developing a disability and to reduce the time of transmissibility of the agent. Nevertheless, the proportions of severe disability found in this review are clear proof that diagnosis of the disease is being made late, increasing the risk of nerve damage [26], producing negative repercussion on the daily life of this young. Undoubtedly, physical disabilities negatively affect child’s development, stigmatizing the individual, causing severe psychological repercussions and affecting his/her social life, and it can also reduce a person’s future ability to enter the job market [14, 26, 33]. On the other hand, the high rates of leprosy reaction found highlight the need to increase the availability of specialist services for the prevention of physical disabilities in young people affected by this disease [24]. Not by chance, one of the principal goals of the new Global Leprosy Strategy 2016–2020 [4], is to eliminate grade-2 disabilities in pediatric patients by implementing quality services aimed principally at women and children.
In addition, the predominance of tuberculoid form indicates that immunocompetent individuals also are being affected, highlighting that Mycobacterium leprae is circulating intensely [3, 17, 24, 40]. Another finding that gives further strength to this affirmative is the proportion of multibacillary cases which cannot be considered low for the age group under 15 years of age. On the contrary, these cases are concerning, since means that the young people most susceptible to the disease are becoming infected at a very early age, representing yet more one evidence that the process of transmission is intense and continuous [21]. The low proportion the healthy contacts vaccinated, measured by the scar presence of the BCG second dose of the contacts, found in the present review, shows possible failure of surveillance to avoid multibacillary forms.
The main limitation of this systematic review was the low quantity of studies available in the scientific literature, highlighting how much leprosy is a neglected disease and, in particular, when it refers to childhood. In addition, most of these studies were descriptive, using surveillance secondary data that, in general show some inconsistency in relation to the quality and quantity of information. These limitations did not allow the analysis of individual and collective risk factors among other epidemiological aspects.
In spite of these limitations, the results of this study will be useful, especially for public health, since they will help to alert the decision makers to the need in identifying the care needs of those under 15 years of age affected by the disease due to social stigma and the leprosy physical repercussions in the children.
Although it has not been the object of evaluation of most of the studies, several authors point out that one of the problems of leprosy control is the fragility of surveillance, since the action of health services is predominantly individualized, with a low proportion of search active contacts. The detection of new cases depends on the spontaneous demand of the individual, who mostly seek the health service in advanced stage of the disease, which increases the risk of permanent damage. [3, 5, 19, 23].
The presence of the contacts and the communities in the greatest risk of disease, should be found of the pillars of collective actions, continues being one of the main force to overcome those in the Brazilian leprosy control program. Besides, the treatment of leprosy patients still not meet the target established (98%) by Brazilian Ministry of Health and WHO [25].
In view of this adverse scenario, we suggested the development of studies in spatial clusters of leprosy, where beyond of the control and surveillance actions established by the Brazilian Ministry of Health, are included news strategies of active search adding social contacts, campaigns and actions of educations inside these areas aiming early diagnoses and treatment. If many new cases in children will be discovered in these communities, it is understood that the search must be expanding to others neighboring area to uncover the primary sources of infection to prevent new cases in childhood.
For this, it is necessary a debate and initiatives needs to be carried out to establish an agenda aimed at strengthening the active surveillance of this disease. This agenda needs to involve the precepts of ethical, humane and supportive care in order to achieve a new level of leprosy control in Brazil.
|
10.1371/journal.ppat.1000572 | A Novel Extracytoplasmic Function (ECF) Sigma Factor Regulates Virulence in Pseudomonas aeruginosa | Next to the two-component and quorum sensing systems, cell-surface signaling (CSS) has been recently identified as an important regulatory system in Pseudomonas aeruginosa. CSS systems sense signals from outside the cell and transmit them into the cytoplasm. They generally consist of a TonB-dependent outer membrane receptor, a sigma factor regulator (or anti-sigma factor) in the cytoplasmic membrane, and an extracytoplasmic function (ECF) sigma factor. Upon perception of the extracellular signal by the receptor the ECF sigma factor is activated and promotes the transcription of a specific set of gene(s). Although most P. aeruginosa CSS systems are involved in the regulation of iron uptake, we have identified a novel system involved in the regulation of virulence. This CSS system, which has been designated PUMA3, has a number of unusual characteristics. The most obvious difference is the receptor component which is considerably smaller than that of other CSS outer membrane receptors and lacks a β-barrel domain. Homology modeling of PA0674 shows that this receptor is predicted to be a bilobal protein, with an N-terminal domain that resembles the N-terminal periplasmic signaling domain of CSS receptors, and a C-terminal domain that resembles the periplasmic C-terminal domains of the TolA/TonB proteins. Furthermore, the sigma factor regulator both inhibits the function of the ECF sigma factor and is required for its activity. By microarray analysis we show that PUMA3 regulates the expression of a number of genes encoding potential virulence factors, including a two-partner secretion (TPS) system. Using zebrafish (Danio rerio) embryos as a host we have demonstrated that the P. aeruginosa PUMA3-induced strain is more virulent than the wild-type. PUMA3 represents the first CSS system dedicated to the transcriptional activation of virulence functions in a human pathogen.
| Pseudomonas aeruginosa is a versatile pathogen; these bacteria are able to cause an infection in humans and other mammals, zebrafish, insects, nematodes and even plants. P. aeruginosa evolved an impressive amount of gene regulation systems to be able to express the right virulence genes under the right circumstances. The best studied examples of these are the two-component systems and the autoinducers. In addition, P. aeruginosa is also able to regulate virulence genes using the pyoverdine cell-surface signaling system (CSS). Genome analysis shows that there are multiple putative CSS systems in P. aeruginosa. In this paper we have studied a novel CSS system with a number of remarkable characteristics and show that this system is involved in the regulation of several putative virulence factors. Induction of this system leads to increased virulence in our zebrafish embryo infection model. Our study provides new insights into the regulation of virulence by P. aeruginosa.
| The human opportunistic pathogen Pseudomonas aeruginosa is known for a high proportion of regulatory genes in its genome [1]. This is not only due to the number of two-component regulatory systems, but this bacterium also contain a large number of different cell-surface signaling (CSS) systems [2],[3]. CSS is a regulatory mechanism used by bacteria to sense signals from the extracellular medium and transmit them into the cytoplasm. CSS systems are generally composed of three different components, an alternative σ70 factor of the extracytoplasmic function (ECF) family, a sigma factor regulator located in the cytoplasmic membrane and an outer membrane receptor [2],[4],[5]. Sigma factors are essential subunits of prokaryotic RNA polymerase, they are involved in promoter recognition and transcription initiation. The primary sigma factor (RpoD), which is responsible for the majority of mRNA synthesis in exponentially growing cells, belongs to the σ70 family. This family also includes many alternative sigma factors that are nonessential proteins required only under certain circumstances [6],[7]. The largest and most diverged group within this family is the one including the ECF subfamily of sigma factors. ECF sigma factors are specially abundant in P. aeruginosa [8].
The outer membrane receptor of CSS systems is usually a member of the TonB-dependent receptor family. These receptors are mostly involved in the transport of iron-siderophore complexes across the outer membrane. To accomplish this task these receptors need to be energized by a protein complex in the cytoplasmic membrane. This protein complex is composed of TonB, ExbB and ExbD, of which the TonB protein is the one that actually makes contact with the outer membrane receptor, hence the name TonB-dependent receptors [9],[10]. TonB interacts with a specific region of the TonB-dependent receptors, generally known as the TonB box [11]. Coupling with the cytoplasmic membrane is necessary because the iron-siderophore complex has to be actively transported across the outer membrane, where there is no source of energy available. All TonB-dependent receptors possess the same structural components: a 22 antiparallel stranded β-barrel, an N-terminal globular domain known as the cork or plug domain that occludes the opening of the β-barrel and a TonB box that extends into the periplasm [10]. However, not all TonB-dependent receptors are involved in CSS, only a subfamily known as TonB-dependent transducers [12]. This subfamily can be easily distinguished from other TonB-dependent receptors on the basis of an N-terminal extension of approximately 70–80 amino acids [13]. This extension determines the specificity of the transduction pathway, but has no effect on the binding and transport of the siderophore [14]. This domain is thought to interact with the sigma factor regulator, which is located in the cytoplasmic membrane.
For P. aeruginosa's own siderophore pyoverdine the signal transduction pathway of CSS starts with binding of the inducing signal Fe-pyoverdine to its outer membrane receptor FpvA, which results in the activation of two ECF sigma factors, PvdS and FpvI. Upon activation, PvdS binds the RNA polymerase core enzyme and directs it to the promoter upstream of the genes required for pyoverdine production and also of the genes encoding the virulence factors exotoxin A and PrpL [15]. Activated FpvI bound to the RNA polymerase initiates transcription of fpvA [16].
In addition to FpvI and PvdS, P. aeruginosa contains another twelve iron starvation sigma factors [17] that are probably part of a CSS pathway [2],[3]. Most of these P. aeruginosa iron starvation sigma factors control iron uptake via haem, via citrate or via heterologous siderophores, such as ferrichrome, ferrioxamine B and mycobactin [3], [18]–[20]. There are also two P. aeruginosa iron-starvation sigma factors that seem to regulate the uptake of a metal ion(s) different than iron, probably zinc or manganese [3]. The last P. aeruginosa iron starvation sigma factor is the one encoded by the PA0675 gene (named pigD in the Pseudomonas Genome Project database). This gene is clustered with a gene encoding a putative sigma factor regulator (PA0676 or pigE) and with one encoding a putative receptor (PA0674 or pigC). In silico analysis of this CSS system, which has been designated PUMA3, showed that it has a number of specific and unusual characteristics. The most obvious difference is the receptor component. The PA0674 receptor is considerably smaller (23 KDa) than that of other CSS outer membrane receptors (75–85 KDa). It contains the N-terminal extension typical of TonB-dependent receptors involved in signaling (Figure S1, Supporting Information), but does not have the C-terminal β-barrel domain typical of these receptors. Moreover, PA0674 seems to form a single operon with the ECF-encoding gene PA0675, while the sigma factor regulator gene PA0676 seems to form a different transcriptional unit. This is in contrast to all other CSS systems in which the genes encoding the sigma factor and the sigma factor regulator are forming an operon [3]. Interestingly, the synthesis of the PA0674 receptor is induced upon interaction of P. aeruginosa with human airway epithelial cells [21],[22], which suggests that this CSS system could be active in vivo.
This work was aimed at characterizing this novel P. aeruginosa CSS system. To get more information about its unusual receptor component, a homology model for the PA0674 protein has been constructed. The PUMA3 target genes were identified by microarray analysis of cells overexpressing the PA0675 ECF sigma factor. These analyses show that this CSS system is involved in the regulation of at least 27 genes, including genes encoding secreted proteins and components of secretion systems. Although the role of most of these regulated genes has not been established yet, we have demonstrated, using zebrafish (Danio rerio) embryos as an infection model, that PUMA3 is involved in the regulation of P. aeruginosa virulence. Therefore, we propose to rename the components of this system VreA (PA0674), VreI (PA0675) and VreR (PA0676) (from virulence regulator involving ECF sigma factor).
Bioinformatic analysis predicts that the VreA receptor contains a signal sequence (SS) of 25 amino acids and separate amino-(N-) and carboxy-(C)-terminal domains (NTD and CTD, respectively) (Figure S2, Supporting Information). The predicted mature domain of VreA was submitted to several secondary structure prediction servers, including PSI-PRED [23]. The consensus results indicate that the NTD consists of residues 29–115 and the CTD of residues 133–238, which are separated by a short linker (residues 116–132).
BLASTp results of the full-length VreA sequence against the Protein Databank revealed that residues 60–115 have a strong structural homology to the Secretin and TonB N-terminus (STN) domain superfamily. The highest ranked homologous structure (30% sequence identity) corresponds to the periplasmic signaling domain (residues 1–117) of the P. aeruginosa ferripyoverdine receptor FpvA. Since the structure of this protein is known, FpvA was used as a template for model building using a structure-based sequence alignment. Superimposition of residues 39–120 of the VreA/NTD structure to residues 1–82 of the signaling domain structure of FpvA reveals that the two structures are nearly identical, with a backbone Cα RMSD of 0.09 Å (Figure 1A). The structure of the VreA/NTD displays a β-α-β fold, with two α-helices positioned side-by-side and sandwiched between two-stranded and three-stranded β-sheets. Despite the high degree of homology between the model and template structures, an interesting difference arises with respect to the location of the expected TonB-box of VreA. In the structure of FpvA, the signaling domain is located before the TonB-box [24], while in the VreA/NTD structure the predicted TonB-box (88-DALTR-92) is found in α-helix 3 of the signaling domain (Figure 1A).
Submission of the C-terminal domain (CTD) sequence to the SUPERFAMILY server [25] revealed a strong structural homology to the C-terminal domain of the TolA/TonB protein superfamily, which was not detected by BLASTp or the Protein Model Portal [26]. The SUPERFAMILY server method is optimized to find homologues of protein sequences with low sequence identity. The crystal structure of the C-terminal periplasmic domain of P. aeruginosa TolA protein was identified as the closest structural domain despite a low overall sequence identity of 9.3%. TolA is part of the Tol-Pal (Tol-OprL in Pseudomonas) membrane complex, which is mainly involved in maintaining the integrity of the outer membrane [27],[28]. TolA is structurally and functionally related to the TonB protein, both of which belong to the TolA/TonB protein superfamily. The final homology model of VreA/CTD encompasses residues 124–233 and includes a portion of the short linker region. Structural alignment of the C-terminal domains of VreA and TolA reveals a backbone Cα RMSD of 7.04 Å (Figure 1B). The VreA and TolA C-terminal domains both adopt the same central secondary structure fold, β(2)-α-β, in which the three-stranded β-sheet is packed against two α-helices. The VreA/CTD homology model differs primarily from the TolA/CTD with respect to its shorter α-helix 1 and β-strand 3, which could have functional implications since both are directly involved in the interaction of TolA with other proteins [29].
These predictions would indicate that this putative VreA receptor is not located in the outer membrane, but in the periplasm. To study this in more detail we generated an influenza hemagglutinin (HA) epitope-tagged version of VreA and expressed this chimeric gene at low levels in the wild-type strain and in the PA0676 mutant, which does not produce the putative inner membrane regulator VreR. Although VreA is partially membrane associated, the majority is soluble (Figure S3) and therefore probably located in the periplasm. Furthermore, the presence or absence of the inner membrane regulator VreR did not affect stability and localization of VreA. Remarkably, the apparent molecular weight of VreA was higher than expected (i.e. 34 kD in stead of 23), which means that VreA is posttranslationally modified or has a secondary structure that affects migration in SDS-PAGE.
To identify genes whose transcription might be regulated by the VreI ECF sigma factor, total RNA from P. aeruginosa cells overexpressing the vreI gene from the pMUM3 plasmid was isolated and subjected to cDNA microarray analysis. Overexpression of ECF sigma factors usually results in the expression of the sigma-dependent genes in the absence of the inducing signal [3],[14],[16],[19]. As listed in Table 1, overexpression of vreI upregulates 30 genes (including the vreI gene itself that was overexpressed, and vreA and vreR that were also partially present on the pMUM3 plasmid and therefore overexpressed). Most regulated genes are located immediately downstream to the PUMA3 locus (Figure 2), as is often the case of genes regulated by ECF sigma factors. These genes encode: components of the Hxc type II secretion system (PA0677-PA0687) involved in the secretion of alkaline phosphatase (PA0688) [30], a putative two-partner secretion system (TPS) (PA0690 and PA0692), a putative transposase (PA0691), exbBD homologues (PA0693 and PA0694), three hypothetical proteins (PA0696, PA0697, and PA0698) two of them containing predicted signal peptides, and a putative peptididyl-prolyl cis-trans isomerase (PA0699) (Table 1). The putative secreted protein of the two partner secretion system (PA0690) belongs to a family of high-molecular-weight surface-exposed proteins involved in cell adhesion and pathogen dissemination [31],[32]. In addition, VreI seems to control the expression of a small number of other genes located in different loci of the P. aeruginosa genome. These include genes encoding an ECF sigma factor (PA0149), a hypothetical protein (PA0532), three putative cytoplasmic membrane proteins (PA1652, PA2404 and PA2784), two putative ATPases of ABC-transport systems (PA0716 and PA4192), two putative lipoproteins (PA2349 and PA5405), a homologue to the Fur regulator (PA2384), and a putative transcriptional regulator (PA5403).
To validate the microarray results, the expression of some VreI-regulated genes was analyzed by RT-PCR. Primers within two VreI-regulated genes, PA0691 and PA0692, were designed to determine the mRNA levels in P. aeruginosa cells overexpressing the VreI ECF sigma factor. As shown in Figure 3A, the expression of both genes was induced by VreI, but not the expression of the control gene PA0636. VreI-mediated induction of PA0691 was also confirmed using a transcriptional fusion of the PA0691 promoter region to lacZ. Overexpression of the sigma factor vreI leads to a 25-fold increase in the PA0691 promoter activity when cells are cultured in presence of 1 mM IPTG (Figure 3B). Since vreI is under control of the Ptac promoter, this inducing condition is expected to result in an increased expression of the PA0675-regulated genes.
Previous experiments have shown that the PUMA3 CSS system appears to be induced in vivo, since interaction of P. aeruginosa with human airway epithelial cells induces the expression of many VreI-regulated genes (Tables S1 and S2, Supporting Information) [21],[22]. In order to determine whether VreI-regulated genes are synthesized in vivo, we analyzed the presence of antibodies against VreI-regulated proteins in the serum of P. aeruginosa infected patients. To this end, predicted highly antigenic internal fragments of the PA0690 (TpsA), PA0692 (TpsB) and PA0697 genes were fused to a glutathione S-transferase (GST) gene and overproduced in E. coli (Figure 4A). The fusion proteins were then purified using Glutathione Sepharose 4B columns. Subsequently, these purified chimera proteins were used to detect the presence of antibodies in the serum of P. aeruginosa infected patients. We tested in total the serum of 25 different patients, 7 with positive blood culture for P. aeruginosa and 18 cystic fibrosis (CF) patients. Antibodies against the secreted component of the TPS system, the PA0690/TpsA protein, were present in the serum of 5 of the 7 patients with positive blood culture for P. aeruginosa (71.4%) and in the serum of 12 of the 18 CF patients tested (66.7%) (Figure 4B). However, antibodies against the second component of the TPS system, the outer membrane transporter PA0692/TpsB could not be detected with any of these sera (Figure 4B). The third protein tested, PA0697, which contains a putative signal sequence, was detected with 4 (57.1%) of the sera from patients with positive blood culture for P. aeruginosa and with 10 (55.5%) of the CF patients sera (Figure 4B). The presence of antibodies against these proteins indicates that they are being expressed during infection. Since mRNA levels of these genes are extremely low under non-inducing conditions, this result also suggests that the PUMA3 CSS system is induced in these patients.
In order to determine the role of the sigma factor regulator VreR in the PUMA3 signaling pathway, we analyzed the stability and activity of the VreI ECF sigma factor in a vreR mutant. To analyze the stability of this sigma factor, we first constructed the pMUM3RσHA-tag plasmid in which the vreI gene is C-terminal tagged with the HA epitope. This plasmid and the control plasmid pMUM3 were then transferred to the P. aeruginosa PAO1 wild-type (WT) strain and the vreR mutant (sigma factor regulator mutant). The presence of the VreI-HAtag protein was analyzed by Western blot using an anti-HAtag antibody. As shown in Figure 5A, the VreI-HAtag protein (22 kDa) could be detected in strains bearing the pMUM3RσHA-tag plasmid, whereas it could not be detected in strains bearing the control plasmid (data not shown). The addition of 1 mM IPTG slightly increases VreI-HAtag production, which is under control of the Ptac promoter (Figure 5A, upper panel). Interestingly, the VreI ECF sigma factor seems to be more stable in absence of the sigma factor regulator as the amount of this protein in the vreR mutant is considerably higher that in the wild-type strain (Figure 5A). Analysis of the cytosol and membrane fractions of both strains showed that VreI is associated to the membrane through the VreR sigma factor regulator since the VreI-HAtag protein could not be detected in the membrane fraction in absence of this protein (Figure 5A, lower panel). Although there is more VreI sigma factor in absence of VreR, this sigma factor is not active in this condition (Figure 5B), since overexpression of vreI in the vreR mutant does not increase PA0691 promoter activity, while it does in the wild-type strain (Figure 5B). Overexpression of the whole PUMA3 system (receptor, ECF sigma factor and sigma factor regulator) from the pMMB-PUMA3 plasmid does not increase PA0691 promoter activity (Figure 5B), possibly due to the simultaneous overexpression of the vreR gene encoding the sigma factor regulator component. In conclusion, VreR is an anti-sigma regulator for VreI that is both required for the function of VreI and inhibits its activity under non-inducing conditions.
Although the role of most P. aeruginosa PUMA3-induced genes has not been established yet, the fact that some of them encode secreted proteins and components of secretion systems suggests that the PUMA3 CSS system could be involved in regulation of virulence. For this reason we decided to analyze P. aeruginosa virulence. Therefore, we used a novel infection model for P. aeruginosa using zebrafish (Danio rerio) embryos as a host. The zebrafish model has a number of advantages over other models of infection [33]. One of them is that zebrafish embryos are transparent, which allows the analysis of bacterial infections in situ, in real time and at a high resolution by using fluorescent microorganisms. Recently, zebrafish embryos have been reported to be a suitable model for P. aeruginosa [34],[35].
In order to set up the model, we analyzed first whether P. aeruginosa could infect 28–30 hours-post-fertilization (hpf) embryos. To this end we introduced P. aeruginosa PAO1 wild-type strain into the zebrafish embryo by microinjection in the caudal vein. We observed that P. aeruginosa was able to lethally infect the embryos in a dose dependent manner (Figure 6A). Embryos were resistant to low doses of bacteria (150–200 colony forming units, CFU), but increased mortality was observed with larger inocula (∼400–1300 CFU) (Figure 6A). These experiments also showed that P. aeruginosa kills the embryos within the first two days-post-infection (dpi); embryos that were alive after this time usually were able to clear the P. aeruginosa infection and developed normally.
Then, we analyzed the virulence of the P. aeruginosa PUMA3-induced strain, by overexpression of the vreI ECF sigma factor, in zebrafish embryos. As shown in Figure 7, infection with the P. aeruginosa PUMA3-induced strain resulted in a significant increase of zebrafish embryo mortality. This effect was repeatedly shown in 5 different experiments using groups of 20 embryos. To demonstrate that this effect is specific for PUMA3 induction, we also infected the embryos with the vreR sigma factor regulator mutant bearing the pMUM3 plasmid that overexpresses the VreI sigma factor. We have shown previously that overexpression of vreI in this mutant does not lead to upregulation of the PUMA3-controlled genes (Figure 5B). As expected, induction of PUMA3 in the vreR mutant did not result in an increase in P. aeruginosa virulence (Figure 6B), which confirms the direct involvement of VreI in the increased P. aeruginosa virulence.
We next assessed the role of different PUMA3-regulated genes in VreI-induced virulence. To this end, the pMUM3 plasmid was introduced in transposon insertion mutants of PUMA3-regulated genes encoding potential virulence factors, such as both components of the TPS system tpsA and tpsB (PA0690 and PA0692, respectively), the tonB homologue PA0695 and the putative secreted protein PA0696. Subsequently, these different mutants were injected in zebrafish embryos. Unfortunately, all mutants, except the vreR sigma factor regulator mutant described previously, were as virulent as the wild-type strain (Figure 6C). This means that none of these potential virulent factors is by itself responsible of the VreI-induced lethality in zebrafish embryos. Possibly a combination of these factors is responsible for this phenotype, or some of the other VreI-regulated genes.
Zebrafish embryos infected with red fluorescent P. aeruginosa (PAO1/RFP) bearing the pMMB67EH (empty plasmid) or the pMUM3 (overexpressing vreI) plasmid were microscopically observed to follow the progression of the infection (Figure 8A). In the first hours after infection, fluorescence was undetectable (data not shown), but with the progression of the infection, fluorescent bacteria were clearly detectable in the embryos at 20–24 hpi (Figure 8A). The amount of fluorescence in the embryos correlated with a slower blood circulation and decreased heartbeat when compared with healthy embryos (in which fluorescence was undetectable), and also with severe damages of the tissues, mainly in the tail (Figure 8A). Affected embryos presented increasing red fluorescence and normally died within the first 24–30 hpi, whereas infected but apparently healthy embryos were able to survive and develop normally. There were no obvious differences between the phenotype of moribund embryos infected with the PUMA3-induced strain or with the non-induced strains (data not shown). Both strains produce a similar necrotic cell death that starts in the tail and extends to other tissues. The difference between both strains lies in the fact that a considerably higher percentage of embryos infected with the induced strain died due to the proliferation of the P. aeruginosa infection (Figure 6B and 8A).
The microscopy studies also showed that P. aeruginosa, which was microinjected in the blood stream, was able to extravasate and infect other tissues, mainly what appears to be brain and spinal cord of the embryos (Figure 8A). By whole mount immunohistochemistry of embryos infected with PAO1/RFP using an antibody that specifically recognizes Acetylated tubulin (AcTub) present in neurons and axons of the embryo, we observed co-localization of bacteria with fluorescent brain and spinal cord tissues (Figure 8B). More in depth analysis with confocal microscopy clearly showed many bacteria in the center and around the nerve bundles of the spinal cord (Figure 8C). No colocalization with axons or neuronal cell bodies was observed (Figure 8C, middle panel), which suggests that P. aeruginosa bacteria reside in non-neuronal cells or extracellularly. In addition single bacteria were observed in the muscles (Figure 8C, middle panel).
In this work, we report the identification and characterization of a novel CSS system, designated PUMA3, which has a number of specific and unusual characteristics, and is involved in the regulation of P. aeruginosa virulence. The most obvious difference of PUMA3 with all other CSS systems is the receptor component VreA. The structural modeling of VreA predicts a bilobal protein (Figure S2), with an N-terminal domain (NTD) that resembles the N-terminal periplasmic signaling domain of TonB-dependent transducers, such as FpvA, and a C-terminal domain (CTD) that resembles the periplasmic C-terminal domain of the TolA/TonB protein superfamily. The signaling domain of TonB-dependent transducers is the domain that interacts with the sigma factor regulator [36]. Therefore, it is likely that VreA/NTD interacts with the sigma factor regulator VreR, as illustrated in Figure S4. Although the structures of VreA/NTD and the signaling domain of FpvA are nearly identical (Figure 1A), an interesting difference is the location of the expected TonB-box of VreA. In FpvA, the signaling domain is located N-terminal to its TonB-box. In the apo-FpvA structure, the TonB-box is buried between the signaling domain and the plug/barrel domains, and forms a β-strand that interacts with the signaling domain. Upon ferric-pyoverdine binding this β-strand is displaced and free to interact with TonB in a β-strand lock-exchange mechanism [24]. In the VreA/NTD structure, the predicted TonB-box is found in α-helix 3 of the signaling domain fold (Figure 1A). If the VreA TonB-box interacts with the TonB protein in a similar mixed four-stranded β-sheet fashion as other reported TonB-dependent proteins a significant conformational change would be required.
The VreA/CTD showed, despite the low sequence identity, strong structural homology to the C-terminal domain of the TolA protein (Figure 1B). The Tol-Pal (Tol-OprL) system is organized into two protein complexes: a cytoplasmic membrane complex that consists of the TolQ, TolR, and TolA proteins, and an outer membrane-associated complex composed of TolB and Pal. TolA plays a central role by providing a bridge between the cytoplasmic and outer membranes via its interaction with the Pal lipoprotein [37]. The Tol proteins are parasitized by filamentous (Ff) bacteriophages and group A colicins [38],[39]. The N-terminal domain of the Ff phage g3p protein and the translocation domains of colicins interact directly with TolA during the processes of import through the cell envelope [40]. The TolA protein has functional analogy with the TonB protein. Especially the interaction of the C-terminal domain of TolA and the Ff phage g3p protein is similar with that of the C-terminal domain of TonB and the TonB-box of TonB-dependent receptors [11],[41]. Since the C-terminal domain of VreA is similar to TolA/TonB, it is tempting to speculate that VreA/CTD could interact with other partner proteins in the outer membrane (Figure S4). Based on bioinformatics analysis, it is clear that the predicted domain architecture of VreA is unique and has yet to be reported. Significantly, both domains of VreA are predicted to resemble proteins that form essential interactions with partner proteins required for signal transduction and bacterial virulence.
By microarray analysis of P. aeruginosa cells overexpressing the PUMA3 ECF sigma factor VreI, we have identified the genes regulated by this novel CSS system. It was shown previously that overexpression of the sigma factors results in the specific induction of the sigma-dependent genes in the absence of the inducing signal [3],[14],[16],[19]. The microarray analysis shows that the PUMA3 system controls the expression of 27 genes (Table 1), most of which are located directly downstream of the PUMA3 locus (Figure 2). As observed previously for other ECF sigma factors [3], overexpression of vreI does not result in an unspecific response and does not affect house-keeping genes. This is probably due to the fact that the RNA polymerase has a higher affinity for the house-keeping sigma factor RpoD (σ70) than for alternative sigma factors [42].
The interaction network of the P. aeruginosa VreA receptor with other proteins using the STRING database [43] shows the physical and functional connectivity of this protein not only with the other two components of the PUMA3 system VreI and VreR, but also with most PUMA3-regulated proteins such as both components of the TPS system, ExbBD2, and the PA0696-PA0697-PA0699 proteins (Figure S5A, Supporting Information). PUMA3 homologues are found in the genome of both Pseudomonas fluorescens Pf5 and Pf01, Pseudomonas entomophila, Burkholderia vietnamiensis, Rhodopseudomonas palustris and Janthinobacterium sp (Figure S5B). Interestingly, in all these bacteria, the PUMA3 gene cluster is associated with homologues of the TPS secretion system, of the ExbBD system and of PA0696. P. fluorescens contains an Hxc-like type II secretion system immediately downstream of PUMA3 and also a tpsA-like gene. In B. vietnamiensis, homologues of some of the hxc-like genes are located upstream of PUMA3, interrupted by a tpsA-like gene. The second component of the TPS pathway is located downstream of PUMA3, associated with exbBD homologues. In R. palustris and Janthinobacterium sp., the PUMA3 cluster is repeated four times. Three of these clusters are followed by a tpsA homologue, and the fourth one by a tpsB-like gene associated with exbBD and PA0696 homologues. This gene association further suggests a role for PUMA3 in the regulation of the genes, especially tpsA.
Although most CSS systems described to date control the expression of their cognate TonB-dependent transducers, this does not seem to be the case of the PUMA3 CSS system. Microarray analysis did show an increase in vreA mRNA levels in cells overexpressing the VreI sigma factor (Table 1). However, because part of the vreA gene is also partially present on the vreI overexpressing plasmid this does not mean that vreA is induced upon activation of PUMA3. Direct analysis of vreA expression using a vreA::lacZ transcriptional fusion showed no differences between cells overexpressing or non-overexpressing vreI (data not shown). This difference between PUMA3 and other CSS systems is probably related to the unusual genetic organization of the PUMA3 system. In contrast to all CSS systems in which the sigma and sigma factor regulator genes are arranged in an operon, the sigma factor gene of the PUMA3 system seems to form an operon with the receptor gene (these genes overlap 4 bp), while the sigma factor regulator seems to be part of a different transcriptional unit (Figure 2). In this situation, regulation of the VreA receptor expression by the VreI sigma factor would imply that VreI also induces its own expression, which is unusual for ECF sigma factors.
Another interesting characteristic of PUMA3 is the role of the sigma factor regulator VreR in the signaling pathway. The function of this integral cytoplasmic membrane protein is to couple the signal perceived by the TonB-dependent transducer to the ECF sigma factor in the cytoplasm. The large periplasmic C-terminal part interacts with the receptor in the outer membrane, whereas the short cytoplasmic tail binds the ECF sigma factor [44],[45]. Currently, there is no structural data available for any member of this protein family, and the molecular mechanism by which these proteins work is not completely understood. It is generally considered an anti-sigma factor, based on the fact that overexpression of the ECF sigma factor results in constitutive induction of the CSS system [3],[19]. In accordance with this, overexpression of the sigma factor regulator results in a strongly reduced induction upon the presence of the extracellular signal [16]. However, for the PUMA3 CSS system the sigma factor regulator is in fact essential for VreI sigma factor activity (Figure 5B). This is also the case for the sigma factor regulator FecR of the E. coli ferric-citrate CSS system [4]. Our experiments also show that the PUMA3 ECF sigma factor is more stable in the absence of the sigma factor regulator and relocates to the cytosol (Figure 5A). A similar situation has been described for the sigma factor PvdS and its regulator FpvR; overexpression of fpvR results in increased degradation of the ECF sigma factor PvdS and possibly also FpvI [45],[46] and PvdS relocates partially to the cytosol in absence of the inner membrane regulator [47]. This means that these sigma factor regulators not only retain the ECF sigma factors at the cytoplasmic membrane in an inactive form, but possibly also deliver these to a specific endoprotease. Future experiments are necessary to show which protease is involved and what the exact role of the PUMA3 sigma factor regulator.
The PUMA3 CSS system appears to be induced in vivo, since the serum of the majority of P. aeruginosa-infected patients, including patients with positive blood culture for P. aeruginosa and cystic fibrosis patients, contains antibodies directed against PUMA3-regulated proteins (i.e. TpsA and PA0697) (Figure 4B). This means that these proteins are synthesized in vivo during infection and that the PUMA3 system is probably induced. This further suggests its activation by a host signal, which is consistent with the previous reports showing that interaction of P. aeruginosa with human airway epithelial cells induces the expression of many PUMA3-regulated genes [21],[22] (Tables S1 and S2).
Although the role of most P. aeruginosa PUMA3-induced genes has not been established yet, we have demonstrated that this CSS system is involved in the regulation of P. aeruginosa virulence. Infection of zebrafish embryos with the P. aeruginosa PUMA3-induced strain by overexpression of the vreI sigma factor results in a significant increase of embryo mortality (Figure 6B). This effect is VreI specific since its overexpression in a mutant in the sigma factor regulator, which is necessary for VreI activity (Figure 5B), did not result in increased virulence (Figure 6B). The induced lethality was visible after the first day of infection as the embryos that were alive at this point usually were able to clear the infection. Zebrafish embryos have been recently reported to be a suitable model for P. aeruginosa [34],[35]. Known attenuated P. aeruginosa mutants, such as mutants in type III secretion or in quorum sensing, are less virulent in zebrafish embryos. Moreover, key host determinants, such as phagocytosis, play an important role in zebrafish embryos pathogenesis, and, as in humans, phagocyte depletion increases the susceptibility of the embryos to P. aeruginosa infection [34]. Neutrophils and macrophages rapidly phagocytosed and killed P. aeruginosa, but if the amount of cells injected exceeds the phagocytic capacity of the embryo bacteria survive and grow causing the death of the embryo [34]. The PUMA3 CSS system seems to play a role in the first hours of infection. Induction of PUMA3 may result in P. aeruginosa resistance to phagocytosis leading to a lower survival of the embryos. Alternatively, the PUMA3-induced strain may replicate faster in the embryo, although the growth rate of these strains in vitro showed no difference (data not shown). The identification of the upregulated factor(s) responsible of the PUMA3-induced virulence will be essential to understand the mechanisms by which this CSS system induces virulence.
Using fluorescence microscopy, we have shown that P. aeruginosa is able to extravasate and infect other tissues, mainly the brain and spinal cord of the embryos (Figure 8). This pattern of infection seems to be specific for P. aeruginosa and completely different to the one caused by, for example, Salmonella typhimurium, which replicates either inside macrophages or extracellularly but always within the vascular system, or Mycobacterium marinum, which is located in clustered macrophages [33],[48].
In summary, in this work we have identified and characterized a novel CSS system that triggers expression of virulence factors, probably in response to a host signal. A similar function for a CSS system has only been described in the plant pathogen Ralstonia solanacearum, which uses a CSS system to regulate virulence in response to a non-diffusible plant signal [49],[50]. However, whereas the Ralstonia CSS system has all the characteristics of a normal CSS system, PUMA3 shows a number of new and unusual characteristics, and represents the first CSS system dedicated to the activation of virulence factors in a human pathogen. Our work also shows that CSS systems can be used for a broader purpose than the regulation of iron uptake.
A homology model of the N-terminal and C-terminal domains of VreA (VreA/NTD and VreA/CTD, respectively) was built using the SWISS-MODEL server [51]. The VreA/NTD homology model was constructed using a structure-based sequence alignment (SBSA) and the crystal structure of the P. aeruginosa ferripyoverdine receptor FpvA (PDB ID: 2O5P) as template. Other structural homologues of FpvA, including the periplasmic signaling domains from the E. coli ferric citrate receptor FecA (PDB IDs: 2D1U and 1ZZV) and P. putida pseudobactin 358 receptor PupA (PDB ID: 2A02) were included in the SBSA. For the VreA/CTD domain, the template for model building was the C-terminal domain of TolA from P. aeruginosa (PDB ID: 1LRO). Other structural homologues including the C-terminal domains of TolA (PDB ID: 1S62) and TonB (PDB ID: 1XX3) from E. coli were used to construct an SBSA upon which the C-terminal domain sequence of VreA was initially threaded using the program Deepview (version 4.0), prior to submission to the SWISS-MODEL server.
Sequence analysis of the PAO1 genome was performed at http://www.pseudomonas.com [52]. Signal peptides were predicted using the SignalP 3.0 Server available at http://www.cbs.dtu.dk/services/SignalP/ [53]. The functional associations of PUMA3 were predicted using the STRING 8 database at http://string.embl.de/ [43].
The bacterial strains and plasmids used are listed in Table 2. P. aeruginosa PAO1 wild-type strain and all the P. aeruginosa transposon insertion mutants used were from the comprehensive P. aeruginosa transposon mutant library at the University of Washington Genome Center [54]. The locations of the mutations were confirmed by PCR with primers flanking the insertion sites. The strain ID (unique identifier) is given on the table, and further information on these mutants can be found at http://www.genome.washington.edu/UWGC/Pseudomonas/index.cfm.
Bacterial strains were routinely grown in liquid Luria-Bertani (LB) medium [55] at 37°C on a rotary shaker operated at 200 revolutions per min. When required, antibiotics were used at the following final concentrations (µg mL−1): ampicillin (Ap), 100; chloramphenicol (Cm), 30; kanamycin (Km), 25 for E. coli and 200 for P. aeruginosa; piperacillin (Pip), 25; tetracycline (Tc), 10 for E. coli and 20 for P. aeruginosa.
Standard molecular biology techniques were used for DNA manipulations [55]. PCR amplifications and DNA sequencing were performed as described previously [19]. The sequences of the oligonucleotide primers used in this study are listed in Table S3 (Supporting Information). Transcriptional fusions to lacZ were made by cloning the promoter regions, amplified by PCR as EcoRI-XbaI or BglII-KpnI fragments, into the EcoRI-XbaI or BglII-KpnI sites of pMP220 [56]. The fusion constructs were confirmed by DNA sequencing, and transferred from E. coli DH5α to P. aeruginosa by triparental mating using the helper plasmid pRK600 as described before [57]. The influenza HA tag was cloned in the XbaI-HindIII sites of the pMUM3Rσno-stop plasmid (Table 2), which carries the vreI gene without stop codon, as an adapter that was the result of a primer dimer formed by the primers HAtagF-X and HAtagR-H (Table S3). This introduces the HA tag epitope YPYDVPDYAC* at the C-terminal end of the VreI protein. The resulting plasmid, in which the XbaI site is not restored, was designated pMUM3RσHA-tag. The HA tag was introduced at the end of the vreA gene, replacing the stop codon, by a PCR with primers PA0674F-E and EndPA06742. This introduces the HA tag epitope YPYDVPDYAC* at the C-terminal end of the VreA protein and an additional EcoRI and HindIII cloning site. These restriction enzymes were used to digest the PCR product after gel extraction and clone it in the brood host range vector pMMB67EH. The resulting plasmid was designated pMMB674HA.
P. aeruginosa cells bearing the plasmid pMMB67EH or pMUM3 were grown in quadruplicate in 300 mL Erlenmeyer flask with 30 mL LB and 25 µg/ml piperacillin at 37°C and 200 revolutions per min. In these conditions, the growth rate of the wild-type strain and the vreI overexpressing strain was similar (not shown), therefore cell density-dependent regulatory circuits are not affected. When the optical density at 600 nm reached 0.7–0.8, cultures were induced with 1 mM IPTG. After 45 min of incubation, a total of 50 ml of cells from two independent cultures were harvested by centrifugation at 4°C, and total RNA was isolated as described before [3]. RNA quantity was assessed by UV absorption at 260 nm in a ND-1000 Spectrophotometer (NanoDrop Technologies, USA). RNA quality was monitored by 1.5% (wt/v) agarose gel electrophoresis containing 2,2 M formaldehyde as denaturing agent. The cDNA probes were prepared according to the protocol supplied by the manufacturer (Affymetrix) as described before [3]. Target hybridization, washing, staining and scanning were performed by the Affymetrix Core Facility using a GeneChip® hybridization oven, a Fluidics station and MICROARRAY SUITE software (Affymetrix) at the Leiden Genome Technology Center (LGTC®) (Leiden, The Netherlands) as described previously [3]. Genes were considered differentially regulated if the relative change (n-fold) was >2.5 and the P-value was <0.05. Microarray data sets are available from the NCBI Geo Database under accession number GSE15697.
RT-PCR analyses were performed by using the Titan One-Tube RT-PCR kit (Roche) in accordance with the manufacturer's recommendations. For each reaction, 1 µg of total RNA was used. The annealing temperature in each reaction was determined according to the composition of the primers included. DNA contamination of the RNA samples was ruled out by inactivation of the reverse transcriptase at 94°C for 4 min prior the RT-PCR reaction. The sequences of the primers used for the RT-PCR are listed in Table S4 (Supporting Information).
ß-galactosidase activities in soluble cell extracts were determined using ONPG (Sigma-Aldrich) as described previously [19]. Each assay was run in duplicate at least three times and the data given are the average. The β-galactosidase activity is expressed in Miller Units.
Overnight cultures of E. coli DH5α cells bearing the pRP270 (empty plasmid), or the pGST-0690, pGST-0692 and pGST-0697 plasmids (containing the indicated GST fusions) in LB liquid medium supplemented with ampicillin were subcultured 1∶10 in 500 ml of the same medium, grown until log-phase and incubated 3 h with 0.1 mM IPTG. The cells were harvested by centrifugation, resuspended in 10 ml of 1% (v/v) Triton X-100 in phosphate-buffered saline (PBS), and ultrasonically disrupted. The total bacterial lysate was centrifuged (12.000×g, 15 min, 4°C) and the supernatants loaded on Glutathione Sepharose 4B column (Pharmacia) equilibrated with 3 column volumes of PBS. GST-fused proteins were eluted with 10 mM glutathione in 50 mM Tris-HCl pH 8.0 in a fraction of 750 µl (3× times with 250 µl). Protein concentration was determined by the BCA protein assay (Pierce) using BSA as a standard.
Purified proteins were separated on a 12.5% (w/v) acrylamide SDS-PAGE gel and electrotransferred onto nitrocellulose and immunodetected with the serum of 25 different P. aeruginosa infected patients. The second antibody, horseradish peroxidase-conjugated goat anti-human, was visualized using 4-chloronaphtol/3,3-diaminobenzidine staining [55].
P. aeruginosa cells were grown in LB until late log phase and cultures were then centrifuged (10 min at 15,000×g). The cell pellets were solubilized in Laemmli sample buffer [58] and heated for 5 min at 95°C (total protein fraction). To separate cytosol and membrane fractions, cell pellets were first ultrasonically disrupted and centrifuged 5 min at 2,000×g to remove unbroken cells. The resulting supernatant was then centrifuged during 45–60 min at 12.000×g, 4°C. The pellet from this centrifugation step (membrane fraction) was solubilized in Laemmli buffer, and the proteins from the supernatant (cytosol fraction) were precipitated with 10% (w/v) trichloroacetic acid. Proteins from cell lysates, membrane and cytosol fractions were separated by SDS-PAGE containing 15% acrylamide. Proteins were electrotransferred onto nitrocellulose and immunodetected with a monoclonal antibody directed against the HA epitope. The second antibody, horseradish peroxidase-conjugated goat anti-mouse, was visualized using ECL detection (Pierce). Quantification was performed on a Fluor-S MultiImager (Bio-Rad) using Bio-Rad multianalyst software, version 1.0.2.
Zebrafish embryos were collected from a laboratory-breeding colony kept at 24°C on a 12∶12 h light/dark rhythm as previously described [48]. Embryos were staged at 28 hours post-fertilization (hpf) dechorionated and anaesthetised in 0.02% buffered 3-aminobenzoic acid methyl ester (MS222, Sigma). Overnight cultures of P. aeruginosa cells bearing the pMMB67EH empty plasmid or the pMUM3 plasmid overexpressing vreI in LB liquid medium supplemented with piperacillin were subcultured 1∶50 in the same medium, grown until log-phase and incubated 1 h with 1 mM IPTG to induce vreI expression from the Ptac promoter. Then, 2 ml of the log-phase bacteria was pelleted by centrifugation, washed twice with PBS and diluted in phenol red containing PBS (Sigma) at the desired bacterial density. Embryos were individually infected by microinjection (2 nl) of P. aeruginosa in the caudal vein near the blood island and the urogenital opening as previously described [48]. To determine the number of CFU microinjected in each set of embryos, bacteria were also microinjected in PBS and plated on LB-agar.
The mCherry variant of DsRed [59] was obtained as derivative of the pRSET-B plasmid (Invitrogen). In order to optimize for high gene expression in bacteria a new Shine/Dalgarno (SD) sequence was introduced upstream of the original ATG initiation codon. This was achieved by cloning an adapter that was the result of a primer dimer formed by the primers WBcherF (5′-GATCCAAGCTTGAGGAGGA-3′) and WBcherR (5′-GATCTCCTCCTCAAGCTTG-3′) in the BamHI site of the pRSET-B mCherry plasmid. This cloning introduced a HindIII site (underlined) and a SD sequence (GGAGGA) (shown in bold) in front of the mCherry gene. In order to centre the new SD on the -10 position from the ATG start codon, a 7 bp fragment between the SD and the ATG codon was then deleted. This was achieved by an inverse PCR reaction using DNA from this new plasmid as template and the primers Cherry-TomatoF (5′-CATGGTCAGCAAGGGCCAGG-3′) and WBcherR. The obtained PCR product was then re-ligated and transformed in E. coli DH5α cells. The resulting mCherry gene was then cloned as a HindIII-EcoRI fragment in the pBBR-PoprF plasmid, a derivative of the broad-host range pBBR1MCS-5 plasmid [60] (Table 2). This plasmid contains the promoter region of the P. aeruginosa oprF gene to maximize mCherry expression. This final construct, designated pBPF-mCherry, was introduced in P. aeruginosa by triparental mating [57].
For whole-mount immunohistochemistry, embryos were fixed in 4% (w/v) paraformaldehyde overnight at 4°C, rinsed in PBS and incubated in blocking solution for 1 hour (PBS+0.1% (v/v) Triton X-100+10% (v/v) normal goat serum). The primary antibody, a monoclonal anti-acetylated tubulin (Sigma, clone 6-11B-1) was diluted 1∶250 in PBS with 0.1% (v/v) Triton X-100 and 1% (v/v) normal goat serum, and incubated overnight at room temperature with slow agitation. After extensive washing with 0.1% (v/v) Triton X-100 in PBS, embryos were incubated overnight at room temperature in PBS with 0.1% (v/v) Triton X-100, 1% (v/v) normal goat serum and 1∶250 diluted secondary antibody, a goat anti-mouse conjugated to Alexa 480 (Invitrogen). After extensive washing in PBS with 0.1% (v/v) Triton X-100, embryos were transferred to Vectashield mounting medium (Vector Laboratories) and examined by microscopy.
Fixed or live zebrafish embryos, anaesthetised in 0.02% buffered MS222, were examined with a Leica MZ16FA stereomicroscope equipped with a DFC420C digital camera. Photographs were taken with the Leica Application Suite software (version 2.8.1 © Leica Microsystems). In addition, whole-mount immuno labeled zebrafish embryos were examined with a Leica DMIRE2 confocal microscope using the Leica confocal software (version 2.6.1 © Leica Microsystems).
|
10.1371/journal.pcbi.1006061 | Material and shape perception based on two types of intensity gradient information | Visual estimation of the material and shape of an object from a single image includes a hard ill-posed computational problem. However, in our daily life we feel we can estimate both reasonably well. The neural computation underlying this ability remains poorly understood. Here we propose that the human visual system uses different aspects of object images to separately estimate the contributions of the material and shape. Specifically, material perception relies mainly on the intensity gradient magnitude information, while shape perception relies mainly on the intensity gradient order information. A clue to this hypothesis was provided by the observation that luminance-histogram manipulation, which changes luminance gradient magnitudes but not the luminance-order map, effectively alters the material appearance but not the shape of an object. In agreement with this observation, we found that the simulated physical material changes do not significantly affect the intensity order information. A series of psychophysical experiments further indicate that human surface shape perception is robust against intensity manipulations provided they do not disturb the intensity order information. In addition, we show that the two types of gradient information can be utilized for the discrimination of albedo changes from highlights. These findings suggest that the visual system relies on these diagnostic image features to estimate physical properties in a distal world.
| Objects in our visual world contain a variety of material information. Although such information enables us to experience rich material impressions, it can be a distraction for the estimation of other physical properties such as shapes, albedos, and illuminations. The coupled estimation of these properties we humans perform in daily situations is one of the fundamental mysteries in visual neuroscience. Here, we show that material and shape perception depend on two different types of intensity gradient information. Specifically, our image analyses and psychophysical experiments show that a human’s material perception relies mainly on the intensity gradient magnitude information, while shape perception relies mainly on the intensity gradient order information. In addition, we show that the intensity order information can be utilized for discriminating albedo changes on an object surface from other physical properties including specular highlights.
| The physical parameters that affect a retinal image are extremely complex. In addition, the same retinal image can be produced from an infinite number of combinations of materials, shapes, and illuminations in the distal world. Therefore, it is a hard ill-posed problem to estimate what exists in the distal world from a single retinal image. This appears to be a chicken-and-egg problem as material estimation requires knowledge about shape (and illumination), while shape estimation requires knowledge about material (and illumination). Nevertheless, (we believe) we can estimate the physical parameters that produce a retinal image. For instance, from a single photograph wherein a metal teapot is placed on a table, we can simultaneously judge the material and shape of the object. This paper concerns the visual processing underlying such simultaneous estimation.
We found a clue for solving this problem in the image-based material editing methods developed in the computer graphics community [1–3]. By changing image parameters, not physical ones in the distal world, these methods can alter the material appearance of an object without significantly affecting its apparent shape or illumination. Among them, a simple yet effective method is to modulate the luminance histogram of an image. For instance, when the histogram of the original image in Fig 1(A) is matched with that of the reference image in Fig 1(B), the material appearance of the original image becomes very similar to that of the reference image (Fig 1(C)). Another example is the use of monotonic nonlinear tone-remapping for print or screen display devices to transform the intensity histogram of an input image and modify its qualitative appearance [4,5].
Successful manipulation of material appearance by histogram transformation suggests that luminance histograms contain critical information for material perception [6–12]. Specifically, Motoyoshi et al. [8] found that a surface tends to look glossier when the luminance histogram of the surface’s image is positively skewed. Although histogram skewness can be affected by various image parameters, it can be a very good predictor of apparent gloss when the images are nearly the same in other respects, as is the case for histogram-transformed images. Motoyoshi et al. [8] also showed that adaptation to textures with skewed statistics alters the perceived glossiness of surfaces subsequently viewed. These findings led them to conclude that human observers may use histogram skewness, or some image features correlated with it, in making judgments about glossiness.
However, when the spatial structure of an image is inconsistent with a natural glossy surface (e.g., a pixel- or phase-scrambled image), the image does not look glossy regardless of histogram manipulation [8]. Kim, Marlow and Anderson [13] further investigated spatial conditions of gloss perception and found that when specular highlights of an object image are inconsistent in position and/or orientation with the diffuse shading component, they look more like white blobs produced by surface reflectance changes (see also, [14–17]). Marlow, Todorovic and Anderson [18] have demonstrated that three-dimensional shape perception of a surface affects gloss perception of the surface.
These findings suggest that the visual system has to simultaneously solve at least three mutually dependent problems: it has to estimate surface material, surface shape (surface orientation), and reflectance changes. As mentioned above, we believe that material editing by histogram manipulation suggests a clue to this complex computation. The histogram-matching method, as shown in Fig 1, successfully changes the material appearance of a surface image, while it seems to have a negligible effect on surface shape. This suggests that the image properties changed by the histogram transformation affect material processing, while those unchanged by the histogram transformation affect shape processing. In what follows, we will show which components in the image are changed and unchanged by histogram manipulation and then consider the effects of each component on the perception of material, shape and reflectance change. The analysis will lead us to a computational strategy the visual system may follow to simultaneously and nearly independently estimate material, shape and reflectance change from a single image of an object.
To anticipate the conclusion, we here propose that the human visual system may use orthogonal features about image intensity gradient to estimate material and shape: the intensity gradient magnitude for material perception, and the intensity gradient order for shape perception. We also suggest that the intensity order structure provides the critical information for discrimination of highlights from albedo changes [13–17]. The intensity gradient magnitude is related to the intensity histogram statistics, which some have suggested are related to material perception [8], while the intensity gradient order is related to the isophote and orientation flow that have been suggested to be important for robust shape estimations [19–27]. Combining thoughtful insights originating from past theories with new image analyses and psychophysical experiments, we attempt to comprehensively understand how the human visual system simultaneously estimates many interdependent object properties from a single picture.
To explore image constraints for discriminating material changes from other property changes, we focused on a histogram-transformation method that has been widely used to edit the material appearance of a surface image [1,3,6]. In this method, to adjust the luminance histogram of an original image to that of a reference image, each histogram of the original and reference images is converted into a cumulative histogram (Fig 1(D) and 1(E)). Then, the bin values of the original histogram are transformed into those of the reference histogram so that each cumulative value of the original histogram is matched to that of the reference one. Consequently, histogram matching does not change the intensity order of the image. When the pixel intensity of the output image is plotted as a function of that of the original image (Fig 1F), the tone-remapping function monotonically increases, or at least does not decrease. Similar features are observed in general tone mapping techniques [5]. These observations suggest that retaining the intensity order of the original image may be the key feature for editing material while keeping other physical properties constant.
When we consider the image generation processing of an object image, there are good reasons to believe that retaining the intensity order information is critical for material editing. A (monochromatic) surface image can be decomposed into albedo, shading, and specular images (Fig 2A). The albedo image of a surface indicates how much illumination is diffusely reflected at each surface point. It is irrelevant to the surface normal and thus independent of the shading and specular images. The shading image of a surface is the interaction map of the surface normal and the illumination. With diffuse Lambertian shading, the shading intensity is a function of the incident angle of light. The specular component is the direct mirror-like reflection of the incident light. The specular intensity of a surface is a function of the incident and viewing angles of light.
Since both specular and shading intensities depend on the incident angle of light, the specular image is dependent on the shading image. Under a collimated illumination, the surface normal directions of highlight regions are nearly uniform, and the intensity of the matte component behind the highlight regions is also nearly uniform (Fig 2B). If the highlight regions are uniformly painted with (or replaced by) the hidden matte intensity, the image becomes something akin to a matte surface image. Although the hidden matte intensity is not known, the specular highlights tend to be produced near the highest intensity of the diffuse shading, but the position could shift slightly depending upon the difference between the incident angle and viewing angle [24]. Because of these constraints, adding specular gloss on a matte image, unlike adding an albedo change, has relatively little effect on the intensity order of the image.
If the luminance order structure is the same, so is the isophote structure (an isophote is a contour of equal intensity in an image). The direction of the luminance gradient is orthogonal to the isophote. Hence, if the intensity order information of an image is kept constant, the direction of the intensity gradient is too. Fig 2C shows how adding gloss affects the intensity gradient structures. To make an intensity gradient map, we computed the horizontal and vertical derivatives of the intensity distribution, and then converted them to the polar coordinate. In Fig 2C, the magnitude and direction of the intensity gradient vector are indicated by the hue and saturation of a color map, respectively. The intensity gradient map shows that highlight regions have larger gradient magnitudes, which implies that adding specular highlights drastically changes the gradient magnitudes. However, when the gradient magnitudes are normalized and only the directional information of intensity gradients is preserved, the map of the gloss image is similar to that of the matte image. The results suggest that adding gloss to a matte image has a negligible effect on the direction map of the intensity gradient.
To test the generality of the observations, we analyzed material images rendered using the MERL BRDF (Bidirectional Reflectance Distribution Function) database [28], which is a set of measured BRDFs of 100 materials, including rubber, plastic, metal, and fabric. There were four illumination conditions: three single point light sources (slant = 0, 20, and 40 degrees) and one HDR (High Dynamic Range) environment map (Fig 3). Fig 4A shows the results of the analysis. Each cell of the panels indicates the correlation coefficient of the magnitude or the direction map of the intensity gradient between the images rendered with the BRDFs of the row and the column. When the point light source lit the objects from the viewing direction, the direction of the intensity gradient consistently showed quite a high correlation (Fig 4(B), upper), whereas the magnitude of the intensity gradient showed correlations that are relatively low and highly variable depending on the comparison pair (Fig 4(B), bottom). These findings can be confirmed from the probability density distribution of the correlation coefficients in Fig 4(B).
When the incident angle of the light deviates from the viewing angle, the position of specular highlights tends to shift from the position of the highest intensity of the diffuse shading [24]. However, the displacements of the lighting direction do not significantly change the pattern of results (Fig 5(A) and 5(B)). The analysis of the surface images rendered under point light sources suggests that material changes (with no changes in shape and illumination) have little effect on the direction of the intensity gradient.
Under the HDR illumination environment, the correlation in the direction of the intensity gradient is reduced for some material combinations (Fig 6(A)). This is because the direction of the intensity gradient is disturbed by the spatially complex illumination that produces spatially non-uniform mirror reflections, especially when the BRDF has low specular roughness. Since we computed the intensity gradient on a small scale (the kernel size was 5 x 5 pix for an image size of 256 x 256 pix), the fine structures of a mirror reflection of the environment affect the direction of the intensity gradient. However, a simple tone operation can reduce the effect of a mirror reflection of the environment. While tone remappings normally change the intensity histogram without changing the intensity order, strong compressive tone remappings in which the output intensity levels off beyond a certain input intensity can remove the intensity gradients in the high intensity range. Since the mirror reflection generally has higher intensities than those of the shading pattern, a strong compressive tone mapping can eliminate the variation caused by the spatially non-uniform mirror reflection. In one analysis (Fig 6(B)), when the magnitude became smaller than a very small threshold value, we excluded the gradient values from computation of gradient directions. The analysis showed that the correlation in the direction of the intensity gradient is markedly improved.
In addition, it should be noted that the strong correlations in the direction of the intensity gradient across different materials can be obtained only under similar illumination conditions. When we compute the correlation across different illuminations, e.g., between the lighting conditions of 0° and 40°, the direction information of the intensity gradient is markedly different (Fig 7).
The present analysis suggests that the material change of a surface strongly modulates the magnitude of the intensity gradient but does not unduly disrupt the intensity order or the direction of the intensity gradient. This explains why the histogram-matching method, which affects the magnitude of the intensity gradient while preserving intensity order, effectively changes the material appearance.
At the same time, our analysis suggests that the intensity order of an object image contains rich information about the surface shape and reflectance pattern. In the context of computer vision, intensity order information is widely utilized as a feature descriptor [31–34]. For instance, Dalal and Triggs [31] utilized the local histograms of image gradient orientation (called histograms of oriented gradient, or HOG). They showed that the descriptor is robust against environmental changes. In addition, shape-from-shading studies suggest that the intensity order information or the directional information of the intensity gradient is useful for shape estimation [19–27].
Fig 8 shows a hypothetical processing scheme that the human visual system may use for simultaneous estimation of a variety of surface properties. The critical idea is that an input surface image is analyzed in two ways. One focuses on the information about the order of intensity. It could be in the form of isophote, gradient direction map, or orientation map. Shape processing mainly relies on this intensity order information. The other image analysis focuses on the information about the magnitude of the intensity gradient. Material processing mainly relies on this gradient magnitude information. To be precise, the important information for material estimation is likely to be the intensity gradient relative to the surface orientation change [18], but we assume this is computed in subsequent stages. The estimation of the remaining properties, i.e., surface albedo and illumination, relies both on the intensity order and gradient information, along with the absolute intensity level. According to this hypothesis, one can tell whether a bright spot is produced by an albedo change or by a specular highlight by checking how it affects the intensity order information.
While previous studies have suggested that luminance histogram manipulation is an effective way to change material appearance [8], as well as pointing out the importance of orientation field or isophote map in shape perception [19–27], to our knowledge, one potential implication of these findings has not been recognized. That is, material perception and shape perception may be based on separate, independent, orthogonal features of the object image, and this is why the visual system can simultaneously estimate material and shape. Although visual estimation of the material and shape appears to include a hard chicken-and-egg problem (material estimation requires shape information, while shape estimation requires material information), the brain may be able to solve it by computing the two attributes, at least initially, based on the independently measurable image features.
Although our hypothesis includes an explanation as to how the visual system robustly estimates the shape for some materials, it does not cover every kind of material. This is because our basic intuition came from a critical observation that luminance histogram matching affects apparent material, but not shape. While luminance histogram matching realized by monotonic luminance re-mapping can produce a wide range of matte and glossy objects, it cannot easily make mirrored objects with a perfectly specular reflectance. Hence, we do not have a strong theoretical basis to assume that our theory is applicable to mirrored objects. Textured objects and line drawings are also outside our scope. Compared to the “orientation field” theory proposed by [23–25], we consider a more specific problem of monocular shape perception (see Discussion for details).
In the five psychophysical experiments reported below, we empirically tested our hypothesis. The first three experiments measured the apparent shape (surface orientation) of object images to see whether it is affected by the intensity order information but not by the intensity gradient magnitude information. Experiment 1 changed the intensity distribution by means of histogram matching that preserved the intensity order information. Experiment 2 changed the intensity distribution by means of non-linear intensity remapping that disrupted the intensity order information under some conditions. Experiment 3 disrupted the intensity order information more naturally by using velvet-like surface reflectance. The last two experiments examined apparent surface gloss and reflectance uniformity to ascertain whether they are affected by the intensity gradient magnitude information but not by the intensity order information. By using objects with veridical and inconsistent highlights, we considered how the two types of intensity information are used to discriminate material features from reflectance changes. Experiment 4 changed the intensity distribution by means of histogram matching, while Experiment 5 changed it by means of compressive remapping.
In the previous section, we showed that perceived shape is sensitive to the intensity order information but not to the intensity gradient magnitude information. In this section, we will consider the perception of materials and surface reflectance properties.
Although we have seen effective modulations of perceived material by changing the intensity gradient magnitude information with no change in the intensity order information, it is hard for the visual system to estimate material only from the intensity gradient magnitude information. This is not only because of the effects of surface shape on material perception [18] but also because albedo/reflectance changes on the surface of the object affect the intensity gradient magnitude information. For example, a white patch with a steep intensity gradient on the object surface could be either a specular highlight or white paint. To distinguish between them, the visual system can use the intensity order information, since the addition of a reflectance change does, while that of a highlight does not, make the intensity order map significantly different from that of the shading pattern of an object with diffuse uniform reflectance.
When a highlight is located and/or oriented in a manner inconsistent with the shading pattern, it is perceived as an albedo change (e.g., white blob) and does not make the pattern look glossy [14–17]. While past studies proposed congruence in brightness and orientation as conditions for highlight consistency, we additionally suggest that if a white patch is a specular highlight, it does not disrupt the luminance order of the shading pattern. This means that when reducing the bright patch intensity by histogram matching to a less skewed intensity distribution or by applying a compressive tone remapping, one can smoothly erase the highlight and obtain a diffuse surface image. This should not happen if a white patch is an albedo change, since an albedo change disrupts the luminance order map. It should remain visible regardless of how the intensity gradient magnitude information is altered by the manipulation of the intensity distribution. If this hypothesis is correct, our predictions are as follows: For consistent highlights, apparent glossiness is reduced for negative skew or by compressive tone mapping, and the uniformity rating is always low. For inconsistent highlights, apparent glossiness is always low, and the uniformity rating is always high. These predictions were tested in the following two psychophysical experiments.
The ultimate goal of our study is to comprehensively understand how human vison estimates the material property together with other object properties such as shapes, albedos, and illuminations. Our opening question was: Why does intensity histogram manipulation affect human material perception to a much greater extent than it does the perception of other physical properties. The image analyses of a variety of materials revealed that typical material changes have little effect on the intensity order information (which defines the isophote map and the direction of the intensity gradient), while they strongly affect the residual information, i.e., the magnitude of the intensity gradient. This led us to a hypothesis that the human visual system may mainly use the intensity gradient magnitude information for material perception, while it uses the intensity order information for shape perception and albedo change detection (Fig 8). The first three experiments confirmed that shape perception was affected little by the intensity gradient magnitude information but was affected strongly by the intensity order information. The last two experiments confirmed that perceptual discrimination of material-related intensity changes (veridical highlights) from albedo-related intensity changes (inconsistent highlights) is dependent on the intensity order information.
Numerous studies have utilized histogram-transformation methods as used in Experiment 1 to modulate the pattern of intensity histogram [1,6,8, 40–42]. As shown in our image analysis, the transformation does not disturb the intensity order information of a surface image, but it does distort the magnitude information of the intensity gradient of the image. In addition to histogram matching, compressive nonlinear tone mapping is widely used for appearance control in printing or screen display devices [4,5]. The mapping also usually retains the intensity order information of an input image. These techniques are consistent with the present finding that the modulation of the gradient magnitude information can be a diagnostic for the material appearance of the surface.
If the intensity order of the image histogram of a surface is kept constant, then so is the isophote structure or the direction of the intensity gradient of the surface image. The effect of the structure of isophotes on surface shape estimation has been traditionally recognized in the context of shape-from-shading [19–27]. For instance, Koenderink & van Doorn [19–20] showed the structural relationships between the pattern of isophotes across a diffuse surface and the geometric structure of the surface. Specifically, they focused on the “Gauss map”, which is a spherical image where a surface in Euclidean space is mapped to the unit sphere. Since in simple stimulus situations (e.g., Lambertian materials under a collimated illumination) the radiance of a point in a surface only depends on the surface normal, each radiance of the surface image with an identical normal can be mapped to the same point on the sphere image. Koenderink and van Doorn showed that when a specific region of a surface image, such as a convex, concave, or saddle-shaped region, is extracted based on the local extrema of the image, its spherical image corresponds to the surface geometry in a one-to-one fashion. Then the isophotes of the Gauss map can have invariant structures related to the surface geometry, irrespective of illumination directions.
Similarly, Breton & Zucker [21] showed that under a diffuse surface illuminated by a point light source, the orientation of the intensity gradient field of the surface only depends on the geometric properties of the surface irrespective of the irradiance and the diffuse reflectance. They computed the “shading flow field” based on the orientation information and showed that the flow field can be used for shape estimation and edge classification. For instance, an attached shadow cast on a corrugated surface produces discontinuity in the continuous shading field of the surface and thus the discontinuity can be a cue for edge classification. More recently, Zucker and his colleagues introduced the idea of constructing a set of local surfaces based on the shading flow field for diffuse surfaces under any point light source [26].
Although the elegant analyses by Koenderink and van Doorn (1980) and Zucker et al. on potential shape information in intensity gradient maps assume Lambertian objects, the present findings indicate that their theories are also helpful for understanding shape perception for non-Lambertian materials.
In this regard, the contribution of the present study is to show that the effect of the intensity order information is considerably robust against material changes. Our image analysis showed that the changes in natural BRDFs (100 types) did not strongly affect the intensity order information of object images. In addition, we showed in Experiment 3 that when a specific material change distorted the intensity order information of an object image, the perceived shape was changed with the distortion. This finding is consistent with previous studies showing that shape constancy across specific materials could not be obtained [38, 39, 43, 44]. The findings suggest that human shape processing strongly relies on the intensity order information and that distortion of the information tends to cause the perceived shape’s modulation even when actual material changes produce the distortion.
Our psychophysical experiments show that keeping the intensity order constant makes the perceived shape constant (Experiment 1 and 2). In addition, when we disturb the intensity order information, the perceived shape changed with the distortion (Experiments 2 and 3). However, we emphasize that the same shape can have different intensity order maps and that different intensity order maps do not always produce different perceived shapes, due to, say, the effect of illumination differences. Our study mainly investigated material and shape perception under the conditions where an object is placed in a specific illumination environment, but as shown in the Image Analysis section (Fig 7), illumination changes can produce large distortions of the intensity gradient information. Nevertheless, identical objects under different illuminations can be perceived as similar in shape even when their intensity order information is markedly different, as shown in Experiment 2b (Fig 14). The previous studies also reported that the perception of shape and material is quite robust across different illumination contexts [45]. Hence, to recover the perceived shape from the intensity order information, the visual system has to discount the influence of the illumination field. While how it does this remains an open question, one possibility is that the occluding contour of a surface image may normalize the mid-level representation of an object obtained from the intensity order information [18]. Another possibility is that the visual system may extract some illuminant-invariant higher order differential structures from the intensity order information (cf., [26]).
A luminance-order map is far from sufficient to recover the geometrical ground truth of an object even when material information is given. Shape estimation solely from luminance-order information must introduce many ambiguities. It is obvious that the shape-from-intensity-order-map problem suffers from bas-relief ambiguity [46], since our theory concerns perception of matte (diffuse) and gloss (diffuse+specular) objects seen without light source information. In addition, in many cases, luminance-order maps must be equated between completely different shapes by adjusting material (BRDF, BSSRDF, BTF) and/or illumination parameters. Although we have not theoretically analyzed this ambiguity, this would seem to be a hard analysis, since it should consider not only geometrical optics, but also natural statistics of reflectance and illumination parameters. Furthermore, in order to understand human vision, the important issue is not only ambiguity in estimation of the ground-truth 3D structure, but also ambiguity in estimation of the perceived shape. The perceptual representation of shape is degenerated in the sense that it does not contain full detailed information about the ground truth structure, though what is perceptually represented about shape remains controversial [47]. According to our experiments, provided we preserved the luminance order map, the observers reported similar shapes. Despite enormous physical ambiguity, we found little evidence of perceptual ambiguity like that observed in the Necker cube. We think this provides an important hint about perceptual shape representations in the human brain.
Consider next the relationships of luminance-order information to orientation information that Fleming and his colleagues proposed were influential in shape estimation [23–25]. Like Zucker and his colleagues, they constructed the orientation field of a surface image. The dominant orientation of the field was determined according to the relative powers of oriented-linear filters’ outputs. They showed that the distortion of the orientation field of a surface image corresponds to the distortion of the perceived shape of the surface. In computer graphics also, the orientation field has been utilized for apparent shape editing [27]. Specifically, Vergne et al. [27] used structure tensors of a surface image to construct the orientation field and showed that the modulation of the field drastically changed the apparent shape. In addition, they showed in their statistical analysis that the orientation information of identical shapes with different materials (four types) or illuminations (four types) can be similar to each other, as in our image analysis.
Fleming and his colleagues have constructed a general framework of shape perception from image orientation information. Their investigation started from perfectly specular (mirrored) objects, and then generalized their theory to shape perception from diffuse shading, texture or contours. In contrast, our theory was based on a critical observation that luminance histogram matching affects apparent material. Since luminance histogram matching realized by monotonic luminance re-mapping can control the material appearance of an object in the range between pure matte (diffuse) and gloss (diffuse+specular), but cannot easily make perfectly specular appearances (we need non-monotonic luminance re-mapping [41]), we do not have a strong theoretical basis to assume that our theory is applicable to mirrored objects. Textured objects and line drawings are also outside our scope. Despite having a scope narrower than that of Fleming et al., our theory has more specific predictions about material and shape perception of objects within the scope of our analyses.
A critical question is what kind of directional information, i.e., a vector map modulo 180 or 360 degrees, or both, the visual system relies on. The orientation field of Fleming and his colleagues is based on a vector map modulo 180 degrees, while the present study used a vector map modulo 360 degrees to explain the perceived shape. In Experiments 2 and 3, we found that the modulation of the tone-mapping curves of a surface image changed the perceived shape of the surface, even though it only distorted the vector map modulo 360 degrees while keeping constant the vector map modulo 180 degrees. The finding suggests that at least for the class of materials we considered, shape perception is different when the orientation map is similar, but the luminance order is different, as predicted by our theory.
Fleming et al. proposed a ground theory for a wide range of monocular shape perception including cases where shape perception is similar even when luminance order is not preserved [48–50]. We also recognize that non-linear tone re-mappings that do not preserve luminance order information are able to change glossy objects into mirrored or translucent objects of similar shapes (e.g., [41]). We speculate the perceived shape distortion may depend on the spatial flow structure where the flow distortion emerges. For instance, in Experiment 3 (Fig 17) a perceived shape distortion for the asperity (a = 0.2) condition was obtained in gauge position 6 where the shading flow produces a cusp. The finding suggests that the effectiveness of the vector map modulo 360 degrees may depend on the diagnostic flow structure.
Although we show the importance of the signed intensity gradient in shape estimation, we agree that unsigned orientation measurements are also useful in shape processing. For instance, the computation based on 180 degrees would be beneficial in the estimation of specular-only images because the processing modulo 180 degrees is tolerant to the first-order modulation due to mirror reflections. Thus, the processing based on the vector map modulo either 180 or 360 degrees has benefits in some situations. This suggests a possibility that two types of processing are adopted by the visual system, and thus further studies are necessary to elucidate how human shape processing codes directional information.
While Fleming et al. [23–25] measured unsigned orientations at multiple scales, the luminance gradient computation normally has a single value at each location. Simultaneous gradient measurements at multiple spatial scales might be more beneficial, but we have not investigated this possibility yet.
In sum, although the relationship of our theory with the theory of Fleming et al. is still to be clarified, we can at least say that a luminance-order map contains richer information than that of an orientation map, and that human shape processing does not refrain from using the extra information when it is available and useful.
As for the effect of gradient magnitude, although we showed that the perceived shape is little affected by modulating the magnitude of intensity gradients, we understand that in some cases the perceived shape, especially perceived volume, can be affected by position/scale specific modulation. When intensity magnitudes are small, the surface tends to appear to have less curvature (i.e., they appear flatter) than when the magnitudes are large, even if the intensity ordering is constant. In particular, Giesel & Zaidi [51] showed that enhancing the amplitude of specific spatial frequency components increases perceived volume, although the modulation does not change the perceived tilt. It has been shown that the perceived volume (slant) of an object image is unstable, compared with its perceived tilt [52], and therefore it might be affected by several factors depending on the context in which the object is placed.
This study suggests that the computation based on two types of intensity gradient information may facilitate a comprehensive understanding of material and shape processing. In addition, we showed that this computation may also be used for the perception of reflectance changes. That is, the present study revealed that the specular-shading consistency could be judged in the shading processing as a problem of discrimination of smooth shadings from reflectance changes. It is noteworthy that in Experiments 4 and 5 the perception of albedo-uniformity for an object image with inconsistent highlights was not changed by histogram modulations (Figs 20 and 23). This finding suggests that processing based on the intensity order information may be sufficient for discriminating an object image with veridical highlights from inconsistent ones.
Although the algorithms of intrinsic image decomposition in the field of computer vision can discriminate smooth shadings from reflectance changes [53–59], it is not easy for many of them to discriminate veridical specular highlights from reflectance changes such as white blobs. For instance, when one of the cutting-edge algorithms [58] is applied to object images with veridical and inconsistent highlights, even for a uniform albedo image with veridical highlights, it incorrectly detects the highlights as regions with different albedos. However, when the same algorithm is applied to a slope-normalized image (Fig 24B, right), it correctly predicts the image with veridical highlights to have a uniform albedo, while the image with inconsistent highlights to have non-uniform albedos. This observation suggests that, if the visual system has a pre-processing stage to extract a luminance-order (slope-independent) image, it can easily discriminate smooth shadings from reflectance changes, and correctly solve the highlight consistency problem.
We do not intend to dispute a previous hypothesis that the position and orientation congruence of specular highlights relative to diffuse surface shading could be critical for discrimination [13]. In terms of our hypothesis, position and orientation incongruences imply non-smooth luminance-order maps, and thus are likely to arise from albedo changes.
The cortical processing of intensity order information, as well as that of intensity magnitude information, remains unclear. One plausible hypothesis is that the brain decodes the direction and magnitude of the local intensity gradient from the outputs of orientation-selective filters. These filters should be located at early stages where local phase information is preserved, not at later stages where local orientation energy is represented. However, since little attention has been paid to intensity order information, we can only speculate on its cortical mechanism at present. Our brain may adopt a completely different strategy to process luminance-order information. We hope the present psychophysical findings will motivate future neurological investigations into the mechanisms of cortical processing of the intensity order and magnitude information. Specifically, it would be interesting to see which cortical areas are more sensitive to intensity order than to intensity magnitude, and vice versa. Some neurological studies have found gloss-selective neurons in the ventral stream of monkeys and common marmosets [60–62]. For instance, Nishio et al. [61] found neurons in the inferior temporal (IT) cortex of the monkey that selectively and parametrically respond to physical gloss enhancements. These neurons are likely to be more sensitive to intensity gradient magnitude information. On the other hand, object-selective neurons in the other parts of the IT cortex may be more sensitive to intensity order information.
While investigating the effects of histogram transformation methods on material perception, we showed that material processing depends on detailed gradient information rather than intensity order information, such as the direction of the intensity gradient. These findings also revealed the image constraints produced by other physical properties such as albedo and shape. The present study suggested that specular-shading consistency could be judged from intensity order information, with which a specular consistency problem becomes a general shading-reflectance separation problem. In addition, our study suggests that human perception of shape from shading is sensitive to the intensity order information of an object image but not sensitive to the detailed intensity gradient information.
All the psychophysical experiments were approved by the Ethical Committees at NTT Communication Science Laboratories and were conducted in accordance with the Declaration of Helsinki.
We applied a variety of nonlinear remappings to three object images (Fig 11). The remapping function was defined as follows.
f4(Lo)=s1(Lo−m)+s2sin(ω(Lo−m)+ϕ)+m,
4)
where m is the mean intensity, s1 is the slope of the remapping function, s2 is the amplitude of a sinusoidal modulation, ω is the angular velocity of the modulation, andφis its phase. Specifically, we made fifteen remapping curves by adding a sinusoidal modulation with one of five different amplitudes (s2 = 0, 0.015, 0.065, 0.115, or 0.165) to a linear tone remapping function with one of three different slopes (s1 = 0.5, 1, or 2). The ω and φ in the experiment were the constant values of 2.857π and π, respectively. When the slope was the steepest (2), the remapping curve always monotonically increased. This implies that the intensity order of the original image was not disrupted even by the largest modulation. When the slope was midway between steep and gentle (1), the remapping curve was non-monotonic, and the intensity order of the original image was disrupted when a1 was 0.115 or 0.165. When the slope was gentle (0.5), the intensity order was disrupted when a1 was 0.065, 0.115 or 0.165. It should be noted that our manipulation did not change the orientation (modulo 180 degrees) map of the image. The geometric models we used in the experiment were two bumpy spheres (the displacement in the normal direction of the surface of each sphere was given by a coarse or fine Gaussian band-pass noise; Fig 11, bottom left and bottom center), and a cylinder (Fig 11, bottom right). Each model was lit by a point light source from the camera direction.
Eight observers were asked to estimate the perceived shape of the objects by setting a gauge probe with the matching apparent surface slant/tilt. The position of the nine gauge probes is shown in Fig 11. Each of the 45 stimuli (3 objects x 15 tone-mapping types including the original image) was tested three times for each observer. In addition, to confirm the performance stability of the gauge task, the gauge matching for the original images of the three objects was again conducted three times in a different session. These data were used for the baseline and showed as no shape changes in Fig 12. The other methods were the same as in Experiment 1.
In the experiment, we applied a variety of non-linear remappings, as in Experiment 2a, on several object images rendered under a point light source place in the same direction as the viewing one or in the upper-right direction for the object where the illumination slant was 45° and the illumination tilt was 30°. Ten observers were asked to estimate the perceived shape of the objects by setting a gauge probe with the matching apparent surface slant/tilt. The position of the six gauge probes is shown in Fig 13. Each of the 10 stimuli (2 illumination direction x 5 tone-mapping types including the original image) was tested ten times for each observer. The other methods were the same as those used in Experiment 2a.
The geometric model was a bumpy object with low spatial frequency bumps (Object 4) (Fig 16). The model was lit by a point light source from the camera direction. The reflection model used in the experiment was the Lambertian or asperity material [38, 39]. The output intensities on the models were determined as follows:
Ll=ρdπll(I⋅N),
5)
La=aπ[a+(I⋅N)(J⋅N)]la(I⋅N),
6)
where Ll and La are the intensity of the Lambertian and asperity materials, respectively. I, and J are the incident and reflected angles, respectively. N is the surface normal, l is the intensity of the light source, and a is the asperity parameter known as the edge brightening factor. When the incident and reflected angles are the same, which is true under the lighting condition we used, the intensity of the asperity material, La, can be described as a function of the intensity of the Lambertian material as follows:
La=alaLlaρdll+π2Ll2ρdll
7)
This could be regarded as an intensity remapping function from Lambertian to asperity materials, the shape of which is dependent on parameter a (Fig 15). As the Ll increases, the remapping function first rises and then falls after a transition point. The transition point (peak) shifts towards the lower range with a decrease in a. When a is moderately small (0.2), the intensity order is preserved in the lower range of Lambertian pixel intensity, while reversed in the higher range. When a is even smaller (0.02), the intensity order is reversed in most of the intensity range. We rendered asperity objects using these two values of a (Fig 16). In the experiment, the parameters ll, and ρd were π and 0.6, respectively. The parameter la was π under the asperity(a = 0.2) condition, and 4π under the asperity(a = 0.02) condition. In addition to a Lambertian object and two asperity objects, we used an object the intensity of which was completely reversed from that of the Lambertian object (Fig 16).
Ten observers participated in Experiment 3. The observers were asked to estimate the perceived shape of the objects by setting a gauge probe. The other methods were the same as those used in Experiment 2.
The same ten observers in Experiment 3 participated in one additional experiment. In the experiment, the geometric model was lit by a point light source in the upper-right direction for the object where the illumination slant was 45° and the illumination tilt was 30° (Fig 18). The same four reflection models and nine gauge probes as those in Experiment 3 were used. The observers were asked to estimate the perceived shape of the objects by setting a gauge probe.
|
10.1371/journal.pgen.0030164 | Gene Expression Programs of Human Smooth Muscle Cells: Tissue-Specific Differentiation and Prognostic Significance in Breast Cancers | Smooth muscle is present in a wide variety of anatomical locations, such as blood vessels, various visceral organs, and hair follicles. Contraction of smooth muscle is central to functions as diverse as peristalsis, urination, respiration, and the maintenance of vascular tone. Despite the varied physiological roles of smooth muscle cells (SMCs), we possess only a limited knowledge of the heterogeneity underlying their functional and anatomic specializations. As a step toward understanding the intrinsic differences between SMCs from different anatomical locations, we used DNA microarrays to profile global gene expression patterns in 36 SMC samples from various tissues after propagation under defined conditions in cell culture. Significant variations were found between the cells isolated from blood vessels, bronchi, and visceral organs. Furthermore, pervasive differences were noted within the visceral organ subgroups that appear to reflect the distinct molecular pathways essential for organogenesis as well as those involved in organ-specific contractile and physiological properties. Finally, we sought to understand how this diversity may contribute to SMC-involving pathology. We found that a gene expression signature of the responses of vascular SMCs to serum exposure is associated with a significantly poorer prognosis in human cancers, potentially linking vascular injury response to tumor progression.
| It has been estimated that the human body contains approximately 200–400 distinct cell types. These estimates are largely based on the morphological characteristics of cells and have yielded, among many others, the category of smooth muscle cells, which have a distinct appearance and are present in a wide variety of tissues. By using DNA microarrays to interrogate the gene expression of anatomically varying smooth muscle cells, we were able to accurately tease apart many of the distinct cell subtypes that are classically categorized as smooth muscle cells. Remarkably, genes expressed by these newly identified, distinct subtypes corroborate many of their known biological properties and give clues about their susceptibility to specific disease states, retained developmental programs, and potential drugable targets. Additionally, from a smooth muscle cell model of vascular injury, we were able to extract a gene expression signature that provides prognostic information for human breast cancers. Of particular interest for modeling tumor progression was the finding that this gene expression signature was associated with tumor hypoxia. This study adds much to our ever-growing depth of understanding of cellular diversity and the contributions of this diversity to normal physiology and disease.
| Smooth muscle (SM) is a morphologically distinct tissue that mediates the contraction of hollow organs in the circulatory, respiratory, gastrointestinal, and urogenital systems. Beyond the viscera, SM is also present in a variety of anatomical locations, such as the hair follicles, irises, and lacrimal ducts. Smooth muscle cells (SMCs) are the main cell type in SM tissue and have a distinct “smooth” appearance because their sarcomeres, the units of contraction and force generation, are arranged with no specific banding pattern. SMCs share many lineage-specific markers, such as smooth muscle α-actin (SM-α-actin), SM myosin heavy chain (SM-MHC), SM22α, calponin, and caldesmon [1].
The differentiation of SMCs is marked by the expression of SMC-specific lineage markers and the acquisition of the contractile function. The discovery of common cis-acting elements in the promoter regions of these lineage markers has provided some insights into how SMC differentiation is initiated and maintained (reviewed in [2,3]). The promoters of many SMC-specific genes contain CArG [CC(A/T)6GG] boxes or CArG box–like sequences; expression of these SMC-specific genes is triggered by cooperative binding of the ubiquitous serum responsive factor (SRF) and the SMC-specific coactivator myocardin to CArG sequences [2–4]. Other transcription factors, such as MEF2B, P311 (also known as C5orf13), MRF2 (also known as ARID5B), and GATA4 also possess the ability to trigger SMC differentiation programs when ectopically expressed [2,5–9], but their relative contributions to the process of SMC differentiation, in comparison to SRF/myocardin, are not known. Also unclear are the temporal and spatial patterns in which these transcription factors act to initiate or maintain SMC differentiation programs.
Although all SMCs share many morphological and molecular features, they carry out distinct functions in different organs and tissues and are, therefore, likely to vary significantly in their contractile and mechanical properties, hormonal control, physiological regulation, and pathological alterations. For example, arterial SMCs must maintain proper vascular tone to ensure adequate tissue perfusion in response to rapid fluctuations in blood volume, pressure, and tissue oxygen demands, as well as hormonal and nervous inputs [10]. Gastrointestinal tract SMCs, in contrast, participate in the periodic peristalsis that facilitates food passage and, therefore, operate at an autonomous and slower pace [10]. These functional differences imply that significant variations may exist in the epigenetic programs of different SMCs. Indeed, several studies have found heterogeneity among SMCs in the visceral organs [11,12], within the blood vessel wall, and in atherosclerotic lesions of the vascular wall [13–15]. But our knowledge about the nature, extent, and molecular details of these differences is limited.
Unlike skeletal and cardiac muscle cells, which are terminally differentiated, mature SMCs retain their ability to undergo large-scale, reversible phenotypic modulations in response to various genetic and environmental influences [16,17]. For example, the transition of vascular SMCs from the physiological quiescent (contractile) to the pathological activated (synthetic) state is associated with vascular injuries, which cause migration of SMCs into the intima, where they proliferate and produce matrix proteins [16,17]. This phenotypic plasticity plays an important role in the pathogenesis of human diseases, including atherosclerosis, hypertension, asthma, and human cancers.
SMCs are commonly found in cancers as components of blood vessel walls. Blood vessels in tumors are often abnormal—they can be greatly enlarged, tortuous, and “leaky”—and the component SMCs often have abnormal morphology and sometimes fail to express the appropriate SMC differentiation marker genes [18]. These structural abnormalities contribute to the spatial and temporal heterogeneity in tumor blood flow, resulting in elevation of tumor interstitial pressure, hypoxia, and acidosis [19,20]. These hostile tumor microenvironments play a major role in tumor progression and treatment failure, but it has been challenging to quantify and dissect these factors [19].
In this study, we investigated molecular details of SMC heterogeneity by systemically examining the global expression profiles of purified, cultured SMCs isolated from various organs and anatomical structures. We found pervasive differences among SMCs from different organs and tissues. An investigation of the gene expression program induced in vascular SMCs by serum exposure, an ex vivo model of vascular injury, identified a signature that proved to be significantly associated with prognosis in various human cancers. This observation suggests a possible link between blood vessel injury and tumor progression and treatment response in cancer patients.
SMCs isolated from bronchus, uterus, cervix, different blood vessels, urinary bladder, ureters, urethra, and pulmonary artery were obtained from Cambrex. The cells were thawed and propagated in SmGM-2 media (Cambrex), according to the manufacturer's instructions, and cultured under standard conditions (5% FCS with fibroblast growth factors, steroids, and epidermal growth factors) or low-growth conditions (1% FCS with no growth factors). Once the cells reached 60%–70% confluency, mRNA was harvested using the FastTrack mRNA Isolation Kit (Invitrogen). The cells were harvested between the third and fifth passages, after approximately 10–15 generations in culture. We confirmed that each cultured cell population consisted of SMCs, free of epithelial, endothelial, or Schwann cells, by immunofluorescent staining, using antibodies against cytokeratins (C-11, Sigma), desmin (Ab-1, NeoMarkers), glial fibrillary acid protein (ab-7, NeoMarkers), vimentin (V9, Sigma), and CD31 (Ab-2, Neomarkers or Pharmingen).
Human cDNA microarray production (from Stanford Functional Genomic Facility) and hybridization were performed as previously described [21]. mRNA was purified using the FastTrack mRNA Isolation Kit (Invitrogen), according to the manufacturer's instructions. Human common RNA reference (Strategene) was used in all experiments as the standard reference. Two micrograms of all RNA samples were fluorescently labeled with amine-reactive dyes after reverse transcription. SMC samples were labeled with Cy5, and common reference samples were labeled with Cy3. The Cy5- and Cy3-labeled samples were mixed together and heated at 95 °C for 3 min before hybridizing with printed cDNA microarrays for 12–16 h in a 65 °C water bath in sealed cassettes. Following hybridization, microarrays were washed and dried prior to high-resolution scan on a GenePix 4000B Array Scanner (Axon). Each element was located and analyzed using the GenePix Pro 5.0 software package (Axon). These data were submitted to the Stanford Microarray Database (SMD) for further analysis. Data were normalized globally per array, such that the average LogRatio was 0 after normalization. Hierarchical clustering with weighted average linkage clustering was performed after indicated data filtering based on spot quality and variations in signal intensity as described [22].
We established 39 primary SMC cultures from 18 different anatomic locations under identical culture conditions (see Materials and Methods). The 36 primary SMCs included cells purified from five different arteries (aorta, coronary artery, pulmonary artery, iliac artery, and umbilical artery) and four different veins (hepatic, renal, saphenous, and popliteal veins) as well as SMCs isolated from the bronchus, gastrointestinal tract (colon), female genital tract (uterus and cervix), and urinary tract (urinary bladder, ureter, and urethra) (detailed information on all SMC samples is available on the supplemental Web site). All SMCs displayed a uniform spindle shape in culture and were positive for SMC-α-actin. All SMC samples were also stained for endothelial cell and epithelial cell markers (CD31 and pan-keratins, respectively) to rule out contamination from other cell lineages. Three SMC samples (one from bronchus, one from colon, and one from saphenous vein) were excluded from further analyses due to significant contamination (>10%) from other cell types. To investigate how the gene expression patterns in diverse SMCs respond to a common physiological stimulus, 24 of the 36 remaining SMC samples were also cultured in low-growth media (basal media [SMBM] supplemented with only 1% serum and no additional growth factors).
The global expression patterns of all 60 samples (36 SMCs grown in standard-growth and 24 SMCs grown in low-growth conditions) were analyzed using cDNA microarrays, containing approximately 42,000 elements, representing 27,291 unique Unigene clusters (Build number 173, released on 28 July 2004) to generate a total of 2.6 million gene expression measurements; these measurements cover a significant portion of known genes. The expression data were submitted to the SMD and globally normalized so that the average LogRatio was 0 after normalization [23]. We further analyzed the 16,352 gene elements that were expressed in at least 80% of the SMC samples. These elements were identified by looking for elements with Cy5 signal more than 2.5-fold above local background. Unsupervised hierarchical clustering of the gene expression patterns of all 60 samples produced consistent groupings of most SMCs according to their sites of origin (Figure 1A), suggesting that SMCs from different anatomic locations have distinct expression patterns that persist with serial passage in vitro. The 60 SMC samples were clustered into two large, distinct branches: a “vascular” branch contained all the cells isolated from blood vessels and airways, and a “visceral” branch contained all the cells isolated from visceral organs (colon, urinary tract, uterus, and cervix) (Figure 1A). Within the visceral branch, there were three distinct subbranches; one containing all the SMCs from the urinary tract (including ureter, urinary bladder, and urethra), one with female reproductive tract SMCs (uterus and cervix), and one subbranch of colon SMCs (Figure 1A). This grouping pattern is not related to the age of the donor since there was no statistically significant difference in the donor ages between the “vascular” and “visceral” SMC groups (Table S2). To rule out the contribution of gender in sample clustering, we performed an unsupervised analysis on the 24 SMC samples from female donors (under high-growth conditions) and were able to generate a similar grouping pattern (detailed in the Web supplement). Interestingly, culturing SMCs in low-growth conditions had only a relatively modest effect on the clustering pattern (Figure 1A). The clustering of all SMC samples into either the vascular or visceral SMC groups was driven mainly by two large clusters of genes, to which we will refer as the vascular and visceral gene clusters (Figure 1B).
The HOX gene family encodes a family of evolutionarily conserved transcription factors known to be involved in determining positional identity and tissue specialization in animals [24]. Previously, we found that fibroblasts could be clustered by anatomic location based on their patterns of expression of a small number of HOX genes [25]. To assess the possible role of HOX genes in SMC topographic differentiation, we identified 63 genes encoding homeodomain transcription factors that had well-measured expression in these experiments. Hierarchical clustering of the cultured SMCs based solely on their patterns of expression of these 63 homeodomain genes recapitulated grouping of the SMCs according to their site of origin (Figure 1C, the dendrogram labeled as “all” is identical to the dendrogram from Figure 1B); the vascular and visceral groupings were also retained. This homeodomain gene set also successfully separated the visceral SMCs into the three previously defined subbranches: urinary tract, female reproductive tract, and colon, suggesting that this family of transcription factors may play an important role in specifying the distinct developmental programs of the SMCs.
To determine the differences in molecular features between vascular and visceral SMCs, we used a supervised method, Significance Analysis of Microarrays (SAM) [26], to identify 3,276 unique genes (represented by 4,870 array elements), whose expression varied consistently between the 31 vascular SMCs and the 29 visceral SMCs, with a false discovery rate (FDR) of less than 0.001%. All SMC samples were then arranged by hierarchical clustering, based on expression of these 3,276 unique genes (Figure 2A), to yield a clustering pattern almost identical to the unsupervised sample groupings (Figure 1A).
Among genes preferentially expressed in vascular SMCs were many that encode proteins in the transforming growth factor-β (TGF-β) pathway, which affects differentiation, proliferation, migration, and the induction of extracellular matrix (ECM) production, as well as genes responsible for ECM biosynthesis and modification [27–30]. The high expression of these genes in the vascular SMCs contributes to their ability to maintain the tensile strength of blood vessels through the synthesis and deposition of connective tissue proteins [31]. Vascular SMCs also expressed many genes involved in inflammatory responses, suggesting an intrinsic ability of vascular SMCs to communicate with inflammatory cells to initiate and modulate the chronic inflammatory and fibroproliferative processes underlying atherosclerosis and other vascular diseases [32].
Vascular SMCs also expressed high levels of transcripts encoding proteins known to be involved in reciprocal signaling with adjacent endothelial cells and in the eicosanoid/prostaglandin signaling pathways that regulate vascular tone (Figure 2B, gene names shown in purple). A more extensive discussion of these genes appears in the supplemental text in the supplemental Web site.
To further search for systematic differences in molecular pathways between the two broad divisions of SMCs (vascular and visceral), we employed a gene-set enrichment analysis [33] that evaluated differential expression of predefined sets of functionally related genes within our dataset. A running statistic (the Kolmogorov-Smirnof or KS statistic) determines how highly the coordinate expression of each gene set ranks. This statistical tool provides a systematic approach to objectively identify gene sets with functional themes that can be correlated with biological phenotypes. We tested 410 gene sets (compiled from previously published expression studies of common cellular pathways and pathological states and curated by Biocarta [34] and KEGG [35]) for their enrichment in vascular or visceral SMCs. Gene sets that achieved enrichment greater than expected by chance alone were identified by permuting the vascular and visceral SMC sample labels 1,000 times. Of the ten pathway-specific gene sets with the highest normalized enrichment scores (the normalized enrichment scores for all 410 pathways are detailed in the web supplement), eight were enriched in vascular SMCs and two were enriched in visceral SMCs (Figure 2C). The gene sets showing enrichment in vascular SMCs were associated with inflammation, tumor necrosis factor (TNF), TGF-β, interleukin 1 receptor (IL1R), and chemokine receptor pathways. Interestingly, genes of the HIF pathway [such as VEGFA, endothelin-1 (EDN1), lactate dehydrogenase A (LDHA), and HIF1A] were also enriched among genes differentially expressed in vascular SMCs. To test whether an expression signature of the hypoxia response was indeed overrepresented in vascular SMCs, we used a previously obtained hypoxia-response gene signature [36] to analyze the SMC expression dataset. When all 60 SMC samples were arranged by hierarchical clustering based on the expression levels of all 71 genes in the common hypoxia-response gene list [36], most vascular SMCs clustered separately from visceral SMCs as a result of their relatively high expression of HIF1A and the common hypoxia-response genes (Figure 2D). This concordant expression of HIF1A and the other hypoxia-response genes was also observed in renal proximal tubule epithelial cells and ovarian cancers, and high HIF1A transcript levels were associated with a vigorous hypoxia response [36]. Interestingly, hypoxia of the blood vessel wall has been linked to the development of atherosclerosis [37,38], so the differential expression of this hypoxia-response gene set and its influence on the transcriptional program under low oxygen tension may have implications for the pathogenesis of atherosclerosis.
All SMCs isolated from visceral organs share features of the gene expression programs that distinguish them from vascular SMCs (Figure 2B, gene names shown in black), notably including several transcription factors involved in the establishment and maintenance of SMC differentiation (MRF2 (also known as ARID5B) [9], PBX1, HoxA10, and HoxA11 [6,39,40]). Disruption of the PBX1 gene in mice leads to a general hypotrophy of many visceral organs, yet no vascular abnormalities are noted [40], in accordance with its visceral SMC-specific expression. Visceral SMCs also preferentially expressed microphthalmia-associated transcription factor (MITF). In addition to its expression in melanocytes and mast cells [41], MITF has been shown to be present in uterus and other tissues [42]. MITF can inhibit the activities of transforming growth factors (TGF-β) by binding to Smad3, a key signaling component of the TGF-β pathways [43]. High levels of MITF may play a role in modulating the activity of the TGF-β pathway in visceral SMCs (Figure 2B), thereby promoting a functional partitioning of visceral from vascular SMCs.
Visceral SMCs also preferentially expressed Histamine N-methyltransferase (HNMT), an essential enzyme that regulates histamine levels in tissues by catalyzing its inactivation by N-methylation [44] (Figure 2B). Histamine is a strong agonist of smooth muscle contraction [45] with an important role in maintenance of contractile tone. The importance of HNMT in regulating histamine signaling is illustrated by the increased incidence of bronchial asthma that is seen in patients with certain HNMT polymorphisms [46]. Preferential expression of HNMT in all visceral SMCs suggests a strong, intrinsic ability of these cells to catalyze the local degradation of histamine. The difference in HNMT levels between visceral and bronchial/vascular SMCs may be at the root of the clinical observation that the respiratory and vascular system responses (hypotension, tachycardia, and bronchoconstriction) are more prominent than those of the visceral organs (urinary or digestive tract) when circulating histamine levels are high as a result of systemic anaphylaxes [47]. We have further confirmed the vascular- and visceral-specific expression in these SMC cells with reverse transcriptase PCR (RT-PCR) (Fig 2E).
In an unsupervised analysis (Figure 1A), SMCs isolated from arteries, veins, and bronchi did not segregate according to their tissue types of origin, unlike their endothelial cell counterparts, whose clustering pattern based on global gene expression closely reflected their arterial or venous origins [21]. Taken together, these results show that endothelial cells have stronger intrinsic arterial versus venous gene expression programs than their SMC counterparts, suggesting that endothelial cells, instead of SMCs, are the main cellular determinants for the arterial or venous identity [48]. To further explore this concept, we used a supervised analysis to determine whether and how the diversity in developmental origins, structures, and functions among bronchi, arteries, and veins were associated with identifiable, tissue-specific molecular features. Using a multi-class SAM analysis, at an FDR of 0.1%, we identified a set of 1,037 genes differentially expressed among SMCs isolated from these three distinct kinds of tissues. The distinct patterns of expression of these genes allow the hierarchical clustering of vascular SMC samples into three separate groups largely based on tissue of origin (bronchus, vein, or artery) (Figure 3A). Gene clusters showing arterial- (red), venous- (blue), and bronchial-specific (light blue) expression are expanded and shown (Figure 3B). Several arterial SMCs were grouped in the venous branch, even with this set of selected genes (Figure 3A and 3B).
Interactions between the immune system and vascular SMCs play important roles in the pathogenesis of many vascular diseases. Arterial SMCs, in particular, express a large set of genes that mediate these interactions—such as cytokines/chemokines (IL6, CCL8, CCL7, CCL2, CXCL1, CXCL2, CXCL3, and CXCL6), complement pathway (complement factor B [CFB]), and surface receptor (ICAM1) (Figure 3B)—suggesting that vascular SMCs may themselves mediate recruitment of immune cells to the vascular walls. Perhaps these specific molecular interactions between the immune system and arterial SMCs may contribute to the preferential occurrence of atherosclerosis in arteries.
SMC samples derived from lung tissue (pulmonary artery and bronchus) expressed a common set of genes (Figure 3C), notably including the genes encoding FOXP1, a transcriptional repressor expressed in lung mesenchyme that modulates gene expression in lung tissue [49], and endothelin receptor A (EDNRA), the high-affinity receptor for EDN1, a peptide hormone that stimulates vasoconstriction and proliferation of SMCs [50].
When the global gene expression patterns of cultured SMCs isolated from urinary tract (ureter, urinary bladder, and urethra), colon, and the female reproductive tract (uterus and cervix) were hierarchically clustered, they consistently grouped into three distinct subbranches according to their anatomic origin (Figure 1A). To identify genes expressed differentially among the three subgroups of visceral SMCs, a multi-class SAM was performed. The analysis identified 3,879 genes (represented by 4,889 array elements) with an FDR of 1.5%. We then performed a hierarchical cluster analysis of all visceral SMCs based on expression of these genes (Figure 4A). The distinct gene expression patterns of SMCs from different visceral organs are likely to be related to the characteristic differences in the developmental fates and physiological functions unique to visceral organs.
For example, colonic SMCs expressed fibrillin-2 (Figure 4B), a component of connective tissue microfibrils that is involved in elastic fiber assembly, organization of the ECM (which allows it to influence the physical properties of connective tissue), and regulation of growth factor signaling (which allows it to also direct a broad spectrum of cellular activities) [51]. The urinary tract gene cluster contained several genes that are known to be essential for the development of the renal and urinary tract systems, including the leukemia inhibitory factor (LIF) and oncostatin M (OSM) receptors (LIF receptor, OSM receptor, and IL6ST [52]) and their downstream signaling molecule SOCS3 (Figure 4C). The activation of these receptors by their respective ligands (LIF and OSM, two members of the IL6 family) induces the interconversion between metanephros mesoderm and epithelium in the urinary tract [53,54]. Inactivation of IL6ST, a component of the functional receptors for LIF and OSM, leads to developmental defects in the kidney [53,54]. Bone morphogenic protein 4 (BMP4) was also selectively expressed by SMCs isolated from urinary tract structures (Figure 4C). This protein has an essential role in the inductive signal between the endodermal epithelium and mesenchyme derived from splanchnic mesoderm [55], and BMP4 haplodeficiency can lead to renal defects [56]. It appears, therefore, that expression of several of the genes involved in key inductive signals during urinary tract development persists beyond embryogenesis and remains a feature of locally specialized SMCs, allowing these cells to selectively detect and respond to local induction cues.
We have also found components of a regulatory network consisting of FOXF1, vascular cell adhesion molecule-1 (VCAM1), and hepatocyte growth factor (HGF) selectively expressed in colon and urinary tract SMCs (Figure 4D). FOXF1 is a transcription factor essential for development of visceral splanchnic mesenchyme [57,58]. Haplodeficiency of FOXF1 abolishes the expression of VCAM1 and HGF and leads to structural defects in the gallbladder and other visceral organs [59]. The simultaneous presence of FOXF1, VCAM1, and HGF in colon and urinary tract SMC gene clusters (Figure 4D) suggests that their regulatory relationship may be preserved in these SMCs.
For hormones that reach their distant target tissues through systemic circulation, an important determinant of the specificity of their biological effects is the differential expression of hormone receptor genes in target tissue cells. Oxytocin, the neuropeptide responsible for triggering uterine contractions and lactation [60], is no exception. Oxytocin receptor is present in many organs, suggesting roles in a variety of biological processes [61,62]. In the present study, we have found that oxytocin receptor is expressed specifically in uterus SMCs but none of the other SMCs examined in this study (Figure 4E). This uterus-specific expression can help to explain the unique sensitivity of the uterus to oxytocin during parturition and the clinical usefulness of oxytocin agonists or antagonists, respectively, to induce or prevent labor.
Uterus SMCs also have especially high expression levels of many genes that encode the contractile machinery of SMCs, including tropomyosin 1 and 2, calponin, caldesmon 1, SM-α-actin, MYH9 myosin, heavy and light chain polypeptide, and phospholamban (Figure 4E). These genes are widely considered as cell-type lineage markers for all SMCs; indeed, expression of SM-α-actin was used in this study as a criterion to insure that all cells in our cultured samples were SMCs. The unusually high level of expression of these proteins in uterine SMCs, however, suggests not only that there are quantitative differences in the force-generating apparatus and the contractile capacities among different SMCs but also that these differences may reflect the distinct mechanical requirements of each SM cell type. The uterus, for example, must generate enough force and pressure within a short time frame to deliver the fetus through the birth canal during labor. The SMCs of the adjacent cervix, which relaxes and expands during parturition, express the genes encoding the contractile apparatus at levels significantly lower than that in the uterine SMCs (Figure 4E), despite the overall similarity in the gene expression patterns of uterine and cervical SMCs (Figures 1A and 2B).
SMCs can undergo remarkable phenotypic modulations in response to environmental stimuli, and the association between the occurrence of these transitions and the onset of various human diseases is well established. An example of this phenomenon is the phenotypic change from the quiescent “contractile” to the activated “synthetic” state of vascular SMCs associated with serum exposure. Vascular SMCs are usually shielded from the circulating blood by a layer of overlying endothelial cells. When the integrity of the endothelial cell barrier is compromised, vascular SMCs come into direct contact with all of the components of blood, including serum constituents generated by activation of the coagulation cascade. Although serum is a complex and not fully defined mixture, exposure to serum, the soluble fraction of coagulated blood, represents a physiologically relevant stimulus associated with various forms of vascular injury. Scenarios that can lead to the exposure of vascular SMCs to serum include inflammation; injury; the development of faulty, leaky blood vessels in tumors [63]; and acute endothelial damage from balloon angioplasty [64].
We investigated the response to serum as a simple, controlled, ex vivo model of the temporal response of vascular SMCs to vascular injury induced by serum exposure. Coronary artery SMCs were first placed under replicative quiescence in DMEM media with 0.1% serum for 48 h, then exposed to fully supplemented media, DMEM with 10% serum [65], and the ensuing temporal program of gene expression was followed by analyzing samples at 1, 3, 6, 12, and 24 h with DNA microarrays. We derived the SMC serum responses by performing zero-transformation against three time zero samples. A comparison of the SMC serum response to a previously defined fibroblast serum response [66] (Figure S1) revealed that there was a high degree of similarity between the serum responses of these two cell types and that several gene clusters share similar induction kinetics (clusters 1, 3, and 5 in Figure S1). There were also some distinct features associated with each cell type. For example, several genes induced by serum only in SMCs were implicated in atherosclerosis. A cluster of genes involved in cholesterol biogenesis (cluster 7) was noted to be repressed by serum exposure in fibroblasts but not SMCs [66].
To systemically define a gene signature reflecting the serum response of vascular SMCs, we used SAM to identify 534 unique genes (653 gene elements), with an FDR of 1%, that distinguish three serum-starved vascular SMC samples from five serum-exposed vascular SMC samples (Figure 5A). All the selected genes were induced upon serum exposure. The SMC serum-response signature shares relatively few features with previously defined gene expression signatures of cellular responses to physiological stimuli: 39 genes in common with the fibroblast serum-response signature [66], 19 genes in common with the proliferation signature [67], and 22 genes in common with the epithelial hypoxia response signature [36].
To investigate the possible contribution of the vascular SMC serum-response program in the progression and phenotypes of human cancers, we defined a quantitative “SMC serum-response score” for each sample by simply averaging the relative gene expression levels (logarithmic scale) of the 534 genes in the vascular SMC serum-response gene signature. Calculating the average expression level allowed a quantitative and unbiased determination of the activity of the SMC serum-response program, which was likely to be more stable than the expression of any particular genes in the expression program. This approach also provided a metric that could be applied to tumor samples based on their gene expression patterns to gauge the degree to which a program similar to the vascular SMC serum response was active in each tumor. We first evaluated the vascular SMC serum-response signature in a Stanford breast cancer study representing expression profiles of 85 samples containing normal breast tissues, fibroadenomas, and 78 locally advanced breast cancers with associated extensive clinical and molecular data [68]. Among the vascular SMC serum-response genes, 68 genes were well measured in 80% of the breast samples and used to determine their SMC serum scores (Figure 5B). When we split all ductal adenocarcinoma samples based on their SMC score, 41 samples had an SMC score lower than zero while 44 samples had an SMC score higher than zero (Figure 5B). To investigate the significance of this separation, we compared samples with respect to overall and relapse-free survival. The patients with high SMC scores had significantly lower overall survival (p = 0.0063) and relapse-free survival (p = 0.0007) based on a Cox-regression model (Figure 5C).
To assess the consistency and prognostic significance of the SMC serum-response signature in an independent set of breast cancer samples, we analyzed a published data set from the Netherlands Cancer Institute (NKI), which consisted of 295 early-stage breast cancer samples (stage I and II). Of the 653 genes that comprised the vascular SMC serum-response gene cluster in our microarray dataset, 640 were also represented in the microarrays used in the NKI study. The expression of these 640 genes in the 295 breast cancer samples allowed the separation of all the tumors into two distinct groups based on the SMC serum-response score (Figure 5D). Tumors in the high SMC serum-response group were associated with poorer overall survival (p = 0.001) (Figure 5C) than those in the low SMC serum-response group, confirming the prognostic value of the serum-response gene expression program of vascular SMCs. While thresholding the score at zero produced a striking discrimination of tumor phenotypes, this cut-off point was somewhat arbitrary. To investigate whether an alternative model based on the SMC serum-response signature might improve predictive power, we fit a multivariate Cox model including the SMC serum score in a quantitative form. These curves estimated the differential contribution of the SMC serum score to the (log) relative risk in a continuous fashion (Figure 5E). The results showed a strong positive correlation between the clinical risks with the SMC scores over a wide range of SMC serum scores in which most of the data occurred. Similarly, this SMC serum-response signature was associated with significantly shorter survival in a separate study of ovarian cancers, separating 72 patients with advanced ovarian cancer into groups with markedly different clinical outcomes (M. E. Schaner, personal communication).
Previous studies have shown that gene expression signatures related to cellular proliferation [67], a fibroblast serum response (wound) [66] and a hypoxia response [36] identified breast cancer patients with significantly poorer survival. Although there is rather little overlap in the genes comprising the gene expression signatures of these three distinct biological processes, their efficacy in providing clinical insights in the same set of tumor samples raises the possibility that they may be related. To investigate their potential relationship, we calculated the scores of 295 NKI breast tumors based on the four different gene signatures (Figure 5F). Then we calculated the pair-wise correlation of signature scores to determine their potential relationship. As reported previously, the hypoxia score is only weakly correlated with the fibroblast wound-response (corr = 0.11) or proliferation scores (corr = 0.215) [36]. In contrast, the SMC Serum score was highly correlated with the hypoxia response score (corr = 0.601). This result suggests that the breast cancers with an activated hypoxia response also tend to have an activated program related to the vascular SMC serum response. The vascular SMC serum response might reflect vascular injury; perhaps the hypoxia response and SMC serum response are both elevated in a subset of breast tumors with vascular abnormalities and poor tumor oxygenation. A similar correlation between the hypoxia and SMC serum-response signatures was observed among the breast cancer samples characterized at Stanford (Figure S2). Interestingly, although the fibroblast wound-response signature was also elicited by ex vivo serum exposure, there was only a weak correlation between the fibroblast wound-response score and SMC serum-response score in these cancers (corr = 0.268).
To test whether the vascular SMC serum response contributes anything new and useful to clinical decision making or complements previously established prognostic factors, we evaluated it in a multivariate Cox model that included other established prognostic and clinical factors in the NKI breast cancer dataset. Although it had significant prognostic value by itself, the vascular SMC serum-response signature contributed little additional predictive power when chemotherapy, ER status, tumor size, grade, angioinvasion, and age were included in the predictive model. Therefore, the prognostic information represented by the vascular SMC serum-response signature was already contained in, and perhaps functionally linked to, the classical prognostic factors (Table S1).
It has been claimed that there are approximately 200–400 cell types in the human body [69]. The aims of this study were to begin to examine how many distinct cell types are truly encompassed by the moniker “smooth muscle cell” by systematically comparing gene expression patterns in smooth muscle cells from diverse anatomical sites, and to search for links between these characteristic gene expression patterns and the differentiation, functional specialization, and contribution to pathology of SMCs. The results show that SMCs native to different anatomic sites have distinct, reproducible gene expression patterns that persist through many generations of ex vivo culture in standard media, suggesting that these SMCs comprise distinct differentiated cell types. Similar observations have been made about fibroblasts and endothelial cells in our previous studies [21,25] and for many other superficially similar cell types, notably lymphocytes, whose underlying diversity has emerged through molecular characterization and has been shown to have implications for our fundamental understanding of disease processes [70]. Thus, despite their apparent similarity in morphology and function, many of the cell types that compose the stroma and the vascular systems of various anatomical locations and structures in the human body have diverse, distinct differentiation programs and molecular specializations. Our results on the diversity of fibroblasts [25], endothelial cells [21], smooth muscle cells (in this study), and blood cells [70] highlight the unexpected large number of cell types in the human body and emphasize the need for a better understanding of the fine specialization of these different cell types. The molecular and physiologic differences among superficially homogenous cells may help explain the variations in the physiological behavior of cell types and provide insights into the genetic networks that lead to regional differentiation and to the final local specialization of the architecture and function of human tissues. This depth of understanding will almost certainly have wide-reaching implications for human biology.
Several biological themes emerged in our analysis of SMC gene expression. First, although important common mechanisms for regulating the expression of SMC lineage markers are shared by all SMCs [3], unique differentiation programs characteristic of SMCs at distinct anatomic sites were reflected in the pervasive differences in gene expression patterns. Several growth factors and transcription factors are able to activate SMC differentiation programs [3]; the relative importance and contribution of each growth or transcription factor may vary by SMC subtype. For example, the expression of vascular SMC-specific genes appears to be partially controlled by high TGF-β activity. It is known that TGF-β can trigger SMC differentiation by activating the expression of many SMC lineage genes [71] through the TGF-β control elements present in their promoters [72]. During blood vessel formation, TGF-β is the crucial molecular signal through which endothelial cells recruit splanchnic mesoderm cells and then induce them to differentiate into SMCs [73,74]. The relatively high level of expression of both TGF-β and its receptor may account for the characteristic biosynthetic phenotype of vascular SMCs, as many genes involved in the biosynthesis, trafficking, and modification of ECM proteins are downstream target genes of TGF-β 27]. The preferential expression of ligands, receptors, and other signaling components of the TGF-β signaling pathway by vascular SMCs [75–77] suggests that this feature of the vascular SMC expression program might be maintained via an autocrine mechanism. In contrast, the low level of TGF-β activity in visceral SMCs, reflected in the low level of expression of TGF-β signaling pathway–related genes, suggests that their differentiation programs may be less reliant on TGF-β. Several visceral SMC-specific transcription factors (such as MRF2 (also known as ARID5B) and PBX1) possess the ability to trigger SMC differentiation [6,9] and are likely to contribute to the differentiation programs of these SMCs instead. This divergence of SMC subtype differentiation programs and molecular mechanisms has been found to account for differential SM–major histocompatibility complex and CRP-1 (also known as CEBPE) activation in various SMC subtypes [78,79].
Second, the anatomic site–specific differentiation of SMCs may help explain the particular susceptibility of specific tissues and organs to specific pathogenic processes. For example, vascular SMCs, but not visceral SMCs, express large sets of genes implicated in fibrogenic matrix deposition, tissue remodeling, hypoxia response, and inflammatory responses. These biological processes are implicated in pathogenesis of human diseases affecting SMCs, such as atherosclerosis, hypertension, and asthma. For example, the unique vulnerability of arterial SMCs to atherosclerosis may be related in part to the high levels at which they express TGF-β and the inflammatory response genes (Figure 3B), and the relatively low expression level of a group of genes involved in glutathione biosynthesis in vascular SMCs might contribute to their relative vulnerability to damage from oxidative stress [80].
Most patients with pulmonary artery hypertension have elevated EDN1 levels, and pharmacological blockage of EDN1 activity leads to clinical improvement and better prognosis [81]. EDN1 receptor A mediates a contractile response to EDN1 in SMCs; the high intrinsic expression of its transcript in pulmonary artery SMCs suggests that pulmonary arteries may be especially sensitive to the vasoconstrictive activity of circulating EDN1 [82]. The differential expression of fibrillin 1 and fibrillin 2 in SMCs may be related to the anatomic specificity of diseases that result from defects of these proteins. Fibrillin 1 is expressed exclusively in vascular SMCs while fibrillin 2 is found also in colon SMCs (Figure 2B). Mutations in fibrillin 1 lead to Marfan's syndrome, in which major defects affect the vascular structures, in agreement with vascular SMC expression of fibrillin-1 (Figure 2B). Mutations in fibrillin 2, on the other hand, lead to congenital contractural arachnodactyly. In addition to vascular defects, congenital contractural arachnodactyly patients suffer from abnormalities in the digestive tract, such as duodenal atresia, esophageal atresia, and intestinal malrotation [83]. Thus, links between differential gene expression and disease phenotypes may provide a basis for the localization of defects in syndromes, paralleling similar findings in fibroblasts [25].
Third, specialized gene expression programs reflecting anatomically specific differentiation may underlie the distinct mechanical properties of the SMCs in each tissue. For example, uterine SMCs express genes encoding contractile proteins and sarcomere units at especially high levels, but cervical SMCs, while otherwise very similar, do not. The uterus-specific expression pattern may be related to the uterus' unique need to generate contractile force during parturition and points to a uterus-specific mechanism for fine tuning expression of genes broadly expressed by cells of SMC lineage. Although SRF/myocardin levels are similar in all SMCs, variations in the expression levels of other regulators may account for these fine-tuned differences. For example, the oxytocin receptor, expressed primarily in uterus SMCs, may play such a role, given its ability to trigger muscle differentiation programs [84].
Fourth, variations in gene expression patterns suggest regulatory mechanisms underlying position-specific differentiation. For example, the patterns of expression of a small number of HOX genes can recapitulate the same anatomic clustering of SMCs that was originally achieved with all of the genes in our study, suggesting that positional information encoded in the pattern of expression of the HOX genes may play an important role in determining the distinct molecular phenotypes of SMCs at different sites. The expression of the BMP4 and LIF/OSM receptor in urinary tract SMCs may be a vestige of their developmental origins and suggests the possibility that these cells may retain some of the developmental plasticity of their progenitors. Corollary to all of our findings on differential gene expression in SMCs is the idea that topographically regulated genes, which can be studied ex vivo in cultured cells, may provide an excellent starting point for dissecting the molecular pathways involved in the development of individual tissues and organs.
Finally, we have found that a gene in vitro expression program elicited in vascular SMCs by short-term exposure to serum, an in vitro model of vascular injury [3], is predictive of elevated risk of progression in a variety of human carcinomas. Expression of the vascular SMC serum-response program in human carcinomas may reflect aberrant properties of the tumor vasculature, which can often have biochemical, structural, and compositional abnormalities that result in defective and leaky endothelial cells. Consistent with this possibility, tumors with a gene expression pattern resembling the SMC serum response tend to also exhibit a strong hypoxia response [36], another characteristic of tumors with defective tumor vasculature [19,85]. Such vascular dysfunction may impede drug delivery and create tumor microenvironments that favor metastasis. The possibility that the SMC serum-response signature could identify patients with defective tumor blood vessels, who might benefit from the emerging cancer therapeutics that target tumor vasculature (e.g., the VEGF-specific antibody, bevacizumab), deserves further investigation [19,86,87].
The functional and regulatory specializations revealed by the global gene expression patterns of SMCs from different tissues are of particular clinical importance because SMCs are among the cells most frequently targeted by drugs—e.g., for the reduction of airway resistance, the regulation of blood pressure, or for the induction or inhibition of peristalsis, urination, or labor. Most current treatments affect SMCs indirectly through the autonomic nervous system, and these treatments sometimes have undesirable off-target side effects. Perhaps more specific interventions, with fewer side effects, might be achieved by directly targeting the intended smooth muscle groups. A notable example of such a specifically targeted treatment is provided by the use of oxytocin to induce uterine contraction, by targeting the oxytocin receptor, whose expression in smooth muscle is largely limited to the uterus. Although this study provides only an initial survey of the molecular diversity and heterogeneity of SMCs, it is clear that the genes encoding many potentially “drugable” targets—ion channels, adhesion molecules, and G-protein signaling receptors—are selectively expressed in different SMCs. This fact makes them particularly interesting as potential targets for cell type–specific therapeutics. A more comprehensive study of global expression patterns in anatomically distinct SMCs is likely to uncover additional potential targets for selective SMC-directed therapeutics.
Our microarray experiment data were deposited in the GEO database (http://www.ncbi.nlm.nih.gov/geo/) and assigned the accession number GSE7195.
Supplemental Web site URL: http://microarray-pubs.stanford.edu/smoothmuscle/
Stanford Microarray Database URL: http://genome-www5.stanford.edu/ |
10.1371/journal.pgen.1007156 | Regulation of circadian clock transcriptional output by CLOCK:BMAL1 | The mammalian circadian clock relies on the transcription factor CLOCK:BMAL1 to coordinate the rhythmic expression of 15% of the transcriptome and control the daily regulation of biological functions. The recent characterization of CLOCK:BMAL1 cistrome revealed that although CLOCK:BMAL1 binds synchronously to all of its target genes, its transcriptional output is highly heterogeneous. By performing a meta-analysis of several independent genome-wide datasets, we found that the binding of other transcription factors at CLOCK:BMAL1 enhancers likely contribute to the heterogeneity of CLOCK:BMAL1 transcriptional output. While CLOCK:BMAL1 rhythmic DNA binding promotes rhythmic nucleosome removal, it is not sufficient to generate transcriptionally active enhancers as assessed by H3K27ac signal, RNA Polymerase II recruitment, and eRNA expression. Instead, the transcriptional activity of CLOCK:BMAL1 enhancers appears to rely on the activity of ubiquitously expressed transcription factors, and not tissue-specific transcription factors, recruited at nearby binding sites. The contribution of other transcription factors is exemplified by how fasting, which effects several transcription factors but not CLOCK:BMAL1, either decreases or increases the amplitude of many rhythmically expressed CLOCK:BMAL1 target genes. Together, our analysis suggests that CLOCK:BMAL1 promotes a transcriptionally permissive chromatin landscape that primes its target genes for transcription activation rather than directly activating transcription, and provides a new framework to explain how environmental or pathological conditions can reprogram the rhythmic expression of clock-controlled genes.
| Circadian clocks in mammals rely on the heterodimeric transcription factor CLOCK:BMAL1 to drive rhythmic gene expression and allow biological functions to perform best at the most appropriate time of the day. Investigation of the mechanisms by which CLOCK:BMAL1 regulates its target genes transcription has led to the paradoxical observation that while CLOCK:BMAL1 DNA binding is rhythmic and occurs during the day for all target genes, its transcriptional output is highly heterogeneous. To address this issue, we analyzed independent genome-wide datasets and found that CLOCK:BMAL1 DNA binding during the day is associated with a reorganization of the chromatin that is favorable to transcription activation for all target genes. However, this diurnal CLOCK:BMAL1 DNA binding and chromatin remodeling is not sufficient to promote a transcriptionally active enhancer, therefore suggesting that CLOCK:BMAL1 cooperates with other factors to control the transcription of most of its target genes. This hypothesis is supported by our finding that ubiquitous transcription factors, but not tissue-specific transcription factors, are differentially recruited at CLOCK:BMAL1 enhancers. Altogether, our data highlight the critical role of transcription factors recruited at CLOCK:BMAL1 enhancers in regulating transcription, and present a new mechanistic framework to understand how changes in the environment can reprogram circadian transcriptional programs.
| Virtually every mammalian cell harbors a circadian clock that regulates rhythmic gene expression to enable biological functions to occur at the most appropriate time of day. Circadian clocks rely on transcriptional feedback loops which are initiated in mammals by the heterodimeric transcription factor CLOCK:BMAL1 [for review, 1]. CLOCK:BMAL1 rhythmically binds to DNA to activate the rhythmic transcription of the core clock genes Period (Per1, Per2, Per3), Cryptochrome (Cry1 and Cry2), Rev-erb (Rev-erbα and Rev-erbβ) and Ror (Rorα, Rorβ and Rorγ). Upon expression and maturation, PERs and CRYs form a repressive complex that rhythmically inhibits CLOCK:BMAL1-mediated transcription first on-DNA and then off-DNA [2–5]. Furthermore, REV-ERBs and RORs rhythmically regulate Bmal1 expression by repressing or activating its transcription, which promotes robustness of circadian oscillations [6, 7]. In addition to activating the rhythmic transcription of core clock components, CLOCK:BMAL1 also regulates rhythmic expression of thousands of clock-controlled genes to generate oscillations in biochemistry, physiology and behavior, and thus control the rhythmic organization of most biological functions [8–10].
Characterizing the mechanisms through which CLOCK:BMAL1 regulates expression of its target genes has largely been carried out by determining how CLOCK:BMAL1 regulates the transcription of core clock genes (Per, Cry and Rev-erb), and target genes (e.g., Dbp, named for D site of albumin promoter binding protein). Results from many laboratories show that the rhythmic binding of CLOCK:BMAL1 to e-boxes located in core clock gene promoters is necessary and sufficient for rhythmic transcription [4, 5, 11–13]. Upon DNA binding during the light phase, CLOCK:BMAL1 promotes chromatin modifications by recruiting histone-modifying enzymes to core clock gene promoters and enhancers. These enzymes include the histone acetyltransferases p300 and CBP, which mediate the acetylation of H3K9 and H3K27, and the histone methyltransferases MLL1 and MLL3 (Myeloid/Lymphoid Or Mixed-Lineage Leukemia 1 and 3), which promote the tri-methylation of H3K4 [3, 14–19]. CLOCK:BMAL1 rhythmic DNA binding was also recently shown to promote rhythmic nucleosome removal, thereby generating a chromatin landscape that is favorable for the binding of other transcription factors at its enhancers [20]. Finally, CLOCK:BMAL1 recruits transcriptional co-activators, including components of the mediator complex and RNA Polymerase II (Pol II) to initiate core clock gene transcription [3, 21, 22]. During the repression phase in the early night, binding of the PER/CRY complex to DNA-bound CLOCK:BMAL1 is accompanied by the co-recruitment of histone deacetylases and demethylases and the removal of the H3K9ac, H3K27ac and H3K4me3 marks [19, 23–27]. While these mechanisms are required for CLOCK:BMAL1-mediated transcription activation of core clock genes, it still remains unclear if the same mechanisms regulate the rhythmic expression of clock-controlled genes.
The recent characterization of CLOCK and BMAL1 mouse liver cistromes revealed that although CLOCK:BMAL1 binds synchronously during the middle of the day to thousands of enhancers and promoters, the transcription of its target genes is highly heterogeneous [3, 28–30]. Indeed, not all CLOCK:BMAL1 target genes are rhythmically expressed, and a large fraction of the rhythmically expressed target genes are transcribed at night, in antiphase to maximal CLOCK:BMAL1 DNA binding [28]. These data therefore suggest that the mechanisms by which CLOCK:BMAL1 regulates transcription of core clock genes differs from the regulation of other clock-controlled genes, and that additional mechanisms account for the activation of rhythmic gene expression by the circadian clock.
To uncover these mechanisms and to delineate the transcriptional logic underlying CLOCK:BMAL1 heterogeneous transcriptional output, we performed a meta-analysis of genome-wide datasets investigating the molecular events occurring at CLOCK:BMAL1 DNA binding sites, including CLOCK:BMAL1 rhythmic DNA binding, epigenetic modifications and transcription activation. Our analysis reveals that while CLOCK:BMAL1 DNA binding is sufficient to decondense the chromatin and prime its enhancers for transcriptional activation, it is not sufficient to generate transcriptionally active enhancers. Our results also indicate that many transcription factors bind to CLOCK:BMAL1 enhancers, and their recruitment likely contributes to CLOCK:BMAL1 clock-controlled transcriptional output. Altogether, our data support that CLOCK:BMAL1 regulation of clock-controlled gene expression relies on the cooperation between CLOCK:BMAL1 and other transcription factors. Furthermore, our data also suggest that a major role of CLOCK:BMAL1 is to generate a permissive chromatin landscape to rhythmically prime its enhancers for the recruitment of other transcription factors, rather than directly promoting transcription activation.
To characterize the mechanisms by which CLOCK:BMAL1 regulates the transcriptional activity of its target genes at the genome-wide level in the mouse liver, we first generated a list of high-confidence CLOCK:BMAL1 DNA binding sites by determining the overlap between CLOCK and BMAL1 ChIP-Seq peaks in the mouse liver [3]. This analysis resulted in a list of 3217 CLOCK:BMAL1 binding sites, of which 2458 peaks can be assigned to a direct target gene (i.e., a CLOCK:BMAL1 peak located between -10kb of a target gene transcription start site and +1kb of a target gene transcription termination site; see S1 Fig, S1 Table, and methods section for details). To determine the extent to which rhythmic CLOCK:BMAL1 DNA binding contributes to rhythmic transcription activation at the genome-wide level, we used a public mouse liver Nascent-Seq dataset that characterized the levels of nascent RNA expression over the course of a 24-hr day [28]. A Nascent-Seq dataset was preferred over RNA-Seq because nascent RNA expression directly reflects transcription activation, and is unaffected by the post-transcriptional regulations that contribute to rhythmic mRNA expression in the mouse liver [3, 28, 31]. We found that only a small fraction of CLOCK:BMAL1 target genes are rhythmically transcribed (~26%; S1 Fig). Noticeably, not all rhythmic target genes are transcribed during the day, i.e., coincidently with rhythmic CLOCK:BMAL1 DNA binding (ZT02-ZT12). Indeed, 38% of the rhythmic CLOCK:BMAL1 target genes exhibit a peak of transcription between ZT12 and ZT02, out-of-phase with the rhythmic DNA binding of CLOCK:BMAL1 (n = 124 CLOCK:BMAL1 peaks) (Fig 1A–1C; S1 Fig). Importantly, our analysis also reveals that the majority of CLOCK:BMAL1 direct target genes are either arrhythmically transcribed (AR; n = 654 CLOCK:BMAL1 peaks) or not expressed (NE; n = 291 CLOCK:BMAL1 peaks) (Fig 1A–1D; S1 Fig). To determine if this result may be due to comparing samples collected in constant darkness (ChIP-Seq) and in a light:dark (LD) cycle (Nascent-Seq), we also analyzed a mouse liver BMAL1 ChIP-Seq rhythm performed under LD condition [29]. BMAL1 binding phase and ChIP-Seq signal under LD condition both exhibit a remarkably high level of similarity to those under DD conditions, and this even for the AR or NE target genes (S2 Fig). This therefore suggests that the large number of arrhythmically transcribed or not expressed CLOCK:BMAL1 target genes is not a consequence of using datasets generated under different lighting conditions. Taken together, these results indicate that the mechanisms underlying CLOCK:BMAL1-mediated rhythmic transcription of core clock genes (i.e., Per1, Per2, Per3, Cry1, Cry2, Rev-erbα, Rev-erbβ and Dbp) are not prevalent at the genome-wide level. They also suggest that the rhythmic recruitment of CLOCK:BMAL1 at its target gene promoters and enhancers is not sufficient to activate transcription for the majority of its target genes.
To investigate the mechanisms underlying CLOCK:BMAL1 heterogeneous transcriptional output, we first examined if differences in the phase, intensity or location of CLOCK:BMAL1 DNA binding might explain the differences in transcription activation. The phase of CLOCK:BMAL1 DNA binding was found to be indistinguishable between all four transcriptional output categories, as both CLOCK and BMAL1 rhythmically bind to DNA with a peak between ZT3 and ZT9 for almost all target genes (Fig 1A, 1B and 1E). We then used CLOCK and BMAL1 ChIP-Seq signal as a readout to determine DNA binding intensity, and found that both CLOCK and BMAL1 ChIP-Seq signals are significantly higher at DNA binding sites targeting the in-phase transcriptional cyclers (Rinφ) when compared to peaks targeting the 3 other groups (out-of phase cyclers, arrhythmically expressed and non-expressed target genes) (Fig 1A, 1B and 1F; Kruskal-Wallis test, p < 0.05). Remarkably, the binding intensity of CLOCK and BMAL1 at non-expressed target genes (NE) is similar to the binding intensity observed at the out-of-phase transcriptional cyclers (Ro/φ) and arrhythmically transcribed (AR) target genes, suggesting that CLOCK:BMAL1 DNA binding alone does not directly activate transcription at most of its target genes (e.g., comparisons between Fig 1D and 1F). To verify that these results are not due to the cut-offs we used to partition CLOCK:BMAL1 transcriptional output, we performed similar analyses using direct correlations between BMAL1 or CLOCK ChIP-Seq signal and the phase of rhythmic transcription, as well as by partitioning rhythmic target genes in five groups of equal sizes. These analyses confirmed our results (S3 and S4 Figs). While rhythmically transcribed target genes peaking from ZT5 to ZT13 exhibit higher BMAL1 and CLOCK ChIP-Seq signal, no differences in DNA binding signal were observed between the rhythmically expressed targets peaking from ZT13 to ZT5 and the AR and NE groups (S3 Fig). In addition, we did not find any significant correlation between either CLOCK or BMAL1 ChIP-Seq signals and the phase of DNA binding or the phase of rhythmic transcription (S4 Fig). Because CLOCK:BMAL1 peaks targeting core clock genes are enriched in the Rinφ and Ro/φ groups and exhibit higher ChIP-Seq signal than clock-controlled (output) genes, we also compared CLOCK and BMAL1 ChIP-Seq signals between groups after removing peaks targeting the core clock genes (i.e., comparing clock-controlled genes only). Whereas BMAL1 ChIP-Seq signal intensity was still significantly higher at the Rinφ target genes compared to the three other groups, CLOCK DNA binding intensity was similar between all 4 groups (Fig 1F). Our data therefore indicate that while higher BMAL1 DNA binding signal may contribute to Rinφ transcription, the different transcriptional output of CLOCK:BMAL1 target genes cannot be explained solely by differences in CLOCK:BMAL1 DNA binding intensity.
We also examined if differences in the location of CLOCK:BMAL1 DNA binding sites are associated with differences in transcriptional output by mapping CLOCK:BMAL1 peaks to either the transcription start site (TSS), gene body or extended promoter (-10 kb to -1 kb from the TSS) of their target genes. While the AR and NE groups were found to be statistically different (chi square test; p < 0.05), we did not observe any differences between the rhythmic target groups (Rinφ and Ro/φ) and the arrhythmically or not expressed groups (Fig 1G). The vast majority of CLOCK:BMAL1 peaks were located within enhancers (i.e., gene body or extended promoter), and only ~10–19% of CLOCK:BMAL1 peaks were mapped to TSS. Finally, we examined if differences in the number of genes targeted by multiple CLOCK:BMAL1 peaks were associated with differences in transcriptional output. We found that in-phase transcriptional cyclers were more frequently targeted by multiple CLOCK:BMAL1 peaks, and that conversely, non-expressed target genes were less frequently targeted by multiple peaks (S5 Fig). However, the lack of differences between the Ro/φ and AR groups indicates that the presence of multiple ChIP-Seq peaks does not directly influence the rhythmicity of CLOCK:BMAL1 target genes.
Taken together, our analysis indicates that CLOCK:BMAL1 heterogeneous transcriptional output can not be simply attributed to differences in the phase, intensity or location of CLOCK and BMAL1 binding to the DNA. While stronger DNA binding intensity may contribute to rhythmic transcription during the light phase, additional mechanisms are likely to contribute to CLOCK:BMAL1 transcriptional output heterogeneity.
Circadian repression in mammals is initiated at the beginning of the night by the recruitment of the PER/CRY repressive complex and its associated histone deacetylases and methyltransferases to CLOCK:BMAL1 on DNA [23–25, 27, 32]. Because a differential recruitment of PERs and CRYs at CLOCK:BMAL1 DNA binding sites could lead to differences in CLOCK:BMAL1-mediated transcriptional output (e.g., decreased recruitment at arrhythmically transcribed target genes, delayed recruitment of out-of-phase transcriptional cyclers, etc.), we investigated the DNA binding profile of PER1, PER2, CRY1 and CRY2 at CLOCK:BMAL1 DNA binding sites for each of the four transcriptional output groups using publically available ChIP-Seq datasets [3].
Our analysis shows that PER1, PER2 and CRY2 are rhythmically recruited at CLOCK:BMAL1 DNA binding sites with little difference between the four transcriptional output groups (S6 Fig). Maximal DNA binding for PER1, PER2 and CRY2 occur at CT12-16 for all groups, and differences were mostly observed for CRY2, where higher ChIP-Seq signal was found for rhythmically expressed target genes (S6 Fig). On the other hand, analysis of CRY1 recruitment to CLOCK:BMAL1-bound enhancers revealed more pronounced differences between all four groups. CRY1 is a potent circadian repressor that is preferentially recruited at the beginning of the light phase just prior CLOCK:BMAL1 transcription activation (i.e., CT0-4), a mechanism proposed to poise CLOCK:BMAL1 for transcription activation [3]. We found that CRY1 recruitment at CT4 is significantly higher for rhythmically transcribed target genes (both Rinφ and Ro/φ) than for arrhythmically transcribed and non-expressed genes (S6 Fig). In addition, CRY1 recruitment was significantly decreased in non-expressed CLOCK:BMAL1 target genes than arrhythmic genes at CT4. These data thus suggest that CRY1 recruitment to CLOCK:BMAL1 DNA binding sites is, in addition to its well-characterized repressive effect, linked to rhythmic transcription activation. Consistent with this hypothesis are the higher levels for Ro/φ at CT12 compared to Rinφ (S6 Fig).
Based on the mechanisms mediating the delayed transcription of the CLOCK:BMAL1 target gene Cry1 [33], a model incorporating the nuclear receptors Rev-erb (repressor) and Ror (activator), and the D-box transcriptional factors E4bp4 (also called Nfil3; repressor), Dbp, Hlf and Tef (activators) has been proposed to explain the different phases of rhythmic gene expression in the mouse liver [33, 34]. In this model, co-binding of D-box transcription factors with CLOCK:BMAL1 is proposed to delay the phase of CLOCK:BMAL1 target genes from the morning to the afternoon (i.e., from ~ZT6 to ~ZT12), and additional binding of REV-ERBs and RORs would further delay the phase of transcription to the night (e.g., ~ZT18). To test if the binding of REV-ERBs and D-box transcription factors contribute to the delay of the out-of-phase CLOCK:BMAL1 target genes, we used publicly available ChIP-Seq datasets to determine REV-ERBα, REV-ERBβ [35], and E4BP4 [36] DNA binding intensity at CLOCK:BMAL1 enhancers. We find that REV-ERBα and REV-ERBβ DNA binding, which peaks at ZT10 for all target genes [35], is significantly higher at CLOCK:BMAL1 peaks targeting genes transcribed during the night [consistent with the model proposed based on Cry1 expression; 33, 34], and no differences were observed between Rinφ, AR and NE target genes (Fig 2A and 2B; Kruskal-Wallis test, p < 0.05). The binding of E4BP4, which is maximal at ZT22 [36], was also enriched at CLOCK:BMAL1 enhancers targeting the Ro/φ genes, but to a lesser extent than what was observed for the REV-ERBs (Fig 2C). In particular, no significant difference in enrichment was observed between the Rinφ and the Ro/φ groups, perhaps because co-binding of both CLOCK:BMAL1 and D-box transcription factors drives rhythmic transcription in the afternoon around ZT12, a time used for our cut-off to differentiate the in-phase from out-of-phase transcription cyclers. In summary, our analysis indicates that the binding of REV-ERBα and REV-ERBβ (and eventually E4BP4) at CLOCK:BMAL1 enhancers may, as suggested by others [33, 34], contribute to the delayed transcription of rhythmically expressed CLOCK:BMAL1 target genes.
Our inability to detect substantial differences in CLOCK:BMAL1 DNA binding that would explain the heterogeneity of CLOCK:BMAL1 transcriptional output suggests that mechanisms other than the recruitment of core clock proteins to target gene promoters control CLOCK:BMAL1-mediated transcription. The recent finding that CLOCK:BMAL1 promotes the removal of nucleosomes when bound to DNA may represent one of these mechanisms [20]. Indeed, by mediating the removal of nucleosomes, CLOCK:BMAL1 would enable other transcription factors to access CLOCK:BMAL1 enhancers (most transcription factors bind better to naked DNA than DNA wrapped around nucleosomes).
To test if CLOCK:BMAL1-mediated nucleosome removal can contribute to the heterogeneity of CLOCK:BMAL1 transcriptional output, we examined mouse liver nucleosome signal over the 24-hr day at CLOCK:BMAL1 DNA binding sites for each of the transcriptional output groups, using a public MNase-Seq dataset (micrococcal nuclease digestion of mouse liver chromatin at 6-time points and high-throughput sequencing of mononucleosomes [20]). Our analysis reveals that nucleosome signal is rhythmic at CLOCK:BMAL1 DNA binding sites for each of the transcriptional output categories, i.e., even at CLOCK:BMAL1 DNA binding sites targeting arrhythmically transcribed and non-expressed genes (Fig 3A–3D; S7 Fig). Importantly, the phase of the rhythms is similar for all groups and minimal nucleosome signal coincides with maximal CLOCK:BMAL1 DNA binding during the light phase. Closer inspection of the levels of nucleosome signal and rhythm amplitude reveals important differences between each of the four transcriptional output categories (Fig 3E). First, the amplitude of the rhythms is significantly decreased for arrhythmically transcribed target genes. While minimal levels of nucleosome signal during the day are similar between the AR and Rinφ groups, nucleosome signal remains low during the night (i.e., when CLOCK:BMAL1 is not bound to DNA) at CLOCK:BMAL1 peaks targeting AR genes (Fig 3E). This suggests that some transcription factors may be still bound to DNA during the night in the AR group (when CLOCK:BMAL1 is not bound to DNA), thereby preventing the reformation of nucleosomes. This may promote transcription at night and thus lead to arrhythmic transcription. Second, the overall nucleosome signal is significantly lower at CLOCK:BMAL1 peaks targeting Ro/φ genes than for Rinφ genes, without any significant effect on the amplitude of the rhythm (Fig 3E). In addition, the time of minimal nucleosome signal is delayed by 4 hours between Rinφ and Ro/φ: while it coincides with the time of maximal CLOCK:BMAL1 DNA binding for Rinφ genes (ZT06), minimal nucleosome signal is observed at ZT10 for Ro/φ genes. This delayed nucleosome signal for the out-of-phase transcriptional cyclers may be explained by the significant recruitment of REV-ERBα and REV-ERBβ (Fig 2A and 2B). Indeed, CLOCK:BMAL1 has been recently proposed to facilitate circadian repression by promoting the recruitment of REV-ERBα through chromatin decondensation [37]. Thus, the increased binding of REV-ERBs at CLOCK:BMAL1 enhancers at ZT10 may promote a further decrease in nucleosome signal. Furthermore, anti-phase binding of the RORs on ROREs during the night would prevent a full nucleosome re-compaction, thereby promoting lower levels of nucleosome signal at CLOCK:BMAL1 peaks targeting Ro/φ target genes. Finally, there are no significant differences of nucleosome signal between CLOCK:BMAL1 DNA binding sites targeting in-phase transcriptional cyclers than those targeting non-expressed target genes (Fig 3E). This intriguing result suggests that although CLOCK:BMAL1 is unable to promote transcription activation at NE target genes, its rhythmic DNA binding still mediates a rhythm in nucleosome signal. One possible explanation for this result is that CLOCK:BMAL1 decondenses the chromatin to facilitate the binding of other transcription factors, but those would not be recruited at NE target genes except under specific conditions (e.g. environmental stressors), thereby preventing activation of transcription under standard conditions.
Our data indicate that CLOCK:BMAL1 rhythmic DNA binding promotes the rhythmic removal of nucleosomes at all four transcriptional output categories. We then asked if CLOCK:BMAL1 can also promote the formation of transcriptionally active enhancers. To address this question, we used public datasets [3, 36, 38] to examine the rhythmic pattern of two independent markers of enhancer activity at CLOCK:BMAL1 DNA binding sites: the post-translational modification H3K27 acetylation (H3K27ac), which positively correlates with enhancer activity at almost all enhancers and TSS [39], and the expression levels of enhancer RNA (eRNA), which are relatively short non-coding RNA molecules (50–2000 nucleotides) transcribed at active enhancer regions [40].
While mouse liver H3K27ac ChIP-Seq signal is rhythmic and high during the light phase at CLOCK:BMAL1 DNA binding sites targeting in-phase transcriptional cyclers (consistent with CLOCK:BMAL1 directly facilitating the acetylation of H3K27; Fig 3F and S8 Fig), significant differences were observed at CLOCK:BMAL1 DNA binding sites targeting the other 3 transcriptional output categories. Rhythmic H3K27ac rhythm is delayed for the out-of-phase transcriptional cyclers, and the amplitude of H3K27ac rhythm is significantly dampened at CLOCK:BMAL1 DNA binding sites targeting arrhythmically transcribed genes (Fig 3F and S8A Fig). Remarkably, levels of H3K27ac are close to background levels at CLOCK:BMAL1 peaks targeting non-expressed genes. Given that CLOCK:BMAL1 rhythmically binds to relatively similar levels for all four transcriptional output categories, our analysis suggests that CLOCK:BMAL1 DNA binding does not directly contribute to the acetylation of H3K27.
To extend on this observation, we then examined another marker of enhancer transcriptional activity by assessing eRNA expression at CLOCK:BMAL1 DNA binding sites using a publicly available GRO-Seq dataset [36]. The analysis confirmed the results obtained with H3K27ac (Fig 3G). Rhythmic eRNA expression is only observed at CLOCK:BMAL1 enhancers targeting rhythmically transcribed genes, and eRNA expression at enhancers targeting non-expressed genes is dramatically decreased to levels close to background (Fig 3G). Importantly, these differences in eRNA expression between the four CLOCK:BMAL1 transcriptional output categories are further corroborated by similar variations in RNA Polymerase II (Pol II) ChIP-Seq signal at CLOCK:BMAL1 enhancers (S8C Fig). Altogether, our analysis therefore demonstrates that, contrary to what has been typically found for core clock genes, CLOCK:BMAL1 DNA binding is not sufficient to promote the activation of its enhancers. Instead, our results suggest that CLOCK:BMAL1 rhythmically opens the chromatin to facilitate the binding of other transcription factors at its enhancers, and that the nature of these transcription factors (e.g., activators, repressors) significantly contributes to CLOCK:BMAL1 transcriptional output.
To test our hypothesis that transcription factors bind at CLOCK:BMAL1 enhancers to contribute to their transcriptional activity and thereby impact on CLOCK:BMAL1-mediated transcription, we assessed if transcription factors were differentially recruited at CLOCK:BMAL1 DNA binding sites within each transcriptional output group. To this end, we performed a DNA binding motif analysis using HOMER Software Suite that we further validated using mouse liver transcription factor ChIP-Seq datasets.
As expected, the motif analysis revealed that CLOCK:BMAL1 DNA binding motif (e-box of the sequence CACGTG) is highly enriched at CLOCK:BMAL1 enhancers for all transcriptional output categories (Fig 4A). Surprisingly however, we found that motifs for liver-specific transcription factors (e.g., Cebp, Hnf1, Hnf4 and Hnf6) were also enriched for all four transcriptional output categories, and thus even at CLOCK:BMAL1 enhancers targeting non-expressed genes (Fig 4B, S9 Fig, and S2 Table). On the contrary, motifs for ubiquitous transcription factors (u-TFs; broadly expressed transcription factors with a transcriptional activity regulated by external factors) were almost always enriched for specific CLOCK:BMAL1 transcriptional output groups (Fig 4C, S9 Fig, and S2 Table). For example, CRE motif was enriched at all CLOCK:BMAL1 enhancers except those targeting out-of-phase transcriptional cyclers, and FXR motif was enriched at all CLOCK:BMAL1 enhancers except those targeting out-of-phase transcriptional cyclers. Noticeably, the motifs for NF-κB [which binds to DNA and becomes transcriptionally active upon infection and inflammation; 41, 42], and CTCF [which establishes discrete functional chromatin domains by promoting DNA looping; 43, 44, 45] were enriched at enhancers targeting non-expressed genes.
To assess the relevance of this difference of motif enrichments between tissue-specific (ts-TFs) and u-TFs, we determined the DNA binding pattern of several transcription factors at CLOCK:BMAL1 enhancers in the mouse liver using publicly available transcription factors ChIP-Seq datasets [46–51]. This in vivo analysis largely confirmed the computational motif analysis: most liver-specific TFs were found to bind at CLOCK:BMAL1 DNA binding sites independently of the transcriptional output, whereas u-TFs were more specifically enriched in specific CLOCK:BMAL1 transcriptional output groups (Fig 4D, E, S10 Fig, and S3 Table). For example, Hnf4a and Hnf1 are the only liver-specific TF to exhibit a differential binding between CLOCK:BMAL1 transcriptional output groups of the six TFs tested (Foxa1, Foxa2, Hnf1, Hnf4A, Hnf6 and Cepba). Conversely, all twelve u-TFs investigated exhibit DNA binding differences at CLOCK:BMAL1 enhancers between categories of transcriptional output (Fig 4D and 4E, S10 Fig). Although each u-TF bound to different subsets of CLOCK:BMAL1 enhancers, u-TF recruitment was generally higher in rhythmically expressed target genes and lower in non-expressed target genes compared to the arrhythmic CLOCK:BMAL1 target group. (Fig 4D and 4E, S10 Fig, and S3 Table). To further characterize the differences in TF DNA binding between CLOCK:BMAL1, u-TFs and ts-TFs, we computed a TF DNA binding variability index by calculating the standard deviation of the ChIP-Seq signal between the 4 CLOCK:BMAL1 transcriptional output groups (see methods for details). We found that the DNA binding variability at CLOCK:BMAL1 peaks is comparable between CLOCK:BMAL1 and ts-TFs, whereas there is significantly more variability for u-TFs than for CLOCK and BMAL1 when peaks targeting core clock genes are removed from the analysis (Fig 4G). While there are variability index differences among ts-TFs and u-TFs, this analysis further supports our finding that u-TF recruitment at CLOCK:BMAL1 peaks is globally more variable than for ts-TF (Fig 4G).
Altogether, our data indicate that the mechanisms by which CLOCK:BMAL1 regulates transcription of clock-controlled genes differ from the well-characterized CLOCK:BMAL1-mediated regulation of core clock gene expression. Specifically, our data show that although CLOCK:BMAL1 mediates rhythmic nucleosome removal at its enhancers, it is not sufficient to generate an active enhancer or drive rhythmic transcription. We thus propose a model whereby CLOCK:BMAL1 regulates transcription of clock-controlled genes by rhythmically opening chromatin to facilitate the binding of other transcription factors at its enhancers (Fig 5A). This possibility is supported by results showing that nucleosome signal is rhythmic at the DNA binding sites of several TFs when those sites are located close to a CLOCK:BMAL1 peak, and not rhythmic when CLOCK:BMAL1 binding is absent (S11 Fig). Consequently, the transcriptional activities of these transcription factors would dictate the transcriptional outcome of clock-controlled genes rather than CLOCK:BMAL1 (Fig 5A). For example, binding of positive transcription factors along with CLOCK:BMAL1 would activate enhancers and lead to transcription activation during the day, whereas binding of transcriptional repressors (e.g., REV-ERBα/β) would inhibit CLOCK:BMAL1 enhancer activity and thereby contribute to rhythmic transcription peaking during the night, in anti-phase with CLOCK:BMAL1 DNA binding (Fig 5A). If no transcription factors are recruited (e.g., inducible transcription factors), CLOCK:BMAL1 enhancers remain silent and target genes are not expressed or are arrhythmically expressed (Fig 5A). Arrhythmically expressed genes at CLOCK:BMAL1 enhancers may also have positive transcription factors bound at all times overriding the absence of CLOCK:BMAL1 DNA binding at night (see result section about rhythmic nucleosome signal and Fig 3E). Our results also suggest that u-TFs regulate CLOCK:BMAL1 transcriptional output more prevalently than ts-TFs. It may be that ts-TFs facilitate the binding of CLOCK:BMAL1 at tissue-specific enhancers rather than contributing to CLOCK:BMAL1 transcriptional output (see discussion). To validate this model experimentally, we investigated how i) Bmal1 knockout, and ii) changes in environmental conditions (that alter u-TFs transcriptional activities) affect CLOCK:BMAL1 transcriptional output.
If the activity of u-TFs contributes to CLOCK:BMAL1 regulation of clock-controlled gene transcription, then, a knockout of Bmal1 (which eliminates CLOCK:BMAL1-mediated transcription [52]) should differentially affect the expression of CLOCK:BMAL1 target genes. Specifically, target gene expression levels in Bmal1-/- mouse should be arrhythmic and low for the in-phase transcriptional cyclers (no recruitment of positive transcription factors at CLOCK:BMAL1 enhancers), while they should be arrhythmic and high for the out-phase transcription cyclers (no recruitment of repressors at CLOCK:BMAL1 enhancers). These effects should also be more obvious during the light phase when CLOCK:BMAL1 binds to DNA. In addition, Bmal1 knockout should have a reduced effect on arrhythmically and non-expressed target genes. These predictions were confirmed by analyzing a public dataset that characterized the genome-wide effect of Bmal1 knockout in the mouse liver using RNA-Seq of rRNA-depleted total RNA (Fig 5B and 5C) [53]. For both intronic and exonic RNA-Seq signal, the expression of Rinφ genes in Bmal1-/- mouse liver is at the trough level of wild-type mice, and at peak levels in Ro/φ genes. Moreover, Bmal1 knockout does not significantly affect the expression levels of arrhythmic and non-expressed CLOCK:BMAL1 target genes (Fig 5B and 5C).
Our model also predicts that the transcriptional output of CLOCK:BMAL1 target genes can be altered by environmental changes that affect u-TF DNA binding capacity. External signals that activate or repress the binding of u-TFs are predicted to impact CLOCK:BMAL1 cooperation with other transcription factors, and thereby change the transcriptional output of CLOCK:BMAL1 target genes. For example, signals that enable the recruitment of positive transcription factors at CLOCK:BMAL1 enhancers could increase the amplitude of rhythmic transcription and/or initiate the rhythmic expression of target genes that are arrhythmic under control conditions. Conversely, signals that inhibit the binding of transcription factors that normally cooperate with CLOCK:BMAL1 could blunt the rhythmic expression of some CLOCK:BMAL1 target genes. To test this hypothesis, we analyzed how fasting, which is known to affect the transcriptional activity of many u-TFs [54, 55], alters CLOCK:BMAL1 target gene expression in the mouse liver using a public dataset [56]. Strikingly, while the expression of Clock, Bmal1 and several direct rhythmic target genes (e.g., Phf17, Slc16a2) are unaffected by fasting, some other targets exhibit a significantly altered gene expression profile (Fig 5D–5G, S12 Fig for additional examples). For example, some rhythmic target genes become arrhythmically expressed under fasting (e.g., Sgk2, Flcn) while other targets exhibit an increased amplitude of expression (e.g., Gnat1, Gm129) (Fig 5E). Remarkably, some direct CLOCK:BMAL1 target genes that are arrhythmically or not expressed under ad libitum condition become rhythmically expressed under fasting condition (Fig 5F and 5G). Because not all CLOCK:BMAL1 target genes are equally affected by fasting, these results cannot simply be explained by a global change in CLOCK:BMAL1 transcriptional activity under fasting condition. One possibility is that fasting enhances or represses the transcriptional capabilities of several u-TFs that cooperate with CLOCK:BMAL1, thereby altering the transcriptional output of many direct CLOCK:BMAL1 target genes. Similar results were found by analyzing a public dataset investigating the effect of high-fat diet on rhythmic gene expression in the mouse liver (S13 Fig) [57].
Based on the mechanisms by which CLOCK:BMAL1 regulates the expression of several core clock genes, it is commonly assumed that the rhythmic binding of CLOCK:BMAL1 to DNA is necessary and sufficient to drive the rhythmic transcription of its target genes. However, the recent characterization of CLOCK and BMAL1 cistromes in the mouse liver revealed that CLOCK:BMAL1 target gene transcription is highly heterogeneous, thereby suggesting that CLOCK:BMAL1 regulation of clock-controlled gene expression relies on more complex mechanisms than those underlying core clock gene rhythmic transcription [3, 28–30]. We report here that CLOCK:BMAL1 heterogeneous transcriptional output does not stem from differences in the DNA binding profiles of CLOCK and BMAL1, or the PER/CRY circadian repressive complex. Instead, we found that while CLOCK:BMAL1 rhythmically promotes chromatin decondensation at its enhancers, it is not sufficient to promote transcription activation. Based on these data and the characterization of transcription factor DNA binding profiles at CLOCK:BMAL1 enhancers, we propose that CLOCK:BMAL1 regulates the expression of clock-controlled genes by generating a permissive chromatin landscape that facilitates the binding of other transcription factors at its enhancers rather than directly promoting rhythmic transcription. Interestingly, analysis of a random set of genes not directly targeted by CLOCK:BMAL1 but exhibiting similar profiles of expression of the four CLOCK:BMAL1 transcriptional output categories suggests that this mechanism is largely specific to CLOCK:BMAL1 (S14 Fig).
The current models describing the regulation of rhythmic gene expression by circadian clocks in other eukaryotes are also based on how core clock components regulate their own transcription via transcriptional feedback loops. For example, the mechanisms underlying transcriptional regulation by CLOCK:BMAL1 orthologs in Neurospora (WCC for White Collar Complex) and Drosophila (CLK:CYC heterodimer) are based largely on how they regulate the expression of the core clock genes frequency (in Neurospora), and period and timeless (in Drosophila) [2, 58–60]. Given that the circadian clock mechanisms are highly conserved in eukaryotes, it is likely that both WCC and CLK:CYC also regulate their target gene expression by remodeling the chromatin and facilitating the binding of other transcription factors. Consistent with this hypothesis, WCC and CLK:CYC transcriptional outputs are also heterogeneous [61, 62], and both recruit chromatin remodelers to promote nucleosome eviction at their binding sites [63–67].
The recent characterization of many transcription factor cistromes revealed that the number of transcription factor DNA binding sites often exceeds the number of anticipated target genes, suggesting that many of these DNA binding sites are non-functional [68, 69]. Although many CLOCK:BMAL1 DNA binding sites could be considered as non-functional because they target arrhythmically or not expressed genes, the observation that CLOCK:BMAL1 rhythmically promotes nucleosome eviction at enhancers targeting both arrhythmically expressed (albeit with a decreased amplitude) and non-expressed genes instead indicates that CLOCK:BMAL1 rhythmic DNA binding is not “silent”. More specifically, our data suggest that the majority of CLOCK:BMAL1 DNA binding events are functional, in that they rhythmically shape the chromatin landscape, and that transcription activation requires additional downstream events to be initiated (e.g., recruitment of other transcription factors). This hypothesis is further supported by our finding that CLOCK:BMAL1 does not directly generate a transcriptionally active enhancer. Indeed, both H3K27ac ChIP-Seq signal and eRNA transcription are minimal at CLOCK:BMAL1 enhancers targeting non-expressed genes, and are delayed at CLOCK:BMAL1 enhancers targeting out-of-phase transcriptional cyclers (Fig 3F and 3G). The observation that H3K27ac ChIP-Seq signal at CLOCK:BMAL1 enhancers correlates with CLOCK:BMAL1 transcription output rather than CLOCK:BMAL1 DNA binding phase/intensity seems inconsistent with the well-described interactions between core clock proteins and histone modifiers [3, 14–19, 26], and thus raises the question on whether or not CLOCK:BMAL1 DNA binding occurs with enzymatically activate histone modifiers. Interestingly, instances of enhancers bound by p300/CBP but lacking H3K27ac (and transcriptional activity) have been described at enhancers targeting developmental genes in human ES cells [70, 71]. Those enhancers, which are termed poised enhancers, share most of the properties of active enhancers, including similar levels of nucleosome depletion, p300, and chromatin remodelers binding. However, these poised enhancers are unable to drive gene expression in ES cells until they acquire H3K27ac signal during differentiation [71]. Here we found that the binding of CBP and p300 at non-expressed target genes is above background levels, and that the differences in CBP and p300 DNA binding between non-expressed and expressed target genes are smaller than those observed for H3K27ac and Pol II ChIP-Seq signal (Figs 3Fand 4F, S10 Fig). It is thus tempting to speculate that the concept of poised enhancers extends to the circadian field, with CLOCK:BMAL1 rhythmically priming the chromatin landscape of “circadian poised enhancers”. While those circadian poised enhancers would share properties of active enhancers (similar CLOCK:BMAL1 DNA binding, nucleosome eviction rhythm, etc.), they would be transcriptionally inactive and require the binding of other transcription-associated factors needed to trigger H3K27ac and rhythmic transcription.
Investigation of the transcription factors that are recruited at CLOCK:BMAL1 enhancers revealed a surprising difference between u-TFs and ts-TFs. In particular, ts-TFs are recruited at similar levels between expressed and non-expressed CLOCK:BMAL1 target genes, suggesting that they do not significantly contribute to the heterogeneity of CLOCK:BMAL1 transcriptional output. Because ts-TFs are known to establish tissue-specific enhancers and enable the binding of u-TFs in a tissue-specific manner [72–75], it is likely that ts-TFs contribute primarily to the binding of CLOCK:BMAL1 at tissue-specific enhancers and thus enable the generation of a tissue-specific circadian transcriptional program [8–10, 76]. Contrary to ts-TFs, u-TFs appear to bind at CLOCK:BMAL1 enhancers targeting specific transcriptional output categories, suggesting that their nature (i.e., activator or repressor, constitutively active or inducible), as well as mode of cooperation with CLOCK:BMAL1, likely contributes to the heterogeneity of CLOCK:BMAL1 target gene transcription. For example, the transcriptional repressors REV-ERBα and REV-ERBβ are enriched at CLOCK:BMAL1 enhancers targeting out-of-phase transcriptional cyclers, agreeing with the recently proposed model of facilitated repression whereby CLOCK:BMAL1 remodels its enhancer chromatin to facilitate the recruitment of REV-ERBs and delay the transcriptional output of some of its target genes [37]. Since rhythmically expressed genes tend to exhibit higher u-TF ChIP-Seq signal than arrhythmic and non-expressed genes (S10 Fig), and given the low expression of in-phase transcriptional cyclers in Bmal1-/- mice, we propose that a major function of CLOCK:BMAL1 is to facilitate the recruitment of both positive and negative transcription factors to drive the rhythmic transcription of clock-controlled genes (i.e., not just to facilitate the binding of the circadian repressors REV-ERBα/β). Although the mechanisms underlying of this cooperation between CLOCK:BMAL1 and other transcription factors are still unknown, nucleosome-mediated cooperation between transcription factors is not unprecedented [77–82], and several papers have shown that two non-interacting TFs can synergistically bind to DNA through a mechanism whereby the first TF leads to partial unwrapping of nucleosomal DNA, thus making the site of the second TF more accessible and thereby increasing DNA binding.
This cooperation between CLOCK:BMAL1 and other TFs may explain why a large fraction of CLOCK:BMAL1 target genes are not expressed: u-TF recruitment is not sufficient to activate CLOCK:BMAL1 enhancers and promote transcription. In support of this idea, CLOCK:BMAL1 enhancers targeting non-expressed target genes are enriched for the NF-κB transcription factor motif, which is known to mediate transcriptional response to immune and inflammatory responses [41]. Because the genome-wide characterization of circadian clock mechanisms has mostly been carried out in healthy mice raised in standard laboratory conditions, NF-κB is likely inactive, sequestered in the cytosol and its target genes are not expressed. CLOCK:BMAL1 may thus prime NF-κB DNA binding upon inflammation or immune response, thereby triggering a rhythmic response to acute infection. Interestingly, such a mechanism may explain, at least in part, why the immune host response oscillates based on the time-of-day bacterial infection [83–86]. We also found that CLOCK:BMAL1 enhancers at non-expressed target genes are enriched for the transcription factor CTCF (CCCTC-binding factor; Fig 4A). CTCF is known to promote long-range interactions between two or more genomic sequences, and thus bring sequences that are far apart in the linear genome into close proximity [44]. This may suggest that some CLOCK:BMAL1 DNA binding sites situated in non-expressed gene loci actually target other clock-controlled genes located hundreds of kilobases apart through long-range interactions, as recently described for one CLOCK:BMAL1 DNA binding site in the mouse liver [87]. Although it is impossible to assess the prevalence of CLOCK:BMAL1 binding sites mediating long-range chromatin interactions without the appropriate experiments, we found a few examples suggesting that this is a likely possibility (S15 Fig).
Transcription regulation in higher eukaryotes relies on the activity of multiple enhancers [88, 89]. It is thus likely that CLOCK:BMAL1 target gene expression results from a complex integration between CLOCK:BMAL1 enhancers and other enhancers. Our results indicate that enhancers targeting the same gene typically share the same transcriptional activity profiles (H3K27ac signal, eRNA levels, and Pol II ChIP-Seq signal; S8 Fig). Based on these observations, we cannot exclude that other enhancers targeting arrhythmically expressed CLOCK:BMAL1 target genes outcompete CLOCK:BMAL1 enhancers, to produce constitutive expression. Further experiments aimed at revealing hierarchical influences of enhancers on the regulation of gene expression at the genome-wide level will be required to directly test this hypothesis.
It was recently proposed that altering the environmental conditions can reprogram circadian transcriptional programs (e.g., high-fat diet and antibiotics treatment in the liver, LPS treatment in the lung [57, 90–92]). Our model that CLOCK:BMAL1 regulates the expression of clock controlled genes by facilitating the binding of other TFs represents a mechanistic framework for explaining how environmental signals can mediate this transcriptional reprogramming. Indeed, activation of new signaling pathways by environmental changes is likely to modulate multiple transcriptional programs, thereby altering how CLOCK:BMAL1 cooperate with those programs to drive rhythmic gene expression. Importantly, this mechanism may also explain, at least in part, why the number and nature of rhythmically expressed genes vary between datasets and laboratories [93–95]. Indeed, differences in diet, light environment and housing may all lead to changes in u-TF transcriptional activity, which may in turn affect clock-controlled gene expression.
In conclusion, our data indicate that the mechanisms by which CLOCK:BMAL1 regulates the transcription of core clock genes do not apply to clock-controlled genes, and suggest that the primary function of CLOCK:BMAL1 is to regulate the chromatin landscape at its enhancers to facilitate the binding of other transcription factors. Our results therefore highlight the emerging role of other transcription factors in regulating the ~15% of genes that are rhythmically expressed in a given mammalian tissue, and suggests that clock-controlled gene expression relies more on the interplay between the circadian clock and other signaling pathways. Given that the majority of CLOCK:BMAL1 target genes are either arrhythmically or not expressed under standard conditions, our data also suggest that these non-oscillating genes may become rhythmically expressed under other environmental and/or pathological conditions, and thus expand the total number of genes under circadian control to more than 50% in mammals [10]. Finally, because the clockwork mechanisms are highly conserved between eukaryotes (e.g., heterogeneous transcriptional output, poor reproducibility between datasets characterizing circadian gene expression, regulation of chromatin landscape by core clock components), it is likely that the mechanisms we uncovered largely apply to all eukaryotic circadian clocks.
Unless notified below, publically available datasets used in this paper were downloaded from the NCBI or EBI websites in either sra or fastq formats (see Table 1 for accession numbers). Files in sra format were converted to fastq files using the sratoolkit (version 2.3.5–2). Fastq files were mapped to the mouse genome (version mm10) using bowtie2 [96] or tophat2 [97]. For all datasets, we only considered reads that mapped uniquely to the mouse genome (i.e., one unique genomic location). Datasets were further filtered to remove duplicated reads using samtools (rmdup function) or a custom-made script. Additional information is provided for each dataset as Supplementary Materials and Methods.
Genomic locations of CLOCK and BMAL1 DNA binding sites in the mouse liver provided in the original paper (supplementary S2 Table) [3] were used to generate our list of high confidence CLOCK:BMAL1 DNA binding sites. Genomic locations were converted to the mm10 version of the mouse genome using UCSC genome browser liftOver tools, and processed as indicated in S1 Fig to generate our list of high confidence CLOCK:BMAL1 DNA binding sites. Overlap between CLOCK and BMAL1 ChIP-Seq peaks was determined using bedtools (intersectBed) and coordinates from BMAL1 ChIP-Seq datatsets were further kept to generate a list of 3217 CLOCK:BMAL1 peaks. We also used the original data provided by the authors in their S2 Table to assign CLOCK:BMAL1 peaks to their putative target genes (original analysis performed using HOMER tools). In particular, we defined a CLOCK:BMAL1 target gene as a gene with at least one CLOCK:BMAL1 peak located between -10kb of the transcription start site and +1kb from the transcription termination site. Using this criteria, 2458 CLOCK:BMAL1 peaks were assigned to a target gene, and the remaining 759 peaks were assigned as an intergenic CLOCK:BMAL1 DNA binding site.
The 2458 CLOCK:BMAL1 peaks assigned to a target gene were then parsed based on the transcription profile of their target genes using the Nascent-Seq datasets from Menet et al., 2012 [28]. We directly used the original Nascent-Seq expression values and the assessment of their rhythmic expression from the original paper without performing new analysis. Details on how genes were determined to be rhythmically transcribed are provided in Supplementary Materials and Methods. Using these data, CLOCK:BMAL1 peaks were parsed into 4 different categories of transcriptional output (see also S1 Fig):
The list of the 3217 CLOCK:BMAL1 peaks parsed into the different transcriptional output categories is provided in S1 Table. The phase of rhythmic CLOCK:BMAL1 DNA binding, ChIP-Seq signal, and genomic location of CLOCK:BMAL1 DNA binding sites were retrieved from the Koike et al., 2012 original paper supplementary S2 Table [3] and processed to generate the analysis presented in Fig 1.
ChIP-Seq, MNase-Seq and GRO-Seq signal was retrieved from bam files containing uniquely mapped reads (and duplicated reads removed) at CLOCK:BMAL1 enhancers using custom-made scripts [20]. Specifically, signal was retrieved at:
and normalized to the sequencing depth. Differences in the window size were calculated based on the width of the ChIP-Seq signal at CLOCK:BMAL1 DNA binding sites (e.g., H3K27ac ChIP-Seq signal is significantly wider than any transcription factor ChIP-Seq signal). Because we aimed at assessing the role of CLOCK:BMAL1 in removing a nucleosome at its DNA binding site, we chose a narrower window size of 150bp (see Fig 3A–3D). All analyses were performed at individual CLOCK:BMAL1 ChIP-Seq peaks, and this even for peaks targeting the same gene. Data presented in S4 Fig examined the role of multiple peaks targeting the same gene on BMAL1 and CLOCK ChIP-Seq signals. For all datasets, ChIP-Seq signal is displayed as the number of reads/bp per 100,000,000 reads.
Enhancers lying into CLOCK:BMAL1 target gene loci (-10kb from the transcription start site to +1kb from the transcription termination site) were identified using a public mouse liver DNAse-Seq dataset (see above) [98] and bedtools (intersectBed function). Enhancers were then parsed based on the presence or not of a CLOCK:BMAL1 ChIP-Seq peak (3155 out of the 3217 CLOCK:BMAL1 ChIP-Seq peaks are located into a DNaseI hypersensitive site). Because a majority of the 104,556 DHS peaks only displayed low levels of ChIP-Seq (transcription factors, Pol II, H3K27ac) and GRO-Seq signals [as shown in the ENCODE project, 99, 100], we filtered the number of DHS lying into a CLOCK:BMAL1 target gene by only considering those being into the top 40,000 DHS list (based on DNase-Seq signal), obtaining the following number of DHS peaks:
H3K27ac and Pol II ChIP-Seq signals, as well as GRO-Seq (eRNA) signal, were retrieved at those DHS sites (as well as those overlapping with a CLOCK:BMAL1 peak) using the DHS peak coordinate and normalized to 100,000,000 reads. Signal was then normalized to the coordinate length (in bp) to obtain the signal displayed as reads/bp per 100,000,000 reads. The coordinates used were, for the same reason as above for CLOCK:BMAL1 DNA binding sites:
Because our analysis revealed the existence of small but significant overall variations of H3K27ac and Pol II ChIP-Seq signal between time points (see S8 Fig), we further normalized the datasets by performing either a mean normalization (H3K27ac) or a ranking analysis (Pol II). For H3K27ac ChIP-Seq datasets [3, 38], averaged H3K27ac signal was calculated at the top 40,000 DHS peaks (the top 40,000 DHS peaks concentrate the majority of TFs DNA binding sites; peak center ± 1 kb; total of 104,556 total DHS peaks; dataset from Ling et al., 2010 [98]) for each time point. This averaged signal was then used to normalize the raw H3K27ac ChIP-Seq signal, by calculating for each time point the ratio between H3K27ac signal for each peak and this averaged signal (see S8 Fig). Pol II ChIP-Seq dataset [22] were normalized by performing a ranking normalization (method similar to a quantile normalization). To this end, Pol II ChIP-Seq signal was calculated at all 104,556 DHS peaks (peaks mapped in Ling et al., 2010 paper [98]), and sorted based on the ChIP-Seq values. The raw values for each DHS peak were then normalized using the sorted averaged ChIP-Seq signal at each of the 104,556 ranks for all time points.
Motif analysis was performed at CLOCK:BMAL1 enhancers (original peak coordinates) for each of the transcriptional categories using the findsMotifGenome.pl script from the HOMER suite. Parameters were as the following: -size given–len8. The resulting table was sorted by the q-value and a q-value less than 0.05 was considered significant. Percent enrichment (percent of target sequences with motifs / percent of background sequences with motif) was then calculated for motifs found to be significant in at least one of the CLOCK:BMAL1 transcriptional output category. Results of the motif analysis are provided as S2 Table.
To determine the variance of each TF DNA binding (CLOCK, BMAL1, ts-TFs and u-TFs) between the four CLOCK:BMAL1 transcriptional output categories, we computed a TF DNA binding “variability index” based on the analysis performed in S10 Fig. The variability index was calculated by summing up the standard deviation of the ChIP-Seq signal between the 4 transcriptional output groups, which was calculated for each decile (0.1 to 0.9) and normalized to the averaged signal for each decile (the standard deviation is higher for upper deciles because ChIP-Seq signals are higher). This index reflects differential DNA binding strength between groups, as similar binding between the 4 groups results in small standard deviation values for each decile, and thus a small variability index. Conversely, differences in DNA binding signal between groups result in larger standard deviation values and thus a larger variability index.
To determine if the results described in this paper are specific to CLOCK:BMAL1, we also performed an analysis on genes not targeted by CLOCK:BMAL1, but exhibiting similar profiles of expression to the 4 CLOCK:BMAL1 transcriptional output categories (Rinφ, Ro/φ, AR and NE). To this end, 125 genes were randomly selected for each of the 4 groups, using criteria similar to those used to define CLOCK:BMAL1 transcriptional output (see above). Levels of expression for each group were not significantly different to those of CLOCK:BMAL1 target genes (Kruskal-Wallis test). Nucleosome signal, H3K27ac ChIP-Seq signal, Pol II ChIP-Seq signal, eRNA expression, tissue specific and ubiquitous transcription factor ChIP-Seq signal were all calculated as described above for CLOCK:BMAL1 target genes. Statistical analysis was also performed similarly to CLOCK:BMAL1 transcriptional output.
Statistical analysis was done using JMP, Version Pro 12.0.1. SAS Institute Inc., Cary, NC, 1989–2007. Differences in sequencing signal, represented in the boxplot graphs, were analyzed for statistical enrichment using the nonparametric Kruskal-Wallis test. Rhythmic analysis of nucleosome signal and ChIP-Seq signal was performed using a Fourier analysis (Fig 3A–3D) (see Supplementary Materials and Methods for details). Differences in the amplitude of nucleosome signal rhythm (Fig 3E) were analyzed using a 2-way ANOVA. Differences in CLOCK:BMAL1 ChIP-Seq peaks genomic location were analyzed using a chi-square test (Fig 1G), and differences in the number of CLOCK:BMAL1 peaks per target genes (S5A Fig) were analyzed by a Fisher’s exact test. Differences were considered significant when p < 0.05.
|
10.1371/journal.pbio.1002013 | Epidermal Growth Factor Signalling Controls Myosin II Planar Polarity to Orchestrate Convergent Extension Movements during Drosophila Tubulogenesis | Most epithelial tubes arise as small buds and elongate by regulated morphogenetic processes including oriented cell division, cell rearrangements, and changes in cell shape. Through live analysis of Drosophila renal tubule morphogenesis we show that tissue elongation results from polarised cell intercalations around the tubule circumference, producing convergent-extension tissue movements. Using genetic techniques, we demonstrate that the vector of cell movement is regulated by localised epidermal growth factor (EGF) signalling from the distally placed tip cell lineage, which sets up a distal-to-proximal gradient of pathway activation to planar polarise cells, without the involvement for PCP gene activity. Time-lapse imaging at subcellular resolution shows that the acquisition of planar polarity leads to asymmetric pulsatile Myosin II accumulation in the basal, proximal cortex of tubule cells, resulting in repeated, transient shortening of their circumferential length. This repeated bias in the polarity of cell contraction allows cells to move relative to each other, leading to a reduction in cell number around the lumen and an increase in tubule length. Physiological analysis demonstrates that animals whose tubules fail to elongate exhibit abnormal excretory function, defective osmoregulation, and lethality.
| Many of the tissues in our bodies are built up around complex arrays of elongated cellular tubes, which permit the entry, exit, and transport of essential molecules such as oxygen, glucose, and water. These tubes often arise as short buds, which elongate dramatically as the organ grows. We sought to understand the mechanisms that govern such transformations of shape using the fly renal tubule as a model. We find that elongation of this tissue is predominantly driven by cell rearrangement. Cells move around the circumference of the tubule, intercalating with each other so that the cell number around the lumen reduces, while increasing along the length of the tube. Our next question was how cells sense the direction in which they should move. We show that cells orient their position in the tissue by reading a signal sent out by a specific pair of cells at the tip of each tube. Cells use this directional information to make polarised movements through the asymmetric activity of the cell's contractile machinery. We find that the activity of myosin—the motor protein that regulates contraction—is pulsatile and polarised within the cell. This activity shortens the cells' circumferential lengths, so that cells move past each other around the tube circumference, thereby intercalating and producing tube elongation. We go on to show that excretory physiology is severely impaired when elongation fails, underlining the importance of sculpting organs with appropriate dimensions.
| Our tissues and organs are built up around arrays of tubes that allow the exchange of nutrients, ions, and gases vital for bodily function. These tubules have precise architectures tailored to their physiological activities. It is important that appropriate tubule dimensions are established during development and maintained throughout life and where this fails, as for example in human polycystic kidney diseases, in which nephron diameters are grossly enlarged [1], physiological function is severely compromised, often leading to organ failure.
Many tissues are sculpted during development by convergent extension (CE) movements. This process describes the concomitant narrowing of a tissue in one axis while it elongates along a perpendicular axis (Figure 1A) [2]–[4]. CE is brought about by changes in cell-neighbourhood relationships produced by cell intercalation. These changes can be driven by a variety of force-generating processes, such as lamellipodial protrusion, that allow cells to crawl over one another [5] or by cell-junction remodelling [5]–[7]. In both cases cell intercalation is highly organised and is polarised in the plane of the tissue [2],[8].
The insect renal or Malpighian tubules (MpTs) eliminate metabolic and foreign toxins and maintain the animal's ionic, acid-base, and water balance [9],[10]. They are long, narrow, single cell-layered epithelial tubes with a distinct distal-to-proximal (D-P) axis in which the distal regions are secretory in function and proximal regions have reabsorptive roles [11]. In Drosophila the tubules evert from the embryonic hindgut as short buds. During mid-embryogenesis they undergo a dramatic transformation in a period of just a few hours—increasing in length approximately 4-fold whilst narrowing substantially around their circumference. Tubule extension occurs in the absence of cell division and is accompanied by substantial rearrangement of cells within the plane of the epithelium [12]. This morphological transformation appears to be a dramatic example of CE and, because it occurs in the absence of cell proliferation that might complicate analysis, it is an attractive model to study the process of CE and its regulation.
How CE is controlled at the tissue level is still poorly understood in terms of the mechanisms and signals that orchestrate local cell behaviours to bring about orderly morphogenesis in the tissue as a whole. During Drosophila germband extension the segmentation genes that pattern the anterior-posterior axis are important in establishing planar polarity [13]. However, it is not known whether the influence of the segmentation genes is direct, nor have the mechanisms by which these genes control cell intercalation been established [4],[14],[15]. In other tissues the core planar cell polarity (PCP) genes regulate both oriented cell divisions [16]–[19] and polarised cell movements that underlie tissue extension [20]–[22], but details of the mechanisms involved remain elusive [23],[24].
Here we address the fundamental question of how cell intercalation is controlled at the tissue level, using the developing fly renal system as a model. We analyse cell movements during tubule extension and show that elongation results from circumferential cell intercalation associated with pulsatile and planar polarised accumulation of the motor protein, Myosin II in the basal cortex of cells. We consider the spatial cues that direct these oriented cell intercalation events as the tubule lengthens. We show that a polarised signal within the tubule organises cell rearrangement during CE. Using a combination of genetic manipulation, laser ablation, and live imaging we provide evidence that the epidermal growth factor (EGF) pathway ligand Spitz provides this cue. Spitz is expressed in the distal tubule tip, activating graded EGF signalling along the tubule, which is required for coordinated cell intercalation. EGF signalling acts to establish an axis of planar polarity in tubule cells at the onset of cell intercalation, which is independent of the activity of planar polarity genes. Perturbation of EGF signalling results in disorganised Myosin II dynamics, failure of cell intercalation, and defective elongation, leading to impaired tubule function and the failure of fluid homeostasis.
By mid embryogenesis (stage 13) the tubules are short and stubby in shape, measuring approximately 80 µm in length (Figure 1A, 1C, and 1E) with between 10 and 12 cells surrounding the lumen (Figure 1A and 1D). Over approximately 5 hours, they undergo a 4-fold elongation to approximately 320 µm in length (Figure 1A, 1C, and 1E; Movie S1) whilst the number of cells surrounding the lumen reduces progressively to just 2 cells (Figure 1D). Imaging this morphogenetic transformation in real time and tracking individual cells reveals that rings of cells around the tubule lumen (as shown in Figure 1G′ and 1G″) intercalate circumferentially, becoming more spread out in the orthogonal, distal-to-proximal axis (Figures 1F and 1G; Movies S2 and S3). Cell intercalations can be followed (Figure 1H and 1H″; Movies S2 and S4), and we were able to confirm previous observations that tubule extension occurs sequentially with CE occurring in the distal half of the tubule earlier than in the proximal half [25]. Focussing on the distal region of the anterior tubules shown in Figure 1B and 1F we found that individual intercalation events took between 24 and 49 minutes, with an average of 42.2 min (±6.1 min, n = 4 intercalations) with cells moving at an average of 1.14 µm min−1 (±0.02 µm min−1, n = 9 cells). Our observations confirm the hypothesis that tubule elongation results primarily from oriented CE movements, in which cell intercalation around the circumferential axis produces orthogonal extension in the D-P axis of the tubule (Figure 1I).
Previous work [26],[27] has shown that tubules cultured in vitro from stage 11 are able to elongate outside their normal environment, suggesting that tubule CE movements are regulated by mechanisms intrinsic to the tubule. Our previous work has also indicated that the distal-most cells in the tubule, the tip cell (TC) and its sibling (SC) (Figure 2A), are important for tubule elongation [12],[26],[28].
Both terminal cells secrete the EGF signal Spitz (Spi) during stage 12 to promote tubule cell proliferation [29]–[31]. However EGF signalling from the TC lineage persists beyond stage 12 throughout the period of tubule elongation, as revealed by the expression of the protease Rhomboid that cleaves Spi to produce its secreted and active form (sSpi) (Figure 2B). Staining for a read-out of signalling, diphosphosphorylated extracellular signal-regulated kinase (dpERK) [32], shows that the EGF pathway is activated in tubule cells through to stage 16 and it appears that activation is stronger in distal (close to the TC) compared to proximal tubule regions (Figure 2C and 2C′). Quantitative analysis of dpERK levels confirms graded activation in response to signalling, highest closest to the TC and declining towards the point of tightest tubule curvature (the kink) approximately half way along its D-P length (Figure 2E and 2F). These observations were confirmed using a second assay for pathway activation. Capicua is localised to the nuclei of quiescent cells but is translocated to the cytoplasm upon activation, where it is processed for degradation [33],[34]. Confirming our findings for di-phosphorylated ERK, expression of a tagged Capicua::Venus construct is diminished distally but remains high in the proximal tubule (Figure 2D and 2D′).
We ablated both the TC and SC either genetically or physically using a laser. In embryos mutant for the proneural genes [35] or components of the JAK/STAT pathway (BD, unpublished data), the TC lineage is not specified, tubules lack TCs and SCs and fail to undergo elongation (Figure 2G and 2G′). However the late phase of tubule cell proliferation also fails in the absence of the TC lineage [26],[35]. To test whether the reduced cell number contributes to elongation defects we laser ablated the TC and SC in late stage 12 embryos when tubule cell proliferation is complete. We used ctB>UAS-CD8-GFP to mark all tubule cells and centred our ablation on the distal three or four cells to ensure TC/SC removal. By stage 16, tubule elongation had failed completely in tubules lacking a TC and SC (Figure 2H and 2H′), whereas the contralateral (non-ablated) tubules underwent normal elongation (Figure 2I and 2I′). Together these experiments show that the TC lineage is required for tubule elongation.
To test the role of EGF signalling in tubule extension we abrogated signalling after the completion of EGF-dependent tubule cell division using the temperature sensitive allele of the epidermal growth factor receptor (EGFRf7) [29],[36]. In the majority of embryos shifted to the restrictive temperature at mid-stage 13, tubule elongation is severely disrupted (87%, n = 15 cf controls raised at the permissive temperature 5%, n = 18). The tubules remain short and the number of cells encircling the lumen fails to reduce as in sibling control embryos (Figure 3A and 3B). Similar defects in tubule elongation occur when EGF signalling is disrupted by expressing a dominant negative receptor [37] using the Gal4 driver ctB, which represses signalling only after all tubule cell divisions have ceased (Figure 3C, 3D, and 3G; Movies S5 and S6) [31]. The analysis of tracked cells from movies of tubules expressing the dominant negative receptor reveals that very few cells complete intercalation in contrast to the wild type (Figure 1H; Movies S2 and S4). Cell movement is strongly reduced; the average speed of movement is 0.5±0.07 µm min−1 n = 42 distal cells (cf wild type intercalating cells 1.14±0.02 µm min−1).
Surprisingly, experiments in which EGF pathway signalling is hyperactivated in all tubule cells, either by expressing the constitutively activated receptor λTop/EGFRact [38] or active ligand sSpi [39], also produce striking defects in tubule extension, strongly reminiscent of the loss of function phenotypes; 64% (n = 28) of tubules expressing the activated receptor fail to elongate, remaining short and thick (Figure 3E–3H; Movies S7 and S8). Tracking cells in activated tubules shows that, as in the loss of EGF pathway function, the number of cell intercalations is reduced, although those in the most distal region close to the source of ligand still occur at near the wild type rate with intercalations taking 37±4.3 min, n = 10 distal cells (Figure 3F and 3F′; Movies S7 and S8). However overall cell movement is much reduced (0.6±0.06 µm min−1, n = 38 cells cf wild type 1.14±0.02 µm min−1). In contrast, enhanced expression of sSpi from the TC lineage, which is likely to induce higher than normal levels of EGF pathway activity whilst retaining its spatial asymmetry, does not lead to defective tubule elongation (Figure 3I).
Together these experiments show that the TC lineage is the source of the EGF ligand, Spitz. They also reveal that asymmetric, signalling from a localised source is crucial for tubule elongation, as either loss of receptor activation or hyperactivation along the whole tubule length disrupts CE movements.
These data suggest the idea that the Spitz signal establishes an axis of polarity along the tubule length, about which directed cell rearrangements, required for orderly cell intercalation, occur. We therefore asked if tubule cells exhibit planar polarised features, which are dependent on EGF signalling. As tubule cells lack visible polarity landmarks we relied on a technique that has revealed polarity in other tissues: the expression of the membrane-associated Slam protein. During germ band extension Slam assumes a bipolar distribution on the vertical cell membranes orthogonal to the A-P axis that presages cell intercalation [6],[13]. Expression of a tagged form of Slam (Slam-HA [40]) in tubule cells (Figure 4A–4D) reveals that it becomes planar polarised at the onset of CE. In wild type tubules prior to CE (before stage 13) Slam accumulates in a single central clump in the basal cortex of each cell (Figure 4B). At stage 13 (Figure 4C), Slam relocates towards the basal proximal cortex. The proximal localisation is maintained throughout the period of intercalation during stages 14 and 15 (Figure 4A and 4D) where it appears to spread down the lateral cell membrane. To show conclusively that Slam localises preferentially to the proximal cortex we induced Slam expression in single cells (Figure 4A) where it accumulates on the proximal side and is virtually absent from the distal cortex (Figure 4A and 4A″).
These data provide the first evidence that tubule cells are polarised within the plane of the epithelium and reveal that aspects of planar polarity are established before the initiation of tubule extension. Slam is not normally expressed in the tubules [41]. Thus while Slam localisation reveals latent polarity in the tissue, it does not contribute to the mechanism by which the tissue is normally polarised.
Using ectopic Slam as a marker for tissue planar polarity, we assessed tissue polarity in tubules in which EGF signalling was perturbed. Before stage 13, Slam-HA localisation is unaltered in conditions of either loss- or gain-of-function EGF signalling; Slam is found in a central cluster in the basal cortex of tubule cells (Figure 4G″). However, Slam fails to redistribute to the proximal cortex and becomes severely disorganised as development proceeds. At the time when Slam would normally relocate proximally, it either fails to relocate or spreads around the entire cell (Figure 4E–4G). These results show that in the absence of EGF signalling, or under conditions of global pathway activation, tubule cells are unable to polarise within the plane of the tissue, supporting the hypothesis that EGF signalling is the source of vectorial information that establishes and/or maintains planar polarity in the tubule.
Other studies have shown that EGF signalling is required for the maintenance of apicobasal cell polarity and also for cell survival [42]. Defects in either could account for defective tubule development and so we analysed these parameters after perturbing EGF signalling in tubules. Neither driving a dominant negative EGF receptor nor activating the pathway in all tubule cells alters their apicobasal polarity, as revealed by the distribution of the apical marker Bazooka and the lateral membrane protein FasII (Figure S1A–S1C). There is no cell death in control tubules during stage 15 detected by cleaved Caspase 3 staining (Figure S1D, see S1G and S1H for positive control), and we find no increase in cell death in the tubules either when the EGF pathway is abrogated or hyperactivated (Figure S1E and S1F). These data indicate that the primary response of tubule cells to asymmetric and graded EGF signalling is the acquisition of PCP.
The so-called PCP genes play key roles in regulating tissue polarity in diverse organisms [43],[44]. Furthermore, PCP signalling has been implicated in some [21],[22],[24], but not all [13], tissues undergoing CE movements. We therefore asked whether the PCP genes contribute to tubule planar polarity and elongation. There is evidence for two independently acting PCP systems, the Dachsous (Ds) and Starry night (Stan) systems [45]. For this reason we examined mutations in genes for each system independently (removing maternal and zygotic contributions) and double mutant combinations that remove the function of both systems together (see Table S1 for the lines examined). No defects in tubule extension were found for any of the mutants or double mutants we tested (Figure 4H and 4H′; Table S1), indicating that the PCP genes are not required for CE movements in the tubules. Furthermore, we found that Slam localises normally in the absence of PCP gene function (Figure 4I and 4J), revealing that the PCP genes we have tested are also dispensable for planar polarity in the developing tubule.
Previous studies have shown that the normal activity of non-muscle Myosin II is required for tubule elongation [46],[47]. However the phenotypes reported were quite weak perhaps because both Zipper (Zip, Myosin heavy chain) and spaghetti squash (Sqh, myosin light chain) are supplied maternally (http://insitu.fruitfly.org/cgibin/ex/report.pl?ftype=1&ftext=CG15792). We therefore assessed the effects of perturbing Myosin II activity in tubule cells by driving the expression either of a dominant negative Zipper (YFP-ZipDN) (Figure 5B) [48] or of constitutively active Sqh (SqhE20E21), which is known to result in an increase in Myosin II activity (Figure 5C) [49],[50]. Compared to wild type tubules, those with altered Myosin II activity fail to extend normally but remain short with more than two cells around the circumference of the lumen (Figure 5A–5C and 5G, Movie S13; YFP-ZipDN 95% and SqhE20E21 75% of embryos showed elongation defects by stage 15 [n = 20 in each case]). These data show that normal levels of Myosin II activity are required for cell intercalation and tubule elongation.
Expression of a tagged Myosin II light chain, Sqh::GFP, in a sqh mutant background allowed us to analyse the activity of myosin in tubule cells during elongation. This construct rescues the embryonic sqh mutant phenotype [6],[49] indicating that endogenous levels of expression are maintained. Movies of the basal-most side of cells during tubule extension reveal regular pulses of Myosin II activity, in which cytoplasmic spots of myosin move to the proximal cortex of cells (Figure 5D; Movies S9 and S10). Analysis of 63 pulses in 37 tubule cells from eight different embryos indicates that the duration of pulses ranges from 0.9 to 4.1 min (average 1.98±0.08 min) with 0.13 to 3.6 min (average 1.78±0.2 min) between pulses (interpulse). Of 61 pulses analysed, 49 showed basal, proximal enrichment and only 12 showed proximal-to-medial or proximal-to-distal movement of myosin. It is striking that although spots of myosin fluorescence can be seen at the orthogonal, circumferential cortices, enriched crescents are almost never seen in these regions.
In order to follow the localisation of Myosin II more precisely we double labelled tubule cells with Sqh::mCherry [51] and GAP43::GFP to label cell membranes. Movies show dynamics identical to those using Sqh::GFP (Figure 5E; Movie S11). When driving a dominant negative EGFR construct, where cells fail to intercalate and tubule extension is lost (Figure 3C, 3D, and 3G), the pulses of active myosin completely fail (Figure 5F; Movie S12).
Actomyosin dynamics are associated in other systems with alterations in cell shape, cell movement, and rearrangement [6],[51]–[53]. We analysed fluctuations in the basal shape of tubule cells by measuring their area, D-P, and circumferential lengths. This analysis reveals that cell shape is in a constant state of flux (Figures 5G, S2, and S3A-S3C). We compared the dynamics of fluctuations during a Myosin II pulse with interpulse periods in individual cells but found no obvious correlation with cell shape (Figure S2). However, averaging measurements from multiple cells (n = 10) revealed that while pulses caused no significant change in the D-P axis, there is a small but significant decrease in circumferential length associated with Myosin II pulses (Figure 5H).
In contrast, driving the expression of YFP-ZipDN in tubules results in a strong reduction in cell dynamics with much reduced fluctuations in cell area (Figures 5G and S3A″) and little change in either the circumferential or D-P axial length of cells (cf shaded areas in Figure S3B, S3C, with S3B″ and S3C″). The dynamic fluctuations in cell shape are also dramatically reduced when EGF signalling is compromised; in tubules expressing EGFRDN the basal area of cells (Figures 5G and S3A′) and their axial lengths (Figure S2B′ and S2C′) scarcely alter compared with the fluctuations seen in wild type tubule cells (Figure S2A–S2C). Tracking cell trajectory over time (Figure 5I) shows that control cells move in a circumferential direction (± approximately 50°). In tubules expressing either EGFRDN or YFP-ZipDN the small cell movements that occur fail to show any bias towards the circumferential axis (Figures 5I and S3D).
Together our analysis of cell behavour in control and mutant tubules in which EGF signalling is deranged or where the normal activity of Myosin II is compromised indicates that proximally directed pulses of cortical Myosin II are essential for cell intercalation but occur only in cells that have been planar polarised by asymmetric EGF signalling. In control tubule cells these pulses produce a transient, small but significant reduction in the circumferential length of cells enabling them to move in this axis (Figure 5J), resulting in the cell rearrangements that produce tubule elongation. Without polarised EGF signalling the pulses fail, cells dynamics are dampened and intercalation is either much reduced (pathway activation) or fails (loss of the pathway activity) so that tubule elongation is compromised.
Our analysis of the response to EGF signalling in tubules has established graded activity but only in the distal half of the tubuels (Figure 2E and 2F). We therefore wondered whether cells in the proximal part of tubules become planar polarised. Expression of Slam-HA in tubules from stage 13–16 embryos reveals that the protein in the proximal tubule, in contrast to the distal half, is not asymmetrically distributed at any stage during elongation (stage 15 shown in Figure S4A and S4A″). As we have correlated the acquisition of planar polarity in cells with the development of dynamic, asymmetric subcellular activity of Myosin II, we wondered whether cells in the proximal region exhibit similar cytoskeletal dynamics. Imaging tubules expressing Sqh::mCherry (Figure S4B and S4B″; Movies S14 and 15) reveals that in contrast to distal regions, where repeated, proximally localised crescents of Myosin form, there is no apparent asymmetric Myosin activty in the proximal half of the tubules.
These data suggest that the mechanism by which cells in the proximal half of the tubules intercalate differs radically from the distal half and might not depend on polarised cytoskeletal activity, resulting in circumferential cell movements.
MpTs are the major organ for excretion, ionic balance, and osmoregulation in the majority of insects [9],[10]. Toxins are cleared from the haemolymph by active transport and primary urine is secreted into the tubule lumen by two cell types in the distally placed transitional and main segments. Homeostasis is accomplished by modification of primary urine as it passes down more proximal regions of the tubule before emptying into the hindgut via the ureters. We asked whether the final tubule shape is important for its function. Embryonic tubules persist through larval life and metamorphosis and so are retained in the adult. As there are some escapers when ctBGal4 is used to drive the activated EGF receptor we examined tubules from adult flies of this genotype. Their tubules are abnormal in shape compared to controls, being shorter and wider with conspicuous bulges (Figure 6A). This shows that embryonic tubule defects are not rectified during later developmental stages. In the distal two-thirds of wild type tubules stellate cells, responsible for anion and water movement in the secretion of primary urine, are regularly interspersed with the cation-transporting principal cells (PCs) (Figure 6B). Stellate cells are present in the abnormally shaped ctB>UAS-EGFRact tubules but their regular spacing is severely disrupted (Figure 6C).
Defects in tubule shape and in the organisation of specialised secretory cell types could well compromise renal physiology. To test this possibility, we compared tubule secretion in control and ctB>UAS-EGFRact tubules using an established in vitro tubule secretion assay [54],[55]. In control tubules the unstimulated, basal rate of primary urine secretion was 0.59 nl min−1 (n = 10). In contrast 4/11 ctB>UAS-EGFRact tubules did not secrete at all. The average basal rate of secretion for the remaining ctB>UAS-EGFRact tubules was 0.2 nl min−1 (n = 7). After stimulation with the diuretic activators cAMP and Leukokinin (LK) control tubules increased their secretory rate to 1.39 nl min−1 while ctB>UAS-EGFRact tubules failed to show any increase in secretory rate (Figure 6D). These data clearly demonstrate that secretory rate, a direct measure of tubule function, is either abolished or significantly reduced when tubule morphogenesis is disrupted.
The impact of defective tubule elongation on the physiology of adults can be seen within 24 hours of eclosion. Compared with control animals, ctB>UAS-EGFRact adults have grossly distended abdomens and mouthparts (Figure 6E), indicative of fluid retention through defective osmoregulation. We confirmed that distention was due to fluid retention (and not gas) firstly by pricking submerged flies, which led to abdominal deflation without gas bubbles. Secondly, we compared wet weight versus dry weight in experimental and control flies. ctB>UAS-EGFRact adults are over twice as heavy as control flies when measured wet, while dry weight measurements are not significantly different (Figure 6E; data for females, equivalent results were obtained for males). These data indicate that osmoregulation is severely compromised in ctB>UAS-EGFRact adults. Together our data illustrate the critical importance of tubule shape both for effective physiological function of the organ system and homeostasis in the whole animal.
In many situations tissue morphogenesis results from orderly cell rearrangements, which require the integration of positional information and oriented cell intercalation. Our data show that in fly renal tubules axial information is provided by an asymmetric EGF signal from a localised source, which acts to polarise cells in the distal half of the tubule just as elongation is about to start. The acquisition of D-P PCP leads to asymmetric, proximally directed pulses of myosin, which in turn result in repeated small contractions of the cell in the circumferential axis. Over time this results in the intercalation of cells around the tubule circumference to produce tubule elongation. While previous reports in other systems have focussed on parts of this sequence of events [6],[7],[13],[21],[56],[57], we have identified the source of polarisation, established the axis of planar polarity at the cellular level, and shown that it is required for the asymmetric behaviour of Myosin II motors that ensure oriented cell rearrangements (Figure 7).
Polarity in the plane of an epithelium is frequently conferred by the activity of PCP genes [43],[44],[58], and in many cases CE movements depend on the expression of these genes [24]. However the patterning of cell movements during tissue morphogenesis or collective migration have also been shown to require the activity of other pathways. In Drosophila the extension of the germ band depends on the expression of the early patterning pair rule genes [8],[13], Drosophila hindgut elongation depends on JAK-STAT pathway activity [56], and border cell migration in the egg chamber is polarised by gradients of receptor tyrosine kinase (RTK) signalling [59],[60]. The sensitivity of border cells to differences in ligand levels across a single cell diameter is enhanced by spatially regulated receptor endocytosis and processing. This depends on Cbl (a RTK-associated E3 ubiqitin ligase) and Sprint (a pathway activated Rab5GEF), which together act to down-regulate RTK receptors asymmetrically, leading to enhanced levels of pathway activation at the leading (higher ligand) cell face [61]. We find that both Cbl and Sprint are required for normal CE movements during renal tubule elongation [62],[63], showing that the enhancement of polarised RTK activation also occurs in this tissue.
The gradient of pathway activation is clear in the distal half of the tubules during elongation, as revealed by dpERK staining or Capicua::Venus expression and quantification suggests that the graded response extends over a considerable distance, approximately 60 µm (ten cell diameters). Such a comparatively long-range effect could result from the secretion of ligand from a localised source, particularly if diffusion of ligand into the haemolymph were restricted either by the extracellular matrix, known to ensheath the tubules [64], or by secretion of ligand through an apical route into the tubule lumen. The ultrastructure of the TC is consistent with apical secretion ([26] and HS, unpublished data), and the distribution of Rhomboid, which is enriched apically in both TCs and SCs, favours this hypothesis (Figure 2B) [31]. Alternatively, short-range signalling from the TC lineage might act to break axial symmetry, followed by local interactions between cells to propogate polarity. We hoped that it would be possible to distinguish between these models by generating clones of cells with altered EGFR activity or ectopic ligand secretion in order to assess non-cell autonomous effects of altered signalling on cell polarity. However clones generated even at syncytial stages of embryogenesis yield tubule clones that do not exceed two to three suitably labelled cells, ruling out the validity of this approach (see Figure S5).
Our analysis also revealed that there is little discernible expression of dpERK or modulation of Capicua::Venus in the proximal half of the tubules and this is reflected in the lack of any polarisation in the distribution of cortical Slam-HA or of asymmetric actomyosin activity. These findings indicate that, although tubule extension in the proximal regions, as in the distal, results from circumferential intercalation of cells, underlying movements must be regulated by different processes in the two halves. Our analysis has focussed on the anterior tubule pair, whose forward movement through the body cavity is regulated by guidance cues expressed by specific target tissues [64], and we suggest that extension of the distal tubule results in sufficient forward movement to deliver the cue-responsive kink region close to target tissues that promote continued forward movement of the whole tubule. As the tubule is tethered both to the ureter/hingut proximally and distally, by TCs/alary muscle contacts [65], forward tubule movement will produce mechanical forces that could promote circumferential intercalation in the proximal tubule half.
In the distal tubule, one important consequence of cell polarisation is asymmetry in the activity of Myosin II to the proximal side of tubule cells, which leads to transient circumferential cell contraction. Oriented Myosin II accumulation has been shown to result from EGF signalling in the tracheal placode; in the absence of signalling, Myosin II accumulation remains punctate and dispersed [66]. We find a similar phenotype. When EGF signalling is perturbed, Slam fails to become localised and Myosin II remains dispersed in unpolarised cells so that pulses fail altogether.
A novel observation of this study is that planar Myosin II pulses are required in the basal cortex of MpT cells for CE. This contrasts with findings in the extending germ-band where adherens junction remodelling resulting from apical planar actomyosin enrichment has been proposed as a major motive force for cell rearrangements [6],[7],[15]. In MpT cells, asymmetric Myosin II activity within 4 µm of the basal surface correlates with cell shape change and is required for CE movements. We have examined the apical side of tubule cells for actomyosin dynamics and do not detect repeated or polarised Myosin II cresents in this region of the cell cortex (see Movie S15), but we have not been able to live image the apical regions deeper in the tubules with sufficient reliability to assess whether junctional remodelling precedes or follows the basal changes. He and colleagues [53] have shown that oocyte elongation depends on pulsatile basal contractions of follicle cells, oriented in the circumferential axis. Similarly elongation of the Caenorhabditis elegans embryo results from the intercalation of hypodermal cells led by basal, medially directed protrusions [67]. It is possible that intercalating cells commonly initiate movements basally and that in epithelia junctional remodelling follows, also contributing actively to tissue morphogenesis [24].
A remaining question is how the acquisition of PCP relates to asymmetry in cytoskeletal activity. Slam localisation during cellularisation of the Drosophila embryo or in the extending germ band is known to highlight sites of Myosin II accumulation [6],[13],[68]. Like the extending germ band, tubule cells do not express Slam endogenously; its localisation therefore must reflect asymmetry in a binding partner, such as RhoGEF2, to which it is known to bind during cellularisation [68]. An antibody against RhoGEF2 revealed that its expression is scarcely detectable in tubule cells and it does not appear to be asymmetrically localised. However, mutants for RhoGEF2 show CE defects in tubule elongation [69]. This suggests that RhoGEF2 might provide a link between EGF signalling, tubule cell polarity and asymmetric cytoskeletal activity.
The duration of myosin pulses and cell circumferential contraction is approximately 2 min, while cell intercalation takes an average of 42 min. During tubule elongation the diameter of cells is around 5 µm (see Figure 5) yet our measurements indicate that cells move an average of 1 µm min−1. Are these measurements consistent with the dynamics and timing of tubule morphogenesis as a whole? If one assumes that cells move in a consistent direction during intercalation this would suggest a serious overshoot so that cells would move past their neighbours. However cell movement relative to neighbouring cells is not uniform (see Movie S4) and the movements we measure result from multiple factors; (a) the displacement of the whole tubule as a result of gut morphogenesis (for example; hindgut elongation during stages 13–16 [70]); (b) the concerted movement of tubule cells as a result of cell rearrangements in more distal regions; and finally (c) movements of individual cells relative to their neighbours that produce cell intercalation. The first two tend to produce distal-to-proximal movement, which would explain the deviation in total cell movement from the circumferential axis seen in Figure 5I. Our observations concerning circumferential movement suggest that cell intercalation results from repeated, transient, and very tiny movements (Figure 7iii)—which must be stabilised perhaps by adhesion either to the basement membrane or to adjacent cells—in which a small but consistent circumferential bias eventually achieves cell rearrangement (Figure 7iv).
Concerning timing; tubules increase in length 4-fold with the reduction of eight to 12 cells around the lumen to just two cells (Figure 1A, 1C, and 1D). Simple calculation suggests that this would require that every cell intercalates twice, with some undergoing a third cell rearrangment. Tubule extension takes approximately 5 hours and each intercalation an average of 42 min, indicating that three intercalation events could be accommodated in the time-frame of elongation.
The mechanisms known to drive tubule elongation include oriented cell division and polarised changes in cell shape, as well as CE cell rearrangements [16],[19],[22],[71]. Here we show that fly renal tubules elongate predominantly by cell intercalation. In frog and mouse embryos renal tubule extension also depends primarily on cell rearrangements as the orientation of cell division is random [16],[22]. As in Drosophila CE depends on Myosin II activity that is polarised in the plane of the tubule epithelium to bring about mediolateral cell intercalation. But in contrast to our findings, nephron elongation results not from cell intercalation directed by graded EGF signalling but from PCP gene-regulated formation and resolution of multicellular rosettes [22]. However, EGF signalling does play an important role during mouse embryonic kidney development in regulating both nephron cell proliferation and morphogenesis and in collecting duct extension [72],[73].
The mature shape of tubular epithelial tissues is critical for their effective function. Abnormalities in the morphogenesis of renal tubules in mammals, for example in cystic kidney disease, results in defective excretory physiology leading to premature death [1],[24],[74]. Flies lacking the normal polarising signals that regulate renal tubule morphogenesis similarly suffer renal malfunction leading to lethality [75]. The identification of a genetically manipulable system in which to study the molecular interactions that lead from cell polarity to asymmetric cytoskeletal regulation, polarised cell movement and tissue shaping provides a powerful model for future analysis of tubule morphogenesis in health and disease.
Flies were cultured on standard media at 18°C or 25°C with ectopic expression at 29°C. Embryos were collected overnight at 25°C (29°C for dual colour imaging) on apple-juice agar plates with yeast paste. The following stocks were used: Oregon-Red (wild type); ctB-Gal4; UAS-RedStinger6; UAS-(EGFP)Stinger2; Capicua::Venus (gift of E. Wieschaus); UAS-EGFRDN (gift of M. Freeman); UAS-λtop4.2/4.4 (UAS-EGFRact, gift of T. Shupbach); EGFRf7; UAS-SqhE20E21; UAS-YFP-ZipDN and sqhAX3, sqh-Sqh::GFP, sqh-Sqh::GFP (gift of T. Lecuit); sqh-Sqh::mCherry [51] (gift of A. Martin); UAS-GAP43::GFP; sqhAX3, sqh-Sqh::GFP, sqh-GAP43::mCherry [76] (gift of B. Sanson); A37-LacZ (nrmLacZ); Df(os)1A; UAS-Slam::HA [40] (gift of J. Zallen); hs-flp122 (gift of J. Castelli-Gair Hombría), tub>stop>Gal4 (gift of M. Landgraf), FRT dsUA071; dsUAO71, stanE5; dsUAO71, stan3; dsh1; fz1; stbm6; ftG-rv (gifts of J. Casal and D. Strutt).
hs-flp122; tub>stop>Gal4/UAS-Slam::HA; embryos were collected for 2 hours at 25°C and heat-shocked in a 37°C water bath for 10 minutes. Embryos were aged to stage 15 at 25°C, fixed and processed for antibody staining.
Embryos were dechorionated in 50% bleach, washed extensively with double-distilled water, and oriented dorso-laterally to visualise anterior Malpighian tubules (aMpTs). Oriented embryos were mounted on type-1 coverslips with an evenly spread layer of glue (3M Scotch tape glue-Heptane). Care was taken not to compress the embryos. Mounted embryos were covered with Voltalef-3S or Halocarbon-10S oil. For dual colour imaging of Myosin II and membrane dynamics, embryos of the following genotypes were used: w−;ctB>UAS-GAP43::GFP, sqh-Sqh::mCherry/sqh-Sqh::mCherry;+; sqhAX3;sqh-Sqh::GFP;sqh-GAP43::mCherry; w−;ctB>UAS-GAP43::GFP, sqh-Sqh::mCherry/UAS-EGFRDN;sqh-Sqh::mCherry/+.
Images were acquired on Leica SP5 or Olympus FV1000 confocal microscopes with 488 nm and 561 nm lasers. An Argon ion laser was used for imaging GFP; dsRed and mCherry were imaged with a 561 nm diode laser. z-Stacks were acquired every 45 seconds for 5–6 hours with a water immersion 20×/0.7 NA objective to capture aMpT elongation (Figure 1C, 1F, 3D, and 3F). 60×/1.4 NA (Figure 5A) or 63×/1.4 NA (Figure 5B and 5C). Oil immersion objectives were used to visualise Myosin II and membrane dynamics in the basal-most 2–4 µm z-sections every 8–15 sec. Dual colour imaging was performed using previously established excitation band-pass settings [51] with Leica-SP5 Hybrid detectors. All images except those in Figure 5A were acquired on a Leica-SP5 confocal microscope. For Sqh::GFP in Figure 5A an Olympus-FV1000 confocal was used. All embryos completed development and hatched as L1 post-imaging.
aMpT lengths and cell shape changes were analysed using ImageJ (http://imagej.nih.gov/ij/). Cell tracking was performed using SIMI-Biocell (SIMI reality motion systems). Origin (OriginLab) and SigmaPlot (Systat Software) were used for statistical analysis, independent t-test, and for generating graphs.
Cell-tracking and speed measurements were performed as described previously [77], with minor modifications. SIMI-Biocell (version 4.0 built 155, SIMI reality motion systems) was used for tracking cell movements and for the generation/colouring of 4-D reconstructions. 3-D positions of fluorescently labelled aMpT nuclei were tracked over time manually. 3-D coordinates of the nuclei were saved every 9 minutes (or every 1 minute in Figure 1H and 1H″) during the course of a movie. Cell speeds were measured by calculating the distance moved by aMpT nuclei every 9 minutes (or every 1 minute in Figure 1H and 1H″). Movies and 4-D reconstructions were annotated and represented in their final form using ImageJ (Rasband WS, National Institutes of Health, http://imagej.nih.gov/ij/; 1997–2012).
Cell shape analysis was performed using ImageJ. Single z-slices at 12–18 second intervals were used to manually trace basal cell outlines with the polygon selection tool. Traces were saved using the ROI manager. The centre and total area for each trace was determined with the in-built “centroid” and “area” measurement tools. Cell length was measured by drawing lines through the centroid that connected edges in axes either parallel (distal-proximal) or perpendicular (circumferential) to the distal to proximal tubule length. Area and lengths of a cell were normalised with their average represented as 1.0.
aMpT lengths were calculated by drawing and measuring a segmented line along the distal to proximal tubule length in ImageJ.
Dechorionated ctB>UAS-CD8-GFP embryos were mounted on double-sided Scotch tape in PBS solution. The TC and surrounding two or three cells were ablated (to ensure removal of both TC and SC) in late 12/early 13 stage embryos. Cell ablation was performed using a 63×/0.9 NA water immersion lens on a Yokogawa spinning disk (CSU-10) confocal microscope fitted with a pulsed nitrogen laser (MicroPoint). Image acquisition and microscope control were by MetaMorph (version 7.0) software (Molecular Devices). Embryos were allowed to develop to stage 16 under humid conditions at 25°C, fixed and processed for immunostaining.
Embryos were fixed in 4% paraformaldehyde and devitellinised by vigorous shaking in 1∶1 heptane/methanol. Immunostaining was performed using standard techniques. For pMLC staining (Figure S5), embryos were fixed in 37% formaldehyde for 3–5 minutes and devitellinised using a fine glass needle.
The primary antibodies used were: mouse anti-FasII (1∶10, DSHB); mouse anti-Cut (1∶50, DSHB); rabbit anti ß-gal (1∶10,000, ICN Biomedicals); rabbit anti-Rhomboid (1∶500, gift of E. Bier); rabbit anti-dpERK (1∶50 Cell Signaling technology); rabbit anti-Bazooka (1∶500, gift of A. Wodarz); rabbit anti-Cleaved Caspase3 (1∶20, Cell Signaling technology); goat anti-GFP (1∶500, Abcam); mouse anti-Futsch/22c10 (1∶200, DSHB); rat anti-HA (1∶200, Roche), rabbit anti-phospho-Myosin Light Chain 2 ([Ser19]; 1∶20, Cell Signaling technology).
Secondary antibodies were used at 1∶200. Appropriate biotinylated secondary antibodies were used with the Vector Elite ABC Kit (Vector Laboratories) for DAB staining. FITC- or Cy3-conjugated secondary antibodies were used for fluorescent labelling. When required, streptavidin-conjugated FITC/Cy3 amplification was used. TSA-Biotin amplification system (Perkin-Elmer) was used for dpERK detection. DNA was stained with DAPI (1∶1,000, Molecular Probes). Embryos and tissue were mounted in Vectashield (Vector Laboratories) and viewed on a Leica SP5 confocal microscope.
Image processing was performed using ImageJ and Adobe Photoshop. To measure dpERK staining levels, stage 13 tubules were traced using the segmented line tool in ImageJ with a line width approximately equal to tubule width. The plot profile tool was used to quantify staining intensity along the line. Values were binned into 1 µm bins and averaged for n = 7 tubules. Figures were assembled in Adobe Illustrator.
Embryos of the appropriate stage were fixed and stained with anti-FasII, dehydrated, and mounted in Araldite resin. Transverse sections approximately 2.5 µm in thickness were made midway along the distal region of aMpTs using a Reichert microtome.
Embryos of the appropriate genotype were collected overnight and aged for a further 6 hours at 29°C. 40 first instar larvae were transferred to a vial of standard food and incubated at 25°C until adults eclosed. Secretory assays were performed as described previously [55] at 23–24°C using 3–5 day old adults. cAMP and LK1 (Sigma) were added to a final concentration of 1 mM and 100 µm at approximately 30 and 60 min, respectively. To measure wet and dry body weights, flies were briefly anaesthetized with CO2, transferred to Eppendorf tubes on ice, three flies were pooled and weighed on a Mettler Toledo precision balance (wet weight). The flies were killed by freezing for 20 minutes and transferred to a 50°C oven containing a tray of silica crystals, allowed to desiccate for ∼24 hours, and weighed again (dry weight).
|
10.1371/journal.pgen.1007763 | Metabolic adaptations underlying genome flexibility in prokaryotes | Even across genomes of the same species, prokaryotes exhibit remarkable flexibility in gene content. We do not know whether this flexible or “accessory” content is mostly neutral or adaptive, largely due to the lack of explicit analyses of accessory gene function. Here, across 96 diverse prokaryotic species, I show that a considerable fraction (~40%) of accessory genomes harbours beneficial metabolic functions. These functions take two forms: (1) they significantly expand the biosynthetic potential of individual strains, and (2) they help reduce strain-specific metabolic auxotrophies via intra-species metabolic exchanges. I find that the potential of both these functions increases with increasing genome flexibility. Together, these results are consistent with a significant adaptive role for prokaryotic pangenomes.
| Recent and rapid advancements in genome sequencing technologies have revealed key insights into the world of bacteria and archaea. One puzzling aspect uncovered by these studies is the following: genomes of the same species can often look very different. Specifically, some “core” genes are maintained across all intraspecies genomes, but many “accessory” genes differ between strains. A major ongoing debate thus asks: do most of these accessory genes provide a benefit to different strains, and if so, in what form? In this study, I suggest that the answer is “yes, through metabolic interactions”. I show that many accessory genes provide significant metabolic advantages to different strains in different conditions. I achieve this by explicitly conducting a large-scale systematic analysis of 1,339 genomes across 96 diverse species of bacteria and archaea. A surprising prediction of this study that in many ecological niches, co-occurring strains of the same species may help each other survive by exchanging metabolites exclusively produced by these different accessory genes. More pronounced gene differences lead to more underlying metabolic advantages.
| Prokaryotes exhibit remarkable genome flexibility, with strains from the same species often containing dramatically different gene content [1–4]. Intraspecific differences in gene content are often characterized by a “core” genome (genes common to all strains) and “accessory” genome (genes found in a fraction of strains) [5]. While the core genome might represent a set of species-specific indispensable genes, we do not yet understand whether the accessory genome of a species is the result of neutral or adaptive evolution. Indeed, this is the subject of an ongoing debate: do the majority of prokaryotic accessory genes have negligible or positive fitness effects, i.e. are they neutral or adaptive?
Recent population genetics arguments support roles for both neutral and adaptive evolution as possible factors driving accessory genome evolution [6–9]. For example, microbial species with more accessory genes also tend to have larger effective population sizes, as expected of genetic variation in a population under neutral evolution [9]. On the other hand, models in which microbial genomes evolve in large, migrating populations, suggest that acquired genes can often be beneficial, as expected under adaptive evolution [7]. However, these studies have only addressed broad aspects of microbial populations such as effective population size, migration, and the fitness effects of gene loss and gain. In response, subsequent criticisms of these studies have strongly expressed the need for more functional, gene-explicit and ecological analyses [10–11]. Here I present the first such systematic analysis of 96 phylogenetically diverse prokaryotic species, which suggests that prokaryotic accessory genomes often provide significant metabolic benefits.
I chose to study metabolism as a possible explanatory factor for three reasons. (1) Metabolic genes dominate the functional content of accessory genomes [12] (S8A Fig). (2) Metabolic interactions between microbes—especially interdependencies—can often be adaptive [13–14]. For instance, microbes that obligately cross-feed, i.e. that critically depend on exchanging metabolites with each other, can grow faster than their wild-type counterparts [13]. Such a fitness benefit can also drive genomes, in many cases, to lose genes and become metabolically dependent [14,15]. If different genes are lost between different conspecific strains, this can lead to both metabolic interdependence, as well as accessory genomes (since different strains will have different metabolic repertoires) [16]. (3) Databases such as KEGG contain already-curated genomes for several fully-sequenced strains. KEGG contains high-quality gene and reaction annotations, allowing us to accurately predict the biosynthetic capabilities of each strain under different conditions [17].
In this study, I ask to what degree accessory genes can metabolically benefit conspecific strains. For this, I have used genome-scale metabolic network reconstructions of 1,339 prokaryotic strains (corresponding to 92 bacterial and 4 archaeal species) from the KEGG database over 59 distinct nutrient environments. In general, my analyses reveal two beneficial roles for the accessory metabolic content of prokaryotes. First, I find that the accessory genome of most species harbours extensive biosynthetic potential, with several accessory genes providing strains with additional nutrient utilization abilities. Second, I find that pairs of strains from the same species often display a remarkable potential for metabolic interdependence, which scales with the amount of accessory genome content. These interdependencies have the ability to alleviate strain-specific auxotrophies in a particular niche through the exchange of secreted metabolites. My results are, from a metabolic standpoint, consistent with a possible adaptive evolution of accessory genomes.
To obtain a large set of species pangenomes, I first collected a list of all prokaryotic species in the KEGG GENOME database, and filtered those that had complete genomes for 5 or more conspecific strains. This gave me 1,339 genomes (96 species), which I used in all my subsequent analyses (S1 Table). To account for potential biases due to uneven phylogenetic sampling, I verified that a more restrictive choice of one species per genus did not significantly impact my results (55 species; S1 Fig). For each strain, I then extracted all annotated genes and metabolic reactions from the KEGG GENES and REACTION databases, respectively. To quantify accessory genome content for every species, I used the well-studied genome fluidity measure, φ [18]. For this, I calculated, across each pair of conspecific strains, the fraction of all genes in the pair that were unique to each strain. The average of this fraction over each species gave me its genome fluidity φ (see Methods).
I constructed metabolic networks for every individual strain, where each network contained the set of reactions corresponding to the strain’s genome in KEGG. I included gap-filled reactions when curated models were available [19], though I verified that their addition did not impact my results (S2 Fig). I used these reaction networks to infer the biosynthetic capabilities of each strain under several different conditions. To define these conditions, I selected 59 different carbon sources, previously shown and commonly used to sustain the growth of diverse microbial metabolisms in laboratory experiments [20–22] (S2 Table). I associated with each carbon source a different nutrient environment or condition. In each condition, I included exactly one of the 59 carbon sources, say glucose, along with a set of 30 commonly available metabolites, which I assumed were always available (for example, water and ATP; S3 Table).
To assess biosynthetic capability, I curated a list of 137 crucial biomass precursor molecules, often essential for growth (hereafter, “precursors”) from 70 experimentally verified high-quality metabolic models [23] (S4 Table). Finally, to calculate what each strain could synthesize in a particular environment, I used a popular network expansion algorithm: called scope expansion [24–25]. This algorithm determines which metabolites each strain can produce—its “scope”—given an initial seed set of already available metabolites. To start with, only those reactions whose substrates are available in the environment can be performed, and their products constitute the initial set of metabolites that can be produced. These metabolites can then be used as substrates for new reactions that can then be performed, and step by step, more metabolites can be produced. When no new reactions can be performed, the algorithm stops, giving the full set of metabolites that could be synthesized in the given environment. Such a calculation sidesteps the need for arbitrary assumptions of binary (yes/no) growth and optimality typically used in more complex metabolic modeling approaches such as flux balance analysis [26] and is well-known for its ability to infer what metabolic networks can synthesize in diverse conditions [27–28].
I first investigated the capabilities of individual strains. Specifically, I was interested in the extent to which the accessory genes in each strain expanded the set of precursors that could be synthesized. For each species, I calculated, via network expansion, the list of precursors that could be produced per strain per condition. I then counted how many unique precursors each strain could synthesize across all conditions, i.e. by the accessory genes alone. From this, I computed, for every species, its accessory metabolic capacity α, defined as the average number of precursors (per strain per condition) produced exclusively due to the accessory genome.
I found that while for 18 species this quantity was zero, for the majority of prokaryotic species (81%), this number lay between 0.1 and 15.0 (mean 3.1; median 2.0). Further, α scaled positively with genome fluidity φ (Spearman's rho = 0.44; P value = 7 x 10−6; Fig 1).
Since I observed that the accessory genome of different strains typically imparted different biosynthetic capabilities to different strains, I wondered if, in the same conditions, metabolic interactions between conspecific strains could further expand these capabilities. This could, for instance, indicate a potential dependence of an auxotrophic strain on another strain, i.e. a strain that cannot produce a crucial precursor in an environment. Such auxotrophies have been previously shown for example, in different strains of Escherichia coli co-inhabiting the human gut [29].
For each pair of conspecific strains in each condition, I calculated a metabolic dependency potential (MDP), defined as the average number of new precursors each strain has the potential to synthesize when grown as a pair versus alone. Here I assessed, in every pair, which metabolites that could be produced and secreted by one strain could subsequently allow the production of a new precursor in the other strain that it would not otherwise be able to make (i.e. was auxotrophic for). Note that this method does not count those metabolic interactions that can provide extra (functionally redundant) pathways to produce a precursor and supplement growth, and is thus more likely to represent actual or obligate dependencies. I verified that my approach can successfully predict such obligate dependencies by comparing with some well-documented intra- and inter-species pairs [13–14,30–32] (see Methods) (S3 Fig).
I found that while I could not detect any dependency potential for 17 species, surprisingly, the majority of species (82%) showed an MDP per strain per condition between 0.1 (for Bacillus thuringiensis) and 3.3 (for Ralstonia solanacearum), with a mean 1.7 and median 1.4. Interestingly, the 17 species for which I could not detect any MDP matched those with zero α (the leftover Legionella pneumophila showed low MDP = 0.5). Over all tested pairs with detected dependency potential (48%), commensal interactions were more common than mutualisms (29% versus 19%; Fig 2B). This is because, in species with detected dependencies, not all pairs show dependency potential (on average 46% conspecific pairs do). The auxotrophies relieved by these dependencies varied from those for amino acids, vitamins, carbohydrates, and organic acids, among others (Fig 2C).
Strikingly, like α, MDP also scaled positively with genome fluidity φ, suggesting that greater amounts of accessory content can potentially sustain more conspecific metabolic dependencies (Spearman's rho = 0.56, P value = 4 x 10−9; Fig 2A). For both α and MDP, considering medians instead of means did not impact my results (for α: Spearman's rho = 0.38, P value = 10−4; for MDP: Spearman's rho = 0.51, P value = 10−7; S4 and S5 Figs). To verify that such potential conspecific dependencies are indeed ecologically realizable, I repeated my analysis, this time restricting it to those genomes, which were known to co-occur in microbial communities (29 strains across 14 species; see Methods). I found that my observed trend was still valid, namely MDP still scaled with φ, suggesting that several auxotrophies may indeed be reduced through within-species metabolic exchanges in nature (Spearman’s rho = 0.56, P value = 0.03; S6 Fig).
Given the extent to which I detected the potential for obligate metabolic interactions between conspecific strains, I wondered whether such interactions are possibly common among prokaryotes. For this, I extended my study to analyze metabolic dependency potential between inter-specific strains (see Methods). I found that, indeed, strains from all species can metabolically depend on strains from at least one other species to alleviate potential auxotrophies across many different environments (with MDP ranging from 1.6 to 3.2, with mean 2.2; S7 Fig). Interestingly, I found that these interspecific metabolic interactions often involve accessory metabolic genes as well.
Taken together, my results suggest that a considerable fraction of prokaryotic accessory genomes contains potentially beneficial metabolic functions (upto 70% of accessory genes per strain across this study, with median 40%; see S8D Fig). Specifically, I found that the accessory “metabolome”: (1) expands a genome’s biosynthetic potential, possibly allowing for niche-specific adaptations [29,33–34]; and (2) reduces potential auxotrophies via obligate metabolic interactions, also explaining how conspecific strains can coexist despite high competitive potential [35–38]. The accessory genes that impart these functions are often different (median overlap 10%), suggesting that these are indeed distinct, non-redundant benefits. Moreover, apart from these additional biosynthetic abilities, the y-intercept for genome fluidity (at φ = 0.03 for both α and MDP) provides an estimate of metabolic redundancies (such as extra pathways).
My findings may additionally help explain the following observations: (1) metabolic functions are enriched in accessory genomes (median 50% in accessory versus 38% in core; S8C Fig); and (2) the variation in accessory metabolic genes exceeds the variation in genes of many other functions (metabolic variation being dominant in 81% of examined cases; S8B Fig).
Previous studies have suggested that the evolution of metabolic dependencies likely occurs via adaptive gene loss [15,39] (e.g. the Black Queen hypothesis). Such a mechanism suggests that metabolic dependency evolution can often lead to reduced genome sizes, but makes no comment on genome flexibility (i.e. gene content variability). My results also indicate that metabolic dependency evolution can impact genome flexibility as well. Specifically, more flexible genomes (with more variable gene content) are more likely to display a potential for metabolic interactions.
Can stochastic accessory gene turnover explain these results? To test this, I repeated my study with randomly assembled pangenomes. Within each species, I retained the core genes in every strain and shuffled the accessory genes between strains (see Methods). During this randomization, I preserved the observed within-species genome size distribution, strain number distribution, and ensured that any change in species’ genome fluidity was insignificant. I found that not only did this significantly diminish the metabolic benefits observed in each species, both measurements of α and MDP yielded non-significant correlations, suggesting that the mere presence of additional accessory genes is unlikely to explain my observed trends (for α: Spearman's rho = 0.01, P value = 0.9; for MDP: Spearman's rho = 0.17, P value = 0.1; S9A and S10A Figs). The measured benefits remained lower than observed, even when I shuffled known operons of genes together instead of shuffling genes one by one (S9B and S10B Figs; see Methods). I believe this is because often, prokaryotic operons do not contain complete metabolic pathways, but instead parts of them (in my data set, each metabolic operon encoded 1.5 reactions on average, while pathways typically had 4 to 5 steps). Collectively, this suggests that accessory gene acquisition is consistent with the coordinated gain of functional and beneficial pathways, which I believe provides further support for the accessory genes being maintained for adaptive reasons.
To summarize, here I addressed the debate on whether the accessory genomes of prokaryotes are beneficial. I found that, indeed, large fractions (about 40%) of the prokaryotic accessory gene pool can contribute to metabolic benefits. Specifically, such genes can allow microbes to produce a larger repertoire of crucial molecules, and facilitate the exchange of others. Since these functions can improve growth in many habitats, my results suggest that adaptation may explain accessory gene maintenance. Note, however that my analyses are only capable of detecting obligate metabolic dependencies and biosynthetic potential, and do not consider signaling, regulation, metabolic redundancy, etc. that could also play important functional roles and might indicate potential benefits due to additional accessory genes. Further work might also explain accessory genomes in those species, where I could not detect additional metabolic functions, if such roles are indeed there. Moreover, even in the context of metabolism, more detailed metabolic models, when available, may be used to probe even more precise fitness effects of intraspecies metabolic variability, including the effect of higher-order interactions. However, these studies would require knowledge of a large number of parameters such as reaction kinetics, thermodynamics, and exact strain biomass compositions before they are feasible. Finally, systematic measurements of the fitness effects of all accessory genes, metabolic and otherwise, are needed for more complete estimates of the fraction of accessory genomes consistent with adaptive versus neutral evolution.
I used the KEGG GENOMES database [17] to extract a list of all prokaryotic species with complete genomes for 5 or more strains. This yielded a list of 1,339 strains or genomes corresponding to 96 species (92 bacteria, 4 archaea), which I used for all subsequent analyses (see S1 Table for the full list of species and strains, along with their taxonomic classification). For each strain with a unique genome abbreviation, I extracted the full set of annotated genes under the KEGG GENES database and reactions under the KEGG REACTION database using an in-house Python script. I also extracted the full list of reactions with their stoichiometries and participating metabolites in the database. The metabolic reaction network for each strain was considered to be the complete set of annotated reactions detected in that strain’s genome in KEGG. Note that my analyses systematically ignore genes without known functions.
For strains for which genome-scale metabolic reconstructions were available in the Model SEED database [19], I also included gap-filled reactions. Specifically, I extracted the list of all gap-filled reactions for 130 genomes from table S3 in ref. 19. I mapped all genomes from this table to KEGG genomes by matching strain names, and all reaction IDs to KEGG reactions by searching the Model SEED database online (https://modelseed.org/biochem/reactions) using a custom Python script. This resulted in a total of 562 gap-filled reactions, spread across 22 genomes (20 out of 96 species; S6 Table). I then added these reactions to the metabolic networks already constructed via KEGG. Separately, I verified that adding these gap-filled reactions did not impact my results (S2 Fig).
For nutrient environments or conditions, I selected a set of 59 diverse carbon sources known to sustain microbial biomass and energy synthesis from previous genome-scale metabolic studies of phylogenetically broad species [20–22] (S2 Table). Every condition was assumed to contain one of these carbon sources (such as glucose and maltose), along with a set of 30 commonly available metabolites (assumed to be present in all conditions, such as water, oxygen and ATP), similar to the aforementioned studies (S3 Table). To infer biosynthetic potential, I separately collected a set of all prokaryotic species biomass compositions and their constituent metabolites from high-quality experimentally verified metabolic models in the BiGG database [23]. I curated from this a list a union of 137 biomass precursors across diverse microbial metabolisms (S4 Table).
To infer what each strain could synthesize in each nutrient environment or condition, I used a well-documented network expansion algorithm—scope expansion [24–25]. Briefly, this algorithm is given a reaction network (one from each genome) and an initial “seed” set of available metabolites (each nutrient environment). It first determines which reactions can be performed by the network using only the nutrients in the environment. I assume that metabolites that are products in this initial set of reactions can be synthesized by the network, and can be subsequently used as reactants in new reactions. Again, I consider that the products of such new reactions can be synthesized by the network, and may allow additional new reactions to be performed. This continues step by step, till no no new reactions can be performed. All metabolites that can be produced over all such steps are defined as the “scope” of the metabolic network, i.e. I assume that these metabolites can be synthesized by the reaction network from the initial nutrients in the environment.
I calculated genome fluidity φ as prescribed in a previous study [17], using a custom Python script. For every genome, I considered each constituent gene’s KO number as its unique identifier. Then, to estimate φ for every species, I calculated, for all conspecific pairs, the ratio of the number of genes unique to a strain in the pair to the total number of genes in their sum. The average over all pairs for a species was considered its genome fluidity φ. Note that though using KEGG orthologous groups underestimates the exact values of φ, my estimates still scale well with previously reported values [9] (Spearman's rho = 0.60, P value = 7 x 10−5; S11 Fig).
I calculated an accessory metabolic capacity α for every species. For each conspecific strain, I first calculated, using the network expansion algorithm described, the scope of each of the 1,339 reaction networks across all 59 conditions. Then, species by species, for every condition, I calculated a “core” metabolome, i.e. metabolites that were present in the scope of every conspecific strain. I then explicitly removed these metabolites within every species from the scope of each strain and counted how many precursors remained in the corresponding “accessory” metabolome of every strain across all conditions. This gave me a number of additional precursors that could be synthesized per strain per condition for each species, and was defined as the species’ accessory metabolic capacity α (S5 Table).
I calculated a metabolic dependency potential (MDP) for every species. For this, I considered within each species, all conspecific pairs across all 59 conditions. For each pair, I calculated the scope for both strains first in “monoculture”, i.e. when grown alone. I then calculated, for every metabolite that could be produced by one strain but not the partner strain, whether or not its secretion could alleviate an auxotrophy in the partner. I specifically considered auxotrophies only for the 137 key precursors I had selected. I then counted each alleviated auxotrophy as a potential metabolic dependency, and the average number of dependencies (per strain per pair per condition) for each species was defined as its metabolic dependency potential, or MDP (S5 Table).
To quantify the extent of metabolic interactions between inter-specific strains, I calculated a separate metabolic dependency potential for every species. For each species, I paired each conspecific strain with 25 randomly chosen strains from other species, also picked at random. For all inter-specific pairs generated this way, I calculated metabolic dependency potential using the same method as described above, for conspecific pairs. In this way, the average number of dependencies identified per strain per condition was defined as the inter-specific metabolic dependency potential, or MDP (S5 Table).
To test whether the conspecific metabolic interactions detected in my MDP analysis could be realized in natural microbial communities, I analyzed genome co-occurrence data from Chaffron et al [40]. These data list all 16S rRNA sequences co-detected across several microbial community samples. Here, sets of sequences are clustered into operational taxonomic units (OTUs) corresponding to different sequence similarity thresholds. To map these OTUs to the genomes in my study, I first obtained 16S rRNA sequences for all the 1,339 genomes I analyzed from KEGG. When multiple sequences were available for a given genome, I used the longest sequence and maped that as the unique 16S identifier for that strain. Then, using BLAST, I mapped OTUs in the co-occurrence data to the genomes in my study (where OTUs were binned with a sequence similarity threshold of 99%). Here, I used the BLAST bit score as my assignment criterion. I used the 689 genomes that could be mapped this way for further analysis. Here, across all microbial community samples, I asked which conspecific genomes co-occurred in at least one sample—from which I found 29 genomes corresponding to 14 species (S7 Table). I then repeated my metabolic dependency potential analysis for these conspecific strains, as described above.
To test if my observed correlations between φ and α as well as φ and MDP could be explained by random accessory gene turnover, I repeated my study with a “randomly assembled” pangenome dataset. I randomized genomes species by species. I first collected all available genomes for a species, and picked a random pair of these. I then shuffled the accessory genes in this pair in two ways: (1) gene by gene, and (2) operon by operon.
When shuffling gene by gene, for each strain pair, I randomly picked two genes, one from each strain in the pair, and swapped them. I repeated this several times before picking another conspecific pair from the same species. The number of swaps per pair was chosen such that each accessory gene was swapped once on average. I verified that the exact number of swaps does not affect my results. By the end of this process, I had a new set of genomes which had undergone “stochastic accessory gene turnover”. Note that in order to avoid any potential biases, this process preserves the observed genome sizes and strain numbers while only slightly affecting genome fluidity. I then repeated my α and MDP calculations for these “shuffled” genomes. This would test if the mere acquisition of extra genes from a species’ accessory genome could allow the expanded biosynthetic potential and metabolic dependencies observed in the data.
To identify operons, I used the ProOpDB database, which lists operon compositions for more than 1200 prokaryotic genomes [41]. I found that this database had operon compositions available for 795 strains across 64 of the species in my study, which I used for the operon shuffling analysis (S8 Table). Here, when shuffling operons, I used a similar method as when shuffling genes, but instead of swapping merely randomly chosen genes from a pair of strains, I identified which operon they belonged to in their respective strain’s genome, and swapped all genes in those operons across the pair. I repeated these operon swaps several times for each strain pair, and for several pairs, at the end of which, I had another new set of randomly shuffled genomes.
To test if my metabolic dependency potential (MDP) measure could accurately predict metabolic dependencies between different pairs, I compared its performance on genome-scale metabolic networks corresponding to some well-studied experimentally verified metabolically dependent pairs. Specifically, I considered 2 conspecific and 4 inter-specific pairs. For conspecific dependencies, I used 2 Escherichia coli cross-feeding pairs [13] and for inter-species, I used (1) a Desulfovibrio vulgaris and Methanococcus maripaludis pair [30]; (2) an E. coli and Acinetobacter baylyi pair [14]; (3) an Lactobacillus bulgaris and Streptococcus thermophilus pair [31]; and (4) a Bifidobacterium longum and Eubacterium rectale pair [32]. In all cases, I obtained the metabolic models for the closest available strains from KEGG and, when needed, modified the genes present to best match those described in the respective studies. I then used my MDP approach as described to infer which potential dependencies were detected in each pair for the specific conditions mentioned in each study. I found that my method could accurately identify the extent of dependencies (number of auxotrophies) and interaction directionality (commensal or mutualistic interactions; S3 Fig).
To calculate correlation coefficients throughout the study, I used Spearman’s nonparametric rho, and for P values, I used a one-way asymptotic permutation test for positive correlation. All statistical tests were performed using standard implementations in the SciPy (version 0.18.1) and NumPy (version 1.13.1) libraries in the Python programming language (version 3.5.2). Linear regression and prediction interval calculations were performed using the Seaborn library function regplot (version 0.7.1).
All computer code and extracted data files used in this study are available at the following URL: https://github.com/eltanin4/pangenome_dep.
|
10.1371/journal.pgen.1006886 | Mouse models of 17q21.31 microdeletion and microduplication syndromes highlight the importance of Kansl1 for cognition | Koolen-de Vries syndrome (KdVS) is a multi-system disorder characterized by intellectual disability, friendly behavior, and congenital malformations. The syndrome is caused either by microdeletions in the 17q21.31 chromosomal region or by variants in the KANSL1 gene. The reciprocal 17q21.31 microduplication syndrome is associated with psychomotor delay, and reduced social interaction. To investigate the pathophysiology of 17q21.31 microdeletion and microduplication syndromes, we generated three mouse models: 1) the deletion (Del/+); or 2) the reciprocal duplication (Dup/+) of the 17q21.31 syntenic region; and 3) a heterozygous Kansl1 (Kans1+/-) model. We found altered weight, general activity, social behaviors, object recognition, and fear conditioning memory associated with craniofacial and brain structural changes observed in both Del/+ and Dup/+ animals. By investigating hippocampus function, we showed synaptic transmission defects in Del/+ and Dup/+ mice. Mutant mice with a heterozygous loss-of-function mutation in Kansl1 displayed similar behavioral and anatomical phenotypes compared to Del/+ mice with the exception of sociability phenotypes. Genes controlling chromatin organization, synaptic transmission and neurogenesis were upregulated in the hippocampus of Del/+ and Kansl1+/- animals. Our results demonstrate the implication of KANSL1 in the manifestation of KdVS phenotypes and extend substantially our knowledge about biological processes affected by these mutations. Clear differences in social behavior and gene expression profiles between Del/+ and Kansl1+/- mice suggested potential roles of other genes affected by the 17q21.31 deletion. Together, these novel mouse models provide new genetic tools valuable for the development of therapeutic approaches.
| The 17q21.31 deletion syndrome, also named Koolen-de Vries syndrome (KdVS), is a rare copy number variants associated in humans with intellectual disability, friendly behavior, congenital malformations. The syndrome is caused either by microdeletions in the 17q21.31 region or by variants in the KANSL1 gene in human. The reciprocal 17q21.31 microduplication syndrome is not so well characterized. To investigate the pathophysiology of the syndromes, we studied the deletion, the duplication of the syntenic region and a heterozygous Kansl1 mutant in the mouse. We found affected morphology and cognition, similar to human condition, with genes controlling chromatin organization, synaptic transmission and neurogenesis dysregulated in the hippocampus of KdVS models. In addition we found that synaptic transmission was altered in KdVS mice. Our results demonstrate the implication of KANSL1 in the manifestation of KdVS and extend substantially our knowledge about altered biological processes. Nevertheless, phenotypic differences between deletion and Kansl1+/- models suggested roles of other genes affected by the 17q21.31 deletion.
| The Koolen-de Vries syndrome (KdVS) has a prevalence estimated at 1/55,000 based upon CNV studies[1–3] and was primary described as a consequence of the 17q21.31 microdeletion. Patients with KdVS present characteristic facial dysmorphisms[4] and clinical features including intellectual disability, friendly behavior, hypotonia, and several brain anomalies[5–7]. Microdeletions and microduplications of genomic fragments in the 17q21.3 region ranging from 400 to 800kb have been found in individuals with intellectual disability[6, 8]; these genomic fragments include five protein-coding genes: CRHR1, SPPL2C, MAPT, STH, and KANSL1. The reciprocal duplication is much more rare than the deletion. To our knowledge, only eight patients have been described in the literature[8–12]. The symptoms are heterogeneous and include craniofacial malformations, microcephaly, psychomotor delay, poor verbal and motor skills, and reduced social interaction[8]. Two cases out of eight have been diagnosed with autism spectrum disorder (ASD).
Loss-of-function mutations and atypical deletions restricted to the KANSL1 gene, encoding the KAT8 Regulatory NSL Complex Subunit 1, have been found in several KdVS patients. Interestingly, phenotypic comparison of both the 17q21.31 microdeletion and KANSL1 heterozygous mutation patients show similar clinical severity, implicating that haploinsufficiency of KANSL1 is sufficient to cause the full manifestation of KdVS phenotype[3, 13–15]. KANSL1 is a member of the evolutionarily conserved nonspecific lethal (NSL) complex that controls various cellular functions, including transcription regulation and stem cell identity maintenance and reprogramming[16, 17]. The NSL complex contains the histone acetyltransferase MOF (males absent on the first) encoded by KAT8 which acetylates histone H4 on lysine 16 (H4K16) and with lower efficiency on lysines 5 and 8 (H4K5 and H4K8, respectively) to facilitate transcriptional activation[18, 19]. Recent studies in flies have shown that KANSL1 acts as a scaffold protein interacting with four NSL subunits including WDR5 which plays a critical role in assembling distinct histone-modifying complexes with different epigenetic regulatory functions[20].
Genes within the human 17q21.31 region are highly conserved on mouse chromosome 11E1. Crhr1, Sppl2c, Mapt and Kansl1 orthologs have all been found in the same orientation as in the human H1 haplotype. To investigate the pathophysiology of KdVS and microduplication syndrome, we generated first a mutant mice bearing deletion (Del/+), and duplication (Dup/+) of the 17q21.31-homologous Arf2-Kansl1 genetic interval and looked for phenotypes related to the human condition. We studied behavior, cognition, craniofacial and brain morphology of single Deletion carried compared to wild type and pseudo-disomic (Del/Dup) controls. Then we compared these data to results obtained with mutant mice for Kansl1 and extended our analysis to gene expression. We found a large phenotypic overlap with altered molecular mechanisms controlling the hippocampus synaptic response.
Del/+ and Dup/+ mice were generated on the C57BL/6N (B6N) genetic background (see supplementary information; Fig 1). In comparison with wild-type (wt) littermates, he Del allele frequency was reduced significantly while the transmission of the Dup allele and the Del/Dup carriers were not affected (Table 1), thus demonstrating that lethality is associated with the deletion on the B6N background.
We generated and characterized a compound Del-Dup cohort with littermates carrying four genotypes: Del/+, wt, Del/Dup, and Dup/+. First, we followed general parameters. Compared to wt mice, Dup/+ mice were underweight, whereas Del/+ and Del/Dup mice were not (Fig 1d). At 20 weeks of age. Del/+, Dup/+, and Del/Dup animals showed significantly reduced body length compared to wt littermates (Fig 1e). In comparison with wt, Del/+ littermates showed lower adiposity levels (Fig 1f) at the same age. Nevertheless, we did not detect any notable differences in feeding behavior between mutant and wt animals during a circadian activity test (S1 Fig).
Patients with 17q21.31 CNVs have impaired intellectual and adaptive functioning[4, 5]. As a primary experiment, we looked at the activity and the Del/+ and Dup/+ mice displayed a normal circadian pattern (S1 Fig). However, in comparison with wt, Del/+ mice showed reduced spontaneous locomotor activity (Fig 2a) as well as reduced rearing behavior during the light phase (Fig 2a). In the open field, no differences in exploration, locomotor activity, rearing behavior, or time spent in the center of the area were observed between mutant mice and wt littermates (Fig 2b; S1 Table; S2a–S2d Fig). In the elevated plus maze, Del/+ and Dup/+ mice explored the same number of arms and spent similar periods of time in open arms as those observed for wt (S1 Table). We evaluated motor coordination and learning using the rotarod test but no differences were observed between mutant Del/+ and Dup/+ mice, and wt mice (S2a–S2c Fig). Similarly, for the grip test, no differences in the muscular strength were observed between mutant mice and wt littermates (S2d Fig).
Spatial learning and memory was assessed in the Morris water maze, but similar acquisition and retention were observed in wt and mutant animals (S3 Fig). In the Y maze test, we found similar working memory parameters in wt and mutant animals (S2 Table). Next, we evaluated our models in the novel object recognition test. During the acquisition session, mice of all genotypes spent an equal amount of time exploring the sample object (S2 Table). After 3h of retention delay, Del/+ mice are able to differentiate the novel versus the familiar object but show object discrimination deficits compared to wt whereas Dup/+ and Del/Dup mice showed similar memory capacities (Fig 2c) compared to control.
Associative memory was evaluated with the fear conditioning test. No significant differences in the baseline and post-shock freezing levels were observed between mice of all genotypes (S2 Table). In the context session, Dup/+ mice displayed a higher level of freezing with significant differences in the last 2 min of the test (F(3,56) = 6.399, P < 0.001; Dup/+ vs wt: P = 0.027; Fig 2d). The cued session was performed 5 h after the context session. During the presentation of the conditioning cue, all genotypes demonstrated higher freezing incidence (Fig 2d). During the second 2-min long cue, Del/+ and Dup/+ animals showed lower and higher levels of freezing, respectively, in comparison with wt littermates (H(3, 56) = 20.609, P < 0.001; Del/+ vs wt: P = 0.002, Dup/+ vs wt: P = 0.036). To challenge cued fear extinction in those animals, we used another fear conditioning/extinction protocol with reinforced conditioned stimuli (CS) and long-term follow-up; we separated cohorts for the Del/+ and for Dup/+ with their own littermate controls. This revealed opposite effects on the capacity of Del/+ and Dup/+ animals to extinguish the fear response (Fig 2e). Dup/+ animals presented with higher levels of freezing as compared to their wt littermates suggesting that the fear trace persists in those animals. In contrast, the freezing levels of Del/+ animals were globally lower compared to those of their wt littermates, indicating that the fear memory trace is less stable in those animals.
To determine whether sociability traits were evident in our mouse models, we first used the three-chamber sociability test. Familiar and unfamiliar animals were of similar sex and genetic background than experimental animals but were of younger age in order to avoid aggressiveness. In the first phase of the test, the social interest session, Del/+ mice spent relatively more time than wt littermates exploring the unfamiliar mouse (F(3,51) = 3.447, P = 0.023; Del/+ vs wt: P = 0.017; Fig 2f, Session 1; S4c Fig). In the second phase, the social discrimination session, Del/+ mice interacted with familiar and novel congeners for longer periods compared to wt littermates as observed in the first session (F(3,51) = 6.034, P = 0.001; Del/+ vs wt: P = 0.002; Fig 2f, Session 2; S4d Fig). Nevertheless no differences in the percentage of time spent to explore the novel congener versus the familiar congener were observed between mutant mice and wt littermates (Fig 2e).
We studied the influence of 17q21.31-homologous CNVs on the mouse craniofacial structure. We analysed computed tomography (CT) cranial scans of animal heads combined with 3D reconstruction of skull images using 39 cranial landmarks (S5a Fig). Separate cohorts of Del/+ and Dup/+ females were used for the Euclidean distance matrix analysis[21] and MorphoJ analysis[22]. Skull size in Del/+ (T = 0.115; S5b Fig) and Dup/+ animals (T = 0.115; S5d Fig) was similar to that of wt littermates. The skull shape measurements were nominally altered in Del/+ (Z = 0.077; S5c Fig) and Dup/+ animals (Z = 0.052; S5e Fig) compared to those in wt littermates. Principal component analysis helped to identify change in the skull shape in Del/+ and the Dup/+ versus control littermates with the three main components (PC1 and PC2; S5f Fig) accounting for 59.2% of the total variance. Differences are more pronounced in the skull shape of the Del/+ mice than in wt controls with a predominantly shorter nasal bone and a broadening of the face at the level of the zygomatic spine and squamosal junction.
In addition to neuropsychiatric features, over 50% of patients with 17q21.31 microdeletion also present with various brain structure changes[4, 5, 15]. Furthermore, 50% of patients with the 17q21.31 microdeletion present with microcephaly[8]. To identify potential morphological alterations of brain regions, we analyzed the brain structure of 8 Del/+, 10 wt, 11 Del/Dup, and 8 Dup/+ mice using magnetic resonance imaging (MRI). Overall, we found significant differences in total brain volume between the genotypes (F(3, 33) = 14.14, p < 0.001; Del/+: 458±23 mm3, wt: 448±10 mm3, Del/Dup: 446±12 mm3, Dup/+: 412±13 mm3, brain volumes given as mean±sd). Dup/+ animals showed a globally reduced brain volume in comparison with that of the other genotypes. Using a segmented atlas that divides the brain into 159 separate brain regions[23–25], we examined the 83 structures of at least 1 mm3 in size. A reduction of the whole brain volume was noticed for Dup/+ animals (Fig 3a). Brain structures significantly affected after a correction for multiple testing included the hippocampus, amygdala, nucleus accumbens, cingulate complex, entorhinal cortex, frontal region, and perirhinal cortex. Notably, for the majority of these structures, we observed opposite absolute volume changes in Del/+ and Dup/+ animals in comparison with values determined in wt littermates (Fig 3b). Relative volumes of the discussed regions are represented in the supplementary information (S3 Table).
To explore if genes from the Arf2-Kansl1 region regulate electrophysiological parameters in mouse neurons as suggested by changes in ChIP-seq profiles, we assessed basal synaptic transmission and synaptic plasticity by measuring field excitatory postsynaptic potentials (fEPSPs) in acute hippocampal slices from Del/+ and Dup/+ mice. In Del/+ mutants, we observed decreased fEPSP slopes in mutant slices, especially in response to higher stimulus strengths (S6a Fig). Mean slopes of fEPSPs invoked by the maximum stimulus strength (4.2 V) were significantly smaller in slices from Del/+ mice (1.46 ± 0.09 mV/ms) than wt littermates (1.87 ± 0.09 mV; F(1,13.34) = 8.31; P = 0.025; two-way nested ANOVA, genotype effect). The mean paired-pulse ratio of slopes of fEPSPs evoked at a 50 ms interpulse interval was also significantly lower in mutant slices (S6c Fig; F(1, 11.04) = 6.506; P = 0.027. No significant changes in LTP elicited by theta-burst stimulation were noted in slices from Del/+ mice (S6 Fig).
Basal synaptic strength was slightly enhanced in slices from Dup/+ mice, as fEPSPmax mean slope was nominally higher in slices from Dup/+ mice (2.12 ± 0.09 mV/ms) than in slices from wt littermates (1.89 ± 0.09 mV/ms; S6b Fig). However, the effect did not reach statistical significance (F(1,8.67) = 3.09; P = 0.114; two-way nested ANOVA, genotype effect). Likewise, paired-pulse facilitation (S6d Fig) and LTP were not significantly different in slices from Dupl/+ and litter-matched wt mice (S6f Fig).
We performed similar examinations of mice with heterozygous ablation of Kansl1 (Kansl1+/- mice) in the same B6N genetic background and compared the outcome with the Del/+ phenotypes.). In comparison with wt mice, Kansl1+/- adult animals were underweight (two-way ANOVA genotype effect F(1,30) = 11.729, P = 0.004; Fig 4a) Kansl1+/- mice had a significantly smaller body size (F(1,15) = 11.516, P = 0.004) and lower adiposity level (F(1,15) = 6.813, P = 0.020; Fig 4a) than wt littermates in 20 week old animals. In aggregate, these data indicate many similarities in basic traits between the Kansl1+/- and the Del/+ carriers.
Next, we examined the behavior of Kansl1+/- mice. In a circadian activity test, Kansl1+/- mice displayed normal patterns of activity (S7 Fig). However their baseline locomotor activity levels differed from those of wt littermates during the dark phase (F(1,16) = 8.482, P = 0.010) and the light phase (F(1,16) = 8.573, P = 0.010; Fig 4b). In the novel open field arena, Kansl1+/- mice demonstrated an increased level of rearing behavior (F(1,15) = 4.846, P = 0.044: Fig 4c). To investigate further this hyperactivity, we performed visual observations of animals in odorless home-cages (Fig 4d). Mutant mice displayed more intensive rearing behavior (F(1,15) = 7.207, p = 0.017) and also showed a decreased level of digging behavior (F(1,15) = 12.268, p = 0.003) in comparison with wt littermates. These results indicate a global alteration of Kansl1+/- activity characterized, in particular, by locomotor hypoactivity and vertical hyperactivity. During the learning phase of the rotarod test, Kansl1+/- mice displayed higher levels of motor coordination and learning than wt mice (two-way ANOVA genotype effect F(1,30) = 115.867, P < 0.001; Fig 4e). In the test phase, Kansl1+/- mice showed improvements for speed higher than 10rpm (Fig 4f), a phenotype not observed in the deletion due to lower power of the tests. Recognition memory was assessed in mice by using the novel object recognition task with a retention delay of 3 h. While no difference was observed in the acquisition session (S4 Table), Kansl1+/- mice displayed a significant memory impairment compared to wt during the choice session (F(1,15) = 22.566, P < 0.001; Fig 5a). Then, we evaluated associative memory with the fear conditioning test. No differences in the baseline and post-shock freezing levels were detected between Kansl1+/- mice and wt littermates in the conditioning session (Fig 5b; S4 Table). In the context session, Kansl1+/- mice displayed a lower incidence of freezing than wt littermates (H(1,16) = 6.419, P = 0.011). In the cue session, a decreased freezing level was detected in Kansl1+/- mice during the second 2-min cue period (F(1,16) = 16.748, P < 0.001). Finally, we evaluated animal social behaviors with the three-chamber sociability test and the social interaction test (Fig 5c and 5d; S8 Fig). In both tests, no differences were observed between Kansl1+/- mice and wt littermates.
To identify potential gene expression differences in KdVS models, we carried out epigenetic profiling in the hippocampus, a brain region implicated in learning and memory processes, isolated from 3 Del/+, 3 Kansl1+/- and 6 wt (3+3 matched littermates). We performed ChIP-Seq of H3K4me3 which is a histone mark located in actively expressed genes. As expected, H3K4me3 marks were lower in the Arf2-Kansl1 region with half peak height values for Arf2, Crhr1, Mapt and Kansl1 in the Del/+ samples compared to those observed in wt samples and for the neighboring genes (Fig 6a). Analysis of H3K4me3 tracks (DESeq2, p<0.01) revealed 788 and 751 misregulated promoters in Del/+ and Kansl1+/-, respectively (Fig 6b). Clustering of Pearson correlations, an unbiased method to measure the degree of similarity between large data sets, showed clear segregation between conditions and high concordance of biological replicates (Fig 6c; data are available in S5 to S10 Tables).
Cell-type marker analysis (Fig 6d; see Methods)[26] revealed that up-regulated promoters observed in Del/+ mostly corresponded to genes expressed in neuronal populations (pyramidal neurons and interneurons). Furthermore, Gene Ontology (GO) analysis revealed that they present significant enrichments for the synapse (p<7.9e-20), and dendrite compartments (p<3e-14) as well as synaptic transmission (p<9.1e-24) or neurogenesis processes (p<1.3e-22; S12 and S20 Tables). In contrast, the term oxidoreductase and mitochondrion were found enriched respectively in Del/+ and Kansl1/+ down-regulated promoters using DAVID [27] but no other GO significant enrichments were found for Del/+ (S11 and S19 Tables).
Of the 470 promoters up-regulated in Del/+, 36% (172) were also up-regulated in Kansl1+/-, whereas the two genotypes shared 67% (211) of genes that were down-regulated (Fig 6e). Among all the promoters up-regulated in either of the genetic conditions, we observed 4 distinct clusters of genes (Fig 6f). For cluster 1, 160 genes enriched in non-neuronal populations (astrocytes, ependymocytes, choroid plexus, mural cells, Fig 6g) are up-regulated in the hippocampus of Kansl1 heterozygotes but not of Del/+ mice. Cluster 2 encompassed 214 genes up-regulated in both genetic conditions and enriched in markers of CA1/CA2 pyramidal neurons and astrocytes, while Cluster 3 contained 212 genes enriched in neuronal markers and whose expression was up-regulated to a greater extent in Del/+. Finally, we noted that cluster 4 comprised 162 genes upregulated specifically in Del/+ and expressed in CA1/CA2 pyramidal neurons and astrocytes. Each cluster showed a specific GO enrichment profile (Fig 6h). Several neuronal processes, including synaptic transmission and neurogenesis were overrepresented GO terms in genes from clusters 2, 3 and 4. Cluster 2 genes (up in both models) were also enriched in DNA-packaging and nucleosomes (Fig 6h). In the Del/+ hippocampi, 8 genes involved in social behavior were up-regulated and only 2 of these, Tbx1 and Nr2e1, were also dysregulated in Kansl1+/- mice (Fig 6i). These results suggest a dominance of Del/+ with respect to Kansl1 for determining social behavior.
To explore if genes from the Arf2-Kansl1 region regulate electrophysiological parameters in mouse neurons as suggested by changes in ChIP-seq profiles, we assessed basal synaptic transmission and synaptic plasticity by measuring field excitatory postsynaptic potentials (fEPSPs) in acute hippocampal slices from Del/+ and Dup/+ mice. In Del/+ mutants, we observed decreased fEPSP slopes in mutant slices, especially in response to higher stimulus strengths (S6 Fig). Mean slopes of fEPSPs invoked by the maximum stimulus strength (4.2 V) were significantly smaller in slices from Del/+ mice (1.46 ± 0.09 mV/ms) than wt littermates (1.87 ± 0.09 mV; F(1,13.34) = 8.31; P = 0.025; two-way nested ANOVA, genotype effect). The mean paired-pulse ratio of slopes of fEPSPs evoked at a 50 ms interpulse interval was also significantly lower in mutant slices (S6 Fig; F(1, 11.04) = 6.506; P = 0.027. No significant changes in LTP elicited by theta-burst stimulation were noted in slices from Del/+ mice (S6c–S6e Fig).
Basal synaptic strength was slightly enhanced in slices from Dup/+ mice, as fEPSPmax mean slope was nominally higher in slices from Dup/+ mice (2.12 ± 0.09 mV/ms) than in slices from wt littermates (1.89 ± 0.09 mV/ms). However, the effect did not reach statistical significance (F(1,8.67) = 3.09; P = 0.114; two-way nested ANOVA, genotype effect). Likewise, paired-pulse facilitation (S6 Fig) and LTP were not significantly different in slices from Dupl/+ and litter-matched wt mice (S6 Fig).
In this study, we described the first mouse models of Koolen-de Vries syndrome (KdVS) and 17q21.31 microduplication syndrome. Del/+ mice showed similar phenotypes observed in KdVS patients: higher level of social interaction, lower level of recognition memory, associative learning and memory and brain malformations [3, 5, 15, 28]. We found a single phenotypic similarity between patients carrying the 17q21.31 microduplication and Dup/+ mice, which is microcephaly that has been reported in 50% of the human individuals with this microduplication [8]. Several SNPs associated with risk for Alzheimer’s disease (AD) were identified near MAPT and KANSL1 in humans and they appeared to be correlated with an overexpression of both genes in different brain regions [29]. This observation was further supported by the description of a familial form of late onset AD due to the microduplication of the 17q21.31 region [30]. Thus it would be important now to follow cognition in ageing cohorts carrying the Dup/+ allele generated here as young individuals analysed in the present study do not shown any cognitive impairment, but rather more some improvement in associative memory.
Overall, the phenotypic comparison observed in the Del/+ and Kansl1+/- models confirms the critical importance of KANSL1 in KdVS[3, 15]. Nevertheless, increased social interaction was not found to be affected in Kansl1 haploinsufficient mice whereas it is predominant in humans with KANSL1 mutations [3, 15]. This discrepancy could reflect either different dosage threshold levels in the mouse and human that governs proper social interaction, with the mice needing more change to induce such friendly phenotype. Indeed we have found more genes associated with social behavior misregulated in the Del/+ brain compared to the Kansl1 heterozygotes. An educated guess would be that the haploinsufficiency of another gene(s) from the Arf2-Kansl1 region would contribute to this phenotype in mouse.
The hippocampal epigenetic analysis of 17q21.31 models unraveled several features. Only the mitochondrion term was enriched in the Kansl1/+ down-regulated genes, a situation partially similar to a recent study where KANSL1 and its partner MOF were observed in mitochondria regulating expression of genes involved in oxidative phosphorylation [31]. Interestingly the oxidoreductase term was enriched in Del/+ down-regulated genes. Thus it would be interested to follow mitochondrial activity in the mouse models. Another common set of dysregulated promoters, were largely affecting CA1 neuronal populations and neuronal functioning and includes many genes with long introns. Common up-regulated genes also appear to be implicated in DNA-packaging and nucleosomes, possibly reflecting the outcome of the misregulated KANSL1 activity. Interestingly, several genes (Adcyap1, Cntnap2, Grid1, Nrxn1, Nrxn3, Ucn, Tbx1, and Nr2e1), that are associated with disorders [32, 33], stress response [34], social behaviors [35–37] or autism spectrum disorders [38–42], are up-regulated specifically in the hippocampus of Del/+ mice. Deregulation of these genes may be a molecular underpinning of the friendly/amiable affect of 17q21.31 deletion patients. Expression of the majority of those genes (except for Txb1 and Nr2e1) was not altered in Kansl1+/- mice. We also emphasize that two overexpressed genes, Ucn and Adcyap1, and one underexpressed gene, Chd1l, found deregulated in Del/+ mice are linked to corticotropin release and are associated with stress response. Those genes may be relevant for the overly friendly social phenotypes observed in 17q21.31 deletion carriers.
Electrophysiological experiments confirmed that dosage of one or several genes within the 17q21.31 syntenic region affects basal synaptic transmission and short-term plasticity of excitatory synapses in the hippocampus. Noted disturbances in the expression level of several genes could contribute to this impairment. For example, dysregulation of Cntnap2 could affect migration of interneurons [37] and inhibitory synaptic function [43], which could, in turn, alter excitatory synaptic responses. Other gene affected by 17q21.31 mutations is Nrxn1 that shapes the balance between excitatory and inhibitory synaptic activity [44, 45]. Such a defect at the expression level may account for the change in synaptic strength and impaired learning and memory.
In conclusion, this study confirms a previously hypothesized role of KANSL1 in the manifestation of KdVS phenotypes and extends substantially our knowledge about biological processes affected by these mutations. With these new genetic tools, we can explore the function of these genes and dissect further the pathophysiological mechanisms to eventually inform potential therapeutic avenues.
The 17p21.31 mutant mice carrying the deletion of the Arf2–Kansl1 region (noted Del/+), or the reciproqual duplication (noted Dup/+), were generated on the C57BL/6N genetic background (see Supplementary information). The Kansl1+/- mutant mice were derived in a C57BL/6N genetic background from the unique IKMP ES cell clone HEPD0766_8_G02. Kansl1tm1b(EUCOMM)Hmgu[46] animals were obtained by breeding Kansl1tm1a(EUCOMM)Hmgu/+ mice with animals expressing the Cre recombinase[47] to generate the Kansl1tm1b(EUCOMM)Hmgu/+ (noted here Kansl1+/-). The local ethics committee, Com’Eth (n°17), approved all mouse experimental procedures, under the accreditation number 2012–069.
Behavioral studies were conducted in 12-20-week old animals. All assessments were scored blind to the genotype as recommended by the ARRIVE guidelines[48, 49]. All the experimental procedures for behavioral assessments have been described[50, 51] and are detailed in the supplementary information.
Craniofacial phenotyping is described in the supplementary data. Magnetic resonance imaging (MRI) was used to identify alterations of brain regions in 17q21.31-homologous CNV mice (8 Del/+, 10 wt, 11 Del/Dup, and 8 Dup/+ mice). MRI scans were acquired from 41 male mice at 43 weeks of age with specimens prepared as described[52] and detailed with the image processing in the supplementary information.
Acute hippocampal slices were used to record field excitatory post synaptic potentials (fEPSPs), by using an electrophysiological suite of 8 MEA60 set-ups consisting of a MEA1060-BC pre-amplifier and a filter amplifier (gain 550×) (Multi Channel Systems, Reutlingen, Germany) as described[50, 53]. All experiments were performed using two-pathway stimulation of the Schaffer collateral/commissural fibers in the CA1 area of 350-μm thick hippocampal slices (see supplementary information).
Adult hippocampi from Del/+, Kansl1+/- and wt mice were dissected and snap-frozen in liquid nitrogen. Tissue samples were ground in a liquid-nitrogen chilled mortar and the resulting powder was used for ChIP. Chipping for H3K4me3 (Diagenode A5051-001P) was performed as in (www.blueprint-epigenome.eu/UserFiles/file/Protocols/Histone_ChIP_July2014.pdf). Libraries were synthetized with KAPA Hyper prep kit (KK8504) following the manufacturer’s instructions. The libraries were pooled (4/lane) and sequenced on the illumina HiSeq. Libraries were mapped with BWA (0.6.2). Peaks were called with a custom C++ script and DEseq2 (R+) was used to perform statistical comparisons. Data are deposited in GEO under accession GSE80311. All enrichment analyses are made from standard hypergeometric tests with Benjamini or Bonferroni correction. GO annotations are updated to 25/6/2015.
ChIP-seq data and ChIP-seq supplementary tables were deposited in GEO and available at the link https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=cnavacaedzgztgz&acc=GSE80311. Cell-types enrichments are based on the single-cell RNAseq data from Amit Zeisel et al., 2015, “cell types in the mouse cortex and hippocampus revealed by single-cell rna-seq” [26]. In this work, single-cell RNAseq was used to measure trascriptomes of >3000 single cells, allowing to define the markers of 11 different cell types in adult (P21-P30) hippocampus and somatosensory cortex. Given that also our ChIP-seq is done on adult (P30) hippocampus, single-cell RNAseq data from Zeisel et al. becomes a highly valuable resource to gain insight at the cell type level. Here we performed standard hypergeometric tests with bonferroni correction against the cell-type markers derived from data of Zeisel et al. in order to evaluate the abundance of specific markers in deregulated gene sets. A significant enrichment (p<0.01) means that a high amount of markers of a specific cell type is found in Kansl1+/- or Del/+ deregulated genes, suggesting that the latter cell type should be particularly affected. The complete statistical data and the lists of markers found in de-regulated genes are fully available in S5 to S20 Tables.
All acquired behavioral data were analyzed using a one-way ANOVA analysis with a post-hoc Tukey’s test, when applicable, non-parametric Kruskal-Wallis test or Mann-Whitney U test using the Sigma Plot software (Ritme, France). The Pearson’s chi-squared test was used for mutant allele transmission. Data are represented as the mean ± s.e.m. and differences were considered to be significant if P < 0.05. When comparing freezing levels between wt, Del/+, and Dup/+ animals over different time points during extinction we used the two-Way ANOVA Repeated Measures statistical test followed by Holm-Sidak post-hoc tests to evaluate for interactions between the groups. Otherwise, when comparing wt data with those obtained from respective Del/+ or Dup/+ animals for a single time point, we used Student’s t-test. When data did not follow a normal distribution, we used the Mann-Whitney rank-based statistical test.
In electrophysiological experiments input-output relationships were compared initially by a mixed model repeated-measures ANOVA and Bonferroni post hoc test implemented in Prism 5 (GraphPad Software, San Diego, USA) using individual slice data as independent observations. Since several slices were routinely recorded from every mouse, fEPSPmax, PPF and LTP values between wt and mutant mice were compared using two-way nested ANOVA design with genotype (group) and mice (sub-group) as fixed and random factors respectively (STATISTICA v.10, StatSoft, USA). DF error was computed using the Satterthwaite’s method and main genotype effect was considered significant if P < 0.05. Graph plots and normalization were performed using OriginPro 8.5 (OriginLab, Northampton, USA). Electrophysiological data are presented as the mean ± s.e.m. with n and N indicating number of slices and mice respectively.
|
10.1371/journal.pgen.1008027 | Genetic signatures of gene flow and malaria-driven natural selection in sub-Saharan populations of the "endemic Burkitt Lymphoma belt" | Populations in sub-Saharan Africa have historically been exposed to intense selection from chronic infection with falciparum malaria. Interestingly, populations with the highest malaria intensity can be identified by the increased occurrence of endemic Burkitt Lymphoma (eBL), a pediatric cancer that affects populations with intense malaria exposure, in the so called “eBL belt” in sub-Saharan Africa. However, the effects of intense malaria exposure and sub-Saharan populations’ genetic histories remain poorly explored. To determine if historical migrations and intense malaria exposure have shaped the genetic composition of the eBL belt populations, we genotyped ~4.3 million SNPs in 1,708 individuals from Ghana and Northern Uganda, located on opposite sides of eBL belt and with ≥ 7 months/year of intense malaria exposure and published evidence of high incidence of BL. Among 35 Ghanaian tribes, we showed a predominantly West-Central African ancestry and genomic footprints of gene flow from Gambian and East African populations. In Uganda, the North West population showed a predominantly Nilotic ancestry, and the North Central population was a mixture of Nilotic and Southern Bantu ancestry, while the Southwest Ugandan population showed a predominant Southern Bantu ancestry. Our results support the hypothesis of diverse ancestral origins of the Ugandan, Kenyan and Tanzanian Great Lakes African populations, reflecting a confluence of Nilotic, Cushitic and Bantu migrations in the last 3000 years. Natural selection analyses suggest, for the first time, a strong positive selection signal in the ATP2B4 gene (rs10900588) in Northern Ugandan populations. These findings provide important baseline genomic data to facilitate disease association studies, including of eBL, in eBL belt populations.
| We present a genome-wide analyses of genetic structure, gene flow, and natural selection in Ghana and Northern Uganda populations, both residing in the Sub-Saharan eBL belt, a region with intense falciparum malaria transmission and high endemic Burkitt Lymphoma (eBL) incidence. These populations are from different ethnolinguistic groups and are located 2400 miles apart in sub-Saharan Africa. We characterized genetic composition of these populations in the context of 22 additional African populations and present evidence for gene flow events that occurred in the last 3000 years, possibly related to regional migrations in Western Africa and major migrations involving Nilotic, Cushitic, and Bantu groups. We identified in Northern Ugandans a strong signal of malaria-driven selection in the ATP2B4 gene coding for a calcium transporter expressed in erythrocytes. Characterization of biological relationships between the ATP2B4 gene and malaria may inform the investigation of complex genomic disease associations in eBL belt populations.
| The endemic Burkitt Lymphoma (eBL) belt is a geographic area spanning 10°N-10°S and altitudes below 1500m above sea level (Fig 1A) in sub-Saharan Africa, where there is a high geographical correlation between malaria and eBL (an aggressive pediatric B-cell non-Hodgkin lymphoma). This correlation has led to the identification of malaria infection as a major driver of eBL [1][2], which was confirmed by the evidence that the sickle cell trait that protects against severe malaria [3] also protects against eBL [4]. Because eBL occurs in areas of sub-Saharan Africa [5] with stable intense Plasmodium falciparum (Pf) malaria (for 7–12 months in the year), eBL burden provides a novel way to identify populations under strong malaria selective pressure. Pf malaria is one of the most important selective pressures that have shaped the African genetic diversity [6], but there are limited reports on the combined effects of malaria-related natural selection and the demographic history of populations in the eBL belt.
The eBL belt was the scenario of several human migrations over the last 3000 years and archaeological and linguistic evidence have described the following historical events: in West Africa: (i) interaction between West and West-Central Africa [7], (ii) cultural interaction between the local kingdoms of West-Central Africa [8,9], and (iii) migrations across the Sahel that include the westward Nilotic expansion [7,10,11]. In East Africa: (iv) Eastern Cushitics migrated from the Horn of Africa [7] into the Great Lakes region ~3000 years ago, maintaining (v) interactions with Nilotic groups that migrated from Southern Sudan [10,11], and subsequently, (vi) with Bantu speakers from West-Central Africa who reached the Great Lakes region ~2000 years ago [7,12–15]. Moreover, malaria imposed an important evolutionary pressure well known for its effect on the genetic structure of affected populations, such as those that settled in the eBL belt.
Datasets representing African populations, such as those included in the 1000 Genomes Project [16], the African Genome Variation Project [17], the Tishkoff laboratory [18][11] and the H3Africa initiative [19][20], have provided an important baseline for genomic studies in Africa. However, due to the high genetic diversity among African populations, reference datasets should closely match populations in which specific scientific questions are explored. For example, the Nilotics in the Great Lakes region on Northern Uganda region, which experience high malaria intensity [21] and high eBL burden (S1 Table), have not been included in previous genomic studies [18].
To determine if the historical migrations described above (i-vi) and intense exposure to malaria have shaped the genetic composition in the eBL belt, we analyzed a new dataset of 945 Ghanaians and 568 Northern Ugandans in whom ~4.3 million single nucleotide polymorphisms (SNPs) were genotyped. These sub-Saharan Africa populations reside on opposite longitudes of the eBL belt (2400 miles apart) (Fig 1A), and are both exposed to high malaria pressure and have published evidence indicating a high eBL burden (S1 Table) [22].
Details of the study populations are given in S2 and S3 Tables. Briefly, the Ghanaian population included approximately 35 tribes, predominantly from the Kwa and Gur Niger-Congo language families (S2 Table). The Ugandan populations included approximately 17 tribes, predominantly of the Western Nilo-Saharan (Nilotic) language family (S3 Table). Because the Ugandan populations were recruited from opposite sides of the deep gorge of the East African Rift Valley, through which the Albertine Nile flows (Fig 1A) and this is a potential physical barrier to gene flow, we designated the populations descriptively as Uganda North West (UNW) for those recruited from the west side of the gorge and Uganda North Central (UNC) for those recruited from the east side of the gorge. We estimated the level of genetic relatedness of our dataset and excluded closely related individuals that may affect population-structure and natural selection analyses [23] (S1 Text and S2–S5 Figs).
Population structure was evaluated using a Pan-African genome-wide dataset (PA dataset, Methods) that included 1.3M SNPs genotyped in 3,102 individuals, including 1,513 from the combined UNW, UNC, and National Cancer Institute (NCI) Ghana datasets, and 1,589 from 22 additional African populations [17,24,25] (S1 Table and S1 Fig). This Pan-African dataset is comprised of populations from five broad geographical regions: West Africa, West-Central Africa, Great Lakes Africa, Horn of Africa, and Southern Africa (Fig 1A and S1 Table). Specifically, the West African region includes Gambian and Ghanaian tribes [17], and the West-Central African region includes Nigerian tribes (Yoruba and Igbo). The Great Lakes African region includes our Northern Ugandan (UNW and UNC) populations and also Southwest Ugandan, Kenyan and Tanzanian populations.
Although our NCI Ghana set included individuals from approximately 35 tribes, ADMIXTURE results showed a homogeneous ancestry pattern (91% of the blue genomic ancestry Fig 1C and S6, S7, S10 and S12 Figs), similar to the Ga-Adangbe tribe, with the blue genomic ancestry being predominant in West-Central Africa (Fig 1C and S6–S8 Figs). We observed similar ancestry composition of Ghanaians and Nigerians, who both share predominant West-Central Africa ancestry (blue). In accordance with their more Western location, Ghanaians shared a minor proportion of West African ancestry (red genomic ancestry in Fig 1C) related to Gambian tribes, while the Yoruba and Igbo shared a minor ancestry proportion (purple, Fig 1C) related to Eastern Bantu populations from the Great Lakes Africa region. This pattern of ancestry in Yoruba and Igbo has been seen in recent studies [17, 26–28]. Our Ghanaian population showed negligible Eurasian admixture (S9 Fig) with mean Eurasian ancestry of 0.4%.
Consistent with ADMIXTURE inferences, both GLOBETROTTER analysis and the three-population test (ƒ3 statistic) inferred episodes of gene flow from Gambian tribes, and also from Nilotics, to Ghana and Nigeria that occurred during the last 4000 years (Fig 2, S13 Fig and S4 Table). The pattern of genetic structure in Ghanaians and Nigerians, and the inferred episodes of gene flow into West-Central Africa show that historical cultural exchanges between West and West-Central Africa [8, 9] and migrations across the Sahel (historical events i-iii of the Introduction) involving populations from East Africa have shaped the genetic composition of West-Central African populations.
The main feature of the genetic structure of Uganda shown by ADMIXTURE and PCA is the dichotomy between Northern Uganda populations, that show a predominantly Nilotic genomic ancestry (cyan ancestry in Figs 1 and 2 and S6–S9 Figs), and Southwest Uganda populations that have predominantly Eastern Bantu ancestry (purple ancestry, in Figs 1 and 2 and S6–S9 and S12 Figs). Within the predominantly Nilotic Northern Uganda populations, the UNW population is more homogeneous (93% Nilotic ancestry, Fig 1B and 1C and S7 Fig), while the UNC population is a mixture of Nilotic (64%) and Eastern Bantu genomic ancestry (Fig 1B and 1C and S7 and S12 Figs). Interestingly, Nilotic ancestry was detected in all Great Lakes African populations (Fig 1C and S7 Fig). In general, ADMIXTURE and PCA showed that the Great Lakes African region, which includes populations from Uganda, Kenya and Tanzania, was the most ancestry diverse region in sub-Saharan Africa (Fig 1B and 1C). Our Ugandan populations showed negligible Eurasian admixture (S9 Fig) with mean Eurasian ancestry of 0.02% in UNC and 0.015% in UNW.
GLOBETROTTER inferences suggest an episode of gene flow from West/West-Central Africa into UNW (849–936 years before present (YBP), 95% confidence interval, S13 Fig), although this was not confirmed by ƒ3 statistics (Fig 2 and S4 Table). In contrast to UNW, both ƒ3 (Fig 2 and S4 Table) and GLOBETROTTER (S13 Fig) consistently inferred several episodes of gene flow into the UNC and Southwest Uganda populations (Baganda, Barundi and Banyarwanda) from different sources: UNW (Nilotic), Southern Bantu, Horn of Africa (Cushitic), and also from West/West-Central African populations. GLOBETROTTER dates for these gene flow events (397–484 and 1499–2659 YBP, 95% confidence interval, S13 Fig) suggest two gene flow events that do not overlap with the inferred gene flow event into the UNW. We also inferred Nilotic-related (UNW and UNC) gene flow into Southwest Ugandan (Banyarwanda), Kenyan (Kikuyu and Kalenjin), Horn of Africa, West and West-Central African populations. Taken together, these results show that historical migrations (events iv-vi of the Introduction) of several human groups (Nilotic, Bantu and Cushitic) have shaped the current genetic composition in the Great Lakes region, and that Nilotic westward migration was accompanied by gene flow (historical event iii of the Introduction) (Fig 2 and S4 Table).
While our studied populations (Ghana and Northern Uganda) share a high incidence of malaria and eBL burden (S1 Table and S1 Fig) [22], our population structure analyses showed that they have distinct patterns of genetic ancestry (Fig 1). In order to understand if they share common signals of natural selection despite their differential genetic history, we searched for genomic signatures of natural selection in Ghana and Northern Uganda populations. The eBL cases were excluded from this analysis to eliminate confounding of natural selection results with disease associations. We applied the population branch statistic (PBS) approach [29] to each of these as a focal population, using the Southern Bantu Sotho and Zulu populations as a sister group and Europeans as the reference population (S15 Fig and see Methods). We used Southern Bantu populations as a sister group because, after the Bantu expansion in the last 2000 years, they have occupied an area outside the eBL belt, where the climate is drier and cooler, and thus not conducive for malaria transmission [30], also supported by a low reported frequency of malaria-associated variants [31]. We compared the PBS outlier values (99.9th percentiles) against those generated by simulations of plausible neutral demographic models (Methods and S15–S17 Figs). In addition to the PBS statistic, we performed cross-population haplotype-based approach (xpEHH) to identify genomic regions under positive selection. We report as candidate selection regions those that showed extreme signal in both PBS and the xpEHH approach (above the 99.9th percentiles for PBS and >2 for xpEHH).
We observed 14, 12 and 11 candidate genomic regions in the Ghanaian, UNW and UNC populations, respectively, (Fig 3, S16 Fig, and S5–S7 Tables), nominated by 32 index SNPs. While the Ghanaian sample yielded the largest number of candidate genomic regions, none of them were significant in the demographic model performed (S5 Table). Of 32 candidate genomic regions, seven are found within/adjacent the same gene and shared between two populations: RARB found in Ghana and in UNC (different index SNPs), and six genomic regions within/adjacent to KLHL20, ATP2B4, NIT2, TENM3, GPHN and HERC2, are found in both UNW and UNC (five of six regions share the same index SNP) (S5–S7 Tables).
The extreme PBS values came from the genomic region at the ATP2B4 gene in UNW (p-value = 0.0011) and UNC (p-value = 0.0021) (Fig 3A), but not in Ghana or other eBL belt populations evaluated (S5 and S8 Tables). Analysis using the xpEHH statistic (based on the pattern of extended haplotype homozygosity (EHH) between populations) corroborates the PBS signal in ATP2B4 gene for both UNC and UNW (Fig 3 and S6–S8 Tables). ATP2B4 encodes the plasma membrane Ca2+-ATPase type 4 protein (PMCA4), the main calcium pump of the human erythrocyte [32]. Six SNPs within the genomic region in the ATP2B4 gene (rs11240734-C, rs1541252-T, rs1419114-A, rs10900588-G, rs3851298-T, rs2228445-T) were detected as PBS outliers in both UNW and UNC. These six SNPs are located within two adjacent linkage disequilibrium (LD, r2 = 0.82) blocks of 6 and 12 Kb (S18 Fig). The intronic SNP rs10900588-G derived allele exhibited the highest PBS values in both Northern Uganda populations (Fig 3A and 3B). This SNP is within a core haplotype observed with high frequency in both Northern Uganda populations (UNW and UNC) and much lower frequency in the South Bantu Sotho and Zulu populations (Fig 3C and S18 Fig). Consistently, the highest frequencies in Africa of the rs10900588-G were observed in UNW (0.72), followed by UNC (0.63), and the lowest frequencies in the Horn of Africa (0.064–0.096), followed by Fula (0.22), Zulu (0.23) and Sotho (0.24) (Fig 3C).
The other five shared signals of candidate selection in Northern Ugandans (UNW and UNC) for the following genes: KLHL20 (p-values 0.0021 UNW, and 0.0043 UNC), NIT2 (p-values 0.0027 UNW and 0.0035 UNC), TENM3 (p-values 0.0022 UNW and 0.0041 UNC), GPHN (p-values 0.0036 UNW and 0.0018 UNC) and HERC2 (p-values 0.0016 UNW and 0.0031 UNC). None of these genes have clear relationship with malaria pressure and are likely related to other selective pressures in Northern Uganda, which are not explored in the current study.
Our study highlighted how the combined effects of demographic history and likely malaria-driven natural selection have shaped the genetic structure of populations in the eBL belt. We found evidence of gene flow events across the eBL belt in the last 3000 years, possibly related to regional migrations in Western Africa and major migrations involving Nilotic, Cushitic, and Bantu groups. Importantly, we identified for the first time in Africa a Northern Uganda-specific strong signal of malaria-driven selection in the ATP2B4 gene.
Our results showed that historical migrations (denoted as i-vi in the Introduction) have left signals in the genome of eBL belt populations. The historical interactions of diverse linguistic groups (pastoral Nilotic, Cushitic and farming Bantu) along lush migratory corridors in the Lake Victoria basin plateau [33] is reflected in the current genomic composition of Uganda, Kenya and Tanzania populations (Figs 1 and 2 and S13 Fig). In the context of the six historical events highlighted in the Introduction (i-vi), the observed pattern of genetic structure is consistent with Nilotic dispersion southward into the Great Lakes region (event v, Figs 1C and 2, and S7 Fig) and westward across the Sahel region (event iii), which may have led to historical contacts with West African populations [11,26,34,35]. Our results showed that Nilotic influence extends into the Great Lakes Africa region, and also to the Western African region, likely in the last 2000 years, as suggested by our GLOBETROTTER inferred dates (Fig 2,S7 and S13 Figs and S4 Table). The dichotomous pattern of ancestry between Northern Uganda (predominantly Nilotic) and Southern Uganda (predominantly Bantu) probably reflects the influence of the Nilotic migration into Northern Uganda, in contrast with the Bantu migration into Southern Uganda.
Our three dozen Ghanaian tribes showed high genetic homogeneity, but also evidence of gene flow from Gambian tribes in West Africa. Historically, the West and West-Central African regions have experienced extensive interactions between local kingdoms and tribes in the last 2000 years [8,9]. For Ghanaians, these interactions led some tribes to change their language due to social or economic motivation [7]. Local historical interactions such as these could explain the observed homogeneous genetic ancestry in Ghanaians. We inferred one gene flow event from Gambian tribes into Ghana and Yoruba about 1337–3022 YBP (S13 Fig). These inferred episodes of gene flow may be the signature of Mande migration into Ghana as part of trading networks [36], as well as of interactions of ancient populations along salt, gold, and slave trade routes [7,13].
To search for natural selection driven by malaria in the eBL belt, we used eBL burden as an indicator of populations exposed to high sustained falciparum malaria transmission in the eBL belt (Fig 1, S1 Fig and S1 Table). By comparing the populations with the highest malaria pressures versus those with no malaria, we identified for the first time a candidate region for malaria-driven selection in the ATP2B4 gene in African, specifically Northern Ugandan populations (Fig 3 and S6–S8 Tables). ATP2B4 is ubiquitously expressed in human tissues, and encodes the plasma membrane Ca2+-ATPase type 4 protein (PMCA4) [37], which is the most commonly expressed Ca2+ transporter in human erythrocytes [32]. We note that seven ATP2B4 intronic SNPs (not present in our data) have been reported to be associated with multiple blood cell-related traits in African American, East Asian, European and Hispanic populations: mean corpuscular volume and hemoglobin concentration [38–41], lymphocyte counts, and red cell distribution width [38]. Furthermore, five of these seven ATP2B4 SNPs (minor frequency alleles: rs10900585-G, rs2365860-C, rs10900589-A, rs2365858-G and rs4951074-A) were associated with resistance against severe falciparum malaria in Western African populations in Ghana and Gambia [42], and rs10900585 has been associated with reduced malarial placental infection and related maternal anemia in Ghana [43]. In an analysis of 11,890 cases of severe falciparum malaria and 17,441 controls from Africa, Asia and Oceania of 55 previously identified SNPs, rs10900585 was significantly associated with severe malaria over all African sites combined, and in the Ghanaian and Gambian samples [44]. Importantly, the protective minor alleles of these five SNPs above (not present in our data) are highly linked (mean r2 = 0.94) with the minor allele (as defined in non-Nilotic populations) of our strongest signal of selection (ATP2B4 rs10900588-G) in the Luhya population (LWK) from the 1000 Genomes Project.
While polymorphisms in the ATP2B4 gene were described as protective against severe malaria in Ghana and Gambia [42], the outlier approach used in the present study did not identify ATP2B4 as a candidate selection gene in Ghanaians and Nigerians (S8 Table). This result is in accordance with the absence of natural selection signals in the ATP2B4 gene reported for previous studies using samples from Western Africa [17, 45–49]. The lack of concordance between association studies and natural selection analysis can be explained by the fact that the frequency of the protective haplotype observed in Ghanaians is sufficient to identify significant disease association (a 6% difference between cases and controls across the protective haplotype) [42], but not sufficient to identify significant positive selection signal (an average 14% difference between West Central African and South African populations, compared to an average 45% difference between Northern Ugandan and South African populations, at rs10900588). In addition, when analyzing the ATP2B4 association studies with cerebral malaria and severe malaria anemia in African, Asian and Oceanian populations, the Malaria Genomic Epidemiology Network [44] noted that the effect of the ATP2B4 ancestral allele rs10900585-G on malaria might be heterogeneous across phenotypes and/or populations. The heterogeneity of effects may indicate presence of biological variation due to epistasis, gene-environment interactions, or that the analyzed SNP is in LD with an unknown causal allele associated with resistance to malaria. As LD patterns vary among populations, replication of the association would only be feasible if the causal SNP were genotyped.
The highest worldwide frequency of rs10900588-G allele and its related core haplotype observed in Northern Uganda populations (UNW and UNC, Fig 3 and S18 Fig) suggests a Northern Uganda- or Nilotic-specific selection in the ATP2B4 gene, although the reasons for specificity are currently unclear to us. Consistent with this, our natural selection analyses using neighboring populations in Southern Uganda and Kenya did not identify signal of selection in the ATP2B4 gene (S8 Table). The most likely explanation for this Northern Uganda-specific selection is that this region has historically experienced one of the highest levels of malaria infection worldwide (400–1,500 infectious mosquito bites per capita per year) [21]. A previous report has identified a signal of malaria-driven natural selection, at rs10900585 in the ATP2B4 gene, by estimating the population-scaled selection coefficient in a time series of allele frequencies [50] in 92 ancient European samples from the Bronze Age (5000 bp) to the Post-Roman era [51], suggesting an ancient role of ATP2B4 in malaria-driven selection.
The biological relationship between ATP2B4 and malaria resistance is mediated by polymorphisms in ATP2B4 changing PMCA4 structure or expression, which leads to a homeostatic disruption of intra-erythrocytic Ca2+ levels that are critical to the development of the Plasmodium parasite [42]. In an expression quantitative trait locus (eQTL) meta-analysis of whole blood gene expression [52], the allele rs10900588-G and linked SNPs were described as significant cis-eQTLs of ATP2B4 (rs10900588-G with Z = -7.30, p-value = 2.91E-13, FDR = 0.00), i.e., the minor allele rs10900588-G is associated with significantly reduced ATP2B4 expression. Recently, in a search for eQTLs enriched in human erythroblasts, Lessard et al. identified an erythroid-specific enhancer region just proximal to exon 2/alternate exon 1 of ATP2B4 [53]. Lessard et al. demonstrated functional effects of the enhancer region through genome editing and in vitro cell culture, suggesting a Ca2+ homoeostasis defect as one possible pathway for the ATP2B4 associations with malaria. The core haplotype we defined in the Northern Ugandan population extends from just proximal to exon 2/alternative exon 1 into intron 2/alternative exon 1. This haplotype overlaps with a minor ATP2B4 haplotype in a European population (defined by the minor alleles in non-Nilotic populations of rs1541252, rs1541253, rs377342347, rs1419114, rs2228445, with mean r2 = 0.96 with rs10900588 in the LWK population) that results in reduced erythrocyte PMC4A expression and reduced Ca2+ export [54]. Both Lessard et al. and Zámbó et al. have suggested mechanisms by which reduced Ca2+ export may be related to reductions in malaria risk: Lessard et al. suggests erythrocyte dehydration as a resistance factor, while Zámbó et al. suggests that reduced Ca2+ export into the invaginated extracellular membrane reduces Ca2+ concentration, which is required for Pfa maturation. Supporting the suggested mechanism, the most recent report [55] showed a significant association between low falciparum malaria parasitemia and the homozygous genotype for the ATP2B4 rs1541255-G allele (not present in our data). Importantly, this allele is in perfect LD (R2 and D’ = 1) with our most important ATP2B4 signal (rs10900588-G) in Kenya.
There are extensive reports in the literature regarding selection pressure driven by malaria in the HBB, ABO, DARC and G6PD genes [44, 56]. It should be noted that, in the present study, the tests used for the detection of positive selection are based on assumptions such as high differentiation between populations (PBS) and hard selective sweeps (xpEHH). Therefore, it is important to emphasize that this is not the case for HBB and ABO, that are evolving under a balancing selection regime [56], nor is this the case for DARC, that despite being under positive selection, is almost fixed and with low differentiation among African populations [57]. Also, as we did not examine the X chromosome, G6PD, found on the X chromosome, was not investigated in the present study.
Although malaria is the presumed major driver of natural selection in the eBL belt populations (S1 Table), we understand that other selection pressures, which were not investigated in our study, might be acting on our study populations. For example, we found significant signal of selection in Northern Ugandans for the OCA2/HERC2 and NIT2 genes (Fig 3, and S6 and S7 Tables). The first is significantly associated with skin, eyes and hair pigmentation [18] and the latter is a potential tumor suppressor [58].
After characterizing the genetic structure of the Ghanaian and Ugandan populations in the eBL belt, we showed that (i) historical interaction between West and West-Central Africa involved episodes of gene flow from West to West-Central Africa; (ii) the documented cultural interaction between the local kingdoms of West-Central Africa, specifically in Ghana, were accompanied by an homogenization of the gene pool of these populations, independently of their linguistic diversity; (iii) the pattern of genetic diversity of the eBL belt populations show the signature of migrations across the Sahel that include Nilotic expansion into West Africa; (iv) the genetic composition of Great Lakes African populations is the result of the interactions between Nilotics, Cushitics and Bantu groups in the last 3000 years; and, (v) the ATP2B4 gene, which was previously associated with erythroid-related traits and malaria susceptibility, shows the signature of malaria-driven natural selection specific to Northern Uganda (UNW and UNC). These results provide important baseline genomic data to facilitate disease association studies, including of eBL, in eBL belt populations.
Ethical approval for EMBLEM was obtained from the Uganda Virus Research Institute Research and Ethics Committee, the Uganda National Council for Science and Technology (H816), and the NCI Special Studies Institutional Review Boards (10-C-N133). The Ghana Prostate Health Survey was approved by the Noguchi Memorial Institute for Medical Research Institutional Review Board (001/01-02) and by the NCI SSIRB (02CN240). Participants in both the EMBLEM and Ghana Prostate Healthy Study gave informed written consent.
The NCI Ghana set included random samples of 964 healthy men from approximately 35 tribes (S2 Table) aged 50–74 years old enrolled for prostate cancer screening into the Prostate Healthy Survey [59]. The Ugandan samples were from 758 children aged 0–15 years old (including 197 eBL cases and 561 controls) from 13 tribes enrolled in the Epidemiology of Burkitt Lymphoma in East-African Children and Minors (EMBLEM) study in two regions of Northern Uganda (Uganda North West [UNW] and Uganda North Central [UNC]). The healthy children were enrolled from 100 randomly selected villages in these regions (S3 Table) [60]. The samples were genotyped using the Illumina Infinium HumanOmni5-4v1 genotyping array in the Cancer Genomics Research Laboratory (CGR) at the National Cancer Institute (NCI); quality control was performed using PLINK 1.07 software [61] and in-house scripts [62].
We calculated the inbreeding (F) and the kinship coefficients (Φij) using the PLINK 1.07 software [61] (S2 and S3 Figs). Following Kehdy et al. [24] a Φij threshold ≥ 0.1 was used to create family networks (S2 and S3 Figs) and we excluded interactively individuals with the highest number of relatives, which allow us to reduce family structure, minimizing sample loss. Following this procedure, we created “unrelated” NCI Ghana and Ugandan datasets (S1 Table).
We merged the NCI datasets (1,513 individuals with >48 tribal affiliations) with public African genome-wide datasets, creating a Pan-African dataset (PA dataset) of 1,287,642 SNPs for 3102 individuals, from 9 countries, and 11 ethnolinguistic groups in Sub-Saharan Africa (S1, S2 and S3 Tables). We also merged the PA dataset with all 1000 Genomes Project Phase 3 populations [24] creating the PA1KGP dataset, to test the extent of Eurasian admixture in the NCI datasets.
Since ADMIXTURE software [63] assumes independence among genetic markers, we used PLINK 1.07 to prune the SNPs in high linkage disequilibrium (LD) using a pairwise linkage disequilibrium maximum threshold of 0.4, a window size of 50, and a shift step of 10, creating the PA non-LD dataset with 727,834 SNPs. Then, we used the PA non-LD dataset to perform ADMIXTURE [63] and Principal Components Analysis (PCA) [64]. To verify possible sample size effects on ADMIXTURE and PCA analysis [65], we resampled the PA non-LD dataset to reach similar number of individuals for each studied population (S8 and S11 Figs).
We phased the PA dataset using SHAPEIT [66]. Using the phased dataset, we performed fineSTRUCTURE [67] analysis (10 million iterations of Markov chain Monte Carlo) to determine the genetically homogeneous groups and GLOBETROTTER [68] to infer historical admixture events.
We also estimated the ƒ3 statistic to infer events of gene flow and their possible directions, as implemented in the software ADMIXTOOLS [69], for all possible combinations of three populations using the PA dataset. All ƒ3 statistics with Z-score ≤ -3 were considered as highly significant evidence of gene flow. For the ƒ3 statistic and GLOBETROTTER analysis of historical gene flow events, we described contributing ethnic groups or populations with the suffix “-like”, representing present day surrogates of the real sources [67]. Masterscripts used for data curation and population structure analyses are available at the EPIGEN-Scientific Workflow (http://ldgh.com.br/scientificworkflow/, [62]).
To search for genomic footprints of selection in Ghana and Uganda, we explored allele frequency differentiation using Population Branch Statistic (PBS) using all the data, i.e., without LD pruning as done during the PCA and ADMIXTURE analysis [29], but excluding the eBL cases in Northern Uganda. PBS estimates were performed using NCI Ghanaians and Northern Ugandan controls as study populations, the Southern Bantu populations (Sotho and Zulu) from the African Genome Variation Project [17] as a sister group, and the Europeans (CEU+TSI+FIN+GBR+IBS) from 1000 Genomes project [24] as reference population.
In addition to PBS, we performed Extended Haplotype Homozygosity (EHH) [70] analysis (SI) using the Cross-population Extended Haplotype Homozygosity (xpEHH) [71] in R package rehh v.2.0.2 [72]. To minimize spurious results of individual SNPs [73], all the selection analyses were performed on windows of 20 SNPs overlapping by 5 SNPs. For the density of SNPs used in the present study (~1,000,000), the average window size of 20 SNPs corresponds to an average ~ 50 Kb. We used ANNOVAR [74] to annotate SNPs found in candidate regions under selection. To consider a candidate region to be under selection, we adopted a conservative approach of filtering those regions that showed extreme signals in both PBS and xpEHH methods (S5–S7 Tables). For the intergenic natural selection signal, we represented the genetic distances from the closest genes (S5 and S6 Tables).
Simulations were carried out using the demographic model [76] (S15 Fig), based on estimated divergence (thousands of years ago, kya) and effective population size (Ne) of African populations performed in Mallick et al. [75]. We used the Dinka population as a proxy for UNC and UNW, and the Luhya population as a proxy for Southern Bantu, with inferred divergence range of 9 and 25 kya (Mallick et al. high and low divergence inference), and current Dinka and Luhya Ne of 3x104 and 3x104, respectively [75]. We used the Yoruba population as a proxy of the Ghanaian population, and the estimated divergence from the Luhya of 5 and 10 kya and current Yoruba Ne of 7x104. We also used the French population as a European proxy, 40 to 60 kya for an inferred divergence time and 3x104 for current Ne. Considering that the study populations were involved in gene flow events, we introduced migration parameters between study populations and Southern Bantu considering the ancestry proportions inferred by ADMIXTURE (Fig 1C), as 4Nemij, where 4Ne is the population effective size and mij the fraction of population i that is made up of migrants from population j (for more details see S15 Fig).
Additional Methods are presented in Supporting Information (S1 Text).
|
10.1371/journal.pgen.1000893 | Incipient Balancing Selection through Adaptive Loss of Aquaporins in Natural Saccharomyces cerevisiae Populations | A major goal in evolutionary biology is to understand how adaptive evolution has influenced natural variation, but identifying loci subject to positive selection has been a challenge. Here we present the adaptive loss of a pair of paralogous genes in specific Saccharomyces cerevisiae subpopulations. We mapped natural variation in freeze-thaw tolerance to two water transporters, AQY1 and AQY2, previously implicated in freeze-thaw survival. However, whereas freeze-thaw–tolerant strains harbor functional aquaporin genes, the set of sensitive strains lost aquaporin function at least 6 independent times. Several genomic signatures at AQY1 and/or AQY2 reveal low variation surrounding these loci within strains of the same haplotype, but high variation between strain groups. This is consistent with recent adaptive loss of aquaporins in subgroups of strains, leading to incipient balancing selection. We show that, although aquaporins are critical for surviving freeze-thaw stress, loss of both genes provides a major fitness advantage on high-sugar substrates common to many strains' natural niche. Strikingly, strains with non-functional alleles have also lost the ancestral requirement for aquaporins during spore formation. Thus, the antagonistic effect of aquaporin function—providing an advantage in freeze-thaw tolerance but a fitness defect for growth in high-sugar environments—contributes to the maintenance of both functional and nonfunctional alleles in S. cerevisiae. This work also shows that gene loss through multiple missense and nonsense mutations, hallmarks of pseudogenization presumed to emerge after loss of constraint, can arise through positive selection.
| Local adaptation is thought to be a driving force in population differentiation and the formation of new species. Yet, there are few examples of ecologically relevant phenotypes that have been mapped to individual genes, making it difficult to know what drives the evolution of such genes and contributes to the molecular mechanisms underlying divergence. Here, we provide a unique case of local adaptation through multi-gene loss. We mapped the genetic basis for natural variation in yeast freeze-thaw tolerance to two water transporters, AQY1 and AQY2. Although tolerant strains harbor functional alleles of both genes, the set of sensitive strains lost aquaporins at least 6 independent times, through missense mutations and frame-shifting deletions. Genome-wide scans reveal several signatures of recent, partial selective sweeps at the aquaporin loci, indicating positive selection for gene loss. This was likely driven by a major fitness advantage of aquaporin loss when cells grow in high sugar concentrations common to many strains' niche. Surprisingly, strains that lost aquaporins also lost the ancestral requirement for these genes during sexual reproduction. This work provides a compelling example of how gene loss through nonsense mutations, a hallmark of pseudogenization, is caused not by loss of constraint but by positive selection.
| Biologists have long sought to understand the process of natural selection and the signatures left behind in extant species. Finding evidence of adaptive evolution has been a holy grail for evolutionary biologists, because it can provide insights into how and why organisms evolve. However, examples of adaptive selection from which to glean insights remain relatively scarce [1]. The recent explosion in the number of genomes available for different organisms provides an exciting opportunity to identify loci with unusual patterns of variation indicative of selection (for example [2]–[7]). However, even for loci with strong signatures of selection, the affected phenotypes are often a complete mystery. In contrast, mapping studies link quantitative trait variation to genomic loci that can then be interrogated for evidence of selection. The challenge in most organisms is identifying responsible SNPs within candidate regions, which are often megabases long and contain hundreds of functional elements, hindering further study [8].
Here, we used the power of yeast genetics and genomics to uncover a unique example of adaptive gene loss, involving multiple paralogous genes and several sequential evolutionary events. We previously surveyed phenotypic variation in Saccharomycetes collected from diverse environments and found that relatively few of those strains (12%) could survive freeze-thaw (FT) stress [9]. Many tolerant strains were isolated from oak soil in the Northeastern United States, whereas sensitive strains were typically isolated from warm environments, often from fruit or fermentations. This suggested that FT tolerance has been selected for in strains from cold climates but lost in other isolates. Several genes have been linked to freeze-thaw tolerance in yeast and other organisms, including water transporters. The paralogous yeast aquaporins (AQYs) AQY1 and AQY2 were implicated in FT stress by the baking industry, which found that AQY over-expression increases yeast viability in frozen bread dough [10]. Rapid export of water through AQYs is thought to increase FT survival by preventing intracellular shearing due to water crystallization [10],[11]. The paralogs may have arisen in the whole-genome duplication (WGD) event in the Saccharomyces lineage [12], since all post-WGD species all have two aquaporins whereas most pre-WGD species have a single ortholog (Dana Wohlbach and A.P.G., unpublished). It has been observed that laboratory and industrial strains as well as several vineyard isolates harbor non-functional alleles of AQY2, while several strains harbor a non-functional version of AQY1 [13]–[16]. However, without population-level analysis or knowledge of potential ecological driving forces, it is difficult to distinguish selection at these loci from neutral gene loss in the progenitor of the related strains. Here, we provide the first evidence of adaptive loss of AQY paralogs in natural populations of S. cerevisiae, leading to incipient balancing selection via spatial variation in selective pressures.
We mapped FT tolerance using a cross between naturally FT-resistant strain YPS163, collected from Pennsylvania oak trees [17], mated to a FT-sensitive lab strain derived from S288c, by phenotyping 44 recombinant strains together genotyped at 198 markers spaced roughly every 60 kb (∼30 cM) [18]. Two loci were identified: one on the left arm of chromosome 12 and one on the right arm of chromosome 16 (Figure 1, see Methods). Each contained one of two paralogous aquaporin transporters, AQY2 and AQY1, respectively, which were previously linked to freeze-thaw tolerance [10],[11],[19]. Together, these genes explained >90% of the phenotypic variation, with AQY2 alone explaining two-thirds of the effect (Figure 1C). This was confirmed by reciprocal translocation experiments (Figure 1D): deletion of either gene from YPS163 diminished FT tolerance according to the QTL effect plots, while deletion of both genes ablated FT survival. Introducing either gene into the S288c-derived lab strain (which harbors non-functional alleles of both genes [13]–[16]) donated partial FT tolerance to the otherwise sensitive strain. Thus, AQY2 and, to a lesser extent, AQY1 are major effectors of natural variation in yeast FT tolerance.
Sequencing AQY2 and AQY1 from the population revealed a near-perfect correlation between FT tolerance and the presence of functional AQY genes. Tolerant strains contained nearly identical and known functional alleles of both genes, while several strains with an intermediate phenotype contained only one functional gene. However, FT-sensitive strains displayed several different non-functional AQY alleles (Figure 2 and Figure S1). There were three distinct frame-shifting deletions in AQY2, including a known 11-bp deletion in laboratory and vineyard strains [14],[16], deletion of G at position 25 (G25) in Asian isolates and several other strains [14],[16], and a G528 deletion in the Malaysian AQY2 that is unable to contribute FT tolerance in our assay (Table S2). Several coding polymorphisms were shared in the recapitulated proteins encoded by the 11 bp-deletion allele or by the G25 allele (Figure 2). There were also three different non-functional AQY1 alleles in the population, including the A881 deletion that renders AQY1 inactive in our context (see Table S2 and Figure S5C), the V121M polymorphism known to inactivate water transport [13], and a 955-bp deletion that removes the first 106 bp of AQY1 and its upstream region in Malaysian strains. The trees for Aqy2 and Aqy1 are distinct from one another, and significantly different from trees based on neutral or genomic sequence that show clear distinction between Asian strains and vineyard isolates [9],[20],[21]. Such discordance between gene and species trees can be a sign of non-neutral evolution. Furthermore, there were five different combinations of non-functional AQY1 and AQY2 alleles, and a higher-than-expected frequency of strains harboring both functional or both non-functional genes (p = 3.5×10−4, Chi-square test). This cannot be simply explained by shared ancestry, which would have produced similar protein trees and a limited combination of alleles, and instead supports the non-random retention or loss of both AQY genes.
We applied several tests to assess if loss of AQY function may have been selected for in some strains. Under the neutral model, the rate of polymorphism within strains should be similar to the rate of divergence across species. Instead, both AQY1 and AQY2 show an excess of replacement polymorphism, assessed by the McDonald-Kreitman test [22] that compares non-synonymous (A) to synonymous (S) codon changes (Table S3). AQY1 showed an A/S ratio of polymorphism (5/4 = 1.25) that was significantly higher than that of divergence (11/49 = 0.22, p = 0.026, Fisher's exact test). AQY2 also showed an excess of polymorphic sites (A/S of 8/20 = 0.4) compared to divergent sites (3/40 = 0.075, p = 0.019), as well as an excess of deletions (3/20 versus 0/40, p = 0.045). AQY2 (but not AQY1) also deviated from neutral evolution at synonymous sites, showing an excess of SNPs compared to 8 intergenic sequences (p = 0.028, multi-locus HKA test [23], Table S4 and Table S5). For the most part, the tests were not significant if subgroups of strains, defined by AQY haplotypes in Figure 2 (see Table S1 for details), were considered separately (Table S3 and Table S5). This result indicates that much of the variation is between strain groups.
Excess polymorphism can result from relaxed constraint in the species, or if local adaptation is driving divergence between populations [24],[25]. To distinguish between these models, we used non-imputed genome sequence data of Liti et al. [21] to characterize sequence variation flanking the AQY genes. We applied several empirical tests, which can handle the missing data in the low-coverage genomic sequences and are less subject to the unusual features of S. cerevisiae populations (including extensive population structure, unknown population dynamics, ambiguous balance between clonal vs. sexual reproduction, and human-associated migration [21], [26]–[29]) that can confound standard tests [2],[30]. To monitor the variation surrounding AQY2, we subdivided 21 strains with data at AQY2 into strains harboring the Asian G25 deletion, the 11 bp deletion, or the full-length AQY2 (clonal Malaysian strains were not considered, see Methods). We calculated the average pairwise nucleotide differences surrounding AQY loci within and between groups, and then compared this variation to other regions across the genome. We tested for several signatures: a recent selective sweep is predicted to reduce variation flanking the selected allele in the affected population, while balancing selection can increase variation between strain groups [25]. Since much of the genome may be evolving neutrally, loci with extreme values display the strongest evidence for non-neutral evolution.
For strains harboring the Asian G25 allele of AQY2, we saw a high correlation in between-group variation and within-group variation across much of the genome, including the right arm of chromosome 12 (Figure 3A, right side). However, a 50 kb stretch on the left arm of chromosome 12 showed below-average variation within the strains (0.76th percentile compared to other similarly sized regions genome-wide, see Methods) but high variation between groups. There was a sharp break in this pattern at ∼72 kb, which may represent the breakpoint of a selective sweep. To further explore this, we calculated the difference in between-group variation minus within-group variation, then calculated the area under contiguous peaks in the difference curve for comparison (see Figure S2 and Methods). This procedure identified a 5.6 kb region spanning the 870 bp AQY2 ORF that ranked in the top 1.2 percentile of 4,600 regions genome-wide with skewed between-group versus within-group variation (Figure 3A). Strains harboring the 11-bp deletion displayed a 4,300 bp region encompassing AQY2 with a significant skew in the between- versus within-group variation (1.8th percentile of 3,238 regions genome-wide, Figure 3B) and below-average within-group variation (6th percentile genome-wide). Strains harboring the full-length AQY2 showed a smaller peak of 1,800 bp with high between-group variation (6.3rd percentile, Figure 3C), but average within-group variation (>50th percentile). These results show that strains harboring either deletion have low variation within those strain groups, and that the high variation at AQY2 distinguishes the three groups from one another. Indeed, a genome-wide plot of FST [2], which measures the population differentiation based on these three groupings, identified a clear peak of 6.4 kb over AQY2 with above-average FST, ranking among the top 1.5th percentile genome-wide (Figure 3D).
A confounding feature is the extensive population structure within S. cerevisiae [21],[29], which can mimic some signatures of selection. Several controls indicate that the observed patterns are unlikely due to demographics. First, these regions were among the most extreme across the genome, which is not expected if population structure is the underlying cause. However, many S. cerevisiae strains have mosaic genomes, for which large regions have distinct lineages [21],[29]. To control for this, we performed a partitioning sampling: strains were partitioned at each of 1,370 randomly chosen SNPs across the genome. The difference in between-group minus within-group variation was scored surrounding the partitioning SNP and compared to the difference profile when strains were partitioned based on AQY2 allele (see Methods). The regions observed for the Asian G25 or 11-bp deletion classes remained among the most extreme (4.7th and 5.4th percentile, respectively). Thus, the profiles we observe in Figure 3A and 3B are uncommonly found at random SNPs, most of which likely reflect neutral variation. In contrast, the skew in variation found in strains with full-length AQY2 was not significant by this assessment (26th percentile).
We conclude that the observed skew in polymorphism observed in strains with the Asian G25 deletion and the 11-bp deletion in AQY2 resulted from two separate partial selective sweeps that reduced variation within each group. The high variation distinguishing strain groups is a signature of balancing selection, which may be maintaining both functional and non-functional AQY2 alleles in the population. Indeed, we observed a positive Tajima's D at AQY2, assessed on a smaller set of high-quality sequences (D = 0.851, p<0.05, Figure S3), indicating an excess of intermediate-frequency polymorphism that is consistent with balancing selection [24].
The patterns at AQY1 were less clear. Strains harboring the aqy1 V121M allele or the A881 deletion showed reduced variation within each group and high variation between groups at the AQY1 region (Figure S4). Although these were highly significant compared to other loci across the genome (0.77th and 1.23rd percentile, respectively), they were not significant compared to random-SNP partitioning described above (16th and 48th percentile, respectively). Thus, the slight skew in between-group versus within-group variation at AQY1 could be due to demographic factors, incorrect strain groupings, or older or weaker selective sweep(s) that have since recovered variation through recombination or mutation.
The above results strongly suggest selective pressure to lose AQY function in some strains, perhaps driven by environmental factors. We previously reported an anti-correlation between FT survival and osmo tolerance across a wide range of S. cerevisiae strains (R = −0.35, p = 0.006) [9]. Furthermore, a lab strain with functional AQYs was shown to be sensitive to hypo- and hyper-osmotic cycling, but not to consistently high osmolarity [13],[14]. Instead, we found that loss of both AQY genes provides a major growth advantage in high osmolar conditions found in nature (Figure 4). A YPS163 mutant lacking both AQYs displayed ∼1.7X greater survival in 1.5 M sorbitol, whereas introducing a functional AQY into the S288c-derived lab strain decreased survival 2–3X. Furthermore, sorbitol tolerance was anti-correlated to both freeze-thaw tolerance (R = −0.38) and the number of functional aquaporins (R = −0.31) in these strains (Table S1). The sugar concentration used here is comparable to that found in the fruit substrates of many wild strains [31]. Thus, AQY function presents a substantial fitness defect in conditions relevant in nature, likely due to passive water loss triggered by the high osmolarity of sugary substrates.
In the course of these experiments, we also discovered that YPS163 lacking either aquaporin had a major defect in spore formation during meiosis (Figure 5). Although AQY1 had been previously implicated in a late step in spore maturation [32], our phenotype is distinct in that it affects spore production. Whereas >70% of the parental YPS163 formed full tetrads within 2 days, only 18–24% of the double or single mutants produced full tetrads. After 9 days, the mutant produced more spores but was still defective compared to the parental strain (<60% full tetrads compared to ∼85%, Figure S5). The AQY requirement is ancestral, since an S. paradoxus aqy1Δ mutant displayed an identical defect (Figure 5A). In contrast, strains without functional AQY genes produce full tetrads (albeit with lower efficiency than YPS163 [33]), consistent with a previous report showing AQY1 is not required for sporulation in vineyard strains [34]. More importantly, introducing the functional YPS163 allele of AQY1 into strains with different combinations of non-functional AQY alleles (including strains M22, K1, SK1, and S288c) did not significantly improve spore production (Figure 5B). Thus, strains lacking AQY function have also lost the ancestral need for AQY during spore production.
This work provides the first clear evidence for adaptive loss of AQY function in subgroups of wild S. cerevisiae isolates. The excess polymorphism at AQY genes (McDonald-Kreitman and HKA tests), high between-group variation surrounding AQY2 that distinguishes strain groups (group variation and FST plots, Figure 3), and skew in the frequency spectrum toward intermediate-frequency AQY2 alleles (Tajima's D) are all consistent with non-neutral evolution. Furthermore, AQY paralogs have been lost at least 6 independent times, through 2 partial selective sweeps at AQY2 and possibly others at AQY1. The high variation between strain groups, and the non-random retention or loss of both paralogs in diverse strains, is consistent with the establishment of balanced polymorphism. We propose that the antagonistic pleiotropy of aquaporin function, coupled with spatial differences in selective pressures, provide pressure to maintain both functional or both non-functional alleles in distinct subpopulations of S. cerevisiae.
FT tolerance may be crucial for survival in cold climates, and along with sporulation efficiency may impart strong pressure to retain AQY genes in strains from wintry niches. Indeed, the ratio of non-synonymous to synonymous differences in YPS163 compared to S. paradoxus is 2 - 6X lower for AQYs compared to the genomic average (Ka/Ks of 0.018 and 0.059 for AQY2 and AQY1, respectively, versus 0.1 across all genes [35]). This is consistent with purifying selection acting to remove deleterious codon changes. The oak strains likely represent the ancestral state, since close relatives S. paradoxus and S. mikatae are also recovered from tree exudates and soil [17],[36], display high FT tolerance [9], and require aquaporins for sporulation (Figure 5 and data not shown). Interestingly, Northeastern-US oak strains display unique phenotypes suggestive of other evolutionary forces as well. AQY2 is expressed on average 14-fold higher in YPS163 compared to 17 other surveyed strains [9],[37]; those levels are doubled in YPS1009, which underwent a duplication of the entire chromosome 12 [9]. Although further studies will be needed, that over-expression of AQY2 is known to enhance FT tolerance in industrial strains [10] hints that the elevated expression may have been selected for, further underscoring the importance of AQY function in these strains.
In contrast, many other strains exist in warm environments that never experience freezing. Most of these were sampled from fruit substrates and distillations, which typically consist of ∼25% sugars [31], in contrast to oak soil [38],[39] from which many cold-climate strains have been isolated. Thus, the significant advantage in osmo-tolerance due to AQY loss likely played a major role in selection at this locus. It is unclear which came first–loss of aquaporin requirement during sporulation, or loss of aquaporin function that drove subsequent loss of the sporulation role. Loss of sporulation dependency on aquaporins, coupled with migration to warmer climates, would have relaxed constraint on the genes and facilitated their adaptive loss when cells moved to high-sugar substrates. This model could have involved a single loss of sporulation requirement followed by multiple independent losses of aquaporin function. Alternatively, strong selective pressure to lose aquaporins could have forced multiple independent losses of the sporulation requirement, just as it lead to multiple independent losses of aquaporin function.
S. cerevisiae strains are thought to have migrated globally through human association, after two domestication events produced sake/distillation strains and vineyard/wine-making lines ∼10,000 years ago [20],[21],[26],[27],[29],[40]. Human-facilitated migration may have significantly increased exposure of S. cerevisiae to diverse climates, which may have imposed new selective pressures when strains encountered new niches. Increased migration may also have facilitated outcrossing of domesticated strains with natural strains, allowing several of these alleles to spread through natural populations. It is important to note that Malaysian strains, not previously associated with domestication events, show unique non-functional AQY alleles, revealing that loss of aquaporins is not strictly driven by domestication.
The selective sweeps of nonfunctional aquaporin alleles appear to have been recent events, given the strength of the signal at AQY2, and may reflect an ongoing process. A remaining question is the fate of the emerged balance in polymorphism. Given sufficient migration of strains between the two niches and unequal fitness costs of the opposing haplotypes (i.e. two functional or two nonfunctional AQY alleles), one haplotype may eventually win out to fixation, eliminating the balanced alleles. On the other hand, long-term balancing selection could result if equivalent selective constraints are maintained in each respective niche. In the extreme case, strongly opposing selective forces could restrict yeast migration between environments to promote ecological speciation [41]. Little is known about S. cerevisiae migration between tree soil and fruits, although oak-soil strains are genetically well separated from vineyard/fermentation isolates [21],[29],[40],[42]. The antagonistic forces driving aquaporin loss at the cost of freeze-thaw sensitivity may be one factor that has limited gene flow between these niches.
Strains and plasmid constructs are described in Table S6. Two S. cerevisiae strains (DY8 and DY9) were isolated from oak-tree soil from Maribel, Wisconsin using the method of [17], and typed by a mating/sporulation assay with a tester S. cerevisiae strain (Dan Kvitek and APG, unpublished). Gene deletions were created by homologous recombination, replacing AQY1 and/or AQY2 with KanMX3 or NatMX3 drug-resistance cassettes, respectively. Homothalic wild strains capable of mating-type switching (including YPS163, M22, and S. paradoxus) were sporulated and dissected, and drug-marked colonies were selected as homozygous diploids. In all cases, homozygous gene deletions were confirmed by diagnostic PCR. The region corresponding to the 870 bp full-length AQY2 ORF plus 971 bp upstream and 393 bp downstream sequence was cloned from YPS163 or BY4741, by homologous recombination replacing a GFP-ADH1-terminator cassette in plasmid BA1924 (provided by P. Kainth and B. Andrews), which is derived from pRS315-based CEN plasmid BA1926 [43] but with the NatMX3 cassette replacing the LEU2 marker. The region corresponding to the 918 bp full-length AQY1 ORF with the flanking 947 bp upstream and 747 bp downstream was similarly cloned. All clones were verified by sequencing. To assess functionality of the different alleles, AQY1 ORFs representing M22, BY4741, or Y55 alleles (identical to the YPS163 allele but harboring the A881 deletion) or the Malaysian AQY2 allele (identical to YPS163 except for the G528 deletion) were cloned between the native upstream and downstream AQY1 sequence from YPS163. This was done to prevent confounding influences on expression through variation in the flanking regulatory regions. Plasmids were introduced into YPS163 aqy1Δ, BY4741, or other naturally AQY-minus strains, and complementation of spore production in the YPS163 aqy1Δ mutant or of FT tolerance in BY4741 was scored (Table S2).
Yeast strains were grown at 30°C in YPD medium to an optical density at 600 nm (OD600) of 0.3–0.4 in 24-well plates. To measure freeze-thaw tolerance, 200 µl of cells was transferred to 1.5 ml tubes and frozen in a dry ice/ethanol bath (<−50C) for two hours or on ice as control. Viability was measured by scoring serial dilutions spotted onto agar plates as previously described [9], or using Live/Dead stain (Invitrogen, Carlsebad, CA) read on a Guava EasyCyte Plus flow cytometer (Millipore, Billerica, MA). Scores in Figure 2 correspond to high (>80% of YPS163 viability, three pluses), medium (50–80% viability, two pluses), low (<50% viability, one plus), or no detectible (minus sign) FT tolerance. Osmotic tolerance was measured by plating cells onto agar plates containing 1.5 M sorbitol. Percent viability was scored as the number of colony-forming units compared to the no-stress control plate.
Cells were grown in YPD rich medium to OD600 nm of 1.0, harvested by centrifugation, resuspended in 1% potassium acetate, and incubated at 25°C for 2 or 9 days. Cells were harvested, diluted and the number of spores per tetrad was counted on a hemocytometer.
QTL mapping strains and analysis were as previously described [18], using the Haley-Knott algorithm implemented in R-QTL [44]. Two additional peaks in Figure 1B (left arm of Chromosome 2 and right arm of Chromosome 8) were not significant when Chromosome 12 and 16 QTL were held as fixed terms, suggesting the additional peaks may be false positives. Sequencing using Big-Dye (Applied Biosystems, Carlsbad, CA) scored at least 3 reads (including forward and reverse) per basepair from 2 independent genomic preparations (GenBank accessions GQ848552-74 and GQ870433-54). The vast majority of sequence data represented homozygous sites. The few base pairs with evidence of heterozygocity were represented by one of the alleles, randomly chosen. MK-tests, Tajima's D, and Ka/Ks were calculated in DNASP 5.0 [45] and ML-HKA was done as in [23] using sequence data from [9],[20] and here.
Genome-wide sequence analysis was performed using unimputed, aligned data from [21] with quality scores > = 40 (generously provided by Alan Moses), treating all gaps as missing data to avoid alignment errors. Strains were grouped according to AQY2 or AQY1 alleles (see Table S1 for details), and the average number of pairwise SNPs was calculated every 1000 bp with a 100 bp step size, for all pairs of strains within each group and for all strains in a given group compared to each strain outside that group. Within-group variation was scored for all 50 kb regions across the genome with a step size of 20 kb, and for all 5 kb regions with step size of 2 kb. These regions were compared to the 50 kb region highlighted in the text (position 22,000–72,000 in Figure 3) for strains with the G25 AQY2 allele and to a 5 kb region centered on AQY2 for other strain groups. All regions were ranked based on the average pairwise within-group variation to calculate the percentile rank of regions in question.
To monitor the skew variation within and between groups, a difference profile of between-group variation minus within-group variation (calculated as described above) was taken across the genome, and all contiguous regions (“peaks”) where the difference value was >1.5X the chromosome-wide average were identified (see Figure S2 and Table S7). The area under each peak was estimated by the trapezoidal method, and compared to the area under the peaks in Figure 3A and 3B spanning AQY2. For the partitioning sampling, we scanned for SNPs with at least 3 strains harboring the minor allele, every 10,000 bp across each of the 16 yeast chromosomes. Strains were partitioned based on that SNP, then the between-group and within-group variation was measured for 20,000 bp centered on the partitioning SNP, based on the average-pairwise differences every 1,000 bp with a 100 bp step size as above. A profile of the between-group variation minus the within-group variation was taken in every window. For each partitioning SNP, a peak in the difference profile was identify by walking outward until the difference value was <3.54, the cutoff used the genomic scan shown in Figure 3B. The area under the curve was calculated as above and compared to that measured at AQY2 by an identical procedure except that strains were partitioned by Asian G25 allele vs. all others strains or by 11-bp deletion vs. all other strains. Very similar percentile rankings were obtained if we scored 5 kb windows centered on each SNP (data not shown).
|
10.1371/journal.pntd.0005630 | First report of naturally infected Aedes aegypti with chikungunya virus genotype ECSA in the Americas | The worldwide expansion of new emergent arboviruses such as Chikungunya and Zika reinforces the importance in understanding the role of mosquito species in spreading these pathogens in affected regions. This knowledge is essential for developing effective programs based on species specificity to avoid the establishment of endemic transmission cycles sustained by the identified local vectors. Although the first autochthonous transmission of Chikungunya virus was described in 2014 in the north of Brazil, the main outbreaks were reported in 2015 and 2016 in the northeast of Brazil.
During 5 days of February 2016, we collected mosquitoes in homes of 6 neighborhoods of Aracaju city, the capital of Sergipe state. Four mosquito species were identified but Culex quinquefasciatus and Aedes aegypti were the most abundant. Field-caught mosquitoes were tested for Chikungunya (CHIKV), Zika (ZIKV) and Dengue viruses (DENV) by qRT-PCR and one CHIKV-infected Ae. aegypti female was detected. The complete sequence of CHIKV genome was obtained from this sample and phylogenetic analysis revealed that this isolate belongs to the East-Central-South-African (ECSA) genotype.
Our study describes the first identification of a naturally CHIKV-infected Ae. aegypti in Brazil and the first report of a CHIKV from ECSA genotype identified in this species in the Americas. These findings support the notion of Ae. aegypti being a vector involved in CHIKV outbreaks in northeast of Brazil.
| The chikungunya outbreaks have become one of the main public health problems in the Northeast Region of Brazil. Since the chikungunya virus (CHIKV) was recently introduced in this country, finding vector associations involved in its transmission to the vertebrate host is primordial to better design efficient intervention against mosquitoes. In the present study, field-caught mosquitoes in Aracaju city, Sergipe State, Brazil were evaluated for the presence of CHIKV and other two endemic arboviruses (Zika and Dengue viruses) in this region. We described here the first identification of the East-Central-South-African (ECSA) CHIKV genotype naturally infecting Aedes aegypti mosquitoes in Brazil. This work suggests the involvement of this species in CHIKV outbreaks in Northeast Region of this country and reveals a possible interaction between this vector and the recently introduced CHIKV ECSA genotype.
| Chikungunya viral disease is caused by an arbovirus (Alphavirus genus) from Togaviridae family [1,2]. The most frequent symptoms during human infection are fever and joint pain, but headache, muscle pain, joint swelling and rash are also observed [3,4]. Chikungunya virus (CHIKV) is transmitted to humans by Aedes mosquitoes in sylvatic (animal-mosquito-man) [5] or urban transmission cycles (man-mosquito-man) [6]. This virus is endemic in Africa and Asia, and has spread to several European countries and more recently to America [7]. In Brazil, the first autochthonous case was described in 2014 in Amapá, a state in the North Region [8]. Since then, CHIKV has expanded its distribution in Brazil and 13,236 cases were confirmed in 2015, most of them occurring in states within the Northeast Region [9–11].
Three CHIKV genotypes have been identified since its discovery in 1952 [2]. The East-Central-South-African (ECSA) and West African genotypes circulate in sub-Saharan Africa, whereas Ae. aegypti is the vector of the Asian genotype in human urban transmission cycles in Southeast Asia and in the Americas. Asian and ECSA genotypes were described in Brazil concomitantly with autochthonous cases caused by both genotypes [12]. Currently, all Brazilian states are infested by both of the main CHIKV vectors, Ae. aegypti and Ae. albopictus [13,14]. The presence of both species raises concern about the establishment of sustained CHIKV outbreaks throughout the entire country [15]. Although neither of these vectors have yet been found to be naturally infected with CHIKV in Brazil, it is known that Brazilian populations of both species are highly competent vectors for CHIKV when infected under laboratory conditions [16]. Moreover, it is not known if other Culicidae species occurring in urban area in Brazil are capable to transmit CHIKV. Therefore, entomological surveillance is essential for understanding the role of vector species in CHIKV transmission, especially in new endemic countries such as Brazil. Recently, Aracaju, a Northeast Region city and the capital of Sergipe state with 571,149 inhabitants [17], has experienced an increase in CHIKV cases since the detection of the first autochthonous transmission in 2015. According to the State Department of Health from the Sergipe government, the highest number of CHIKV cases in Aracaju has been reported in 2016, being 6,810 suspected and 4,743 confirmed until epidemiological week 33. The same epidemiological bulletin reported 30 confirmed cases for Zika virus (ZIKV) and 1,457 for Dengue virus (DENV) in 2016 [18].
In this alarming scenario, we hypothesize that the city of Aracaju is a potential site for detecting vector species naturally infected by CHIKV, ZIKV and DENV. During the third week of February 2016, we collected adult mosquitoes, inside and outside homes situated in urban areas of Aracaju, where residents were complaining of symptoms consistent with CHIKV or related arboviruses (DENV and ZIKV). Our results revealed that the ECSA genotype of CHIKV is naturally infecting Ae. aegypti in the Americas and showed the first identification of a field-caught mosquito infected with CHIKV in Brazil. These findings confirm the vectorial capacity of Ae. aegypti for transmitting CHIKV and suggests that this species is an important CHIKV vector which can be involved in the recent outbreaks since this virus was introduced in this country.
During a multidisciplinary coordinated investigation involving the ZIKV São Paulo task force (Zika Network) and the Central Public Health Laboratories (LACEN—Aracaju, Sergipe, Brazil), medical interviews, clinical diagnosis and mosquito collections were performed during one week in Aracaju. Houses in six different neighborhoods were selected based on the locations of clinically ill patients (suspected symptoms of arboviral infection) provided by LACEN. All neighborhoods were mostly residential, composed by heterogeneous house sizes of very low- to low-income, except by São José neighborhood, with middle-income dwellings. After an authorized employee from LACEN obtained permission of the residents, mosquitoes were collected by the entomological team. Peridomiciliary and intradomiciliary areas (all the rooms) of these houses and adjacent residences were aspirated during the morning after the dawn or afternoon until the dusk period using vacuum aspirators manufactured by Horst Armadilhas. Captured mosquitoes were placed into a collection vial previously identified by date and neighborhood and for each home inspected, one vial was used.
The collected mosquitoes were ice anesthetized and sorted by species using a morphological identification key as reference when necessary [19], sex and collected region. Males were pooled in 4 to 11 individuals of the same species and both engorged and non-engorged females were individually divided into head, thorax and abdomen using a McPherson-Vannas Scissors #501234 (World Precision Instruments, Sarasota, FL) to avoid contamination from possible blood inside the mosquito midgut. Samples were stored individually into a 1.5 mL microtube and immediately frozen at -80°C for further analyses.
Total RNA was extracted from each individual female thorax and from pooled whole male mosquitoes using QIAmp Viral RNA Mini Kit (Qiagen, USA) following manufacturer’s recommendations. About 5 to 10 μL of extracted thoracic RNA from each female was pooled in a species-specific manner and the remaining volume of each RNA sample was stored at -80°C for subsequent individual confirmations and further analyses.
The species-specific RNA pools were tested for CHIKV, ZIKV and DENV1-4 presence using QuantiTect Probe RT-PCR kit (Qiagen, USA) and Mastercycler Realplex 2 thermocycler (Eppendorf, Germany). Primers/probes used for detection of emerging arbovirus were previously described for DENV1-4 serotypes (DEN FP and DEN RP primers and DEN P probe) [20], ZIKV (ZIKV 835 and ZIKV 911c primers and ZIKV 860 probe) [21] and CHIKV (CHIKV FP1 and CHIKV RP1 primers and CHIKV P2 probe) [22]. CHIKV primers and probe applied in this study detect the 3 CHIKV genotypes.
The one step qRT-PCR conditions, including total RNA sample volume and cycling for ZIKV and CHIKV detection reactions were previously described [2]. For DENV detection, 20μL reactions were used as following: 10 μL of QuantiTect Probe RT-PCR Master mix, 0.25 μL of QuantiTect RT mix, 1 μL of 10 μM of each primer, 0.4 μL of 10 μM probe, 2.4 μL of DNA RNA free water and 5 μL of total RNA sample. The thermocycler conditions for DENV were 50°C for 20 min and 95°C for 15 min followed by 45 cycles of 95°C for 15 sec, 60°C for 15 sec and 40°C for 30 sec. For each qRT-PCR experiment, a positive control consisting of a specific viral RNA (DENV2, ZIKV or CHIKV) was added for cDNA synthesis and detection validation. Each sample, as well as negative (DEPC treated water as template) and positive controls, were analyzed in technical duplicates.
For positive pools, individual RNA samples were retested to confirm which mosquitoes from the positive pool were infected.
Ae. aegypti RNA from CHIKV positive female was prepared for NGS sequencing using a sequence-independent single-primer amplification (SISPA) method as described previously [23]. Libraries were sequenced with an Illumina MiSeq desktop sequencer using a version 3 kit (2 x 150 cycles). Additionally for mosquito sample that was positive for CHIKV we used the Illumina TruSeq RNA Access enrichment method to get coding-complete genomes as described previously [24]. CHIKV specific probes were designed against Brazilian isolate KP164569.
For phylogenetic characterization of sequenced CHIKV detected in positive mosquitoes from Sergipe-Brazil, additional CHIKV sequences from Brazil and other countries, representing all known genotypes, were recovered from Genbank (http://www.ncbi.nlm.nih.gov) (S1 Table). All sequences used in this work are presented in the format: genotype/accession number/country/year of isolation in the phylogenetic tree. Sequences were aligned using Clustal X2 [25] and the alignment was manually edited using jalview.
Viral phylogenies based on full-length nucleotide sequences were estimated using Maximum Likelihood (ML) implemented in FastTree 2 [26] and Bayesian Inference (BI) analysis with BEAST package v.1.8.2 [27], using the general time-reversible with gamma-distributed rate variation substitution model (GTR+G), as described by Akaike’s information criterion (AICc) in jModelTest 0.1 [28]. For the BI, a relaxed lognormal molecular clock model and the nonparametric Skygrid coalescent model were employed. The evolutionary analysis were computed using two independent runs for 100 million MCMC steps and the convergence of the MCMC chains was inspected using TRACER v.1.6 (http://tree.bio.ed.ac.uk). Posterior trees were summarized discarding the first 10% of the sampled trees and choosing the Maximum Clade Credibility (MCC) was summarized using TreeAnnotator v.1.8.2. The final trees were then visualized and plotted using FigTree v.1.4.2 (http://tree.bio.ed.ac.uk).
Six neighborhoods were inspected in Aracaju city during February 16th to 20th, 2016 (Fig 1).
A total of 248 mosquitoes were collected in 39 properties and we identified 4 species during the triage procedure and identification (Table 1).
Culex quinquefasciatus were the most frequently collected species (78.2%,194/248), 3.88 times more represented than Ae. aegypti (20.2%, 50/248). Two other species, Ae. scapularis and Ae. taeniorhynchus were also captured, but with the same low proportion (0.8%, 2/248 each) in relation to the total number of mosquitoes collected.
Fifteen properties presented the co-occurrence of multiple species. The co-occurrence of Ae. aegypti-Cx. quinquefasciatus was the most registered, but we also found other combinations of concurrences as well. For instance, Cx. quinquefasciatus/Ae. aegypti/Ae. taeniorhynchus or Ae. scapularis were found in the same home. However, the co-occurrence of more than three species in the same property was not observed.
We tested the mosquitoes for DENV, ZIKV and CHIKV because Aracaju is endemic for DENV and human cases of ZIKV and CHIKV infection were recently confirmed in this city in 2016. All eight male and fifteen female pools of Cx. quinquefasciatus, as well as the one Ae. scapularis female pool and the Ae. taeniorhynchus female pool, were negative for CHIKV, ZIKV and DENV serotypes 1–4. Together, these pools represented a total of 198 mosquitoes (Table 2). Negative results were also observed for ZIKV and DENV1-4 detection in all male and female Ae. aegypti pools. However, one Ae. aegypti female pool was positive for CHIKV from a total of thirteen pools tested. All male pools presented negative for CHIKV (Table 2).
Following the pool analysis, we tested the individual thoracic samples that composed the CHIKV positive pool. We were able to detect one positive female, confirming our initial analysis (Table 2). The positive female was collected in São Conrado neighborhood (Fig 1), and the mean Ct values obtained in qRT-PCR was 21.76 for the individual thorax from the positive sample. Therefore, of thirty-eight females and twelve males of Ae. aegypti analyzed, only one female was infected with CHIKV.
The total RNA from CHIKV positive sample was selected for sequencing using Illumina sequencing platform. The whole sequence of the CHIKV genome was obtained and deposited in GenBank with accession number: KY055011.
Identity analysis of the whole CHIKV genome generated in this study with representative sequences of all known CHIKV genotypes obtained in GenBank (S1 Table) had percentages identities ranging from 84.24% to 99.99%, while between CHIKV sequence previously isolated in 2014–2015 in Bahia state showed percentages identities ranging from 99.91% to 99.99% (S2 Table). The mutations previously associated with CHIKV fitness increase in Ae. aegypti and Ae. albopictus [29,30] were not detected in envelope protein 1 (E1) from sequenced CHIKV.
Maximum likelihood analysis and Bayesian Inference comparing the CHIKV sequence isolated in Sergipe with sequences previously characterized (S1 Table) produced trees with similar topologies (Fig 2A and 2B, respectively).
The CHIKV-mosquito sequence was closely-related to other sequences from Bahia-Brazil isolated in 2014–2015 and with sequences from Africa, characterized as ECSA genotype (Fig 2). The consistency of the results is supported by the high bootstrap values and posterior probability observed in the trees (Fig 2A and 2B, respectively).
The identification of the vectors in endemic areas is essential to better design surveillance programs and defines efficient vector control strategies. In Brazil, it is well known that Ae. aegypti is the main DENV vector and although Ae. albopictus is also a potential vector in this country, ecological characteristics and adaptations favor the first species to sustain DENV outbreaks in urban areas [31]. Nevertheless, the recent introductions of CHIKV and ZIKV in Brazil and severe epidemics reported, have raised fundamental questions regarding the establishment of viral circulation in relation to the mosquito species occurring in endemic places [32].
A recent study showed that American populations of Ae. aegypti and Ae. albopictus, including Brazilian ones, are highly competent to transmit different CHIKV genotypes when orally infected in the laboratory [16]. Based on their results, the authors highlighted the imminent risk for an expansion of CHIKV outbreaks in tropical and subtropical areas of this continent [16]. In fact, Ae. aegypti transmitting CHIKV Asian genotype in the Americas was first described in Mexico, during an outbreak in Ciudad Hidalgo, Chiapas State [33].
CHIKV was first isolated in Brazil from infected patients in 2014 and the circulation of two genotypes was observed: Asian genotype in Oiapoque (Amapá state), North Region of Brazil and ECSA genotype in Feira de Santana (Bahia state), Northeast Region of Brazil [12]. It is important to highlight that Aracaju is in the Northeast Region and is approximately 300 km away from Feira de Santana. Coherently, we report here the first detection of a natural infection of Ae. aegypti mosquito by CHIKV genotype ECSA in Brazil and the Americas.
In the present study, Ae. albopictus was not found in the inspected houses although this species is endemic in Aracaju city [14]. However, we found two unusual species intra domiciliary, Ae. scapularis and Ae. taeniorhynchus, but the occurrence was much less frequent than Ae. aegypti and Cx. quinquefasciatus. The collected females from Ae. scapularis (n = 2) and Ae. taeniorhynchus (n = 2) were negative for ZIKV, CHIKV or DENV 1–4 serotypes. These females were engorged suggesting they are biting humans inside their homes. Moreover, it is known that Ae. scapularis presents vectorial competence for Rocio encephalitis, Ilheus and Melao viruses [34–36] while Ae. taeniorhynchus for Venezuelan equine encephalitis and West Nile viruses [37,38]. Nevertheless, it is unknown if these mosquito species are competent vectors in Brazil for the three arboviruses assayed in this study and if they can act as secondary vectors during outbreaks.
Recently, Ae. aegypti was shown to be a ZIKV vector in Rio de Janeiro city. In this study ZIKV-positive Cx. quinquefasciatus or Ae. albopictus were not identified [39]. We found Ae. aegypti as a potential vector of CHIKV in Aracaju city, as seen in outbreaks already reported elsewhere [5,40–42]. None of the other species examined were positive for CHIKV, ZIKV or DENV serotypes 1–4, mainly Cx. quinquefasciatus that was abundant in our sampling effort.
The surveillance of the Aedes mosquitoes should be expanded in order to prevent new CHIKV outbreaks in Brazil, since this country presents adequate conditions for the establishment of an endemic situation, which can also exposes other countries at risk. The mutations that enhance the fitness of the CHIKV genotype ECSA in Aedes mosquitoes were previously described [29,30] and they were not found in the genotype characterized in our study. However, new mutations that improve vector competence can be acquired since the ECSA genotype is being detected in other regions in Brazil [43], which suggests that different populations of Aedes mosquitoes are interacting with this recently introduced genotype, mainly Aedes aegypti in urban areas.
Our findings constitute the first description of Ae. aegypti-CHIKV genotype ECSA interaction in Brazil. These results reinforce the role of this species as an important vector of CHIKV in urban areas of northeast regions in Brazil.
|
10.1371/journal.pcbi.1002944 | Collective Cell Motion in an Epithelial Sheet Can Be Quantitatively Described by a Stochastic Interacting Particle Model | Modelling the displacement of thousands of cells that move in a collective way is required for the simulation and the theoretical analysis of various biological processes. Here, we tackle this question in the controlled setting where the motion of Madin-Darby Canine Kidney (MDCK) cells in a confluent epithelium is triggered by the unmasking of free surface. We develop a simple model in which cells are described as point particles with a dynamic based on the two premises that, first, cells move in a stochastic manner and, second, tend to adapt their motion to that of their neighbors. Detailed comparison to experimental data show that the model provides a quantitatively accurate description of cell motion in the epithelium bulk at early times. In addition, inclusion of model “leader” cells with modified characteristics, accounts for the digitated shape of the interface which develops over the subsequent hours, providing that leader cells invade free surface more easily than other cells and coordinate their motion with their followers. The previously-described progression of the epithelium border is reproduced by the model and quantitatively explained.
| Living organisms, from bacteria to large mammals, move not only as single entities but also in groups. This is true for cells in multicellular organisms. The group or collective motion of cells is an important component of development as well as processes like cancer and wound healing. To better understand this phenomenon, we have recorded the displacement of cells as they move collectively on a substrate and invade free space. The results can be accurately described by modelling the motion of cells as random but with a tendency to move at the same velocity as their neighbors. This allows us to analyze conditions under which the invasion of free space takes place, guided by a few cells that have become different of the others, as observed in the experiments. The developed model should serve as a useful basis for the description of other processes that involve collective cell motion.
| Interactions between moving entities correlates their motions. This takes place at all scales, from atoms and molecules, as evidenced by the familiar experiences of wind and fluid vortices to the astronomical scales of stars and galaxies. In the biological realm, collective movements are observed from colonies of bacteria [1] to herds of animals [2]. They underlie the fascinating motions of bird flocks [3], [4] and fish schools [5] as well as pedestrian track patterns and traffic flows [6]. In these different cases, the motion of the individual organism is very complex and difficult to describe in a detailed way. However, simple models that captures important features of the interaction have proven useful for the description of collective movements. For instance, that car drivers reduces their speed when car density increases is a key property for traffic jam formation. At the level of cells, collective motion is an important component of different biological processes in multicellular organisms [7]. It is an integral part of development [8], as illustrated for instance by dorsal closure in Drosophila embryo, maintenance processes such as wound healing [9], and disorders with cancer as a prime example [10]. It has been studied in vivo, in model systems such as border cell migration in drosophila oogenesis [11], [12] or lateral line migration in zebrafish [13], [14], as well as in simpler and more controlled ex vivo experiments where the motion of cells is simpler to record [15]–[21].
Many aspects of the migratory behavior of cells in two dimensions have thus been studied by using the classical “wound healing” scratch assay, in which a confluent epithelium is scratched with a tool such as a pipette cone or a razor blade, so as to mechanically remove a “strip of cells” from the monolayer. The progression of the remaining cells during the “healing” of this “wound” is then observed under the microscope for up to a few days. In previous works [17], [19], [22], we developed and studied a very reproducible version of these experiments in which a portion of the culture plate is masked by microfabricated stencils. Stencils removal unmasks surfaces free of cells. This produces well-defined “wounds” with rectilinear edges and precisely controlled widths and it triggers cell movements. In the subsequent hours, cells invade the free surface under the apparent guidance of “leader” cells [15], [17], [22].
Our understanding of the mechanisms that coordinate the behavior of multiple cells in these different processes is far from complete. A model of collective cell motion should be useful to try and precisely describe these diverse phenomena. It should also allow to test and quantify the effect of different perturbations [18], [23]. A pioneeringly simple description of the collective behavior of self-propelled particles in general has been proposed by Vicsek et al [24] based on interacting and stochastically moving particles. Several authors have since carefully analyzed this model [25]–[27] as well as related ones [28]–[30]. Coordinated motion of active cells has been modelled along this line [31]–[34] or with more extended cell descriptions [35], [36] as well as with continuum descriptions [37]–[39]. While these previous models provide insights in coordinated cell motion, continuum models do not account for the stochastic character of individual cell motion and simplifying assumptions, such as the use of discrete time and/or velocities of fixed-modulus, in other models [24], [25], [31], [40], prevent detailed comparisons to experimental data. Thus, our aim here is to obtain a minimal model that quantitatively describes coordinated cell motion. We compare the developed model to motion of cells recorded in our experiments, as obtained from Particle-Image Velocimetry (PIV). We find, using numerical simulations that the model accounts quite precisely for the collective cell movements studied at early times before the appearance of leader cells. We further incorporate fast-moving leader cells and determine the conditions under which they guide collective motion as in the experiments.
Our aim is to describe in the simplest quantitative fashion the collective motion of cells in an actively moving epithelium. We wish in particular to take into account that cells are actively motile [16]–[18], that the motion of a cell is stochastic [16], [41], [42] and that it is influenced by interactions with its neighbors. We also wish to obtain a model of minimal computational complexity that can provide a description of a large population of moving cells. A particle-based model appears best suited to this task. We thus propose and study a model of stochastically moving objects biased by their interactions with their neighbors. Each cell is reduced to its center point, the dynamics of which is described by a Langevin-like equation in continuous-time. The velocity of the cell is a real two-dimensional vector which evolves as,(1)where the summation on the right-hand-side (r. h. s.) is performed on the cells that are the nearest neighbors of cell . Thus, the behavior of a cell in our model is only influenced by by its closest neighbors, that is cells that are supposed to be in direct contact with it. We describe here the model main characteristics (see Methods for implementation details). We consider cells that are actively motile and explore their environment in a random fashion [16], [41], [42]. This is modelled in a classical way by a noisy drive (), here described as an Ornstein-Uhlenbeck process with correlation time . The noise amplitude is first taken to be a constant, . It is then generalized to a decreasing function of the local cell density to describe the dependence of the mean cell speed on cell density. The linear damping term () is meant to account in an effective way for dissipative processes coming from rupture of adhesive contacts or friction with the substrate or other cells. Finally, the motion of cell is influenced by its interaction with a neighbor in two ways. First, its velocity tends to become equal to the velocity of the neighboring cell [24], [25] with a strength determined by the coupling constant . Second, the fact that cells do not overlap and have a maximal extent is taken into account by the force between neighbor cells and , which is repulsive with a hard-core at short distances and attractive at longer distances [40], [43]. These interactions are sketched in Figure S1.
Stencil removal in the experiments rapidly increases cell motility in the whole epithelium. Complex displacement fields are observed that can precisely be measured by PIV analysis as described in previous works [17], [19] and shown in Figure 1A. The histogram of the velocity component normal to the epithelium border is identical to the histogram of , the velocity component parallel to the epithelium border as seen in Figure 2 A,B, showing that cell motion is isotropic at early times. After a couple of hours, leader cells appear and guide cell invasion of the free surface. Cell motion then becomes dissymmetric along the and axes.
The model with a constant noise amplitude, , was simulated with a number of model cells (N = 4000) comparable to the number of cells in the experiments. Its parameters were adjusted to the experimental data at early time (30 min after stencil removal) by fitting correlation functions computed from model simulations to experimental velocity fields provided from PIV analysis, as described in Methods. With the obtained parameters, the simulated and experimental velocity field appeared very similar as can be seen in Figure 1. This was quantitatively assessed by comparing different statistical quantities for the model and experimental data. The and histograms closely match and are both well described by the same gaussian (Figure 2 A, B and Figure S2 A, B which displays the data plotted in log-linear coordinates, to better show the histogram tails). Similarly, the distribution of the velocity moduli is close to the corresponding Maxwellian distribution (Figure 2 C and and Figure S2 C). The experimental equal-time spatial velocity correlation (Figure 2 G, H and Figure S2 D, E) as well as the velocity field auto-correlation (Figure 2 I, J and Figure S2 F, G) are both well fitted by the model. The similarity of the correlation of the and velocity components in these plots makes further apparent the isotropy of the cell dynamics at early times. The velocity correlation length is remarkably long of the order of 150 or about ten cells, as noted previously [17], [19], [21]. The cell velocity auto-correlation decays with a time scale of about one hour which quantifies the time during which cells maintain their velocity. This time scale is comparable to the characteristic organization time of microtubules [44] and to the 50 min that we previously measured [22] for the reorientation of the microtubule organizing center relative to the nucleus.
The model provides a good description of the statistical structure of the cell velocity field, while correctly accounting for the spatial relations between neighboring cells (Figure 2 D, E, F). It reproduces the correlation function of cell positions, that is the probability density of finding a cell center at a distance of another cell (Figure 2 D) as well as the distribution of distances between neighboring cells (Figure 2 F). Finally, taking the center of a cell as origin, the rotation angle between the positions of two of it successive neighbors is shown in Figure 2 E. The average angle is as it should, and the whole angle distribution is seen to be very similar for the real and model cells (Figure 2 E). In summary, we find that the interacting particle model described by Eq. [1] with a constant noise amplitude succeeds in capturing quite precisely cell dynamics in the epithelium bulk. As shown in Figure S3, the model continues to describe well the distribution and correlation of the cell velocity component parallel to the band border in the epithelium bulk for a few hours. However as time passes, the border motion influences more and more the motion of cells in the epithelium bulk as shown by the progressive departure of distributions and correlations from their initial values (see Figure S3).
In spite of the model simplicity, it is difficult to obtain exact expressions for the statistical quantities displayed in Figure 2. In order to better understand the influence of the different parameters, we considered the analytically solvable approximation of the model obtained by computing the time evolution of cell velocities as given by Eq. [1], but with the cell positions fixed at the vertices of a triangular lattice, as described in Text S1. The obtained expressions for the distribution of cell velocities and for the velocity correlation functions approximate that of the full model and describe their dependence on different parameters. As shown in Text S1, the noise amplitude determines the cell speed scale but does not influence the shape of the cell speed distribution or the normalized velocity correlation function. Figure S4 illustrates the influence of the parameters and on the velocity correlation functions. An increase of diminishes both the spatial and temporal extent of the velocity correlations. An increase of increases correlation spatially but has almost no effect on temporal correlations. On the contrary, an increase of increases correlations in time but has a very weak influence on their spatial extent.
In the above-described experiments, cell density does not strongly vary in the epithelium bulk at early time. The mean cell speed however depends on the cell density [45] as shown by its decrease as cells reach confluence before stencil removal (see Figure 1A in ref. [19]), as well as by the observed correlation between cell speed and cell density in migrating bands that exhibit large density heterogeneities. In order to account for this effect, we generalized the model to include a dependence of the noise amplitude on the local cell density (see Methods), as written in Eq. [1], since is the main model parameter that controls the mean cell speed. As shown in Figure S5, the inclusion of this dependence does not significantly change the agreement between model and experimental data at early times. It plays however an important role in the epithelium motion at later times, as described below.
Since the proposed model described well the coordinated motion of cells at early times, we investigated whether it could also reproduce the behavior of the epithelium during the whole duration of the experiments. After a couple of hours, leader cells appear and are observed to guide MDCK cell motion at the epithelium border in the form of “fingers” that invade free space, as described in previous works [15], [17], [22] and shown in Figure 3. Different suggestions have previously been made as to the origin of fingers and leader cells. Proposed mechanisms include diffusion and chemo-attraction [37], [39] as well as an intrinsic instability [46] of the cell border hypothesized to be driven by an increased border speed in its outward curving parts [47]. Leader cells are about three times larger than their followers, with a size of the order of 50 as compared to 15–25 for other cells. They also move faster than other cells, do not divide and are often binucleated (see Figure S6). Thus, we here take the more conservative viewpoint that leader cells have acquired different characteristics from other cells in the epithelium. We study whether the observed fingers and border movement can be reproduced by introducing a few modified cells in our model.
The dynamics of leader cells differ in several ways from those of other cells. The velocities of leader cells are found to be constant in modulus and direction to a good approximation (see Figure 7 in ref. [17]) with velocity moduli peaked around . Contrary to other cells, leader cells display a very active lamellipodium along their whole membrane in contact with the free surface. They actively invade free surface while other cells do not. This difference in explorative behaviors may stem from the actin cable that follows the epithelium border [48], [49] and is only interrupted in leader cells. Finally, the motion of a leader cell is not independent of that of other cells. Cutting a leader cell from the following cells strongly perturbs its dynamics. Its motion becomes erratic and it regains its characteristics only after re-adhesion to the epithelium [22].
We introduced model leaders cell in our simulations that took into account these different properties in a simple way (see Methods for details). Leader cells were created as faster cells at the epithelium border with a fixed outward predetermined velocity. The difference in explorative behaviors between leader and other cells was accounted for, in an effective manner, by a repellent force felt by non-leader cells at the epithelium border upon exploration of free surface. This repulsion disappeared as soon as the surface had been explored by a leader or another non-leader cell (see Methods for details). In addition, a leader cell was assumed to coordinate its motion with the cells directly following it. Namely, a leader cell was assumed to slow down when it was too fast for its followers.
Fingers produced with these prescriptions resemble those observed in experiments as shown in Figure 3 A, B and in Figure S7. In the observed experimental fingers, the cell density is lower than in the epithelium bulk. It decreases continuously from the finger bases to their leader cell tip, as quantified previously [22] (Figure 3 C). A very similar trend is observed in the fingers produced in the model as shown in Figure 3 D. The agreement between the experimental and model finger densities is actually surprisingly close given that the model only accounts for the increased spreading on the surface of fast moving cells but does not explicitly include more specific facts such as the more elongated shapes of cells in the fingers.
The effects of the different assumed properties was assessed by relaxing or modifying some of them. Feedback on the leader cell from its followers was found necessary to prevent detachment of the leader cell from the epithelium bulk. Narrow finger-like protrusions on the interface were only present when differences in explorative behaviors between leader cells and other cells were taken into account. Similarly, the noise amplitude dependence on local cell density provided a mean for follower cells to autonomously reach a high speed. In the model with a constant noise amplitude, the leader cell was observed to slow down. It adapted to the “unrestrained” border progression speed that follower cells would adopt in absence of free surface repulsion which is of the same order as cell speed in the bulk (i.e. about ). The increase of cell speed with decreasing cell density greatly increased the unrestrained border progression speed, as shown in Figure S8. This allowed leader cells to maintain a high speed when coordinating their motion with follower cells. A high border progression speed could in principle come from other mechanisms. We noted for instance that with a density-independent noise term, it could arise from the addition of cell division to the model which created an internal pressure in the epithelium. In the experiments, this alternative mechanism is probably not dominant since it is opposed by the fact that, in the epithelium bulk at high density, dividing cells have smaller areas and undergo contact inhibition [45].
Having obtained a quantitative model of cell motion in the epithelium bulk and of cell entrainment by a leader cell, we investigated whether the appearance of a number of leader cells would be sufficient to account for the motion of the whole epithelium border. The rate of appearance of leader cells was measured in time and space along the epithelium border, in the experiments. It was found to be approximately constant in time and uniform along the border with a value of in the first 20 hours after stencil removal and a lower value of at later times (see Figure S9). It was also approximately uniform in space except that the probability of a leader cell appearance within a lateral distance of of another leader cell was found to be very low.
Simulations of the cell model were thus performed with the measured rate of leader cell creation (see Methods).The direction of leader cell velocities was taken normal to the initial epithelium border. The moduli of their velocities were drawn according to a gaussian distribution with parameters determined by the measured leader cell velocity distribution. With a fixed number of cells, the epithelium border speed initially increased. However, the border progression then slowed down and eventually stopped when cells reached their maximal size (given in the model by the attractive part of the cell-cell interaction potential). Cell division was thus incorporated in the model. In order to avoid creating spurious internal pressure in the epithelium, a potential division was implemented only when it did not increase the cell density above the initial one (see Methods).
As shown in Figure 4, the simulated border shapes and movements closely resemble the experimentally observed ones (see also Video S1 and Video S2). The mean border progression also quantitively agrees with the experimentally measured ones as shown in Figure 4. Both display an early regime in which the mean border position grows as where is the time elapsed since the unmasking of the free surface, as reported previously [17]. This is followed by a later regime in which the epithelium border mean position moves approximately linearly in time i.e. at a constant speed. The previous results and model actually provide a simple explanation of both regimes. Without leader cells, the epithelium border invades the free surface at a low speed. Each new leader cell that appears entrains at its higher progression speed a portion of the border the lateral extent of which is of the order of the velocity correlation length. Therefore, the mean speed of the border progression increases with the appearance of each new leader. This speed increase is linear in time for a constant rate a leader cell appearance, resulting in the time progression of the border (see Text S1 for mathematical details). Crossover to the second regime takes place when the leader cell creation rate becomes low and the number of fingers increases much more slowly (see Figure S9).
Finally, we computed the mean component of the cell velocity normal to free border, at different times and at different distances from the mean border position. As shown in Figure 5, the experimental and model velocity profiles change as time evolves but both sets are very similar both in amplitude and scale over the whole time course of the experiment.
In summary, the developed model of cell motion in the epithelium bulk reproduces well the epithelium motion over the whole time course of our experiments, upon addition of leader cells with suitable properties.
Collective motion is a remarkable feature of the dynamics of different organisms, and simple models that appear to capture the essence of this phenomenon have attracted a lot of interest [24], [26]. Data collection is actively pursued for various types of coordinated movements (e. g. [3]) but detailed comparisons between models and experiments [2], [50] are still relatively scarce. We have developed a simple model to try and describe collective cell motion in an epithelium. The quantitative agreement between simulations and the experimental data demonstrates that our description based on the stochastic motion of interacting particules is indeed able to accurately capture the coordinated movement of cells. This agrees with a previously made analogy between cell motion in an epithelium and the dynamics of a complex fluid [21]. We have furthermore shown that adding to these interacting cells, model leader cells with suitable properties, reproduce the motion of the epithelium and of its border over the whole duration of our experiments. The results provide a simple explanation for the previously-reported different regimes of border progression at early and late times. The model should therefore prove helpful to better analyze the consequences of interactions between cells and of their perturbations in different contexts. It will hopefully also be of some use in more complex in vivo situations in which cell motion can be monitored [32], [51], [52].
Existing models of collective cell motion can be classified into three broad categories : model that include an extended description of the cell membrane, based for instance on a Potts model description [35], [36] or a vertex description [45], particle-based models [31]–[33] as the one here studied, and finally continuum descriptions [37]–[39] of tissue movement which do not explicitly describe individual cells. These three levels of description are complementary and each one has its own merits. Detailed cell descriptions allow for a more easy inclusion of biochemical and biophysical mechanisms while reduced ones are computationally more efficient, usually make use of less parameters and are easier to analyze mathematically. The model of interacting particles that we have analyzed in the present work, shares several general features with previously proposed models. The description is based as in refs. [31]–[33] on self-propelling particles that repel their neighbors when they are too close and attract them when they become more distant. The model includes a velocity alignment term as initially proposed in ref. [24]. However, to match the statistical properties of cell velocities in our experiment, the model departs from previous ones in significant ways. The motion of particles is not deterministic [33] but stochastic. Moreover, the model does not include a preferred cell speed and without added nonlinearity, a symmetry broken phase with spontaneously aligned velocities is precluded in our model. In this respect, it qualitatively differs from the family of models studied in ref. [24]–[27], [29]. It remains to be seen whether further nonlinearities will be needed to describe cell motion in different biological conditions. Some directions studied in extended cell models appear worth studying in extensions of the present model. For instance, inclusion of a cell area variable in our particle description would allow one to take advantage of the detailed model of cell division and contact inhibition proposed in [45]. Similarly, it has been found worth distinguishing cell velocities and cell polarization in extended cell models [35], [36]. Our model instead includes a memory of past velocities as in some single cell models [42]. A comparison of these two approaches should prove useful to the understanding of cell behavior in collective migration modes.
Several other features of cell motion in the proposed model deserve further attention. It should prove interesting to see how the model effective parameters emerge from more basic properties and, for instance, to elucidate whether the velocity alignment between a cell and its neighbors arises from adhesion, repeated encounters, signalling or a mix of these different processes. We have found that fingers comparable to experiments are produced when leader cells more actively invade free environment than following cells and also regulate their motion according to their contacts with following cells. The first property is reminiscent of the known role of leader cancer cells in three-dimensional geometry in degrading and remodelling the surrounding matrix to generate tracks for their followers [53], [54]. The second appears to accord with photo-ablation experiments which show that following cells provide important feedback for proper leader cell motion [22]. However, both properties need to be further investigated. We have introduced leader cells without specifying what induces a cell to become a leader. Determining the role in this transformation of chemical signalling and interface geometry and mechanics would allow one to relate the present model to previous proposals [37], [46]. The frequent appearance of bi-nucleated leader cells and the marked change of their creation rate after 20 hrs also point toward a role of the cell cycle that needs to be further investigated. Finally, experiments have started to classify on a large scale how different genes affect collective motion and to cluster them in different modules [18], [23], [55]. Further work is needed to see how these correlate with the few parameters of a simple model such as the one presented here.
The model describes a collection of particles moving according to Eq. [1] of the main text. The noise term that drives the motion of cell is taken to be an Ornstein-Uhlenbeck process with correlation time (2)with a delta-correlated white noise independently drawn in each cell (). The force exerted by cell onto cell is taken under the form(3)with and the potential chosen as the sum of a repulsive short-range gaussian potential and an attractive part acting at longer distances,(4)with , and the Heaviside function for and otherwise. In the simulations of cell motion in the epithelium bulk (Figure 1 and 2), the attractive part of the potential played no role and was omitted. In some exploratory simulations, the velocity alignment coupling was chosen to be a decreasing function of the velocity difference between adjacent cells. However, this did not bring significant improvement to the fit between model and data and a constant coupling term was chosen, as described in Eq. [1].
Neighbors of a cell/particle needed to be defined to implement Eq. [1]. For computational efficiency, this was done as follows. The neighborhood of a particle was split in 6 equal sectors (with the -axis taken as one of the sector boundaries). The particle closest to particle in each sector was taken to be a neighbor of particle if its distance to particle was less than 100 . The local density at cell was computed as , with the average distance between a cell and its 6 neighbors (for sectors without neighbors the cut-off distance of is taken). The distribution of angle between neighbors of a cell in Figure 2E was obtained by computing the absolute value of the angle between the vectors and for each cell center position , and every pair of center positions and of its successive neighbors (i.e. neighbors of the chosen cell that are also themselves neighbors).
As described in the text and supplementary material, simulations were performed either with a constant noise amplitude or with a noise amplitude that increased with decrease in local cell density where is the initial cell density in the band.
After free surface unmasking, the epithelium border was straight and parallel to the y-coordinate axis. The epithelium border was defined in the simulations at later times, by taking the particle with the largest -coordinate in successive bands of width in the direction, covering the simulation space. Leader cells were introduced in the simulations, by randomly transforming into a leader cell a particle within of the epithelium border and the y-coordinate of which was not within of an already created leader cell. The leader cells were created along the border length at an otherwise uniform rate in space and time of during the first 20 hours and at later times. The leader cell velocity was chosen parallel to the -axis (i.e. normal to the epithelium initial border) with a modulus drawn according to a gaussian distribution of mean 18 m/h with a standard deviation of . The velocity of a created leader cell was maintained constant at its initial velocity as long as it had 4 neighbors or more. Otherwise, it was given the mean velocity of its neighbors (with a maximum speed of ) until the threshold number of 4 neighbors was reattained.
As described in the text, normal cell explore free surface less easily than leader cells. This was modelled by introducing a repulsive force exerted on normal cells upon invasion of unexplored surface, as follows. The unexplored surface was covered by surface particles with a repulsive force between pairs of particles (representing cells) and surface particles, closer than . This force was added to the r. h. s. of Eq. [1] for the cell particles. The surface particles were assumed to disappear upon exploration of free surface by a particle associated to a leader or another cell. This was implemented as follows. A scalar ‘damage’ variable was associated to the surface particle . It was chosen to obey(5)with . The summation on the r.h.s. of Eq. [5] was over the forces applied on the surface particle by the cells in its neighborhood , namely cells at a distance of the surface particle smaller than . A surface particle was chosen to disappear when its damage variable reached the threshold value with .
The surface particle-normal cell interaction was modeled by(6)where was taken to be of gaussian form with values of parameters and . The same form was used for the force exerted by leader cells upon surface particles but with and a much larger amplitude such that surface particles were quickly damaged by leader cells. Note that as the dynamics of leader cells was prescribed, there was no need to consider the force exerted by upon them by surface particles.
In the simulations with free-surface, cell division was implemented with a cell doubling time of . However, in order to avoid creating overpressure in the epithelium, a cell division was accepted only if it did not increase the local cell density above the initial cell density . This resulted in modest increases in the number of cells during the course of the simulations. For instance in the simulation displayed in Figure 4 B, D, F the number of cells was initially , at and finally, at .
Simulations were performed with custom computer codes. For parameter fitting and early time simulations, particles were used on a square with periodic boundary conditions, so that the cell density matched that of the experiments. For leader cells and border progression, particles were used on a rectangle with periodic condition in the short -direction.
In order to determine the parameters that best fitted the experimental data at early time ( after mask removal; Figure 1 and 2), simulations served to compute, for given model parameters, both the equal time spatial velocity correlation function and the particle velocity autocorrelation normalized by their value at 0 (as indicated by the subscript ). This was compared to the experimental data for the normalized cell equal-time velocity correlation and for the normalized autocorrelation of the cell velocity field as provided by PIV (see below) by computing the mismatch .(7)with and i.e. time and space intervals over which the experimental correlations significantly differed from zero. The minimization was performed with the Nelder-Mead simplex Amoeba algorithm [56]. The starting parameters were obtained from fitting in the same way the analytical expressions obtained for the model approximation with fixed particle positions (SI text section I). In this approximate model, noise amplitude does not influence the normalized correlations and it was determined from the mean cell speed after the determination of the other parameters. In the full model, the starting noise amplitude was taken equal to the one determined in the approximate model. It was recomputed to fit the mean cell speed after the determination of the other parameters by the Amoeba algorithm. The procedure was iterated until convergence. We compared the cell-velocity auto-correlation in the model to the auto-correlation of the velocity field in the experiment because the first is the most natural quantity for the model and the second a direct output of the PIV analysis of the experiment. In principle, these two quantities can differ since the first one corresponds to following a given cell and the second one to measure cell velocity at successive times at a fixed time of space (i.e. the first is a Lagrangian quantity whereas the second is an Eulerian one). To check that they were not significantly different in the present case, we computed the velocity field in the simulation by assigning to the center of each square of a space-covering grid ( or , the mean velocity of particles in the considered square. As shown in Figure S10, for the two grids, the model velocity field auto-correlation was not found to be significantly different from the particle velocity auto-correlation. We also computed the cell velocity auto-correlation in the experiment by tracking individual cells after stencil removal. The obtained cell auto-correlation was also not found to significantly differ from the velocity field auto-correlation (Figure S11).
MDCK cells [57] were cultured in Dulbecco's modified Eagle's medium supplemented with 10% FBS (Sigma), 2 mM L-glutamin solution (Gibco) and 1% antibiotic solution (penicillin (10,000 units/mL), streptomycin (10 mg/mL)). Cells were seeded and maintained at , 5%CO2 and 90% humidity throughout the experiments.
Microstencils were made of PDMS elastomer (Sylgard 184, Dow Corning) and prepared by classical microlithography as described elsewhere [17]. Experiments were performed in plastic six-well plates on the bottom of which microstencils were previously deposited. Cells were plated on the microstencils and cultured in the incubator until they reached confluence. At this time, the microstencils were peeled off.
Time lapse acquisitions were performed on an automated inverted microscope (Olympus IX71) equipped with temperature, humidity and CO2 regulation. Displacements of the stage (Prior Scientific), and image acquisition (CCD camera (Retiga 400, QImaging), shutter (Uniblitz)) were computer-controlled through Metamorph (Universal Imaging). The delay between two successive frames was 5 min or 15 min. Experiments were performed for typically 40 hours.
The images were processed with ImageJ [Rasband W.S. 2007. ImageJ (National Institutes of Health, Bethesda), available at http://rsb.info.nih.gov/ij/] using a watershed plugin (available at http://bigwww.epfl.ch/sage/soft/watershed/index.html) to extract the contours when needed. The velocity field in the monolayer was mapped by PIV analysis using the MatPIV [58] software package version 1.4 for MatLab (MathWorks Inc.) [Matpiv is a GNU public license software (www.math.uio.no/jks/matpiv/)] as previously described in [19]. The data in Figure 1 and Figure 2 were obtained from squares of in the center of bands of cells of .
|
10.1371/journal.pgen.1008155 | Enhancing face validity of mouse models of Alzheimer’s disease with natural genetic variation | Classical laboratory strains show limited genetic diversity and do not harness natural genetic variation. Mouse models relevant to Alzheimer’s disease (AD) have largely been developed using these classical laboratory strains, such as C57BL/6J (B6), and this has likely contributed to the failure of translation of findings from mice to the clinic. Therefore, here we test the potential for natural genetic variation to enhance the translatability of AD mouse models. Two widely used AD-relevant transgenes, APPswe and PS1de9 (APP/PS1), were backcrossed from B6 to three wild-derived strains CAST/EiJ, WSB/EiJ, PWK/PhJ, representative of three Mus musculus subspecies. These new AD strains were characterized using metabolic, functional, neuropathological and transcriptional assays. Strain-, sex- and genotype-specific differences were observed in cognitive ability, neurodegeneration, plaque load, cerebrovascular health and cerebral amyloid angiopathy. Analyses of brain transcriptional data showed strain was the greatest driver of variation. We identified significant variation in myeloid cell numbers in wild type mice of different strains as well as significant differences in plaque-associated myeloid responses in APP/PS1 mice between the strains. Collectively, these data support the use of wild-derived strains to better model the complexity of human AD.
| Despite the rise in incidence of Alzheimer’s disease (AD), it has been over a decade since a new drug treatment has been introduced. Recently, a number of pharmaceutical giants have shut down their AD research units. One issue that these companies and researchers have struggled with is the lack of translatability of preclinical studies to the clinic. One aspect that has come under heavy scrutiny is whether the mouse can be an appropriate model for a complex human disease such as AD. Current mouse models of AD have incorporated well-known early onset AD mutations on a single genetic background, C57BL/6J, which does not develop all features of human AD- namely marked neurodegeneration. Here we sought to improve the utility and translatability of mouse models through the use of three genetically distinct, wild-derived inbred mouse strains, CAST/EiJ, WSB/EiJ and PWK/PhJ. These mice encompass millions of genetic differences that have never before been explored in the context of modeling AD. Wild-derived mice that carried the early onset AD mutations exhibited robust differences in immune response to amyloid, evidence of mixed pathology and early neurodegeneration, better recapitulating what happens in human AD than previous models.
| Alzheimer’s disease (AD) is the most common cause of adult dementia, with approximately 6 million Americans diagnosed with either clinical AD or mild cognitive impairment in 2017[1]. Age is the greatest risk factor and currently we have the largest aging population that has ever been on this planet [2]. Globally, there are 50 million people living with dementia and this number is expected to reach 152 million by 2050. Low- and middle-income countries are the hardest hit, comprising 66% of global cases [3]. AD is pathologically characterized by the accumulation of beta amyloid (β-amyloid) plaques, neurofibrillary tangles, and widespread neuronal loss. Another prominent feature is the neuroinflammatory response by a variety of cells including astrocytes and microglia. Multiple studies have identified two forms of AD: familial AD (FAD, also known as early-onset AD) and sporadic AD (also known as late-onset AD). Widely-used mouse models of AD utilize FAD mutations in amyloid precursor protein (APP) and presenilin 1 and 2 (PSEN1 and PSEN2). However, a recent review on the current status of AD clinical trials has suggested that the failure of these trials, in part, is due to the inability of current AD mouse models to translate to humans [4]. While FAD mouse models have been vital to understand early drivers of amyloidosis, to date, they do not effectively model all hallmarks of AD, particularly frank neurodegeneration. This has led some to question the utility of mouse models as preclinical models for AD and other diseases of complex etiologies.
Studies including the Dominantly Inherited Alzheimer’s Network (DIAN) and the Religious Order Study and Memory and Aging Project (ROSMAP) show significant variation in age of onset and rate of disease progression in individuals who inherit the same FAD mutations[5, 6]. Furthermore, new work performing a genome-wide association study (GWAS) [7] on individuals with a family history of AD identified multiple novel variants. This suggests that the underlying genetic contribution of many cases of FAD are also due to multiple interacting variants, not simply the single strong variants such as those in APP, PSEN1 and PSEN2. Therefore, the same is likely true in mouse models.
Murine models relevant to AD have been almost exclusively developed on a single genetic background, C57BL/6 (B6). Few studies have assessed FAD mutations in a limited number of alternative genetic backgrounds including 129S1/SvImJ [8], A/J and DBA/2J (D2) [8–10]. These studies showed genetic background altered β-amyloid deposition and seizure incidence, but modifications to neuronal cell loss were not reported. Supporting the potential of incorporating genetic variation in AD mouse models, a recent study used F1 crosses between B6 and thirty classical inbred strains to show that the phenotypes observed from a heterozygous null mutation related to neurological function were not generalizable across strain [11]. Interestingly, there were multiple cases in which there were inverse effects of the same allele on phenotypic outcomes. Another recent publication showed greater transability of the mouse to human Alzheimer’s through the development of a new mouse panel known as the AD-BXDs. This panel was developed by crossing congenic B6 5xFAD mice with BxD males (B6xD2), greatly increasing the genetic diversity in the context of 5 aggressive familial mutations. Aging and characterization of these mice indicated a greater range in AD related phenotypes such as plaque pathology and cognitive deficits, and a greater transcriptional overlap with human AD [12]. These studies highlight the likely huge potential for generating more translatable AD mouse models through the use of different genetic contexts.
Therefore, to take full advantage of the level of natural genetic variation available in mice, we employed genetically distinct wild-derived strains. Historically, the lineage of commonly used classical laboratory strains can be traced to domesticated fancy mouse stock developed on a farm in Massachusetts in the early 1900s [13]. Due to this, classical laboratory strains are undefined genomic mixtures of two or more subspecies of Mus musculus (including Mus musculus domesticus and Mus musculus molossinus). They exhibit limited inter-strain polymorphisms (less than 5 million differences between a classical inbred strain when compared to B6/J ([14] and Fig 1), and do not represent any animal that exists in nature. To overcome the limitations of classic laboratory strains, ‘wild-derived’ strains were introduced as research models in the 1980s. Wild-derived strains are genetically distinct subspecies of Mus musculus (e.g. Mus musculus musculus and Mus musculus castaneous). Founders of each strain were caught from well-established wild mice populations from around the world (see methods), and then inbred [15]. Wild-derived strains show a much greater degree of genetic variation compared to B6 than other classical inbred strains do (between 6 and 17 million differences) including millions of private variations. Importantly, the genetic variation encompassed in these strains and interactions of different gene networks evolved, thus, are likely physiologically relevant to the natural world. This variation includes genes previously associated with AD including Apoe, Trem2 and Tyrobp, and these strains also show variation in phenotypes relevant to AD risk factors including cardiovascular health [16], insulin secretion [17–19], gut microbiota [19, 20] and circadian rhythm [21].
In this study, we hypothesized that incorporating FAD mutations into genetically distinct, wild-derived mouse strains would establish more clinically-relevant AD mouse models compared to those on classic laboratory strain backgrounds. To test this, two commonly used FAD mutations (APPswe and PSEN1de9, herein referred to as APP/PS1) were introduced into three wild-derived strains representative of the three Mus musculus (mus) subspecies: WSB/EiJ (WSB, M. mus domesticus), PWK/PhJ (PWK, M. mus musculus), and CAST/EiJ (CAST, M. mus castaneus). Assessment of AD-relevant phenotypes showed that the effects of the APP/PS1 transgenes are strain-dependent and sex-dependent, with significant differences in amyloid deposition, neuronal cell loss and cerebral amyloid angiopathy (CAA). Transcriptional profiling and neuropathological assessment suggested myeloid cell responses are major contributors to the variation in AD phenotypes we observed in the wild-derived AD strains.
Three wild-derived AD mouse models were created by backcrossing for at least six generations the APP/PS1 transgenes from B6 to the genetically distinct substrains WSB, PWK and CAST (Fig 1). The presence of both the APPswe and PSEN1de9 transgenes was confirmed by PCR (S1 Fig). For each strain, balanced cohorts of female and male wild type (WT) and APP/PS1 mice were established and aged to 6 months–an age window when the majority of plaques have seeded and are in an exponential growth phase in B6.APP/PS1 [22–24]. APP/PS1 and randomized WT litter mate controls from each strain were tested sequentially in the following order: (1) PWK, (2) WSB, (3) CAST and (4) B6. For this first characterization of these new strains, a set of metabolic and functional assays were selected that spanned across a wide-range of AD-relevant phenotypes. Significant strain-, sex- or genotype-specific differences were observed in body weight, body temperature and body composition (S2 Fig). In addition, significant differences were observed in activity measured using both piezoelectric floor monitoring and open field arenas (S3 Fig). WT mice from all three wild-derived strains were significantly more active than B6 WT mice. Also, irrespective of genetic context, all APP/PS1 strains showed the previously reported increase in activity [25] compared to their WT counterparts.
Cognitive function was assessed in wild-derived strains and B6 using spontaneous alternation (working memory) and novel spatial recognition (short-term memory) in a Y-maze. Given the increased behavioral ‘wildness’ [26, 27] of the wild-derived strains, the Y-maze was modified to include specially fabricated covers (see Methods) to minimize likelihood of escape. This was the first time that these tasks had been employed by us for either aging B6 mice or wild-derived strains of any age. However, these tasks had been previously validated using young B6 male mice [28] and further validated here using PWK (the first wild-derived strain to be tested). For spontaneous alternation (S4A–S4C Fig), percent alternation exceeded 50% for all strains irrespective of genotype. Despite hyperactivity phenotypes observed in open field in wild-derived mice, there were no transgenic-related differences in activity levels as measured by total arm entries, thus, increased activity does not confound the interpretation of this task. Furthermore, we found no correlation between number of arm entries and performance. Therefore, these data suggest working memory was not affected by the APP/PS1 transgenes. For novel spatial recognition (S4D–S4F Fig), strain-, sex- and genotype-specific differences were observed. For the PWK strain, a robust preference for the novel arm after a 30-minute delay was shown for both male and female WT and APP/PS1 mice indicating an intact short-term memory. In contrast, for WSB females and CAST males, WT but not APP/PS1 mice showed a preference for the novel arm suggesting working memory was impaired in both female WSB.APP/PS1 and male CAST.APP/PS1 mice. Highlighting the challenges of identifying tasks that can be performed by diverse strains, short-term memory using this task could not be determined for male WSB.APP/PS1, female CAST.APP/PS1, and male and female B6.APP/PS1, as the strain-matched and sex-matched WT counterparts were unable to perform the task.
Next, to assess neurodegeneration, NEUN+DAPI+ cell counts were performed across all strains, sexes and genotypes in a region of the superior cortex and in the CA1 region of the hippocampus, two brain regions commonly affected early in human AD (Fig 2, S1 Table). Interestingly, even in the absence of the APP/PS1 transgenes, strain background was a significant driver of the overall neuronal cell number in the CA1 region. Importantly, there was a significant loss of NEUN+DAPI+ cells in female WSB.APP/PS1 in the cortical region and CA1 compared to WT WSB females. There was also significant loss of neurons in male and female CAST.APP/PS1 mice in the CA1 region. There was no detectable NEUN+DAPI+ loss in either B6.APP/PS1 (as previously published in [10, 29, 30]) or PWK.APP/PS1 strains in the two regions studied. Despite the presence of neurodegeneration in CAST.APP/PS1 and female WSB.APP/PS1, there was no evidence of increased tau pathology using AT8, a marker of early tau hyperphosphyloration (S5 Fig).
Amyloidosis was assessed in all four strains using ThioS staining, ELISA and Western blotting. Surprisingly, numbers of cortical ThioS+ plaques were significantly decreased in all three of the wild-derived APP/PS1 strains in comparison with B6.APP/PS1 (Fig 3A–3C, S2 Table). Numbers of hippocampal ThioS+ plaques were also significantly decreased with the exception of WSB.APP/PS1 females. No plaques were observed in WT mice from any of the four strains in any brain region. Plaque morphology appeared different between B6.APP/PS1 and wild-derived APP/PS1 strains. Specifically, there was an absence of small ThioS+ plaques in wild-derived APP/PS1 compared to B6.APP/PS1 mice. Despite the reduced numbers of plaques, there was a significant increase in Aβ42 (measured by ELISA) in both female CAST.APP/PS1 and WSB.APP/PS1 compared to B6.APP/PS1 (Fig 3D). This increase cannot be accounted for simply by differences in mutant APP production as Western blotting using 6e10 (antibody to human mutant APP) showed similar APP protein levels across all strains (Fig 3E) with the exception of male PWK.APP/PS1 (significant difference between male B6.APP/PS1 and male PWK.APP/PS1, p ≤ 0.01). Therefore, our data suggest that at 8 months, plaques in the wild-derived APP/PS1 strains may be further along in the rapid growth period previously defined for B6.APP/PS1 mice [24].
Another prominent amyloid phenotype observed in the wild-derived strains was ThioS+ vessels, suggesting the occurrence of cerebral amyloid angiopathy (CAA). Brain sections from all strains, sexes and genotypes were examined for the presence of ThioS+ vessels and by silver staining. CAA was pronounced in vessels of CAST.APP/PS1 and WSB.APP/PS1, but not B6.APP/PS1 or PWK.APP/PS1 mice (S6 Fig). There was no evidence of vascular staining of ThioS in WT animals. CAA has been associated with cerebrovascular damage in human AD and recent studies support a more prominent role of cerebrovascular decline in AD pathogenesis [31, 32]. To test the relationship between CAA and cerebrovascular integrity, brain sections from WSB.APP/PS1 were assessed as they showed the greatest percentage of ThioS+ vessels. Cerebrovascular integrity was determined using antibodies to fibrin(ogen), a protein that is ordinarily present in blood but its presence in the brain is indicative of blood brain barrier compromise. Fibrin was present outside of the microvessels in brain sections from WBS.APP/PS1, but not in B6.APP/PS1 (Fig 4).
To provide insight into the strain-specific differences that may be driving the phenotype differences observed between strains, transcriptional profiling by RNA-seq was performed on the left brain hemispheres from WT and APP/PS1 male and female mice from all strains (93 samples in total). Sequencing depth (S7 Fig) and expression levels of the APPswe and PSEN1de9 transcripts in APP/PS1 mice (S8 Fig) were consistent between strains. Principle Component Analysis (PCA) identified strain as the greatest driver of gene expression variance across samples, consistent with the genetic distinctness of strains (Fig 5A). To identify modules of genes that were differentially expressed between groups, Weighted Gene Co-expression Analysis (WGCNA) was performed. The majority of modules were driven by strain, independent of APP/PS1 genotype (S9 Fig). However, one module (termed ‘light yellow’) was driven by APP/PS1 genotype and seen in all strain backgrounds (Fig 5B). This module contained 35 genes that are enriched for the Lysosome and Osteoclast Differentiation KEGG pathways (Fig 5C). The light yellow module included App and Psen1 supporting the fact that this module is likely an amyloid response module. The majority of other genes in the module are expressed in myeloid cells (either resident microglia and/or monocytes/macrophages). Many of these genes have been previously implicated in AD-relevant processes such as amyloid deposition and synaptic loss including C1qa, Csf1r, Tyrobp, Cx3cr1, Cd68 and Ctsz. Importantly, DNA variations in two genes in the light yellow module, Trem2 and Cd33, have previously been associated with human AD suggesting these genes may be early drivers of AD pathogenesis.
Assessment of the eigenvalues for the light yellow module revealed two major findings. First, there was great variation in the eigenvalues when comparing WT samples between strains. For instance, eigenvalues were lowest for WT samples from WSB and CAST. This was reflected in the normalized expression levels of genes in the module. Trem2, Tyrobp and Ctss showed the lowest expression in WT samples from WSB and CAST (Fig 5E)–strains that showed neuronal cell loss in the presence of amyloid (Fig 2). The second major finding was that there were marked differences in eigenvalues comparing WT to APP/PS1 samples. The greatest difference between WT and APP/PS1 samples was observed in PWK. Again, these differences were also observed at the level of individual genes within the light yellow module (Fig 5B). This suggests the ability of myeloid cells to respond to amyloid is strongly influenced by genetic context. Together, these data suggest that there are intrinsic differences between myeloid cells in WT samples from different strains and that these cells respond differently to amyloid deposition. Both these factors are likely critical in determining whether or not a strain is susceptible to amyloid-induced neurodegeneration.
A major and unexpected finding from the transcriptional profiling was that transcript levels of myeloid cell genes were significantly lower in WT WSB and CAST mice compared to B6 and PWK mice. This suggests that myeloid cells vary between strains, even in the absence of amyloid. To test this, IBA1+ myeloid cell numbers were determined. There was a significant difference in the numbers of IBA1+DAPI+ cells in WT mice of different strains (Fig 6). WT mice from CAST and WSB mice showed significantly fewer IBA1+ cells compared to B6 (Bonferroni’s multiple comparison test vs B6: Male WSB p ≤ 0.0001 and CAST p ≤ 0.01; Female WSB p ≤ 0.01 and CAST p ≤ 0.05). This supports the transcriptional profiling data (Fig 5). As expected, there was a significant sex and region-specific increase in IBA1+ cells in mice carrying the APP/PS1 transgenes compared to their WT counterparts (Fig 6, S3 Table).
Transcriptional profiling also predicted that plaque-mediated myeloid cell responses would differ between strains. To assess this, the numbers of myeloid cells surrounding plaques were determined for each APP/PS1 strain. For each mouse, the numbers of IBA1+DAPI+ cells (myeloid) were determined around five plaques of similar relative size in 6 mice per strain (a total of 30 plaques/strain, Fig 7A and 7B). The median number of IBA1+ myeloid cells per section was averaged per animal and then compared across strains. For male animals, CAST.APP/PS1 had the greatest number of plaque-associated IBA1+ cells, while WSB.APP/PS1 had the least. For female animals, CAST.APP/PS1 exhibited the greatest number of plaque-associated IBA1+ cells.
One myeloid cell response that has recently been highlighted as important is proliferative capacity, and there remains a debate regarding whether this is helpful or harmful in response to injury or in progression of neurodegenerative diseases [33, 34]. CAST.APP/PS1 mice showed the greatest numbers of myeloid cells around plaques despite having the fewest numbers of myeloid cells in WT animals (Fig 6). To determine whether this could be due to myeloid cell proliferation, the proliferative marker KI-67 was used. KI-67+IBA1+ cells in CAST.APP/PS1 mice were compared to B6.APP/PS1 mice. There were significantly more plaque associated KI-67+IBA1+ cells observed in CAST.APP/PS1 compared to B6.APP/PS1 mice (t(14) = 3.73, p = 0.002) (Fig 7D). This suggests underlying differences in the proliferative capacity of myeloid cells between strains, and may be a factor in neuronal cell loss exhibited by CAST.APP/PS1.
The work presented here highlights the value and power of increased genetic diversity within mouse models in order to gain insight into the complex etiologies of human disease. Our work shows that in contrast to the B6.APP/PS1 strain that has been widely used historically, CAST.APP/PS1, WSB.APP/PS1 and PWK.APP/PS1 represent models that provide a new lens to understanding central features of human AD including amyloid-induced neurodegeneration, neuroinflammation, cerebrovascular integrity and cerebral amyloid angiopathy. Overall, there were three major findings: (1) Female WSB.APP/PS1 showed significant hippocampal and cortical neuronal cell loss, whole brain elevated levels of Aβ42, and cognitive impairment in a short-term memory task. This was accompanied by substantial vascular amyloid deposition in the form of cerebral amyloid angiopathy accompanied by vascular compromise. (2) CAST.APP/PS1 showed hippocampal cell loss and females exhibited whole brain elevated levels of Aβ42. (3) Transcriptional profiling corroborated by neuropathology identified strain-dependent baseline differences in the expression of neuroinflammatory (primarily myeloid-related) genes and in the magnitude of their response to amyloid. Based on these findings, we predict that a major driver of the phenotype differences observed between strains (e.g. neuronal cell loss and CAA) is due to differences in neuroinflammation, particularly myeloid cells. A substantial part of the observed variation in myeloid cell-driven inflammation has been linked to genetic differences between human populations [35].
Wild-derived AD mouse models appear to show important strain- and sex-dependent differences in behavior and pathology that are similar to the human clinical population that show both sex-specific and ethnic differences in terms of prevalence and progression [36]. However, a major challenge of this study was to develop a functional battery that could be used across the strains as they exhibit formidable differences in wildness compared to classical laboratory strains. Wildness score is comprised of measurements of jumping, escape, struggle, squeaking and biting, and strains like B6 and D2 earn a score ranging between 0.21 and 0.66, while wild-derived strains range from 1.35 for CAST [26] to 2.5 or greater for WSB [27]. While it was important to include a range of functional assays, we anticipated there could be issues as traditional behavioral assays have been primarily optimized for typically behaving mice (e.g. young male B6) and likely would not be optimal for testing wild-derived strains. While B6.APP/PS1 mice have previously been shown to exhibit deficits in assays such as Contextual Fear Conditioning as early as 6 months [37], we chose to avoid aversive tasks due to the inherent difficulty in handling wild-derived mice and stress caused to the animals with repeated handling. Instead, we chose tasks that utilized the animal’s natural exploratory drive such as y-maze tasks that assess spatial memory. While spatial memory deficits in B6 or B6/C3H mice carrying the APP/PS1 transgenes are typically over 12 months of age [38, 39], it would be expected that earlier impairment would be identified in a sensitized genetic context. To our knowledge, this is the first time that many of the tasks included in our battery were used to assess wild-derived mouse behavior, thus, all data is left intact (no outliers removed) and presented as individual data points. Unfortunately, in this study, not all WT strains were able to perform the novel spatial recognition task, which may have been due to the length of the memory delay chosen (30 minutes). This meant that short-term memory could not be assessed for some APP/PS1 strains. Therefore, more extensive functional assays are still required at multiple ages to determine the utility of these strains for studies of cognitive impairment. Given our experiences in this study, it may be necessary to develop and validate strain-specific cognitive assays due to inherent differences in wildness and age-dependent cognitive abilities. This will be of particular importance in tasks that may require food restriction as these strains have vastly different metabolic rates.
In the age window tested, there were significant differences in plaque numbers and Aβ42 levels between strains. Taken alone, plaque counts would suggest less amyloid in the wild-derived APP/PS1 mice in comparison with B6.APP/PS1. However, the size distribution of plaques varied across the strains, with B6.APP/PS1 exhibiting many smaller proximal deposits that may correspond to initial seeding of Aβ. This is in contrast to plaques in the brains of wild-derived APP/PS1 mice that were of moderate size. This could be indicative of a more advanced stage of amyloid deposition in 8 month wild-derived mice carrying APP/PS1 in comparison with B6.APP/PS1 of the same age; and/or, suggest the presence of different conformations of amyloid fibrils. Previous work has shown that identical peptide sequences are capable of forming into different conformations of amyloid fibrils, and that this difference can be detected by seeding efficiencies [40].
The development of cerebral amyloid angiopathy (CAA) in WSB.APP/PS1 coupled with fibrin leakage (Fig 4) suggests compromised vascular integrity and/or deficits in amyloid clearance. It is possible that cerebrovascular damage (measured here by fibrin leakage) is downstream of amyloid. Conversely, an inherent weakness in cerebrovascular structures in WSB mice may dispose mice to CAA. While CAA has been reported many times before in AD models carrying APP mutations on a B6 background, typically it does not appear with complete banding until mice are over 14 months of age [41–43]. Furthermore, the severity of CAA and risk of associated microhemorrhage progresses with age. There is strong evidence to suggest that the earliest predictors of AD-susceptibility and onset are related to vascular and blood-brain-barrier integrity [44]. The presence of severe CAA and neurodegeneration in WSB.APP/PS1 will allow mechanistic dissection of the relationship between CAA and neurodegeneration.
We found that strain is the greatest driver of gene expression variation in these mouse models, even more so than sex or APP/PS1. This is representative of the inclusion of millions of genetic differences created from wild-derived strains that have never before been explored in context of modeling AD. Mus musculus, also known as the house mouse, are characterized as being commensal animals, meaning that they live in close association with humans, and even though they are able to adapt to a wide-range of environments, are dependent on human shelter or activity for their survival [15]. Each distinct subspecies is from different geographical regions (CAST was trapped in Thailand, WSB was trapped in eastern shore Maryland, USA and PWK was trapped in the Czech Republic), and evolved separately to survive alongside humans in the face of similar region-specific pressures (exposure to pathogens or infection, climate, diet etc.). Therefore, some of the genetic differences between mouse substrains may correspond with genetic variants in different populations of humans. More likely however, the variations driving the phenotype differences will impact similar genes/pathways that are modified by genetic risk variants in the human population. In support of this, many of the genes in the module identified by WGCNA (Fig 5D) have previously been implicated in human AD–including sporadic AD–either through genetic association, gene expression studies or functional studies. Therefore, despite the presence of the APP/PS1 transgene artificially driving amyloid accumulation in these strains–the responses appear to be directly relevant to human AD. This suggests that interventions tested in these new AD strains that target factors downstream of amyloid deposition but upstream of neurodegeneration would be expected to be clinically relevant to FAD and LOAD.
PWK.APP/PS1 is particularly intriguing as transcriptional data suggest it is the greatest responder to amyloid at 8 months and appears to be a resilient strain (no neuronal cell loss detected). These data may be consistent with a slower progression/transition from amyloid deposition to neuronal cell dysfunction which could become apparent at older ages, or representative of a neuroprotective signature. Two additional and striking phenotypes may also be reflective of the substantial neuroinflammation in PWK.APP/PS1 mice. First, during generation of the experimental cohort, PWK.APP/PS1 had to be separated from WT littermates at 3 months of age due to increased aggression. Second, changes in activity in the piezoelectric chambers were observed in female PWK.APP/PS1 (S3 Fig). These may be behavioral manifestations of increased neuroinflammation in response to amyloid. Agitation and circadian disruption are clinical symptoms that directly interfere with the ability of caregiving to occur in the home and have both been linked with neuroinflammation in humans [45, 46].
Transcriptional profiles in WT animals suggested that in comparison with B6 and PWK, CAST and WSB show lower baseline expression of the primarily myeloid-related genes in this module. Cell counts confirmed that there was ~50% reduction in the number of IBA1+ cells in both CAST and WSB. Natural inherent differences in neuroinflammation is important given the lack of studies into how genetic variation impacts glial cell development and homeostasis in the human population–an area that might be critical in predisposing to age-related diseases such as AD. Similarly, while there have been renewed efforts to characterize the immune systems in wild-caught mice [47], there still remains a dearth of knowledge regarding how genetic variation impacts myeloid cell and astrocyte function in inbred wild-derived strains. This may be starting to change as recent work by Christopher Glass and colleagues [48] analyzed macrophages from five inbred mouse strains, including PWK. As in our study, strain was the greatest driver of differences in gene expression in these macrophages.
Much of the foundation of mouse genetics has been focused on examination of a single genetic difference while holding all other genetic (i.e. strain background) and environmental influences constant. Somewhere along the way, limited resources and a wise desire for standardization restricted this examination to only one or two laboratory strains, despite efforts more than 4 decades ago to develop mouse resources such as the wild-derived strains and periodic suggestions of researchers past to expand beyond one strain [49–51]. Our study represents one of few studies to utilize natural genetic variation in mice to gain further insight in human AD. For the first time, we show neurodegeneration and mixed pathology in wild-derived strains carrying the APP/PS1 transgenes. Interestingly, our data suggests B6 is a ‘resilient’ strain when considering neurodegeneration. This ‘resilience’ may be specifically driven by differences in myeloid-related neuroinflammation, and we predict that differences in myeloid cell biology in these new wild-derived AD mouse models will provide a much-needed platform for identification of novel genes/variants modifying susceptibility to neuronal cell loss.
One caveat of these new strains is that amyloid is driven by the APP/PS1 transgenes. Transgenic overexpression of proteins can provide additional side effects and mutations in APP and PSEN1 which may not be ideal to uncover the mechanisms of sporadic AD. A second caveat is that, despite neuronal cell loss, they appear to lack overt TAU pathology (S5 Fig). Therefore, further work is still needed to improve both the construct validity and the face validity of these new mouse models. Research is now focused on inducing sporadic AD in mice in the absence of transgenic overexpression of familial AD mutations. Mice differ from humans in both the APP and TAU proteins. The human APP protein is generally considered to be more amyloidogenic than the mouse and the ratios of the 3R and 4R isoforms of TAU are balanced in human adults, but not in adult mice. This may be a contributing factor to the apparent lack of TAU pathology in wild-derived APP/PS1 strains. Multiple efforts, including our own, are improving the construct validity of AD mouse models by humanizing the App and Mapt loci and incorporating sporadic AD-relevant variants such as APOEE4 and TREM2R47H. The development of gene editing technologies such as CRISPR make these approaches feasible. However, our study and others [12] show it will be important that these efforts incorporate genetic diversity and natural genetic variation to improve both the face and predictive validity of these new mouse models.
All research was approved by the Institutional Animal Care and Use Committee (IACUC) at The Jackson Laboratory (approval number 12005). Animals were humanely euthanized with ketamine/xylazine mixture. Authors performed their work following guidelines established by the “The Eighth Edition of the Guide for the Care and Use of Laboratory Animals” and euthanasia using methods approved by the American Veterinary Medical Association.”
All mice were bred and housed in a 12/12 hours light/dark cycle on pine bedding and fed standard 6% LabDiet Chow. Experiments were performed on four strain genetic backgrounds: C57BL/6J, CAST/EiJ (JAX Stock #000928), WSB/EiJ (JR#001145), and PWK/PhJ (JR#003715). B6.Cg-Tg(APPswe, PSEN1dE9)85Dbo/Mmjax (JAX stock #005864), and referred to in this study as B6.APP/PS1 mice, were obtained from the Mutant Mouse Resource and Research Center (MMRRC) at The Jackson Laboratory and backcrossed with the three different mouse strains for at least 6 generations to produce: CAST.APP/PS1 (JAX Stock #25973), WSB.APP/PS1 (JAX Stock #25970) and PWK.APP/PS1 (JAX Stock #25971).
Generation of experimental cohorts consisted of 12 mice of each sex and genotype (APP/PS1 carriers and littermate wild-type controls). Due to increased pup mortality in the wild-derived strains, once determined to be pregnant, female mice were removed from the mating and housed individually. During this time, they were also given BioServ Supreme Mini-treats (Chocolate #F05472 or Very Berry Flavor #F05711) in order to discourage pup cannibalism. Animals were initially group-housed during aging and then individually housed at the start of the behavioral testing battery. Due to severe aggression in PWK.APP/PS1 mice, these mice were individually housed earlier at 3 months of age. There was also some cohort loss throughout the behavioral battery (i.e. seizure lethality mainly in B6.APP/PS1), so individual data points are shown for all assays where appropriate. Due to a substantial loss in the first cohort of male CAST.APP/PS1 during the behavioral battery, a second cohort was generated and tested. Data were analyzed independently and combined if there were no significant differences in groups. For post-mortem characterization of AD phenotypes, brains from 6 males and 6 females at 8 months (±2 weeks) were assessed, with the exception of female B6.APP/PS1, where only an n of 4 survived to harvest.
A behavioral and physiological battery was designed in to order to test a wide range of AD-relevant phenotypes (S1 Fig). All testing was conducted by trained technicians in the Center for Biometric Analysis at The Jackson Laboratory. APP/PS1 strains were scheduled as they became available and on average, the battery took about 6 weeks to complete. Testing always involved blinding and randomization of all littermates. An animal’s data was excluded from analysis if there was an indication from the technician that it should be (i.e. due to animal escaping prior to placement in assay, equipment failure, etc.). Summary tables report n’s used in analyses and individual data points are shown in all plots. Animals were then directly taken from the facility to the laboratory for harvesting.
Piezoelectric floor monitoring (Signal Solutions, Lexington, KY) is a non-invasive, high throughput method of assessing sleep patterns through measurement of breath rate to classify animals as either awake or asleep [52, 53]. The piezoelectric pad is located on the cage floor and is sensitive to respiratory patterns. Pressure on these sensors converts analog input into an electric/digital signal. The system monitors activity by measuring the amplitude of the electrical signal, and comparing this to a signal threshold in order to classify the animal as either awake or asleep. Data was exported as either hourly percent activity or as hourly activity bout length over 5 days. Activity for the first day was excluded to allow for animal acclimation, and then averaged across sex and genotype.
Stress induced Hyperthermia is an assay developed to detect typical stress responses determined by elevation in body temperature as the sympathetic nervous system is activated. Disruptions in this response can be indicative of metabolic dysfunction and/or an anxiety phenotype. The day prior to testing, mice were individually housed in standard cages. On testing day, animals were brought into the testing room and allowed to habituate for 60 minutes. Body temperature was taken at two time points separated by a 10-minute delay via a glycerol lubricated thermistor rectal probe (Braintree Scientific). In between readings, mice were placed back into the home cage.
Open field is a measure of exploration and motor activity. Introduction of APP/PS1 on the B6 background has been correlated with hyperactivity, thus, this is an important measure for the battery to ensure that strain-specific differences in other tasks cannot be accounted for by hyperactivity alone. The apparatus used for this test was a square chamber (~40 x 40 x 40 cm) fabricated from clear Plexiglas and illuminated at 400 lux. Data are recorded via a sensitive infrared photobeam system in 5-minute time bins for a 60-minute trial length.
Spontaneous Alternation Y-maze task is a widely used task to assess spatial working memory, and relies on the animal’s natural exploratory behavior [54]. Sequence of entries into each arm of the Y-maze is tracked to assess if animals demonstrate intact working memory. The maze is a y-shaped arena constructed of Plexigas with equal arm lengths (~30 cm), arm lane width (~6 cm) and wall height (~15 cm). Special arm covers were fabricated to ensure that wild-derived mice could not escape and were used for all strains. A black curtain surrounded the perimeter of the maze in order to minimize additional room cues. A camera mounted above recorded mouse exploration and tracking software (Noldus Ethovision) allowed the export of multiple variables such as sequence of arm entries. Mice were allowed to habituate to the testing room for 60 minutes prior to testing. Activity was recorded over an 8-minute period. Each maze was wiped out with 70% ethanol between animals. A correct alternation represented when animals entered three different arms of the Y-maze without returning to a previously visited arm. The initial two arm entries were subtracted from the total to account for the placement of the animal in arm A at the start of the trial. Percent correct was determined by dividing the number of correct alternations by the adjusted total arm entries throughout the trial. This means for our assay, performance at chance is calculated at 22%. Typical performance of a young B6 male is approximately 50% [55].
Spatial working memory y-maze task was conducted at least 2 weeks after spontaneous alternation in the same y-maze arena. This task consisted of two trials separated by a 30-minute delay period. During trial 1, the start and familiar arms were available for the animals to explore for 10-minutes and featured two distinct intramaze cues. Animals were then returned to their home cage for the 30-minute delay. In trial 2, all three arms were open for exploration and time spent in each of the arms over the 5-minute period was calculated. An animal was determined to have intact short term memory if it spent significantly more time in the novel arm in comparison with the start and familiar arms. Between trials and between animals the maze was cleaned with 70% ethanol. An additional exclusion criteria for this task was exclusion for failure to explore both arms during trial one (≤20% time percentage per arm).
Body composition measurements were collected at the conclusion of the battery if a significant weight difference was detected. Animals were weighed and placed into a Plexiglas tube of 2.5 inches diameter and 8 inches in length. This tube is placed in a nuclear magnetic resonance device (EchoMRI, Houston, TX) that uses a 5 gauss magnet in order to pulse a magnetic field in a gradient across the animal to determine body composition consisting of lean muscle, fat and water. Each scan lasts approximately 1–3 minutes. Upon conclusion, animals are returned to their home cage and the tube is cleaned with 70% ethanol.
A lethal dose of ketamine/xylazine was administered to mice by intraperitoneal injection, in accordance to IACUC protocols. After transcardial perfusion with 1X PBS (Phosphate buffered saline) brains were removed. The left hemisphere was snap frozen for RNA/protein isolation, and the right hemisphere was fixed in 4% paraformaldehyde for sectioning. Protein was extracted with Trizol Reagent (Life Technologies cat#15596–018) following manufacturer's guidelines. Pellets were resuspended in a solution of 1:1 8M urea and 1% SDS.
For tissue sectioning, following 48 hours in 4% paraformaldehyde, half brains were kept at 4°C and placed in 10% sucrose for 24 hours. The tissue was then placed in 30% sucrose for an additional 24 hours, or until it sank. Brains were then embedded in optimal cutting temperature (OCT) compound, sectioned at 25μm and stored at −80°C until required.
Primary antibodies were applied to 1xPBS washed brain sections and incubated for two nights at 4°C. The following primary antibodies were used to characterize neuronal and glial cells phenotypes in the brain: rabbit polyclonal anti-NeuN (1:200, Cell signaling), chicken polyclonal anti-GFAP (1:200, Acris Antibodies), rabbit polyclonal anti-IBA1 (1:200, Wako), and mouse polyclonal anti-Ki-67 (1:200, eBioscience). Primary antibodies were diluted in PBT (1X PBS, 1% TritonX-100) containing 10% normal donkey serum. After primary incubation, sections were washed three times in PBT and incubated with their respective secondary antibody (donkey anti-chicken Alexa Fluor 633 or donkey anti-rabbit Alexa Fluor 488/594, 1:1000 dilution, Life Technologies) for 2 hours at room temperature. All sections were then counterstained with DAPI and mounted with Aqua PolyMount. For Thioflavin S staining, sections were incubated with 1% Thioflavin S (diluted in a 1:1 water: ethanol ratio) for eight minutes at room temperature, followed by three washes in 80% ethanol, 95% ethanol, and finally in dH2O, and mounted. For assessment of cerebral amyloid angiopathy, X34 (100 uM, Sigma Aldrich) occurred first. Slides were brought to room temperature, washed with 1XPBS for 5 minutes and then incubated with 500 ul of the X34 solution. Slides were then dipped in deionized water and incubated with 500 ul of 0.02M NaOH for 5 minutes. After an additional wash with 1XPBS, primary antibodies rabbit polyclonal anti-fibrin (1:200, Millipore) and goat polyclonal anti-IBA1 (1:300, Abcam) were diluted in PBT and applied to the slide. The remainder of the protocol overlaps with that described above, with the exception of DAPI staining. For assessment of tau, X34 steps occurred first, followed by the use of Mouse on Mouse Basic Kit (Vector Laboratories). Tissue was then incubated ON in primary antibody Mouse Phospho-Tau (Ser202, Thr205) monoclonal antibody (AT8, 1:250, Thermofisher), goat anti-IBA1 (same as above) and rabbit Anti-NEUN (Abcam, 1:500) diluted in 10% normal donkey serum and PBT. Secondaries used overlap with description above. To ensure AT8 staining worked, a positive control slide of a 13 month hTau (B6.Cg-Mapttm1(EGFP)Klt Tg(MAPT)8cPdav/J, JR# 005491) was stained alongside APP/PS1 strains. Images of IHC were taken using either the Leica SP5 confocal microscope, Leica SP8 confocal microscope or the Zeiss Axio Imager Z2.
Quantitative analyses of plaques, microglial cells, astrocytes and neurons were performed in WT and APP/PS1 mice of each of the four strains. The number of plaques present in the entire cortical region from three central sections for each mouse was determined. To quantify the number of IBA1+ microglia and GFAP+ astrocytes, 12 images (20X, 1388 X 1040 microns) were taken for each brain (for cortex: 3 images/section for 3 sections– 9 images in total; for hippocampus: 1 image in CA1/section for 3 sections– 3 images in total) with a Zeiss Axio Imager fluorescent microscope, and cells were manually counted using the cell counter plugin from ImageJ (1.47d) software. For counting NEUN+/DAPI neurons in the parietal cortex, three images (20X, 447 X 335 microns) were randomly taken in similar areas for each brain from each mouse, images were stacked using ImageJ and cropped altogether to 274.13 X 225.75 microns (including only cortical layers II and III). For quantification of pyramidal neurons in the hippocampus, images of the CA1 region were taken at 20X (447 X 335 microns) and cropped to 225.75 X 129.00 microns. NEUN+/DAPI+ cells in the cortex and hippocampus images were manually counted with the cell counter plugin from ImageJ (1.47d) software. All image analyses were performed blind to the experimental conditions. For IBA1+ cells surrounding plaques, five plaques per brain were imaged (using 20x optical lens). Images were processed and cells counted using the cell counter plugin for ImageJ/FIJI. For each mouse, IBA1+ cells around each plaque from the three images were totaled and then averaged across mice.
Protein levels of human APP were assessed using western blotting. Briefly, DC assay (Bio-Rad) was used to determine protein concentration. Protein samples were heated to 95°C for 5 minutes and a total of 10 ng of protein was loaded onto a 12% TGX stain-free gel (Bio-Rad). Protein samples were transferred to a nitrocellulose membrane (Life Technologies) using an iBlot (Thermo Fischer Scientific) following manufacturers’ instructions. Blots were then incubated overnight with 6E10 antibody (1:2000, Covance/BioLegend) and 5% milk/PBS-tween at 4°C. After incubation with the appropriate secondary antibodies (Anti-Mouse IgG 1:30,000, Millipore) for 2hrs at room temperature, ECL detection reagents (GE Healthcare) were used to develop the chemiluminescence signal. Blots were further probed with anti-GAPDH (1:1000, Millipore) after treatment with 0.1% sodium azide and incubation with secondary antibody (Anti-mouse IgG 1:30,000, Millipore) for 2 hours at room temperature, washed, and detected. Quantification of blots was determined using Fiji ImageJ software.
Human amyloid β-42 (Aβ42) levels were determined using the ELISA detection kit from Life Technologies (cat#KHB3442) following the manufacturer’s instructions. To ensure that urea and SDS levels in the protein samples were compatible with the ELISA kit (see protein isolation), protein samples from 8 months old mouse brains were diluted 1:50 in standard diluent buffer. Samples were then compared to a standard curve and Aβ42 concentrations were established against the samples’ protein concentrations following manufacturers recommendations.
Left brain hemispheres (n = 93) were snap frozen at harvest and then samples corresponding to mice for pathological assessment were sent to The Jackson Laboratory Genome Technologies core for further processing. RNA was isolated from tissue using the MagMAX mirVana Total RNA Isolation Kit (ThermoFisher) and the KingFisher Flex purification system (ThermoFisher). Tissues were lysed and homogenized in TRIzol Reagent (ThermoFisher). After the addition of chloroform, the RNA-containing aqueous layer was removed for RNA isolation according to the manufacturer’s protocol, beginning with the RNA bead binding step. RNA concentration and quality were assessed using the Nanodrop 2000 spectrophotometer (Thermo Scientific) and the RNA Total RNA Nano assay (Agilent Technologies).
Libraries were prepared by the Genome Technologies core facility at The Jackson Laboratory using the KAPA mRNA HyperPrep Kit (KAPA Biosystems), according to the manufacturer’s instructions. Briefly, the protocol entails isolation of polyA containing mRNA using oligo-dT magnetic beads, RNA fragmentation, first and second strand cDNA synthesis, ligation of Illumina-specific adapters containing a unique barcode sequence for each library, and PCR amplification. Libraries were checked for quality and concentration using the D5000 assay on the TapeStation (Agilent Technologies) and quantitative PCR (KAPA Biosystems), according to the manufacturers’ instructions. Libraries were pooled and sequenced by the Genome Technologies core facility at The Jackson Laboratory, 100 bp paired-end on the HiSeq 4000 (Illumina) using HiSeq 3000/4000 SBS Kit reagents (Illumina) at an average sequencing depth of ~100 million reads per sample (S1 Fig). 2x100 base length paired end reads were quality trimmed and filtered using Trimmomatic tool [56] and reads passing the quality filtering were mapped to the mouse mm10 reference genome using ‘STAR’ aligner [57]. Custom genomes were generated for CAST, PWK and WSB strains by incorporating REL-1505 variants into the reference genome [58]. RSEM software package [59] was used to estimate expression levels for all genes in Transcripts Per Million (TPM) unit based on Ensembl Release 84 transcriptome. HTSeq Python package was used to calculate raw read counts per transcript.
Genes that have an HTSeq estimated raw read count of less than 10 in more than 90% of the samples were considered noise and excluded from the analysis. Downstream analyses were performed on 17408 genes that passed the read count threshold. We applied variance stabilization transformation (vst) on raw read counts using DESeq2 R package. Principal Component Analysis (PCA) was applied on vst transformed read counts to identify clusters of samples and any potential outliers.
We then applied Weighted Gene Co-expression Network Analysis (WGCNA) algorithm [60] to identify co-expressed gene modules in our dataset. We extracted module’s eigengenes, which are equivalent to the first principal component, to represent the overall expression profiles of the modules. GO pathway enrichment analysis was performed using Homer tool [61].
For behavioral analyses, groups of 12 mice per sex per strain per genotype allowed us to detect effects of greater than 1.15 standard errors at 80% power. Due to some premature death, not all groups contained 12 mice but sample sizes are stated clearly in the associated tables and all graphs show individual data points. For neuropathology and biochemistry, six biological replicates were assessed for all groups with the exception of B6 female APP/PS1 (n = 4) and PWK female WT (n = 5). Individual data points are shown on all graphs and associated tables include sample sizes. For data within strain and sex were analyzed as non-parametric using Mann-Whitney Rank Sum Test to account for departure for normality or using unpaired t-tests. One-way multifactorial analysis variance (ANOVA) followed by Bonferroni post-hoc tests for multiple comparisons were utilized for across strain differences. All statistical tests are labeled within data tables. Data were analyzed using GraphPad Prism software. P values are provided as stated by GraphPad and significance was determined with P values less than 0.05. When significance values were more than two decimal values, P values were presented as follows: p ≤ 0.01 (**),p ≤ 0.001 (***) or p ≤ 0.0001 (****).
These new strains are available through The Jackson Laboratory and all associated data is being made available through GEO and Accelerating Medicines Partnership-Alzheimer’s disease (AMP-AD) knowledge portal.
|
10.1371/journal.pgen.1003498 | RNA–Mediated Epigenetic Heredity Requires the Cytosine Methyltransferase Dnmt2 | RNA–mediated transmission of phenotypes is an important way to explain non-Mendelian heredity. We have previously shown that small non-coding RNAs can induce hereditary epigenetic variations in mice and act as the transgenerational signalling molecules. Two prominent examples for these paramutations include the epigenetic modulation of the Kit gene, resulting in altered fur coloration, and the modulation of the Sox9 gene, resulting in an overgrowth phenotype. We now report that expression of the Dnmt2 RNA methyltransferase is required for the establishment and hereditary maintenance of both paramutations. Our data show that the Kit paramutant phenotype was not transmitted to the progeny of Dnmt2−/− mice and that the Sox9 paramutation was also not established in Dnmt2−/− embryos. Similarly, RNA from Dnmt2-negative Kit heterozygotes did not induce the paramutant phenotype when microinjected into Dnmt2-deficient fertilized eggs and microinjection of the miR-124 microRNA failed to induce the characteristic giant phenotype. In agreement with an RNA–mediated mechanism of inheritance, no change was observed in the DNA methylation profiles of the Kit locus between the wild-type and paramutant mice. RNA bisulfite sequencing confirmed Dnmt2-dependent tRNA methylation in mouse sperm and also indicated Dnmt2-dependent cytosine methylation in Kit RNA in paramutant embryos. Together, these findings uncover a novel function of Dnmt2 in RNA–mediated epigenetic heredity.
| The possibility of a mode of inheritance distinct from the Mendelian model has been considered since the early days of genetics. Only recently, however, suitable experimental models were created. We now see the development of new experimental systems detecting non-Mendelian inheritance in a variety of organisms, from worms to mice. We have previously shown that RNA molecules act as transgenerational inducers of epigenetic variations in mice. We are currently using Mendelian genetics to dissect the factors involved in RNA–mediated transgenerational signalling. By showing an absolute requirement for Dnmt2 in this process, our study extends our knowledge of this still somewhat enigmatic protein. We confirmed that RNA rather than DNA methylation by the protein is involved in epigenetic heredity, and our genetic results indicate a requirement during an early step in the reproductive process, between parental gametogenesis and the preimplantation stage.
| Experimental results on model animals ranging from Caenorhabditis and Drosophila to the mouse have recently provided support for a mode of epigenetic heredity distinct from the canonical Mendelian rules [1]–[10]. These findings may help in understanding unexpected epidemiological results showing paternal transmission of pathological states over several generations [11]–[13] and provide at least partial solutions to the ‘missing heritability’ problem raised by genomic analyses [11], [14]. Several of the current experimental systems point to RNA as the transgenerational signalling molecule, sperm RNA [15] in the case of paternal heredity.
One important example of RNA-mediated inheritance is provided by the mouse paramutation, where transcriptional activation of a locus is mediated by small non-coding RNAs (sncRNAs). These epigenetic variations were first detected by the hereditary maintenance of the white-tail phenotype of the Kit mutation in Kit+/+ offspring of heterozygotes carrying an inactivated allele (KittmlAlf1/+), which was associated with an accumulation of aberrant Kit transcripts in germ cells [8]. These RNAs were thought to play a role in the transgenerational transfer of the phenotype, a conclusion strengthened by microinjection assays in naive fertilized eggs. More specifically, oligoribonucleotides with sequences of the transcripts and Kit-specific microRNAs generated the hereditary modification. Similarly, microinjection in eggs of microRNA miR-1 resulted in overexpression of its target Cdk9 and that of miR-124 in increased expression of Sox9 during the preimplantation period. The miR-1/Cdk9 paramutants developed cardiac hypertrophy [4] and the miR-124/Sox9 variants a giant phenotype and twin pregnancies [10]. In all three cases, the epigenetic variations were stable over at least three generations of outcrosses and paternal transmission was explained by the transfer of sequence-related molecules in the spermatozoal RNA fraction.
A search for genes involved in paramutation led us to consider a role of the Dnmt2 methyltransferase in RNA mediated epigenetic inheritance. In contrast to other members of the Dnmt family, the Dnmt2 protein catalyses cytosine methylation in RNA substrates, an activity which was at first enigmatic, homozygous null mutants of Drosophila, Arabidopsis and mouse being viable and fertile under laboratory conditions [16]. Methylation by Dnmt2 was reported to protect tRNAs from cleavage under stress conditions [17] and to be involved in upholding steady state levels of tRNAs [18]. We now report that a homozygous loss-of-function mutation of the Dnmt2 locus prevents the appearance of epigenetic variants of the Kit and Sox9 loci. Our results indicate that the methyltransferase is not required for expression of the variant phenotype during development. Our data further indicate a Dnmt2-dependent initiation step and suggest a role for Dnmt2 in the homeostasis of sncRNAs in the early embryo.
The white tail and feet of the KittmlAlf/+ heterozygotes (Figure 1A) are immediately recognizable, thus allowing for quantitative studies on relatively large numbers of mice. A non-Mendelian mode of transmission detected in their progenies had initially allowed us to identify a hereditary epigenetic modification of expression of the Kit+ allele (paramutation), which is determined by cognate sncRNAs [8]. We then initiated a search for genes that would affect the establishment and/or maintenance of the paramutated state and considered the Dnmt2 RNA methyltransferase as a possible candidate. We generated 129/Sv mice carrying the heterozygous Kit locus and a Dnmt2 null mutation [16]. The results of crosses between KittmlAlf1/+, Dnmt2−/− parents are summarized in Figure 1A, with a more detailed presentation in Table 1 and Figure S1. The Dnmt2+/+ control crosses yielded the expected frequency of Kit paramutants (Kit+/+ genotype with the white-spotted phenotype of the mutant). In contrast, in the progeny of two Dnmt2−/− parents, segregation of the phenotypes strictly corresponded to the Kit genotype. Crosses with Kit+/+, Dnmt2+/+ mice of the wild type, full tail color Kit+/+, Dnmt2−/− offspring failed to restore the modified state. A role of the genetic background was excluded because the results were reproduced in C57BL/6 and in B6D2F1 hybrids (Table S1).
The regular segregation of the Kit+ phenotype in Dnmt2−/− crosses could have been explained by the selective mortality of variant embryos during development. However, further analysis argued against this possibility. As shown in Table 1 (Exp #2), all the embryos generated in 10 crosses between KittmlAlf1/+ Dnmt2−/− males and Kit+/+ Dnmt2−/− females were transplanted at the one-cell stage into Dnmt2+/+ foster mothers and 75 living births were obtained from 77 transplants. None of the Kit+/+ progenies showed the variant phenotype under these conditions, thus excluding embryonic lethality.
Genetic analysis identified an initial period of establishment of the epigenetic variation. In crosses between KittmlAlf1/+, Dnmt2+/− males and either Dnmt2−/− or Dnmt2+/− females, a fraction of the Kit+/+ Dnmt2−/− offspring showed the white-spotted phenotype (Table 2). This Dnmt2-negative Kit paramutant progeny was generated with a frequency identical to that in Dnmt2+/+ crosses. However, when these mice were subsequently crossed to wild-type partners, they did not further transmit the white tail phenotype. In other words, the epigenetic state was initially maintained in the Dnmt2−/− genotype during somatic development but was heritable only from a parent with an intact Dnmt2 allele. We conclude that Dnmt2 activity is critical during parental gametogenesis and/or in fertilized eggs. The resulting change in Kit expression in early stem cells can then be maintained in melanoblast stem cells in a Dnmt2-independent manner and results in the defect in their migration during early development responsible for the pigmentation of the adult tail.
Dnmt2 is known to be expressed in oocytes and preimplantation embryos [19] and we detected both Dnmt2 RNA and protein in fractionated male germ cells in spermatocytes, round and elongated spermatids (Figure S3). We also analysed the methylation patterns in mouse sperm of the C38 target site in two established Dnmt2 substrates. The results showed high levels of C38 methylation for tRNA(Asp) and tRNA(Gly) in sperm from wild type mice (Figure S4). This methylation was substantially reduced in sperm from Dnmt2−/− mice (Figure S4), which provided confirmation for the enzymatic activity of Dnmt2 in the male germline.
Further support for a role of Dnmt2 in the inheritance of epigenetic variation was obtained from RNA microinjection experiments. We had previously shown that microinjection into naive fertilized eggs of either RNA extracted from KittmlAlf1/+ tissues, or the cognate microRNAs, or oligoribonucleotides with transcript sequences induced the heritable phenotype modification. We then used these assays to compare the efficiency of RNA preparations from Dnmt2−/− and Dnmt2+/+ KittmlAlf1/+ heterozygotes. The results showed that RNA from the brains and testis of Dnmt2-deficient Kit heterozygotes did not induce the modified phenotype (Table 3). In subsequent experiments, an oligoribonucleotide with a sequence from the Kit mRNA (nt 2123–2150, Table S2) also induced the white-spotted phenotype when microinjected into wild-type one-cell embryos (Table 4). We also tested a form of the same Kit oligoribonucleotide in which all cytosines were methylated. This RNA, indicated as ‘Kit2123–2150met’ in Table 4, was more efficient in inducing the modified phenotype in Dnmt2+/+ embryos but inefficient in the Dnmt2-deficient background, indicating a requirement for Dnmt2 expression in the embryo. We conclude that, while methylation of the inducer RNA is required for optimal efficiency, the methyltransferase is still needed in the most early embryonic period.
To extend our analysis to a second example of a mouse paramutation, we tested whether the lack of Dnmt2 would affect the epigenetic modulation of Sox9 which can be induced by microinjection of the cognate microRNA miR-124 and of Sox9 transcript fragments in Dnmt2+/+ embryos [4]. The miR-124/Sox9 variants were characterized by augmented numbers of blastocyst stem cells and, as a result, overgrowth of the embryo and adult body and frequent twin pregnancies. Following microinjection of miR-124 into Dnmt2−/− fertilized eggs, E7.5 embryos were identical to controls (Figure 1B) and not oversized as the Dnmt2+/+ Sox9 paramutants. We concluded that the paramutation of Sox9 is also dependent on Dnmt2 expression.
Modified patterns of DNA methylation have been reported in various instances of epigenetic variation [ref. 20 for review] including the maize paramutation [21]. We used methylated DNA immunoprecipitation (meDIP) to determine the DNA methylation status of the Kit locus in testis DNA from wild type, KittmlAlf1/+ and Kit+/+ paramutant mice. Assays were developed for three distinct regions covering the Kit promoter, exon 2 and exon 14, respectively (Figure 2A). The results indicated only background levels of methylation in the promoter, and substantial methylation in the two intragenic regions (Figure 2B). This pattern was observed for all three genotypes (Figure 2B), indicating that the Kit paramutation is not associated with altered DNA methylation profiles of the locus – although we cannot exclude a localized change in an unknown element at a distance, as described for the b1 paramutation of maize [21]. In parallel experiments, we also used this approach to determine the DNA hydroxymethylation status of the locus and found that hydroxymethylation levels were invariably low in all genotypes and regions tested (Figure 2B). The meDIP findings were subsequently validated by DNA bisulfite sequencing of testis DNA. The results demonstrated that the Kit promoter was completely unmethylated and that the exon 14 region was completely methylated (Figure 2C). This pattern was again observed for all three genotypes (Figure 2C), which further suggests that paramutation is not associated with an altered DNA methylation profile of the Kit locus.
We also used RNA bisulfite sequencing to analyze the possibility that Dnmt2 might methylate Kit transcripts. To this end, we induced the Kit paramutation by microinjection of an oligoribonucleotide (Kit2123–2150) into fertilized eggs obtained from either two Dnmt2−/− or two wild-type parents. In parallel, we also prepared control embryos that were injected with buffer. RNA was prepared from E9.5 embryos and methylation analysis was performed on the 45 cytosines of a region amplified from Kit RNA (nt 2100–2336) that covers the inducer oligoribonucleotide (nt 2123–2150) and overlaps the exon 14∶15∶16 junctions. The results revealed two closely associated cytosines (cytosines #4 and #8, Figure 3) that remained unconverted in a higher fraction of reads, specifically in the microinjected Dnmt2+/+ embryos. Methylation of mRNA by Dnmt2 has not been reported so far and it is possible that our results have been influenced by deamination artifacts. However, the same two methylation sites were identified in three independent biological replicates and were not observed in the control embryos or in the oligoribonucleotide-treated Dnmt2−/− embryos (Figure 3), which suggests that they might represent genuine methylation marks. Contamination by DNA was excluded by the spliced structure of the sequence. Furthermore, we also tested the methylation pattern of the corresponding genomic sequence. The results showed methylated CpG sites that were clearly distinct from the sites detected in RNA and that were not dependent on the Dnmt2 genotype (Figure S5).
The physiological role of the RNA methyltransferase activity of Dnmt2 has been enigmatic for a long period of time. Dnmt2-mediated tRNA methylation has recently been linked to tRNA stability [17], [18]. However, the widespread occurrence of 5-methylcytosine in RNA [22] may reflect a variety of functions, most of which still remain to be identified. The recurrent general considerations on a regulatory role of noncoding RNAs [reviewed in refs 23], [24] led us to consider a possible physiological function of the methyltransferase in epigenetic regulation. The three instances of RNA-mediated hereditary variation that we reported as paramutations provided suitable experimental models.
We now report that Dnmt2 is required for establishment and hereditary transmission of the epigenetic variation at the Kit and Sox9 loci. This was first revealed for the visible color phenotype of the Kit variants, a most classical approach in genetics. It was further confirmed and extended to Sox9 by microinjection experiments. Our data show that the parental RNAs and synthetic oligoribonucleotide inducers of the epigenetic variations were inefficient in Dnmt2-negative embryos. Evidence for RNA methylation in the inducer oligonucleotide sequence was observed in embryos undergoing the Kit paramutation. Furthermore, while the modified Kit phenotype was never observed in Dnmt2 −/− homozygotes born from two parents with the same genotype, it was, however, expressed by genetically identical homozygotes when at least one of their parents was a Dnmt2-positive heterozygote (Table 2). We concluded that the protein is required only during the parental gametogenesis or in the early embryo and not at later developmental stages – except for subsequent transgenerational transfer.
At least two general explanatory models can be considered for the absolute requirement in Dnmt2 in the establishment of the epigenetic change. One model would be based on the knowledge that tRNAs are bona fide substrates of Dnmt2, and that tRNA fragments are highly abundant in mouse sperm [25]. Our data show that at least two tRNAs are methylated in mouse sperm in a Dnmt2-dependent manner (Figure S4), which raises the possibility that methylation-dependent processing of tRNAs [17] could result in the generation of paramutation-inducing sncRNAs. However, we have so far been unable to detect any recognizable phenotypes after microinjection of various tRNAs and tRNA fragments (data not shown). A second model would consider that the inducer small RNAs are maintained only in the Dnmt2+/+ genotype, possibly because they are methylated or complexed with methylated tRNAs. Such a model would also account for the increased efficiency of the methylated synthetic oligoribonucleotides (Table 4). Current preliminary results suggest that exogenous small RNAs introduced in the early embryo are stably maintained only in Dnmt2-positive embryos, leading us to the hypothesis of a protection against endonucleolytic cleavage by methylation in a manner analogous to tRNAs [17]. A control of the maintenance of parental small RNAs at the maternal-zygotic transition would be reminiscent of the mechanisms that, at the same developmental stages, eliminate part of the parental mRNAs [26]. In such a model, the new individual would actively constitute its own set of functional RNAs, both large and small, from the parental stocks.
The experiments here described were carried out in compliance with the relevant institutional and French animal welfare laws, guidelines and policies. They have been approved by the French ethics committee (Comité Institutionnel d'Ethique Pour l'Animal de Laboratoire; number NCE/2012-54). KittmlAlf1/+ heterozygotes were maintained in parallel in the original 129/Sv genetic background and in C57BL/6×DBA/2 F1 hybrids (B6D2). The Dnmt2−/− homozygote [16] was kindly provided by T. Bestor. Originally maintained on a mixed genetic background, the mutation was backcrossed onto 129/Sv, C57BL/6 and B6D2 genetic backgrounds, in each case for more than ten generations. Genotypes were determined by PCR analysis of Neo and LacZ expression and by Southern blot hybridization using a genomic probe.
Total brain and testis RNA and oligoribonucleotides with Kit and miRNA sequences were adjusted to a concentration of 1 µg/ml and microinjected into B6D2 fertilized eggs according to established methods of transgenesis [27]. Quality of RNA preparations from the mouse organs was checked by spectroscopic analysis using the Bioanalyzer 2100 apparatus (Agilent Technologies, Santa Clara, CA) (Figure S2). Oligoribonucleotides were obtained from Sigma-Prolabo (sequences provided in Table S2).
Northern blot analysis was performed by standard methods [28]. For analysis, RNA was extracted with Trizol Reagent (Invitrogen).
Protein extracts for Dnmt2 Western blot were prepared from snap-frozen enriched germ cell populations obtained by homogenization in RIPA Buffer. Testicular fractions were purified by elutriation as described [29]. 20 µg of protein was fractionated onto a 15% denaturing SDS-polyacrylamide gel and transferred to nitrocellulose. The following antibodies were used for immunodetection: rabbit anti-Dnmt2 antibody (Santa-Cruz, Rabbit polyclonal IgH sc-20702, lot: B1903) 1∶100 and rabbit anti-ß-actin antibody (Santa-Cruz, sc-47778, lot: D0907) 1: 250 with peroxidase-coupled goat anti-rabbit secondary antibody (Santa Cruz Biotechnology) 1∶10,000.
Methylated DNA immunoprecipitation was performed as described previously [30]. Sequences of PCR primers are shown in Table S3. DNA bisulfite sequencing analysis was performed by using the EpiTect Bisulfite Kit (Qiagen), in combination with 454 sequencing technology (Roche). Sequences of 454 bisulfite sequencing primers are shown in Table S3 and S4. Methylation maps were generated by BISMA [31].
Analysis of cytosine methylation in Kit RNA was performed as described [32], with minor modifications. RNA isolated using TRIzol (Invitrogen) was digested with DNase (Promega). An aliquot of 6 µg of RNA dissolved in 20 µl of RNase-free water was mixed with 42.5 µl of “Bisulfite Mix” and 17.5 µl of “DNA Protect” buffer. The RNA was denatured at 70°C for 5 min, followed by 1 h incubation at 60°C. This cycle was repeated 5 times. RNA was isolated from the bisulfite reaction mix using the RNeasy Purification Kit (Qiagen) and treated with 0.5 M Tris-HCl, pH 9 at 37°C for 1 h. Finally, RNA was precipitated and further processed for sequencing, as described previously [32]. This included random barcoding during the reverse transcription reaction to confirm that the sequenced DNA molecules represented different RNA molecules. Sequences of PCR primers are shown in Table S3 and S4. Sperm RNA was prepared as described [10] and analyzed as described previously [18].
Data are expressed as means ± s.e.m. A p-value of less than 0.05 was considered statistically significant.
|
10.1371/journal.pbio.0050090 | Transcript Specificity in Yeast Pre-mRNA Splicing Revealed by Mutations in Core Spliceosomal Components | Appropriate expression of most eukaryotic genes requires the removal of introns from their pre–messenger RNAs (pre-mRNAs), a process catalyzed by the spliceosome. In higher eukaryotes a large family of auxiliary factors known as SR proteins can improve the splicing efficiency of transcripts containing suboptimal splice sites by interacting with distinct sequences present in those pre-mRNAs. The yeast Saccharomyces cerevisiae lacks functional equivalents of most of these factors; thus, it has been unclear whether the spliceosome could effectively distinguish among transcripts. To address this question, we have used a microarray-based approach to examine the effects of mutations in 18 highly conserved core components of the spliceosomal machinery. The kinetic profiles reveal clear differences in the splicing defects of particular pre-mRNA substrates. Most notably, the behaviors of ribosomal protein gene transcripts are generally distinct from other intron-containing transcripts in response to several spliceosomal mutations. However, dramatically different behaviors can be seen for some pairs of transcripts encoding ribosomal protein gene paralogs, suggesting that the spliceosome can readily distinguish between otherwise highly similar pre-mRNAs. The ability of the spliceosome to distinguish among its different substrates may therefore offer an important opportunity for yeast to regulate gene expression in a transcript-dependent fashion. Given the high level of conservation of core spliceosomal components across eukaryotes, we expect that these results will significantly impact our understanding of how regulated splicing is controlled in higher eukaryotes as well.
| The spliceosome is a large RNA-protein machine responsible for removing the noncoding (intron) sequences that interrupt eukaryotic genes. Nearly everything known about the behavior of this machine has been based on the analysis of only a handful of genes, despite the fact that individual introns vary greatly in both size and sequence. Here we have utilized a microarray-based platform that allows us to simultaneously examine the behavior of all intron-containing genes in the budding yeast S. cerevisiae. By systematically examining the effects of individual mutants in the spliceosome on the splicing of all substrates, we have uncovered a surprisingly complex relationship between the spliceosome and its full complement of substrates. Contrary to the idea that the spliceosome engages in “generic” interactions with all intron-containing substrates in the cell, our results show that the identity of the transcript can differentially affect splicing efficiency when the machine is subtly perturbed. We propose that the wild-type spliceosome can also distinguish among its many substrates as external conditions warrant to function as a specific regulator of gene expression.
| The coding regions of most eukaryotic genes are interrupted by introns, which must be removed for proper gene expression. The removal of introns requires single-nucleotide precision in order to faithfully convert genomic information into functional protein. The process of intron removal is performed by the spliceosome, a large ribonucleoprotein that catalyzes two sequential transesterification reactions [1,2]. The spliceosome itself is highly conserved across all eukaryotes, consisting of five small nuclear RNAs (snRNAs) and well over 100 proteins [3,4]. A combination of genetic and biochemical experiments have revealed conformational rearrangements in both RNA and protein components that are required for the spliceosome to accurately process pre–messenger RNA (pre-mRNA) transcripts [5]. About 50 of these proteins can be considered core components of the spliceosome in that their activity is required for cell viability in budding yeast, whereas the remainder of the factors can be deleted with little or no effect on cell growth under standard laboratory conditions.
To date, the vast majority of what is known about the mechanism of pre-mRNA splicing has been deduced from experiments using, at most, a handful of transcripts. Yet it remains unknown how well these transcripts represent the behavior of the entire complement of spliceosomal substrates. In higher eukaryotes, where genes are often interrupted by multiple introns, it is known that the spliceosome can utilize specific sequences present in individual transcripts to regulate both quantitative and qualitative aspects of gene expression [6,7]. Two large groups of proteins, the SR and hnRNP families, are known to modulate splicing activity, allowing the spliceosome to generate multiple distinct proteins from a single genomic locus, significantly increasing proteomic diversity. In particular, the sequence-specific RNA-binding members of the SR family improve the processing efficiency of introns containing suboptimal signals at their 5′ or 3′ splice sites. Binding of the SR proteins to enhancer sequences, which may be located in exons or introns, facilitates recruitment of the core machinery to the suboptimal splice sites.
By comparison to higher eukaryotes, splicing in the yeast Saccharomyces cerevisiae appears much simpler in a number of ways. Whereas more than 95% of human genes are interrupted by an intron [8], only about 250 yeast genes, or less than 5% of all genes, contain an intron [9]. Moreover, the vast majority of the intron-containing genes in yeast contain only a single intron, differing significantly from humans, where the average gene is interrupted by eight introns [8]. As seen in Figure 1A, yeast introns tend to conform to very strong consensus sequences at the branch point and at both the 5′ and 3′ splice sites. Perhaps accordingly, only three SR-like proteins have been identified in yeast, none of which is known to regulate splicing activity. On the other hand, while relatively few yeast genes are interrupted by introns, those that are tend to be highly expressed and tightly regulated genes. It has been previously shown, for example, that intron-containing genes account for nearly one-third of total cellular transcription [10]. Notably, several functional categories are overrepresented among intron-containing genes. The most dramatic of these corresponds to transcripts that encode ribosomal protein genes (RPGs): 102 of the 139 RPGs in yeast contain introns. Interestingly, all yeast introns tend to be small relative to those found in humans; the largest yeast intron contains only 1,002 nucleotides, whereas many human introns stretch to well over 10,000 nucleotides. Nevertheless, there is a bimodal distribution of intron lengths that has a strong correlation with gene function; introns found in genes encoding ribosomal proteins tend to be longer, while those in nonribosomal proteins are generally shorter (Figure 1B).
Given the relative simplicity and similarity of intron-containing transcripts in yeast and the limited role of the SR proteins, it has been a reasonable expectation that the yeast spliceosome would have a restricted capacity to distinguish among its different intron-containing substrates. In this case, the behavior of a few individual transcripts could be extrapolated to accurately describe the behavior of most spliceosomal substrates. The advent of splicing-specific microarrays, which facilitate a simultaneous examination of all intron-containing transcripts [11], allows us to directly ask whether the spliceosome interacts similarly with all pre-mRNAs, or whether transcript-specific differences can be identified. Here we have used this approach to examine the splicing responses of all ∼250 intron-containing transcripts resulting from inactivation of 18 different core components of the spliceosome and five factors involved in up- or downstream steps in pre-mRNA processing. Far from being homogeneous in their effects on the different intron-containing transcripts, these mutants reveal a remarkable ability of the core components to differentiate among transcripts. Interestingly, most ribosomal proteins are encoded at two distinct genomic loci in yeast. Whereas the coding sequences are nearly identical at these two locations, the sequences of the introns that interrupt them are generally quite dissimilar. For some of these paralogs, dramatic differences can be seen in response to the different spliceosomal mutations. The capacity of the spliceosome to differentiate among substrates suggests that splicing offers an important opportunity to regulate gene expression in a transcript-specific manner.
To examine genome-wide changes in splicing in yeast, we have designed microarrays similar to those previously described [11] that allow us to distinguish between the spliced and unspliced isoforms of transcripts derived from intron-containing genes. As seen in Figure 2A, three different oligos are used to target each intron-containing gene (see also Materials and Methods). One oligo targets a portion of the last exon and is used to detect changes in total transcript level (probe T). A second oligo targets a region of the intron, allowing us to detect changes in pre-mRNA levels (probe P), while the third oligo targets the junction of the two exons, allowing for a determination of changes in levels of mature mRNA (probe M).
Because there are multiple, independent probes describing the behavior of each intron-containing gene, analysis of a splicing-specific microarray is inherently more complicated than that of a standard expression array. In order to assess the splicing behavior of any given transcript it is important to simultaneously examine the changes in signal on all three feature types. In analyzing data from splicing-specific microarrays, others have generated splicing indices for each gene by compressing the data derived from the different feature types. For example, splicing changes can be represented as a precursor/mature (PM) index by dividing the ratio of the movement of the pre-mRNA feature by the ratio of the mature mRNA feature [11,12]. Although this manipulation simplifies the data, allowing them to be more easily examined in the context of traditional microarray analyses, for reasons described below we find that a more detailed understanding of individual transcript behavior is achieved by considering the independent movement of all three feature types simultaneously.
Figures 2B and 2C show an example of such an analysis derived from a comparison of RNA from a wild-type strain and a strain harboring the temperature-sensitive prp2-1 mutation after both were shifted to the nonpermissive temperature of 37 °C. The spliceosomal Prp2 protein is a member of the DEAD/H box family of RNA-dependent ATPases and plays an essential role in catalyzing an as yet unidentified structural rearrangement required for activation of the spliceosome prior to the first chemical step [13]. As such, the primary result of a defect in Prp2 activity is expected to be a broad decrease in pre-mRNA splicing. The splicing profile resulting from this experiment, where each horizontal line describes the behavior of a single gene, is displayed for both a single timepoint (Figure 2B) and across the time course of inactivation (Figure 2C). A simple prediction for the behavior of a given transcript in an experiment where the activity of a core spliceosomal component is impaired is that the precursor species would accumulate with a concomitant loss of the mature species. Indeed, for many of the transcripts, including those shown with a red bar in Figure 2C, this is precisely the pattern that is observed: precursor species are seen to accumulate to nearly ten times the normal level, and up to 90% of the mature mRNA is lost. For most of these genes, the behavior of the total mRNA probe closely follows that of the mature species, suggesting that the majority of the transcripts for these genes are present in their spliced forms. However, different profiles are apparent for many other genes. For example, many show robust accumulation of precursor species with very little decrease in mature mRNA, including those shown with a green bar in Figure 2C.
The different behaviors displayed by these two groups of transcripts presumably reveal important differences in the complex interplay of cellular machineries that define the steady-state levels of both the precursor and mature forms of each of these transcripts. For example, the rate of precursor accumulation observed for a given transcript depends upon, among other things, the transcription rate of that gene, the degradation rate of the transcript, the rate at which the transcript is normally spliced, and the extent to which its splicing is inhibited. Likewise, the decrease in mature mRNA is a function of both the extent of splicing inhibition and the decay rate of the mature form of that transcript. For those genes marked by the red bar in Figure 2C, the dramatic decrease in levels of mature mRNA may be explained by either a strong block to the production of mature species, a rapid rate of decay of the mature species, or some combination of both. By comparison, a less complete block to splicing or a slower rate of mRNA decay may explain the failure to detect significant decreases in mRNA for those genes marked with the green bar. Importantly, the process of creating splicing indices to describe the behavior of these transcripts in many cases eliminates the ability to differentiate between these scenarios. For example, genes which show both a modest increase in precursor levels and a modest decrease in mature mRNA levels generate nearly identical PM index values as do genes that show a strong precursor accumulation but little decrease in mature levels.
By simultaneously examining each of the feature types as described above, the global splicing profile resulting from the prp2-1 mutation shows that precursors rapidly accumulate for about 80% of the intron-containing genes. While the detailed behaviors of the different transcripts are varied, this result appears to satisfy the simple prediction that defects in this factor would result in defective splicing of all actively transcribed genes. At least two different scenarios may explain the failure to detect precursor accumulation for the remaining 20% of the transcripts. It is almost certainly true that some of the intron-containing genes are transcriptionally quiescent during logarithmic growth conditions. Naturally, no precursor accumulation would be expected for such genes. Alternatively, these transcripts may be insensitive to the particular defect introduced by the prp2-1 mutation.
In order to further distinguish between transcript behaviors, we examined the RNA from strains containing mutations in two additional core spliceosomal components: PRP5 and PRP8. Like PRP2, PRP5 encodes an essential member of the DEAD/H box family of RNA-dependent ATPases. By comparison with Prp2, the activity of Prp5 is required to promote a molecular rearrangement earlier in the splicing pathway, catalyzing the stable addition of the U2 small nuclear ribonucleoprotein to the branch point sequence [14]. As with mutations in Prp2, loss-of-function mutations in Prp5 result in a block to splicing prior to completion of the first chemical step. Notably, Prp5 and Prp2 are only transiently associated with the spliceosome, whereas Prp8 is a stable component of the U5 small nuclear ribonucleoprotein and is thought to form the physical core of the spliceosome [15,16]. As with Prp5 and Prp2, most mutations in Prp8 block spliceosomal activity prior to the first chemical step; however, mutants have been isolated that can perform the first but not the second transesterification reaction [17,18].
As seen in Figure 3A, strains containing the loss-of-function mutations prp2-1, prp8-1, or prp5-1 all exhibit conditional growth phenotypes, showing nearly wild-type growth at 25 °C but an inability to support growth at 37 °C. A comparison of growth at intermediate temperatures allows these mutations to be ordered according to the strength of the defect they impart: prp2-1 is the most severe defect as it is unable to support growth above 25 °C; prp8-1 is slightly less severe as it can support weak growth at 30 °C; and prp5-1 is the weakest as it shows nearly wild-type growth even at 33 °C. Figure 3B shows the time-resolved splicing profiles derived from shifting each of these strains from 25 °C to 37 °C and comparing them to a similarly treated wild-type strain. Importantly, the ordering of the genes is consistent for all three profiles, meaning that the behavior of a single intron-containing gene can be seen by following a single horizontal line across all three profiles.
As expected, inactivation of each of these factors results in inhibition of splicing of a large number of intron-containing transcripts. Interestingly, while the overlapping set of transcripts that are affected by inactivation of all three factors is quite large, a closer examination suggested important differences in the particular transcripts whose splicing was affected and prompted us to more carefully compare the profiles of these mutants. Microarray experiments were performed that directly compared the pairwise behaviors of these mutant strains (Figure 3C), revealing three distinct classes of transcript behavior. The first class is composed of transcripts for which splicing is equally affected by each of these mutations. Approximately 100 transcripts behave in this fashion, exemplified by those indicated with a red bar in Figure 3B and 3C and highlighted in Figure 3E. For these transcripts, the comparison of mutant versus wild-type behavior seen in Figure 3B shows a strong splicing defect for all three mutations. Further, the direct comparisons of the mutants in Figure 3C show that each of the mutants causes a nearly identical level of precursor accumulation and mature mRNA reduction. Interestingly, found within this class is the transcript encoding Act1, which is widely used as a reporter for both in vivo and in vitro splicing assays. Figure S2 shows the results of quantitative RT-PCR experiments validating a subset of these findings.
The second class is also composed of transcripts that exhibit a canonical splicing defect in response to inactivation of each of the single mutants, but for whom the magnitude of the defect is different for each of the mutants. About 100 transcripts behave in this fashion, exemplified by those indicated with a green bar in Figure 3B and 3C and highlighted in Figure 3F. As with the first class, the experiments shown in Figure 3B demonstrate that the splicing of each of these transcripts is affected by each of the mutations; however, the direct comparisons of the mutants in Figure 3C show a larger precursor accumulation in the prp2-1 strain than in either the prp8-1 or prp5-1 strains, and a larger accumulation in the prp8-1 strain than in the prp5-1 strain. Interestingly, the molecular splicing defects observed for this class of transcripts correlate with the conditional growth defects: the prp2-1 mutation produces the strongest phenotype, followed by the prp8-1 mutation, and finally the prp5-1 mutation. Notably, as indicated in Figure 3D, the vast majority of the transcripts in this class encode RPGs, and most RPGs belong to this class.
The final class is composed of a smaller number of transcripts, those that exhibit splicing inhibition in response to one or two of the mutant strains but not all three. Examples of several transcripts belonging to this class are shown in Figure 3G. Interestingly, the splicing of the Rps30a transcript is dramatically affected by the prp2-1 mutation, but shows little defect in response to either the prp8-1 or prp5-1 mutations. Conversely, the splicing of the Rpl19b transcript is affected by both the prp8-1 and prp5-1 mutations, but is largely unaffected by the prp2-1 mutation. Likewise, the Hnt1 transcript shows little defect in response to the prp8-1 mutation, while splicing of the Cox4 transcript is strongly affected by the prp5-1 mutation but is only mildly affected in the prp2-1 strain, the inverse of the ordering seen for the second class of transcripts described above.
Given the predominance of RPG transcripts in the second class of transcripts, we sought to identify properties of these transcripts that might distinguish them from the others. As an initial attempt to identify such differences, we asked whether RPG transcripts showed a difference in splicing efficiency relative to other transcripts under standard laboratory growth conditions in wild-type cells. Quantitative RT-PCR was performed using primers specific to both intron and exon regions of 12 different intron-containing genes. Using genomic DNA to generate standard curves, relative copy numbers of both the precursor and total mRNA species were then calculated and compared. As seen in Table 1, the RPG transcripts tend to show exceptionally high levels of splicing efficiency, particularly relative to non-RPG transcripts. The relative copy numbers for the total mRNA features is in good agreement with previously published mRNA abundance values [19]. Interestingly, while the mature RPG transcripts are quite abundant, there is in fact less detectable precursor species for most RPGs than for non-RPGs. Nevertheless, as seen in Figure S3, the ability to detect large increases in precursor levels on our microarrays is independent of the initial level of precursor mRNA.
The differences that we observed between mutations in different spliceosomal factors led us to ask whether distinct mutations in a single factor might also cause such transcript-specific responses. Shown in Figure 4A and 4B are the time-resolved splicing profiles obtained from shifting strains harboring either the prp8-1 mutation or the temperature-sensitive prp8-101 mutation, respectively, to the nonpermissive temperature and compared to similarly treated wild-type strains. Strikingly, the splicing of most intron-containing transcripts is almost completely unaffected by the prp8-101 mutation. Instead, this mutation appears to affect the splicing of only a small number of transcripts. As it had previously been shown that the prp8-101 mutation results in a defect in catalyzing the second transesterification reaction [18], we examined a defect in another factor necessary for the second step, the helicase Prp16 [20]. Figure 4C shows the time-resolved splicing profile derived from shifting a strain harboring the cold-sensitive prp16-302 mutation to the nonpermissive temperature of 16 °C. The broad splicing defect seen for this mutant suggests that the differences between the prp8-1 and prp8-101 splicing profiles are not simply the result of comparing mutants defective for the first or second chemical steps, but rather truly reflect allele-specific differences in substrate specificity.
The transcript-specific phenotypes that we observed with this small subset of conditional mutants compelled us to examine the effects of a much wider variety of spliceosomal defects. To facilitate these experiments, a set of high-throughput methods was developed that allowed us to automate many of the steps in a microarray experiment (see Materials and Methods). Using these methods, we examined the effects of 18 different mutations in core spliceosomal components and five additional mutations in up- or downstream processes in mRNA processing. Figure S4 shows the different mutants that were examined using these methods, the step in pre-mRNA processing at which they are presumed to be defective, and their conditional growth phenotype. Two different views of the data resulting from these experiments are presented in Figure 5 and Figure S5. By clustering the data across all of the experiments, Figure 5 tends to highlight those pre-mRNAs for which splicing is negatively affected by most of the mutants. By comparison, the clusters derived from each individual mutant shown in Figure S5 allow a more detailed analysis of the behavior of the pre-mRNAs in response to each individual mutation. As expected, many of the spliceosomal mutations result in splicing defects over a broad range of transcripts. Interestingly, the splicing profile resulting from inactivation of the brr5-1 mutant is quite similar to that seen for many other canonical splicing mutants. While this mutant was originally isolated as a splicing mutant [21], it has been subsequently shown that Brr5 functions during 3′ end processing and is the yeast homolog of mammalian CPSF-73 [22,23]. By comparison, the rna14-64 mutant, which is also defective in 3′ end formation [38], produces a noticeably different phenotype.
Importantly, a careful examination of the response of individual transcripts to this panel of mutations makes clear the fact that all transcripts are not equally affected by mutations in core components. To exemplify this point, Figure 6 shows the behavior of a select group of transcripts in response to all of the mutations examined. For some transcripts, such as the U3 small nucleolar RNA, splicing is rapidly blocked by nearly all spliceosomal mutations, while remaining largely unaffected by mutations in other mRNA processing factors. Interestingly, while the prp19-1 mutation does affect the splicing of many pre-mRNAs, it has little effect on the splicing of this transcript. A comparison of the two U3 paralogs, SNR17A and SNR17B, whose mature products are nearly identical but whose introns share little sequence homology, also shows some differences in their responses to some of the splicing mutants. For example, the U3a transcript shows a stronger defect in response to the prp5-1 mutation than the U3b transcript.
For some transcripts, the subset of mutations that affect their splicing is much different. A dramatic example of this is provided by the RPS30 paralogs. Whereas splicing of the Rps30b transcript is affected by most of the splicing mutants examined, the Rps30a transcript is not. The splicing of the Rps30a transcript shows a strong defect in response to the prp2-1 and prp16-302 mutations, a somewhat weaker response to the prp28-1, brr2-1, brr1-1, and brr5-1 mutations, and very little defect in most of the other mutants. The RPL19 paralogs also display divergent behaviors: whereas the Rpl19a transcript shows a similar defect in response to each of the mutants, the Rpl19b transcript shows strong defects in response to the prp8-1, prp5-1, and prp4-1 mutants, but less severe defects in response to the other mutants. Differences can also be seen between non-RPG transcripts by comparing the behavior of the Act1 and Tef4 transcripts, for example.
Here we describe the results of genome-wide experiments designed to examine the in vivo responses of all intron-containing transcripts in S. cerevisiae to mutations in core spliceosomal factors. For largely technical reasons, most experiments designed to study the efficiency of pre-mRNA splicing to date have focused on a relatively small number of transcripts, leaving open the question of how much variety there is among different spliceosomal substrates. The experiments presented here overcome this limitation by utilizing splicing-specific microarrays that allow for the simultaneous examination of all spliceosomal substrates. Splicing-specific microarrays are particularly amenable to studies in budding yeast, where there are a limited number of intron-containing genes and there is very little alternative splicing. By using this approach to examine the time-resolved effects of inactivating 23 different factors involved in pre-mRNA processing, the work presented here reveals a complex relationship between the activity of the core spliceosome and the full complement of transcripts with which it must interact.
A large collection of data such as this offers a unique opportunity to both compare and contrast the behaviors of all 250 intron-containing transcripts according to their individual responses to the different spliceosomal mutations. From this perspective, the detailed comparison of mutations in PRP2, PRP8, and PRP5 shown in Figure 3 provides important information about the different spliceosomal substrates, and suggests that all transcripts are not equally affected by particular spliceosomal mutations. Each of these three spliceosomal proteins plays an important role in catalyzing events necessary for the first chemical step in splicing. As such, mutations in these factors that reduce their activity at this step can be expected to negatively affect the splicing efficiency of all actively transcribed intron-containing genes. Indeed, while the majority of pre-mRNAs show such a defect in response to the prp2-1, prp8-1, and prp5-1 mutations, the strengths of the splicing defects imparted by these mutations are variable. Whereas the splicing of many transcripts encoding factors other than ribosomal proteins is equally inhibited by mutations in any of these three factors, the strength of the defect observed in the splicing of most ribosomal protein gene transcripts is dependent upon the particular defect in the spliceosome. For these mutations, the strength of the molecular splicing defect correlates with the temperature sensitivity that the mutation imparts upon the strain: prp2-1 causes the strongest defect and prp5-1 causes the weakest. While the fraction of new transcript from any single gene whose splicing is inhibited cannot be determined from these experiments alone, these results nevertheless demonstrate that a greater fraction of most RPG transcripts is blocked by the prp2-1 mutation than by either the prp8-1 or prp5-1 mutations. This difference is unlikely to simply reflect differences in the severity or onset of the protein mutations because nearly identical fractions of newly transcribed Act1, for example, are blocked by all three of the spliceosomal mutations. Rather, it suggests that the RPG-encoding transcripts are fundamentally different from the other transcripts in their susceptibility to these mutations.
We have not yet been able to identify any single feature of intron-containing RPGs that explains this altered susceptibility. While it is true that RPG introns tend to be longer than non-RPG introns, and that RPGs tend to be among the more highly transcribed genes in the genome, neither of these features alone seems sufficient to explain the different responses. The Act1 transcript, for example, is interrupted by an intron that is similar in length (308 nucleotides) to an average RPG intron (mean length of 402 nucleotides), and is transcribed at a similar rate as many RPGs [24], yet its behavior is distinct from that of most of the RPGs.
In an effort to identify properties that might differentiate RPG transcripts from other intron-containing transcripts, we used quantitative RT-PCR to determine the amount of precursor and mature mRNA species present in wild-type cells for a variety of different intron-containing genes. While all 12 of the transcripts that we examined were efficiently spliced (>80%), the RPG transcripts showed remarkably high partitioning toward mature species. It is noteworthy that, in spite of the high levels of total transcript present for most RPGs, the amount of precursor species present is remarkably low relative to the other intron-containing transcripts. For example, while there is more than ten times as much total Rps21b transcript in a cell as there is total Nmd2, 50 times less Rps21b pre-mRNA can be detected as compared to Nmd2 pre-mRNA.
It is tempting to speculate that the different splicing efficiencies and the different susceptibilities to spliceosomal mutations displayed by the RPGs may be mechanistically related. For example, while much recent progress has been made in identifying protein components of the spliceosome [4], it remains unknown whether the entire set of components is required for splicing of all transcripts, or whether different transcripts will utilize the activities of different subsets of factors. Given that the promoter elements driving transcription of RPGs are often different from non-RPGs, it is easy to imagine that the complement of proteins present during their splicing could also be different. The low levels of precursor species detected under steady-state conditions for the RPG transcripts suggest that the time between transcription and splicing is shorter for these transcripts than for non-RPG transcripts. In this light, it will be important to determine whether unique spliceosomal complexes are present during the splicing of RPG transcripts that allow for enhanced catalytic rates of either the first or second chemical steps of splicing. Alternatively, the rate of cotranscriptional loading of spliceosomal components onto RPG transcripts could be enhanced. In this case, genome-wide chromatin immunoprecipitation experiments might be expected to show higher levels of spliceosomal association with the RPGs than the non-RPGs. In either case, the presence or absence of unique, auxiliary spliceosomal components on the RPG transcripts might also change the susceptibility of these transcripts to defects in the core spliceosomal components.
Importantly, not all RPG transcripts behave similarly in these experiments. Whether considering the simple comparisons of the prp2-1, prp8-1, and prp5-1 mutations, or the entire panel of factors as shown in Figures 5 and 6, several RPG transcripts show significantly different behaviors from the others. Compelling examples of these differences can be seen among certain pairs of RPG paralogs. While many paralogs display similar defects in response to the mutations, some pairs show dramatically different responses to these mutants (Figure 6). The RPS30 paralogs offer the most dramatic example of these differences. Whereas splicing of the Rps30b transcript is adversely affected by almost all of the spliceosomal mutations that we examined, splicing of the Rps30a transcript is only affected by a small subset of these mutants. The particular features unique to each of these transcripts that underlie this differentiation remain unclear. Presumably the ability of the Rps30a transcript, for example, to be efficiently spliced in the prp8-1 mutant does not indicate that Prp8 function is unnecessary for the splicing of this transcript. Rather, it suggests that the unique set of interactions that this transcript forms with the spliceosome are less sensitive to the defect imparted by this particular mutation in Prp8. Importantly, because the Rps30a and Rps30b transcripts behave so differently in response to so many mutations, they should prove to be extremely useful as tools for more traditional biochemical experiments designed to understand the function of individual spliceosomal factors.
Most of the mutants that we have examined here were isolated from conditional lethal screens. As such, this subset of mutations may be biased towards those with strong growth defects and hence broad transcript specificity. Perhaps because of this, we have found it difficult to associate a particular mechanistic role with any of these factors based simply on the subset of pre-mRNAs whose splicing is affected in these experiments. Rather, our findings seem to indicate that the signature of introns that are affected by a particular spliceosomal mutation is unique to that mutation. Nevertheless, it is tempting to imagine isolating other mutations in these core components that affect the splicing of a smaller subset of transcripts. Such mutations may be more difficult to isolate, as they may be aphenotypic for growth under standard laboratory conditions. Nevertheless, a genome-wide splicing analysis of such a mutant may provide important insights into the activity of the protein. Indeed, for the prp8-101 mutation, which appears to represent such a mutant, knowing the identity of transcripts that are either strongly affected or unaffected provides important information for designing more traditional biochemical experiments in the future to examine the mechanism of its activity in the spliceosome. Likewise, the genome-wide splicing profiles derived from similar mutations in other factors might prove more revealing about the mechanistic role of those factors.
Interestingly, the capacity of core spliceosomal components to elicit transcript-specific changes in splicing activity does not appear to be limited to yeast. Recent RNAi-mediated depletion experiments targeting core spliceosomal components in Drosophila melanogaster show differential effects on splicing activity at alternative splice sites [25]. Given the level of conservation of the core spliceosomal components across eukaryotes, it seems highly likely that they could also play important roles in augmenting the regulatory roles of the SR and hnRNP families of proteins in higher eukaryotes. The surprising finding that the macular degenerative disease retinitis pigmentosa is associated with mutations in core components of the spliceosome may in fact reflect such a role [26–28]. As seen here for the prp8-101 mutation, the mutations associated with retinitis pigmentosa may cause defects in the splicing of a distinct group of transcripts, presumably uniquely associated with retinal function.
We have recently shown that the yeast spliceosome can rapidly and specifically alter the splicing efficiency of distinct subsets of transcripts in response to at least two unrelated environmental stresses (J.A.P., G.B.W., M.B. and C.G., unpublished data). Notably, in response to amino acid starvation, the splicing of virtually all RPG transcripts is specifically downregulated. Given that S. cerevisiae lacks the large SR and hnRNP protein families responsible for modulating transcript-specific splicing activity in higher eukaryotes, the observations presented in the current work suggest that the core spliceosomal components themselves could be targets of environmental regulation. As seen with stable genetic modifications here, post-translational modifications of these components might similarly facilitate transcript-specific regulation of pre-mRNA splicing. Such modifications would allow the cell to both rapidly and specifically regulate the production of translatable mRNA for particular transcripts in a reversible manner. Whether or not the complex relationships between transcripts and core spliceosomal components revealed by these mutant studies represent features of a system that has evolved to be the target of biological regulation is a provocative question that remains to be fully addressed.
The work presented here does however suggest that the relationship between the spliceosome and the full complement of transcripts with which it must interact is much more complex than previously believed. Important questions for the future concern both the mechanism by which the spliceosome distinguishes among these substrates and the biological rationale for this specificity. The simple hypothesis that all mutations in splicing factors that inhibit the growth of a yeast cell would also inhibit the splicing of all (expressed) intron-containing transcripts is inconsistent with our observations. The splicing efficiency of a given intron can not yet be ascribed to obvious cis features of the intron, nor does the biochemical activity presumed to be affected by a mutation appear to dictate which introns will be affected. Instead, in much the same way that transcription is now known to be regulated by the combinatorial control of its holoenzyme, it appears that the efficiency of pre-mRNA splicing may be dependent upon the complexes present in the spliceosome as well as the particular transcript upon which it assembles.
Unless otherwise indicated, cultures were grown according to standard techniques in rich medium supplemented with 2% glucose at 25 °C. Both mutant strains and the corresponding wild-type strain were grown in parallel, 100 ml cultures until their optical densities were between A600 = 0.5 and A600 = 0.7. An initial 15 ml sample was collected at 25 °C prior to initiation of the time course. Cells were collected by filtration using Millipore HAWP0025 filters (http://www.millipore.com). The filters were immediately frozen in N2(l). The culture flasks were then transferred into water baths at either 16 °C or 37 °C, with additional 15 ml aliquots removed and collected at the appropriate times. Total cellular RNA was isolated as previously described [29] with a few exceptions. Tubes containing PhaseLock gel (Eppendorf, http://www.eppendorf.com) were used during phase separation steps. Also, RNA samples were precipitated using isopropanol.
For each microarray sample, cDNA was prepared from 25 μg of total RNA in a 50 μl reaction mixture containing 50 mM TrisHCl (pH 8.3), 75 mM KCl, 3 mM MgCl2, 10 mM DTT, 0.5 mM ATP, 0.5 mM CTP, 0.5 mM GTP, 0.3 mM TTP, 0.2 mM aa-dUTP, 12.5 μg dN9 primer, and 5 ng murine Moloney leukemia virus (M-MLV) RT. Primers were annealed to the RNA by heating to 65 °C for 5 min in the presence of buffer and salt alone. Reactions were allowed to incubate at 42 °C for at least 2 h. Remaining RNA was then hydrolyzed by incubation for 15 min at 65 °C in the presence of 0.1 M NaOH and 10 mM EDTA. This was neutralized with HCl, then purified using a Zymo25 DNA purification column according to the manufacturer's protocol (http://www.zymoresearch.com). Typical yields ranged from 10 to 15 μg of cDNA from 25 μg of starting material. The purified cDNA was then conjugated to the appropriate fluorescent dye in a 10 μl reaction containing 50 mM sodium bicarbonate (pH 9.0), 50% DMSO, and ∼20 μg of NHS-derivatized fluorophore. Reactions were incubated in the dark at 60 °C for 60 min, after which time the cDNA was repurified using a Zymo25 DNA purification column.
Cultures were grown in sets of four mutant strains along with their corresponding wild-type strains in rich medium in flasks at 25 °C until their optical densities were between A600 = 0.5 and A600 = 0.7. After the strains were synchronized, the cultures were transferred into the wells of a 96-well growth plate at 25 °C. A reverse time course was then initiated using a multichannel pipettor at the appropriate times to transfer the cultures into a second 96-well plate submerged in a 37 °C water bath (Figure S6). The entire time course was then collected by centrifugation of the plate at 5,000g for 5 min. The cell pellets were then frozen in N2(l). With the help of a Biomek FX liquid handling system (Beckman Coulter, http://www.beckmancoulter.com), total cellular RNA was then isolated using largely standard procedures. Typical yields from a single 1.8 ml well ranged from 15 μg to 25 μg of total RNA. For each well, cDNA synthesis was performed in a 50 μl reaction as described above. After cDNA synthesis, the remaining RNA was hydrolyzed by addition of 25 μl of 0.3 M NaOH/0.03 M EDTA and heating to 60 °C for 15 min. This solution was neutralized by addition of 25 μL of 0.3 M HCl. To purify the cDNA from this solution, 0.5 ml of cDNA binding buffer (5 M GdnHCl, 30% isopropanol, 60 mM potassium acetate, 90 mM acetic acid) was added to each well. This material was passed over a 96-well glass-fiber filtration plate (Nalgene Nunc, http://www.nalgenenunc.com). The filters were then washed once with 0.5 ml of Wash I (10 mM TrisHCl [pH 8.0], 80% EtOH), and once with 0.5 ml of Wash II (80% EtOH). Purified cDNA was then eluted in two steps of 100 μl water. The appropriate dyes were then coupled to the cDNA as described above. Fluorescently labeled cDNA was then repurified using 96-well glass-fiber filtration plates as described above.
Total cellular RNA for quantitative RT-PCR experiments was prepared as described above. Prior to conversion to cDNA, the RNA was treated with DNase I (Fermentas, http://www.fermentas.com) according to the manufacturer's protocol. The cDNA was then produced from 2 μg of total RNA in a 20 μl reaction mixture containing 50 mM TrisHCl (pH 8.3), 75 mM KCl, 3 mM MgCl2, 10 mM DTT, 0.5 mM each dNTP, 25 nM each gene-specific primer, 2.5 μg dN9 primer, and 5 ng M-MLV RT. Quantitative PCR was then performed using an Opticon from MJ Research (http://www.mjresearch.com). Primer sets used in this study are shown in Table S1. Relative copy numbers shown in Table 1 are the result of triplicate measurements from a single biological sample. Standard deviations were only slightly higher when comparing biological replicates. A series of 10-fold dilutions of genomic DNA covering a total range of 106 molecules was used to generate the standard curves. Genomic DNA was purified using a ZR Fungal DNA kit (Zymo Research) according to the manufacturer's protocol. The copy numbers presented in the Table have been normalized to the least abundant species detected in these experiments, the precursor isoform of the RPS21b transcript. Importantly, these copy numbers do not represent cellular copy numbers.
In an initial set of experiments, we examined a variety of probe designs for their ability to specifically and efficiently identify the mature mRNA species derived from intron-containing genes. Whereas other splicing-specific microarrays have utilized fixed-length probes that are geometrically centered about the exon–exon junction, our best results were obtained when the probe was thermodynamically balanced between the two sides of the junction. Accordingly, oligonucleotides were designed for every exon–exon junction such that the hybridization energies were equivalent on either side of the junction (Figure S7). Because of the different GC contents in the transcripts, the junction oligos range in length from 28 nucleotides to 40 nucleotides.
Total mRNA feature probes (expression oligos) and premature mRNA feature probes (intron oligos) 32 nucleotides in length were designed using a previously described algorithm [30]. Intron and junction oligos were designed to target 256 intron features in 246 different transcripts. Expression oligos were designed to target the last exon of these 246 intron-containing transcripts, as well as 689 non–intron-containing transcripts and 99 noncoding RNAs (including snRNAs and tRNAs).
Probes were printed on poly-lysine-coated glass slides using standard techniques. Each probe was print on the array three times by each of two different printing pins (for a total of six replicate probe spots for each feature on each array; Figure S8). Prior to use, arrays were preprocessed by 60 s of incubation at 25 °C in 3× SSC, 0.2% SDS. Arrays were then blocked using standard procedures. Mutant and wild-type samples were then competitively hybridized against one another using the Agilent Hybridization Buffer (Agilent, http://www.agilent.com) at 60 °C for a period ranging from 12 to 16 h.
Microarray images were acquired using an Axon Instruments GenePix 4000B scanner, reading at wavelengths of 635nm and 532nm (Axon Instruments, http://www.axon.com). Image analysis was performed using Axon Instruments GenePix Pro version 4.0 (number 2500–137, Rev E, 2001). Ratio values derived from the median pixel intensities for the 635 nm and 532 nm images of each spot were used to represent probe behaviors in data preprocessing analysis. A standardized qualitative assessment of array quality was performed using the Bioconductor arrayQuality package (Paquet AC, Yang YH, “arrayQuality: Assessing array quality on spotted arrays,” version 1.4.0) [31,32]. Arrays were manually removed from further analysis if they showed an above average spot intensity to ratio bias or a strong spatial ratio bias (Figure S9). Spot ratio data were log2 transformed and normalized within each array using global loess regression implemented in the Bioconductor marray package [33]. All experiments were performed as dye-flipped pairs using biological replicates. Both within-array and between-array data replication were analyzed for data quality (Figure S10) using the Bioconductor limma package [34]. Linear analyses of the data were performed using correlations of within-array and between-array spot replicates to assess evidence for differential expression of each RNA species across all of the experimental factors [35]. Results from these linear analyses pertaining to conclusions highlighted in the text are shown in Tables S2 and S3.
For downstream hierarchical clustering, feature ratio values were calculated by averaging the median within-array spot replicate measures between replicate arrays. Where appropriate, replicate variation was used to weight feature measurements. Hierarchical clustering was performed using the C Clustering Library version 1.32 [36]. Data were clustered using average linkage and Pearson correlation as the distance measure.
All microarray data are available at the Gene Expression Omnibus (GEO) repository at the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov/geo).
|
10.1371/journal.pntd.0006194 | Inferring the risk factors behind the geographical spread and transmission of Zika in the Americas | An unprecedented Zika virus epidemic occurred in the Americas during 2015-2016. The size of the epidemic in conjunction with newly recognized health risks associated with the virus attracted significant attention across the research community. Our study complements several recent studies which have mapped epidemiological elements of Zika, by introducing a newly proposed methodology to simultaneously estimate the contribution of various risk factors for geographic spread resulting in local transmission and to compute the risk of spread (or re-introductions) between each pair of regions. The focus of our analysis is on the Americas, where the set of regions includes all countries, overseas territories, and the states of the US.
We present a novel application of the Generalized Inverse Infection Model (GIIM). The GIIM model uses real observations from the outbreak and seeks to estimate the risk factors driving transmission. The observations are derived from the dates of reported local transmission of Zika virus in each region, the network structure is defined by the passenger air travel movements between all pairs of regions, and the risk factors considered include regional socioeconomic factors, vector habitat suitability, travel volumes, and epidemiological data. The GIIM relies on a multi-agent based optimization method to estimate the parameters, and utilizes a data driven stochastic-dynamic epidemic model for evaluation. As expected, we found that mosquito abundance, incidence rate at the origin region, and human population density are risk factors for Zika virus transmission and spread. Surprisingly, air passenger volume was less impactful, and the most significant factor was (a negative relationship with) the regional gross domestic product (GDP) per capita.
Our model generates country level exportation and importation risk profiles over the course of the epidemic and provides quantitative estimates for the likelihood of introduced Zika virus resulting in local transmission, between all origin-destination travel pairs in the Americas. Our findings indicate that local vector control, rather than travel restrictions, will be more effective at reducing the risks of Zika virus transmission and establishment. Moreover, the inverse relationship between Zika virus transmission and GDP suggests that Zika cases are more likely to occur in regions where people cannot afford to protect themselves from mosquitoes. The modeling framework is not specific for Zika virus, and could easily be employed for other vector-borne pathogens with sufficient epidemiological and entomological data.
| Since May 2015, when Zika was first reported in Brazil, the virus has spread to over 60 countries and territories, and imported cases of Zika have been increasingly reported worldwide. However, there is still much uncertainty behind the mechanisms which dictated the rapid emergence of the epidemic. This work introduces a novel modeling framework to improve our understanding of the risk factors which contributed to the geographic spread and local transmission of Zika during the 2015-2016 epidemic in the Americas. The model is informed by data on regional socioeconomic factors, mosquito abundance, travel volumes, and epidemiological data. As expected, our results indicate that increased presence of mosquitoes, human hosts, and viruses increase the risk for mosquito-borne virus transmission. Passenger air travel, however, was less impactful, suggesting that travel restrictions will have minimal impact on controlling similar epidemics. Importantly, we found that a lower regional GDP was the best predictor of Zika virus transmission, suggesting that Zika is primarily a disease of poverty.
| Prior to 2015, local cases of Zika virus had only been reported in Africa and Asia, most prominently in the Pacific Islands [1]. Phylogenetic analysis suggest the virus was introduced into the Americas as early as 2013 [2, 3], but it was not detected in Brazil until May 2015. By this time, Zika virus had already silently spread throughout most of the Americas [3, 4]. As of 2016, local Zika virus transmission has been established in over 60 countries and territories, with the number of estimated cases exceeding 750 thousand [5]. Zika virus infection typically presents with mild flu like symptoms, and in many cases the infection is asymptomatic. However, the potential harm posed by Zika is now known to be substantially greater since it has been associated with a rare congenital disease, microcephaly [6–13], and Guillain-Barre syndrome [13]. The unprecedented size of the outbreak and links to severe disease prompted the WHO to declare the current Zika virus outbreak a public health emergency of international concern [14]. The emergency status lasted until November 2016, at which point Zika virus was recognized to remain a significant enduring public health challenge [15].
Like dengue and chikungunya, Zika is a vector-borne virus primarily transmitted by Aedes aegypti [16–22]. Geographic spread of the viruses occurs through global transport systems, such as passenger air travel, cruises, and maritime freight, where infected travelers depart affected regions for destinations where competent vector species have established populations [2, 23–26]. The numbers of recent travel-related Zika cases diagnosed around the world (e.g., United States, Europe, Australia, New Zealand and China) [27] demonstrates how these networks facilitate virus emergence in new areas. It is not always clear, however, what factors are necessary for successful establishment and outbreaks. For example, in many countries, imported cases did not result in local transmission, while phylogenetic analysis shows that some regional outbreaks were initiated by multiple Zika virus introductions [3, 4, 24, 28].
The objective of our work is to better understand the risk factors which contributed to the spread of Zika virus during the 2015-2016 epidemic in the Americas. Our work complements and builds upon several recent studies investigating the potential spread of Zika into new regions by utilizing a substantially different framework. Monaghan et. al. [29] overlaid simulated Ae. aegypti and Ae. albopictus mosquito abundances, travel capacities, and socioeconomic factors to estimate the cities in the United States with the highest expected cases of travel-imported Zika. In two studies, Bogoch et. al. [30, 31] presented the potential for Zika virus spread into the rest of the Americas, Africa, and the Asia Pacific region using air travel and vector abundance risk maps. Nah et. al. [32] used survival analysis and publicly available epidemiological and air travel data to predict the risk of importation and local transmission of Zika virus at the country level. Zhang et. al. [33] applied a stochastic epidemic model to simulate the spatiotemporal spread of the virus at a global scale, and estimated both the number of Zika infections and microcephaly for several countries in the Americas. Ogden et. al. [34] demonstrated that risk to travelers in Zika virus affected countries correlates with estimates of R0 using human case surveillance data. Others have also estimated R0 to assess transmission risk across a variety of environments [35–37]. We previously mapped the variations in the geographic risk profile under different assumptions of vector species capacities for Zika virus transmission [38], and illustrated the geographic spread of Zika virus to be driven primarily by Ae. aegypti [39]. In collaboration with a large international research team, we also analyzed sequenced Zika virus genomes, reported cases, mosquito abundance, and travel patterns to track the spread of the epidemic (i.e., genomic epidemiology) [3, 4, 24].
The proposed methodology we present in this paper is what we refer to as the Generalized Inverse Infection Model (GIIM), initially introduced in [40], and further expanded upon in [41]. GIIM is a network optimization model originally designed for static environments, which we extended to a dynamic framework. Here, we define the network structure by the passenger air travel movements between all pairs of regions in the Americas. For each region, we provide the model data regarding socioeconomic factors, available health infrastructure, vector habitat suitability, passenger air travel, and reported Zika case numbers. Our model estimates the relative contribution of each risk factor in the spread and local transmission of Zika virus using a multi-agent based optimization search method to estimate the parameters of a link risk function. Specifically, our results provide quantitative time-dependent estimates for the likelihood of Zika virus spreading between regions, resulting in local transmission at the destination. At a regional level, exportation, and importation risk profiles are also provided. The risk function parameters are estimated through iterative refinement, the direction of which is driven by an error function between predicted and observed properties of an infection process, similar to what we previously implemented for dengue virus [25]. Our results indicate that the most common transmission routes were to countries with some of the lowest gross domestic product (GDP) per capita in the Western Hemisphere [42].
The proposed model is evaluated at a country level, and therefore dependent on country level data for input. The socioeconomic data, epidemiological data, travel data, and vector suitability data for the principle spreading vectors species are all aggregated to the country level for all countries and territories in the Americas as well as the individual U.S. States. Each variable considered in the model is described in further detail below. All data that we can make publicly available are listed in S1 Data.
For the purpose of estimating the risk of Zika virus spread and local transmission in the Americas, we expand upon a previously developed method based on the GIIM [40, 41]. The GIIM model is referred to as an inverse infection model because, instead of attempting to simulate an infection process directly, it uses observations from an actual outbreak on a network and seeks to estimate the parameters of this process, e.g., transmission probability functions on the links of the network. This objective is complementary to a related problem which seeks to infer the set of links most likely responsible for explaining an observed transmission process [54–56]. To accomplish the prescribed task, GIIM relies on information such as the status of confirmed local transmission for each node (or a subest of nodes) in the network, as well as the structure and properties of the network itself. The transmission status is used as a reference point, and GIIM sets the edge weights such that the underlying stochastic simulation model replicates the actual outbreak scale as closely as possible, i.e., GIIM seeks to minimize the error between the observed and reference point. The GIIM model was originally designed for static environments, and is extended to handle dynamic inputs of the application proposed in this work. Both the original model and the extensions are discussed in detail in this section.
The inputs for the GIIM consist of the underlying network structure, attributes of both the nodes and links of the network, and observations on the actual transmission process. In this work the network structure is defined by the passenger air travel movements between regions in the Americas. We denote graph G as G(V, E), where VG is a set containing all the vertices of the graph, while EG contains all the edges of the graph. Edges are denoted as eu,v, where u is the source and v is the destination of a travel route represented by the edge. In this paper, VG contains all countries in the Americas, the French, Dutch, British and American overseas territories, and the individual states of the US, which also defines the spatial resolution of the model. The temporal resolution of the model implemented in this work is at a monthly level, which corresponds to the available travel and suitability data. The edges of the graph exists between a pair of regions if there were passenger air travel movements between them during any month in 2015. The travel patterns between countries is asymmetric, therefore the resulting graph is directed. We denote this graph as GA.
We assign a number of attributes to the vertices and edges of the graph to capture the potential risk factors that may influence the regional spread of Zika virus over time. These attributes correspond to the variables described in the data section. The attributes take the form of normalized real values between zero and one, and include the following:
A subset of the attributes are dynamic, i.e. their value changes over time, and therefore are denoted with an additional time index, t. The dynamic attributes include the passenger volumes, the incidence rates in each region, and the vector suitability. The monthly airport travel volumes, V u v t, are aggregated to the state and country level. The mean and standard deviation of the vector suitability for each region for each month is computed as described in the data section. The remaining attributes are static, and are assumed not to vary over the course of the year.
The GIIM defines a transmission model on the input graph. Graph-based transmission models require a real value w e t ∈ [ 0 , 1 ] , e ∈ E G A to be present on all edges of the graph, these are called edge transmission probabilities. In this application of the GIIM, the attributes are incorporated into a functional form to represent each time-dependent edge transmission probability. The function takes the form as shown below:
w e t = C + α V u v t + β I u t + γ S v t + δ S V v t + ω E v + ρ H v + λ P v (1)
The variables in the function correspond to the set of attributes previously listed, and the coefficients of each attribute are the parameters to be estimated by the model. Values w e t are bounded between 0 and 1. We will denote the surjective assignment of edge transmission probabilities to the edges as W G A : E G A ↦ [ 0 , 1 ].
In addition to the network attributes, GIIM requires a reference observation of the real-life transmission process it seeks to estimate. In the current application, the observation used is the date of the first reported local Zika cases in each region for the time period considered. The reference point is therefore given as a set of 18 binary vectors; each binary vector corresponds to a month of the observed Zika virus outbreak starting from July 2015 to December 2016, and assigns a value of 0 or 1 to each node of the graph indicating its transmission state. A value of 1 indicates that the presence of local transmission of Zika virus in the region was reported within or before the corresponding month, and a value 0 indicates that Zika cases have yet to be reported from the region.
The GIIM relies on an underlying stochastic simulation to model the spreading process. The compartmental model that is used in this paper is the SI model, which has two states: susceptible (S) and infected (I). The graph-based SI infection model is an iterative discrete-time model that assigns states to the nodes of the graph: each node can only be in a single state at any time step, and all nodes must have a state at all times. While GIIM can accommodate a more complex compartmental model, e.g., SEIR, the SI model is selected to fit the Zika virus application based on the assumption that once Zika virus becomes locally established, that region remains a potential risk of furthering the spread of the virus for the time frame being considered. The reason for allowing the option of non-zero outgoing risk after the reported case count reduced to zero is that there could still be infected individuals in the region, especially given the high rate of asymptomatic cases and reporting error. However, it is important to note that the use of the SI model does not enforce a region to have a positive transmission probability over the entire period modeled, it simply allows a non-zero transmission risk value to be estimated by the model. The actual time-dependent transmission probability is defined as a function of the incidence rate at the origin country among other factors; a probability of zero is a feasible solution of the model, and is actually assigned to many of the links over the course of the outbreak. Given the estimated transmission probabilities, in each step of the simulation, “infected” nodes try to “infect” all their susceptible neighbors according to the transmission probability we connecting them. If the attempt is successful, the neighbor will be infected in the following iterations. If the attempt is unsuccessful, the neighbor remains in a susceptible state, and the infected node can continue to make attempts in the following iterations indefinitely.
More formally, the transmission process starts from an initially infected set of nodes A 0 ⊂ V G A at iteration t0. The rest of the nodes V G A \ A 0 are susceptible at the beginning of the process. Let A i ⊆ V G A be the set of infected nodes at iteration i. At each iteration, t, each node u ∈ Ai tries to infect each susceptible neighbor v ∈ V G A \ A i according to the probability w e i , e = ( u , v ) ∈ E G. If the attempt is successful, v becomes infected starting from the following iteration: v ∈ Ai+1. If more than one node is trying to infect v in the same iteration, the attempts are made independently of each other in an arbitrary order within the same iteration. By definition, the transmission process stops at iteration t if A t = V G A. Since we only simulate a finite portion of the Zika virus epidemic, we stop all transmission models after 18 iterations, corresponding to the observed months of the epidemic at the time of analysis.
Within each model run, the transmission process is repeated 10000 times to produce a real value indicating the likelihood of each node being in an infectious states at each iteration, t. The value is calculated by counting the number of repetitions when nodes were in an infectious state, and dividing by the total number of repetitions i. e.10000.
The GIIM [41] implemented in this work formulates the estimation of edge transmission probabilities as a general optimization task. The model relies on knowledge of the underlying graph and (at least a subset of) observations from a transmission process taking place on the network. The real observations take the form o → t ∈ O, where o → is a binary vector and t is a time stamp. Each vector represents a point in time and o → t assigns a binary value to all v ∈ V G A indicating the observed transmission (or lack there of) of Zika virus in the region at time t. Set O contains all observations, while set T contains all sample times, i.e. |O| = |T| = 18.
The inputs of GIIM are: an unweighted graph G, a transmission model I, the set of sample times T, and the set of observations O, where O = I n f ( G , W , I , T ). In this work I is the SI stochastic simulation transmission model defined in the previous subsection, WG: EG ↦ [0, 1] is the unknown weight assignment, and Inf is a procedure that generates observations at sample times T based on transmission process I taking place on graph G with assigned edge weights we ∈ W, e ∈ E(G). The set of observations, O′, is a time-dependent vector of real values equal to the probability each node is infected at each timestep, computed from the set of runs. The task of GIIM is to find an estimation W′ of W so that the difference between O and O ′ = I n f ( G , W ′ , I , T ) is minimal. Due to the need to compare a set of binary vectors with a set of real vectors, we compare observations O and O′ using ROC evaluation by pairwise comparing vectors o → i ∈ O and o ′ → i ∈ O ′ for i = 1…18, computing the AUC value for each pair and averaging over all pairs.
The formal definition of the GIIM is as follows:
General Inverse Infection Model: Given an unweighted graph G, and transmission model I, the set of sample times T and observations O = I n f ( G , W , I , T ), we seek the edge transmission probability assignment W′ such that the difference between O and O ′ = I n f ( G , W ′ , I , T ) is minimal.
The GIIM defines the estimation of W as an optimization problem, which is solved using an iterative refinement algorithm. The procedure begins with an initial weight configuration W 0 ′, runs transmission model I, makes observations O′ and computes the error between O and O′. Based on the error, W′ is refined and the process is repeated, until the error becomes less than an accuracy constant a selected by the user. The search strategy used in this paper is the Particle Swarm Optimization (PSO) method [57]. According to the findings in [41], the method is stable and is able to produce outputs close to the reference, and therefore the solution method chosen for this work as well. It can also be implemented in a parallel environment speeding up computations considerably. Algorithm 1 summarizes the iterative GIIM algorithm.
Algorithm 1 Generalized Inverse Infection Model
1: Inputs: G, I, T, O, a
2: Choose initial edge infection probability assignment W′
3: repeat
4: Compute O ′ = I n f ( G , W ′ , I , T )
5: Compute d(O,O′)
6: if d(O,O′)≤a then
7: return W′
8: else
9: Choose new W′ according to the PSO search strategy.
10: end if
A major modification was necessary in order to adapt the GIIM method to be applied in the context proposed here. The original GIIM method estimates the edge transmission values of the graph directly as real values that are static, i.e., they do not change over time [41]. In this work the edge weights are given as functions of known attributes on the nodes and edges of the graph, and the task becomes the estimation of these functions, or more specifically, the coefficients of these functions. Several attributes in this application are dynamic, i.e., they change with time. Thus, it is necessary to further extend the function estimation method to account for dynamic attributes, and to adapt the simulation model to handle dynamic edge transmission values.
In this work the edge transmission values are defined using a linear function of known attributes, as defined in eq (1). More generally, if a i t ( e ) , e ∈ E G A is the set of attributes, where i represents the i-th attribute and t represents the time period, then the time-dependent edge transmission probabilities w e t are given as w e t = g ( f 1 ( a 1 t ( e ) , c 1 → ) , f 2 ( a 2 t ( e ) , c 2 → ) , … , f ℓ ( a ℓ t ( e ) , c ℓ → ) , c g → ) for all e ∈ E G A, where ℓ is the number of available attributes, f1, …, fℓ and g are functions and c 1 → , … , c ℓ , c g → → are coefficients of functions f1, …, fℓ, g. Following the notation proposed in [41], the functions used to compute the edge transmission probabilities are given as a set of attribute functions f1, …, fℓ assigned to each invididual attribute, an aggregator function g with the role of creating a single value from the result of the attribute functions and a normalization function ensuring that the result falls between 0 and 1. This formulation makes implementation of the method easy while retaining the flexibility of the model. Let C denote the set of all coefficient vectors. The values of the attributes can change over time, however the functional form and estimated coefficients are assumed to remain constant over the time period examined.
The optimization process of GIIM changes from the estimation of the direct weight assignment W to the estimation of C. This provides a means to identify the key factors contributing to the spread of Zika virus throughout the regions in the Americas. A second advantage is technical; |C|<< |W|, therefore the optimization process is much easier because we are only looking for a limited number of function coefficients as opposed to the edge transmission probabilities for all edges of the graph. For the implementation of the model in this paper, all parameters are bounded between -0.5 and 0.5, in order to reduce the solution space for the PSO method. Finally, the resulting edge transmission values are trimmed above 1 and below zero by taking w e t = M A X ( 0 , M I N ( 1 , ∑ i = 1 ℓ f ( a i t ( e ) , c i ) + c g ) ) ).
By the time Zika virus was first detected in Brazil in May, 2015, the virus had already rapidly spread to most regions of the Americas [3, 4]. The goals of our analyses were 1) to identify the relative contribution of each risk factor in the spread and local outbreaks of Zika virus and 2) to compute the risk of spread (or re-introductions) between each pair of regions during 2016. We aggregated the route level risks to provide a relative ranking of total importation and exportation risk posed by and to each region per month during the outbreak. Our final network, representing feasible air travel routes in the Americas, is a directed, weighted graph structure with 103 nodes and 2946 edges.
The first task of our study is to identify the set of attributes (and corresponding contribution of each) to be included in the model. We consider the entire set of attributes previously presented in the data section. A linear weighted sum function as defined in eq (1) is used to compute transmission probabilities, and the dynamic GIIM method is implemented to estimate the coefficients of the function. Different variable configurations were considered, and the model that produced the best fit was selected. The model fit is based on the quantifiable performance metric, ROC AUC averaged over the entire time period. To evaluate and ensure stability of the proposed estimation method, we ran the algorithm 20 times with the same set of inputs and computed the mean and variance of the estimated model coefficients over all runs. The results of the final model are presented in Fig 5.
w e t = 0 . 040 + 0 . 079 V u v t + 0 . 247 I u t + 0 . 174 S v t - 0 . 372 E v + 0 . 285 P v (2)
The results of the model are highly robust. The estimated coefficients vary minimally across runs, and, more significantly, the ranking of all risk factors remained consistent across all runs (Fig 5A). The expected value of the estimated coefficients represent the relative influence of each attribute in the risk of Zika virus spread between a pair of regions. When interpreting these coefficients it is important to first note that this model is designed to estimate the risk of Zika virus spread between two regions resulting in local, vector-borne transmission. This is different from the risk of Zika infected passenger arrivals. For example, there are multiple locations where travel cases of Zika were continually reported, yet no local cases resulted [27]. The lack of local transmission in these examples could be due to many explanations, including insufficient populations of competent vectors and/or intense surveillance and vector control programs implemented at the destination. Thus, a high number of travel-reported cases does not necessarily translate to a high local transmission risk, and for this reason, the risk of Zika infected passenger arrivals and local transmission risk should be modeled separately. Due to the potential harm posed by local outbreaks of Zika, local transmission risk is the primary focus our analyses.
The important distinction between modeling travel-reported cases and local transmission risk is perhaps most evident from the low coefficient estimated for travel volume. In fact, we found infected travelers to be significantly less influential than all the other risk factors. This can be explained by the fact that there is a high level of connectivity between most pairs of regions, and more importantly, the highest volume of air travel exists between and within the U.S. states. Yet, with the exception of Florida [24] and Texas [58], Zika was not broadly established in the U.S. Thus, we do not identify travel volume to be a driving forces in the spread and transmission of Zika virus in the Americas; travel is a necessary, but not sufficient condition. It is worth noting that a model seeking to estimate the risk of infected passenger arrivals alone would likely find this variable to play much more substantial roles.
In contrast, we found that Zika virus spread and local transmission is largely driven by regional attributes at the origin (incidence rate) and destination: Ae. aegypti suitability, human population density, and the GDP, with GDP being the most significant (Fig 5A). As expected, a higher incidence rate at the travel origin, which can act as a proxy for the likelihood of an infected individual arriving at the destination, significantly increases the risk posed to the travel destination. Similarly, the results indicate a higher vector suitability at the destination corresponds to an increased risk of transmission. High human population density at the destination is also revealed to increase the risk of transmission, consistent with the required presence of both vectors and hosts for mosquito-borne virus transmission. The health indicator variables were excluded from the final model, as they were not found to have a significant impact.
Based on our model results, the most dominant and only negative risk factor is a region’s GDP, i.e., a higher GDP at the destination corresponds to a lower risk of transmission. Based on the magnitude of the coefficient, the destination’s GDP contributes more than any other risk factor considered. The highly negative coefficient of GDP can be explained by the substantial delay (or complete lack) of local transmission in the wealthier U.S. states and certain territories and islands in the Caribbean. GDP is obviously not directly involved in Zika virus transmission, but it may indirectly influence the interactions between components of the cycle: hosts and vectors. Poorer nations likely have lower housing qualities and inhabitants may be exposed to more mosquito bites, e.g., a lack of screens on windows and doors allowing mosquitoes to enter. They may also have more debris around their homes acting as breeding containers for Ae. aegypti. Lastly, GDP may be a proxy for the available surveillance and vector control resources at the destination, an increase of which would aid in reducing local transmission.
Our final model captures the relative contribution of both static and dynamic risk factors to explain the spread and local transmission of Zika virus in the Americas. The AUC ROC for this model averaged over the 18 months is 0.923, which indicates an excellent fit (Fig 5B). For the first month we see perfect classification, since the epidemic was only reported in Brazil, which proves trivial for the estimation method. The second half of 2015 corresponds to a major increase in the number of regions reporting local transmission (as illustrated in Fig 1). Specifically, between November and December the number of regions reporting transmission nearly doubled from 9 to 16, and then increased to 32 in January. This proved to be the most challenging part of the estimation process, and the reason for the performance drop below 0.9 during the period from October to January. The minimum AUC value of 0.83 occurred in November 2015, which is still considered to be good performance for any classifier. The estimation task gets easier again for the last 9 month characterized by a low but constant rate of spreading between the regions, and performance goes around 0.95 again for these months. We can conclude, that even though sudden bursts in the reported local outbreaks decrease prediction accuracy slightly, the method is able to provide accurate estimation. The estimation process took between 3 to 4 hours with the number of iterations between 250 and 350. A parallel version of the algorithm was implemented in C++, and the results were computed on a PC with an 4-core i7-7700k 4,2 GHz processor.
Using the final estimated model, we computed the probability of Zika virus spread between any pair of regions in the Americas resulting in subsequent local transmission (Fig 6). The link probabilities reveal the highest risk travel routes connecting regions at discrete points in time over the outbreak, as well as how the relative ranking changes over time (Fig 6A). Out of the 2946 feasible edges in the network, only 711 edges have nonzero transmission probabilities, and only 58 of the edges has a value greater than 5% at any point of the observation period. The time-dependent data for the top 100 transmission links, exportation risk, and importation risk are provided in S1 Data.
In general, the high risk travel routes are dominated by ones outbound from the Caribbean Islands with the highest incidence rates (earlier in the outbreak), and pointing to the less developed countries and territories of the Caribbean and Central America. These results are corroborated by the early estimated introduction times into countries like Haiti and Honduras relative to Puerto Rico, Mexico, and the U.S. [3, 4, 24], see also www.nextstrain.org/zika. The risk on routes departing a given country, e.g., Brazil, behave similarly, but vary in magnitude. They also display different behavior than the outgoing risk posed by other high risk countries, e.g., Martinique. The estimated transmission probabilities fluctuate over time due to changes in the dynamic attributes at both the route origins (outbreak scale) and destination (vector suitability), as well as variations in the monthly travel volumes between regions. The effect of these risk factors, for example, the lower Zika incidence rates in Puerto Rico relative to Martinique (Fig 1B), is illustrated by the corresponding temporal differences in transmission risk into Haiti (Fig 6A).
The route level risk can be aggregated to provide import and export risk profiles at a regional level. This type of spatial and time-dependent information can help guide policy decisions, such as where to allocate available resources at different stages during an outbreak. The regional level risk is achieved by computing the node strength statistic for all nodes of the network. Node strength is defined as the sum of all weights incident to a node. In the case of a directed network, out-strength, i.e., the sum of all outgoing link weights, and in-strength, i.e., the sum of all incoming link weights, are calculated to provide the relative export and import risks. Node strength values can be used to rank the regions according to outgoing (export, Fig 6B) and incoming risk (import, Fig 6C and 6D), and more critically, observe how the ranking and magnitudes of the risk change over the course of the outbreak.
To determine the regions most likely to contribute to the spread of Zika virus during the epidemic, we estimated the the dynamic exportation risk using the route-level network (Fig 6B). Martinique, Brazil, Colombia, Puerto Rico and Mexico are identified by the model to pose the highest risk of spreading Zika to new regions. Intuitively, the export risk is dominated by the set of counties infected earlier in the outbreak and those with high incidence rates. Martinique stands out as having the highest exportation risk, which peaks during March, corresponding to the month with the highest incidence rate (Fig 1B). The highest exportation risk through 2015 is posed by Brazil and Colombia, which were the two countries reporting early and large outbreaks (Fig 1A). Brazil was identified as the likely source of spread to the first few Caribbean Islands, which is consistent with the phylogenetic data [3, 4]. Our model estimates that Brazil and Colombia became less prominent in their roles of seeding new Zika virus outbreaks from December, 2015, to May, 2016. This time period corresponds to the significant rise in the number of new regions reporting local Zika virus transmission (Fig 1C) and the rise of Zika incidence rates in many Caribbean Islands (Fig 1B). The increased exportation risk posed by Martinique during this time captures this behavior (Fig 6B).
We similarly aggregated the route-level network at each destination to determine temporal and regional importation risk (Fig 6C and 6D). The regions at highest risk of local transmission are dominated by the less developed countries and territories in Central America such as Belize, and the islands in the Caribbean, such as Haiti, Saint Lucia, Grenada, and Dominica. The high ranking of these regions is due to their low GDP (Fig 4A), high human population density (Fig 4B), and high vector suitability (Fig 3).
For the U.S., the states with the highest importation risks were Florida, Georgia, and South Carolina, mostly due to their high Ae. aegypti suitability (Fig 3) and incoming travel volume from the affected regions [24]. Compared to much of the Americas, however, these importation risks were low, predominantly due to their high GDP (Fig 4A). In fact, Florida was the only one of those three states (along with Texas) that reported local Zika virus transmission. These findings correspond with and reinforce previous route-level risk rankings, that is less developed regions are more likely to see local Zika virus transmission, given they meet the basic requirements for Ae. aegypti-borne virus transmission.
Much of the dynamic aspects of our network model are based upon reported Zika virus cases, which are likely biased and vastly underestimated. Moreover, there is often a substantial delay (3-12 months) between actual introduction of the virus and the first reported case [2–4, 24]. The reporting inaccuracies can be attributed to the high percentage of asymptomatic Zika cases and insufficient surveillance methods, among other factors. While we cannot feasibly correct for all reporting inaccuracies, we conducted sensitivity analyses to account for delays in case reporting (Fig 7). Both three- and six-months reporting delays were considered, and the model was re-run with corresponding shifts in the data to represent each assumption. While the coefficient rankings slightly changed, likely representing better fits between local vector suitability and Zika incidence rates, the general conclusions were unchanged—low GDP was still the best predictor of local Zika virus transmission. Thus, our results appear to be robust to some reporting inaccuracies.
Our work takes a major step towards improving our understanding of the risks associated with Zika virus spread and local transmission; however, there are certain limitations of this analysis that must be noted here and addressed in future research:
Our work enhances our understanding of and ability to investigate the risk factors which contributed to the spread and local transmission of Zika virus during the 2015-2016 epidemic in the Americas. For each region, our model is informed by data on regional socioeconomic factors, vector habitat suitability, passenger air travel data, and epidemiological data. We constructed and implemented a dynamic extension of the GIIM to estimate the contribution of each risk factor to the likelihood of Zika virus transmission. Our model relies on a multi-agent based optimization method to estimate the parameters and a data driven stochastic-dynamic epidemic model for evaluation. The GIIM was shown to perform well based on quantitative metrics.
Our results from the final model indicate the spread and local transmission of Zika virus was quite multifaceted. As expected, regional attributes influencing vectors (Ae. aegypti suitability), hosts (human population density), and viruses (Zika incidence rates at origin of travel) all contributed to the likelihood of establishing local mosquito-borne transmission. Passenger air travel volume, however, was notably less impactful that the other attributes. Therefore, rather than travel restrictions, we predict that mosquito control will be more effective at reducing Zika virus introductions leading to local transmission. This debate recently arose during the 2016 Rio Summer Olympics where some wanted to ban the games to prevent further Zika virus spread [62]. Our results suggest that additional travel for the Olympics was highly unlikely to make a significant impact.
From our model, the coefficient most associated with Zika virus transmission was the regional GDP per capita, where a lower GDP corresponded to higher transmission risk. Although GDP does not directly influence transmission, it likely serves as a proxy for mosquito-host interactions [63] and surveillance activities. For example, people living in poverty often do not have the means to protect themselves from host seeking mosquitoes, such as air conditioning and screened windows common in higher income areas. Findings by Netto et al. [64] of higher Zika virus seroprevalence in areas with lower socioeconomic status further support our association. This, now evidence based conclusion that Zika and other Ae. aegypti-borne viruses should be considered diseases of poverty, is also consistent with other expert opinions [37, 65, 66].
The two most significant risk factors identified in our work, namely GDP and population density, are often excluded in geographic risk profiling of Aedes vector-borne diseases, and should be considered in future analysis. Our model is not specific for Zika virus and could easily be employed for other mosquito-borne viruses, such as dengue and chikungunya, with sufficient epidemiological and entomological data. Furthermore, the model could be adapted as a tool to inform real-time policy decisions regarding resource allocation for destination-based surveillance and vector control.
|
10.1371/journal.pgen.1007774 | How many individuals share a mitochondrial genome? | Mitochondrial DNA (mtDNA) is useful to assist with identification of the source of a biological sample, or to confirm matrilineal relatedness. Although the autosomal genome is much larger, mtDNA has an advantage for forensic applications of multiple copy number per cell, allowing better recovery of sequence information from degraded samples. In addition, biological samples such as fingernails, old bones, teeth and hair have mtDNA but little or no autosomal DNA. The relatively low mutation rate of the mitochondrial genome (mitogenome) means that there can be large sets of matrilineal-related individuals sharing a common mitogenome. Here we present the mitolina simulation software that we use to describe the distribution of the number of mitogenomes in a population that match a given mitogenome, and investigate its dependence on population size and growth rate, and on a database count of the mitogenome. Further, we report on the distribution of the number of meioses separating pairs of individuals with matching mitogenome. Our results have important implications for assessing the weight of mtDNA profile evidence in forensic science, but mtDNA analysis has many non-human applications, for example in tracking the source of ivory. Our methods and software can also be used for simulations to help validate models of population history in human or non-human populations.
| The maternally-inherited mitochondrial DNA (mtDNA) represents only a small fraction of the human genome, but mtDNA profiles are important in forensic science, for example when a biological evidence sample is degraded or when maternal relatedness is questioned. For forensic mtDNA analysis, it is important to know how many individuals share an mtDNA profile. We present a simulation model of mtDNA profile evolution, implemented in open-source software, and use it to describe the distribution of the number of individuals with matching mitogenomes, and their matrilineal relatedness. The latter is measured as the number of mother-child pairs in the lineage linking two matching individuals. We also describe how these distributions change when conditioning on a count of the profile in a frequency database.
| Human mitochondrial DNA (mtDNA) has long been a useful tool to identify war casualties and victims of mass disasters, the sources of biological samples derived from crime scenes or to confirm matrilineal relatedness [1–3]. The autosomal genome is much larger and has higher discriminatory power, but the mitochondrial genome (mitogenome) has multiple copies per cell, allowing better recovery of sequence information from degraded samples [1, 3], including ancient DNA [4, 5]. Some biological samples such as fingernails, old bones, teeth and hair have mtDNA but little or heavily degraded autosomal DNA. In addition, because of the lack of recombination, mtDNA can be used to confirm relatedness over many more generations than is possible using autosomal DNA, though only in the female line.
It has now become widely feasible to sequence all 16,568 mitogenome sites as part of a forensic investigation [6–8]. For autosomal short tandem repeat (STR) profiles, there are two alleles per locus and because of the effects of recombination, the alleles at distinct loci are treated as independent, after any adjustments for sample size, coancestry and direct relatedness [9]. In contrast, the maternally-inherited mitogenome is non-recombining, behaving like a single locus at which many alleles, or haplotypes, can arise. Due to relatedness and limited population size, the variation in mitogenomes in any extant population is greatly restricted compared with what is potentially available given the genome length. Whereas a match of two mitogenomes without recent shared ancestry is in effect impossible, there can be large sets of individuals sharing the same mitogenome due to matrilineal relatedness that is distant compared with known relatives but much closer than is typical for pairs of individuals in the population.
This limited variation has important implications for the use of mtDNA to help identify individuals or establish relatedness. A match between the mtDNA obtained from bones found under a Leicester UK carpark and a living matrilineal relative of the former King of England, Richard III, played an important role in establishing the bones as those of the king. However, in contrast with popular reports of genetic evidence “proving” the identification, the mtDNA evidence was not decisive, contributing a likelihood ratio (LR) of 478 towards an overall LR of 6.7 million in favour of the identification [10]. Although that mitogenome was at the time unobserved in the available databases, its observation in both the skeleton and a contemporary individual meant that it was expected to exist in hundreds and perhaps thousands of others. The public interest in the story led to multiple matches being subsequently observed in contemporary individuals, raising the question of how many humans alive today share this “royal” mitogenome?
We recently addressed similar questions for paternally-inherited Y chromosome profiles [11]. Forensic Y profiles focus on a few tens of STR loci, but these can have a combined mutation rate as high as 1 per 7 generations [11, 12], much higher than the mutation rate for the entire mitogenome, for which estimates range up to around 1 per 70 generations (see Methods). We showed that the high mutation rate of Y profiles has dramatic consequences for evaluating weight of evidence. For example, males with matching Y profiles are related through a lineage of up to a few tens of meioses. Further, the number of males with a matching Y profile varies only weakly with population size, and since the population size relevant to a forensic identification problem is typically unknown, it follows that the concept of a match probability that can be useful for autosomal DNA profiles is of little value for Y profiles.
Because of the lower mutation rate for the mitogenome, the situation is less extreme for mtDNA profiles than for Y profiles. Here we describe the distribution of the number of individuals with the same mitogenome as a randomly-chosen individual under three demographic scenarios and two mitogenome mutation models, finding that the number is typically of the order of hundreds rather than the tens that share a Y profile. The number of mitogenome matches is consequently more sensitive to demographic factors than is the case for Y profiles, but it remains a small fraction of the population relevant to a typical crime scenario. As we did previously for Y profiles, we also describe the conditional distributions given database frequencies for the observed mitogenome, assuming that the database is randomly sampled in the population. We show for example that a mitogenome that is unobserved in a large database can nevertheless exist in hundreds of individuals in the population. We also show that individuals sharing a mitogenome are related, typically within up to a few hundred meioses, which is much more distant than recognised relationships but still much closer than the relatedness of random pairs of individuals in a large population. Therefore the matching individuals may not be well-mixed in the population so that database statistics can be an unreliable guide to the number of matching individuals in the population.
See Methods for details of our two mutation models, based on [13] and [14], and three demographic scenarios which we denote 1.2M growth, 1.2M constant and 300K constant (suffix M for 106, i.e. millions, and suffix K for 103, i.e. thousands).
As for Y profiles, it is difficult to rigorously check our simulation models against empirical databases because real-world databases often result from informal sampling schemes that are far from random samples. They are often drawn from a much larger population than is relevant to a specific crime scenario, and sometimes from a number of different administrative regions such as states. However, broad-brush comparisons are useful, because while the databases are not scientific in their design, the resulting deviations from population values may not be very large. For this purpose we identified a US Caucasian database of 263 mitogenomes [15], which includes 259 distinct haplotypes, a very high level of diversity (259/263 = 98%) that reflects sampling from many US states. Most of our simulated databases of size 263 show less haplotype diversity than this database, but those under the 1.2M constant model come close (Fig 1 and S1 Fig). We also considered an Iranian database [16] of size 352 with 315 distinct haplotypes (89% diversity). This total included several distinct ethnic identities: Persians (181, 91% diversity), Qashqais (112, 84% diversity) and Azeris (22, 100% diversity). The simulated databases of size 352 under the 1.2M growth and 300K constant models show mtDNA diversity close to that of the Iranian database.
Low mitogenome diversity has been reported in three Philippines ethnic groups with 39, 43 and 27 mitogenomes yielding a diversity of 51%, 58% and 81% [17], which may reflect low population size and isolation. These lower levels of diversity may be appropriate in some forensic contexts, and can be analysed with our methods using a smaller population size than the examples presented here.
For both mutation schemes, Fig 2 (black curves, which are the same in each row) shows the cumulative distribution of the number of mitogenomes in the live population matching that of the PoI (person of interest). The distributions (see Table 1 for quantiles) are similar for the 1.2M and 300K constant models (middle and right columns), with the number of sequence matches with the PoI almost always < 1,000, but for 1.2M growth model some PoI have > 5,000 matches.
These distributions are altered by conditioning on an observation of m matches in a randomly-sampled database of size n (Fig 2, coloured curves). For the largest database we now see a clear difference between the two constant-size populations. For example m = 10 represents 0.1% of the database, consistent with 300 matches in the smaller population, a value that is well supported by the unconditional distribution and so the conditional distribution is centred around 300. However, 0.1% of the larger population is 1,200, which is not supported by the unconditional distribution and so the conditional distribution is shifted towards lower values, with most support between about 600 and 1,200. There is a similar effect for the m = 10 conditional distribution in the 1.2M growth population (note the different x-axis scale).
Estimated quantiles for the solid curves in the middle column of Fig 2 are given in Table 2. For the other two demographic scenarios under the Översti mutation scheme [13], see S1 Table (300K constant) and S2 Table (1.2M growth). Corresponding quantiles for the Rieux mutation scheme [14] are given in S3 Table (1.2M constant), S4 Table (300K constant) and S5 Table (1.2M growth).
The number of meioses separating individuals with matching mitogenomes ranges up to a few hundred, and is almost never larger than 500 (Fig 3). This is close to unrelated for most practical purposes, but random pairs of individuals are very unlikely to be related within 1,000 meioses, and so pairs with matching mitogenomes are much more closely related than average pairs of individuals. Key quantiles for the distributions of matching pairs are given in Table 3. As a guide for comparison, a coalescent theory approximation [18] for the mean numbers of meioses separating a random pair are 100K and 400K for our small and large constant-size populations, respectively.
Empirical mitogenome databases do not in practice represent random samples from a well-defined population, so that detailed comparisons with our simulation models are not meaningful. However, we have verified here that the haplotype diversity generated by our simulation models is broadly comparable with that observed in two real databases from large populations.
In our related paper on Y profile matching [11], we showed that because of the high mutation rates of contemporary Y profiles, the numbers of males with Y profile matching a PoI (person of interest) are low, typically up to a few tens, and that this number is little affected by population size or growth. Moreover the clusters of matching males are related within a few tens of meioses and so are unlikely to be randomly distributed in the population relevant to a typical crime scene. We argued that it was therefore not appropriate to report a match probability (a special case of the likelihood ratio) to measure the weight of evidence, even though likelihood ratios are central to the evaluation of autosomal DNA profiles.
In the present paper we have shown that the situation for mtDNA evidence is intermediate between Y and autosomal profiles. Because the whole-mitogenome mutation rate is an order of magnitude smaller than the mutation rate for contemporary Y profiles, the number of individuals matching a PoI is correspondingly larger, and varies more with demography. The unconditional distribution (Table 1) is very similar for the two constant-size populations that differ in size by a factor of four, but for the growing population the median number of matches is about twice as big. As for the case of Y profiles, our simulation-based approach can easily take into account information from a frequency database, although this requires the assumption that the database is a random sample from the population, which is rarely the case in practice.
The mitolina software that we have presented here can be used to inform the evaluation of the weight of mtDNA evidence in forensic applications, similar to our recommended approach to presenting Y-profile evidence: simulation models are used to obtain an estimate of the number of individuals sharing the evidence sample mitogenome, with conditioning on a database frequency if available. Current methods for evaluating mtDNA evidence rely directly on a database count of the observed mitogenome [2, 3], and are affected by poor representativeness of the databases, and its limited informativeness when there are many rare mitotypes. Our approach can also make use of a database count of the haplotype, but this information is used to adjust an unconditional distribution and so is less sensitive to the database size and sampling scheme.
Limitations of our analysis include the range of demographic scenarios that we can consider, and the difficulty in assessing which demographic scenario is appropriate for any specific crime. Our assumption of neutrality is unlikely to be strictly accurate [19], nor our assumption of a generation time of 25 years, constant over generations. We used two mutation rate schemes [13, 14] based on phylogenetic estimates, as no pedigree-based mutation rates were available for the entire mitogenome. Some discrepancy has been noted between the two estimation methods [20], and the rate may have changed over time [21]. If contemporary pedigree-based mutation rates become available we could improve our mutation model, but that would not address mutation rate changes over time. We have not here addressed the case of mixed mtDNA samples or heteroplasmy (multiple mitogenomes arising from the same individual).
While we have focussed our examples on human populations because of the important role of the mitogenome in human identification and relatedness testing, with appropriate modifications of the demographic model, mitolina and the methods described here can be used for non-human applications of mtDNA. Examples include tracking the source of ivory [22], other areas of wildlife forensics [23] and inferences about the demographic histories of natural populations [24]. Our software may be useful for generating simulation data in approximate Bayesian computation and related methods, and the number of matching sequences may also provide a useful summary statistic for such methods.
We simulated the mitogenome as a binary sequence subject to neutral mutations, using the rates estimated by both Rieux et al. (2014) [14] and Översti et al. (2017) [13], shown in Table 4. They both partitioned the mitogenome into four regions: hypervariable 1+2 (HVS1 + HVS2), protein coding codon 1+2 (PC1 + PC2), protein coding codon 3 (PC3), and ribosomal-RNA + transfer-RNA (rRNA + tRNA). However, the HVS1 + HVS2 region of [14] consisted of 698 sites whereas that of [13] had 1,122 sites, although their total mutation rate estimates for the region are similar.
We simulated populations of mitogenomes under three demographic scenarios. Two constant-size Wright-Fisher populations [25], of 50K and 200K females per generation, were simulated for 1,200 generations. The third scenario started with a constant female population size of 10,257 for 1,000 generations, followed by growth at a rate at 2% per generation over 150 generations to reach a final generation with 200K females. Following [11], individuals in the final three generations are considered to be “live”, and in those generations males were also simulated making total live population sizes of 300K, 1.2M and 1.2M. All the females in any generation had the same distribution of offspring number (no between-female variation in reproductive success).
We assigned mitogenomes to the founders randomly with replacement from a US Caucasian database of 263 mitogenomes (259 distinct haplotypes, see Fig 1) [15], coding each site as 0 if it matched the rCRS reference sequence [8], and 1 otherwise. Each mother-child transmission was subject to mutation, which changed a 0 to a 1, and vice versa. The same mutation rate was assigned to each site within each region, sampled from a normal distribution with 95% interval from Table 4.
The mean whole-mitogenome mutation rate per generation was 0.0135 for [13] and 0.0110 for [14], or about 1 mutation per 74 generations and 1 per 90 generations, respectively. Therefore, following one line of descent over 1,200 generations, the expected numbers of mutations to affect the mitogenome are 16.3 using [13] and 13.2 using [14]. The probabilities that there is any site affected by two mutations and so reverts to its original state during those 1,200 generations are 0.024 and 0.033, respectively.
We simulated five population under each of the three demographic scenarios. For each population simulation and both mutation models, we conducted five replicates of the sequence evolution process: assigning sequences to the founders and then mutations at each meiosis. Thus, for each mutation model and demographic scenario, 25 live populations of mitogenomes were created. In each live population, a PoI (person of interest) was randomly drawn 10,000 times, and we recorded how many live individuals had the same mitogenome as the PoI. Thus, a total of 5 × 5 × 10K = 250K PoIs were sampled for each mutation and demography combination. Further, for 10% of the PoI, the number of meioses between the PoI and each matching individual was recorded.
Following the methodology of [11], in addition to the unconditional distribution of the number of mitogenome matches between a PoI and another live individual, we use importance sampling reweighting to approximate the distribution conditional on observing the PoI mitogenome m times in a database of size n, assumed to have been chosen randomly in the population.
Software to perform these simulations is implemented in the open-source R packages mitolina [26, 27], based on Rcpp [28], and malan [29], previously used for Y profile simulations [11].
|
10.1371/journal.pcbi.1000485 | FLORA: A Novel Method to Predict Protein Function from Structure in Diverse Superfamilies | Predicting protein function from structure remains an active area of interest, particularly for the structural genomics initiatives where a substantial number of structures are initially solved with little or no functional characterisation. Although global structure comparison methods can be used to transfer functional annotations, the relationship between fold and function is complex, particularly in functionally diverse superfamilies that have evolved through different secondary structure embellishments to a common structural core. The majority of prediction algorithms employ local templates built on known or predicted functional residues. Here, we present a novel method (FLORA) that automatically generates structural motifs associated with different functional sub-families (FSGs) within functionally diverse domain superfamilies. Templates are created purely on the basis of their specificity for a given FSG, and the method makes no prior prediction of functional sites, nor assumes specific physico-chemical properties of residues. FLORA is able to accurately discriminate between homologous domains with different functions and substantially outperforms (a 2–3 fold increase in coverage at low error rates) popular structure comparison methods and a leading function prediction method. We benchmark FLORA on a large data set of enzyme superfamilies from all three major protein classes (α, β, αβ) and demonstrate the functional relevance of the motifs it identifies. We also provide novel predictions of enzymatic activity for a large number of structures solved by the Protein Structure Initiative. Overall, we show that FLORA is able to effectively detect functionally similar protein domain structures by purely using patterns of structural conservation of all residues.
| Understanding how the three-dimensional (3D) molecular structure of proteins influences their function can provide insights into the workings of biological systems. Structural Genomics Initiatives have been set up to investigate these structures on a large scale and make the data available to the wider biological research community. However, in a significant number of cases, there is little known about the functions of the structures that are solved. To address this, computational methods can be used as a predictive tool to guide future experimental investigations. One such approach is to exploit global structural comparison to assign the protein in question to an evolutionary family, which has already been functionally characterised. However, this is problematic in some large evolutionary families, which contain a number of different functional sub-families. We have developed a new method (FLORA) which is able to calculate 3D “motifs” which are specific to each of these sub-families. Any new protein structure can then be compared against these motifs to make a more accurate prediction of its function. Our paper shows that FLORA substantially outperforms other standard approaches for predicting function from structure. We use our method to make confident functional predictions for a set of proteins solved by the structural genomics projects, which could not have been assigned reliably by global structure comparison.
| The prediction of protein function from structure has become of increasing interest as a significant proportion [1] of structures solved by the structural genomics initiatives (SGI) lack functional annotation (for a review see [2]). Furthermore, structure-based approaches are of particular interest for predicting binding sites and/or catalytic sites for the purposes of protein engineering and pharmaceutical development (for reviews see [2],[3]). Many current methods focus on encoding a “template” of functional residues and then aligning this template to whole structures. The problems with taking this approach are deciding what qualifies as a functional residue (e.g. one directly involved in catalysis or ligand binding) and creating biologically-accurate templates for the ever increasing number of available protein structures being deposited in the PDB [4]. Resources such as the Catalytic Site Atlas [5] are carefully curated by hand and restricted to residues directly involved in catalysis, whereas MSDSite [6] and PDBSite [7],[8] generate templates based on active site residues defined in the PDB file by the authors. Although these resources are undoubtedly extremely valuable, it is questionable whether sufficient coverage of the PDB can be maintained when manual intervention is required.
To address the problem of generating templates for all protein structures, there are a number of methods that aim to do this automatically. For example, the reverse template method [1] (available as part of the PROFUNC suite [9]) decomposes a query structure into tri-peptide fragments (putative catalytic triads), which are then matched against a non-redundant set of PDB structures using the search algorithm JESS [1]. Hits are evaluated according to the sequence similarity of the local environment of the template. The GASP method [10] uses a genetic algorithm to construct templates based on their ability to discriminate between different protein families against a background of representatives from the SCOP database [11]. Similarly, DRESPAT [12] implements a graph theoretical approach to discover structural patterns associated with a given family of proteins to locate ligand binding motifs (the PINTS method [13] uses a related approach). MultiProt [14] can provide template of structures through a multiple structure alignment. A recent extension of the Evolutionary Trace method for binding site prediction was used to create structural templates based on predicted functional residues [15]. SiteEngines [16] produces templates by matching the geometry and physico-chemical properties of residues in binding site clefts. As well as atom or residue-level templates, other non-template-based approaches seek to compare the electrostatic properties of binding sites (ef-Site, [17], SURF's UP [18]) or surface accessible clefts which often co-locate with active sites (pvSOAR (CASTp) [19]).
One inherent complexity of using PDB structures to transfer annotations between enzymes is the binding state in which the protein is crystallised — for example, structures crystallised with non-cognate ligands, substrate analogs, transition states or apo-enzymes [20]. As a consequence, precise geometric matching in the active site region can be problematic due to the conformational changes that occur on ligand binding. To address this issue, the methods mentioned above use a variety of approaches such as graph matching or geometric hashing with various tolerance levels. The SOIPPA method [21],[22] takes the alternative approach of using a “geometric potential” to characterise the shape formed by a given set of Cα atoms, to account for both local and global relationships between residues across the protein structure. In a recent ligand-binding site comparison analysis, SOIPPA was able to detect distant similarities between very different protein folds binding a range of adenine-containing ligands [21].
Despite the many template methods present in the literature, very few are publicly available to the general user. Hence, the first step in assigning function by structure is often to use global structure comparison methods (e.g. CE [23], DALI [24], CATHEDRAL [25], MAMMOTH [26], FatCat [27], MSDFold [28]), which can detect distant evolutionary relationships even where sequence similarity is weak. These methods have been specifically applied to function prediction (ProKnow [29], Annolite [30]) to assign confidence values when inheriting GO terms between related structures. However, detecting very distant relatives remains a challenge as structure comparison methods generally give an absolute measure (or score) of structural distance, such as RMSD, and applying a cut-off at which one can deduce that two proteins perform related functions results in many missed relationships.
Analyses of CATH [31],[32] have shown that although function and structure are well conserved in the majority of superfamilies, there are a significant number of highly diverse superfamilies where this is not the case [31]. Moreover, the latter superfamilies are disproportionately represented in both the PDB and in the genomes and tend to exhibit a wide range of core biological functions across a large range of species [33]. An analysis by Reeves et al. [31] showed that relatives within these superfamilies tend to share a common evolutionary core, but this core is embellished with different insertions of secondary structure elements that often correlate with changes in function. However, although structural embellishments might change some facet of function (e.g. ligand specificity, protein-protein interactions), others have found that relatives can still retain other aspects in common (e.g. catalytic mechanism, such as kinase activity) [34],[35]. Therefore, calculating a global measure of structural similarity or distance (e.g. RMSD) between two proteins can be less informative than focussing on the structural motifs relevant to a given aspect of function.
The FLORA algorithm presented here was designed to derive structural templates for functional sub-groups (FSGs) within diverse CATH superfamilies. FLORA first performs global structure alignment across the superfamily to recognise the distinctive structural patterns associated with each FSG and builds templates based on these patterns. New functional homologues are then detected by using the global structural alignments to relatives in each FSG again, but only scoring the similarity over positions identified by the FLORA motif. This approach performs very well in discriminating between different enzymatic functions, compared to global methods and another motif-based approach. Although we benchmark here on enzyme superfamilies, the method is applicable to superfamilies containing non-enzymatic relatives. To test FLORA, we have automatically generated a large data set of domains from 29 diverse superfamilies (containing multiple FSGs). Our data set allows us to look at the variation of FLORA results between superfamilies and to stress the importance of using a large test data set for benchmarking methods. We have benchmarked FLORA against CE [23], CATHEDRAL [25] and Reverse Templates (RT) [1] to give an indication of how it performs in comparison to other standard methods of function prediction. We also present some examples of structural motifs identified by FLORA and explain their functional relevance. Finally, we use FLORA to make novel predictions of function for proteins solved by the Protein Structure Initiative (PSI).
In order to benchmark FLORA as a protein function prediction method, it was important to generate a relatively large and unbiased data set. We focussed on functionally diverse superfamilies (≥3 functions at the third E.C. [36] level) in the CATH database, where global fold similarity and evidence of homology is not necessarily indicative of a functional similarity. An overview of the protocol is shown in Figure 1.
All protein chains from PDB structures classified in CATH v3.1 were annotated with an E.C. number using PDBSprotEC [37], which maps PDB chains to corresponding entries in the SwissProt database [38]. E.C. annotations were then transferred from the whole chain level to each constituent domain in a chain. Assigning functional annotation to individual domains is not a straight-forward process, as other domains in the chain (or indeed, residues from other chains in the protein) may be required for the enzyme to be catalytically active. This problem is dealt with more extensively in the PROCOGNATE resource [39]. However, we were only interested in finding domains that were “associated” with proteins of a given enzymatic function, as FLORA was designed to consider all residues for inclusion in a template and not just those in the active site.
To simplify the benchmark data set, all domains from enzymes assigned more than one E.C. (i.e. multifunctional enzymes) were removed. This exclusion criterion removed less than 8% of enzymatic chains in the PDB. In addition, any domains with an incomplete E.C. number (e.g. 2.7.-.-) were also excluded.
All annotated domains in CATH were clustered at 60% sequence identity and a representative taken from each cluster (S60Rep). This threshold was applied as 60% has been found to be an appropriate sequence cut-off for functional similarity [40],[41]. Discovering homologous domains sharing more than 60% sequence identity is trivial using BLAST [42] and other sequence-base methods and we wished to generate a benchmark data set that contained more challenging cases.
S60Reps were then grouped within the superfamily if they shared at least the first three E.C. numbers; to create what we will subsequently refer to as a functional sub-group (FSG). A CATH superfamily was then included in the data set if it contained at least 3 FSGs, where each enzyme family contained at least 4 S60Reps. These criteria were chosen to create a sufficiently diverse data set, which could be effectively assessed using leave-one-out benchmarking.
The final domain data set (Dataset S1) comprised: 82 FSGs from 29 different CATH superfamilies (a total of 911 S60Reps domains), covering all 3 major protein classes (α, beta and mixed α-beta). Although the data set comprises ∼2% of the total number of superfamilies in CATH, these superfamilies account for ∼48% of domain sequences from functionally diverse superfamilies in Uniprot. Furthermore, they represent a set of domains where global fold similarity does not necessarily correlate with functional similarity.
An outline of the FLORAMake algorithm is shown in Figure 2. The aim was to select a set of conserved vectors from a given domain in a given FSG which when compared against relatives of different functions/FSGs would produce a low score and similarly a high score to relatives with the same function.
As FLORA is essentially a pattern discovery method, it was vital to assess its performance in an unbiased fashion. We took a standard leave-one-out (or jack-knifing) approach as is often used to test machine learning methods. For each superfamily, one test domain was removed, while training on the remaining domains. The test domain was then scored against all the resulting templates. The aim of this process to was accurately reproduce a situation where a novel domain is classified into a CATH superfamily and then needs to be assigned to a functional group.
The performance of FLORA, CATHEDRAL [25], CE [23] and Reverse Template (RT) [1] were analysed by plotting sensitivity (i.e. tp/(tp+fn)) versus precision (tp/(tp+fp)). We compared the performance on individual superfamilies by calculating AUC value (area under ROC curve).
In order to examine where residues identified by FLORA overlapped with known functional residues, we compared the location of FLORA positions to those in the Catalytic Site Atlas [5] (v2.2.9).
For each functional sub-group (FSG), we selected the domain that had the highest mean global structural similarity (measured by CATHEDRAL) to all other members of the FSG as a representative. All residues, from each relative within an FSG, identified by FLORA and CSA annotations were then mapped onto this representative using the CATHEDRAL structural alignment. Consequently, for each FSG we had a representative structure where all residues were annotated as FLORA positions, catalytic residues, or neither. The CSA provided annotations for 61 out of 82 FSGs (74%). We then calculated the average distance between the FLORA residues to the catalytic residues and the average distance between non-FLORA and the catalytic residues.
FLORA produces a set of inter-residue vectors for each domain in a given FSG that are considered to be specific to its enzymatic function, in the context of its evolutionary superfamily. In order to visualise where these vectors lay, we took each set of domain templates for a given enzyme family and mapped them onto the most representative structure — i.e. the structure with the greatest cumulative global structural similarity to all other domains in the family. A given residue was then coloured if it was involved in the top 30% of FLORA template vectors. Residues that are conserved across the whole superfamily (in 75% of relatives) were also identified and those which overlapped with FLORA residues were coloured gold.
Despite targeting proteins with no significant sequence similarity to existing structures in the PDB, Protein Structure Initiative (PSI) structures can often be classified into one of the large, diverse superfamilies in CATH by structure comparison methods once their structure has been solved. However, these superfamilies contain a significant number of relatives with different functions and therefore to be able to further assign these proteins to a specific functional sub-group is of great use for guiding future functional studies. We took all PSI structures solved up to January 2008 that had been newly classified in v3.2 of the CATH database and selected the 276 domains which fell into one the superfamilies in our data set. These 276 were further clustered at 60% sequence identity to produce a non-redundant test set of 104 domains, which was then scanned against the FLORA templates for each FSG in order to predict their function. To exclude hits that could have been fairly confidently assigned using global structure comparison, we removed any structures that matched a CATH domain in v3.1 library with a SIMAX score<1.5 [25].
FLORA was designed as a generic method to create structural motifs that can discriminate between different functional sub-groups (FSGs) within diverse domain superfamilies, purely using patterns of structural conservation — FLORA makes no assumptions as to the physico-chemical properties of functionally important residues and uses a purely structure-based conservation score (i.e. sequence similarity is not used to select or score equivalent motif vectors, see Methods). We created a benchmark data set of diverse enzyme superfamilies in the CATH database [44], although FLORA can be applied to protein structures grouped by any function or superfamily annotation scheme.
We tested the performance of FLORA against global structure comparison methods (CE [23], CATHEDRAL [25]) and the Reverse Template (RT) method [1]. The residue positions identified by the FLORA templates were examined to determine whether they co-located to functional regions in the protein structures. Finally, we used FLORA to predict broad enzymatic functions for a set of structural genomics targets solved by the Protein Structure Initiative [45].
To fairly benchmark any function prediction algorithm, it is important to compare against current methods. Unfortunately, the vast majority of function prediction methods are not publicly available, however here we compare against CE as this method has been used as a benchmark for other structure-based function prediction methods (e.g. [10],[21]). We also compare the performance of FLORA against a more sensitive structure comparison method (CATHEDRAL [25]) and a leading function prediction method (RT [1]).
Initially, we investigated to what extent global structure comparison could be used to reliably assign function. The graph of sensitivity versus precision (Figure 3) shows the ability of CE and CATHEDRAL to discriminate between domains in the same enzyme family across our entire data set. It can be seen that at high precision (∼90%), CATHEDRAL outperforms CE, although the sensitivity is still very low (18%). We suspect that the superior performance of CATHEDRAL over CE is due to the fact that it is able to generate improved alignments of homologous structures by aligning more equivalent residues (as shown in [25]). The performance of both methods shown here is fairly poor for correctly classifying domains into FSGs, but it is obviously important to note that neither of the methods was designed to detect functional relationships.
FLORAMake and FLORAScan were applied to the domain data set and the performance was assessed using a leave-one-out approach (described in the Methods section). It can be seen from Figure 3 that even at high precision, FLORA significantly outperforms CATHEDRAL, CE and RT — e.g. 90% precision, CATHEDRAL detects only 15% of true functional homologues, versus 27% for RT and 61% for FLORA. These results show that the FLORA algorithm significantly outperforms global structure comparison. This can be explained by the fact that although FLORA uses the same alignments as CATHEDRAL, it only scores those positions which have been identified as functionally-relevant (i.e. captured by the FLORA template) within a given FSG. Furthermore, FLORA uses data from multiple structures and is able to accurately discover functionally-relevant structural motifs and discover more than twice the number of functional homologues at 90% precision than RT. This suggests that where the data are available, exploiting multiple structures with similar functions can improve the sensitivity of function prediction methods. However, where these is not available, methods such as RT [1] can be very valuable.
FLORA was benchmarked on 29 functionally diverse enzyme superfamilies and the performance quoted thus far refers to an average calculated over the entire data set. Figure 4 shows the performance per superfamily (as measured by the Area Under ROC Curve (AUC)) for FLORA and CATHEDRAL. It can be seen that where FLORA is able to perfectly discriminate between domains in different functional sub-groups (i.e. AUC = 1.0), CATHEDRAL is also able to do so as functionally-similar domains must share high global structural similarity. However, for all but one (CATH code: 3.30.830.10) of the superfamilies in the data set, FLORA out-performs CATHEDRAL. Superfamily 3.30.830.10 comprises two FSGs (aminopeptidases and carboxypeptidases), which contain domains that are part of larger multi-domain complexes. For example, the protein chain 1hr6A actually contains two homologous yet non-identical domains (<30 sequence identity), both of which are members of this superfamily — i.e. a domain duplication has produced the multi-domain architecture 3.30.830.10::3.30.830.10. As a consequence, it is more biologically meaningful to align this superfamily at the chain level, which indeed improves the performance of FLORA (AUC increases from 0.32 to 0.88, see next section and Figure 5). Although there is only one example of this case in our data set, it will be important to account for domain duplications when building templates in the future. For example, we encountered similar problems in a superfamily of periplasmic binding domains (CATH 3.40.190.10), where a domain duplication creates a receptor of two halves involved in the transportation of small ligands (unpublished data).
At this point, it can be seen that simply focussing at the domain level FLORA is able to very effectively improve the recognition of structures in the same FSG. This is interesting given that the majority of structure-based function prediction methods tend to use the whole protein chain. A possible explanation of the power of FLORA could be that the domains in our data set form a core part of the enzymatically active region of the whole protein. Alternatively, it could be that the selected vectors for each template also contain residues that interact with other enzymatic domains within the chain, and it is these interaction sites that FLORA is detecting.
To see whether any improvement could be achieved by using the whole protein chain, we used CATHEDRAL to re-align the corresponding PDB chain for each of the domains in the data set and performed an identical benchmark as before. Figure 5 shows that the performance increase of using whole chains over using the component domains is minimal. This suggests that there is enough of a structural signal at the domain level and adding vectors from other domains in the protein chain does not seem to be advantageous. It also means that FLORA could be used to transfer functional annotation between relatives with different multi-domain architectures, therefore expanding the scope of the method.
The benchmarking analysis presented above shows that FLORA is indeed able to correctly discriminate between homologous domains from different FSGs better than global structure comparison, despite using global alignments to determine residue correspondence. This suggests that although a global alignment may not be perfect, especially between very distant relatives, it still aligns enough residues that are important for maintaining different functions. To examine where these function-specific residue lay, we chose a representative structure for each enzyme family and visualised the conserved FLORA residues (see Methods section).
We have analysed these motifs further in domains from the HUP superfamily (CATH 3.40.50.620 [46]), which is the subject of particular attention within our group. HUP domains are very diverse in terms of sequence, structure and function, and are involved in various essential biological processes (e.g. protein translation). In addition, several proteins with HUP domains have attracted attention due to their medical importance (e.g. [47]). Domains in this superfamily adopt a Rossmann-like fold with a central parallel β-sheet surrounded on both sides by α-helices. The main active site is always located in the C-terminal half of the central β-sheet and is generally involved in nucleotide-binding. HUP domains in the FLORA dataset divide into 3 major FSGs when clustered using the first three digits of the E.C. numbers. In the following section, we consider one representative member of each of these FSGs to describe motifs identified by FLORA.
The first FSG consists of the catalytic domain of class I aminoacyl-tRNA synthetases (EC 6.1.1.-). These enzymes are essential for protein translation as they catalyse the ligation of amino-acids to their cognate tRNAs in a two-step mechanism that involves ATP. The HUP domains of aminoacyl-tRNA synthetases are found in many different multi-domain contexts in CATH, which appear to partially depend on the amino-acid substrate (data not shown). In representatives from this group, (S. cerevisiae arginyl-tRNA synthetase, PDB: 1f7u), FLORA identifies two major motifs, one of which is located in the amino-acid and ATP binding site, whereas the other covers residues in loops that bind the tRNA (Figure 6A).
The next FSG in the HUP superfamily is a group of metabolic enzymes called nucleotidyltransferases (EC 2.7.7.-), which transfer nucleotidyl groups from nucleotide tri-phosphates to other compounds. The nucleotidyltransferase we have analysed further (Th. Thermophilus pantetheine phosphate adenylyltransferase PDB: 1od6), is a relatively small protein and consists of a homo-hexamer of single HUP domain subunits. FLORA identifies two motifs in this domain, one of which locates in the main active site in the C-terminal half of the central β-sheet, whereas the other maps to the inter-subunit interface (Figure 6B).
Finally, the third FSG consists exclusively of argininosuccinate synthases (EC 6.3.4.5), which catalyse the ATP-dependent synthesis of argininosuccinate from citrulline and aspartate. These enzymes are homo-tetramers in which each subunit is comprised of a nucleotide-binding HUP domain and an additional domain involved in multimerisation and catalysis. Three motifs are identified by FLORA in E. coli argininosuccinate synthase: one is located in the nucleotide-binding site (C-terminal half of the central β-sheet), another consists of residues at the interface with other subunits of the tetramer, whereas the third motif is comprised of residues from N-terminal α-helices that are not involved in any identified interactions to our knowledge (Figure 6C). The location of these α-helices on the outward surface of the tetramer cannot exclude the possibility that these FLORA residues might be involved in interactions that have yet to be described in the literature.
Analyses of residues identified by FLORA in these domains and others in this superfamily (data not shown) suggest that FLORA is generally able to target motifs known to be involved in different aspects of molecular function, like binding interfaces or catalytic sites. This behaviour is somewhat expected from FLORA, which was specifically designed to detect such function-related signatures in homologous domains. By mapping catalytic residues from the CSA onto each FSG representative (see Methods), we found that in 59% of cases the FLORA residues were closer to the functional site than other residues in the domain. This is interesting as it means that in a significant number of FSGs, FLORA is identifying other positions in the protein, for example those involved in interaction sites as demonstrated by the examples discussed above. In the particular case of the HUP superfamily mentioned above, it is noteworthy that in each FSG, FLORA not only identifies functional regions which are unique to the FSG (e.g. the tRNA binding site in aminoacyl-tRNA synthetases), but also residues in the main nucleotide-binding active site which is shared by HUP domains from all FSGs at the C-terminal half of the central β-sheet. Although this would require further investigation, it suggests that FLORA is able to detect relatively small differences in residue positions and orientations between similar active sites in different FSGs.
Examining similar representatives from the Class I aldolase superfamily (3.20.20.70) reveals that FLORA template residues (Text S1) tend to cluster around the active site of the enzymes (data on active site residues from the Catalytic Site Atlas [5]), which suggest that it is where the majority of structural features characteristic of each FSG occur.
Our analysis thus far has shown that FLORA is able to substantially improve on the performance of global structure comparison for reliably assigning domains to functional sub-groups. We therefore sought to use it to make novel predictions for structural genomics targets from the PSI. As a data set, we took structures that had been assigned to superfamilies in the latest version of CATH (v3.2) and scanned these against the FLORA templates. Using the benchmark curve from the leave-one-out benchmark, we took a score cut-off corresponding to a precision of 95% (Z-score>3.4) to ensure high confidence in our assignments. All hits above this cut-off were collated, rather than simply taking the top hit so that we could account for bi-functional enzymes and observe any conflicting predictions (i.e. those structures which hit more than one FSG template). A complete table of results is shown in Text S1.
104 domains from our v3.2 PSI set correspond to 94 PDB structures. Of these 94, we were able to make predictions for 66 (70.4%) with FLORA. To assess the added value of using FLORA over global structure comparison, we took out any PSI structures that matched a domain in CATH with a SIMAX score<1.5 (see Methods). This left us with 51/66 (78%) predictions that could not be easily assigned with CATHEDRAL. This supports the earlier benchmark of FLORA, which shows that scoring structural similarity over all FSG-specific residues can dramatically increase the number of functional homologues we are able to detect.
Figure 7 shows the structure of 2pbl (a putative thiol esterase from the Joint Center For Structural Genomics) superposed against its best hit 1epx. A closer superposition of the active site shows conservation of the surrounding secondary structures and even the positions of the catalytic residues. FLORA finds significant hits to all members of the FSG (E.C. 3.1.1-, Carboxylic ester hydrolases) in superfamily 3.40.50.1820, despite none of the domains superposing with an RMSD less than 4, indicating that 2pbl is a distant relative of other superfamily members. The other FSG in the superfamily corresponds to E.C. 3.4.16.-, which is a group of Serine-type carboxypeptidases to which FLORA assigns no significant hits. FLORA predicts 2pbl to be a carboxylic ester hydrolase, as opposed to a Thiolester hydrolase (E.C. 3.1.2) as suggested by the authors. However, given that there are no examples of thiolesterases currently in the superfamily it is possible that they are in fact closely related to the carboxylic ester hydrolases. Biochemically, this function is certainly closer than the peptidase function of FSG (EC 3.4.16.-).
FLORA predicted NESG structure 2bdt with the E.C. number 2.7.1.-, which is a group including enzymes such as fructose 1-,6 bisphosphate. When this structure was published, it was assigned as a putative gluconate kinase but currently has no official E.C. annotation.
PDB 1vm8 from the JESG consortium was functionally characterised when the structure was solved as UDP-n-acetylglucosamine pyrophosphatase and given the E.C. number E.C. 2.7.7.23. Again, FLORA correctly predicts the E.C. number as 2.7.7.-, despite low global structural similarity to any domains in the template data set.
1ylo is a hypothetical protein solved by the MCSG consortium in 2005. FLORA predicted the E.C. number 3.4.11.-, which comprises a group of amino-acid specific peptidases, with significant hits (Z-score>4) to three domain templates in our data set. A BLAST search indeed reveals significant hits (>99% sequence identity) to annotated amino peptidases, as the protein has now been functionally characterised since its structure was solved. Again, these trivial hits were not in the data set we used, which demonstrates the power of FLORA to find functional homologues even after significant evolutionary divergence.
FLORA is a novel algorithm which exploits patterns of structural conservation to derive templates for different functional sub-groups (FSGs) within diverse domain superfamilies. Unlike many other methods which focus on generating templates based on known or predicted functional residues [1],[10],[15], FLORA considers all residues to provide a more discriminating functional fingerprint. We have shown it is able to use these templates effectively to discriminate between domains with different functions better than global structure comparison (CATHEDRAL), CE and RT.
By generating a superfamily-specific Z-score, we found that the performance of FLORA increases significantly. This suggests that the degree of structural variation that confers a change in function is specific to each superfamily and the absolute structural similarity must be compared to a background distribution. Therefore, as has also been identified at the sequence level [40],[41], function prediction methods should account for the divergence of the superfamily, rather than adopt one similarity measure that applies to all superfamilies. However, we acknowledge that a representative distribution can only be obtained in sufficiently populated superfamilies.
Another important novelty in our approach was to create a large data set comprising 29 superfamilies (which is made publically available). Although FLORA performed well across the majority of superfamilies, this was not universally true, which suggests that function prediction methods should be benchmarked across as diverse a data set as possible. We have also shown that CATHEDRAL outperforms CE, probably due to producing superior alignments outside of the conserved structural core. Although global structure comparison is not always able to reliably find distant functional relatives, we feel it is appropriate for benchmarking new methods to give a guide of the value they add to structure-based function prediction.
As detailed in the methods, FLORA calculates vectors based on the geometry of Cβ side chain atoms. However, a re-implementation using just Cα co-ordinates produces almost identical performance on the data set (data not shown). This is encouraging as it increases applicability of our method to theoretical and homology-based models.
One of the major ways in which FLORA differs to other methods is by focussing on the domain, rather than at the whole chain or protein complex level. Simply because a domain is present in a given enzyme does not necessarily mean it contributes to or confers catalytic activity. Indeed it might be responsible for protein-protein interactions or other aspects of function, such as locating the protein in a given part of the cell. We have shown that except in the case where there has been a domain duplication (superfamily 3.30.830.10), deriving structural motifs at the domain level performs as well as aligning whole multi-domain chains. Our hypothesis is that where FLORA does not locate conserved positions around the active site, it is able to find parts of the domain that interact with other catalytic domains. We intend to undertake more detailed analysis of other CATH superfamilies to confirm this.
FLORA makes no assumptions about the physico-chemical (e.g. solvent accessibility or polarity) or sequence conservation properties of residues in the templates it derives, only that they show high structural conservation within a given functional sub-group. As a consequence, we observed residues both around the enzymatic active sites and in other locations in the protein. In two of the example superfamilies presented here, we have shown that FLORA template vectors co-locate around the active site. This is possibly due to structural changes in the protein that allow for different relatives to bind different ligands. However, this trend is not observed across the whole data set, where only 59% of FLORA template vectors are on average closer to the active site than other residues in the protein. This suggests that it is not only the enzymatic site that is important for discriminating between different FSGs, but other locations in the structure related to domain-domain or protein-protein interfaces.
The substantial improvement in performance of FLORA over global structure comparison has allowed us to assign 70% of structural genomics targets, assigned to superfamilies in our data set to functional sub-groups, in this case predicting the type of catalytic reaction they perform. Of our FLORA predictions, 78% could not have been reliably made by standard structure comparison techniques, as we were able to transfer annotation from far more distant relatives (RMSD>4 Å). Although some of the predictions we made are supported by experimental work that occurred after the structure was solved, the accuracy of the rest remains for future functional characterisation work.
Taken in the context of our previous analysis of functional divergence across large domain superfamilies in the CATH database [31], we have shown that it is indeed possible to derive structural templates that can be used to characterise these different functional sub-groups, without explicitly focussing on known or predicted catalytic residues. Both CATHEDRAL and FLORA exploit the same algorithm to align structures, but the performance increase observed by FLORA is due to the fact that it identifies those positions which are distinctive to a function group and only scores the structural similarity over these positions, whereas CATHEDRAL calculates a global score. Although we have benchmarked here using CATH enzyme superfamilies, FLORA can be applied to any other functional or superfamily classification (both enzyme and non-enzyme) where there are sufficient structural data. We are currently implementing FLORA as a web service for the structural biology community.
|
10.1371/journal.pcbi.1005243 | Hierarchical Post-transcriptional Regulation of Colicin E2 Expression in Escherichia coli | Post-transcriptional regulation of gene expression plays a crucial role in many bacterial pathways. In particular, the translation of mRNA can be regulated by trans-acting, small, non-coding RNAs (sRNAs) or mRNA-binding proteins, each of which has been successfully treated theoretically using two-component models. An important system that includes a combination of these modes of post-transcriptional regulation is the Colicin E2 system. DNA damage, by triggering the SOS response, leads to the heterogeneous expression of the Colicin E2 operon including the cea gene encoding the toxin colicin E2, and the cel gene that codes for the induction of cell lysis and release of colicin. Although previous studies have uncovered the system’s basic regulatory interactions, its dynamical behavior is still unknown. Here, we develop a simple, yet comprehensive, mathematical model of the colicin E2 regulatory network, and study its dynamics. Its post-transcriptional regulation can be reduced to three hierarchically ordered components: the mRNA including the cel gene, the mRNA-binding protein CsrA, and an effective sRNA that regulates CsrA. We demonstrate that the stationary state of this system exhibits a pronounced threshold in the abundance of free mRNA. As post-transcriptional regulation is known to be noisy, we performed a detailed stochastic analysis, and found fluctuations to be largest at production rates close to the threshold. The magnitude of fluctuations can be tuned by the rate of production of the sRNA. To study the dynamics in response to an SOS signal, we incorporated the LexA-RecA SOS response network into our model. We found that CsrA regulation filtered out short-lived activation peaks and caused a delay in lysis gene expression for prolonged SOS signals, which is also seen in experiments. Moreover, we showed that a stochastic SOS signal creates a broad lysis time distribution. Our model thus theoretically describes Colicin E2 expression dynamics in detail and reveals the importance of the specific regulatory components for the timing of toxin release.
| Gene expression is a fundamental biological process, in which living cells use genetic information to synthesize functional products like proteins. To control this process, cells make use of many different mechanisms. A well-studied example is the binding of expression intermediates by a cellular component in order to delay the synthesis. This mechanism is known to regulate the stress-induced release of the toxin colicin E2 by E. coli bacteria. However, experimental studies have shown that this system is not regulated by just one component, but the interplay of several cellular components, in which the hierarchically ordered main components interact. Here, we create a mathematical model for the interaction network of colicin E2 release, and study how the component levels evolve. We show that the system is able to delay the release of the toxin. Additional components allow to fine-tune the delay and dampen fluctuations in gene expression that would lead to premature toxin release. A comprehensive analysis of the time evolution reveals a broad distribution of toxin release times, which is also observed in experiments. This rich dynamical behavior emerges from the interplay of regulatory components, and, due to its generality, may also be transferred to similar regulatory networks, in particular toxin expression systems.
| Regulation of gene expression occurs at transcriptional and post-transcriptional levels, and has been studied intensively both experimentally and theoretically [1–10]. Bacterial stress responses, such as the well-studied production and release of the toxin colicin E2 in Escherichia coli, represent one setting in which post-transcriptional control is crucial [11–15].
Colicins are toxic proteins produced by certain E. coli strains in response to stress as a means to kill bacteria that compete with them for the same resources. More specificly, colicin E2 is a bacteriocin, which damages the DNA of bacterial cells that absorb it (a DNAse). Once synthesized, colicin E2 forms a complex with an immunity protein, thus protecting its producer from its otherwise lethal action [14, 16, 17]. The toxin is released only upon cell lysis, which is triggered by the synthesis of a dedicated lysis protein [15, 18–20]. As this inevitably entails the death of the producer cell [19], it is vital for the persistence of the population that only a fraction of its members actually releases the toxin [14]. The genes for the colicin, immunity protein and lysis protein are organized into the colicin E2 operon, which is depicted in Fig 1, together with the interaction network that controls colicin E2 expression and release.
Each of the three components is encoded by a single gene—the colicin by cea, the colicin-specific immunity protein by cei, and the lysis factor by cel—and three regulatory regions control their transcription: an SOS promoter upstream of the cea gene [21], and two transcriptional terminators T1 and T2, located upstream and downstream of the cel gene, respectively [22]. The key transcriptional regulator of the SOS operon is the LexA protein (reviewed in [23]), marked in orange in Fig 1. LexA dimers repress the SOS promoter region of the ColE2 operon, but also block the transcription of over 30 other SOS genes [24, 25], many of which play an important role in DNA repair [26]. In the event of DNA damage, the LexA dimer undergoes auto-cleavage upon interaction with RecA [27], and the transcription of SOS genes begins. The presence of the two transcriptional terminators in the ColE2 operon results in the production of two different mRNAs: A shorter transcript (short mRNA, marked purple in Fig 1) that encompasses only the genes for the toxin colicin E2 and the immunity protein, and a longer transcript (long mRNA, marked green in Fig 1), which additionally includes the lysis gene [14, 28, 29–32]. Hence, lysis can only be initiated after the translation of long mRNA [18], and this crucial operation is regulated at the post-transcriptional level, as described below.
Post-transcriptional regulation makes use of many different mechanisms. Recent studies emphasize the particular importance of non-coding sRNAs [33] for various processes in E. coli, especially because of their ability to introduce delays and set up thresholds for translation [34–37]. This is done either directly, by sRNAs pairing with their target mRNA (sRNA-mRNA interaction), or indirectly, by sequestering of specific mRNA-binding proteins (mRNA-protein interaction) [2, 38, 39]. For the latter form of regulation, recent studies highlighted the importance of the production rates of regulatory components [40]. In the case of the ColE2 system, the translation of the long mRNA is regulated by the carbon storage regulator protein CsrA [28], marked red in Fig 1. CsrA dimers destabilize target mRNAs by binding to a region that includes the ribosome-binding site (Shine-Dalgarno sequence) [41]. Masking of the ribosome-binding site by CsrA thus not only represses translation of the lysis gene but also promotes degradation of the long mRNA. However, CsrA is also recognized by two specific sRNAs, CsrB and CsrC [42], marked blue in Fig 1. These sRNAs can therefore sequester CsrA dimers, preventing them from binding to target mRNAs [43–45]. Thus, translation of the ColE2 lysis gene is indirectly regulated by sequestration of CsrA. This process, also known as “molecular titration”, exhibits ultrasensitive thresholds and has been extensively studied [46, 47].
The basic interaction network that controls the ColE2 regulatory network has been studied in great detail in previous works [48–51], and many of its functional characteristics, in particular the threshold behavior, were described for a wide range of both bacterial and eukaryotic systems [52]. However, a detailed theoretical description of the dynamics leading to the release of colicin is still missing, in particular the role of the hierarchically ordered regulation involving CsrB and CsrC. In this work, we have formulated this post-transcriptional network in a detailed mathematical model, constructed by analogy to studies of simpler sRNA-regulated systems (for example, [33, 34, 36]). We then simplified the model by assuming fast complex equilibration, and combining the sRNAs CsrB and CsrC into a single, effective sRNA (see S1 Text for details). This reduced the regulation network to three relevant components: free long mRNA, free CsrA and the effective sRNA (see Fig 2). We then analyzed this simplified network in detail. In contrast to previous work [36], we give a general analytical solution for the three component system, and derive a precise approximation for fast and clear analysis. This analytic solution exhibits a pronounced threshold in mRNA production due to CsrA-dependent regulation, which was also confirmed using numeric simulations. We investigated, how this threshold depends on system parameters, and how it affects the actual biological system. Furthermore, we have analyzed the role of fluctuations in the post-transcriptional regulation network and how fluctuations in long mRNA expression may be dampened by sRNA. Finally, we extended our model by including the transcriptional regulation, and analyzed how the system behaves during a realistic SOS response. Previous studies have shown discrete activation peaks in LexA-repressed promoters [26] that can lead to large fluctuations close to the threshold of mRNA expression [9]. In a stochastic simulation of the complete model, we were able to reproduce this phenomenon. Comparison with experimental data on lysis time distributions [48] also shows that our model can explain the delayed and broadly distributed release times of colicin complexes. This underlines the importance of stochasticity for the heterogeneous expression of colicin E2 in E. coli populations.
For our theoretical analysis, we initially developed a detailed mathematical model for the post-transcriptional regulation of colicin E2 release. To this end, we derived a set of coupled, deterministic rate equations from the interaction network depicted in Fig 1, with the corresponding rates for transcription, degradation, binding interactions etc. as parameters. In the following, we briefly review how we reduced the network to its core components, which comprise the theoretical model. The interaction scheme underlying the complete model is presented in S1 Fig and further explanations can be found in the Supporting Information, where we also detail how our model can account for sequestration by other targets of the global regulator CsrA.
As we wished to study the post-transcriptional regulation of colicin E2 expression, we included in the model only those components that are relevant at that stage. The model therefore omits the short mRNA and its products. However, the rate of transcription of the long mRNA is a crucial parameter, which is influenced by the kinetics of activation of the SOS promoter, and thus by the processing of its repressor LexA. Upon DNA damage, RecA promotes auto-cleavage of LexA dimers, thus removing inhibition of the SOS response (marked in red in Fig 1). The LexA-RecA interaction network has recently been modeled stochastically [53]. Before including this detailed network in our final model, we focused on understanding the post-transcriptional dynamics. To this end, we initially assumed that activation of the SOS promoter occurs rapidly relative to the rates of production and degradation of the long mRNA [54], which allowed us to approximate the transcription rate of long mRNA by an effective rate αM (Materials and Methods). With respect to post-transcriptionally relevant components, we were then left with long mRNA, CsrA, and the two sRNAs CsrB and CsrC, and the mRNA-CsrA-, CsrA-CsrB-, and CsrA-CsrC-complexes.
CsrB and CsrC regulate CsrA by forming complexes with it. The two sRNAs each have several (on average: N) CsrA binding sites, and if every occupation state of the sRNAs were to be modeled as a separate component, a large number of coupled rate equations would need to be added to the model. However, due to the fast dynamics of the CsrA-CsrB- and CsrA-CsrC-complexes, and their virtually identical biochemical behavior, we were able to reduce the sRNA interaction to a single equation for an effective sRNA, with only one binding site and transcription rate NαS (see Materials and Methods). As a result, the mechanisms of complex formation, dissociation and degradation are replaced by an effective coupled degradation of complex partners. Despite the different processes that are integrated to effective ones, the effective sRNA still resembles the dynamical behavior of CsrB/CsrC. A detailed derivation of the simplified system of rate equations can be found in S1 Text. The final post-transcriptional model is thus reduced to a set of three coupled, deterministic rate equations that capture the behavior of the free long mRNA (M), free CsrA dimers (A), and an effective free sRNA (S) component with a single CsrA binding site:
M ˙ = α M - δ M M - k M M A , (1) A ˙ = α A - δ A A - k M p M M A - k S p S A S , (2) S ˙ = N α S - δ S S - A k S S , (3)
where (1 − pM) and (1 − pS) are the probabilities for CsrA to survive the coupled degradation. A graphical illustration of this differential equation system is depicted in Fig 2. Note that in the model the quantities M, A and S represent the abundance of the corresponding free components. Once a long mRNA, sRNA, or CsrA dimer binds to some other component, it loses its function and is thus removed from the model system.
For the analysis of our model, we had to determine production, degradation and binding rates. The particular values used are listed in S1 Table. As far as possible, we chose values that are measured in studies on either the same or comparable systems (see S1 Text for details). In the other cases, we tried to derive plausible parameters from known factors that influence the particular rate. A detailed motivation and derivation of these rates is given in chapter 2 of S1 Text.
We analyzed the reduced post-transcriptional model by first calculating its steady state. In order to obtain a cleaner and simpler result, we derived an approximation (see Materials and Methods) for the steady state solution, which agreed very well with the results of numerical simulations (see S2 Fig). Using these simplified equations, we then investigated the impact of the rates of production of long mRNA (αM) and sRNA (αS) on the levels of the three components. The results (see Fig 3) reveal a linear threshold that appears at the same position for all three components. The threshold divides the parameter space into two regimes, in which either CsrA or long mRNA and sRNA have a non-zero abundance. This is due to the coupling between the degradation of CsrA and the abundance of both long mRNA and sRNA, such that the presence of CsrA dimers excludes that of long mRNA and sRNA, and vice versa. This mechanism in turn controls the release of colicin-immunity complexes, since a sufficiency of CsrA dimers ensures reliable repression of the long mRNA and prevents synthesis of the lysis protein.
From the aforementioned analytic solution we calculated the threshold position as a function of the system parameters (S1 Text). We found that the threshold for non-zero levels of long mRNA lies exactly at the point where the production rate of CsrA αA is equal to the sum of transcription rates for long mRNA αM and sRNA αS (S1 Text). Thus, we observed no expression of long mRNA in the regime αM + αS < αA, as shown in Figs 3A and 4. We find the threshold to be sharp, and attribute this to the very slow degradation of CsrA compared to long mRNA and sRNA [55, 56].
Apart from the threshold itself, we find that the levels of free CsrA and free sRNA predicted by our steady state analysis are consistent with experimental in-vivo values determined by previous studies [43, 57]. Moreover, our results are also consistent with the total amount of CsrA as well as its ratio to sRNA (S1 Text).
So far, we have demonstrated that our three-component system is capable of producing a threshold behavior. However, it has been shown previously that a mutually exclusive production of sRNA and a target mRNA is possible with just two components [36]. The question thus arises why a third component is needed at all. One possible explanation is that the sRNA makes it easier to trigger lysis, as an increase in sRNA production induces an increase in the abundance of long mRNA (Fig 3).
After SOS signals, the sRNA controls and accelerates the degradation of CsrA (see section on expression dynamics below), eventually leading to the expression of the lysis protein.
In a next step, we analyzed the stochastic dynamics of the post-transcriptional regulation network. To this end, we switched to a stochastic description, calculated the Fano factor (VarM/〈M〉) for the abundance of long mRNA (see Materials and Methods), and depicted it as heatmap in Fig 4. The Fano factor measures the relative magnitude of fluctuations, and has already been applied to gene regulatory networks in previous studies [58]. It can also be understood as a quantified comparison with the pure birth process (Poisson process), which has the Fano factor F = 1.
We found that fluctuations in mRNA were most pronounced close to the threshold position, with the largest fluctuations occurring slightly above the threshold (Fig 4). Moreover, Fig 4 also shows that the fluctuations became larger as sRNA production decreases. Thus, the third component (sRNA) in the post-transcriptional regulation network also enables significant dampening of fluctuations in long mRNA.
To understand why the fluctuations are localized to the region near threshold, one must take the characteristics of this parameter regime into account. Around the threshold, molecule numbers are close to zero, which has a direct affect on the relative size of fluctuations: the lower the abundance, the larger the fluctuations (stochastic regime). Moreover, the threshold is the only regime in which all three components, CsrA, mRNA and sRNA, can coexist and interact with each other: An increase in the level of CsrA will lead to a decrease in the abundance of long mRNA and sRNA, owing to increased complex formation and subsequent degradation. Analogously, an increase in long mRNA and sRNA molecule numbers leads to a decrease in CsrA abundance. Therefore, the abundance of CsrA dimers is anti-correlated with the abundance of both long mRNA and sRNA. It has been shown for a two-component system, that anti-correlated components can create anomalously large fluctuations [59] if degradation rates are small compared to turnover (ratio of production rate to abundance). For long mRNA, this is exactly the case close to threshold, where the long mRNA abundance is still very low.
These results show that a third component can reduce intrinsic fluctuations of a hierarchically ordered regulatory network.
To study the dynamical response of the ColE2 system to an SOS signal, we extended the post-transcriptional network by including the LexA-RecA regulatory network [53] (Fig 1). LexA not only represses the SOS promoter, it is also an auto-repressor, as well as being a repressor of RecA production. As outlined in the Introduction, RecA forms filaments after DNA damage, which then induce auto-cleavage of LexA dimers. Consequently, the levels of RecA, LexA and the colicin mRNAs increase, as repression due to LexA is relaxed. A stochastic model of this network has been introduced recently [53]. In that study, promoter activity in the LexA-RecA system was found to occur in ordered bursts that result from fluctuations and the particular structure of the RecA-LexA feedback loop.
In our analysis of the ColE2 post-transcriptional regulation network so far (see above), we have assumed the dynamics of SOS promoter activation to be so fast that we could use an effective transcription rate αM for long mRNA. To link the LexA regulatory network to the post-transcriptional regulation network, we must drop this assumption and explicitly model the dynamics of LexA dimers, which connect the two networks. In the biological system, this involves the binding and dissociation of LexA dimers to and from the SOS promoter in the ColE2 operon. Long mRNA and short mRNA are transcribed only from the derepressed promoter at rates αMl and αMs, respectively. Thus, the transcription rates of long mRNA and short mRNA are proportional to the number of open SOS promoters in the bacterium. The majority of transcripts are short mRNAs. The mathematical implementation of the integrated regulation network is again a system of coupled rate equations, which we describe in S1 Text. The additional parameters of the LexA-RecA regulation network are to be found in S2 Table.
We simulated the SOS signal by temporarily up-regulating the coupling parameter cp, which quantifies the ability of RecA to induce cleavage of LexA (Fig 1). In the uninduced state before and after the SOS signal, the auto-cleavage parameter was set to cp = 0. Under SOS stress cp was increased to cp = 6. This increase in cp subsequently boosts the long mRNA production, and therefore relates to a transition from a sub-threshold state (gray area below the white line in Fig 3A) to a super-threshold state (green area above the white line in Fig 3A). Due to the stochasticity in the LexA-RecA network and the resulting stochastic promoter dynamics, the overall transcription rate αMl of long mRNA is not constant, but fluctuates about a mean value. The production rate of sRNA was held constant at αS = 57.5. Fig 5 shows the dynamics of short and long mRNA levels and the abundance of CsrA dimers and sRNA in response to transient SOS signaling. When we compared a stochastic realization using Gillespie simulations (Materials and Methods) with a numerical solution of the deterministic rate-equation system, we observed significant qualitative and quantitative differences. First, the stochastic realization exhibited significant fluctuations that manifested themselves in abrupt, short-lived changes in the abundance of short mRNA over the whole time-course (Fig 5A). Second, the average over 500 stochastic realizations deviated from the deterministically predicted value. Both phenomena arise from the intrinsic stochasticity of the LexA-RecA-regulatory network, as explained by Shimoni [53]. Fluctuations may lead to a spontaneous dip in the number of LexA dimers which releases all LexA-regulated genes, including the lexA gene itself, from repression. This consequently leads to a sudden rise in the abundance of short mRNA. The open lexA and recA promoters will then generate a burst of newly produced LexA and RecA proteins, which block and regulate the promoters for the next burst.
Focusing on the dynamics of mRNA transcription, we found that, due to initial simulation parameters, only small numbers of the short mRNA are produced in the uninduced state. After up-regulation of the LexA auto-cleavage parameter cp at t = 200 min, the abundance of short mRNA rises and the aforementioned large bursts appear. The amount of long mRNA, however, follows a completely different trajectory, conditioned by post-transcriptional regulation. Before the SOS signal, expression of long mRNA is almost completely repressed by CsrA (Fig 5B). Even the bursts of SOS promoter activity reflected in fluctuating amounts of the short mRNA have little or no impact on the long mRNA. This filtering effect is biologically relevant, as it ensures that noisy promoter activity does not erroneously trigger lysis. After induction of the SOS signal, the deterministic dynamics of the underlying rate equations predicted that, after a delay of about 40 min, the abundance of long mRNA should rapidly rise to a saturation value (black dashed line in Fig 5B). However, a mean of 500 realizations deviated significantly from this prediction (Fig 5B). In particular, the average number of long mRNA molecules increased more slowly than predicted by deterministic dynamics. Hence the abundance saturated at a much lower value. An appreciable delay between SOS signal induction and expression of long mRNA was still observed, but lasted for only 15 min.
Studying the dynamics of a single stochastic realization, we observed that the number of long mRNA molecules underwent large fluctuations, which were followed by periods of no expression at all. Moreover, the timing of these bursts varied considerably between different realizations. This constitutes a significant qualitative difference compared to the average over 500 realizations and to the deterministic dynamics (Fig 5), both of which exhibit a smooth and continuous temporal behavior. Fig 5B and 5C indicates the origin of this behavior: The abundance of long mRNA can only grow if the number of free CsrA dimers is low. The same holds for the abundance of sRNA, which supports the degradation of CsrA and also can only reach non-zero abundances if there is no CsrA left (Fig 5D). Thus, before any long mRNA can be expressed, the free CsrA concentration must drop to very low values due to degradation or complex formation. The delay between SOS signal induction and the first burst of long mRNA synthesis therefore depends on the amount of CsrA available. We went on to study the precise timing of the first burst in long mRNA abundance, since it is crucial for the time-point of release of colicin-immunity complexes. To this end, we calculated the probability distribution for the first peak from an ensemble of 500 stochastic realizations. The probability of a peak in long mRNA abundance rose quickly and reached its maximum approximately 60 min after induction of the SOS signal (Fig 6A). This phenomenon is also seen in experimental systems: time-lapse studies with colicin-producing bacteria revealed that their lysis time is broadly distributed [48]. The distribution depicted in Fig 6A matches qualitatively with comparable datasets from these experiments. Moreover, our model is able to numerically predict average lysis times in dependence on different SOS signal strengths (see S5 Fig). From the probability distribution of the timing of the initial peak in long mRNA abundance we calculated the survival function, i.e. the probability with respect to time that a cell will not release toxin. Here we assumed that this first burst provides enough long mRNA in the cell to produce the lysis protein, which then induces its lysis with concomitant release of colicin-immunity complexes into the surrounding medium. The function of lysed cells plotted in Fig 6B shows that the number of cells that release the toxin rises with the duration of the SOS signal.
Incorporation of the LexA-RecA regulatory network allowed us to model the colicin E2 expression dynamics in response to a realistic SOS signal, and the results presented above highlight the importance of CsrA for colicin release.
Gene expression is a process that allows for various forms of regulation at all levels. In theoretical studies of post-transcriptional regulation of several biological systems, modulation of mRNA production by proteins or sRNA has been shown to create, for instance, temporal thresholds for mRNA translation [9, 35, 36]. Proteins have also been shown to regulate the expression of the toxin colicin E2 [28] in the context of an SOS response to environmental stress. Experimental studies have elucidated the detailed interaction network responsible for the production and release of the colicin [28]. However, the dynamics of this system, in particular at the post-transcriptional level, have remained elusive. In close analogy to previous two-component models, we developed a mathematical model for this hierarchically ordered post-transcriptional regulation of colicin E2 release. Interestingly, the known interaction network for this system necessitated the modeling of three, not two, components: the long mRNA which is necessary for colicin release, its negative regulator CsrA, and sRNA, which in turn negatively regulates CsrA. Contrary to previous studies [9, 35, 36, 60], the sRNAs do not regulate the mRNA directly, but control the level of the regulator protein CsrA. Thus, the sRNA acts as the “regulator’s regulator”.
In our analysis of the model, we used rate constants that were determined from experimental systems (see chapter 2 of S1 Text for details). Comparing the predicted CsrA levels before the SOS signal (see Fig 5C) with in-vivo measurements of E. coli [57] shows that our model results in a pre-SOS free CsrA abundance that agrees with actual bacterial systems (for other abundances, see S1 Text). Moreover, the model is not just able to predict steady state abundances, but also reproduces the reaction to varying external stress levels as seen in experiments (see S5 Fig).
Investigation of the dynamics revealed that the model exhibits a time delay in the production of free long mRNAs. This delay is due to the high abundance of CsrA in the non-SOS state of the cell, which causes CsrA to quickly bind to free long mRNA and thus prevents its transcription. Only during an SOS signal, which indicates external stress for the cell, the level of CsrA gets steadily reduced. The time this process takes to get CsrA levels so low that fluctuations in long mRNA production result in free long mRNA, causes a delay in colicin release. As colicin release is coupled to cell lysis, the delay is therefore a mechanism for filtering out transient SOS signals that might erroneously lead to synthesis of the lysis protein. Moreover, also intrinsic fluctuations, for instance in sRNA production, are filtered out by this mechanism: Even if a large and sudden burst in sRNA were strong enough to drop CsrA abundance close to zero, the CsrA buffer gets restored quickly due to the large production rate of CsrA. This rate is only effectively lowered during a SOS signal, which increases the production of the CsrA-sequestering long mRNA. The fact that lysis is regulated by a threshold mechanism of a global regulator protein like CsrA might also be a guarding mechanism for the cell: only prolongued extreme situations will cause the abundance of these regulators to drop to low molecule numbers.
However, delays and similar threshold behavior also emerge in two-component systems, raising the question why a third component is necessary here. Strikingly, we found that the third component (sRNA) in the post-transcriptional interaction network enables the cell to tune the duration of the delay by sequestering CsrA. In the case of the ColE2 system, this means that cells are able to adjust the (average) time between a SOS signal and the onset of cell lysis leading to colicin release.
Furthermore, previous studies of systems with slow, bursting promoter kinetics have also uncovered a major limitation of two-component sRNA-based regulation compared to regulation based on transcription factors: Two-component systems are subject to significantly higher levels of intrinsic noise [9]. However, Fig 4 (panels A,C,D) shows that, in the post-transcriptional regulation of colicin E2 release, fluctuations become smaller at higher values of αS. The sRNA might therefore allow for significant dampening of these fluctuations. This idea is supported by the fact that the relatively high degradation rate of sRNA makes it less susceptible to induced fluctuations.
In bacteria, these mechanisms could have several functions: First, a comparison of different sRNA production rates (S4 Fig) indicates that the sequestration of CsrA by the sRNA could indeed be crucial for fast release of the colicin, as CsrA degradation rates cannot be arbitrarily increased in bacterial systems. Second, they can tune the reaction to external stress at the population level. Experimental studies have shown that, in the absence of stress, 3% of colicin producing cells release the toxin during the stationary phase; but this fraction can be increased up to eventually 100% if an external SOS stress is applied [14, 48]. Previous experimental studies also found that colicin systems exhibit heterogenous expression times, which originate from the stochasticity of the SOS signal [49, 50]. Recent time-lapse experiments with colicin E2 producing bacteria showed that this lysis time distribution also depends on the strength of the SOS signal [48]. We reproduced these experiments with stochastic simulations, in which we created different stress levels by different values of the RecA degradation rate parameter cp. Our predictions for lysis time distributions (Fig 6A and S5 Fig) show qualitative agreement with these time-lapse experiments. Moreover, the ability of the sRNA to tune the average duration of the delay might serve as a mechanism to adjust the cell lysis to different stress levels. Altering the sRNA level could be an additional mechanism, apart from the stochastic SOS signal, by which bacterial populations can adjust the fraction of cells releasing the toxin depending on the strength and duration of the external stress. Finally, the co-option of sRNA makes the cells less susceptible to lysis due to adventitious fluctuations in promoter activity. This is particularly important considering the bursting behavior and large-scale fluctuations seen in the LexA-RecA-regulatory system, which are readily observed in experiments and reproduced by stochastic models [53].
In order to focus on the interplay between the LexA-RecA system and the hierarchical regulation of long mRNA by CsrA and sRNA, we kept the plasmid number constant. If we considered random, Poisson-distributed plasmid numbers instead, the effect would be very small, as shown in S4B Fig. This fact demonstrates that the colicin plasmid copy number only has minor influence on the lysis time distribution (see S1 Text for details).
In conclusion, we have provided here the first detailed theoretical description of colicin E2 production and release, and used it to study the dynamical behavior of this system. Moreover, the general three-component model described here should be applicable to many other systems of toxin production in microorganisms.
In most models of prokaryotic gene expression, it is assumed that promoter kinetics are fast compared to RNA production and degradation rates. In that case, the promoter state is well approximated by its steady state [54]. In the analysis of the post-transcriptional regulation network, the promoter status affects the transcription rate of the (long) mRNA. Thus, we replaced it by an effective transcription rate for (long) mRNA, which takes into account the probability of a gene being blocked. In the literature this procedure is referred to as “adiabatic elimination of fast variables” (see for example [61]). For this effective rate we also took into account that the colicin operon is located on a plasmid [62], of which approximately 20 copies exist in each cell [14] (see S1 Text).
The two sRNAs CsrB and CsrC regulate CsrA via complex formation. More specifically, each CsrB molecule has approximately 22 binding sites for CsrA, with 9 CsrA dimers being attached on average [63, 64]. CsrC interacts in the same way, but has fewer CsrA binding sites [63]. As a first step, we therefore replaced the two sRNA types by a single effective one, which has N binding sites. However, all of the N + 1 sRNA configurations still enter the interaction network as separate components, since the binding and dissociation probabilities change with the number of free binding sites. By investigating the dynamics of the CsrA-sRNA complexes, we discovered that the probability distribution for occupied CsrA binding sites on the sRNAs reaches its stationary state on a time scale that is proportional to the rate of complex-(un)binding. Since binding and unbinding events are biochemically much simpler processes than transcription, translation or degradation, it is very likely that the dynamics of CsrA-sRNA complexes is much faster than all other reaction rates in the system. Following the line of Levine [36] and Legewie [34], we therefore assumed rapid complex dynamics, and replaced the different binding site occupations by an effective sRNA, with only one binding site and transcription rate NαS (see S1 Text for details on the calculation).
For the calculations of the abundances of the three components (for example, to obtain the plots of Fig 3), we began by assuming the stationary state. Solving for the abundance of one component then gives a cubic equation, for which the exact, general solution is very lengthy and cumbersome to analyze. Therefore, we considered the cubic equation for the cases of very large and very small molecule numbers, and ignored terms that became negligible. This resulted in two easily solvable quadratic equations. Comparisons with numerical solutions of the cubic equations proved that the quadratic solutions approximate the general solution well in their respective abundance regime. Equating the terms omited in the approximation yields a criterion for the transition between the two approximations (see S1 Text). The transition is depicted as a white line in Fig 3. That this transition lies close to the threshold is coincidental. Comparison with the exact, numerical solutions showed that the threshold is not an approximation artifact. S2 Fig illustrates the precision of the approximation by comparing its prediction for long mRNA abundance to that from numerical simulations.
We started the analysis of noise properties by reformulating the simplified three-component system as a Master equation. As Master equations are typically impossible to solve analytically, we performed a general van Kampen expansion in multiple variables (components). Our analysis included all higher orders, and not only lowest order terms as is commonly found in textbooks [61, 65]. With van Kampen’s expansion we were able to derive general formulas for the first up to the fourth moment of the random variable representing the fluctuations of the system around the stationary solution of the rate equations. The terms of each equation were classified in first order terms (dominant terms) and higher order terms (second order, third order, etc), according to the scaling behavior of each term with the system size. We used different methods to calculate the Fano factor for long mRNA. The most reliable results were obtained by implementing only first order terms in the calculations of second moments. This reproduced the shape of the Fano factor well, but it overestimates fluctuations in the vicinity of the threshold. S3 Fig illustrates the degree of agreement between analytical calculations of the Fano factor agree with the results from Gillespie simulations.
To verify how well our analytical results of the deterministic rate equations coincide with the actual mean molecule numbers, we set up a Gillespie simulation [66]. The Gillespie algorithm generates a statistically correct realization of the master equation behind the rate equations. The core of the algorithm lies in using random numbers to determine which next reaction will occur and the waiting time prior to the succeeding reaction. The reactions simulated by the Gillespie approach are listed in S1 Text. To quantify the delay between SOS signal induction and the first burst in long mRNA abundance, we defined the beginning of the first peak as the point when the number of long mRNA molecules exceeds 8 for the first time. The time of the peak itself was set to the point at which that number reached a maximum. We then calculated the probability distribution from an ensemble of 500 stochastic realizations, using the parameters defined in S1 and S2 Tables.
|
10.1371/journal.pbio.0050047 | LACHESIS Restricts Gametic Cell Fate in the Female Gametophyte of Arabidopsis | In flowering plants, the egg and sperm cells form within haploid gametophytes. The female gametophyte of Arabidopsis consists of two gametic cells, the egg cell and the central cell, which are flanked by five accessory cells. Both gametic and accessory cells are vital for fertilization; however, the mechanisms that underlie the formation of accessory versus gametic cell fate are unknown. In a screen for regulators of egg cell fate, we isolated the lachesis (lis) mutant which forms supernumerary egg cells. In lis mutants, accessory cells differentiate gametic cell fate, indicating that LIS is involved in a mechanism that prevents accessory cells from adopting gametic cell fate. The temporal and spatial pattern of LIS expression suggests that this mechanism is generated in gametic cells. LIS is homologous to the yeast splicing factor PRP4, indicating that components of the splice apparatus participate in cell fate decisions.
| The selection and specification of the egg cell determine the number of eggs produced by an animal or plant, which in turn dictates how many offspring that organism can produce. In most higher plants, the egg cell forms in a specialized structure consisting of four different cell types. Two cells, the egg cell and the central cell, are fertilized by sperm cells and develop into the embryo proper and the nutritive tissue (endosperm), respectively. These two gametic cells are flanked by accessory cells; but why do some cells become gametic while others differentiate into accessory cells? To answer this question, we looked for mutants in which this process is disturbed. In the lachesis mutant, accessory cells become extra egg cells. Interestingly, it seems that the misspecification of these accessory cells results from defects in the gametic cells. This suggests that accessory cells monitor the state of the gametic cells to act as a backup if required, ensuring the formation of the key reproductive cells.
| The formation of gametes is a key step in the lifecycle of any sexually reproducing organism. In flowering plants, the egg and sperm cells develop within haploid gametophytes (Figure 1). The female gametophyte of Arabidopsis originates from a single haploid spore (Figure 1A) through three nuclear division cycles. The resulting syncytium of eight nuclei (Figure 1B) cellularizes and differentiates four distinct cell fates [1–4] (Figure 1C and 1D). Both the egg and central cells are fertilized by one sperm cell each to form the embryo and the surrounding endosperm, respectively. These gametic cells are flanked by accessory cells. Two synergids lie at the micropylar pole, the entry point of the pollen tube. Synergids are necessary for the attraction of the pollen tube and induce the subsequent release of the sperm cells [5–7]. The opposite pole is occupied by three antipodal cells that degenerate prior to fertilization and whose function is unclear [3].
Although collections of female gametophytic mutants have been reported [8,9], mechanisms that underlie the specification of gametic versus accessory cell fate are unknown. In the present study, we took advantage of an egg cell–specific marker that we isolated in a screen for enhancer detector (ET) and gene trap (GT) lines, to examine the regulation of gametic cell fate. Our results indicate that a combinatorial mechanism operates to ensure maximum likelihood that the key reproductive gametic cells are formed, while at the same time deleterious excess gametic cell formation is prevented.
We performed a screen for ethyl methanesulfonate (EMS)-induced mutants that alter the expression of the enhancer detector line ET1119, which in wild type confers specific β-glucuronidase (GUS) expression to the egg cell (Figure 2A).
In lachesis-1 (lis-1) mutants, expression of the egg cell marker was expanded to the synergids and the central cell, suggesting that the restriction of egg cell fate to a single cell is compromised (Figure 2B and 2C). lis-1 is a loss-of-function mutation for which no homozygous plants were recovered (see below). Therefore, all analyses were performed on heterozygous plants in which only 50% of the ovules contain lis-1 mutant female gametophytes. Heterozygous lis-1/LIS plants produced fertilized seeds and aborted ovules at a 1:1 ratio (50.7%:49.3% in lis-1/LIS, n = 631; 96.6%:3.4% in wild type, n = 655; Figure 1E and 1F), consistent with a female gametophytic defect [10]. Reciprocal crosses with wild-type plants confirmed that lis-1 was rarely transmitted maternally (transmission efficiency through the female [TEF] = 8.6%, n = 347). Paternal transmission was also affected, but less severely (transmission efficiency through the male [TEM] = 59.4%, n = 367).
To determine whether lis-1 female gametophytes are indeed defective in cell specification, we examined morphological, molecular, and functional characteristics of the different cell types in lis-1 gametophytes (Figures 1, 2, and 3, respectively). Until cellularization, gametophytes were indistinguishable between lis-1/LIS and wild-type plants (unpublished data), indicating that LIS is not required for any previous step, including mitotic divisions, migration of nuclei, or cellularization. The first defects in lis-1 female gametophytes were consistently observed only after cellularization, corresponding to the wild-type stage at which the different cell types establish distinct morphological and molecular characteristics, as described below. Wild-type synergids differ morphologically from egg cells by two features. First, the synergid nuclei are smaller than the egg cell nucleus. Second, the polarity of synergids is reversed with respect to nuclear position [11,12] (Figure 1G). In lis-1/LIS plants, however, synergids differentiated the morphological attributes of egg cell fate and were often indistinguishable from egg cells (Figure 1J; Table 1). Additionally, the expression of the synergid marker ET2634 was down-regulated in many lis-1 gametophytes (Figure 2D–2F). To test whether the pollen tube–attracting activity of synergids was affected as well, we pollinated lis-1/LIS and wild-type plants with a marker line expressing GUS in the pollen tube (ET434G) (Figure 3A). The number of ovules without GUS staining was strongly increased in lis-1/LIS plants as compared to wild type (Figure 3A–3C), implying that pollen tube attraction was compromised in the majority of lis-1 mutant female gametophytes. These data, together with the ectopic expression of the egg cell marker in the synergids, indicate that lis-1 mutant synergids differentiate egg cell attributes at the expense of synergid cell fate. A different synergid marker (ET884) was ectopically expressed in lis-1 gametophytes (Figure 2G–2I). This intriguing expression shows that not all aspects of accessory and gametic cell fate are mutually exclusive. However, the reduced pollen tube attraction in lis-1 indicates that this marker is unlikely to reflect fully differentiated synergid cell fate.
The central cell differs from the egg cell by both its size and the presence of two polar nuclei, which fuse prior to fertilization [1–4] (Figure 1H). In lis-1 mutant gametophytes, the polar nuclei rarely fused and often cellularized separately, a process never observed in wild type (Figure 1K and 1M; Table 1). The resulting uninucleate cells were morphologically indistinguishable from an egg cell. These morphological changes were reflected by the down-regulation of a central cell marker pMEA::GUS in most lis-1 gametophytes (Figure 2J–2L). Whereas the wild-type central cell develops into endosperm after fertilization, approximately one half of the lis-1 mutant gametophytes that received a pollen tube (compare Figure 3C) failed to develop endosperm (Figure 3D–3F) although embryo formation was not affected. The failure to develop endosperm is unlikely to be related to an unfertilized central cell, because embryo formation can initiate autonomous endosperm formation in an unfertilized central cell [13]. Together with the ectopic expression of the egg cell marker, our results indicate that in lis-1 gametophytes, the central cell differentiates egg cell attributes at the expense of central cell fate.
A further striking phenotype was observed for the antipodal cells that degenerate prior to fertilization in wild type (Figure 1I). In lis-1 female gametophytes, the antipodal cells were often enlarged and protruded into the center (Figure 1L and 1M; Table 1). Additionally, in about one third of lis-1 gametophytes (16.9% in lis-1/LIS), the enlarged antipodal cells eventually disintegrated their cell membranes, allowing the fusion of antipodal nuclei into one large nucleus (Figure 1N). In wild-type gametophytes, fusion of nuclei was only detected in central cells (Table 1). Consistently, we observed ectopic expression of the central cell marker pMEA::GUS in the protruding antipodal cells (Figure 2M–2O), whereas the expression of GT3733, an antipodal marker line, was down-regulated (Figure 2P–2R). Thus, antipodal cells in lis-1 mutant female gametophytes can adopt a central cell fate. Remarkably, both the central cells and antipodal cells changed not only their molecular profile according to the newly adopted cell fate, but they also adjusted their nuclear status (uninucleate versus binucleate) accordingly. The originally binucleate central cell cellularized ectopically, resulting in two uninucleate cells, whereas the uninucleate antipodal cells fused, producing a binucleate cell. These findings suggest that an intracellular signaling mechanism senses the number of nuclei in a given cell, and reveal a tremendous, previously unrecognized plasticity of the female gametophyte. In summary, whereas wild-type gametophytes differentiate accessory and gametic cell types, accessory cells of lis-1 mutant gametophytes frequently adopted gametic cell fate. These observations suggest that all cells in the female gametophyte are competent to differentiate gametic cell fate and that LIS is involved in a mechanism that represses gametic cell fate in the accessory cells. Interestingly, the gametic central cell in lis-1 gametophytes additionally adopted egg cell fate, suggesting that a further, LIS-dependent mechanism suppresses egg cell fate in the central cell. Thus, the lis-1 mutant phenotype reveals two levels of cell fate regulation, one between gametic and accessory cells, and one between egg and central cell.
The late initiation of cell-specific marker genes indicates that in Arabidopsis distinct cell fates are only manifested after cellularization. Studies in several multicellular systems have shown that cell specification is often preceded by the asymmetric distribution of fate determinants (for review see [14]), and an analogous mechanism could be defective in lis-1 gametophytes, resulting in an instant misspecification of accessory cells. Alternatively, the lis-1 phenotype could result from defects that occur after cellularization when distinct cell fates become manifest. We analyzed the time course of egg cell and central cell marker gene expression. Interestingly, we found that the number of ovules that ectopically expressed gametic cell fate increased over time (Figure 4A and 4B), indicating that accessory cells are not instantly misspecified as gametic cells. In line with a successive misspecification of accessory cells, we found that several morphological features that distinguish lis-1/LIS plants from wild type became more pronounced over time (Figure 4C–4G). Our results suggest that during cell specification in lis-1 mutant gametophytes, accessory cells become gradually recruited as gametic cells.
We mapped the lis-1 mutation to the At2g41500 locus, which encodes a protein with seven WD40 repeats. The lis-1 mutation creates an in-frame stop codon after three WD40 repeats (Figure 5A). The LIS cDNA driven by a 2.6-kilobase (kb) upstream promoter sequence complemented the lis mutant phenotype (Figure 5B and 5C), indicating that lis-1 is a loss-of-function mutation.
The LIS protein is strongly conserved among eukaryotes [15] (Figure S1), showing an overall similarity to Homo sapiens, Caenorhabditis elegans, and Saccharomyces cerevisiae of 62%, 56%, and 54%, respectively. The yeast homolog PRP4 is associated with the U4/U6 complex of the spliceosome [16,17]. PRP4 is an essential splicing factor, and loss-of-function mutants accumulate unspliced pre-mRNA [18]. PRP4 function depends on its interaction with a second splicing factor, PRP3, through its WD40 domain, and the deletion of two WD40 repeats abolishes this interaction [19]. We thus conclude that lis-1 represents the null phenotype, which is consistent with the observation that the lis-2 T-DNA insertion allele (Figure 5A) causes a very similar phenotype (Figure S2).
The Arabidopsis genome contains a second LIS-related sequence, At2g05720. The deduced protein (accession no. AAD25639; Figure S1) shares an overall similarity of 70%, but is only half the size of the LIS protein and notably contains only four complete WD40 repeats (Figure S1). Hence, At2g05720 is unlikely to be functionally redundant to LIS.
To determine the temporal and spatial expression pattern of the LIS gene, we performed RT-PCR and analyzed the expression of a pLIS::NLS_GUS construct containing the same LIS promoter fragment as the LIS cDNA rescue construct that had fully complemented the mutant phenotype (Figure 5B and 5C). LIS expression was detected at moderate levels in all tissues examined, with strongest expression in reproductive tissues (Figure 6A). GUS expression driven by the LIS promoter was detected at all stages of female gametophyte development (Figure 6B–6E). Intriguingly, shortly after cellularization, expression in the accessory cells is down-regulated, whereas expression in the gametic cells is strongly up-regulated (Figure 6E). This suggests that the mechanism, which prevents accessory cells from adopting gametic cell fate, is not cell-autonomous, but is generated in gametic cells, which is consistent with a lateral inhibition model (Figure 6F). We propose that the gametic cells upon differentiation generate a LIS-dependent signaling molecule that is transmitted to the adjacent accessory cells to inhibit their gametic cell competence, thereby preventing excess gametic cell formation. In this view, the lis mutant phenotype is a result of impaired lateral inhibition, i.e., the gametic cells fail to identify themselves to their neighboring cells, resulting in the recruitment of all gametophytic cells as gametic cells. The observation that not all cells are synchronously specified as gametic cells implies some initial bias and suggests that the proposed lateral inhibition operates to maintain rather than to establish different cell fates.
Although this mechanism can perpetuate a binary decision between gametic and accessory cell fate, additional factors are needed to explain the generation of four distinct cell fates.
The surprising nature of the LIS protein as a splicing factor suggests the participation of components of the splicing machinery in cell fate decisions and, potentially, the generation of a lateral inhibition signal. A possible mode of action could be that LIS is involved in splicing this very signal or, much less direct, some upstream regulator.
Our data suggest that a combinatorial mechanism operates to pattern the female gametophyte: The competence of all cells to differentiate gametic cell fate, together with lateral inhibition from the gametic cells, can ensure maximum likelihood that the key reproductive gametic cells are formed, while at the same time, excess gametic cell formation is prevented. Both the expression of the LIS gene and its distinct function in regulating gametic cell fate are surprising, given that LIS is the Arabidopsis homolog of yPRP4, an integral part of the U4/U6 complex. In the future, identification of LIS target(s) and functional analyses of other tissue-specific splicing factors [20] will help to clarify the mechanistic role of the spliceosome in the regulation of distinct developmental processes.
Plants were grown on soil in growth chambers under long-day conditions at 18 °C. Enhancer detector and gene trap lines were generated using the system of Sundaresan and colleagues [21]. (Send requests for ET and GT lines to UG, [email protected].) The lis-1 allele was isolated from ET1119 in the Landsberg erecta (Ler) accession after mutagenesis; seeds were mutagenized by incubation in 0.15% EMS for 10 h. A total of 5,200 M1 plants were screened for deviating GUS expression in the female gametophyte. The lis-2 allele (SALK_070009) was obtained from the SALK T-DNA insertion collection (http://signal.salk.edu). For the pMEA::GUS reporter construct, a gift from D. Page, a 1.6-kb promoter fragment (upstream of the ATG of MEDEA) was cloned into pCAMBIA1381z using EcoRI/NcoI restriction sites.
Mapping of lis-1 was done as described [22] using the polymorphisms annotated by CEREON [23]. The Columbia accession (Col) was used as a crossing partner. lis-1 was located to Chromosome 2 in an area of 126 kb between the polymorphisms CER448978 and CER446310 on BACs F13H10 and T32G6, respectively. We amplified and sequenced lis-1/LIS genomic DNA spanning 15 open reading frames and identified a single heterozygous locus in At2g41500. The LIS cDNA was isolated according to the annotated open reading frame from a cDNA library using primer 5′-GATTGAGGATCCATGGAACCCAACAAGGAT-3′ and 5′-GATTGAGGATCCAAACAAAGTTCATTCATTTGC-3′ and cloned into pDRIVE (Qiagen, Hilden, Germany). The cDNA was released as an XhoI/KpnI fragment, and cloned into pMDC134 (M. D. Curtis, unpublished data). Upstream of the cDNA, a Gateway recombination site (Invitrogen, Karlsruhe, Germany) was introduced using the XhoI/SacI sites. The LIS promoter region was amplified by PCR from genomic DNA using primers 5′-AAGAAACAGCCAAATAGATAAGCA-3′ and 5′-GTTTCCCTTAAATCCTCAAAAGAAAACACC-3′ and cloned into pENTR 1A (Invitrogen) to generate a Gateway compatible promoter fragment. This promoter fragment was cloned upstream of the LIS cDNA via Gateway recombination.
Plants were genotyped for the lis-1 allele using primers 5′-CTACAAGCTATGACAAGACGTGGAGACT-3′ and 5′-TTTTGCTTGGATACGAGGAGGACCAATGGA-3′. The resulting 264 base pair (bp) fragment was digested with BfmI yielding two fragments of 244 bp and 20 bp from the product of the lis-1 allele, whereas the wild-type product is not digested.
Total RNA was isolated from wild-type tissues using the E.Z.N.A. Plant RNA Kit (Peqlab Biotechnologie, Erlangen, Germany). Ten micrograms of total RNA were used for mRNA-isolation via Dynabeads mRNA DIRECT Micro Kit (Invitrogen). Superscript II (Invitrogen) was used for reverse transcription. Intron-spanning primers were used for PCR to prevent the amplifica-tion of genomic DNA. LIS cDNA was amplified using: 5′-CACTGCCTCATACGACATGAAAGTC-3′ and 5′-ACGAGCTATCTGCTGTGATATCTAGAG-3′. ACTIN2 was amplified using 5′-CCTGAAAGGAAGTACAGTG-3′ and 5′-CTGTGAACGATTCCTGGAC-3′. To generate the pLIS::NLS_GUS construct, the LIS promoter region was amplified by PCR from genomic DNA using primer 5′-AAGAAACAGCCAAATAGATAAGCA-3′ and 5′-GTTTCCCTTAAATCCTCAAAAGAAAACACC-3′, and cloned into pDRIVE. The 2.6-kb LIS promoter fragment was isolated by BamHI/XhoI digestion, and ligated into the binary vector pGIIBAR-EcoRV/XhoI. A GUS construct with N-terminal nuclear localization site [24] was inserted downstream of the promoter.
For analysis of mature gametophytes, the oldest closed flower bud of a given inflorescence was emasculated and harvested two days later. Cytochemical staining for GUS activity was performed as described [25], without additional clearing of the tissue prior to observation. Whole-mount mature gametophytes were prepared as described by Yadegari et al. [26]. For whole-mount embryos, siliques were dissected and cleared with chloral hydrate:glycerol:water-solution 8:3:1 (w:v:v) without prior fixation. GUS-stained tissue and cleared whole mounts were visualized using a Zeiss Axioscop Microscope (Zeiss, Oberkochen, Germany).
The National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov) accession numbers for proteins discussed in this paper and the supporting information are cPRP4 (NP_492363), hPRP4 (AAC51925), LIS (AAW80862), LIS homologous sequence (AAD25639), and yPRP4 (P20053).
|
10.1371/journal.pgen.1006360 | Local Transcriptional Control of YUCCA Regulates Auxin Promoted Root-Growth Inhibition in Response to Aluminium Stress in Arabidopsis | Auxin is necessary for the inhibition of root growth induced by aluminium (Al) stress, however the molecular mechanism controlling this is largely unknown. Here, we report that YUCCA (YUC), which encodes flavin monooxygenase-like proteins, regulates local auxin biosynthesis in the root apex transition zone (TZ) in response to Al stress. Al stress up-regulates YUC3/5/7/8/9 in the root-apex TZ, which we show results in the accumulation of auxin in the root-apex TZ and root-growth inhibition during the Al stress response. These Al-dependent changes in the regulation of YUCs in the root-apex TZ and YUC-regulated root growth inhibition are dependent on ethylene signalling. Increasing or disruption of ethylene signalling caused either enhanced or reduced up-regulation, respectively, of YUCs in root-apex TZ in response to Al stress. In addition, ethylene enhanced root growth inhibition under Al stress was strongly alleviated in yuc mutants or by co-treatment with yucasin, an inhibitor of YUC activity, suggesting a downstream role of YUCs in this process. Moreover, ethylene-insensitive 3 (EIN3) is involved into the direct regulation of YUC9 transcription in this process. Furthermore, we demonstrated that PHYTOCHROME INTERACTING FACTOR4 (PIF4) functions as a transcriptional activator for YUC5/8/9. PIF4 promotes Al-inhibited primary root growth by regulating the local expression of YUCs and auxin signal in the root-apex TZ. The Al–induced expression of PIF4 in root TZ acts downstream of ethylene signalling. Taken together, our results highlight a regulatory cascade for YUCs-regulated local auxin biosynthesis in the root-apex TZ mediating root growth inhibition in response to Al stress.
| The phytohormone auxin, which is synthesized mainly through TRYPTOPHAN AMINOTRANSFERASE OF ARABIDOPSIS 1 (TAA1) and YUCCA (YUC) flavin monooxygenase-like proteins, has an important role in the inhibition of root growth induced by aluminium (Al) stress. TAA1 was recently shown to be locally induced in the root-apex transition-zone thus involves in aluminium-induced Arabidopsis root growth inhibition. Here, we report that YUCCA (YUC) regulates local auxin biosynthesis in the root apex transition zone (TZ) and controls root growth in response to Al stress. Furthermore, EIN3 and PIF4 were found to transcriptionally regulate Al-induced YUCCA expression and thus involve in Al-induced auxin accumulation in root TZ and root growth inhibition in Arabidopsis thaliana
| Aluminium is highly abundant in the soil, but only presents toxicity problems to plants in acid (pH≤5) soils, where it becomes solubilized into the Al3+ ion. Relatively modest levels of Al3+ in the soil are sufficient to inhibit root growth in most species [1,2,3,4]. As a result, gaining an understanding of how Al compromises root growth has preoccupied a substantial body of crop research. The root tip is recognized as the major target of Al stress [5,6,7]; in maize, common bean and Arabidopsis thaliana, the distal portion of the root-apex transition zone (TZ), located between the apical meristem and the basal elongation region, has been shown to be the most sensitive part of the root [7,8,9,10,11,12]. The same part of the root in sorghum is the site for reactive oxygen species production, a class of molecules, which cause root growth inhibition under Al stress [13].
A variety of plant growth and developmental processes are mediated by the phytohormone auxin, which is present for the most part as indole-3-acetic acid (IAA). IAA is mainly synthesized via a tryptophan-dependent pathway [14]. The YUCCA (YUC) family of flavin-containing mono-oxygenases and the TRYPTOPHAN AMINOTRANSFERASE OF ARABIDOPSIS (TAA) family of aminotransferases are both key enzymes in this pathway [15,16,17,18,19]. The Arabidopsis genome harbours three TAA genes, and eleven YUC genes. TAA enzymes catalyse the conversion of tryptophan to indole-3-pyruvate (IPyA), while YUCs are involved in the conversion of IPyA to IAA, a rate-limiting step in the IPyA pathway [18,19,20,21,22].
Local auxin biosynthesis mediated tissue or cellular auxin responses control many plant growth and developmental responses [15,17,18,19,23]. During plant embryogenesis, YUC1, YUC3, YUC4, YUC8 and YUC9 were found to be involved in the control of localized auxin biosynthesis in early embryos [24]. Local auxin biosynthesis mediated plant growth and development is also regulated by tissue or cellular specific transcription factors. For example, AP2 PLETORA transcription factors have been implicated as regulators of lateral organ out-growth through the regulation of localized auxin synthesis controlled by YUCs [25]. The basic helix–loop–helix transcription factors, phytochrome-interacting factor 4 (PIF4) and PIF5, through the regulation of TAA1 and YUC8, integrates temperature into the auxin signaling pathway to control Arabidopsis hypocotyl growth [26,27,28,29].
Phytohormones, particularly auxin and ethylene, play critical roles in modulating root growth in response to Al stress. Previous study indicated that ethylene enhanced Al induced inhibition of root elongation, and exogenous application of aminoethoxyvinylglycine (AVG) and Co2+, ethylene biosynthesis inhibitors, or ethylene signaling mutants such as ein2 and ein3-1 eil1-1 mutants markedly alleviates the Al induced inhibition of root elongation [7,30,31]. Auxin was also found to play a positive role in the Al-induced root growth inhibition [7,32]. Consistently, blocking auxin signaling with the auxin antagonist a-(phenylethyl-2-one)-indole-3-acetic acid (PEO-IAA), a molecule that blocks the auxin binding sites of TIR1/AFB auxin receptors [33,34], or in auxin related mutants such as slr-1 and arf10 arf16 greatly enhances root growth inhibition in response to Al stress [7]. TAA1-mediated localized auxin synthesis has been invoked to explain the basis of both shade avoidance and Al-induced root growth inhibition [7,19]. Exposure of Arabidopsis roots to Al stress has been shown to enhance auxin signalling in the root-apex TZ, a process which is dependent on TAA1-regulated auxin synthesis [7]. This study explores the roles of YUCs (YUC3, YUC5, YUC7, YUC8 and YUC9), which were recently shown to regulate root development [35], in the Al stress-induced inhibition of root growth. The results show that YUCCA controls Al-inhibited primary root growth through the regulation of local auxin biosynthesis in the root-apex TZ in Arabidopsis thaliana. Ethylene-insensitive3 (EIN3) [36] and PHYTOCHROME INTERACTING FACTOR4 (PIF4) [29] functions as a transcriptional activator for YUC5/8/9 in this process.
To understand the roles of YUCs, which regulate the rate-limiting step of auxin biosynthesis via the conversion of IPyA to IAA, in Al stress-regulated root growth inhibition, we examined the phenotypes of YUC mutants such as yuc3, yuc7, yuc8, yuc9, yuc8 yuc9 and yucQ (yuc3/5/7/8/9) under Al stress treatment. Al-dependent root growth inhibition was least severe in yuc9 and yuc8 yuc9 double mutants, while its extent in the other four single mutants was similar to that of the wild type (WT) control (Fig 1A and 1B, S1 Fig). Given the functional redundancy among these five YUC proteins, attention was focussed on the multiple yucQ mutant, in which all five YUC genes have been silenced [35]. Consistent with the previous study which showed the defective root phenotypes [35], we also observed the short root length phenotype in yucQ in the presence of sucrose. However, in the absence of sucrose which was used for the Al-treatment, we didn’t see the clear root length phenotype (S2 Fig). When the roots were exposed to levels of Al >1μM, yucQ is defective in Al-induced growth inhibition (Fig 1C). The more pronounced phenotype in yucQ demonstrated the redundant action of YUC in Al stress-regulated root growth inhibition. When yucasin (5-(4-chlorophenyl)-4H-1,2,4-triazole-3-thiol), an inhibitor of YUC activity [37], was introduced in the growth medium, the effect of Al stress on root-growth inhibition was strongly reduced (Fig 1D). These data indicate that the YUCs-dependent auxin production forms part of the cellular machinery responsible for Al stress-induced root growth inhibition.
To test if YUC-regulated auxin biosynthesis is involved in the local auxin response in the root-apex TZ, and thus in root growth inhibition in response to Al stress, we first examined the expression of the auxin responsive reporter DR5rev:GFP in yuc9, yuc8 yuc9, yucQ mutants (Fig 2). The results show that the strong Al stress-induced DR5rev:GFP signals produced in the root-apex TZ of a AlCl3 treated-WT root was remarkably attenuated in yuc9, yuc8 yuc9 and yucQ mutant backgrounds (Fig 2). The presence of yucasin ensured that YUC activity was effectively compromised [37], thereby alleviating the Al stress-induced inhibition of root growth (Fig 1D); it also strongly reduced the level of Al treatment–induced DR5 expression in the WT root-apex TZ (Fig 2). Thus this indicates that YUC activity is involved in the auxin response in the root-apex TZ when plants are subjected to Al stress.
To address how YUCs regulate the local auxin signaling in the root-apex TZ in the presence of Al stress, the effect of exposure to Al stress on the spatial expression of the YUC genes was analysed by monitoring the expression of the YUCp:eGFP-GUS (YUC3, YUC5, YUC7, YUC8 and YUC9) transgenes. Under non-stressed conditions, GFP signal was detected in the root tips of YUC9p:eGFP-GUS transgenic plants, but not initially in those of either YUC3p:eGFP-GUS, YUC5p:eGFP-GUS, YUC7p:eGFP-GUS or YUC8p:eGFP-GUS; however, after a 2 hours-exposure to Al stress, GFP signals did develop in the root-apex TZ of each of the transgenic lines (Fig 3 and S3 Fig). The GUS-staining assay showed the similar results when exposed to Al stress (S4 Fig). Furthermore, decreasing the pH of the medium from 5.5 to 4.2 did not affect the expression of YUCp:eGFP-GUS (YUC3, YUC5, YUC7, YUC8 and YUC9) transgenes (S5 Fig), suggesting that exogenous Al, not protons, up-regulates YUCs, driving up the accumulation of auxin in the root-apex TZ and finally inhibiting root growth. In summary, similarly to what was demonstrated for TAA1 [7], the YUC genes are also specifically induced in the root-apex TZ in response to Al stress.
Auxin acts downstream of ethylene to regulate Al-induced root growth inhibition has been reported [7]. A comparison of the Al stress-induced expression of the YUCp:eGFP-GUS (YUC3, YUC5, YUC7, YUC8 and YUC9) transgenes in the presence of either 1-aminocyclopropane-1-carboxylic acid (ACC) (the precursor of ethylene synthesis) or AVG (an inhibitor of ethylene synthesis) showed that the induction of GFP signals by Al treatment in the root-apex TZ was intensified by the former, but repressed by the latter in every case (Fig 4 and S6 Fig). These results indicate that the Al stress-regulated up-regulation of the YUC genes in the root-apex TZ is modulated by an ethylene-dependent process.
Al-induced root growth inhibition was alleviated in the double mutant ein3-1 eil1-1 as compared with the WT control, and inhibition of root growth in response to Al stress depends on ethylene signalling [7]. Thus, EIN3 and EIL1 may play an important role in the Al-regulated local up-regulation of YUCs. Therefore, we examined the expression patterns of EIN3 and EIL1 in response to Al stress using EIN3p:GFP and EIL1p:GFP transgenic lines. The results showed that Al stress induced a clear local up-regulation of EIN3p:GFP and EIL1p:GFP in the root-apex TZ (S7 Fig). EIN3/EIL1 transcription factors were reported to bind to a consensus DNA sequence of A[CT]G[AT]A[CT]CT [38,39]. Up to 3.0 Kb fragments upstream of the start codon (ATG) in the YUC3/5/7/8/9 genes were analyzed with the Promomer tool database (http://bar.utoronto.ca/ntools/cgi-bin/BAR_Promomer.cgi) to identify potential EIN3 binding sites (EBS) on these promoters. Two putative EBS have been identified in the promoter of YUC9 (S1 Material). Therefore, EIN3 may directly regulate YUC9 expression through binding to its promoter. To test this possibility, we used transient dual-luciferase assays. By fusing YUC9 promoter with Luciferase gene and testing its activation by overexpressing EIN3 in Arabidopsis protoplasts, we found that EIN3 could dramatically increase YUC9 promoter activity (Fig 5A). To identify whether EIN3 has DNA binding activity to the YUC9 promoter, we carried out yeast one-hybrid assay. Bait constructs containing promoter fragments of YUC9 was prepared and the effector AD-EIN3 construct was generated. The results showed that EIN3 can physically bind to the YUC9 promoter (Fig 5B). Consistently, chromatin immunoprecipitation–quantitative PCR (ChIP-qPCR) assay revealed the association of EIN3 protein with YUC9 promoters in 35S:EIN3-GFP transgenic Arabidopsis (Fig 5C). The strong Al stress-induced YUC9:eGFP-GUS signals produced in the root-apex TZ in Al treated-WT roots were remarkably attenuated in the ein3-1 eil1-1 mutant backgrounds (S8 Fig). Therefore, EIN3 may actively regulate the Al-regulated local expression of the YUC9 by directly binding to the YUC9 promoter.
To address if YUCs-mediated local auxin biosynthesis and thus root growth inhibition in response to Al exposure acts downstream of ethylene signaling, the DR5rev:GFP expression in the WT and yuc mutants (yuc9, yuc8 yuc9 and yucQ) root-apex TZ upon Al stress and in the presence of ACC was compared. The AlCl3/ACC co-treatment resulted in an intensification of the GFP signals throughout the root tip, but the level of enhancement was more modest in the mutant’s root-apex TZs. Furthermore, in the presence of yucasin, the ACC effect on auxin signalling was strongly attenuated (Fig 6). This result implies that the YUCs are involved in the ethylene positive modulation of auxin signalling in the root-apex TZ.
The attenuated response of Al stressed yuc mutants after ACC treatment was monitored to determine whether YUCs-mediated auxin synthesis in the root-apex TZ (and thus root growth inhibition) acts downstream of ethylene signalling. The supply of a low concentration (50 nM) of ACC had no effect on the root growth of either WT or yuc mutant plants (Fig 7A and 7B), but Al co-treatment had a pronounced inhibitory effect on the WT root but not on the yuc mutant roots (Fig 7A–7C). Moreover the addition of 10 μM yucasin lifted the inhibition to root growth imposed on WT plants exposed to 50 nM ACC and 6 μM AlCl3. Note that this concentration of yucasin had no effect on the root growth of non-stressed WT plants (Fig 7D). We conclude that a low auxin level in root-apex TZ caused by a reduced auxin biosynthesis lead to a resistance to ACC-enhancement of Al stress-induced root growth.
Previous studies demonstrated a link between PHYTOCHROME-INTERACTING FACTOR 4 (PIF4) and auxin signalling by showing that PIF4 directly regulates the auxin biosynthetic gene TAA1 and YUC8 [27,28,29]. To assess the potential role of PIF4 protein in Al stress, we examined the phenotypes of the overexpression (35S:PIF4) and knockout (pif4-101) mutants of PIF4 under Al stress treatment. We found that the root growth inhibition in the presence of Al stress was also significantly alleviated in pif4-101 as compared with the WT, while 35S:PIF4 exhibited stronger inhibition of root growth in the presence of Al than the WT plants (Fig 8A). To further confirm whether PIF4 promotes Al-inhibited primary root growth via activation of YUC genes expression, we examined YUC3, YUC5, YUC7, YUC8 and YUC9 expression in 35S:PIF4 transgenic seedlings roots. Compared to the WT control, transcript levels of YUC5, YUC8 and YUC9 were elevated while the expression levels of YUC3 and YUC7 were not altered in the 35S:PIF4 seedlings roots (Fig 8B). Since PIF4 regulated YUC8 expression has been demonstrated [28,29], here we only investigated the potential regulation of YUC5 and YUC9 by PIF4. Using transient expression assay in Arabidopsis leaves, we analysed the activation effect of PIF4 on the expression of a reporter containing the YUC5 or YUC9 promoters fused with the LUC gene. Co-expression of YUC5p-LUC or YUC9p-LUC with the 35S:PIF4 construct led to an obvious induction in luminescence intensity (Fig 8C), suggesting that ectopic expression of PIF4 can activate YUC5p-LUC and YUC9p-LUC expression in this transient expression assay. Together, our assays confirmed that PIF4 not only activated YUC8 expression [28,29], but also activated YUC5 and YUC9 expression. In summary, our data suggest that PIF4 is involved into Al-inhibited primary root growth by regulating YUC5, YUC8 and YUC9 expression.
We showed that YUCs are locally induced in the root-apex TZ in presence of Al stress (Fig 3 and S3 Fig). Here, PIF4 was found to be involved into root growth inhibition under Al stress. To explore the potential regulation of YUCs by PIF4 in this process, we analysed YUC8p:eGFP-GUS and YUC9p:eGFP-GUS expression in root-apex TZ in pif4-101 upon Al stress. Compared with WT, the pif4-101 mutant displayed reduced YUC8p:eGFP-GUS and YUC9p:eGFP-GUS expression in root-apex TZ upon Al stress (Fig 9A and 9B), supporting a role of PIF4 in Al stress–induced local up-regulation of YUCs in root-apex TZ. This result is also consistent with the local up-regulation of PIF4 which was shown by the increased PIF4p:GFP signals in root-apex TZ after a 2 hours-exposure to Al stress (Fig 9C).
To further test whether PIF4 affects Al-induced auxin response in the root TZ, we examined the expression of auxin responsive reporter DR5rev:GFP in the pif4-101 mutant roots (Fig 10). The results showed that the strong Al stress-induced DR5rev:GFP signals produced in the root-apex TZ of WT root was remarkably attenuated in the pif4-101 mutant (Fig 10). This result indicates that the reduced sensitivity of pif4-101 root growth inhibition to the Al stress is due to an reduced auxin response in the root-apex TZ under Al stress.
To address if ethylene signaling is involved in Al-induced up-regulation of PIF4 in root TZ, we examined the PIF4p:GFP signals with the co-treatment of Al and ACC or AVG. The results showed that Al stress-induced up-regulation of PIF4p:GFP in the root-apex TZ was intensified by the ACC co-treatment, but repressed by the AVG co-treatment (Fig 11A). However, the supply of ACC or AVG without Al stress treatment had no effect on the expression of PIF4 in the root tips (Fig 11A). This result suggests that the Al-induced local expression of PIF4 is regulated by ethylene signaling. We also examined PIF4 expression in 35S:EIN3 or 35S:EIL1 transgenic seedlings roots. Compared to the WT control, the transcript levels of PIF4 were elevated in the 35S:EIN3 or 35S:EIL1 seedlings roots (Fig 11B). EIN3 or EIL1 can activate PIF4 expression which was shown by the transient expression assay (Fig 11C). Though PIF4 has been shown to regulate both ethylene biosynthesis and signalling in leaf senescence [40,41], ethylene signalling may also involve in the regulation of PIF4 expression by EIN3/EIL1 in the root under Al stress. Together, our assays demonstrated that the up-regulation of PIF4 in the root-apex TZ under Al stress is regulated by ethylene signalling.
Although the role of auxin in root growth and development was well established [42], its contribution to the response to Al stress has not yet been clarified. Auxin has been demonstrated to mediate many adaptive growth responses [43,44], and we recently have shown that exposure of the Arabidopsis root to Al stress induced a local up-regulation of auxin biosynthesis and thus a burst of auxin signalling in the root-apex TZ and root growth inhibition in response to Al stress. The process is dependent on TAA1, a tryptophan aminotransferase which converts Trp to IPyA, the abundance of which in the root-apex TZ rises in roots challenged by Al stress [7]. Here we demonstrate that members of the YUC family of flavin-containing mono-oxygenases, which catalyse the conversion of IPyA to IAA, a rate-limiting step in the tryptophan-dependent auxin synthesis pathway [17,20,21,45], are induced in the root-apex TZ in response to Al stress (Fig 3). Al stress-induced root growth inhibition of yuc mutants was strongly reduced compare with the WT control (Fig 1). Though the yucQ mutant was reported to have very short roots [35], both yucQ and WT have similar root length in MGRL solution without sucrose, which was used in this study. In addition, the yucQ mutant seedlings, which were grown on MS medium without sucrose, also had normal root length (S2A Fig). However, the supplementation of 1% or 5% sucrose severely inhibited root growth of the yucQ mutant seedlings as compared with the WT control (S2A Fig). Consistently, the highly reduced auxin response, which was shown by decreased DR5rev:GFP signals, in the root tips of yucQ mutant [35], was also alleviated in seedlings which were grown on MS medium without sucrose as compared with seedlings grown on MS medium supplemented with 1% or 5% sucrose (S2B Fig). The increased auxin response in the yucQ mutant in the absence of sucrose might be result from the up-regulation of Trp-independent auxin biosynthesis genes, which may be suppressed by sucrose. This study indicates that Al stress induces local up-regulation of YUCs and thus auxin accumulation in the root-apex TZ and root-growth inhibition (Fig 12). The implication is that plants can exploit various steps within the auxin synthetic pathway to generate the extra auxin required to reprogram the expression of the genes involved in the response to Al stress. This study also shows a good example about how environmental cues control root growth plasticity through the regulation of different steps of auxin biosynthesis and thus influencing local auxin accumulations in root-apex TZ.
In addition to auxin, ethylene, another plant hormone, also plays an important role in controlling root growth and development through cross-talk with auxin [46,47,48,49,50]. The role of ethylene in Al-stress induced root-growth inhibition has been reported [30,31]. Ethylene was found to act as a negative regulator of Al-induced malate efflux by targeting TaALMT1 (ALUMINIUM-ACTIVATED MALATE TRANSPORTER 1)-mediated malate efflux [51]. However, the exact mechanism of ethylene involvement in Al-induced inhibition of root growth is still not understood very well. And the potential participation of auxin-ethylene crosstalk in regulation of the Al-induced root growth inhibition has been suggested [31,51]. In previous reports, auxin biosynthetic genes encoding enzymes, such as WEI2/ANTHRANILATE SYNTHASE α1 (ASA1), WEI7/ASB1, TAA1, and TAR1, are regulated by ethylene [18,52], and ethylene signaling component EIN3 was suggested to directly regulate the ASA1 based on the data of EIN3 ChIP-Seq experiments [53]. Our recent study showed that, in response to Al stress, ethylene induces local TAA1 up-regulation and then auxin accumulation in the root-apex TZ, thus causing root growth inhibition [7]. Here, we observed that Al stress-induced local YUCs up-regulation in the root-apex TZ is an ethylene-dependent process (Fig 4). EIN3 acts as a regulator of Al-induced root growth inhibition [7], and Al stress induces a local expression of EIN3/EIL1 in root-apex TZ (S7 Fig). In this study, EIN3 was found to bind to specific regions of the YUC9 promoter (Fig 5) and Al-induced local expression of YUCs in the root-apex TZ is remarkably attenuated in the ein3-1 eil1-1 mutant (S8 Fig). Al stress-induced and ethylene-dependent local up-regulation of YUCs contributes to auxin accumulation in the root-apex TZ and to the control of root growth inhibition (Figs 6 and 7). This study together with the previous reports suggests that auxin acts downstream and mediates ethylene regulated root-growth inhibition in response Al stress (Fig 12). It greatly improves our understanding about how ethylene mediates environmental cues and regulates plant growth adaptation through the crosstalk with auxin under stressful conditions.
PIF4, a member of the phytochrome-interacting factor (PIF) family of bHLH proteins, affects auxin-mediated growth by directly controlling the expression of TAA1 and YUC genes [27,28,29]. In this study, we uncover a role for PIF4 in Al-induced root growth inhibition (Fig 8A). The increased expression of PIF4p:GFP was located predominantly in the root-apex TZ under Al stress, consistent with its role in Al-induced local up-regulation of YUCs and auxin accumulations in root-apex TZ (Fig 9A). Further time-course analysis of GFP signals in the presence of Al revealed that PIF4p:GFP and YUC8p:GFP signals in the TZ appeared as early as 0.5 hours (S9 Fig), suggesting that YUC may involve in an early Al-responsive signal to regulate the Al-induced root growth inhibition. Thus, as a molecular integrator, the PIF4 transcription factor links Al stress to the auxin pathway in regulating Al-induced root-growth inhibition (Fig 12). Though PIF4 has been shown to regulate both ethylene biosynthesis and signalling in the leaf senescence [40,41], ethylene signalling was shown to involve in the regulation of PIF4 expression through EIN3/EIL1 in the root under Al stress (Figs 11 and 12). The Al stress-regulated expression of PIF4 in the root-apex TZ is modulated by an ethylene-dependent pathway (Figs 11 and 12).
TAA1 mediated local auxin biosynthesis contributes to auxin accumulation in root TZ and regulates root growth under Al stress [7]. In this study, YUCs, which act downstream of TAA1 in auxin biosynthetic pathway [20], were also found to involve in the auxin accumulation in root TZ and controls root growth under Al stress. This process was regulated by EIN3/EIL1 and PIF4 in an ethylene signalling dependent manner. Since it is well known that both auxin biosynthesis and polar auxin transport [16,54] control auxin gradient formation and regulate many plant growth and development [55,56,57]. And initial studies have been shown that polar auxin transport were also important for Al-induced root growth [31,58]. Therefore, it will be interesting to study in the future how polar auxin transport contributes to auxin gradient formation in root TZ and thus regulates root growth in response to Al stress.
The genetic stocks used in this study are Arabidopsis ecotype Col-0 (WT in the text), mutants yuc3 (GABI_376G12), yuc7 (SALK_059832), yuc8 (SALK_096110), yuc9 (SAIL_871G01),yuc9/DR5rev:GFP, yuc8 yuc9/DR5rev:GFP, the transgenes YUC5p:eGFP-GUS, YUC7p:eGFP-GUS, YUC8p:eGFP-GUS, YUC9p:eGFP-GUS [24], DR5rev:GFP [59], DR5rev:GFP/yuc9, DR5rev:GFP/yuc8 yuc9 [24], DR5rev:GFP /yucQ [35], ein3-1 eil1-1 [36], pif4-101 and 35S-PIF4 [29]. The materials were grown on Murashige and Skoog (MS) medium[60] or hydroponically grown, as described by [61], in 2% strength modified MGRL solution (pH 5.0), without inorganic phosphate and with the concentration of calcium adjusted to 200 μM as previously described [7]. This pH was chosen to maintain the Al3+ in solution. The plants were held at 22°C under a 16 h photoperiod.
Arabidopsis (WT and mutant) seeds were germinated for seven days on polypropylene mesh floating over the modified MGRL solution supplemented with various concentrations (0 to 6 μM) of AlCl3, with or without 5–(4–chlorophenyl)-4H-1,2,4–triazole-3–thiol (WAKO, 352–12001), 1-aminocyclopropane-1-carboxylic acid (ACC) (SIGMA, A3903) or aminoethoxyvinylglycine (AVG) (SIGMA, A6685) as previously described [7]. The solution was renewed every two days. Although care was taken to maintain the pH of the medium at 5.0 by regular monitoring, it was unavoidable that it dropped below this level between monitoring time points; this may explain some of the experiment-to-experiment variation in Al-induced root growth inhibition. At the end of the period, the roots were scanned and their length measured from digitized images using Image J software.
Confocal micrographs were captured using a LSM-700 device (Zeiss, Germany). To visualize the stress-induced expression of DR5rev:GFP, DR5rev:GFP/yuc9, DR5rev:GFP/yuc8 yuc9, DR5rev:GFP/yucQ, YUCp:eGFP-GUS, PIF4p:GFP, EIN3p:GFP and EIL1p:GFP transgenes, seedlings were grown in non-supplemented nutrient solution for five days, then transferred to 25 μM AlCl3 for 2 hours as previously described [7]. The roots were stained in propidium iodide to distinguish between living and dead cells. Roots were imaged in water supplemented with propidium iodide (PI, 10 mg/L). Propidium iodide and green fluorescent protein (GFP) were viewed at excitation wavelengths of 488 nm and 561 nm, respectively. Fluorescence emission was collected at 575 nm for propidium iodide and between 500 and 530 nm band pass for GFP. The confocal microscopy assays were detected at least 30 seedlings in each experiment with or without treatments.
Staining of seedling roots for GUS activity was carried out by incubation at 37°C in 0.05M NaPO4 buffer (pH 7.0), 5mM K3Fe(CN)6, 5mM K4 Fe(CN)6 and 2mM X-glucuronide. Once the color had developed, the material was passed through an ethanol series (70%, 50% and 20%) before mounting in 70% chloral hydrate in 10% v/v glycerol.
Yeast one-hybrid (Y1H) assays were carried out by using the Matchmaker Gold Yeast One-Hybrid Library Screening System (Clontech). To prepare constructs for the yeast one-hybrid assay, the promoter region of YUC9 (2 kb upstream of ATG) was amplified by PCR and cloned into the pAbAi vector. To generate AD-EIN3 the coding sequence of EIN3 was amplified by PCR with the respective primers and cloned into the pGAD-T7 vector (Clontech). The yeast one-hybrid assay was performed according to the Yeast Protocols Handbook (Clontech). Briefly, the bait vector was linearized and introduced into yeast strain Y1HGold to make a bait-reporter strain, then prey vector transferred into the aforementioned bait-reporter yeast strain. Transformants were grown on SD/-Leu dropout plates containing 50 or 100 ng ml-1 Aureobasidin A. Primers used for generating various clones are listed in S1 Table.
The seedlings of 7-days-old 35S:EIN3-GFP and 35S:GFP [62] plants were harvested and cross-linked with 1% formaldehyde. ChIP was carried out using the EpiQuik Plant ChIP Kit (Epigentek, Brooklyn, NY, USA) with the antibody against GFP (ab290; Abcam). Input samples and immunoprecipitated samples were analyzed by qPCR. The primer sequences are listed in S1 Table. ChIP-qPCR results were first normalized with input sample. Relative enrichment was then calculated by the ratio of normalized results from 35S:EIN3-GFP plants and the 35S:GFP control.
For qRT-PCR analysis, 7-days-old seedling roots were harvested and frozen in liquid nitrogen for RNA extraction. Total RNA was extracted using RNeasy Mini Kit (Qiagen). Two microgram of total RNA was used to synthesize cDNA using Transcriptional First Stand cDNA Synthesis Kit (Roche). PCR amplification was performed with FastStart Universal SYBR Green Master (Roche) on a CFX Connect Real-Time PCR Detection System according to manufacturer's instruction (Bio-Rad). ACTIN2 (AT3G18780) was used as an internal reference. The primers used for qRT-PCR are listed in S1 Table. The expression of each transcript was normalized against the amount of ACTIN2 control transcript in each sample.
For dual luciferase assays, promoter fragments of YUC5, YUC9 and PIF4 were amplified by using specific primers and cloned into the pGreen0800-LUC vector [63]. The EIN3, EIL1 and PIF4 effector constructs were the 35S:EIN3, 35S:EIL1 and 35S:PIF4. For these constructs, the EIN3, EIL1 and PIF4 coding fragments were amplified by PCR and inserted into pDONR221 (Invitrogen). Then the fragments were cloned into the GATEWAY-compatible vector pB7WGF2.0 (Plant Systems Biology, VIB, University of Gent) by LR reaction. Protoplasts were isolated from Arabidopsis Col-0 plants as described [64] and transformed with effector constructs together with reporter constructs by the poly (ethylene glycol)-mediated method. Firefly and Renilla luciferase activities were quantified by using a dual-luciferase assay kit (Promega, USA) and detected by use of a Centro XS³ LB 960 Microplate Luminometer (BERTHOLD TECHNOLOGIES) according to the manufacturer’s instructions.
The genomic fragment upstream of the YUC3, PIF4, EIN3 or EIL1 translation start codon were amplified by PCR and cloned into to pDONR221 (Invitrogen). Subsequently, the fragment was cloned into GATEWAY-compatible vector pKGWF7.1 (Plant Systems Biology, VIB, University of Gent) by LR reaction. The resulted plasmid were sequenced, introduced into Agrobacterium strain GV3101 or PMP90, and transformed into Col-0 plants using the floral dip method [65]. Five independent transgenic lines were examined. Primers used for the vector construction are shown in S1 Table.
Data sets were analyzed using Prism 6 software (GraphPad Software). Comparisons between two groups were made using Student’s t test. Comparisons between multiple groups were made using one-way or two-way ANOVA tests depending whether one or two different variables were considered, respectively. All values were presented as mean ± SD, values of p less than 0.05 were considered significant. *, **, *** denote differences significant at, respectively, P < 0.05, < 0.01 and < 0.001.
Sequence data from this article can be found in the Arabidopsis Genome Initiative database and the GenBank/EMBL database and under the following accession numbers: YUC3 (AT1G04610), YUC5 (AT5G43890), YUC7 (AT2G33230), YUC8 (AT4G28720), YUC9 (AT1G04180),TAA1(AT1G70560), EIN3 (AT3G20770), EIL1(AT2G27050), PIF4 (AT2G43010).
|
10.1371/journal.pbio.0060172 | Humans Lack iGb3 Due to the Absence of Functional iGb3-Synthase: Implications for NKT Cell Development and Transplantation | The glycosphingolipid isoglobotrihexosylceramide, or isogloboside 3 (iGb3), is believed to be critical for natural killer T (NKT) cell development and self-recognition in mice and humans. Furthermore, iGb3 may represent an important obstacle in xenotransplantation, in which this lipid represents the only other form of the major xenoepitope Galα(1,3)Gal. The role of iGb3 in NKT cell development is controversial, particularly with one study that suggested that NKT cell development is normal in mice that were rendered deficient for the enzyme iGb3 synthase (iGb3S). We demonstrate that spliced iGb3S mRNA was not detected after extensive analysis of human tissues, and furthermore, the iGb3S gene contains several mutations that render this product nonfunctional. We directly tested the potential functional activity of human iGb3S by expressing chimeric molecules containing the catalytic domain of human iGb3S. These hybrid molecules were unable to synthesize iGb3, due to at least one amino acid substitution. We also demonstrate that purified normal human anti-Gal immunoglobulin G can bind iGb3 lipid and mediate complement lysis of transfected human cells expressing iGb3. Collectively, our data suggest that iGb3S is not expressed in humans, and even if it were expressed, this enzyme would be inactive. Consequently, iGb3 is unlikely to represent a primary natural ligand for NKT cells in humans. Furthermore, the absence of iGb3 in humans implies that it is another source of foreign Galα(1,3)Gal xenoantigen, with obvious significance in the field of xenotransplantation.
| Identification of endogenous antigens that regulate natural killer T (NKT) cell development and function is a major goal in immunology. Originally the glycosphingolipid, iGb3, was suggested to be the main endogenous ligand in both mice and humans. However, recent studies have challenged this hypothesis. From a xenotransplantation (animal to human transplants) perspective, iGb3 expression is also important as it represents another form of the major xenoantigen Galα(1,3)Gal. In this study, we assessed whether humans expressed a functional iGb3 synthase (iGb3S), the enzyme responsible for lipid synthesis. We showed that spliced iGb3S mRNA was not detected in any human tissue analysed. Furthermore, chimeric molecules composed of the catalytic domain of human iGb3S were unable to synthesize iGb3 lipid, due to at least one amino acid substitution. We also demonstrated that purified human anti-Gal antibodies bound iGb3 lipid and mediated destruction of cells transfected to express iGb3. A nonfunctional iGb3S in humans has two major consequences: (1) iGb3 is unlikely to be a natural human NKT ligand and (2) natural human anti-Gal antibodies in human serum could react with iGb3 on the surface of organs from pigs, marking these tissues for immunological destruction.
| Identification of endogenous antigens that regulate NKT cell development and self-recognition represents a major goal in immunology. This unique population of T cells is characterised by expression of an invariant Vα14Jα18 TCR—Vα24Jα18 in humans—and the recognition of glycolipid antigens presented by CD1d [1]. When activated, natural killer T (NKT) cells regulate immune responses through their ability to produce large amounts of cytokines such as interferon (IFN)-γ and interleukin (IL)-4 [2]. NKT cell deficiencies are associated with a range of diseases, including cancer, autoimmunity, and infection, in mice and humans [2]. This, combined with the fact that NKT cell numbers vary widely in humans [3], highlights the importance of understanding the endogenous antigens in humans that regulate NKT cell development and function.
Initial work demonstrated that α-galactosylceramide (α-GalCer), a glycosphingolipid originally derived from a marine sponge [4], was a potent agonist for NKT cells in a CD1d-dependent manner in both mice and humans [5,6]. However, the physiological relevance of this in mammalian systems was difficult to understand because α-GalCer is not a mammalian product. Zhou et al. [7] demonstrated that a deficiency in the lysosomal enzymes β-hexaminidase A and B selectively abrogated NKT cell development, suggesting that glycolipid(s) downstream of these enzymes are responsible for NKT cell selection. Experiments to directly test which of the candidate glycolipids were capable of stimulating NKT cells pointed to the glycosphingolipid, isogloboside 3 (iGb3). Both mouse and human fresh NKT cells, and NKT cell hybridomas and lines, responded to iGb3, and furthermore, specific inhibition of iGb3 on human cells, using isolectin B4 (IB4) that should selectively target iGb3 via its terminal Galα(1,3)Gal sugars, suggested that iGb3 was also a primary human self-antigen for NKT cells [7]. These data led the authors to suggest that iGb3 was the main endogenous ligand responsible for NKT cell development and self-recognition in both mice and humans. Subsequent studies from independent groups have confirmed that iGb3 is an agonist ligand for at least a subset of mouse and human NKT cells [8–13], and furthermore, that this glycosphingolipid appears to be important for shaping the NKT cell TCR repertoire in mice [12]. However, recent studies have challenged the hypothesis that iGb3 is the primary ligand responsible for NKT cell development in mice [14–16]. One of these studies [16] failed to detect iGb3 in mouse or human thymus, although this study could not exclude the existence of low levels of iGb3, or higher levels of iGb3 expressed by a minor subset of the thymus. Another study [15] demonstrated, by using iGb3 synthase (iGb3S) knockout mice, that NKT cell development was apparently normal, which more strongly suggested that iGb3 is at least not essential for this process in mice. Lastly, two papers have provided evidence that the defect in NKT cell development in Hex-b–deficient mice may be the lysosomal storage disease that occurs with this mutation, thus causing a nonspecific defect in glycolipid processing and presentation [14,17,18]. The ability of iGb3 to activate human NKT cells is not in dispute; what remains in question is the role of iGb3 in human NKT cell biology.
Another issue of major importance involving iGb3 is from the perspective of xenotransplantation, in which expression of this glycolipid on the cell surface of pig tissue could represent a major problem if it is not present in humans. iGb3 is synthesized by iGb3S, a member of the α1,3Gal/GalNAc transferase or Family 6 glycosyltransferases. Other family members include α1,3galactosyltransferase (α1,3 GT), and the B blood group transferase, which like iGb3S, transfer αGal. In contrast, the two other members of the family, A blood group transferase and Forssman synthetase, transfer αGalNAc. Members of Family 6 are the only known mammalian glycosyltransferases that transfer either αGal or αGalNAc in an α1,3 linkage to their respective acceptor molecules. Analysis of the human genome shows the genes for the α1,3GT, iGb3S, A/B blood group transferase, and the Forssman synthetase are present [19]. Recently, GT6m7, a new Family 6 member, was reported [19]; however, in humans, the gene for this glycosyltransferase contains a premature stop codon in the last exon. Indeed, mutation appears to be common in this family, with varying effects (see Table 1).
In a similar fashion to ABO blood groups, in which natural antibodies are made to specificities that individuals lack, humans produce anti-αGal antibodies as a consequence of nonfunctional, or nontranslated enzymes. The evolutionary event that led to selection of the αGal-ve phenotype in humans is not clear, but selective pressure of the αGal+ve protozoan parasites has been postulated [20].
It is well known that the presence of the Galα(1,3)Gal xenoepitope, synthesized by α1,3galactosyltransferase (α1,3GT), causes hyperacute rejection of donor organs in pig-to-human xenotransplantation [21]. To avoid this problem, the α1,3GT gene has recently been deleted in pigs [22]. However, there is still low-level expression of Galα(1,3)Gal [23], presumably synthesized by iGb3S. Thus, organs from GT−/− pigs transplanted into humans may still potentially be subject to rejection by human natural antibodies to Galα(1,3)Gal in the form of iGb3. The study from Zhou and colleagues [7] provided data suggesting that this is unlikely to pose an immediate problem, because they showed that human anti-Gal antibodies did not react with iGb3, presumably because it was a self-ligand that caused deletion of iGb3-reactive lymphocytes in humans. Thus, although there is clearly a significant level of controversy surrounding iGb3, the fact remains that there are compelling results both for and against a role for iGb3 in NKT cell development. For the sake of understanding the factors that regulate this process, as well as whether iGb3 poses an additional problem for xenotransplantion, further studies are required to resolve this issue.
We recently characterized the mouse iGb3S cDNA [24] that encodes the enzyme that synthesizes the Galα(1,3)Gal xenoepitope on iGb3 by catalysing the transfer of donor sugar from UDP-Gal in an α−1,3 linkage to its acceptor molecule Galβ(1,4)Glc-ceramide [25]. This reaction is the first step in the isoglobo-series pathway, which also results in the generation of iGb4 and isoForssman glycolipids (Figure 1).
Here, we have examined iGb3S expression and functional potential in human tissues. Moreover, to assess whether iGb3 might represent a xenoantigen that remains in α1,3GT knockout pigs, we investigated whether iGb3 glycolipid is recognized by natural human anti-Gal antibodies present in normal human serum, and we also determined whether cells expressing this glycolipid on the cell surface are readily targeted for complement-mediated lysis.
Several lines of evidence from our studies of the Galα(1,3)Gal epitope suggest that iGb3S is not expressed in humans. In contrast to both rat and mouse, in which the iGb3S gene is transcribed and the RNA processed [24,25], analysis of a human multiple tissue northern blot did not detect iGb3S mRNA (unpublished data). Furthermore, anti-Gal monoclonal antibodies (mAbs) that detect both rat and mouse iGb3 on tissues and cell lines do not react with a range of both normal and malignant human tissues and cell lines [24] (and our unpublished data). To examine expression of iGb3S mRNA in greater detail, reverse-transcription PCR (RT-PCR) was used to analyse several human tissues. Oligonucleotide primers for these experiments (Table S1) were designed based on the exon arrangement of the human iGb3S gene, established by the analysis of Genbank DNA sequences. RNA from heart, kidney, spleen, lung, and thymus (the latter two tissues express iGb3S in both rat and mouse) generated a product of the correct size (∼550 bp) with forward and reverse primers within exon five (unpublished data). A product was also obtained from cDNA from human dendritic cells (Figure 2B, lane 6). Some products were confirmed as human iGb3S by direct sequencing (unpublished data). However, generation of products within an exon may be due to genomic DNA or heteronuclear RNA; therefore, amplification across exon boundaries is required to show the presence of mRNA. We have previously shown that iGb3S mRNA can be successfully used as a template for cross-exon RT-PCR from mouse RNA [24]. Despite exhaustive attempts (at least 50 times) to amplify human iGb3S using a combination of primers spanning all five exons in different tissues (Figure 2A), products were either not obtained from human template or the size of several of the PCR products corresponded to that expected from genomic DNA rather than spliced mRNA. The data shown are the amplification from dendritic cells (Figure 2B); however, similar results were also obtained from all tissues examined. Primers across exon 1 to 3, exons 1 to 4, and exons 2 to 4 yielded products expected from amplification of genomic DNA rather than spliced RNA (lanes 1, 2, and 4, respectively, Figure 2B and Table S2). The products observed in lanes 6 and 7 are within a single exon and could represent either genomic or mRNA products as there is no splicing over this region of the iGb3S gene. Despite using numerous primer combinations, including forward primers from exons 2, 3, or 4 with a reverse primer from exon 5 (Figure 2C), we were unable to detect any products of the correct size to suggest spliced iGb3S mRNA in any human tissue examined (Figure 2D), even when a high cycle number (up to 40) was used (unpublished data). As expected, products were not observed when template was omitted.
From our experience, mouse iGb3S mRNA is expressed at low levels and is difficult to amplify. Therefore, our inability to detect human iGb3S mRNA was not conclusive evidence that it was absent. Transfection of CHOP cells with mouse iGb3S cDNA results in high level expression of its product Galα(1,3)Gal [24]. We used the same approach to determine whether humans express functional iGb3S. In vitro functional studies indicate that the catalytic domain of iGb3S (which represents 75% of the entire molecule) is encoded by two exons, a small exon (exon 4) and a larger one (exon 5) encoding the major part of the functional domain [26]. Soluble forms of the truncated catalytic domain of several members of the glycosyltransferase family have been shown to be enzymatically active. Using splice overlap extension PCR, we initially generated a chimeric molecule in which exon 5 of the functional rat iGb3S was substituted with that of the human iGb3S homolog (generated from human genomic DNA) (Figure 3A and Table S3). The other rat exons encode the cytoplasmic tail, transmembrane domain, and stalk that anchors the molecule in the lipid bilayer. This approach of exchanging catalytic domains to examine function has been successfully used with Forssman synthetase, another member of this glycosyltransferase family [27]. The ability of this chimeric rat/human(exon5)-iGb3S molecule to synthesize Galα(1,3)Gal was determined by analysis of transfected CHOP cells. As expected, cells transfected with DNA encoding rat iGb3S displayed strong cell surface expression of the Galα(1,3)Gal epitope on glycolipid as determined by binding of the monoclonal antibody 15.101 [28] and human anti-Gal immunoglobulin (Ig) purified from normal human serum (Figure 3B). The 15.101 mAb has been shown to bind preferentially to Galα(1,3)Gal on iGb3 lipid [28]. The chimeric molecule containing the majority of the catalytic domain of human iGb3S (rat/human(exon5)-iGb3S) was unable to synthesise the Galα(1,3)Gal epitope as staining was not observed with 15.101 or human anti-Gal Ig (Figure 3B). A second chimeric molecule comprising the entire human catalytic domain, exon 4 together with exon 5 (rat/human(exon4,5)-iGb3S), was also unable to synthesize Galα(1,3)Gal (Figure 3B). Data from several other mAbs and Bandeiraea simplicifolia IB4 lectin that bind the Galα(1,3)Gal epitope (Figure S1) support the conclusion that the human iGb3S catalytic domain is not functional. Detection of the FLAG epitope in both chimeric enzymes confirmed that the absence of Galα(1,3)Gal synthesis was not due to impaired translation or expression (Figure S2A). As glycosyltransferases are integral membrane proteins of the Golgi complex where oligosaccharides are synthesized, perinuclear staining (Golgi-like) confirmed correct trafficking of the chimeric enzymes (Figure S2B). Staining was not observed with cells transfected with vector alone.
To explore the unlikely possibility that iGb3 staining was not observed due to antibody inaccessibility, an alternative detection method was used. Synthesis of iGb3 is the initial step for the formation of the isoglobo-series glycolipid pathway, and iGb3 is the precursor to iGb4 and, ultimately, isoForssman [25] (see Figure 1). To examine whether iGb3 was synthesized, a complementation assay in CHOP cells, which lack both iGb3S and Forssman synthetase (FS), was used to determine whether coexpression of chimeric iGb3S with FS results in expression of isoForssman glycolipid. As expected, cells transfected with FS alone did not stain for isoForssman (unpublished data), whereas cells transfected with both rat iGb3S and FS were positive for isoForssman (Figure 3C). Expression of Galα(1,3)Gal was confirmed by 15.101 binding. Cells cotransfected with either of the chimeric molecules (rat/human (Exon5)-iGb3S or rat/human (Exon4,5)-iGb3S) and FS did not show any detectable isoForssman staining (Figure 3C). As expected, no Galα(1,3)Gal was observed following staining with 15.101. Thus, the human catalytic domain appears to be incapable of generating detectable iGb3 and does not initiate the downstream synthesis of the iGb4 structure required for FS to function.
Using site-directed mutagenesis, we analysed which amino acid(s) contributed to the loss of function we observed in human iGb3S. Despite an overall similarity of approximately 72%, there are 77 differences within the catalytic domain of the functional rat iGb3S and nonfunctional human iGb3S, any of which, either alone or in combination, may be involved in the loss of function observed with human iGb3S. The targeted residues were selected by comparison of the aligned amino acid sequences of the iGb3S catalytic domains (exon 4 and 5 encoded) of species known to synthesize iGb3 (rat, mouse, and dog) with that of human. To identify more precise candidates, amino acids were excluded: (1) if the human amino acid was identical to either the mouse or dog, (2) the amino acid residue was different in all four species, or (3) the substitution was with an homologous amino acid. Four of these amino acids in rat exon 5 were selected in these initial studies and mutated to their human equivalent (Figure 4A). The single isolated substitution of rat Y252N resulted in the complete elimination of Galα(1,3)Gal staining (Figure 4B), showing that this asparagine in human iGb3S is sufficient to ablate enzymatic function. Rat L187P showed a significant reduction (typically 70%–95%) in Galα(1,3)Gal staining, whereas both the rat A221S and rat E280A substitutions showed strong Galα(1,3)Gal expression that was comparable with that observed following transfection with rat iGb3S (Figure 4B). As expected, a complementation assay with FS resulted in strong isoForssman staining with both rat A221S and rat E280A substitutions (Figure 4C). A similar high level of isoForssman staining was also observed with the rat L187P substitution, despite there being minimal Galα(1,3)Gal expression, thus demonstrating the sensitivity of this method. IsoForssman staining was not observed with cells cotransfected with rat Y252N (Figure 4C).
It is possible that the Y252N and L187P substitutions are not the only ones in humans that influence function. This was examined by reverse mutation of the nonfunctional chimeric rat/human(exon 5)-iGb3S to their rat equivalents with either point mutation alone (i.e., P187L or N252Y), or in combination (P187L+N252Y). A gain of function would suggest these are the primary residues involved in determining whether the transferase is functional. Staining with mAb 15.101 showed no Galα(1,3)Gal expression following transfection of CHOP cells with either the single or combined reverse-mutated chimeric cDNA molecules (Figure 5). The implications of these data are that human iGb3S must have multiple mutations that have resulted in its inactivation. Typical strong Galα(1,3)Gal expression was observed with cells transfected with rat iGb3S.
High-performance thin layer chromatography data from Zhou et al [7] suggested natural mixed human serum antibodies did not recognize iGb3, suggesting that iGb3-reactive B cells had been deleted from the human repertoire, further evidence that iGb3 lipid is present in humans. To test this ourselves, we used a lipid ELISA, and by this approach, demonstrated clear binding of both natural human anti-Galα(1,3)Gal antibodies and the mAb 15.101 to purified iGb3 lipid over several antibody dilutions (Figure 6A). Binding was not observed with either anti-CD17 (lactosylceramide) or anti-CD77 (Gb3) mAbs. Furthermore, as a specificity control, treatment of iGb3 lipid with α-galactosidase, which specifically removes the terminal α(1,3)Gal moiety, resulted in a significant inhibition (up to 60%) of binding by natural human anti-Galα(1,3)Gal antibodies (Figure 6B), yet had no effect on anti-CD17 binding to lactosylceramide (no αGal moiety) in a parallel assay. Antibody specificity was further demonstrated by the lack of binding of natural human anti-Galα(1,3)Gal antibodies to either Gb3 (Figure 6C) or lactosylceramide (Figure 6D). However, as expected specific binding of both anti-CD77 and anti-CD17 mAbs (used at the same dilutions as in Figure 6A) were observed (Figure 6C and 6D, respectively).
A key question that remains to be answered is, if human cells were to express iGb3, would they be susceptible to antibody-dependent complement-mediated lysis due to natural human anti-αGal antibodies present in normal human serum (NHS)? In contrast to nontransfected human cells (αGal−ve) that do not undergo lysis with NHS, human cells expressing iGb3 (αGal+ve) were lysed by NHS (in the presence of rabbit complement) in a dose-dependent manner (Figure 7A). Removal of anti-αGal antibodies from NHS by absorption with Galα(1,3)Gal coupled to glass beads abolished lysis to background levels (Figure 7A). However, lysis was not affected by NHS absorption with uncoupled glass beads (unpublished data). Furthermore, lysis of human cells expressing iGb3 was re-established when anti-αGal IgG antibodies were purified from NHS and used in the cytotoxicity assay (Figure 7A). In addition, this activity could be removed by absorption with Galα(1,3)Gal coupled to glass beads (unpublished data). Inhibition experiments verified that Galα(1,3)Gal is the epitope that the antibodies recognise, as a significant dose-dependent reduction in lysis was observed by preincubation of both NHS and anti-αGal IgG antibodies with Galα(1,3)Gal disaccharide (Figure 7B). No inhibition was observed when lactose (Galβ(1,4)Glc) was used (Figure 7B).
Glycolipids represent one of the last molecular frontiers in immunological recognition. Whereas glycolipids are known to be synthesized in the Golgi and are typically expressed on the cell surface, the exact transport pathway(s) for newly synthesized glycolipids is not well defined. However, it is assumed to be similar to glycoproteins and involve vesicular flow from the endoplasmic reticulum through the Golgi complex to the plasma membrane. Glycolipids, particularly exogenous glycolipids, can localize to lysosomal compartments via endocytosis. Similarly, our knowledge of how glycolipids control immune responses and the context in which they are presented by CD1d and recognized by NKT cells is also still very limited. Since the glycolipid, α-GalCer, was originally shown to potently stimulate NKT cells in a CD1d-dependent manner, there has been an enormous effort to identify other ligands. Several classes of natural CD1d-binding ligands for NKT cells have been identified, including microbial-derived α-linked glycosphingolipids from the nonpathogenic Sphingomonas bacteria and phosphatidylinositol mannoside from Mycobacteria (reviewed in [29]). Recently, a diacylglycerol glycolipid from Borrelia burgdorferi, a human pathogen responsible for Lyme disease, was shown to directly stimulate both human and mouse NKT cells [30]. Although these ligands are all candidates for NKT cell recognition of non-self, none of these are present in normal mammalian cells.
The main candidate self glycolipid-antigen is iGb3. The original collective data, primarily based on the use of β-hexosaminidase-B–deficient mice that are incapable of degrading iGb4 into iGb3 in lysosomes, supported the claim that iGb3 lipid was a principle endogenous ligand for Vα14 NKT cells in mice and, albeit indirectly, in humans [7,12,31]. The interpretation of data using β-hexosaminidase-B–deficient mice was contested by Gadola et al. [14], where it was argued that these mice have a generalised lysosomal storage disease that indirectly impaired CD1d loading in lysosomes. Their interpretation was that it was the accumulation of glycolipids in lysosomes, rather than the lack of iGb3, that abrogated NKT cell development. Some of the data in this paper [14] simply conflicted with that of the earlier study of β-hexosaminidase-B–deficient mice [7], making it difficult to determine which interpretation was correct [14]. Similar suggestions were raised in an independent study of mutations leading to lysosomal storage diseases [18]. Recently, Porubsky et al. and Speak et al. [15,16] failed to detect iGb3 in mouse and human thymus using a biochemical approach. Furthermore, Porubsky et al. [15] reported normal development and function of invariant NKT (iNKT) cells in iGb3S−/− mice. Although the lack of biochemical evidence for iGb3 in thymus might simply be an issue of insufficient sensitivity, the results from the iGb3S−/− mice more strongly challenge the significance of iGb3 in mouse NKT cell development. There is no easy interpretation that incorporates and integrates the findings from the studies for, and against, a role for iGb3 in mouse NKT cell development. In our opinion, this represents one of the most important and controversial issues in the NKT cell field that requires additional input from independent research groups.
In humans, synthetically derived iGb3 can stimulate human NKT cells to proliferate and produce cytokines [7,8,32] and recognition of human dendritic cell self-antigen can be blocked by IB4 lectin [7]. However, direct biochemical evidence to show that human iGb3 is an endogenous NKT cell ligand has been lacking. Although iGb3 was not detected in human thymus or human dendritic cells using a high-performance liquid chromatography (HPLC) assay, this assay had a detection limit of 1% iGb3 to 99% Gb3, which does not exclude the presence of iGb3 at low but still biologically significant levels [16]. Indeed, during review of this manuscript, two publications from Li et al., claimed to be able to discriminate iGb3 from Gb3 (in artificial mixtures and from rat cells) and identified iGb4 from human paediatric thymi, using electrospray ionisation-ion trap mass spectrometry [33,34]. Although these analyses are at odds with our own, they have yet to conclusively demonstrate immunologically significant levels of iGb3 in human tissue. Specifically, the formal possibility remains that the minor MSn mass spectral signature for iGb4 detected in these studies is derived from related tetraglycosylceramides, as acknowledged by these investigators. Alternatively, the very low levels of iGb4 detected in these analyses may be derived from dietary sources and distributed throughout the body via lipoprotein particles.
The presence or absence of iGb3 in humans has potential major implications for xenotransplantation. If humans express iGb3S, iGb3 lipid present on transplanted pig tissues will not be “seen” as foreign and therefore would not represent a drawback for xenotransplantation. However, as humans do not express functional iGb3S (reported herein), then the presence of lipid-linked Galα(1,3)Gal in pigs, synthesized by iGb3S, may pose a serious risk to successful xenotransplantation, even when using α1,3GT knockout pigs as donors (which were specifically generated to eliminate Galα(1,3)Gal epitopes for xenotransplantation purposes). What are the implications of this in a transplant setting? Currently, we know that expression of iGb3 does not mediate hyperacute rejection of pig tissues transplanted into baboons [35,36]. However, human serum has at least a 4-fold higher level of natural anti-Galα(1,3)Gal antibodies (∼1% of human IgG) than other primates [37], so this may not directly represent the human situation. Furthermore, iGb3 expression in pigs may have more serious consequences in the later phases of graft rejection. Firstly, changes in the affinity/avidity of the elicited antibodies may cause tissue damage by complement fixation. It is clear that the level of anti-Galα(1,3)Gal antibody is critical for the speed of rejection in experimental models [38]. Alternatively, elicited anti-Galα(1,3)Gal antibodies may contribute to the acute vascular rejection observed when hyperacute rejection is eliminated, such as in knockout pig-to-primate transplants, by activation of endothelial cells via cross-linking of the lipid itself. Pathological features similar to acute vascular rejection are seen in humans when the Gb3 lipid (closely related to iGb3) is cross-linked by bacterial toxins [39]. Secondly, because iGb3 activates human NKT cells [7,32] (in which synthetic, purified. and enzymatically derived iGb3 were all tested), consequently the expression of iGb3 on pig cells could lead to NKT cell activation resulting in destruction of the xenograft. Furthermore, our data clearly show that the anti-Gal antibodies in NHS can lyse iGb3 expressing cells (Figure 7) and therefore any remaining iGb3 on pig cells may be a target for antibody-mediated destruction. Whereas it is clear that the use of heavy immunosuppression can control the later phases of xenograft rejection, the major advantage of xenotransplantation over allotransplantation is the ability to genetically modify the donor. The ultimate goal is to engineer a donor pig such that minimal, or indeed no immunosuppression is required for long-term graft survival. It is likely that genetic modification of pigs may be required to eliminate any effects of iGb3. Only at that stage will other obstacles be revealed. Thus, in formally demonstrating the lack of functional iGb3S in humans, this study alerts transplantation immunologists to a previously unrecognised risk associated with expression of iGb3 glycolipid on α1,3GT knockout pig tissues. Expression of this glycolipid could act as a secondary source of Galα(1,3)Gal xeno-antigen capable of binding natural human anti-Gal antibodies present in normal human serum and marking these cells for destruction by complement mediated lysis.
In a perspectives article, Godfrey, Pellicci, and Smyth [40] asked whether the search for the elusive NKT cell antigen is over. In mice, in view of several recent publications [14–16], the possible existence of NKT cell-selecting ligands other than iGb3 remains an important consideration [17]. It had generally been assumed that experimental data obtained from mice would be directly relevant to humans, and Zhou et al. [7] provided indirect evidence that iGb3 is also a self-ligand for human NKT cells. However, our new data demonstrate that there appears to be critical differences between the two systems, and suggests that we are a long way from calling off the search for NKT cell-selecting antigens in humans. This remains one of the most important objectives in the field, and will ultimately lead to a better understanding of the factors that regulate NKT cell development and function in health, and in developing novel therapies for the treatment of disease.
The human genomic iGb3S sequence was obtained from the National Center for Biotechnology Information Web site (http://www.ncbi.nlm.nih.gov) and searching the human genomic database. The nucleotide sequence of the gene A3GALT2 (accession number NT 032977) was used, with the exon/intron boundaries for the human iGb3S gene as listed with the sequence.
We were unable to clone human iGb3S from total RNA from adult human tissues (heart, lung, kidney, spleen, and thymus) (Stratagene) or cDNA from dendritic cells using the TITANIUM One-Step RT-PCR Kit (Clontech) with a series of degenerate primers (Tables S1 and S2). The chimeric rat/human molecules included the exchange of rat exon 5 (rat/human(exon5)-iGb3S) and rat exons 4 and 5 (rat/human(exon4,5)-iGb3S) with the equivalent human exon(s). The rat/human chimeras were generated using splice overlap extension PCR. The specific primer combinations used are shown in Table S3. The single amino acid substitutions in rat iGb3S (L187P, A221S, Y252N, and E280A) and the reverse mutations in rat/human(exon5)-iGb3S (P187L, N252Y, and the combined P187L+N252Y) were introduced using the QuikChange site-directed mutagenesis kit (Stratagene) (Table S4). Sequence fidelity, orientation of the insert, and presence of the desired mutation(s) were confirmed by DNA sequencing (Big Dye 3.1; PE-Applied Biosystems).
CHOP cells (Chinese Hamster Ovary cells transformed with Polyoma Large T antigen) [41] and E293 cells (human kidney fibroblasts) were cultured in DMEM (CSL) supplemented with 10% FCS overnight at 37 °C. Transfections were with LipofectAMINE Plus (Life Technologies) as recommended by the manufacturer. Cells were examined after 48 h for either cell surface or intracellular expression of Galα(1,3)Gal using purified natural human anti-Gal antibodies (0.49 mg/ml), and the anti-Galα(1,3)Gal mAbs 15.101, 22.121, 24.7, 25.2, and 8.17 (supernatants) [24,42] and Bandeiraea simplicifolia IB4 lectin. Expression of the FLAG epitope was revealed by staining with the anti FLAG M2 mAb (Sigma). Expression of IsoForssman glycolipid was revealed using an anti-Forssman mAb, FOM-1 (BMA Biomedicals). Expression of lactosylceramide and Gb3 were revealed with anti-CD17 (ascites) and purified anti-CD77 (0.15 mg/ml) mAbs, respectively (Pharmingen). Antibodies were detected with FITC-labelled sheep anti-mouse or human IgG (Dako) or HRP conjugate Sheep anti-human Ig (Silenus) and rabbit anti-mouse Ig (Dako), and analysed either by fluorescence microscopy, flow cytometry (Becton Dickinson FACS Canto II), or lipid ELISA.
Porcine lactosylceramide (Calbiochem) and iGb3 (Alexis Biochemicals) were dissolved in methanol at 1 mg/ml and stored at −20 °C. The ELISA was performed in 96-well Maxisorb plates (Nunc). Lipids were diluted in n-hexane and used at 500-ng/well, incubated for 1 h in a fume hood to dry; plates were then blocked with 3% BSA/PBS for 2 h and washed ×1 with PBS. Primary antibodies, diluted in blocking buffer, were added and incubated for 1 h. After washing ×5 with PBS, secondary antibodies, diluted in blocking buffer, were added and incubated for 1 h before washing ×8 with PBS. All incubations were carried out at room temperature (RT) on a rocking platform. TMB peroxidase substrate (KPL) was used to develop the plate. Colour development was stopped with 0.18 M H2SO4 and quantitated at an optical density at 405 nm (OD405nm) on an ELISA plate reader. For the enzyme digestion, α-galactosidase (Sigma-Aldrich) was diluted in 0.1 M citrate/phosphate buffer (pH 6) and incubated with the lipids overnight at RT, after which an ELISA was performed as described above.
Human E293 cells expressing rat iGb3 [28] were tested for lysis with rabbit complement and normal human sera (NHS, pooled from ten healthy individuals and heat inactivated) or purified human anti-Gal IgG antibodies (prepared by fractionation of the NHS pool on a Protein G Sepharose column (Pharmacia) followed by affinity chromatography on Galα(1,3)Gal-coupled macroporous glass beads (Syntesome) as described previously [43]. In brief, 50 μl of antibody at doubling dilutions were added to 2.5 × 105 cells per well in round-bottomed 96-well plates (Greiner), resuspended, and incubated on ice for 30 min. After two washes, 50 μl of rabbit complement, at an appropriate dilution, was added to the cell pellet, resuspended, and incubated at 37 °C for 30 min (NHS) or 60 min (purified anti-Gal IgG). Cells were pelleted and resuspended in 400 μl of DMEM/0.5% BSA containing 1 μg/ml propidium iodide (PI; Sigma) and analysed by flow cytometry. Percentage lysis (cytotoxicity) was determined by analysis of 10,000 cells.
The importance of anti-Gal antibodies for lysis was determined by serum absorption and carbohydrate inhibition: (1) Absorption; 200 μl of NHS or human anti-Gal IgG was added to an equal volume of Galα(1,3)Gal-coupled macroporous glass beads or non-coupled beads (control) at 4 °C for 30 min; the beads were removed by centrifugation and the absorption step repeated with another aliquot of beads. (2) Carbohydrate inhibition; 25 μl of 20 mM Galα(1,3)Gal disaccharide or lactose (control) was serially diluted and mixed with an equal volume of NHS or human anti-Gal IgG at an appropriate dilution (two dilutions less than the 50% titre of the antibody) and incubated at 4 °C for 16 h. After both of these treatments, the sera were analysed for complement-mediated lysis.
|
10.1371/journal.pgen.1002209 | Gamma-Tubulin Is Required for Bipolar Spindle Assembly and for Proper Kinetochore Microtubule Attachments during Prometaphase I in Drosophila Oocytes | In many animal species the meiosis I spindle in oocytes is anastral and lacks centrosomes. Previous studies of Drosophila oocytes failed to detect the native form of the germline-specific γ-tubulin (γTub37C) in meiosis I spindles, and genetic studies have yielded conflicting data regarding the role of γTub37C in the formation of bipolar spindles at meiosis I. Our examination of living and fixed oocytes carrying either a null allele or strong missense mutation in the γtub37C gene demonstrates a role for γTub37C in the positioning of the oocyte nucleus during late prophase, as well as in the formation and maintenance of bipolar spindles in Drosophila oocytes. Prometaphase I spindles in γtub37C mutant oocytes showed wide, non-tapered spindle poles and disrupted positioning. Additionally, chromosomes failed to align properly on the spindle and showed morphological defects. The kinetochores failed to properly co-orient and often lacked proper attachments to the microtubule bundles, suggesting that γTub37C is required to stabilize kinetochore microtubule attachments in anastral spindles. Although spindle bipolarity was sometimes achieved by metaphase I in both γtub37C mutants, the resulting chromosome masses displayed highly disrupted chromosome alignment. Therefore, our data conclusively demonstrate a role for γTub37C in both the formation of the anastral meiosis I spindle and in the proper attachment of kinetochore microtubules. Finally, multispectral imaging demonstrates the presences of native γTub37C along the length of wild-type meiosis I spindles.
| Proper chromosome segregation during cell division is essential. Missegregation of mitotic chromosomes leads to cell death or cancer, and chromosome missegregation during meiosis leads to miscarriage and birth defects. Cells utilize a bipolar microtubule-based structure known as the meiotic or mitotic spindle to segregate chromosomes. Because proper bipolar spindle formation is critically important for chromosome segregation, cells have many redundant mechanisms to ensure that this structure is properly formed. In most animal cells, centrosomes containing γ-tubulin protein complexes help organize and shape the bipolar spindle. Since meiosis I spindles in oocytes lack centrosomes, the mechanisms by which a meiotic bipolar spindle is assembled are not fully understood. In Drosophila oocytes it was not clear whether γ-tubulin played a role in bipolar spindle assembly or if it was even present on the meiotic spindle. We demonstrate that γ-tubulin plays vital roles in bipolar spindle formation and maintenance, as well as in aligning the chromosomes on the oocyte spindle. Additionally, we show that γ-tubulin is present on the bipolar spindle in Drosophila oocytes. More importantly, we demonstrate that γ-tubulin plays a critical role in the formation of the kinetochore microtubules that are required to properly orient chromosomes on the meiotic spindle.
| In mitosis and male meiosis in animals, the establishment of spindle bipolarity is mediated by centrosomes that act as microtubule organizing centers (MTOCs). These structures serve to organize and focus the growing microtubules to form a bipolar spindle. γ-Tubulin is a primary component of MTOCs and is required for mitotic spindle assembly in many organisms (reviewed in [1]). However, in most animal species, including Drosophila melanogaster, female meiosis is acentrosomal and the mechanisms by which a bipolar spindle is formed during meiosis I have not been fully elucidated.
Despite the absence of centrosomes, a role for γ-tubulin in female meiosis has been implicated in many organisms. Schuh and Ellenberg [2] have presented strong evidence that the spindle in mouse oocytes is formed by the action of a large number of γ-tubulin-containing MTOCs that are self-organized from a cytoplasmic microtubule network. These authors propose that the progressive clustering of MTOCs, along with the action of a kinesin-5 motor protein, facilitates the formation of a bipolar spindle. This mechanism of acentrosomal spindle assembly is fully consistent with mammalian studies of γ-tubulin during meiosis I that show localization of γ-tubulin throughout the meiosis I spindle [3] and with work by Burbank et al. [4] demonstrating the existence of the minus ends of the microtubules throughout the meiosis I spindle.
These observations lead to a model of spindle assembly in which microtubules are initially nucleated in the region around the chromosomes (possibly by γ-tubulin) and then moved poleward. However, despite the evidence in other female meiotic systems for a role of γ-tubulin in meiosis I spindle assembly and function, the role (if any) of γ-tubulin in the formation of the meiosis I spindle in Drosophila oocytes has remained highly controversial [3], [5].
Drosophila has two genes encoding γ-tubulin: γtub37C and γtub23C [6]. γTub23C is expressed in all somatic tissues once embryos become cellularized [6], [7]. However, in ovaries γTub23C is only expressed in the mitotically dividing germ cells [6]. After meiosis is initiated, γTub37C accumulates rapidly in the oocyte and nurse cells for use during the rapid embryonic cell divisions [6]. In embryos, γTub37C localizes primarily to the centrosomes, but does show some localization over the length of the mitotic spindle and at the midbody [6], [8]. Although γTub37C is present in the female germline, whether it plays a role in spindle formation during meiosis I has been controversial.
Using similar sets of γtub37C mutants, different investigators have obtained highly divergent results with respect to the role of γTub37C in the assembly and function of the first meiotic spindle [8], [9]. Wilson and Borisy [9] examined the effects of a number of γtub37C mutants (including a null allele) on female meiosis I and observed some normal-looking bipolar spindles, leading them to conclude that γTub37C was not essential for either microtubule nucleation or the assembly of the female meiotic spindle. Endow and Hallen [10] reached similar conclusions using a weak loss-of-function allele of γtub37C. However, Tavosanis et al. [8] observed significant defects in both spindle morphology and chromosome arrangement during meiosis I in Drosophila oocytes. Indeed, in the Tavosanis et al. [8] study, ∼80% of oocytes from mothers hemizygous for two null mutations of γtub37C showed abnormal meiotic figures, including chromosomes randomly arranged across the spindle, spindles that were less dense and less uniform then those observed in wild-type oocytes, and spindles that were not focused at the poles [8]. We will show below that these divergent conclusions with respect to the role of γTub37C in spindle assembly were the result of methodological differences in the manner in which oocytes were collected. The data presented here show that γTub37C is indeed required for spindle assembly and function during prometaphase I (the stage primarily studied by Tavosanis et al. [8]) and that the spindle defects are often ameliorated by metaphase I (the stage primarily studied by Wilson and Borisy [9]).
Even the presence of γTub37C in the meiosis I spindle has been highly contentious. Wilson and Borisy [6], Tavosanis et al. [8] and Matthies et al. [11] all failed to detect γTub37C on the meiosis I spindle by indirect immunofluorescence microscopy. The inability to detect γTub37C on the meiosis I spindle leant support to the genetic data suggesting that γTub37C was not required for spindle formation in meiosis I. However, Endow and Hallen [10] have recently demonstrated the localization of an overexpressed and green fluorescent protein (GFP)-tagged version of γTub37C to the microtubules and poles of the meiosis I spindle. Although this observation shows that γTub37C is capable of localizing to the meiosis I spindle when overexpressed, there are numerous examples of proteins that mislocalize when overexpressed [12], [13]. Thus, it remained to be determined whether endogenous γTub37C is a native component of the meiosis I spindle.
To both resolve these controversies and to explore the role(s) of γTub37C in the acentrosomal spindle, we have characterized defects caused by both a novel point mutation in γtub37C, γtub37CP162L, and a null mutation of γtub37C, γtub37C3, during prometaphase I and metaphase I using both fixed and live oocyte methods. We find that both mutations cause spindle and chromosome defects, demonstrating that γTub37C plays important roles in spindle formation, maintenance, and positioning, as well as chromosome alignment and morphology during prometaphase I. Indeed, mutations in γtub37C lead to loss of kinetochore biorientation and altered kinetochore microtubule attachments. Finally, using multispectral imaging we detect endogenously expressed γTub37C on the microtubules of the meiosis I spindle, suggesting that γTub37C acts within the meiotic spindle to execute these essential functions.
We examined Drosophila oocytes carrying either of two alleles of γtub37C (Genbank AY070558.1) to understand its function during meiosis I. The first mutant, γtub37CP162L, was isolated in the course of a screen for EMS-induced recessive female sterile mutants. This mutation resulted in a C to T transition at position 834 that results in a P to L change at amino acid 162 in exon 3 of the γtub37C gene (data not shown). We also examined oocytes carrying a previously characterized early-stop mutation (γtub37C3), that removes the C-terminal 106 amino acids of the 457 amino acid γTub37C protein, over a deletion that removes the entire γtub37C gene (γtub37C3/Df) [9]. The γtub37C3 mutation is a presumed null allele since the truncated protein could not be detected by Western blot [9].
Drosophila females were maintained under two different sets of conditions which yield preparations enriched for either prometaphase I or metaphase I oocytes that were established by Gilliland et al. [14]. For prometaphase I-enriched preparations, mated females were held two to three days post-eclosion on wet yeast paste. Metaphase I-enriched preparations females were collected as virgins and held four to five days on yeast paste in the absence of males.
During prometaphase I, in the majority of wild-type oocytes the autosomal bivalents and X chromosomes are aligned together at the midzone of the tapered, bipolar spindle with the achiasmate 4th chromosomes either out on the spindle (Figure 1A) or associated with the chiasmate chromosomes (Table 1) [15]. However, upon examination of prometaphase I oocytes from γtub37CP162L and γtub37C3/Df females we observed a wide array of aberrant chromosome configurations (Figure 1B–1E, Figure S1A–S1B, Table 1). As exemplified in Figure 1B, in γtub37CP162L mutant oocytes the chiasmate chromosomes often failed to form a single chromosome mass at the spindle midzone, but rather were stretched across the length of the spindle. The most notable feature of Figure 1B is the physical separation of the chiasmate autosomes on the spindle (see arrowheads), a phenomenon that we refer to as autosomal slippage. Autosomal slippage that results in a near total disruption of the overlap between the autosomes on the spindle is not commonly observed in wild-type oocytes [15]. Slippage of the autosomes was observed in 42% of γtub37CP162L mutant oocytes and in 23% of γtub37C3/Df mutant oocytes (Figure 1B, 1D, Table 1). The lower rate of slippage in the γtub37C3/Df mutant oocytes is likely due to the higher rate of severely abnormal chromosome masses in γtub37C3/Df mutant oocytes that are described below (Table 1).
We also observed severe defects in chromosome morphology in mutant prometaphase I oocytes. As exemplified in Figure 1C, the chromosomes from the γtub37C mutant oocytes sometimes appeared rounded and misshapen. Additionally, these chromosome masses did not appear to be properly condensed based on the DAPI staining (Figure 1C). The chromosome morphology defects observed in γtub37C mutant oocytes was rarely observed in wild-type oocytes.
As exemplified in Figure 1E and Figure S1A–S1B, the chromosomes of some mutant prometaphase I oocytes failed to align in any direction and displayed both abnormal invaginations and projections away from the main chromosome mass (Figure 1E, Figure S1A–S1B). This lack of obvious alignment was observed in 27% of the γtub37CP162L and 41% of γtub37C3/Df mutant oocytes (Table 1). Chromosome configurations that appeared more similar to wild-type configurations were observed for a few γtub37CP162L and γtub37C3/Df mutant oocytes (Figure 1F, Figure S1C–S1D, Table 1). Finally, in 18% of γtub37C3/Df oocytes the chromosomes had separated far enough apart not to be contained within the same set of microtubules (Table 1).
Thus, as was originally noted by Tavasonis et al. [8], chromosomes fail to align properly on the prometaphase I spindle in γtub37C mutant oocytes. These phenotypes suggest that γTub37C is involved in properly aligning and orienting the chromosomes on the prometaphase I spindle. Additionally, γtub37C mutant oocytes show defects in chromosome morphology, which had not been described previously. Based on these defects, γTub37C appears to function in regulating chromosome alignment and morphology during meiosis I.
Our ability to recognize chromosomes on the prometaphase I spindle was greatly enhanced by the use of an antibody recognizing histone H3 phosphorylated on serine 10 (phH3S10). Histone H3 serine 10 phosphorylation increases during prophase and peaks at metaphase of mitosis and meiosis (reviewed in [16]). The phH3S10 antibody allowed for unambiguous identification of oocyte nuclei that have progressed to at least prometaphase I even in the absence of normal-looking chromosomes (Figure 1B–1E, Figure S1A–S1B).
While using the phH3S10 antibody in control oocytes, we discovered that this antibody also robustly highlighted the DNA threads that connect achiasmate chromosomes [15]. These threads are hypothesized to be involved in the mechanism by which achiasmate chromosomes can reassociate during their dynamic prometaphase I movements on the meiotic spindle and the subsequent congression of the chromosomes to the metaphase plate during metaphase I [15]. Analyzing DNA threads using DAPI alone is difficult. Often chromosomes will show evidence that threads are present, such as spurs on the ends of the chromosomes, but the full-length thread will be below the level of detection [15]. The phH3S10 antibody allowed for the visualization of complete threads connecting achiasmate chromosomes, such as the thread projecting from the achiasmate 4th chromosomes in Figure 1A in a wild-type oocyte and from both X and 4th chromosomes in the FM7/X oocytes shown in Figure S2.
Using the phH3S10 antibody to examine thread number and morphology in wild-type oocytes, we primarily observed phH3S10 threads connecting well-separated 4th chromosomes or the absence of threads when achiasmate chromosomes were part of the main chromosome mass. phH3S10-staining threads that failed to project toward another achiasmate chromosome were seen in only 11% of wild-type oocytes, and in these cases only one or two very short threads were observed (Table 1). However, in many γtub37C mutant oocytes multiple abnormal threads were observed projecting away from the chromosomes (Figure 1E and Figure S1A–S1B). These aberrant thread-like structures frequently co-localized with α-tubulin (Figure 1E and Figure S1A–S1B). Such aberrant threads were observed in 70% of γtub37C3/Df and 42% of γtub37CP162L mutant oocytes (Figure 1E, Figure S1A–S1B, Table 1) indicating that functional γTub37C is required for normal DNA thread morphology.
In wild-type oocytes at prometaphase I the meiotic spindle has two tapered poles and the spindle is approximately the width of the autosomes on the spindle midzone at its widest point (Figure 1A). Tapered, bipolar spindles were observed in 89% of wild-type oocytes (Table 2).
However, when spindle structure was examined in prometaphase I γtub37C mutant oocytes, we observed either abnormal spindles or the absence of a recognizable spindle in 70% of γtub37CP162L mutant oocytes and 86% of γtub37C3/Df mutant oocytes (Table 2). Less than a third of the spindles in γtub37CP162L mutant oocytes, and even fewer in γtub37C3/Df mutant oocytes, could be classified as tapered and bipolar (Figure 1F, Figure S1C, Table 2), though some of these bipolar spindles showed minor defects in width and microtubule density.
Monopolar spindles were observed in 22% of prometaphase I γtub37CP162L mutant oocytes (Figure 1B, Table 2) and 21% of spindles in γtub37C3/Df mutant oocytes (Table 2). Barrel-like spindles that lacked tapered poles, but still displayed bidirectionality, were observed in 23% of γtub37CP162L mutant oocytes and 25% of γtub37C3/Df oocytes (Figure 1C–1D, Table 2). The width of the spindle was not constant in these barrel-like spindles. In the image of the γtub37CP162L mutant oocyte that is shown in Figure 1C the spindle is approximately the width of the chromosomes, but in some oocytes the barrel-like spindles were narrower than the chromosomes (Figure S1D).
Finally, in 15% of γtub37CP162L and 32% of γtub37C3/Df mutant oocytes the microtubules displayed no clear directionality and simply projected in all directions (Figure 1E, Figure S1A–S1B, Table 2). Typically this microtubule morphology was also associated with the most abnormal chromosome morphologies. Three γtub37CP162L mutant oocytes failed to show any microtubule association with the chromosomes despite robust antibody penetration. The multitude of aberrant spindle structures suggests that γTub37C is required for both formation of a proper prometaphase I spindle and for maintenance of a tapered bipolar spindle.
To investigate the defects in chromosome alignment in γtub37C mutant oocytes, we examined the position of kinetochores using an antibody to Centromere Identifier (CID), the Drosophila CENP-A homolog [17]. In 93% of wild-type oocytes CID localizes to eight foci—four foci on each end of the chromosome mass oriented in opposite directions (Figure 2A, Table S1). One oocyte appeared to be in spindle assembly before kinetochores had bioriented, and one oocyte had five CID foci on a single side of the chromosome mass, most likely indicating an achiasmate chromosome was caught in the process of moving dynamically on the spindle as observed by Hughes et al. [15] (Table S1).
In γtub37C3/Df mutant oocytes only 17% showed eight CID foci in the wild-type configuration (Figure 2B, Table S1). In 51% of oocytes CID foci were maloriented with more than four foci pointing towards a single direction or with foci pointing in more than two directions, indicating a frequent failure to properly biorient kinetochores during prometaphase I (Figure 2C, Table S1). We also observed CID foci that appeared to be oriented towards a single pole (Figure 2D, Table S1) or clustered together in the middle of the chromosome mass suggesting a complete failure of chromosome biorientation (Figure 2E, Table S1). Finally, in 14% of γtub37C3/Df mutant oocytes more than eight CID foci were observed suggesting precocious sister chromatid separation (Figure 2F, Table S1).
Kinetochores were bioriented in only 38% of γtub37CP162L mutant oocytes. In the majority of γtub37CP162L mutant oocytes the CID foci were maloriented, while more than eight CID foci were detected in the remaining oocytes (Table S1). These results clearly illustrate that chromosome kinetochores fail to correctly biorient in γtub37C mutant oocytes and in a few cases sister chromatid cohesion may be lost.
To better understand the defects in kinetochore orientation, we examined the location of CID foci in comparison to α-tubulin. In wild-type oocytes CID foci appear to interact directly with the kinetochore microtubules as shown in Figure 3A. In γtub37C mutants we observed that many CID foci lacked these direct connections to microtubules, despite careful analysis of all sections of the imaged Z stacks (Figure 3B–3F). While we observed some CID foci that appeared to lack clear interaction with any microtubules (Figure 3C–3F), it appeared that other CID foci may potentially have lateral interactions with microtubules (Figure 3B, 3E).
These data suggest that γTub37C plays an important role in allowing microtubules to attach properly to kinetochores. The disruption of microtubule attachments to some kinetochores in γtub37C mutants may result in kinetochores failing to undergo proper biorientation [18], as well as contributing to the observed spindle and chromosome alignment defects.
To better understand the defects observed in the γtub37CP162L mutant oocytes, we examined chromosome movement and spindle dynamics in live oocytes as described in Hughes et al. [15]. Video S1 shows a prometaphase I spindle from a wild-type oocyte. The chromosomes are aligned on the spindle midzone and the spindle is bipolar, consistent with fixed images and previous live-imaging studies [15]. The spindle and chromosomes remain stable for over an hour of observation.
We were able to image nuclear envelope breakdown and successful formation of a spindle-like structure in five γtub37CP162L mutant oocytes. In all five the resulting spindles were either barrel-like or failed to maintain a tapered bipolar shape throughout the period of imaging (Video S2). The ends of the spindles were frequently observed to wave back and forth, and the entire spindle often moved in different directions. The chromosomes also moved quickly into different configurations, at times showing no alignment on the spindle midzone.
For an additional 16 γtub37CP162L mutant oocytes the prometaphase I spindle was already present at the start of live-imaging. These spindles displayed many of the same phenotypes that were observed for the spindles that formed after observation of nuclear envelope breakdown. Video S3 shows one such prometaphase I spindle and Figure 4 shows selected still images from this video. At the beginning of Video S3 the DNA is associated with the spindle midzone but the spindle is abnormally wide and barrel-like (Figure 4A). In Figure 4B and 4C the chromosomes stay associated but change in shape, suggesting movement of the chromosomes with respect to one another. The spindle rotates almost 60° clockwise and individual microtubules move rapidly. By 37.8 minutes after imaging began the microtubules appear to be shed into the cytoplasm and the bivalents begin to separate on the spindle (Figure 4D). At the end of imaging (45.7 minutes), the chromosomes have split into two distinct masses (Figure 4E). The spindle has turned another 30° clockwise and starts to drift out of the field of view. The dynamic movement of individual microtubules and the shedding of multiple microtubules into the cytoplasm is not commonly observed in wild-type oocytes after the completion of nuclear envelope breakdown [15]. These microtubule movements imply that individual microtubule bundles are not maintained in an organized bipolar spindle in γtub37CP162L mutant oocytes, suggesting that γTub37C is required for stabilizing microtubules relative to one another within the spindle.
Spindles in wild-type oocytes typically remain fairly stationary for long periods of imaging. However, spindles in γtub37CP162L mutant oocytes frequently shifted position and orientation quickly which hampered imaging for long time periods. As shown in Video S4, movement of the ends of the thin and long barrel-like spindle eventually leads to the spindle moving toward the top edge of the focal area. After re-centering the spindle within the focal plane, the spindle and the chromosomes within the spindle continue to move rapidly. The spindle becomes progressively thinner during these movements until it eventually dissolves and the moving chromosomes disperse. These changes in spindle position may be due to the rapid movement of the spindle microtubules or the possible role of γTub37C in maintaining the cortical microtubules that anchor the meiosis I spindle.
When examining single time points from live-imaging of γtub37C mutant oocytes most, if not all, of the spindle and chromosome configurations observed in fixed images could be identified. The reason we observe both normal and aberrant spindles in fixed images is that they represent transient intermediates in dynamically changing but structurally flawed spindles. For example, Video S2 shows a spindle that gains and loses tapered bipolarity during the course of imaging. Shortly after nuclear envelope breakdown a tapered bipolar spindle is formed. Around 40.3 minutes the chromosomes begin to spread out across the spindle midzone, followed by the spindle forming into a wide barrel. Despite continued reshuffling of the chromosomes across the spindle midzone the spindle regains tapered bipolarity by 50.9 minutes. The bottom pole of the spindle then widens and refocuses a second time before a bipolar spindle reforms and is maintained for the remainder of imaging.
Due to problems inherent in finding the meiotic spindle, imaging of γtub37C3/Df mutant oocytes proved very difficult. We were unable to acquire any quality recordings of γtub37C3/Df mutant oocytes despite numerous efforts.
The results from live-imaging suggest that γTub37C is involved in bipolar spindle assembly and maintenance as well as chromosome and spindle positioning during prometaphase I. Additionally, live-imaging demonstrates that mutants can both lose and regain wild-type spindle and chromosome morphology. This suggests that the array of phenotypes observed in fixed preparations are not due to the spindle progressively deteriorating during the progression to metaphase I arrest but rather represents different stages of the dynamic spindle recovery and dissolution process.
Fixed preparations of virgin females four to five days post-eclosion are highly enriched for oocytes arrested at metaphase I (Table 3) [14]. In wild-type oocytes the achiasmate chromosomes congress back to the metaphase plate and the chromosomes form a lemon-shaped configuration for metaphase I arrest at stage 14 (Figure 5A) [14]. Chromosomes that had congressed into a lemon-like shape on the metaphase plate were observed in 96% of wild-type oocytes (Table 3).
Despite the abnormal chromosome morphologies observed during prometaphase I, the chromosomes in 62% of γtub37CP162L mutant oocytes successfully congressed to the metaphase plate to form lemon-like DNA structures (Figure 5B, Table 3). However, 8% of γtub37CP162L mutant oocytes displayed chromosome masses that had split into 2 or more parts (discussed below). In 18% of γtub37CP162L mutant oocytes chromosomes were associated but formed irregular, non-lemon-shaped configurations (Table 3).
Chromosome morphology at metaphase I arrest was also examined in oocytes from virgin γtub37C3/Df mutant females. Metaphase I DNA configurations were observed in 35% of γtub37C3/Df mutant oocytes, demonstrating that chromosomes can successfully congress to the metaphase plate in some oocytes lacking γTub37C (Figure 5E, Table 3). For γtub37C3/Df mutant oocytes, we noted a propensity for the chromosome mass to split into multiple pieces. This phenotype was observed for 56% of oocytes (Figure 5C–5D, Table 3). In some oocytes the chromosome masses were near each other (Figure 5D) but in many others the chromosome masses were dispersed throughout the cytoplasm (Figure 5C). The phH3S10 antibody facilitated the identification of all the chromosomes, as some chromosomes were not associated with clear spindles (See Figure 5C for an example).
Although chromosomes successfully congressed to the metaphase plate in most γtub37CP162L mutant oocytes, they failed to properly orient in preparation for segregation at anaphase I. Using Fluorescent In-Situ Hybridization (FISH) probes that recognize the X and 4th chromosomal heterochromatin, we observed in wild-type oocytes that the X and 4th chromosomes were properly positioned on each half of the chromosome mass in 100% of oocytes (Figure 6A), which is consistent with previous studies [14]. This configuration was observed in only 28% of γtub37CP162L mutant oocytes (Figure 6B). In 40% of γtub37CP162L mutant oocytes the 4th chromosomes were associated on the same side of the chromosome mass while the Xs were properly segregated (Figure 6C). In 28% of oocytes the X chromosomes were associated on the same side of the chromosome mass while the 4th chromosomes were segregated correctly (Figure 6D). Finally, in 4% of γtub37CP162L mutant oocytes the X and 4th chromosomes were on opposite ends of the chromosome mass (Figure 6E). These results show that while γTub37C is not involved in chromosome congression, it does play a role in ensuring that the chromosomes are packaged correctly when they do congress. The aberrant threads observed during prometaphase I were mostly absent during metaphase I in γtub37C mutant oocytes (15% for γtub37CP162L and 6% for γtub37C3/Df, Table 3). Whether the aberrant threads were resolved or simply packaged into the chromosome mass during congression is unknown.
In wild-type oocytes the spindle becomes shorter as the chromosomes congress, which results in a small, bipolar, tapered spindle, similar to the spindle in mouse oocytes [14], [19]. Bipolar spindles were observed in 87% of wild-type oocytes from four to five day-old virgin wild-type females (Table 4). While the γtub37CP162L mutation caused spindle abnormalities during prometaphase I in 70% of oocytes (Table 2), many mutant oocytes were able to recover to form spindles resembling wild-type metaphase I-arrested spindles. Short, tapered, bipolar spindles were observed in 54% of γtub37CP162L and 8% of γtub37C3/Df mutant oocytes (Table 4). In 35% of γtub37C3/Df oocytes the spindle was reduced to a large microtubule bundle projecting from each end of the chromosome mass (Table 4). This spindle phenotype was also observed in 21% of γtub37CP162L mutant oocytes (Figure 5B, Table 4). In 33% of γtub37C3/Df oocytes and 5% of γtub37CP162L mutant oocytes, no apparent microtubules were associated with the chromosomes. This phenotype was often associated with chromosome masses that had split apart. The remaining γtub37CP162L and γtub37C3/Df mutant oocytes displayed spindles that were abnormal in ways similar to those from prometaphase I-enriched preparations, such as barrel-like spindles, monopolar spindles, and microtubule aggregations lacking directionality (Table 4).
These results suggest that γTub37C plays a less pivotal role in spindle maintenance and chromosome positioning during metaphase I. The ability of chromosomes to congress to the metaphase plate in many γtub37C mutant oocytes suggests that chromosome congression during metaphase I is mediated by a mechanism that is different from chromosome positioning during prometaphase I and that does not require a robust prometaphase I spindle.
In Drosophila oocytes the spindle pole protein D-TACC is required for maintaining the bipolarity of the meiosis I spindle [20]. Spindles fail to maintain bipolarity in d-tacc mutant oocytes [20]. In Caenorabditis elegans embryos, γ-tubulin plays a role in localizing TACC (called TAC-1 in C. elegans) to the spindle poles [21]. Drosophila oocytes with a mutation in the Cdc2 subunit, Cks/Suc1, display defects in chromosome alignment and spindle morphology as well as mislocalization of D-TACC [22]. These results suggest that D-TACC mislocalization can also cause spindle defects. We examined whether D-TACC localization was disrupted in γtub37CP162L mutant oocytes. Using an antibody to D-TACC, Cullen and Ohkura [20] reported that D-TACC primarily localizes to the poles of the meiosis I spindle with diffuse staining of spindle microtubules in a few oocytes. In contrast, we observed this additional diffuse localization in the majority of wild-type prometaphase I and metaphase I spindles we examined (Figure 7A and 7B). Moderate to strong diffuse spindle and polar staining was observed in 23/36 (64%) prometaphase I and 34/45 (76%) metaphase I wild-type oocytes. Weak diffuse spindle and polar staining was observed in an additional 7 (19%) and 4 (9%) prometaphase I and metaphase I spindles, respectively. In the remaining 6 (17%) prometaphase I and 5 (11%) metaphase I wild-type oocytes D-TACC could either not be detected on the spindle or the background staining was too high to assign a definitive localization pattern. One metaphase I oocyte showed staining primarily to the poles and one showed patchy D-TACC staining (described below).
In γtub37CP162L mutant oocytes, D-TACC localization was frequently abnormal. In 38/63 (60%) prometaphase I oocytes, the D-TACC localization was patchy with large, bright foci in some regions and no staining in other regions of the spindle (Figure 7C). In 8/63 (13%) prometaphase I γtub37CP162L mutant oocytes, small, punctate spots of staining were observed along only part of the spindle, often close to the chromosomes (Figure 7D). In 12/63 (19%) of prometaphase I γtub37CP162L mutant oocytes, D-TACC was absent or highly reduced on the spindle (Figure 7E). Only 5/63 (8%) prometaphase I oocytes displayed the diffuse microtubule and polar D-TACC localization observed in wild-type oocytes. For the 49 γtub37CP162L mutant oocytes examined from metaphase I-enriched preparations, 12 (24%) displayed patchy D-TACC localization, 6 (12%) displayed punctuate localization, 29 (59%) showed reduced or absent staining (Figure 7F), and only 2 (4%) displayed D-TACC localization similar to wild-type. Since D-TACC is required for the proper formation of spindle poles, mislocalization of this spindle assembly factor could contribute to the lack of defined spindle poles in γtub37C mutant oocytes.
During our attempts to examine living γtub37C mutant oocytes, we noticed that the oocyte nucleus was mislocalized within the cytoplasm with respect to the dorsal appendages. The mislocalization of the oocyte nucleus was not surprising since cortical microtubules are required for nuclear positioning [23]. Additionally, mutations in γ-tubulin ring components (γTURCs) have been reported to affect bicoid RNA localization to the anterior cortex of Drosophila oocytes, which is a process dependent on microtubules [24].
In DAPI-only fixed preparations, 98% of wild-type oocyte nuclei were localized to the anterior third of the oocyte near the nurse cells during stages 11 and 12 (Table S2). In contrast, oocyte nuclei were mislocalized in 22% of γtub37CP162L mutant oocytes during stages 11 and 12 (Table S2). In γtub37C3/Df mutant oocytes, only 65% of oocyte nuclei were in the anterior third of the oocyte while 26% were located in the middle of the oocyte and 9% of oocyte nuclei were located in the posterior third of the oocyte (Table S2).
In wild-type stage 13 and 14 oocytes, 100% of the oocyte nuclei were located near the dorsal appendages in the anterior one-third of the oocyte (Table S2). Meanwhile, in γtub37CP162L mutant oocytes, 60% of the oocyte nuclei were located in the anterior portion of the oocyte and 40% were located in the middle or posterior of the oocyte (Table S2). In γtub37C3/Df mutant oocytes, only 37% of the oocyte nuclei were in the anterior third of the oocyte at stages 13 and 14 (Table S2). In 43% of γtub37C3/Df mutant oocytes, the nucleus was in the middle third of the oocyte and in 20% the oocyte nucleus was in the posterior portion of the oocyte (Table S2). Our data suggest that γTub37C plays a role in nuclear positioning, most likely by affecting the cortical microtubules required to stabilize the oocyte nuclei in the anterior part of the oocyte during prophase I and after spindle formation.
As noted above, multiple laboratories have failed to detect endogenous γTub37C on the meiosis I spindle [6], [8], [11], [25]. More recently, Endow and Hallen [10] reported that an overexpressed transgenic GFP-tagged γTub37C protein localized to a subset of meiosis I spindles. This experiment had the caveat that the γTub37C-GFP construct was driven not by the genomic γtub37C promoter but rather by the ncd promoter, and endogenous γTub37C was still present in these oocytes. Thus, the γTub37C-GFP construct was likely not present at endogenous levels, raising the possibility that the localization of γTub37C-GFP does not represent the localization of endogenous γTub37C. In order to address these issues and demonstrate the presence of endogenous γTub37C on the meiosis I spindle, we re-examined γTub37C localization using multi-spectral imaging.
We examined the localization of endogenous γTub37C using an antibody raised against the C-terminal 17 amino acids of γTub37C (DrosC) [26]. Processing and acquisition of the images was optimized for the highest possible detection efficiency of γTub37C staining and the maximum possible rejection of signal from other fluorescent labels and autofluorescence. The method of choice for these applications is multispectral imaging [27], [28], [29]. This method provides both high signal-to-noise imaging of multiple fluorophores, as well as residuals-based verification of fitting quality. This allows for detection and removal of even small amounts of bleedthrough or autofluorescence in confocal images. In this case, it permits accurate determination of the presence or absence of γTub37C in the company of α-tubulin, which is in much higher abundance.
We observed γTub37C antibody localizing to all the wild-type meiosis I spindles examined while Endow and Hallen [10] failed to detect γTub37C-GFP on the meiosis I spindle in some oocytes (Figure 8A–8C). Rather than staining predominantly at spindle poles with weaker staining on the remaining spindle as is observed in embryos [6], [30], a uniform signal was observed over the entire spindle for both prometaphase I (Figure 8A–8B) and metaphase I spindles (Figure 8C). This staining is similar to γ–tubulin staining in mitotic and meiotic cells in other systems [31]. Indeed, γTub37C co-localized with the α-tubulin antibody signal, although γTub37C localization was sometimes seen to extend slightly past the α-tubulin signal to form more defined spindle poles (Figure 8B).
To ensure that the DrosC anti-γTub37C antibody is specific for γTub37C we examined γtub37C3/Df mutant oocytes. We focused our analysis on those oocytes that formed a chromatin-associated microtubule structure to rule out the possibility of failing to detect γTub37C simply due to lack of a “spindle.” All γtub37C3/Df mutant oocytes examined failed to show detectable DrosC antibody staining on the spindle (Figure 8D–8E). This result suggests that the DrosC antibody is specific for the C-terminus of γTub37C since the γtub37C3 allele results in a 160 amino-acid C-terminal truncation of the γTub37C protein. These data clearly demonstrate, contrary to previous studies using other γTub37C antibodies, that endogenous γTub37C is present on the meiosis I spindle of Drosophila oocytes.
We also examined the localization of γTub37C in γtub37CP162L mutant oocytes and failed to detect γTub37C on the meiosis I spindle in all γtub37CP162L mutant oocytes examined (Figure 8F). Based on results from Western blot analysis, γTub37C does appear to be expressed in γtub37CP162L mutant ovaries (Figure S3). While the γtub37CP162L mutation appears to be a strong loss-of-function allele, it causes phenotypes that are slightly weaker at prometaphase I and noticeably weaker in terms of the split chromosome mass phenotype at metaphase I compared to the null allele (γtub37C3). For these reasons, we expected to see only a reduced level of γTub37C on the meiotic spindle. A small amount of γTub37C may be present on the meiosis I spindle in γtub37CP162L mutant oocytes, but it may be below our level of detection. Another possibility is that γTub37C has a function not directly associated with the spindle microtubules to regulate meiosis I, and the γtub37CP162L mutation does not completely abrogate this function.
Our analysis both confirms the conclusion of Tavosanis et al. [8] that γTub37C is required for the organization of the female meiotic spindle and extends that conclusion in several very important ways. First, γtub37C mutant oocytes show strong defects in kinetochore biorientation during prometaphase I. Indeed, in γtub37C mutant oocytes many kinetochores appear to lack typical kinetochore microtubule attachments, suggesting that γTub37C plays an essential role in initiating or maintaining the kinetochore microtubule attachments that are required for properly positioning the chromosomes on the meiotic spindle. Our demonstration of a role for γTub37C in mediating kinetochore microtubule interactions in meiosis is consistent with the observation by others that in γ-tubulin depleted S2 cells subjected to cold-induced microtubule depolymerization, kinetochore-driven microtubule regrowth is delayed [32]. Additionally, in HeLa cells the γ-tubulin Ring Complex is recruited to unattached kinetochores and is required for nucleation of kinetochore microtubules [33]. These studies suggest that γ-tubulin may play a role in kinetochore microtubule nucleation during acentriolar meiosis similar to that in mitosis.
Second, although the defects in spindle assembly exhibited by γtub37C mutants are first observed soon after nuclear envelope breakdown and remain severe throughout prometaphase I, normal spindle morphology and chromosome alignment can be lost and regained throughout prometaphase I. Indeed, spindle microtubules in living γtub37CP162L mutant oocytes often appeared to be rapidly moving and microtubule bundles were frequently seen being shed into the cytoplasm. These observations suggest that γTub37C plays an important role in stabilizing existing microtubules within the spindle. The visualization of spindles splitting apart and the subsequent dissociation of the chromosomes from the spindle during prometaphase I by live imaging suggests a mechanism for the split chromosome mass phenotype that is often observed in fixed images.
Third, and quite surprisingly, many metaphase I arrested oocytes appear to partially or even fully recover normal spindle and chromosome morphologies, but the chromosomes fail to orient correctly in the metaphase I chromosome mass. The fact that the highly morphologically abnormal prometaphase I spindles in γtub37C mutant oocytes can nonetheless progress to form metaphase I spindles that are relatively normal in appearance may reflect the existence of redundant mechanisms that can facilitate spindle assembly. The existence of such redundant mechanisms of spindle assembly is supported by the observation that when γ-tubulin is decreased in Drosophila mitotic cells, spindle assembly is delayed but a mitotic spindle does eventually form [34], [35].
Fourth, we also uncovered a previously unidentified role for γTub37C in nuclear positioning. The spindles in living γtub37CP162L mutant oocytes rapidly changed position and orientation within the cytoplasm. Additionally, fixed preparations showed that the oocyte nucleus was sometimes mislocalized to the middle and posterior portion of the oocyte. Cortical microtubules likely play a role in maintaining spindle position within the cytoplasm and a role for γTub37C in regulating these microtubules seems probable [23]. A similar defect in spindle positioning was observed in live oocytes carrying a mutation in ncd, a gene encoding a kinesin motor protein required for bundling the microtubules of the meiotic spindle [11], [23].
Finally, we observed two previously undescribed defects in chromatin morphology in γtub37C mutant oocytes at prometaphase I. Chromosomes at this stage were often obviously morphologically abnormal and thread-like chromatin projections emanating along microtubules were frequently observed. The observed disruption in chromatin morphology, as well as the potential disruption of the chromatin threads that normally connect achiasmate chromosomes [15], may well underlie the failure of chromosomes to properly package into the chromosome mass at metaphase I in γtub37CP162L mutant oocytes. Whether the unusual threads and chromosome morphology defects are an indirect effect of the abnormal spindles or whether γTub37C plays a more direct role in mediating chromosome morphology remains to be elucidated. Abnormal movement of the chromosomes due to a lack of kinetochore microtubules could potentially result in homologs moving so far apart that the chromatin threads sever or that the threads from different chromosomes become entangled and pulled out when chromosomes move around aberrantly.
While consistent with the observations of Tavosanis et al. [8] and Jang et al. [36], our conclusion that γTub37C is required for the assembly and function of the meiosis I spindle conflicts with those of Wilson and Borisy [9] and Endow and Hallen [10]. Both groups concluded that γTub37C did not play an essential role in spindle assembly and maintenance during meiosis I. We argue that there are two major causes for this discrepancy. First, Wilson and Borisy [9] and Endow and Hallen [10] based their conclusions on the observation of apparently normal spindles in some mutant oocytes. The formation of even a few bipolar spindles in γtub37C mutant oocytes, even for those homozygous for the weak allele of γtub37C used by Endow and Hallen [10], led these authors to conclude that γTub37C must not be essential for spindle formation. However, we have shown above that such apparently normal spindles are transient intermediates in a process of highly dysfunctional spindle assembly. Second, while we examined prometaphase I and metaphase I oocytes separately, Wilson and Borisy [9] mainly observed newly laid eggs or activated oocytes, often from virgin females. These preparations would be highly enriched in metaphase I-arrested oocytes and later stages of meiosis. Since many of the spindle defects observed in prometaphase I are partially or fully rectified by metaphase I, the failure to observe spindle defects in a population of oocytes enriched for metaphase I figures is not surprising. Moreover, although metaphase I appeared relatively wild-type in γtub37C mutant oocytes based on DAPI and α-tubulin staining, FISH revealed that chromosome alignment was still defective within the chromosome mass.
Based on the data presented above, we suggest a speculative model for the role of γTub37C in meiosis I. We propose that during meiosis I in oocytes, γTub37C may be controlling spindle assembly and maintenance through several different mechanisms. As outlined in Figure 9, γTub37C localizes to the microtubules in wild-type oocytes. Loss of γTub37C would result in changes in microtubule nucleation and stability. The kinetochore microtubules required for aligning and orienting the kinetochores would be especially affected by these changes. According to our model, these defects would cause spindles to have abnormal morphology, which explains our frequent observation of extremely morphologically abnormal spindles at prometaphase I and metaphase I spindles that appear to be two thick microtubule bundles (see Figure 5B). The mislocalization of the spindle pole component D-TACC in γtub37C mutant oocytes would likely further disrupt spindle morphology.
Decreased kinetochore microtubules in γtub37C mutant oocytes would result in a failure of chromosomes to attach to opposite spindle poles and properly biorient, explaining the chromosome alignment and co-orientation defects that we have observed. These defects in spindle assembly would likely also inhibit the ability of other proteins, such as the Ncd kinesin, to bundle microtubules together, which would further impair spindle assembly and chromosome alignment [37], [38]. As predicted by such a hypothesis, we observed clear second-site noncomplementation between an allele of ncd and a small deficiency uncovering γtub37C in terms of defects in spindle structure, such as non-tapered spindle poles. Such defects were observed in 56% of these doubly heterozygous oocytes. Spindle defects were also observed in 38% of oocytes heterozygous for both γtub37CP162L and ncd mutations (data not shown). Spindle defects were observed in only 11% of ncd heterozygotes, 5% of γtub37CP162L heterozygotes, and 10% of oocytes heterozygous for the deficiency uncovering γtub37C. These results suggest that some of the defects we observe in γtub37C mutant oocytes could be partially mediated by an impairment (perhaps secondarily) in the ability of Ncd to properly function on the microtubules.
Our model also proposes that the defects created by the loss of γTub37C would also impede the functioning of Nod, a chromokinesin-like protein whose plus-end polymerization function is required for the polar ejection force [39], [40], [41]. Indeed, we can imagine that the impairment of Nod function might lead to the defects in chromatin morphology. Such a proposal is consistent with our observation that chromosomes on the most aberrant spindles typically displayed the most severe morphology defects. Abnormal microtubule bundles, impaired attachment of Nod to the chromosomes arms, and a lack of proper kinetochore microtubule attachments would lead to a cascade of defects that affect various aspects of chromosome and spindle structure.
In summary, our analysis of staged mutant oocytes and sophisticated microscopy demonstrate that γTub37C is present in the meiosis I spindle of Drosophila oocytes and plays important roles in spindle assembly, maintenance and positioning, as well as in chromosome positioning, orientation and morphology. Multispectral imaging allowed for detection of endogenously expressed γTub37C in the meiosis I spindle. Furthermore, a point mutation that disrupted localization caused severe spindle defects, strongly suggesting that correct localization of γTub37C to the spindle is necessary for this role. γ-Tubulin also appears to play an important role during meiosis in mammalian oocytes and knock down of γ-tubulin by siRNA in mouse oocytes leads to chromosome misalignment and changes in spindle structure [3], [5]. Our work shows that Drosophila can be used as a model for understanding the function of γ-tubulin in acentriolar bipolar spindle assembly.
Flies were maintained on standard food at 25°C. Wild-type flies were yw; pol. γtub37CP162L flies were yw; γtub37CP162L and the mutation was generated by EMS mutagenesis in the Hawley laboratory (see below). γtub37C3/CyO and w1118; Df(2L)Exel6043, P{XP-U}Exel6043/CyO (deficiency uncovering γtub37C) were obtained from the Bloomington Stock Center and γtub37C3/Df flies were created by crossing the two stocks.
The γtub37CP162L mutation was isolated in the course of a screen for EMS-induced recessive female sterile mutants. Although females homozygous for the γtub37CP162L mutation exhibited complete sterility, homozygous males were unaffected. γtub37CP162L homozygous females laid externally wild-type looking eggs that failed to hatch. Embryos from homozygous γtub37CP162L mothers arrested with one or a few spindle-like structures (data not shown). The spindles were often large, wide, and had chromosomes or chromosome fragments distributed across the spindle structure rather then being aligned on the metaphase plate (data not shown).
Deficiency mapping using the sterility phenotype narrowed the location of the γtub37CP162L mutation to region 37C-D, which includes γtub37C. By sequencing the γtub37C gene from γtub37CP162L homozygous flies we identified a C to T transition at position 834 that results in a P to L change at amino acid 162 in exon 3. A complementation test using the γtub37C3 null mutation confirmed the sterility of γtub37CP162L was due to the mutation in γtub37C.
For sequencing the γtub37C gene, genomic DNA was prepared from single flies by standard protocol [42]. The following gene primers were used for amplification of the γtub37C gene and for sequencing: 5′ CCTACCTCGTTCAGAGTTATTT, 5′ TAATGACTTCCACTTCCATC, 5′ TGGTCTTTCGAACGCTTGTC, 5′ CCACCGCCGTGCTTGGAGAG, 5′ GACAAGCGTTCGAAAGACCA, and 5′ CTCTCCAAGCACGGCGGTGG.
Oocytes were fixed by one of two methods. For all samples except one replicate of the D-TACC prometaphase I experiments, ovaries were dissected from yeasted females in 1× Robb's media (55 mM sodium acetate, 8 mM potassium acetate, 20 mM sucrose, 2 mM glucose, 0.44 mM MgCl2, 0.1 mM CaCl2 and 20 mM HEPES, pH 7.4) containing 1% Bovine Serum Albumin (BSA). For prometaphase I-enriched preparations, females were yeasted for 2–3 days with males [14] (see Table 1 for level of enrichment). For metaphase I-enriched preparations, virgin females were yeasted for 4–5 days post-eclosion [14] (see Table 3 for level of enrichment). Ovaries were fixed using a 1× fix buffer (100 mM potassium cacodylate, 100 mM sucrose, 40 mM sodium acetate and 10 mM EGTA) and 8% formaldehyde (Ted Pella) for 4–5 minutes. After fixation oocytes were washed three times in PBS plus 0.1% triton-X-100 (PBST) and vitelline membranes were removed manually using the rough end of two frosted slides. After further washing with PBST, oocytes were blocked with 5% Normal Goat Serum (NGS) for at least one hour. Oocytes were incubated overnight in primary antibodies in PBST and 5% NGS at 4°C. After several washes with PBST, oocytes were incubated at room temperature for 4–5 hours or 4°C overnight with secondary antibodies in PBST and 5% NGS. 1.0 µg/mL 4′6-diamididino-2-phenylindole (DAPI) or 2.5 µg/mL Hoechst 34580 (Invitrogen) DNA dye was added during the last 10–20 minutes of incubation. Oocytes were washed three times in PBST and then mounted in ProLong Gold (Invitrogen).
For fixed preparation using only DAPI, ovaries were fixed under the same conditions as above and washed three times in PBST. Ovaries were teased using gentle pipetting and 2.0 µg/mL DAPI was added for 20 minutes. After three washes in PBST oocytes were mounted in ProLong Gold (Invitrogen).
To ensure the mislocalization of D-TACC was repeatable under different fixation conditions, the prometaphase I experiments with the anti-D-TACC antibody were replicated using the Buffer A protocol described in McKim et al. [43]. Females were dissected in 1× Robb's media plus 1% BSA and fixed for 10 minutes in 1× Buffer A (15 mM PIPES, pH 7.4, 80 mM KCl, 20 mM NaCl, 2 mM EDTA, 0.5 mM EGTA), 1 mM DTT, 0.5 mM spermidine, 0.15 mM spermine and 4% paraformaldehyde. Samples were washed three times in a Buffer A solution lacking formaldehyde but containing triton-X-100. Vitelline membranes were removed manually using the rough end of two frosted slides and washed three more times. Oocytes were blocked for 30 minutes in a Buffer A solution containing 10% NGS. Antibodies were spun for 10 minutes at 4°C in the same solution and then added to oocytes overnight at 4°C. Oocytes were washed in a Buffer A solution containing 0.2% BSA and then incubated with secondary antibodies in a Buffer A solution with 10% NGS for 3–5 hours. DAPI was added for 10–15 minutes and samples were washed in a Buffer A solution before mounting in ProLong Gold.
The primary antibodies were used at the following concentrations: rat anti-α-tubulin (AbD Serotec, NC 1∶250), mouse anti-α-tubulin DM1a (Sigma-Aldrich 1∶100), rabbit anti-γtub37C ([26] 1∶100), rabbit anti-D-TACC ([44] 1∶250), rat anti-CID ([45] 1∶1000) and rabbit anti-phosphorylated-histone H3 at serine 10 (Millipore 1∶500 or 1∶250). Secondary Alexa-488 or Alexa-555 conjugated antibodies (Molecular Probes) were used at a dilution of 1∶400.
FISH was performed as described by Xiang and Hawley [46], with the following modifications. Incubation and hybridization temperature was 30°C and annealing temperature was 91°C. Part of the 359-bp repeat on the X chromosome conjugated to Alexa Fluor 488 and the AATAT repeat primarily on the 4th chromosome conjugated to Cy3 were chosen as probes as previously described [47], [48].
For fixed experiments not requiring spectral unmixing, the DeltaVision microscopy system was used (Applied Precision, Issaquah, WA). The system is equipped with an Olympus 1×70 inverted microscope and high-resolution CCD camera. The images were deconvolved using the SoftWoRx v.25 software (Applied Precision).
For spectral unmixing experiments images were acquired with an LSM-710 confocal laser scanning microscope (Carl Zeiss Microimaging, Inc., Jena, Germany) using either a 40× 1.3 NA Plan-Neofluar or a 40× 1.3 NA Plan-Apochromat oil objective. Images were collected using a pinhole of one airy unit and a pixel dwell time of 1.6 µs. Each line was averaged eight times in the acquisition. Z stacks were obtained at 0.5 µm intervals. All antibody imaging was performed using the spectral detection channel of the microscope with the MBS 488/561 excitation dichroic with 9.8 nm resolution and collecting from 494 to 660 nm. All focusing and zooming operations were performed with excitation at 561 nm only (for visualization of the Alexa Fluor 555 α-tubulin staining) to avoid photobleaching of the weak Alexa Fluor 488 (AF488) staining of γ-tubulin. All confocal imaging was performed with 488 nm excitation only to maximize the AF488 signal relative to AF555, as we found that 488 nm excitation was sufficient to observe the α-tubulin-AF555 signal. Reference spectra for secondary antibodies were obtained every day that imaging was performed under identical imaging conditions except with small gain and laser power changes. Variation in reference spectra between the two objectives used from day to day was negligible. The reference spectra were obtained using the secondary antibodies mounted in ProLong Gold (identical to biological sample mounting). Hoechst or DAPI imaging was accomplished with 405 nm excitation and a MBS 405 excitation dichroic and a 415–480 bandwidth collection channel. Hoechst or DAPI imaging was done under identical zoom and Z stack settings to the visible light imaging for each sample.
All image processing was accomplished using ImageJ functionality as well as custom-written plugins for binning and spectral unmixing. Spectral images were linear unmixed using standard linear least squares algorithms. Residuals images were generated for each wavelength at each Z position. These residuals were carefully inspected, with close attention paid to channels containing AF488 to ensure that signal from those channels conformed to the expected spectrum for AF488. In all images, these residuals were completely random in the AF488 channels and showed minimal variation in AF555 channels. Maximum projections of selected slices containing spindles were performed for presentation purposes for both the unmixed images as well as the Hoechst or DAPI images. Maximum projection of the entire collected Z stack was avoided due to non-specific staining above and below the spindle.
Live-imaging was performed as describes in Hughes et al. [15]. Briefly: approximately stage 13 oocytes were dissected from ovaries of 2–3 day-old, well-fed adult females and the oocytes were aligned in halocarbon oil 700 (Sigma) in a well made on a no. 1 ½ coverslip. Oocytes were injected using standard micro-injection procedures with an approximately 1∶1 ratio of bovine or porcine rhodamine-conjugated tubulin minus glycerol (Cytoskeleton) and Quant-iT OliGreen ssDNA Reagent (Invitrogen) diluted 0.7 fold with water. After injection, oocytes were covered with a piece of YSI membrane.
The well slides were placed on a temperature-controlled bionomic controller (Technology, Inc) set at 23.5°C. Oocytes were imaged using an LSM-510 META confocal microscope (Zeiss) with a Plan-APO 40× objective (1.3 NA) with a zoom of 2–2.5 or an alpha plan-fluar 100× (1.4 NA) with a 1.5 zoom. Images were acquired using the AIM software v4.2 by taking a 10 series Z stack at 1 micron intervals with 20 seconds between acquisitions which resulted in a set of images approximately every 45 seconds. Images were transformed into 2D projections and concatenated into videos using the AIM software v4.2.
For each genotype, 50 pairs of ovaries from 2–3 day-old, yeast-fed females were dissected in 1× PBS and homogenized in 50 µL of cold lysis buffer containing 150 mM NaCl, 50 mM Tris (pH 6.8), 2.5 mM EDTA, 2.5 mM EGTA, 0.1% Triton-X, and protease inhibitor cocktail (Sigma-Aldrich). Ovary lysates were cleared by centrifugation twice at 14,000 rpm for 15 minutes at 4°C. Equivalent volumes of ovary lysates per genotype were combined with 2× SDS sample buffer, boiled for five minutes, and the solubilized proteins were analyzed by Western blot using standard techniques. The primary antibody used for Western blot was rabbit anti-DrosC γTub37C at a dilution of 1∶500. Immunoreactivity was detected using an alkaline phosphatase-conjugated rabbit secondary antibody (Jackson ImmunoResearch) and the nitroblue tetrazolium and 5-bromo-4-chloro-3-indolyl phosphatase (NBT/BCIP, Invitrogen) reagents.
|
10.1371/journal.pntd.0002130 | Methodology of Clinical Trials Aimed at Assessing Interventions for Cutaneous Leishmaniasis | The current evidence-base for recommendations on the treatment of cutaneous leishmaniasis (CL) is generally weak. Systematic reviews have pointed to a general lack of standardization of methods for the conduct and analysis of clinical trials of CL, compounded with poor overall quality of several trials. For CL, there is a specific need for methodologies which can be applied generally, while allowing the flexibility needed to cover the diverse forms of the disease. This paper intends to provide clinical investigators with guidance for the design, conduct, analysis and report of clinical trials of treatments for CL, including the definition of measurable, reproducible and clinically-meaningful outcomes. Having unified criteria will help strengthen evidence, optimize investments, and enhance the capacity for high-quality trials. The limited resources available for CL have to be concentrated in clinical studies of excellence that meet international quality standards.
| Solid evidence is needed to decide how to treat conditions. In the case of cutaneous leishmaniasis, the diversity of clinical conditions, combined with the heterogeneity and weaknesses of the methodologies used in clinical trials, make it difficult to derive robust conclusions as to which treatments should be used. There also other imperatives - ethical (not exposing patients to treatments that cannot be assessed adequately) and financial (optimize use of limited resources for a neglected condition). This paper is meant to provide clinical investigators with guidance for the design, conduct, analysis and report of clinical trials to assess the efficacy and safety of treatments of this condition.
| It is important to harmonize and improve clinical trial methodology for cutaneous leishmaniasis (CL); currently, treatment options are few and the quality of the supporting evidence is generally inadequate, making the strength of recommendations for the treatment of this disease inadequate.
To improve on the case management and control of CL, better treatment modalities with reliable evidence of the efficacy, safety, tolerability and effectiveness is required. High-quality clinical trials are essential to determine which therapeutic interventions can confidently be recommended for treating which form of CL. Today, this is unfortunately not the case in numerous instances.
The inadequacies of trials of different treatments of CL has been documented by two WHO-supported Cochrane systematic reviews [1], [2] which included 97 randomized controlled trials on treatments for Old World and American CL. They revealed critical issues related to the methodological quality of the design and reporting of these clinical trials, which make it difficult to compare results, meta-analyse the studies, and draw generalizable conclusions. Weaknesses ranged from the inadequacy of study design (including appropriate controls, endpoints, outcome measures, follow-up times), execution (randomization, allocation concealment, blinding), analyses and reporting (e.g. use of disparate endpoints) [3]. They also found a large number of trials that did not meet basic criteria, and could not be included in the analyses.
This makes a highly compelling and cogent case for defining and harmonizing elements related to the design, conduct, analysis, clinical relevance, and reporting of trials, and ultimately study acquiescence by regulatory agencies. Improving the quality of studies and harmonizing protocols will make meta-analysis more informative and thus strengthen evidence for recommendations on treatment and case management. Furthermore, conducting inadequate trials may lead to inappropriate conclusion, is both unethical and an inefficient use of the limited resources available for research into this neglected disease.
As heterogeneity is an inherent feature of CL (reflecting the variety of species and manifestations), there are obvious challenges in designing and interpreting trials to assess interventions for CL which will allow deriving generalizable results and recommendations.
The objective of this paper is twofold:
This paper focusses on CL trial-specific issues; it only touches upon more general aspects of clinical trial conduct, which are extensively addressed in a number of relevant papers and documents. For instance the Global Health Trials website [4] offers several resources including a trial protocol tool [5].
The collective name of CL comprises several manifestations caused by different Leishmania species in the Old and the New World (OWCL and NWCL) and clinical trial methodology should be adapted to this spectrum of conditions. CL is caused by organisms of the L. mexicana complex and Viannia sub-genus (L. braziliensis and L. guyanensis complex) in the New World and L. major, L. tropica and L. aethiopica in the Old World. L. infantum in both Worlds and L. donovani in the Old World can also cause CL. The wide spectrum of clinical manifestations, natural histories and responses to treatment observed in CL patients is accounted for by the combination of parasite's intrinsic differences and patient's genetic diversity.
The time required for natural cure (“self-healing”) is poorly defined and varies widely; it is generally accepted that lesions caused by L. mexicana in the New World and L. major in the Old World heal spontaneously in a time varying from a few weeks to several months in the majority of patients – except new foci (where the disease tends to be aggressive and self-healing is uncommon), and as opposed to other species (where spontaneous healing barely occurs or requires years). Bacterial super-infections are also frequent and can interfere with healing.
The natural history of the disease must be accounted for when designing a clinical trial. Good knowledge of the disease characteristics at the trial site is essential; it is not possible to extract generalizable data from the published literature. For instance, when considering the placebo arms of randomised controlled trials (RCTs) from the Cochrane systematic review of OWCL1, 3-month cure rates for L. major were 21% in Saudi Arabia and 53% in Iran with oral placebo. With a topical placebo, they varied from 13% to 63% at 2 months in Iran and were 61% in Tunisia at 2.5 months. For L. tropica, cure rates were 0%–10% with oral placebo. In the New World, the information is scarce and more variable, ranging from 0% cure rate at one month in Panama [6] to 37% at 12 months in Colombia [7] for lesions most probably caused by L. panamensis. In Guatemala, using topical or oral placebos a 68% cure rate was reported at 3 months for lesions due to L. mexicana and only 2% for lesions due to L. braziliensis [8], while other studies have reported cure rates of 27% and 39% in the general population at 3 and 12 months respectively [9], [10]. In Ecuador, in a small group of 15 patients, a cure rate of 75% at 1.5 months (no speciation but likely L. panamensis) was reported without any treatment [11].
The examples above illustrate the need to acquire and factor in local data on the natural history of disease in order to assess more accurately treatment performance.
A wide variety of treatment modalities has been reported for CL, but none has been shown to be universally effective. Treatment response varies according to a range of factors, including the Leishmania species, the patient immune status and age, the number and localization of the lesions, the severity of the disease, the treatment given and the route of administration, etc.
Treatment would benefit both the individual patient but also reduce the burden of human reservoirs in the case of anthroponotic CL, and prevent super-infection and the resulting complications. The choice of treatment, either local or systemic, is usually based on the size, number and localization of lesions, lymphatic spread or dissemination, patient's immune status, cost, risk-benefit and the availability of the treatment itself in the country. Currently available treatment options (systemic and topical) can be found in the WHO 2010 technical report [12].
The characteristics of the participants to be included must be adapted to the specific purpose of each clinical trial and must be representative of the typical patients seen in practice. The relevance of the spectrum composition of the study population to the range of patients seen in practice is of paramount importance especially in phase 3 an 4 trials. The factors that allow or disallow someone to participate in a clinical trial (“inclusion” and “exclusion” criteria, respectively), are used to identify appropriate participants and ensure both their safety and sound conclusions of the study. Establishing common grounds for entry criteria is also important in order to harmonize study populations across trials and facilitate comparability of trials and meta-analyses. It is also important to indicate the encatchment characteristics in terms of area and population, which would help in deciding as to the applicability of the findings of a trial, and ensure that the enrolled patients are representative of the larger patient population in that site (“spectrum composition”).
One of more of the following criteria may apply.
Table 1 illustrates how to apply the different entry criteria based on the type of study (Phase 2–4) and treatment being tested (systemic or topical). The final decision, however, must be taken based on prior knowledge accrued during the pre-clinical and phase I studies.
The protocol must identify clearly primary and secondary endpoints for efficacy and safety. The primary efficacy endpoint must be both accurate and robust; the protocol should clarify how and when cure is defined. It is advisable to focus the research on few endpoints that are feasible and attainable within the study, and avoid multiple, diffuse endpoints. Harmonizing efficacy endpoints is essential to allow comparing study results and conducting meta-analyses.
Any procedures applied which may interefere with healing should be standardised upfront and reported in sufficient details. Such would be the case, for standard of care, including dressing, debridement and cleaning of ulcers before and during treatment.
The assessment and reporting of the safety, toxicity and tolerability of treatments, while an essential component of the evaluation, is often overlooked in CL clinical trials. Topical treatments may produce local events at the site of the lesion (like irritation); systemic treatments may cause generalised signs or symptoms, including changes in laboratory values.
Events should be reported and graded using standard nomenclature and criteria of severity. Whenever possible, events must be combined under a syndrome or diagnosis.
It is important to comply with regulations for filing serious events; specific requirements exist for timely reporting accoriding to national regulations (health authorities, regulatory authorities, ethics committees). However, investigators must be alerted to the fact that definitions and rules for reporting may evolve with time and are not fully harmonised between countries.
This section treats of study design with a specific focus on issues of special relevance to comparing treatments for CL. In this context, we delve more into types of design (such as non-inferiority trials, adaptive designs) that the typical CL investigator might be less familiar with.
According to the recent WHO treatment recommendations for leishmaniasis, including CL, there are cases (e.g. uncomplicated L. major) where an unfavourable risk-benefit ratio (resulting from the combination of a self-curing lesion and the lack of an effective and safe treatment) means that no treatment may currently be recommended (and thus no standard treatment exists to which to compare) [12]. In other cases, cure rates up to or above 90% have been reported following different treatments, though results depend also on the duration of follow-up [1], [3]. However, even when efficacy is high, the risk-benefit of some such treatments is not always well-established, or in favour of the intervention (e.g. systemic toxicity associated with the use of parenteral antimony).
These elements must be accounted for when designing a clinical trial for any specific form of CL. These trials will belong to either of the following types: Phase 2 (safety and dose-finding studies to select the dose and duration of treatment which is safe and effective to be tested further in larger efficacy studies); Phase 3 (randomized controlled trials (RCTs) to establish the value and support the registration of a new intervention with superiority design (over reference treatment or placebo) or non-inferiority design (against a reference standard treatment); or Phase-4 trials (post-registration, when the new treatment is being implemented in the field in conditions that are closer to real life). All studies, whether with or without a direct external comparison, should have at least two arms and be randomized, with few exceptions.
All trials should be registered (see: the WHO International Clinical Trials Registration Platform (WHO-ICTRP) and reported, whether the results are favourable, unfavourable or inconclusive – both for ethical and scientific reasons. Traditionally, the importance of negative results has been underestimated both by researchers and publishers; publishing only positive results will bias knowledge. The CONSORT checklist (study design, analysis and interpretation) and flow diagram (patient attrition throughout the study) should be followed [63]. All major journals today do not publish papers on trials that have not been registered and do not follow the CONSORT guidelines (see example in Figure 9).
The protocol must be clear as to the population for analysis – typically: intent-to-treat (ITT), modified ITT (mITT) and per-protocol (PP). The basis for exclusion of patients from the analysis must be provided. Patients withdrawn because they could not tolerate treatment or because they required rescue treatment must be accounted for. The analytical plan should be finalised before freezing the data for analysis.
Like any other trial, an appropriate data management process is critical in order to have high-quality data, statistical analyses and results. For this purpose, the data management software adopted must provide a secure location for the clinical data, user rights and profiles along with password protection, as well as an audit trail. Capacity for data management is often scarce in CL-endemic countries, including both the availability of appropriate software with auditable track, and trained data managers. In these countries there is also a general shortage of statisticians to help design and to analyse and report on trials. Capacity building efforts should be organized to increase competences of research teams in this important area.
Clinical trials must be conducted in accordance with current international standards of Good Clinical Practice (GCP), an international ethical and scientific quality standard for designing, conducting, recording and reporting trials that involve the participation of human subjects. Compliance with this standard provides public assurance that the rights, safety and well-being of trial subjects are protected, consistent with the principles that have their origin in the Declaration of Helsinki, and that the clinical trial data are credible. When GCP standards are followed, the quality of data from clinical trials is adequate to make informed clinical and policy decisions.
There is a belief among some that GCP guidelines are only for “registration” studies and not for all clinical trials. However, the principles of GCP should be applied to all clinical studies with any intervention conducted at any stage of development that may have an impact on the safety and well-being of human subjects. Implementation of GCP procedures requires initial training and practice and is best served when trial personnel at a site accept and understand a culture of GCP. Maintaining a GCP environment requires constant training and reinforcement and is a process that requires continuous growth in a site and personnel. Accepted GCP standards include those published by the International Conference on Harmonization (ICH) and the World Health Organization (WHO). The ICH GCP guideline is published under Efficacy (E6) and is often referred to as ICH E6 GCP guideline [64]. A summary review of the principles of GCP are found in the WHO handbook [65].
At the same time, it should be clear that GCP is not about dogma, but rather patient's care and reliability of data, and that the context within which trials occur should be accounted for. A proper balance between the goals of the clinical study and the documentation required has been proposed [66]. The amount of written documentation and the degree of detail required by GCP procedures can be a shock to investigators not used to working in this environment. Although the conduct of clinical trials under GCP with external monitors and proper data management will inevitably increase the cost of studies, it is imperative that higher quality studies in CL be conducted.
For all trials involving human subjects, ethics review and approval must be sought from appropriate boards/committees at the institution (local and/or international) and/or country level as required. It is imperative that all clinical studies are conducted in accordance to the international and country regulations and laws.
The opinions expressed in this paper are those of the authors; the authors alone are responsible for the views expressed in this publication and they do not necessarily represent the decisions, policy or views of the WHO.
Material has been reviewed by the Walter Reed Army Institute of Research. There is no objection to its presentation and/or publication. The opinions or assertions contained herein are the private views of the author, and are not to be construed as official, or as reflecting true views of the Department of the Army or the Department of Defense.
|
10.1371/journal.pbio.1001106 | Coupled Evolution of Transcription and mRNA Degradation | mRNA levels are determined by the balance between transcription and mRNA degradation, and while transcription has been extensively studied, very little is known regarding the regulation of mRNA degradation and its coordination with transcription. Here we examine the evolution of mRNA degradation rates between two closely related yeast species. Surprisingly, we find that around half of the evolutionary changes in mRNA degradation were coupled to transcriptional changes that exert opposite effects on mRNA levels. Analysis of mRNA degradation rates in an interspecific hybrid further suggests that opposite evolutionary changes in transcription and in mRNA degradation are mechanistically coupled and were generated by the same individual mutations. Coupled changes are associated with divergence of two complexes that were previously implicated both in transcription and in mRNA degradation (Rpb4/7 and Ccr4-Not), as well as with sequence divergence of transcription factor binding motifs. These results suggest that an opposite coupling between the regulation of transcription and that of mRNA degradation has shaped the evolution of gene regulation in yeast.
| The regulation of mRNA levels in the cell is important to ensure, for instance, timely cellular responses to changes in the environment. mRNA transcription and mRNA degradation directly affect mRNA levels and it would make sense to have a system in place that would coordinate these opposing processes. Previous studies suggested that regulation of transcription in the nucleus may be linked to regulation of mRNA degradation in the cytoplasm, yet the details of this connection are poorly understood. In this study, we took an evolutionary approach to address this question by comparing both transcription and mRNA degradation between two yeast species. We found that evolution of these distinct processes is coordinated, as genes that diverged in mRNA degradation tend to also diverge in transcription. Interestingly, the coordination is counterproductive, as increased transcription is linked to increased mRNA degradation. We analyzed a hybrid between the two yeast species to classify evolutionary differences according to the type of underlying mutation (cis or trans). This analysis indicated that coordinated changes in transcription and mRNA degradation are likely to be driven by the same individual mutations, and thus directly coupled. Finally, we suggest several mechanisms that may mediate this coupling, including complexes which are involved in both processes (Rpb4/7 and Ccr4-Not) and promoter regulatory regions. These results suggest that a direct coupling between the regulation of transcription and mRNA degradation is a common phenomenon employed by approximately 10% of the genes in yeast.
| Work on the regulation of mRNA levels has traditionally focused on transcription, although mRNA levels reflect the balance between transcription and mRNA degradation. Recent studies have shown that regulation of mRNA degradation also has a central role in control of gene expression, and in certain systems might be as important as transcription regulation [1]–[10], underscoring the importance of systematically studying the patterns of mRNA degradation and their regulation. While the basic machinery of mRNA degradation is well established [2],[11], very little is known regarding gene-specific and condition-specific regulation by RNA-binding proteins (RBPs), which bind to subsets of mRNAs and coordinate their post-transcriptional regulation [12],[13]. Notably, hundreds of RBPs are predicted in each eukaryotic genome, yet the subsets of bound mRNAs and the impact on mRNA degradation are known only for a selected few [14]–[18].
Although both transcription and mRNA degradation individually contribute to the regulation of mRNA levels, they are ultimately integrated to form a coherent regulatory system, and several studies provided evidence for crosstalk between the regulation of transcription and mRNA degradation. First, two conserved and general regulatory complexes, the Rpb4/7 dimmer, which is composed of two subunits of RNA polymerase II [19], and the Ccr4-Not complex [20],[21], have been shown to control both transcription and mRNA degradation and thus may serve to coordinate their regulation. Second, recent work in the fission yeast has described a feed-forward loop whereby a transcription factor activates a regulator of mRNA degradation and both factors jointly control the expression of a common subset of genes [22]. Such interplay between factors that control transcription and mRNA degradation might in fact be a common property of regulatory networks [23]. Third, several studies examined the response of S. cerevisiae to environmental perturbations and found coordinated changes in mRNA degradation and transcription [5],[6],[8],[9],[24]. For example, Shalem et al. [24] found that transcriptional regulation is coordinated with changes in mRNA stability and that the mode of this coordination is condition-dependent, such that induced genes are stabilized in one condition (during DNA damage) and destabilized in another (during oxidative stress).
Taken together, these observations suggest that transcription and mRNA degradation are often coordinated. However, this coordination remains poorly understood, raising several important questions. What is the scope of this coordination? What mechanisms underlie this coordination and are they directly or indirectly influencing both processes? What is the mode of coordination—is transcriptional induction mostly coordinated with decreased degradation, increased degradation, or both? What is the functional significance of such coordination?
To address these questions, we set out to examine the coordination between transcription and mRNA degradation from an evolutionary perspective, by comparing two closely related yeast species, S. cerevisiae and S. paradoxus. These species diverged from a common ancestor ∼5–10 million years ago and maintained similar physiology and genomic sequences (∼90% identity), yet as we have shown previously [25], most of their orthologous genes have diverged in mRNA levels. Comparing the mRNA degradation rates of these species, we find significant differences at ∼11% of the orthologs. Remarkably, around half of these evolutionary differences in mRNA degradation are coupled to evolutionary differences in transcription, indicating a widespread coordination. This coordination involves almost exclusively opposite effects of transcription and degradation such that transcriptional induction is coupled to increased mRNA degradation. Furthermore, classification of transcription and degradation changes into cis and trans, by allele-specific analysis of the interspecific hybrid, suggests a direct mechanistic coupling whereby individual mutations influence both transcription and mRNA degradation. These mutations seem to involve Rpb4/7, Ccr4-Not, as well as additional unknown factors.
To compare the mRNA degradation rates of the two species, we monitored mRNA levels following transcriptional arrest using 1,10-Phenantroline [7],[26]. mRNA levels were measured at 0, 20, 40, and 60 min after addition of the drug using a two-species microarray [25]. As expected, the profiles of most genes were well approximated by an exponential decay, which is reflected by a linear decrease of the log2 mRNA levels with time (Figure 1a). Degradation rates were estimated as the slope of the linear fit for 78% of the genes that had an R2 value (goodness-of-fit) above 0.94, while genes with lower R2 were excluded from further analysis. The calculated mRNA degradation rates of S. cerevisiae genes were highly reproducible among two biological repeats and between probes that were designed for different positions of the same genes, and were consistent with previous measurements of mRNA degradation that utilized a PolII mutant strain to block transcription (Figure 1b) [24].
Degradation rates were largely conserved among the two yeast species, with a genome-wide correlation of 0.78 (Figure 1c), yet we identified considerable differences at ∼11% of the orthologs, in which the difference was both statistically significant (p<0.05) and above 1.4-fold (i.e., the higher degradation rate exceeded the lower degradation rate by at least 40%, see Figure S1 for results with other thresholds). Differential mRNA degradation rates of six genes were validated by real-time PCR (Figure S2). These results indicate that, even among such closely related species, considerable differences in mRNA degradation rates are common, although much less common than differences in mRNA levels, which were observed for approximately half of the genes in this and in previous work (Figure S1) [25]. Differential degradation was observed for genes with various functions but was particularly enriched among respiration-related genes. Notably, degradation rates of these genes were consistently higher in S. paradoxus than in S. cerevisiae, as shown in Figure 1d for the 12 oxidative phosphorylation genes included in our analysis.
We next turned to systematically compare the changes in mRNA degradation rates to the changes in mRNA levels, as measured here in the zero time-point (before transcription arrest), or in a previous work [25]. Sorting the genes by the degree of inter-species differential degradation rate, we observed that differential degradation is associated with inter-species differential mRNA level (Figure 2a). This might seem expected, as mRNA levels are partially determined by mRNA degradation. Surprisingly, however, the direction of differences in mRNA levels is opposite to that expected purely from the difference in mRNA degradation: genes with higher mRNA degradation rate in one of the species tend to have higher mRNA levels in that species, although the increased degradation would be expected to decrease their mRNA levels (Figure 2b). This indicates that apart from the differences in degradation rates, there are also differences in the transcription rates of these genes that exert opposite effects on mRNA levels. For example, oxidative phosphorylation genes have significantly faster mRNA degradation in S. paradoxus than in S. cerevisiae, yet 11 out of 12 of these genes in fact have significantly higher mRNA level in S. paradoxus than in S. cerevisiae (Figure 2b, blue dots). Strikingly, in close to 80% of the genes with differential mRNA level and differential degradation, the difference in mRNA level is opposite to that expected from the difference in mRNA degradation, thus implying opposing effects of transcription and degradation (red section in Figure 2c).
Technical biases do not seem to have a significant effect on the observed coupling. First, the coupling is observed for large differences in mRNA degradation (red section), but not for genes with very small changes in degradation, which are more dependent on technical variations (green section in Figure 2c). Second, we used different datasets to compute mRNA levels and mRNA degradation, thus avoiding potential artifacts that might generate the observed coupling. Third, our microarray contains different probes for the same genes with widely different hybridization intensities (which serve to calculate mRNA levels), but these differences do not affect the estimation of mRNA degradation rates (see Materials and Methods). Fourth, the observed coupling cannot be accounted by microarray artifacts or residual transcription (see Methods and Figure S3).
Notably, the above analysis in fact underestimates the scope of the coupling between transcription and mRNA degradation, since mRNA levels are used instead of transcription rates. For example, some genes displayed a difference in mRNA degradation rates but no significant difference in mRNA levels (e.g., PRP9, see Figure 2b). This again implies an opposite difference in transcription that compensates for the difference in mRNA degradation (thus resulting in similar mRNA levels in the two species), yet these genes were not considered in our previous analysis. To account for this effect we can estimate the transcription rates of the two species by integrating the measures of mRNA levels and degradation (see Materials and Methods). This analysis indeed increases the proportion of coupled genes (gray curves in Figure 2a,c), although calculated transcription rates should be taken with caution and may artificially overestimate the coupling (see Materials and Methods). We thus predict the true scope of opposite coupling to be within the range indicated by analysis of mRNA levels and that of estimated transcription rates (e.g., among genes that differ both in transcription and in mRNA degradation ∼80%–90% have opposite effects; see Figure 2c). Nevertheless, in subsequent analyses we took a conservative approach and considered coupling only among those genes identified by both mRNA levels and estimated transcription rates.
Taken together, a large fraction of the evolutionary changes in mRNA degradation were coupled to opposite evolutionary changes in transcription (44%–80%, as derived from our conservative and relaxed analyses, respectively; see Figure 2d). Note, however, that this coupling constitutes only 10%–20% of the evolutionary changes in transcription (Figure 2d), as transcriptional changes were much more frequent and typically independent of those in mRNA degradation; this might explain why previous studies failed to notice such coupling.
Transcription and mRNA degradation are controlled by different mechanisms and are thus expected to diverge through a separate set of mutations. However, the strong coupling that we observe suggests the intriguing possibility that individual mutations may influence both transcription and mRNA degradation, generating opposing effects on mRNA levels. Although we cannot identify the effect of individual mutations, this possibility can be examined by differentiating between the contributions of cis- and trans-mutations to evolutionary changes in mRNA degradation and transcription. Cis-mutations occur within the affected gene or in its flanking regulatory sequences (e.g., promoter or 3′-UTR motifs), while trans-mutations occur in other loci and indirectly influence the affected gene through the activity of another protein (e.g., RNA-binding protein). Importantly, the genome-wide contributions of cis- and trans-mutations can be uncovered by analysis of inter-species hybrids: cis-mutations discriminate between two hybrid alleles that reflect orthologs from the two species, while trans-mutations do not discriminate between the two hybrid alleles, as the alleles are in the same nucleus and thus exposed to the same set of trans-regulators. This approach has previously been used to assess the contribution of cis- and trans-mutations to total mRNA levels [25],[27]–[29] and recently also to nucleosome positioning [30], while here we extend it to study mRNA degradation rates.
We measured allele-specific mRNA degradation rates for the hybrid of S. cerevisiae and S. paradoxus, with two biological repeats and using the same method as described above for the two species. For each gene whose mRNA degradation rate differs between the species, we examined whether this difference is maintained (cis) or abolished (trans) between the corresponding two hybrid alleles. This analysis indicated that ∼60% of the differences in mRNA degradation reflect primarily cis-mutations, while ∼40% reflect trans-mutations (Figure 3). Six cis-differences were further validated by real-time PCR of the hybrid alleles (Figure S2).
If coupled changes in transcription and degradation are due to independent mutations, then each change can be either in cis or in trans, and thus the coupling should be observed for all combination of cis- and trans-effects; for example, cis-effects in mRNA degradation should be coupled both to cis-effects in transcription (cis-cis combination) and to trans-effects in transcription (cis-trans combination). However, if transcription and degradation changes are mechanistically coupled and the observed opposite effects are generated by the same individual mutations, then these coupled changes would be generated by a single effect, either in cis (cis-cis combination) or in trans (trans-trans combination), but not by cis-trans or trans-cis combinations. Consistent with this, a strong coupling is observed only for cis-cis and trans-trans combinations but not for cis-trans or trans-cis combinations (Figure 4).
Coupling between trans-changes in mRNA degradation and trans-changes in transcription (trans-coupling) suggests that divergence of upstream regulator(s) has influenced both processes. We thus searched for enrichment of 85 high-confidence trans-coupled genes with targets of 116 transcription factors (TFs) [31], 46 RNA-binding proteins (RBPs) [14],[16], Rpb4/7 [32], and Ccr4-Not [33]. Fifteen of the 173 target gene-sets were enriched (p<0.05) among the trans-coupled genes compared to uncoupled genes (Figure 5a). Notably, these include an Rpb4/7 dataset (Rpb4 [32]) and three datasets of Ccr4-Not (Ccr4, Not5, Caf1 [33]), which were among the five most enriched datasets. Furthermore, while the combined target gene-sets of Rpb4/7 and Ccr4-Not include only 12% of all genes examined here and 18% of the genes with uncoupled transcriptional changes, they include 41% of the trans-coupled genes (p = 2×10−7). Thus, our results are consistent with previous studies showing that these two complexes influence both transcription and mRNA degradation.
Target gene-sets of nine TFs and two RBPs were also enriched with trans-coupled genes (Figure 5a). However, excluding the targets of Rpb4/7 and Ccr4-Not completely abolished the enrichment of four of these TFs (Figure 5b), suggesting that their enrichment was due to high overlap with targets of Rpb4/7 and Ccr4-Not and may not reflect the function of these TFs. The remaining enriched gene-sets included targets of three TFs involved in respiration (Hap1, Hap4, and Hap5), two TFs involved in amino-acid biosynthesis (Gln3, Met31), the poly(A) binding protein (Pab1), and the SR-like protein Npl3. Interestingly, both Pab1 [34] and Npl3 [35] are known to shuttle between the nucleus and the cytoplasm, Npl3 was previously implicated in regulation of transcription [36] and translation [37], and Pab1 was previously implicated in regulation of mRNA degradation [38]. These results suggest that, in addition to Rbp4/7 and Ccr4-Not, coordination between transcription and mRNA degradation may also involve Pab1 and Npl3.
The enrichment of trans-coupled genes among targets of specific regulators suggests not only that these regulators control both transcription and mRNA degradation, but also that the activity of these regulators diverged among the two species. Consistent with this possibility, the expression level of Rpb4 is ∼3-fold higher in S. paradoxus than in S. cerevisiae, while the expression of other RNA Pol II subunits is much more conserved (Figure S4). Increased activity of Rpb4/7 in S. paradoxus would be expected to increase both transcription and mRNA degradation in S. paradoxus (compared to S. cerevisiae), and indeed we find that targets of Rpb4/7 are highly enriched among coupled trans-effects with higher S. paradoxus transcription and degradation but not among those with higher S. cerevisiae transcription and degradation (Figure S4).
Coupling between cis-changes in mRNA degradation and cis-changes in transcription (cis-coupling) suggests that mutations in a gene's promoter, coding-region, terminator or untranslated regions influenced both processes. This may reflect mutations that affect the recruitment of specific proteins to the loci of that gene, which then influence both transcription in the nucleus and degradation of the resulting mRNA following its export to the cytoplasm. To examine this possibility, we first searched for enrichment of 92 high-confidence cis-coupled genes with targets of the various regulators, as described above for the trans-coupled genes. Only one of the 170 datasets was enriched among the cis-coupled genes with a p value below 0.01 (Figure 5c). This dataset included genes upregulated upon deletion of Rpb4 and was significantly enriched with cis-coupling (p = 7×10−5), suggesting that cis-mutations may have influenced the recruitment of Rpb4/7 to many genes. At a p value of 0.05, only one additional target gene-set was enriched (Hap3), while ∼9 sets would be expected by pure chance (0.05×173).
Despite the significant enrichment of Rpb4/7 targets, these include only 13% of the cis-coupled genes, suggesting the existence of other mechanisms for cis-coupling. We next examined the sequence divergence between S. cerevisiae and S. paradoxus at various predicted and known cis-regulatory elements. Analysis of diverged 3′-UTR sequences that were predicted to influence mRNA stability [15] or to be bound by RNA-binding proteins (RBPs) [14],[16] did not identify a significant association with cis-coupled genes (Figure 5d). In contrast, diverged transcription factor (TF) binding sites [39] were significantly enriched at cis-coupled genes, compared to uncoupled genes that diverged only in transcription (Figure 5d, p<10−3). This enrichment was found both for known S. cerevisiae TF binding sites [31] that are not conserved in S. paradoxus and for predicted S. paradoxus TF binding sites that are not conserved in S. cerevisiae (Figure 5d). Notably, diverged TF binding sites were enriched at cis-coupled target genes of Rpb4/7, suggesting that these mutations may have influenced the recruitment of Rpb4/7, but also at cis-coupled genes not targeted by Rpb4/7, implying that the effect of these mutations on transcription and mRNA degradation is also mediated by additional mechanisms. This analysis of diverged binding sites included various TFs and we could not detect any TF with specific overrepresentation. As expected, diverged TF binding sites were not enriched among trans-coupled genes (Figure 5e), further supporting their direct association with cis-coupling.
Our systematic comparison of mRNA degradation among two yeast species demonstrated the following: (i) Degradation rates differ among ∼11% of the orthologs, compared to ∼50% that differ in transcription or in mRNA levels. (ii) Differences in mRNA degradation are often coupled to opposite differences in transcription, and this coupling constitutes around half of the changes in mRNA degradation but only ∼10% of the changes in transcription. (iii) Coupled changes in transcription and degradation are generated by the same type of mutations (cis or trans) suggesting a mechanistic coupling. (iv) Trans-coupling is associated with regulators that are known (Rpb4/7 and Ccr4-Not) to control both transcription and mRNA degradation, while cis-coupling may be associated with diverged TF motifs.
The association of trans-coupled genes with Rpb4/7 and Ccr4-Not suggests that altered activity of these complexes influenced, in parallel, both transcription and mRNA degradation of target genes. This possibility of parallel coupling (see Figure 6), whereby an upstream regulator controls multiple regulatory steps and may coordinate them, is consistent with known functions of Rpb4/7 and Ccr4-Not and, more generally, with the notion that RBPs often coordinate multiple steps in the regulation of their target genes [13]. Trans-coupling is also associated with two other RBPs known to shuttle between the nucleus and the cytoplasm (Pab1 and Npl3), suggesting that these may also serve as coordinators of transcription and mRNA degradation, and possibly of additional steps.
Notably, divergence of individual trans-regulators can cause similar evolutionary changes across many co-regulated target genes. Indeed, trans-coupling includes a set of respiration-related genes, all with higher transcription and mRNA degradation rates in S. paradoxus than in S. cerevisiae, likely reflecting a module that coherently diverged through one or few trans-mutations. While this module is known to be transcriptionally co-regulated, these results suggest that it is also post-transcriptionally co-regulated, thus representing an “RNA regulon” [13]. Divergence of this module may have been part of the domestication of S. cerevisiae and an associated optimization of anaerobic fermentation [40]. Notably, although high-confidence trans-coupled genes are highly enriched with the respiration module (p = 10−10), this enrichment accounts only for a quarter (21/85) of these genes, suggesting that additional RNA regulons might have evolved by parallel (and opposite) changes in their transcription and mRNA degradation.
While trans-regulators may affect transcription and mRNA degradation in parallel, cis-acting sequences are likely to be more specific to one of these processes, for example, by mediating the binding of TFs to promoters or that of RBPs to mRNAs. We thus propose that cis-coupling may work by sequential coupling (Figure 6), whereby mutated cis-acting elements affect one process (transcription or degradation) and this in turn signals to the other process, thereby causing an additional effect. The enrichment of cis-coupling with diverged TF motifs, but not RBP (i.e., stability) motifs, suggests a mode of sequential coupling that is directed from transcription to mRNA degradation. This possibility is consistent with a shuttling mechanism, as previously proposed for Rpb4/7 [19], whereby transcription-related molecules bind to the transcribed mRNA and are exported with it to the cytoplasm where they influence its degradation. Rpb4/7 targets are indeed enriched among cis-coupled genes, but this accounts only for a small proportion of cis-coupling, suggesting the existence of additional factors for sequential coupling by a similar shuttling mechanism or by other mechanisms.
Alternatively, the enrichment of TF motifs, but not stability motifs, may reflect the bias in current knowledge, as fewer motifs are known for RNA-binding proteins and these may rely more heavily on structural properties. Sequential coupling may thus initiate by binding of RBPs to yet unknown motifs and regulate mRNA degradation, followed by signaling back to the nucleus that influences transcription of that gene or perhaps of a set of genes. This possibility is consistent with the notion that RBPs are highly inter-connected and coordinate multiple regulatory events [13]. However, the observation that coupling typically involved larger changes in transcription than in mRNA degradation appears to support a transcription-to-degradation directionality. Interestingly, both of these models make the intriguing and testable prediction that experimental manipulation of individual cis-regulatory elements would affect both transcription and mRNA degradation of the associated genes.
The results presented here reflect the specific evolutionary divergence of two yeast species and hence might not be sufficient to infer general conclusions regarding the scope and mode of coupling. For example, few trans-mutations may have driven the evolution of many target genes (e.g., respiration module) and by that bias our results. Importantly, however, cis-coupled genes are each affected by distinct sets of mutations; the only exception is of neighboring genes which may diverge through the same mutations in cis, but these encompass only up to 5% of the observed cis-coupled genes. Therefore, our results imply ∼140 independent cases in which cis-acting mutations affected both transcription and mRNA degradation, generating opposite effects on mRNA levels (Figure 2d). At the same time, ∼1,700 genes diverged by cis-acting mutations only in transcription, and ∼160 genes diverged by cis-acting mutations only in mRNA degradation (Figure 2d). These results demonstrate that coupling is not a global phenomenon, as it does not affect the majority of genes, nor is it a rare event.
It is tempting to further speculate that cis-divergence is not strongly biased towards certain mechanisms and thus that observed patterns of cis-divergence may provide a rough estimate for the frequencies of possible mutational outcomes and regulatory mechanisms. Accordingly, we would predict that (i) transcriptional regulation is much more prevalent than regulation of mRNA degradation, although the exact proportion is difficult to quantify as differential mRNA degradation is more difficult to identify than differential transcription; (ii) coupling constitutes approximately 10% of the regulation of transcription but almost half of the regulation of mRNA degradation. (iii) Coupling occurs almost exclusively between opposite effects on mRNA levels (increased transcription is associated with increased mRNA degradation and vice versa).
This last prediction is especially surprising given that previous studies have highlighted a coherent mode of coupling whereby changes in mRNA levels may be driven by both transcription and mRNA degradation acting in the same direction [5],[7],[8],[22],[41]. These views may be reconciled if one mode (coherent changes) reflects coordination of distinct pathways for transcription and mRNA degradation that have co-evolved to support certain responses to environmental perturbations, while the other mode (opposite changes) reflects a mechanistic coordination whereby the same pathway affects both processes. Since these closely related species differ in the regulation of approximately half of the genes, and these differences are small in magnitude (∼1.5-fold), we suspect that they primarily reflect neutral drift and as such they expose the mechanistic (opposite) coupling that is presumably “built in” to regulatory mechanisms, but does not reveal coherent coupling as these primarily evolved prior to the divergence of these species and may not be continuously evolving.
This proposed mode of opposite coupling appears counterintuitive and inefficient, as transcription and degradation effects would compensate one another. What then may be the benefits of such coupling? One possibility is that an opposite coupling may enable transient responses to environmental changes: upon stress conditions, cells cease to grow and mount an transcriptional response, but at the same time increase the degradation rates of upregulated genes, thereby facilitating their return to basal expression levels and normal growth [4],[24]. Such transient responses may have been particularly important for thriving in fluctuating environments, and coupling mechanisms may have thus become “built-in” components of gene regulation that are active also in the absence of stress and are thus exposed by genetic mutations.
Another plausible advantage of such coupling is that it may decrease the effect of genetic or environmental perturbations on mRNA abundance, as changes in one level of regulation would be compensated by another level. Such intrinsic “negative feedback” could increase the robustness of gene regulation and thus reduce cell-to-cell variability. Surprisingly, however, we observe the exact opposite: genes that display coupled evolution in our data or that are targets of coupling mechanisms (i.e., Rpb4/7 and Ccr4-Not) have a considerably higher cell-to-cell variability in protein abundance (expression noise [42]) than other genes (Figure S5). Notably, this effect is comparable in magnitude to other factors that were previously implicated in increasing noise (i.e., TATA-box [43] and promoter nucleosome occupancy [44]) and remains significant after controlling for these factors. This may indicate that coupling between transcription and mRNA degradation is further associated with additional regulatory effects. Given the recent demonstration that Rpb4/7 also influences translational regulation [45], and the interplay between mRNA degradation and translation [46]–[48], it is tempting to speculate that the coupling that we observed is further linked to translation bursts that give rise to high cell-to-cell variability [49].
To facilitate comparison to the diploid hybrid, we generated diploid homozygote yeast strains of the two species, thus avoiding both potential differences between haploids and diploids, and potential heterozygosity within normal diploid strains, which could confound inter-species comparisons. Diploid homozygote strains were generated from the haploid S. cerevisiae (BY4741) and S. paradoxus (CBS432) strains, by transient HO activation and selection for diploid strains. The hybrid strain was generated by mating the same parental haploids. These three diploid strains (S. cerevisiae, S. paradoxus, and hybrid) were grown to log-phase at rich media (YPD medium at 30°C).
Two to five different 60-mer probes were designed for most genes in each of the two species, and each probe was placed at two different positions (duplicates) on an Agilent custom (two-species) microarray. Probes were selected both by general criteria for probe selection (intermediate %GC, no self-hybridization or low complexity regions, distance from the gene 3′-end) and by preference for low sequence similarity between the two species in order to avoid cross-hybridization (all probes reflect genomic positions with lower than 90% sequence identity between the two species).
S. cerevisiae, S. paradoxus, and their hybrid were subjected to 150 µg/ml of 1,10-phenanthroline at log-phase and sampled after 0, 20, 40, and 60 min. Total RNA was extracted using MasterPure Yeast RNA purification Kit (EPICENTRE), amplified with Agilent's Low RNA Input Amplification Kit and hybridized with Agilent's standard protocols and kits to the two-species microarrays. S. cerevisiae and S. paradoxus samples were pooled and hybridized together and the hybrid was hybridized separately, both with biological repeats. Arrays were scanned using Agilent microarray scanner and feature extraction software. Raw and processed microarray data are available at the GEO database (GSE28849).
During the time-course, transcription is arrested and total mRNA levels are decreasing, but this decrease is masked by the experimental protocol, as equivalent amounts of total RNA are extracted from each sample. Previous studies that used a PolII mutant strain could circumvent this problem since mRNAs constitute only a minor fraction of the total RNAs in a yeast cell, and the transcription of other RNAs (by PolI and PolIII) was not inhibited [3],[24]. However, Phenanthroline appears to inhibit all three RNA polymerases to approximately the same extent and we did not detect a decrease in the relative levels of mRNA (unpublished data). We therefore decided to scale the entire data at each time point according to an overall exponential decay with half-life of 25 min, consistent with previous studies [3],[24]. Accordingly, log2 of the total (or average) abundance of all mRNAs should decrease linearly by 1 unit every 25 min, and thus decrease by 0.8 every 20 min (the interval between consecutive time-points). We thus scaled the data by centering the four consecutive time points (0, 20 min, 40 min, and 60 min) on 0, −0.8, −1.6, and −2.4, respectively.
For each probe, we averaged the hybridization intensities from the duplicate microarray spots, and fitted a linear slope to the log2-intensities. All probes with an R2 value smaller than 0.94 were excluded from further analysis. For each gene, the absolute value of the median slope of all remaining probes was defined as its degradation rate.
Since the four time-points are evenly spaced (0, 20, 40, and 60 min) the difference between mRNA levels at consecutive time-points should be approximately constant and reflect the mRNA degradation rates. To identify differential degradation rates among orthologous probes, we thus performed a two-sample t test, comparing the three estimates of each probe (M20–M0, M40–M20, and M60–M40, where Mi is the mRNA level at time i) between the two species. Genes for which the median p value from the t tests of the different probes was below 0.05 were further examined. p values reflect both the degree of differential degradation and the consistency among the three estimates (even a negligible difference can be identified as significant if the three measures are highly similar within each species). We thus further examined the extent of differential degradation and retained only those genes in which the ratio between the faster and lower degradation rates (from the two species) is higher than 1.4.
The first time-point reflects mRNA levels during exponential growth and before transcriptional arrest. It therefore reflects an approximate steady-state mRNA level. A potential caveat is that if the first time-point is used to measure both mRNA levels and mRNA degradation, then measurement errors could generate artificial coupling between mRNA levels and degradation. For example, if the first time point is increased due to technical noise, then estimates of both mRNA level and mRNA degradation would increase and result in apparent coupling. To avoid this problem, we used only one time-course to derive estimates of mRNA degradation rates and the first time-point of the second time-course to derive an estimate of mRNA level. As additional control, we used mRNA levels as measured in a previous work and obtained similar results (unpublished data) [25]. Differential expression was defined as above 1.5-fold difference between the species (or hybrid alleles).
For each gene, we assume that the production rate of mRNAs (transcription rate) is approximately equal to the overall degradation of mRNAs, and therefore given by the steady-state level of mRNAs multiplied by their constant degradation rate. Hence, TR = D×L, where TR, D, and L are the transcription rate, degradation rate, and mRNA level, respectively. The difference in transcription rates between S. cerevisiae and S. paradoxus can thus be estimated from the respective differences of degradation rates and mRNA levels: log(TRcer/TRpar) = log(Dcer/Dpar)+log(Lcer/Lpar).
We note that this estimation may not be accurate as a result of possible violation of the steady-state assumption, spurious correlations with mRNA degradation due to the method of calculation, and the integration of nuclear and cytoplasmic mRNAs in our measurements. Our main conclusions do not require these estimates of transcription rates and can be inferred from direct comparison of inter-species differences in mRNA degradation to those in mRNA levels. However, since mRNA levels are inherently affected by mRNA degradation in a manner that is opposite to the observed coupling, such analysis would underestimate the scope of the coupling (as illustrated in Figure 2c by PRP9). We thus argue that analysis of mRNA levels underestimates the scope of the coupling, while analysis of estimated transcription rates may overestimate it and that the two analyses are complementary. Nevertheless, we defined coupled genes for further analysis based on consensus of mRNA levels and transcription rates analyses in order to avoid cases of spurious coupling.
Our experimental design may be susceptible to two confounding effects. First, the use of two-species microarrays, whereby the two species are co-hybridized to a single array that contains species-specific probes, may result in cross-hybridization such that mRNA from one species hybridizes to probes of the other species. Second, inhibition of transcription with 1,10-phenanthroline may not be enough to completely block transcription and residual transcription activity may hinder our calculation of mRNA degradation rates. However, as described below, both of these effects are likely to have only a minor influence on our results and, in particular, are not expected to cause the observed coupling between transcription and mRNA degradation.
Classification into cis and trans is based on whether the inter-species difference in mRNA degradation rates (Δspecies) is retained (cis) or abolished (trans) within the hybrid (Δhybrid), while intermediate cases are excluded from the analysis. Cis changes were defined as significant inter-species differences for which Δhybrid has the same sign as Δspecies and is larger than 1.2-fold for each of the two repeats, and the residuals (Δhybrid–Δspecies) are smaller than 1.3-fold. Trans changes were defined as significant inter-species differences for which Δhybrid has either a different sign than Δspecies or is smaller than 1.2-fold for each of the two repeats, and the residuals (Δhybrid–Δspecies) are larger than 1.3-fold. This definition is clearly threshold dependent, and other thresholds or criteria that we used led to similar proportions of cis and trans changes, typically with the percentage of cis differences between 50% and 75% (unpublished data).
High-confidence sets of cis/trans-coupled genes were defined as those with a significant cis/trans mRNA degradation difference above 1.5-fold and a cis/trans mRNA level difference above 1.5-fold (in the opposite direction to that expected by the degradation difference).
Targets of 116 TFs were defined based on Chromatin Immuno-precipitation and sequence analysis, taken from MacIsaac et al. [31] (p<0.005 and no conservation criteria). Targets of RNA-binding proteins were defined based on RNA Immuno-precipitation, taken from Hogan et al. [14]. Targets of seven subunits of Ccr4-Not were defined as genes whose expression decreased by at least 2-fold upon deletion of the respective subunits in rich media [33]. Targets of Rpb4/7 were defined as genes whose expression decreased by at least 2-fold upon deletion of Rpb4 in rich media [32].
TF binding motifs were taken from MacIsaac et al. [31] (p<0.005 and no conservation criteria).
Diverged binding sites were defined as follows:
Total RNA was extracted with the MasterPure Yeast RNA purification Kit (EPICENTRE). One microgram of each RNA sample was reverse transcribed with Moloney murine leukemia virus reverse transcriptase (Promega, Madison, WI) and random hexamer primers (Applied Biosystems). Real-time PCR was performed with StepOne real-time PCR machine (Applied Biosystems, Foster City, CA) with Syber Green PCR supermix (Invitrogen). The primers used are described in Table S1.
|
10.1371/journal.pcbi.1003460 | VarWalker: Personalized Mutation Network Analysis of Putative Cancer Genes from Next-Generation Sequencing Data | A major challenge in interpreting the large volume of mutation data identified by next-generation sequencing (NGS) is to distinguish driver mutations from neutral passenger mutations to facilitate the identification of targetable genes and new drugs. Current approaches are primarily based on mutation frequencies of single-genes, which lack the power to detect infrequently mutated driver genes and ignore functional interconnection and regulation among cancer genes. We propose a novel mutation network method, VarWalker, to prioritize driver genes in large scale cancer mutation data. VarWalker fits generalized additive models for each sample based on sample-specific mutation profiles and builds on the joint frequency of both mutation genes and their close interactors. These interactors are selected and optimized using the Random Walk with Restart algorithm in a protein-protein interaction network. We applied the method in >300 tumor genomes in two large-scale NGS benchmark datasets: 183 lung adenocarcinoma samples and 121 melanoma samples. In each cancer, we derived a consensus mutation subnetwork containing significantly enriched consensus cancer genes and cancer-related functional pathways. These cancer-specific mutation networks were then validated using independent datasets for each cancer. Importantly, VarWalker prioritizes well-known, infrequently mutated genes, which are shown to interact with highly recurrently mutated genes yet have been ignored by conventional single-gene-based approaches. Utilizing VarWalker, we demonstrated that network-assisted approaches can be effectively adapted to facilitate the detection of cancer driver genes in NGS data.
| A cancer genome typically harbors both driver mutations, which contribute to tumorigenesis, and passenger mutations, which tend to be neutral and occur randomly. Cancer genomes differ dramatically due to genetic and environmental factors. A major challenge in interpreting the large volume of mutation data identified in cancer genomes using next-generation sequencing (NGS) is to distinguish driver mutations from neutral passenger mutations. We propose a novel mutation network method, VarWalker, to prioritize driver genes in large scale cancer mutation data. Applying our approach in a large cohort of lung adenocarcinoma samples and melanoma samples, we derived a consensus mutation subnetwork for each cancer containing significantly enriched cancer genes and cancer-related functional pathways. Our results indicated that driver genes occur within a broad spectrum of frequency, interact with each other, and converge in several key pathways that play critical roles in tumorigenesis.
| Next-generation sequencing (NGS) technologies have enabled genome-wide identification of somatic mutations in large scale cancer samples. One major challenge in interpreting the large volume of mutation data is to distinguish ‘driver’ mutations from numerous neutral ‘passenger’ mutations to facilitate the identification of targetable genes and new drugs. So far, the most widely adopted method is to search for highly frequently mutated genes within one cancer type [1], [2]. Although effective in many cases, frequency-based approaches suffer from disadvantages such as lack of power to detect infrequently mutated driver genes and failure to incorporate functional interconnections and regulations among genes. Recently, many new methods have been reported. For a more comprehensive review, please refer to [3], [4].
The complex features of mutations derived from NGS data present great challenges for computational approaches, both genetically and technically. First, the probability that a gene is mutated in a sample, i.e., the gene-based mutation rate, is influenced by both genetic and environmental factors. In this study, we only consider single nucleotide variants (SNVs) and small insertions and deletions (indels), and we define a mutant gene (abbreviated as MutGene) if it harbors at least one non-silent deleterious SNV or indel (see Materials and Methods). Assuming that mutations occur randomly across the genome, long genes have a better chance of harboring mutations (e.g., the gene TTN). Other factors, including sequence context, GC content, replication timing, chromatin organization, and alterations in mutation repair systems [5], [6], [7], as well as personal lifestyle and mutagen exposure period and level, have an impact on the gene-based mutation rate in an individual. Second, mutation ‘hotspot’ families, among other factors, often contribute many genes to the list of top candidate genes that are ranked by frequency. For example, genes from the olfactory receptor family are frequently mutated in many cases [1], including both normal and disease samples [8]. However, it remains unknown whether these mutations, or only some of them, are disease-related. Finally, sequence errors exist; however, large scale validation is still a challenge in NGS projects that involve hundreds of cancer samples. Since all of these factors accumulate non-clinically related events in mutation data, these biases should be considered when developing new approaches to prioritizing driver mutations.
An alternative approach to detect possible driver genes overlays the mutation genes in the context of biological pathways or protein-protein interaction (PPI) networks and then performs integrative analyses to identify significantly altered pathways or subnetworks. In cancer, functional pathways or biological networks are frequently interrupted in many patients [9], and their gene components present mutually exclusive or co-occurring patterns [10]. To date, only a few studies have searched the cooperative mutation modules underlying cancer [11], [12]. Notably, the incorporation of other large scale genetic and/or genomic data, such as mRNA abundance [13] and methylation data [14], can greatly improve the detection of driver genes. However, these datasets are not always available for the same patient cohort in large-scale sequencing projects, creating both challenges and a high demand to develop comprehensive approaches that can prioritize driver genes from mutation data.
In this work, we propose VarWalker, a network-assisted approach that aims to prioritize potential driver genes and better interpret mutation data in NGS studies. Our goal is to develop a tool that can address the huge variations among cancer samples as well as implement conventional approaches in modern NGS data analysis. VarWalker performs sample-specific filtering and implements the Random Walk with Restart (RWR) algorithm to search for frequently interrupted interactions between MutGenes and their interactors. We argue that if an interaction is interrupted by mutations in one or two of its linking proteins across many samples, this interaction has a higher chance to be biologically important than an interaction in which only one protein is disrupted by mutations. We demonstrated VarWalker in two recent large-scale NGS benchmark studies: one involving 183 matched tumor/normal LUAD samples [15] and the other involving 121 matched melanoma samples [16]. In each cancer, we derived a consensus mutation network, which was shown to be significantly enriched with known cancer genes and cancer-related functional pathways. Importantly, we not only identified highly recurrently mutated genes, but also well-known yet infrequently mutated genes, thereby demonstrating the usefulness of VarWalker to prioritize driver genes from NGS data.
The detailed description of the VarWalker algorithm is provided in Materials and Methods. It has four steps (Figure 1). The first three steps are implemented within each single sample, and the last step is across multiple samples. In step 1, for each sample, VarWalker assesses the mutation probabilities of all human genes by fitting them to a generalized additive model based on the patient- (or sample-) specific mutational profile. A weighted resample-based test is then performed to filter passenger genes that occur largely due to random events across the genome. Genes occurring with a frequency of ≥5% in random datasets were suggested for filtration. Step 2 includes the execution of the RWR algorithm in each sample to search for the interactions among the filtered MutGenes in the human interactome. RWR has been proven to be sensitive in identifying disease candidate genes and has been successfully applied in disease-phenotype analyses [17], [18]. Here, the introduction of RWR in mutation data analysis reinforces the recognition that driver MutGenes tend to converge in functional pathways and interrupt the same biological processes frequently, while passenger MutGenes are more likely to occur randomly in the genome (as do their interactors in the whole interactome). This recognition enables us to consult both MutGenes and their close interactors and prioritize MutGenes based on their joint frequency. In step 3, considering the complex topological features of human interactome data, we introduce a randomization-based test to evaluate the candidate interactors utilizing 100 topologically matched random networks. Candidate interactors that remain significant (i.e., pedge<0.05) are collected to form a universal candidate pool. This step is also implemented in each sample, respectively. Finally, a consensus mutation subnetwork is constructed (step 4) by collapsing all sample-specific results.
Using the overall implementation principles described above, we rigorously examined several factors that may influence the results as well as several parameter tunings that can potentially improve the performance. Text S1 in the Supporting Information provides a detailed description of these evaluations. We implemented our method in the network data from the Human Protein Reference Database (HPRD), which serves as a good balance between completeness and biological inference.
The Cancer Gene Census (CGC) [19] is a continuous effort to collect cancer genes with mutations that have been causally implicated in cancer. CGC genes are widely used in many cancer studies for benchmark evaluation. We first explored the topological features of CGC genes in HPRD. In our downloaded version (03/15/2012), a total of 487 CGC genes are included, and 369 of them have protein interactions in HPRD. The examination of the distance (measured by the shortest path) among CGC genes and others showed that CGC genes tend to be located more closely to each other than other genes. Specifically, 263 out of the 369 (71.27%) CGC genes are directly connected, 96 (26.02%) have a shortest path of 2 from other CGC genes, and only 10 (2.71%) have a shortest path ≥3 from other CGC genes. In contrast, in the whole HPRD network, 2931 (33.43%) genes (including 263 CGC genes) directly interact with CGC genes, 4657 (53.11%) genes have a shortest path of 2 from CGC genes, 1038 (11.84%) have a shortest path of 3, and the remaining 142 (1.62%) genes have a shortest path >3 from CGC genes. In summary, 97.29% CGC genes are located within two steps from other CGC genes, whereas 86.54% of all human genes are located within this distance. Based on this observation, we conclude that known cancer genes such as CGC genes show a strong tendency to be more closely connected, which is consistent with previous observations that proteins involved in the same disease have an increased tendency to interact with each other [20]. Therefore, we implemented a filtering step to remove genes that are located far away from CGC genes (e.g., those with a shortest path ≥3).
We explored the number of MutGenes that are retained after each step. The largest proportion of MutGenes was removed during mapping of MutGenes onto the HPRD network. This removal resulted from a limitation of the current human PPI data knowledge. Specifically, during removal of genes located two steps away from CGC genes, an average of 88.06% (range: 66.67–100%) were kept in LUAD compared to the previous step. Similarly in melanoma, an average of 86.86% (range: 72.22–100%) were retained compared to the previous step. These results indicate that gene filtration based on distance from CGC genes does not filter a significant proportion of the MutGenes (Figure S2).
We first explored long genes in the two working datasets: a LUAD patient cohort using mutation data from whole-genome sequencing (WGS) and whole-exome sequencing (WES) [15] and a melanoma patient cohort using WES data [16]. The LUAD dataset contains 183 samples, among which 182 had at least one non-silent deleterious mutation. This dataset involves a total of 11,306 MutGenes. A detailed mutational profile can be found in Figures S3 and S4. We manually examined the MutGenes in these samples and observed the frequency-based approach has a strong preference towards long genes. As shown in Figure S5, of the 10 most frequently mutated genes in the LUAD samples, with the exception of TP53 and KRAS, the remaining eight genes are relatively long when compared to the distribution of all human CCDS gene lengths. In contrast, we examined the least frequently mutated genes, i.e., those mutated in one LUAD sample, and surprisingly pinpointed several important cancer genes, including MDM2, RAC1, AKT1, and CDK4. These observations suggest cancer genes could mutate in a broad range of frequency spectrums, making it difficult for the frequency-based filtering approach to be effective.
We then systematically examined the 11,306 MutGenes in the 182 LUAD samples. Among these MutGenes, 6878 were mutated in at least two samples (i.e., “recurrent MutGenes”) regardless of the mutation sites in these genes. Here, recurrent MutGenes differ from recurrent mutations, where the latter are defined as mutations that occur more than once at the same site. We hypothesize that genes that were mutated in only one sample are more likely to have their mutations attributable to random events. We then built two sets of MutGenes. Set one included all 11,306 MutGenes, and set two included all the recurrent MutGenes. We examined the gene length effects in these two MutGene sets by plotting the proportion of MutGenes versus their cDNA length. As shown in Figure 2, both sets have positive correlations with the cDNA length, but the trend was relatively weaker in set two. This analysis revealed that (i) the probability of observing MutGenes is indeed positively correlated with cDNA length, with longer genes more likely to be MutGenes; and, (ii) the correlation is reduced in recurrent MutGenes, yet is nontrivial (Figure 2A), indicating that even in recurrent MutGenes, random mutations still exist.
The same pattern was observed in melanoma samples (Figure 2B). A total of 121 melanoma patients had at least one non-silent deleterious mutation, involving 11,030 MutGenes that have CCDS IDs, 6852 of which were recurrent MutGenes. As shown in Figure 2B, both sets of MutGenes were positively correlated with cDNA length, and the recurrent MutGenes were less correlated, further supporting the necessity to perform gene length-based filtering.
The same procedure that was used in LUAD was applied to the 121 melanoma samples, all of which had MutGenes. Using the same criteria, we constructed a melanoma consensus mutation network, which contains 331 MutGenes involved in 301 interactions. We found that 65 of these 331 MutGenes are CGC genes, indicating a significant enrichment of cancer genes in the network (p-value<2.2×10−16, Fisher's Exact test). Further examination showed 15 kinase proteins in the network, most of which overlapped with CGC genes.
We also validated the melanoma consensus mutation network using somatic mutation data from the TCGA Skin Cutaneous Melanoma (SKCM) project. Many genes in the discovery consensus network were replicated (Table S1). In particular, 86 overlapping genes that account for 25.98% in the discovery dataset and 73.50% in the evaluation dataset were identified, which is significantly higher than expected by chance (p-value<1×10−3, Figure S6). Similar to the case of LUAD, these results demonstrated that cancer-related genes are effectively prioritized by VarWalker.
Functional enrichment analysis of the mutation network revealed many cancer-related signaling pathways (Table S8) and biological processes (Table S9), further indicating that the resultant network is enriched with cancer-related genes and regulation. For example, 12 of the 19 top significant KEGG pathways (pBonferroni<10−6) are cancer-related (Table S8).
We compared our results with those from the single-gene-based strategy. In our application of VarWalker in LUAD, we selected interactions that occurred in ≥10 samples. This approach resulted in 367 genes, 70 of which are CGC genes (70/367 = 19.07%). Using the single-gene-based strategy, we also selected genes that were mutated in ≥10 samples. This step resulted in 426 genes, 16 of which are CGC genes (16/426 = 3.76%), much less than those observed in the consensus mutation network. In melanoma, we also selected interactions that occurred in ≥10 samples, generating a consensus mutation network with 331 genes, 65 of which are CGC genes. Using the single-gene-based strategy, we obtained 404 mutated genes in ≥10 melanoma samples, 23 of which are CGC genes. The proportion of CGC genes obtained by the single-gene-based strategy (23/404 = 5.69%) is also smaller than the proportion obtained by VarWalker (65/331 = 19.64%). These comparisons clearly proved that our network-based approach is superior to the single gene frequency based strategy.
In cancer research, distinguishing between driver mutations, which contribute to tumorigenesis, and passenger mutations, which are mostly neutral and occur randomly, is extremely important to understand and design targeted therapies and treatments. We proposed an approach to prioritize candidate driver MutGenes and biological networks using individual or cohort NGS data. Our method VarWalker estimates the occurrence of mutation events in the genome according to approximated probabilities based on coding gene length. It implements gene-based filtering such that it can exclude genes that are mutated largely due to random events. VarWalker utilizes the Random Walk with Restart algorithm to search for interaction partners that are close to the mutation genes and assesses the resultant interactions using a comprehensive randomization test, thereby greatly reducing potential random interactors (e.g., those with high degrees). In summary, this method has the advantages of both filtering random mutation genes and detecting possible driver genes along with their functional interactions. Hence, it is promising for driver gene prioritization in the era of personalized medicine.
The applications of our method to both LUAD samples and melanoma samples revealed a mutation network for each of them. These mutation networks include a large proportion of known cancer genes and show the interconnections among the protein products of mutant genes. Interestingly, in each of the subgraphs within the consensus mutation network, we observed key components involved in cancer-related signaling pathways and biological processes. For example, in the LUAD mutation network, the three largest subgraphs focused on (i) the EGF receptor signaling pathway, the regulation of nuclear SMAD2/3 signaling pathways, and the p53 signaling pathway; (ii) transmembrane receptors and receptor protein signaling pathways; and, (iii) the cell cycle and DNA repair systems, respectively. The subgraphs in the melanoma mutation network revealed featured pathways such as the Raf/MEK/ERK pathway and receptor signaling pathways (e.g., EGF/EGFR, FGF, PDGFR-beta signaling pathways). The diversity of the component mutation genes in the mutation networks confirms the multifactorial and multigenic mechanisms underlying cancer. These observations also demonstrated the advantages of network-based approaches over frequency-based approaches in prioritizing cancer genes and revealing their functional impacts.
Comparison of the consensus mutation networks of LUAD and melanoma revealed 94 overlapping genes, 33 of which are also CGC genes (Figure S10). We performed a functional enrichment test of these 94 genes (Table S11) and found that most of them are enriched in protein binding categories or cancer-related signaling pathways. The most highly enriched GO terms are involved in enzyme binding (pBonferroni = 2.16×10−13), receptor binding (pBonferroni = 3.03×10−13), phosphatase binding (pBonferroni = 5.85×10−9), and kinase binding (pBonferroni = 1.76×10−6). The most significant pathways include the pathway of “influence of Ras and Rho proteins on G1 to S transition” (pBonferroni = 1.26×10−9), “signaling events mediated by VEGFR1 and VEGFR2” (pBonferroni = 1.74×10−8), and “tumor suppressor Arf inhibits ribosomal biogenesis” (pBonferroni = 1.01×10−7). Collectively, these results suggested that the overlapping genes between LUAD and melanoma mainly function in cell signaling.
The advantages of our approach are threefold. First, in contrast to single-gene-based mutation frequency, our method is based on the joint frequency of two interacting proteins; thus, at the same threshold of frequency, our method can detect moderately or even rarely mutated genes that fail the threshold individually. Second, our interaction-based method helps to filter out many randomly occurring passenger genes, as these genes are expected to be randomly distributed in the network and the chance that their interactors are mutation genes is smaller. Third, our mutation network shows the interactions and context of mutation genes, providing an interpretation to facilitate biological functional analysis in the future, such as further investigation of the novel gene RAC1 in melanoma.
The limitations of our work, which could be improved in future investigations, are reflected in several factors that may impact the results. First, the method is sensitive to the reference network, though it could be flexibly selected. Currently, PPI network resources are comprehensive, but most of them are collected from large scale experiments [28], [29], [30], [31]. Functional correlation networks are valuable when representing biological knowledge and correlations among genes but are generally limited to genes that have already been annotated. As shown in Figure S1, a condensed mutation network was generated from the functional correlation network. This consensus network recruited 22 known LUAD genes, fewer than the 31 known LUAD genes that were recruited in the HPRD-based mutation network. Future expansion of biological networks is expected to improve the detection of mutation networks.
Second, the threshold we used to select interactions, i.e., 10 for both LUAD and melanoma samples, is a trade-off between accuracy and recall rate. Decreasing this threshold value would recruit more cancer genes, but it would also introduce false-positives. Currently, we propose to fit a linear regression model between the number of edges and the edge recurrence. This strategy works in most cases, including the independent TCGA LUAD and SKCM datasets. In practical applications, expert guidance could help to further refine the selection of candidate genes, e.g., in the case of LUAD dataset [16]. In future work, we plan to optimize the selection of interactions by making it threshold-free.
Third, we may improve the mutation recurrence (mr) index through the use of more sophisticated statistical tests and by including protein domain information (details of the mr index is described in Methods and Materials). In our work, we examined the resultant mutation networks either with or without applying the criterion of mr<1.05 in LUAD samples. In the latter case, the recall rate of LUAD genes increased by 4%; however, this application also led to 20% more proteins recruited in the final mutation network and, correspondingly, greatly decreased the specificity. Taken together, the parameter mr performs satisfactorily in our work.
In summary, we present a sample-specific mutation network analysis method to prioritize cancer driver genes using the mutation profiles generated in NGS projects. Our method will be useful for investigators who explore cancer genes through rapidly emerging NGS applications in cancer research and personalized medicine. It can also be applied to explore functional mutations in other complex diseases or traits. The source code in R is available at http://bioinfo.mc.vanderbilt.edu/VarWalker.html.
Figure 1 shows the workflow, which has the following four steps.
Step 1. Patient-specific assessment of MutGenes. The aim of this step is to filter out potential genes whose mutations likely occur by chance based on the patient- (or sample-) specific mutational profile. Note the data and model fitting in this step are both performed for each single sample. As aforementioned, the likelihood of a gene to be mutated in a sample relies on many factors, including both genetic and environmental factors, which makes it impractical to accurately estimate the mutation rate for each gene. Here, instead of a direct estimation, we tackled this problem by formulating a generalized additive model and estimated a relative mutation rate for each gene. Given a cancer sample with MutGenes, let the vector Y denote the mutation status of each CCDS gene, i.e., yi = 1 if the ith gene is a MutGene in the sample and yi = 0 if it is not. A vector of X represents cDNA gene length. We formulate the following model to estimate the probability of a gene to be mutated as a function of its cDNA length, i.e.,where π is the proportion of MutGenes in the investigated samples (i.e., ) and f(.) represents an unspecified smooth function. The function is then solved using a monotonic cubic spline with six knots. Based on the successful fitting of the function, each gene is assigned a weight, which represents its relative probability to be a MutGene (hereafter denoted as probability weight vector, or PWV) and is used as the gene-specific weight in the follow-up weighted resampling process. PWV retains the relative weight of each gene in a particular patient genome and this relative weight changes in different samples.
We then resampled random gene sets to build the null distribution of MutGenes occurring at random. Each random gene set has the same number of MutGenes. In this random selection procedure, each gene was selected from the genome following its probability weight as defined in the sample-specific PWV. The resampling process thus resembles the way in which MutGenes occur in a specific genome in random cases. The weighted resampling process was performed 1000 times in each sample, and a mutation frequency was computed for each gene using . Here, a freq≥5% indicates the gene likely occurs at random and a frequency <5% indicates the gene is highly unlikely to be mutated due to random events. Accordingly, we filter genes with freq≥5%. Upon completion of this step, we obtained a list of significant MutGenes for each sample.
We attempt to fit sample-specific models using MutGenes for each sample such that the heterogeneous background of cancer patients can be properly considered. However, a practical challenge is to determine the minimum number of observations for reliable model fitting. For example, samples with very few MutGenes may not accomplish successful model fitting. Determination of the minimum number of observations remains an open question in statistics. In our case, to avoid arbitrary selection, we compared the results that were obtained using the sample-specific model with those obtained using the universal model. Here, the universal model was generated by using all MutGenes from the cohort. As shown in Figure S11, the difference in the retained MutGenes was large when samples had more MutGenes. We therefore selected 50 as the cutoff. For samples with ≥50 MutGenes [128 (70%) LUAD samples and 110 (91%) melanoma samples, Figure S4], we fitted a sample-specific model and obtained a sample-specific PWV. For other patients with fewer MutGenes, we performed a resample-based test using the universal PWV.
As a positive control, we examined the performance of the resample-based strategy on CGC genes, which are well-studied cancer genes. We found that 96.30% CGC genes had a frequency <5% in random datasets. Only 3.70% CGC genes had a frequency ≥5%. This result indicates our resample-based strategy retains a high sensitivity as evaluated by CGC genes; thus, the filtered genes are more likely randomly-occurring genes. Based on this observation, we created a manual adjustment to always retain CGC genes, even if they were occasionally observed with ≥5% frequency in random datasets. In practice, the users may remove this inclusion criterion.
Step 2. Sample-specific application of the Random Walk with Restart algorithm to search candidate interactors and MutGenes. The RWR algorithm simulates a random walker's transition in the network from a starting node (or a few starting nodes), with pre-defined starting probabilities, to its neighbors until it reaches a stable status. RWR allows for revisiting of the starting node(s) with revisiting probabilities. Given a network G with n nodes, we denote W as the column-normalized adjacency matrix for G; therefore, W is an n×n matrix. The RWR algorithm is formulated as:where r is the restart probability (e.g., r = 0.5 in this study), and p0, pt, and pt+1 are vectors of size n. Each of the three parameters, p0, pt, and pt+1, represents a vector in which the ith element holds the probability that the walker is at node i at time steps 0, t, and t+1, respectively. In general, assuming that there are k initial genes from which the walker would start with equal probability, the initial vector p0 is defined as a vector, with the initial nodes having a probability of 1/k and the remaining nodes having a probability 0, such that the sum of the probabilities equals 1, i.e., , where i = 1,…,n. The RWR function is solved using this iteration process when the difference between pt and pt+1 is below a predefined threshold (e.g., 10−6 in our analyses).
In each patient, we iteratively took each MutGene as the starting point to initiate the random walk and retained the top 1% (i.e., 10) of nodes that have the highest probabilities with which the walker would stay at a stable status as the highly accessible nodes for the initial node. Previous studies suggested various ways to select candidate nodes, e.g., the most accessible node (i.e., top 1) [18], top 5 [38], top 10 [39], [40], top 20 [40], and top 100 [41], but no consensus rules have been made. In this work, we chose to retain the top 10 accessible nodes. Although this selection criterion is arbitrary, our strategy is based on the observation that, in real biological networks, especially PPI networks, each node often has more than one important interactor. For example, TP53 is inhibited by the protein MDM2, but it is activated by ATM, both of which have a direct interaction with TP53 [36]. In such cases, consideration of only the most accessible interactor would overlook other important interactors. Taken together, the number of candidate interactors should not be too small (e.g., 1), as it may miss many important interactors; however, it should not be too large either, as many irrelevant genes may be included. We tested the selection of the top 1, top 5, and top 10 interactors using the data in this study. Based on the assessment, we selected 10 as a balance between choosing too few informative genes (e.g., top 1) and too many genes. However, this criterion can be adjusted depending on the specific data. It is worth noting that these 10 nodes (genes) that are most highly accessible from the starting node (gene) may not always be statistically significant compared to mere chance and will be evaluated in the next step.
Step 3. Randomization-based evaluation of the candidate interactors. To evaluate whether the subnetworks generated by RWR in step 2 do not occur by chance, we generated 100 random networks, each of which maintains the topological characteristics of the original network (e.g., degree of each node). We adapted the switching algorithm proposed by Milo et al. [42], which starts from the observed network and preserves the degree distribution in the generated random network.
We also performed RWR for MutGenes in each of the 100 random networks and we extracted the top 10 nodes with the highest probabilities. For each node encoded by a MutGene, the 10 candidate interactors in the observed network, g1, g2,…, g10, were assessed by computing an empirical p-value: , where π(gi) is a random network in which gi, i = 1,…,10, was found as the top 10 candidate genes to the same initial node. The empirical p-value indicates the probability of a candidate interactor to be selected by chance. The interactors with pedge<0.05 are retained and denoted as significant interactors for the MutGene (see Figure 1).
Step 4. Construction of a consensus mutation subnetwork. After detecting MutGenes and their interactors in each sample, all significant interactions were pooled together, forming a universal candidate pool. This pool enabled us to better incorporate the information across multiple samples. After tabulating all edges, we explored the number of edges versus the edge occurrence (Figure 3). By fitting a linear regression model, we observed that the number of high frequency edges occurred more often than expected. A cutoff was selected according to the distribution (e.g., 10 for melanoma) and was manually adjusted based on expertise (e.g., 10 for LUAD) when necessary. Furthermore, we required both proteins involved in an interaction to be encoded by MutGenes. After pooling all the sample-specific MutGenes and their interactions, we implemented this step such that a pair of MutGenes and its interactor could be either mutated in the same patient or in different patients. In either instance, the interaction would be interrupted.
Next, we defined a parameter called the mutation recurrence (mr) index for each gene, or a pair of genes whose proteins interact, to control the false positive rate. The mr index is defined as , where ‘# all mutations’ refers to mutations occurring in the gene across all samples, and ‘# unique mutations’ refers to the non-redundant set of ‘all mutations.’ Redundancy was determined if two mutations shared the same genomic coordinate regardless of the derived alleles. The introduction of the mr index is based on the observation that mutations in driver genes typically occur in important domains (e.g., kinase domains) and tend to cluster around ‘hotspots’ [43]. In contrast, mutations in passenger genes do not have particular features and may occur randomly across the whole gene. We removed interactions involving MutGenes whose mr<1.05. This cutoff of mr (<1.05) corresponds to MutGenes with >20 non-silent deleterious mutations in the cohort but none shared with any other (i.e., all are unique mutations). This filtering procedure resulted in a pool of high confidence interactions. Then, a consensus mutation network that was frequently mutated or revisited across many samples was derived by selecting the highly recurring interactions according to the overall distribution of the interaction pool.
We used the online tools DAVID [23] and ToppGene [44] for functional analyses. Both tools provide comprehensive resources for biological pathway annotation (e.g., canonical pathways from KEGG [24]) and biological processes (e.g., GO [25] terms). ToppGene also collected information from other databases, including BioCarta, BioCyc, Reactome, GenMAPP, and MSigDB. Wherever applicable, multiple testing correction using the Bonferroni method was performed to control the false discovery rate.
|
10.1371/journal.ppat.1005003 | Existing Infection Facilitates Establishment and Density of Malaria Parasites in Their Mosquito Vector | Very little is known about how vector-borne pathogens interact within their vector and how this impacts transmission. Here we show that mosquitoes can accumulate mixed strain malaria infections after feeding on multiple hosts. We found that parasites have a greater chance of establishing and reach higher densities if another strain is already present in a mosquito. Mixed infections contained more parasites but these larger populations did not have a detectable impact on vector survival. Together these results suggest that mosquitoes taking multiple infective bites may disproportionally contribute to malaria transmission. This will increase rates of mixed infections in vertebrate hosts, with implications for the evolution of parasite virulence and the spread of drug-resistant strains. Moreover, control measures that reduce parasite prevalence in vertebrate hosts will reduce the likelihood of mosquitoes taking multiple infective feeds, and thus disproportionally reduce transmission. More generally, our study shows that the types of strain interactions detected in vertebrate hosts cannot necessarily be extrapolated to vectors.
| Very little is known about how malaria parasite strains interact with each other inside mosquitoes. In this study we show that mosquitoes that have already been infected with one strain of malaria parasites are more likely to become infected with a new strain. Moreover, the presence of an existing infection enhances the replication of malaria parasites with no obvious impact on mosquito survival. Our results illustrate that interactions between strains are important factors in parasite survival and transmission across the whole of their life cycle.
| Interactions between pathogen strains within hosts can be profound and affect many aspects of infectious disease biology, including disease severity and infectiousness, as well as the evolution of virulence and the spread of drug resistance [1–7]. Yet for medically important vector-borne diseases, very little is known about the nature and implications of strain interactions within the vector. This is in striking contrast to what is known about strain interactions in the vertebrate host.
For example, malaria parasites in mixed strain infections experience significant competitive suppression within the vertebrate host [8–17]. Whether competitive suppression also occurs in their mosquito host is unknown. The progression through the vector is relatively long and complex [18] and involves severe population bottlenecks [19]. Parasite density also influences both the development of the parasite and the probability of the vector surviving for long enough to infect a new host [20–22]. Therefore, strain interactions that increase or decrease parasite density are likely to alter the probability of transmission to a new vertebrate host.
Mixed strain (genotype) infections in mosquitoes are common [23,24] and there are three distinct but non-exclusive routes by which they could arise. First, multiple parasite strains could be taken up from a host during a single blood meal. Mixed strain infections are the norm in areas of high transmission [25], and multiple parasite strains can be transmitted to a vector from a single infective feed [26]. Second, mosquitoes that are disturbed during feeding may move to a new host, resulting in multiple hosts contributing blood to one feeding cycle [27–29]. Finally, mosquitoes could feed on different hosts in successive blood feeding cycles. Studies on human and bird malaria parasites have suggested that mosquitoes that take multiple infective feeds have higher oocyst burdens and parasites at different stages of development, which is suggestive of the accumulation of infections over multiple feeding cycles [30–33]. What impact this has on parasite development or on vector survival not been previously tested. If secondary infections are equally likely to be acquired, then of the mosquitoes surviving to become infectious, up to ~40% of infectious mosquitoes could have oocysts, and up to ~17% could have sporozoites originating from multiple feeds (Fig A in S1 Text). The possibility that mosquitoes can acquire mixed infections from multiple feeds is interesting in its own right, but experimentally, infection from successive blood meals would also provide a way to analyse the competitive interactions between strains without the confounding problems of strain recombination. Parasites in the same blood meal freely recombine in the mosquito gut. There can be no recombination between strains acquired in different feeding cycles because zygotes are formed within a few minutes of a blood meal. When a successive meal takes place several days later, all gametes from the first meal are gone [34].
Here we show that mosquitoes can accumulate mixed strain infections from feeding on multiple hosts, and that the presence of oocysts from an existing parasite infection make subsequent infections more likely and more productive. Additionally, we show that vector mortality was no higher for double infections than for infections with a single parasite strain.
An initial study (experiment 1) was conducted to test whether it is possible for mosquitoes to pick up multiple infections from multiple bloodmeals. Six cages, each containing ~100 three to five day old Anopheles stephensi female mosquitoes were used. Half of the cages fed on mice infected with the rodent malaria parasite Plasmodium chabaudi (strain ER), and half received an uninfected blood meal (control). Four days after their initial feed, all cages of mosquitoes received a second blood meal containing P. chabaudi strain AJ parasites. This 4 day schedule corresponds to the preferred blood-feeding frequency for female mosquitoes [35–37]. Seven days after the second blood meal (experimental day 11) when parasites from the second feed were expected to have established as mature oocysts, ~30 mosquitoes per cage were removed, dissected and tested for the presence and density of each of the parasite strains by genotype specific PCR on infected midguts (see Table 1 for treatment groups and sample sizes). A comparison with mosquitoes dissected four days earlier confirmed that our ability to detect infections from the first feed did not decline over this time period (S1 Fig).
We found that mosquitoes become doubly infected with parasites from successive blood meals. A total of 31%(±5.3 SEM) of mosquitoes became infected with ER parasites during their first feed and of these infected mosquitoes 50%(±10.4 SEM) additionally became infected with AJ parasites during their second feed (Fig 1).
We then conducted a second larger study (experiment 2) with 21 cages, again each containing ~100 female mosquitoes. Six cages received two infective blood meals with one each of our two parasite strains (3 x AJ-ER and 3 x ER-AJ), six cages received an infective blood meal only on their first feed (3 x AJ-C and 3 x ER-C), six cages received an infective blood meal only on their second feed (3 x C-AJ and 3 x C-ER), and finally three cages received two uninfected blood-meals (C-C) (Table 1). All cages received two blood meals with mosquitoes in single infection treatments being given an uninfected feed in place of one of the infective blood meals. This was done in order to control for any effect of a second blood meal on parasite replication [38]. This fully factorial study design allowed us to examine how the presence of a co-infecting strain affects parasites that enter the vector first and second, and to test whether co-infection impacts vector survival.
The six cages that received two infective feeds were all found to contain mosquitoes infected with parasites of both strains. For cages which fed on ER first and AJ second, 75%(±4.6 SEM) of mosquitoes became infected on their first feed, and of these 25%(±5.3 SEM) additionally became infected with AJ. For cages that fed on AJ first and ER second, 25%(±4.7 SEM) of mosquitoes became infected on their first feed, and of these 78%(±8.8 SEM) additionally became infected with ER (Fig 1). Therefore both parasite strains were able to establish in already infected vectors.
It was not possible to determine which feed individual oocysts originated from, but by using quantitative PCR we were able to determine the genome count (total number of potential sporozoites produced) for each of our strains within each infected mosquito midgut. The production of sporozoites within the oocyst requires the acquisition of (presumably limited) nutrients from the mosquito [27,39] and has previously been shown to be negatively related to oocyst density [20]. Due to anaemia and immune factors from the vertebrate host, infective bloodmeals are also likely to be lower quality. Therefore, we predicted that the host infection status and/or the establishment of a new infection during oocyst development would negatively impact parasite replication (competitive suppression). However, the host infection status of the second bloodmeal (infective or control) did not affect the number of genomes from the first infection for either of our focal strains (Treatment (infective or control): χ2 = 0.01, p = 0.77; Treatment*Focal strain: χ2 = 0.20,p = 0.66; Fig 2; Table B in S2 Text). When we split our infective treatment group by whether the second infection established or not, we found no effect of secondary infection on AJ (Control vs. Infected: z = 0.24, p = 0.99; Fig 2; Table B in S2 Text), and for ER infections genome numbers were actually slightly higher in mosquitoes which were subsequently infected with AJ (Control vs. Infected: z = 3.49, p = 0.01; Fig 2; Table B in S2 Text). This suggests that the development of established malaria infections is not negatively impacted by secondary infections.
In our first experiment, AJ was used as our focal strain and was more than five times as likely to infect mosquitoes already infected with ER than mosquitoes which had previously received a control feed or had been exposed to ER on the first feed but had not become infected (previous infection status Χ2 = 21.38, p<0.0001; Fig 3; Table C in S2 Text).
In our second experiment, we measured how previous infection affected the establishment of parasites received during the second bloodmeal for both our strains. In agreement with experiment 1, mosquitoes which had become infected during their first feed were much more likely to then become infected on their second feed (infection with focal strain ~ previous infection status Χ2 = 7.09, p<0.01; Fig 3; Table C in S2 Text).
Infection probabilities varied with focal strain and experiment (Fig 3), which was likely due to mice having lower gametocyte densities for AJ infections in experiment 2 (Table A in S2 Text). However, the relative increase in infection probability during a second feed for previously infected mosquitoes remained consistent (previous infection status*focal parasite strain in experiment 2: Χ2 = 0.44, p = 0.80; previous infection status*experiment for AJ: Χ22,7 = 0.99, p = 0.32). Therefore the presence of parasites from a previous infection increased the probability of a new infection for both our focal strains and in replicate experiments.
The observed increase in infection probability during the second bloodmeal for mosquitoes infected during the first could be due to (1) mosquito variation in susceptibility, so that some individuals had a higher likelihood of infection during both feeds, (2) blood-meal quality of the first feed having knock on effects for the second feed (for example, feeding on an anaemic mouse for the first blood-meal could result in mosquitoes taking a larger second blood-meal), or (3) the first infection facilitating the establishment of the secondary infection (either through physical damage to the midgut, changes in resource availability, or immune depletion).
In each of our experiments, mosquitoes where randomly allocated to experimental cages from the same cohort of inbred mosquitoes. It is therefore unlikely that there would be variation in susceptibility between cages, although it is possible that there could be variation in susceptibility between mosquitoes within cages. If there were a subset of mosquitoes refractory to infection in each cage we would expect i) the total number of mosquitoes in each cage to remain constant ii) mosquitoes which failed to become infected during their first feed would be less likely than controls to become infected during a second feed. In both our experiments, cages which received two infectious feeds had an overall higher prevalence of infection from the second feed than in control cages (Χ21,4 = 6.07, p = 0.034), suggesting the increase in susceptibility in these cages was occurring over and beyond the background level of infection. Additionally, previously exposed but uninfected mosquitoes were just as likely to become infected on their second bloodmeal as mosquitoes from control cages (Experiment 1: X2 = 2.04, p = 0.1; Experiment 2: X2 = 2.05, p = 0.2; Fig 3; Table C in S2 Text) and therefore did not represent a refractory subset of individuals.
Differences in blood-meal quality per se are also unlikely to explain increased transmission to already infected mosquitoes: mosquitoes that had previously received a control feed or had received an infective feed but remained uninfected were equally likely to become infected during their second bloodmeal (Control vs. exposed: Experiment 1: X2 = 2.04, p = 0.1; Experiment 2: X2 = 2.05, p = 0.2; Fig 3), and there was no effect of the mouse red blood cell density on probability of infection (Experiment 1: Χ2 = 0.01, p = 0.99; Experiment 2: Χ2 = 0.10, p = 0.75).
By a process of elimination, it seems most likely that the presence of a primary infection directly increases the chance of a secondary infection establishing. In order to determine how this occurs (e.g. whether through interactions with vector immunity, resources, or physical damage to the mosquito midgut) more experiments are needed.
As expected, overall oocyst burdens were higher in mosquitoes that were infected during both bloodmeals compared to mosquitoes infected only on their second bloodmeal. However, the magnitude of this effect depended on the order of strains in the double infections. The highest oocyst burdens were found in mosquitoes with AJ infections followed by ER infections (oocyst density ~ infection status*focal parasite: X2 = 9.22, p<0.005; Fig 4; Table D in S2 Text).
It was not possible to reliably determine which infection individual oocysts resulted from, but we were able to compare genome counts for our focal infections developing in double infections those in matched single infections (controls). Infections that established in already infected mosquitoes had higher genome counts than those that established in previously uninfected (naïve) mosquitoes (X2 = 8.15, p<0.005; Fig 5; Table E in S2 Text). The magnitude of this effect depended on the focal strain (genome count 6 x higher for ER but over 300 x higher for AJ; Fig 5).
Higher genome counts in already infected mosquitoes could have been due to some mosquitoes being more susceptible to both infections, but genome counts from the first and second infections for double infected mosquitoes were unrelated (Χ21,8 = 0.002, p = 0.97; S2 Fig). Therefore, the presence of parasites from a prior infection increases both the chances that subsequent infection will establish, and the density that subsequent infection will reach in the mosquito.
The probability that parasites will be transmitted to a new vertebrate host depends both on the ability of the parasite to establish and replicate within the vector and the potential number of infective bites a vector can take, which will depend on how many blood feeding cycles the mosquito survives for. We performed a comprehensive examination of the impact of infection status on vector survival. A total of 1631 mosquitoes across 21 cages were monitored twice daily until death (our longest lived mosquito died 72 days after receiving its first bloodmeal). Three cages fed on uninfected mice during both blood meals (C-C), 12 cages fed on control mice for one bloodmeal and infective mice for the other (C-AJ, C-ER, AJ-C or ER-C), and 6 cages fed on infective mice during both bloodmeals (AJ-ER or ER-AJ) (Table 1). Dead mosquitoes were tested for the presence of infection and identity of the infecting strain(s) using PCR.
There was no significant difference in survival between control uninfected mosquitoes and exposed but uninfected mosquitoes (Χ21,615 = 0.003, p = 0.96), therefore these groups were analysed together giving us 4 groups for comparison (uninfected; infected with AJ; infected with ER; infected with both strains). While PCR of mosquito cadavers allowed us to directly determine infection status (uninfected, infected with AJ, infected with ER, or double infection) for mosquitoes used in survival analysis oocyst counts from dead mosquitoes are not possible. Therefore, a mean oocyst density was calculated from a subset of ~30 mosquitoes per cage which were removed and dissected 7 days after each infective bloodmeal. Dissected mosquitoes were counted as censored points in the survival analysis. Total gametocyte densities were taken as the summed gametocyte density from the two feeds taken by each mosquito and red blood cell density was the mean of the two feeds. Across all groups there were no significant relationships between mosquito survival and red blood cell density in the blood-meals (Χ2 = 0.001, p = 0.97), mean oocyst density (Χ2 = 0.84, p = 0.36), or gametocyte density (Χ2 = 3.04, p = 0.08), therefore these factors were dropped from the statistical models (Table F in S2 Text).
There was a significant effect of infection status on mosquito survival (4 level factor; uninfected, AJ infection, ER infection, double infection; Χ23,891 = 9.53, p = 0.024). However the only significant pairwise comparison was between uninfected mosquitoes and those infected with AJ alone (AJ vs. uninfected: Χ21,673 = 6.5, p = 0.01; ER vs. uninfected: Χ21,810 = 1.05, p = 0.31; AJ vs. ER: Χ21,253 = 0.24, p = 0.62; Double infection vs. uninfected: Χ21,638 = 0.002, p = 0.99; Double infection vs. AJ: Χ21,81 = 0.15, p = 0.70; Double infection vs. ER: Χ21,218 = 0.002, p = 0.97; Fig 6; Table F in S2 Text), and so we conclude that while there was some evidence of clone differences in virulence, there was no evidence that double infections had a greater virulence to the mosquito than single infections (Fig 6). While this initially seems surprising, given that double infections contained more parasites than single infections, it is likely that all the densities within our experiment where low enough to not have a detectable impact on vector survival, particularly under laboratory conditions with ad libitum access to glucose and water [20–22,40].
So far as we are aware, our experiments provide the first conclusive evidence that mosquitoes are capable of accumulating multiple infections over successive blood meals. We found that they are (Fig 1), and furthermore that the presence of parasites from a previous infection facilitates both the establishment and density of subsequent malaria parasite infections (Fig 3, Fig 5) without negatively impacting the replication of the primary infection (Fig 2) or mosquito survival (Fig 6).
Facilitation of establishment and density of secondary infections contrasts with the competitive suppression seen during mixed strain infections in the vertebrate host [9,41]. Previous studies have shown negative density dependence in the production of sporozoites by oocysts, presumably due to resource limitation or apparent competition mediated by the vector immune response [20]. However, parasites in our study are unlikely to have reached the threshold for negative density dependence to impact development (estimated at ~200 oocysts [20]). It is possible that the facilitation we observed is because primary infection leads to structural changes in the mosquito midgut making it easier for a second infection to invade, and/or that the vector’s anti-parasite immune response may be depleted or suppressed by the primary infection, thereby leading to lower ookinete mortality. Another interesting possibility is that parasites respond to cues signalling the presence of another genotype and alter their replication schedules, as can apparently occur in vertebrate infections [9,42]. Changes in vector biting behaviour induced by the primary infection [36], or trade-offs between the duration of oocyst development and sporozoite production, may mean that the fitness-maximizing intrinsic incubation period for malaria parasites is different for parasites sharing the vector with parasites from an existing infection. If this were the case, the higher genome counts from secondary infections could be due to parasites speeding up their replication when entering an already infected mosquito, in order to maximise representation in the salivary glands when the mosquito bites new hosts. Further experiments are required in order to determine how the within-vector environment changes with the establishment of a previous infection and why this increases the probability of a new infection and its density. A good first step would be to track the ookinetes invasion and establishment of oocysts, using fluorescently marked parasites within a previously infected mosquito, and therefore determine at which stage facilitation occurs.
At first glance, our discovery that a primary malaria infection facilitates a subsequent infection contrasts with the finding by Rodrigues et al. that midgut bacteria introduced into the mosquito haemolympth by invading ookinetes prime the vector immune response, reducing the density of subsequent malaria parasite infections [43]. Several differences in experimental protocols may account for the apparent contradiction. For example, overall oocyst loads in our experiments were close to natural infection densities [27,44,45] and much lower than those of Rodrigues et al. (mean ~5 oocysts per midgut in our single infections compared to means of ~15 & ~60 [43]). Perhaps a large number of ookinetes must cross the midgut to generate sufficient bacterial infection to prime a protective anti-Plasmodium effect. Alternatively, our challenge infections were four days after our primary infections. Rodrigues et al. [43] challenged their mosquitoes 7 and 14 days later; perhaps anti-malaria immunity elicited by bacterial invasion takes a week or more to develop. The elegant experimental protocols of Rodrigues et al. were not designed to look at direct interactions between the priming and challenge parasites because they induced early death of primary infections. Some combination of their protocols and ours would make possible the analysis of the outcomes of co-infections initiated further apart in time and at higher parasite densities. We concentrated on infections acquired from successive blood meals because mosquitoes rarely live long enough to transmit infections acquired two or more gonotrophic cycles after the first [35,46,47].
Combined, our results suggest that mosquitoes taking multiple infective bites will disproportionally contribute to onward malaria transmission of individual strains. How often mosquitoes would be expected to take multiple infective feeds in natural transmission settings depends on many other parameters (e.g. biting rate, proportion of infectious hosts, vector survival). Using parameters from Killeen et al. [35] we estimate that without facilitation, ~10–41% of infectious vectors would have oocysts originating from more than one feeding cycle and ~8–17% of infectious mosquitoes would have salivary gland sporozoites originating from multiple blood meals (S1 Text). These estimates are lower bounds; with facilitation these proportions could be much higher. They will be even higher if mosquitoes feed on multiple hosts within a gonadotrophic cycle [27–29], if infected mosquitoes are more likely to blood feed [48], and if infected hosts are more attractive to mosquitoes [49], as has been recorded. Our data are in keeping with the observation that mixed species infections in the field appear to be higher in mosquitoes than would be expected from the single constitutive species prevalence’s, or from the prevalence of mixed infections in humans [4]. Additionally, accumulation of infections multiple feeds could partially explain the lower than expected rates of heterozygous oocysts observed in field studies of P. falciparum [45](as parasites from multiple feeds will not be able to mate).
The controlled experiments reported here are not feasible in natural transmission settings as they require replicate infections in vertebrate hosts with known infection densities, matched time since infection (to control for transmission blocking immunity) and parasite strains which can be tracked by PCR through the mosquito. However, if mosquitoes in the field are accumulating multiple infections over the course of their lives, we predict that older mosquitoes would have a higher prevalence of mixed infections than younger mosquitoes [4]. With tools now available for determining infection diversity [25,45] and rapid estimation of age of field caught mosquitoes [50], this can be tested.
If the facilitation we have demonstrated here occurs in natural transmission settings, there could be significant epidemiological consequences. Control measures reducing prevalence in the vertebrate host, and therefore reducing the likelihood of mosquitoes taking multiple infective feeds, could disproportionally reduce transmission of individual strains – for example of drug resistant parasites. By increasing the proportion of infectious mosquitoes with mixed strain infections it is also likely that the facilitation reported here will increase the rates of mixed infections in vertebrate hosts which could have implications for infection virulence and the spread of resistant strains [1,51].
More generally, our results point to contrasting effects of mixed strain infections during the malaria lifecycle – while different parasite strains competitively suppress each other in the vertebrate host [6,9,52,53], we have found that they facilitate each other in the mosquito. The potential epidemiological and evolutionary consequences of this antagonism and synergy could be investigated using mathematical models of malaria populations.
The two wild type Plasmodium chabaudi parasite strains (AJ and ER) used here were originally collected from thicket rats (Thamnomys rutilans) in the Congo [54], maintained as part of the WHO Registry of Standard Malaria Parasites (The University of Edinburgh) before transportation to Penn State University where they are stored in liquid nitrogen.
Mice in our experiments were 6–10 week old female C57Bl/6 kept on a 12:12 L:D cycle. The mice were fed on Laboratory Rodent Diet 5001 (LabDiet; PMI Nutrition International, Brentwood, MO, USA) and received 0.05% PABA-supplemented drinking water to enhance parasite growth [55]. Infections were established via intraperitoneal (IP) injection with 5x105 parasites. For each transmission, double the number of mice needed were infected 14, 15 or 16 days prior to mosquito bloodmeal. On the day of transmission gametocytemia (proportion of red blood cells containing gametocytes taken from thin blood smears) and red blood cell density (from 2 μL of blood examined by Flow Cytometry, Beckman Coulter Counter; see [56]) was used to calculate the gametocyte density per μL of blood. The mice with infections containing the highest density of gametocytes were selected and anaesthetized with a 5μL IP injection of Ketamine (100 mg/kg) and Xylazine (10 mg/kg) and placed on top of individual mosquito cages for 30 minutes. One mouse was used per feed per cage (experiment 1: 12 mice used for 6 cages; experiment 2: 42 mice used for 21 cages; see Table 1 for treatment groups). As each cage was fed on a different mouse, the density of transmission stages in the blood of each mouse was compared across treatment groups within each experiment, confirming that focal gametocyte densities did not significantly differ (AJ in experiment 1: F1,4 = 2.22, p = 0.21; AJ in experiment 2: F1,4 = 0.05, p = 0.84; ER in experiment 2: F1,4 = 0.71, P = 0.44; see Table A in S2 Text for gametocyte densities in each of the relevant pairwise comparisons). In order to maximise power without increasing the number of animals used, mosquitoes from the cages receiving two infective feeds were used to examine both the effect double infections on both the first and second infection to establish (see Table 1).
Anopheles stephensi larvae were reared under standard insectary conditions at 26°C, 85% humidity and a 12L:12D photo-period. Eggs were placed in plastic trays (25 cm × 25 cm × 7 cm) filled with 1.5 L of distilled water. To reduce variation in adult size at emergence, larvae were reared at a fixed density of 400 per tray. Larvae were fed on ground TetraFin fish flakes and from 10–11 days after egg hatch, pupae were collected daily and placed in emergence cages. The adults that emerged were fed ad libitum on a 10% glucose solution supplemented with 0.05% paraaminobenzoic acid (PABA). Adult female mosquitoes between 3 and 5 days old were equally distributed across all experimental cages with 100–120 female mosquitoes per cage. Experimental cages were given Ad lib access to 10% glucose solution supplemented with 0.05% paraaminobenzoic acid (PABA) apart from in the 24 hours prior to feeding on mice where they were deprived of glucose to increase propensity to blood feed. After both blood-feeds, any visibly unfed females were removed and discarded and mosquitoes were provided with bowls for oviposition. Sample sizes in Table 1 reflect the number of mosquitoes that took full bloodmeals on both occasions they were offered a host.
In order to ensure densities were comparable our focal infections were always assessed after 7 days. This means that when we were testing for an impact on the first infection mosquitoes were dissected at experimental day 7 and when we were testing for an impact on the second infection mosquitoes were dissected at experimental day 11 (7 days after the second bloodmeal on experimental day 4). To determine infection status and density ~30 mosquitoes per cage were removed, killed with chloroform and dissected. Midguts were examined for oocyst presence and intensity and infected guts were then placed individually into 30 μL of chilled PBS within 1.5 mL microtubes. Tubes were maintained on ice prior to storage at -80°C. DNA was extracted from individual mosquito midguts using the E.Z.N.A MicroElute Genomic DNA kit (Omega Bio-Tek) as per manufacturer’s instructions, eluted in a total volume 20 μL and stored at -80°C. Clone specific genome numbers were determined by PCR following the methods in [57].
Cages were checked for dead mosquitoes twice daily until all mosquitoes had died (72 days after receiving their first blood meal). Mosquito cadavers were stored individually in 1.5mL microtubes and immediately frozen at -20°C for short-term storage before being moved to -80°C within two weeks. Parasite DNA was extracted for the mosquito cadavers and the presence and genome count for each strain was quantified using the same methodology as for dissected midguts except for the addition of 2.5μL of BSA per reaction well prior to PCR analysis (10mg/mL Bovine Serum Albumin, New England BioLabs Inc.). BSA was used as pigment found in the eyes of insects has previously been shown to inhibit DNA amplification [58]. A pilot study confirmed previous studies [59], showing BSA was successful at preventing this inhibition. Infection prevalence in dead mosquitoes from each cage strongly correlated with prevalence from dissected mosquitoes confirming our ability to reliably detect parasite infection through this method (R2 = 0.99 for AJ; R2 = 0.96 for ER prevalence; R2 = 0.95 for the mean number of strains per mosquito; S3 Fig).
All analysis was performed using R version 3.0.2 (R core team (2013) http://www.R-project.org). Gametocyte densities in the mice used for transmission were calculated by multiplying the gametocytemia by the red blood cell density and were log10 transformed and analyzed using general linear models. The proportion of mosquitoes infected with the focal strain for each group was analyzed using generalized mixed effect models (glmer) with a binomial error structure and cage fitted as a random effect (lme4. R package version 1.0–6). For analysis of infection density within the mosquito, only infected mosquitoes were included and host gametocyte density was fitted as a random effect in models. Oocyst densities were analysed using glmer with a poisson error structure and sporozoite densities were log10 transformed and analysed using lmer models. Survival analysis was performed using Cox proportional hazard mixed effect models (Terry Therneau (2012) coxme: Mixed Effects Cox Models. R package version 2.2–3) with experimental cage fitted as a random effect and infection status, estimated total red blood cells in bloodmeals and the mean oocyst density from mosquitoes dissected from the same cage fitted as fixed effects. Total red blood cell density in bloodmeals was estimated from red blood cell densities in the two mice each cage fed on (one per feed) and was included to account for any variation in the quality of bloodmeals received. For all analyses we followed model simplification by sequentially dropping the least significant term and comparing the change in deviance with and without the term to Chi-square distributions until the minimum adequate model was reached. Full details of statistical models can be found in S2 Text and data are deposited in the Dryad repository: (doi:10.5061/dryad.8nr13) [60].
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 protocol was approved by the Animal Care and Use Committee of the Pennsylvania State University (Permit Number: 35790).
|
10.1371/journal.ppat.1004643 | Recognition of Aspergillus fumigatus Hyphae by Human Plasmacytoid Dendritic Cells Is Mediated by Dectin-2 and Results in Formation of Extracellular Traps | Plasmacytoid dendritic cells (pDCs) were initially considered as critical for innate immunity to viruses. However, our group has shown that pDCs bind to and inhibit the growth of Aspergillus fumigatus hyphae and that depletion of pDCs renders mice hypersusceptible to experimental aspergillosis. In this study, we examined pDC receptors contributing to hyphal recognition and downstream events in pDCs stimulated by A. fumigatus hyphae. Our data show that Dectin-2, but not Dectin-1, participates in A. fumigatus hyphal recognition, TNF-α and IFN-α release, and antifungal activity. Moreover, Dectin-2 acts in cooperation with the FcRγ chain to trigger signaling responses. In addition, using confocal and electron microscopy we demonstrated that the interaction between pDCs and A. fumigatus induced the formation of pDC extracellular traps (pETs) containing DNA and citrullinated histone H3. These structures closely resembled those of neutrophil extracellular traps (NETs). The microarray analysis of the pDC transcriptome upon A. fumigatus infection also demonstrated up-regulated expression of genes associated with apoptosis as well as type I interferon-induced genes. Thus, human pDCs directly recognize A. fumigatus hyphae via Dectin-2; this interaction results in cytokine release and antifungal activity. Moreover, hyphal stimulation of pDCs triggers a distinct pattern of pDC gene expression and leads to pET formation.
| While plasmacytoid dendritic cells (pDCs) are known to be important immune cells involved in protection from viruses and tumors, their role in protection against fungal infections is less clear. Our laboratory has been studying the interplay between pDCs and the fungal pathogen, Aspergillus fumigatus. Our previous work demonstrated that human pDCs bind to and inhibit the growth of A. fumigatus hyphae. Moreover, depletion of pDCs rendered mice very susceptible to experimental infection with A. fumigatus. Here, we show that Dectin-2, a receptor on pDCs, recognizes A. fumigatus hyphae and contributes to cytokine release and antifungal activity. In addition, using confocal and electron microscopy, we demonstrate that upon contact with hyphae, some human pDCs die and form antimicrobial structures called extracellular traps. Finally, using microarrays, we analyzed human pDC gene expression upon A. fumigatus infection and found distinct patterns including the activation of genes previously associated with viral infections and apoptosis. These results provide new insights into the mechanisms by which pDCs might help the immune system when confronted with a fungal invader.
| Aspergillus fumigatus is an opportunistic fungal pathogen with a worldwide distribution. Exposure typically occurs when airborne spores (conidia) are inhaled into the lungs. If the conidia are not contained, they may swell and germinate into hyphae. Invasive aspergillosis (IA) is seen predominantly in immunocompromised patients and is characterized by hyphal invasion associated with tissue destruction [1]. The relatively weak fungicidal activity of the available therapeutic options contributes to the high mortality rates seen in patients with IA [2]. Other clinical manifestations of aspergillosis result from allergic responses to the fungus. Innate immune responses of phagocytes, particularly neutrophils, are essential for effective host defenses against A. fumigatus. Toll-like receptors (TLRs) and C-type lectin receptors (CLRs) on phagocytes recognize surface ligands on A. fumigatus [3], [4], [5], [6]. Although hyphae grow too large to be phagocytosed, phagocytes spread over the hyphal surface and antifungal activity proceeds via both oxidative and non-oxidative mechanisms. Moreover, dying neutrophils can release DNA and antimicrobial proteins, including calprotectin, as extracellular traps (ETs), which are able to trap hyphal elements [7]. Thus, larger fungal morphotypes, including tissue-invading Aspergillus hyphae, can still be controlled [8]. Macrophages, eosinophils, and mast cells also release ETs [7], [9] although it is unknown whether these cell types can form ETs in response to Aspergillus.
Plasmacytoid DCs (pDCs) rapidly produce copious amounts of type I interferon (IFN) upon stimulation with viruses [10]. In humans, pDCs comprise 0.2%–0.8% of the total peripheral blood mononuclear cells (PBMCs) and express the endosomal Toll-like receptors (TLRs) 7 and 9, but not TLR2, TLR3 or TLR4 any of the cell surface TLRs. Activated pDCs link innate to adaptive immunity by secreting cytokines such as IFN-α and tumor necrosis factor (TNF-α) and by differentiating into mature pDCs with upregulated MHC and costimulatory molecules capable of priming naive T cells [11].
pDCs are widely described to have roles in viral defenses, tumor immunity, autoimmunity, allergy and some bacterial infections [12], [13], [14], [15], [16], [17]. Our group recently described that pDCs detect and respond to A. fumigatus. We demonstrated that unmethylated CpG-rich motifs in A. fumigatus DNA stimulate human pDCs to produce IFN-α [18]. In addition, when incubated with hyphae, human pDCs directly inhibit fungal growth via a mechanism that involves A. fumigatus-induced pDC death and the release of antifungal mediators including calprotectin. Moreover, following stimulation with A. fumigatus hyphae, pDCs release IFN-α and TNF-α via a mechanism that appears to be TLR-independent. Importantly, depletion of pDCs renders mice hypersusceptible to pulmonary and intravenous challenge with A. fumigatus [19]. In another model of fungal infection, mice resistant to pulmonary paracoccidioidomycosis expanded a subpopulation of pDC that secreted TNF-α, TGF- β and IL-6. This resulted in expansion of interferon-γ-, IL-4-, and IL-17-positive effector T cells [20].
In the present study, we further investigated the interaction between human pDCs and A. fumigatus hyphae. As fungal recognition appears to be TLR-independent, we investigated the possible involvement of two C-type lectin receptors, Dectin-1 and Dectin-2, which have been demonstrated to bind to A. fumigatus hyphae [5], [6], [21], [22], [23]. Moreover, as A. fumigatus induces pDC death, we examined whether pDC ETs (pETs) formed following incubation with hyphae. Finally, to gain further insight into the nature of the pDC response to A. fumigatus hyphae, we took an unbiased systems biology approach by profiling pDC gene expression following hyphal challenge. We found that human pDCs directly recognize A. fumigatus hyphae via Dectin-2; this interaction triggers antifungal activity and cytokine release. Following incubation with hyphae, pDCs formed ETs containing citrullinated histone H3. In addition, A. fumigatus stimulation elicited a distinct pattern of pDC gene expression including up-regulation of genes involved in cell activation, cell migration, cytokine and chemokine production, apoptosis and other biological processes.
Initial experiments focused on determining which PRRs contributed to the recognition of the hyphal morphotype of A. fumigatus by human blood pDCs. pDCs were incubated with A. fumigatus hyphae for 2 hr at 37°C in the presence of mannans (which blocks mannose receptors) and/or laminarin (which blocks β-1,3-D-glucan receptors). Control wells contained no added polysaccharides or the α-glucan, dextran. pDCs treated with mannans inhibited the association between pDCs and A fumigatus hyphae by greater than 50% (Fig. 1A). In contrast, laminarin or dextran treatment did not significantly alter binding of pDCs to the hyphae although there was a trend towards less binding in the presence of laminarin. There was also a trend towards reduced binding when comparing the combination of laminarin and mannan with mannan alone. The above results suggest that mannose receptors are largely (albeit not solely) responsible for the recognition of A. fumigatus hyphae by pDCs. As human pDCs reportedly express the mannose receptor Dectin-2 but not the β-glucan receptor Dectin-1 [24], [25], we hypothesized that Dectin-2 is a major pDC receptor for A. fumigatus hyphae. Indeed, blocking antibodies directed against Dectin-2 significantly decreased the number of pDCs found in association with hyphae (Fig. 1B). In contrast, blocking antibodies directed against Dectin-1 were not inhibitory. Representative photomicrographs of pDCs incubated with A. fumigatus hyphae in the presence or absence of blocking antibodies to Dectin-2 are shown in Fig. 1C.
To assess the contribution of Dectin-2 to pDC-mediated antimicrobial activity, pDCs were incubated with A. fumigatus hyphae for 2 hr at 37°C in the presence or absence of blocking anti-Dectin-2 antibody. Antifungal activity was measured by the XTT assay. We found that blocking Dectin-2 resulted in a profound reduction in antifungal activity against A. fumigatus hyphae (Fig. 2A). Next, we examined whether Dectin-2 recognition of A. fumigatus hyphae could impact immune responses by triggering cytokine release. pDCs were stimulated with A. fumigatus hyphae for 6 hr at 37°C in the presence of anti-Dectin-2 or anti-Dectin-1 blocking antibodies. As negative and positive controls, pDCs were left unstimulated or stimulated with the TLR9 ligand CpG. Concentrations of TNF-α (Fig. 2B) and IFN-α (Fig. 2C) were measured in the supernatants. We found that release of TNF-α and IFN-α was reduced when the pDCs were blocked with anti-Dectin-2 but not with anti-Dectin-1 antibody.
Transfected B3Z cells were utilized to further examine the role of Dectin-2 in hyphal recognition. Dectin-2 can couple to the Syk-CARD9 innate signaling pathway to activate DCs and regulate adaptive immune responses to fungal infection. Unlike Dectin-1, Dectin-2 couples to Syk indirectly, through association with the FcRγ chain [26]. To assess the ability of Dectin-2 to recognize A. fumigatus and trigger cell activation, we used B3Z cells containing a reporter plasmid for NFAT coupled to the β-galactosidase gene. These cells were also transduced with Dectin-2 alone, Dectin-2 and FcRγ, FcRγ alone or Dectin-2 and a signaling-deficient mutant FcRγ chain. The four cell lines were then stimulated with zymosan (a ligand for Dectin-2) [26], A. fumigatus conidia or A. fumigatus hyphae.
Following 2 hr of hyphal or zymosan stimulation, a significant increase in NFAT reporter activity was seen in B3Z cells that were co-transduced with Dectin-2 and FcRγ chain (Fig. 3). The other B3Z cell lines, including the line expressing the mutant FcRγ chain and Dectin-2, did not increase their signal in response to either A. fumigatus hyphae or zymosan, as determined by NFAT reporter activity. Conidia did not stimulate significant increases in β-galactosidase activity in any of the cell lines tested. In two independent experiments, each performed in duplicate, A. fumigatus hyphae did not stimulate detectable β-galactosidase activity in the absence of cell lines.
It was recently reported that neutrophils sense microbe size and selectively release neutrophil extracellular traps (NETs) in response to large pathogens such as C. albicans hyphae and extracellular aggregates of Mycobacterium bovis [27]. In addition, it was demonstrated that netting neutrophils are major inducers of type I IFN production [28]. This, along with our demonstration that pDC produced IFN-α after binding A. fumigatus hyphae (Fig. 2C and [19]), led us to ask whether pDCs can make extracellular traps following contact with A. fumigatus hyphae. Two complementary techniques, confocal microscopy and scanning electron microscopy (SEM), were used to determine whether extracellular traps are formed by pDC following incubation with A. fumigatus hyphae. For the confocal studies, following a 4 or 6 hr incubation of pDCs with hyphae, the samples were stained for DNA, the pDC specific receptor CD123, and citrullinated histone H3. Unstimulated pDCs had intact nuclear DNA as measured by DAPI staining, labeled brightly with anti-CD123 antibody, and had no detectable staining with antibodies directed at citrullinated histone H3 (Fig. 4A). In contrast, following incubation with A. fumigatus, pDCs that were associated with hyphae exposed disrupted, extracellular DNA that co-localized with citrullinated histone H3 (Fig. 4B-D). When the interactions of pDCs with hyphae were examined by SEM, areas colonized by A. fumigatus showed ET-like structures spread over fungal surfaces (Fig. 5). Following incubation with A. fumigatus hyphae, ET formation was observed in approximately 1% of the pDCs. More precise quantification proved to be problematic; some ETs appeared to be in the process of being formed and for well-formed ETs, it could be difficult to tell whether the ET was from one or more pDCs. The appearance of the observed structures is very similar to that described for NETs [29], [30], [31]. Taken together, these observations strongly suggest that pDCs are able to make ETs upon A. fumigatus infection in vitro and we propose the term pETs (pDC extracellular traps) for these structures.
Activated pDCs link innate to adaptive immunity by secreting cytokines such as IFN-α and TNF-α and by differentiating into mature pDCs with up-regulated MHC and costimulatory molecules capable of priming naive T cells [11]. To begin to better understand the full role of pDCs in defense against fungal infections, we took an unbiased approach by determining the human pDC transcriptome upon challenge with A. fumigatus hyphae. The spectrum of changes in gene expression was examined in pDCs from three blood donors at 2 and 4 hr following incubation with hyphae. Comparative controls included unstimulated and pDCs at 4 hr following stimulation with CpG.
Discriminant microarray analysis demonstrated significant changes in the pDCs transcriptome after 2 and 4 hr of interaction with A. fumigatus hyphae. Of the 53,617 gene probe sets represented on the expression array, we identified a total of 2,309 up-regulated and 1,638 down-regulated genes for pDCs from at least one donor. When we looked for concordant expression for pDCs from all three donors, statistical analyses found 79 regulated genes (44 up-regulated and 35 down-regulated) after 2 hr and 250 regulated genes (179 up-regulated and 71 down-regulated) after 4 hr of pDC-Aspergillus hyphae interaction (S1 Table). Of the 44 genes up-regulated at 2 hr, 12 continued to be up-regulated at 4 hr; of the 35 genes down-regulated at 2 hr, 10 continued to be down-regulated at 4 hr.
In addition, 966 regulated genes in CpG-stimulated pDCs were found, of which 855 were up-regulated and 111 down-regulated. The Venn diagrams (Fig. 6A and B) show the number of up- and down-regulated genes found in each experimental group as well as the overlap between groups. Regulated genes were classified in immune related categories, cell metabolism and other biological process according to the NetAffx (Affymetrix) program (Fig. 6C and D). A heat map of the 250 genes differentially expressed following 4 hr of hyphal stimulation demonstrates the disparate patterns of gene activation following stimulation with hyphae compared to CpG (Fig. 6E). The hierarchical cluster shows a similar pattern of gene expression among the donors but different patterns of gene expression when comparing the unstimulated, Aspergillus-infected and CpG-stimulated groups. In addition, examination of the heat map reveals a large number of genes which were up-regulated in the 4 hr Aspergillus-infected group and CpG-stimulated group but not in the 2 hr Aspergillus-infected group and unstimulated control group.
Within categories such as innate immune receptors, signaling pathways, cytokine and chemokine production, antigen processing and presentation, and cell activation and migration activation, we next examined which individual genes were up- or down-regulated following a 4 hr hyphal stimulation and compared the fold response to that seen with hyphal stimulation for 2 hr as well as CpG stimulation (Table 1). Two genes encoding C-type lectin receptor expression were up-regulated. The highest expression was found for the CLECL1 gene, which encodes a C-type lectin-like protein (also known as DCAL-1). DCAL-1 is highly expressed by DCs and B cells and may act as a T-cell costimulatory molecule [32]. In addition, the gene CLEC2D, which reportedly encodes a natural killer receptor and is also induced on B cells upon viral infection [33], [34], was also up-regulated. In contrast, the gene CLEC12A, previously reported as a negative regulator of granulocyte and monocyte function that is restricted to immature DCs, was down-regulated in pDCs after Aspergillus infection, suggesting the pDCs were activated [35], [36].
There were up-regulated genes involved in STAT (Signal Transducers and Activators of Transcription) pathways, including STAT1, STAT2 and STAT4. In response to type I IFN stimulation, STAT1 forms a heterodimer with STAT2 that can bind the ISRE (Interferon-Stimulated Response Element) promoter. Binding the promoter element leads to an increased expression of interferon-stimulated genes (ISGs) [37]. Expression of type I IFN genes markedly increased in response to CpG stimulation but not to Aspergillus infection (Table 2). However, we found several up-regulated genes involved in type I IFN signaling and/or regulation such as IRF2, DHX58, and HERC5. Besides, we found several up-regulated genes known to be induced in response to either IFN-α or IFN-β stimulation (Table 3).
In addition, the gene MAPKAP3, involved in the TLR signaling pathway, was down-regulated, although TLR7 gene expression was up-regulated upon Aspergillus stimuli. The expression of pDC genes involved in cytokine and chemokine production changed following hyphal stimulation. While the expression of CXCL10, CXCL9, CCR7 and CCL22 was up-regulated, the expression of CXCL3 and CCL20 was down-regulated. In addition, two TNF cytokine family genes were at higher levels in the Aspergillus-infected samples compared with the unstimulated pDCs.
Following antigen recognition and phagocytosis, DCs process antigen and usually migrate to the lymph nodes where the antigen is presented to naive T cells. After Aspergillus infection, the pDCs up-regulated some genes involved in antigen processing and presentation via MHC such as LAMP3, TAP1, TAP2 and SEC16B. Moreover, several genes involved in cell shape, spreading control, cell adhesion and migration were regulated as well (Table 1). Finally, the transcriptome profile of Aspergillus-infected pDCs included many regulated genes involved in apoptosis (Table 4).
Several PRRs have been reported to recognize ligands on A. fumigatus including Dectin-1 (β-glucan), DC-SIGN (galactomannans), and Dectin-2 (α-mannan) [38], [39]. TLR2 and TLR4 also participate in signaling responses against this fungus [40]. The recognition receptors expressed on pDCs have not been well studied. Human pDCs have been shown to express Dectin-2, Siglec-H and DC immunoreceptor (DCIR), but not Dectin-1, mannose receptor, DC-SIGN, Mincle, TLR2 and TLR4 [24], [25], [41], [42]. In addition, human pDCs express some complement and Fc receptors [43]. In the present study, we demonstrated that Dectin-2 is involved in the recognition of A. fumigatus hyphae by human pDCs and that this recognition leads to TNF-α and IFN-α release as well as enhanced antifungal activity by pDCs. While human pDCs express TLR9, we previously demonstrated that release of these cytokines following hyphal stimulation occurred in a TLR9-independent manner [19]. Although Dectin-1 is involved in the recognition of A. fumigatus ligands by other cell types [22], [23] we did not find evidence of its involvement in pDC. The putative involvement of murine Dectin-2 in the release of IFN-α was showed by Seeds et al. [44]. Mannan, a broad blocking reagent against mannose receptors including Dectin-2, inhibited murine pDC IFN-α production in response to inactivated influenza virus. Similar to our findings with human pDCs, an anti-Dectin-1 monoclonal antibody had no effect on IFN-α production by pDC. Moreover, experiments with transfected B3Z cells indicate that Dectin-2 works in cooperation with FcRγ to trigger signaling responses against A. fumigatus hyphae. Similar cooperative interactions between Dectin-2 and FcRγ have been demonstrated using zymosan [26]. In addition, it was recently demonstrated that hyphae stimulated increased IL-17RC expression in neutrophils in a Dectin-2-dependent manner [45]. Taken together, our data strongly support a central role for pDC Dectin-2 in hyphal recognition, antifungal activity and cytokine release.
Notably, when the pDCs were treated either with mannan or anti-Dectin-2 antibody, their association with hyphae was only decreased by about half. This suggests that Dectin-2 is not the only PRR participating in the recognition of A. fumigatus hyphae by human pDCs. Future studies will be needed to determine the identity of the other receptors involved. In addition to Dectin-2, numerous candidate receptors are expressed by pDCs, including Siglec-H and DC immunoreceptor (DCIR) [20], [25], [41]. Interestingly, in the microarray experiments, we found that following A. fumigatus stimulation of pDCs, the highest up-regulated gene encodes for CLECL1 (also known as DCAL-1). Similarly to Dectin-1 and Dectin-2, CLECL1 is a C-type lectin molecule. CLECL1 expression reportedly is restricted to hematopoietic cells, including pDCs [32]. However, it is unknown whether CLECL1 functions as a recognition receptor. A caveat to interpretation of the microarray studies is gene expression of the receptors responsible for hyphae recognition may not be up-regulated following contact of pDCs with A. fumigatus. Indeed, we found this was the case for Dectin-2 as up-regulation of Dectin-2 expression was not seen following fungal stimulation.
In the absence of activating signals, pDCs reportedly undergo spontaneous apoptosis [46], [47]. Our previous report demonstrated that the interaction between human pDCs and A. fumigatus hyphae results in the accelerated death of the pDCs by a mechanism that was partly mediated by fungal gliotoxin secretion but still resulted in antifungal activity [19]. Thus, we asked if the recently described mechanism of cell death known as ETosis, largely described in neutrophils [48] but also reported in other cell types, occurred following the recognition of A. fumigatus hyphae by human pDCs. The dying cells form ETs composed of chromatin decorated with antimicrobial proteins that are able to trap and kill pathogens, including bacteria and fungi, and thus, contribute to extracellular anti-microbial host defense [49], [50], [51]. The different ETs have several features in common, regardless of the type of cells from which they originated, including a DNA backbone with embedded antimicrobial peptides, proteases, and citrullinated histones [7]. The morphotype of the pathogen also appears to influence NET formation. A recent study found that while large hyphae of C. albicans induced NETosis, a mutant of C. albicans that is unable to form hyphae failed to induce NETosis [27]. In our study, following incubation of pDCs with Aspergillus hyphae, many pDCs that spread over hyphae had disrupted DNA and stained strongly positive for citrullinated histone H3. On the other hand, unstimulated pDCs had intact nuclear DNA with no detectable staining with antibodies directed at citrullinated histone H3. These observations suggest that pET formation occurs by mechanisms similar to that described for other types of immune cells, including chromatin decondensation mediated by histone citrullination [52]. Histone hypercitrullination mediates chromatin decondensation and NET formation. When the interactions of pDCs with hyphae were examined by SEM, areas colonized by A. fumigatus showed ET structures that engulfed fungal surfaces. The appearance of pETs is very similar to that described for NETs [29], [30], [31]. A recent study reported that <5% of neutrophils undergo NETosis following incubation with Candida albicans hyphae [27], which is somewhat comparable to our observation that ~1% of pDCs form pETs following incubation with Aspergillus hyphae. While these studies establish that ET formation occurs following stimulation of pDCs with A. fumigatus hyphae, future studies will be needed to determine whether pETs contribute to the antifungal activity of pDCs. Antifungal effects of NETs, albeit not robust, have been reported for A. fumigatus [8], [30], [31].
To better understand the full role of pDC in antifungal defenses, we took an unbiased systems biology approach and investigated the pDC transcriptome profile upon A. fumigatus infection. Comparing 2 and 4 hr time points after infection, we found more genes were regulated at the latter time point. Unfortunately, limitations on yields of pDCs from individual blood donors precluded examination of additional time points. Of interest, several genes that were initially described as being involved in viral infections or virus-induced leukemia [53], [54], [55], [56], [57] were up-regulated in pDCs both after CpG stimulation and A. fumigatus infection. The expression of several genes involved in dendritic cell activation, chemokine production, and antigen presentation and processing supports the hypothesized involvement of pDC in host defense against A. fumigatus. In addition, many genes involved in the apoptotic process were also up-regulated after both CpG stimuli and A. fumigatus infection. Our previous article presented two lines of evidence strongly suggesting that the high rate of pDC cytotoxicity following incubation with A. fumigatus hyphae is at least partially due to secreted factors released by the fungi. First, pDC cytotoxicity was observed when the pDCs and hyphae were separated by a transwell. Second, pDC cytotoxicity was significantly reduced following incubation with hyphae from A. fumigatus strains genetically engineered to be deficient in gliotoxin production [19]. Finally, while early apoptotic cells normally preserve their cell membrane integrity, apoptosis can also progress to secondary necrosis and membrane leakage [58]. Thus, apoptotic gene up-regulation presumably is contributing to the antifungal activity of the dying pDCs and to pET formation as presented in this report.
We previously showed that small quantities of IFN-α are released by pDCs upon stimulation with A. fumigatus hyphae and that mice null for the type I IFN receptor are hypersusceptible to intravenous A. fumigatus challenge [19]. In the present study, we confirmed that IFN-α is released by pDCs upon stimulation with A. fumigatus hyphae. While we did not see significant up-regulation of genes encoding type I IFNs by A. fumigatus-infected pDCs, we did see up-regulation of numerous type I IFN-induced genes. This suggests that hyphae-induced type I IFN release is regulated post-transcriptionally [59], [60] or the microarrays lack the sensitivity to detect relatively small changes in gene expression. In contrast, CpG robustly stimulated up-regulation of genes encoding type I IFNs.
Therefore, our data show for the first time that: 1) Dectin-2 participates in the recognition of A. fumigatus hyphae by pDCs, 2) the interaction between pDCs and A. fumigatus results in the formation of pETs, and 3) a distinctive transcriptional profile is seen following stimulation of pDCs by hyphae. These data add significantly to our knowledge of how pDCs contribute to host defenses in non-viral infections. The challenge will be to apply these findings to infected patients.
All research involving human participants was approved by the University of Massachusetts Medical School’s Institutional Review Board. Written informed consent was obtained from all human participants and all clinical investigations were conducted according to the principles expressed in the Declaration of Helsinki.
RPMI-1640 was obtained from GIBCO (Invitrogen). pDC and Aspergillus media consisted of RPMI-1640 supplemented with 100 U/ml penicillin, 100 U/ml streptomycin, 2 mM L-glutamine, 0.5 mM HEPES, and 1 mM sodium pyruvate. Mannan, laminarin, dextran (molecular weight 473,000) and zymosan were purchased from Sigma-Aldrich. Rat monoclonal anti-Dectin-1 and anti-Dectin-2 blocking antibodies were obtained from Serotec and R&D Systems, respectively. The immunostimulatory CpG 2336 oligonucleotide was synthesized with phosphothioate linkages by Integrated DNA Technologies.
The wild-type A. fumigatus clinical isolate Af293 [61] was obtained from the Fungal Genetics Stock Center. Cultivation of A. fumigatus, harvesting of conidia and growth into swollen conidia and hyphae were performed as in our previous studies with slight modifications [40], [19]. Briefly, fungi were grown on Sabouraud Dextrose Agar slants and conidia were harvested with PBS containing 0.05% Tween 20. The conidia were then vortexed, filtered through a 30-μm nylon mesh, washed, counted and stored in water at 4°C for up to a week. To generate hyphae, conidia were incubated at 21°C for 16 hr in Aspergillus media to swell the conidia, and then an additional 3 hr at 37°C to promote germination.
Human pDCs were isolated from healthy donors as described [19], [62]. Peripheral blood was collected by venipuncture. The blood was anticoagulated with heparin, and the peripheral blood mononuclear cells (PBMCs) were purified by Ficoll-Hypaque density gradient centrifugation. Highly purified human pDCs were obtained from PBMC by two rounds of positive selection using CD304-coated magnetic beads (Miltenyi Biotec, cat 130–090–532) [19]. For the XTT and cytokine release studies, highly purified human pDCs were obtained from PBMC by negative selection using magnetic beads (Miltenyi Biotec, cat 130–097–145). No contaminating PMNs were observed.
A. fumigatus conidia (5 × 104) were plated in flat-bottom 96-well half area plates and grown in Aspergillus media to hyphae of 10–20 μm average length. pDCs (5 × 104) were then added to the hyphae in a final volume of 100 μl pDC media for 2 hr at 37°C. The pDCs were previously treated with laminarin (0.5 mg/mL), yeast mannan (1 mg/mL) or both for 30 min. Dextran (0.5 mg/mL) was used as an irrelevant polysaccharide control. Additional experiments were performed by using anti-Dectin-1 (0.1 mg/mL) and anti-Dectin-2 (0.1 mg/mL) blocking antibodies. Binding was quantified using an inverted microscope (Zeiss) by counting number of human pDCs tightly associated with hyphae in 10 different fields. The cell association index was then calculated by dividing the number of pDCs in association with hyphae by the total number of pDC counted and multiplying this fraction by 100.
Antifungal activity was measured by the XTT assay as described [19], [63]. Briefly, A. fumigatus conidia (5 × 103) were plated in 96-well, half-area plates and grown in pDC media to hyphae of 10–20 μm average length. pDCs (5 × 104) were left untreated or incubated with anti-Dectin-2 antibody for 30 minutes at 4°C and then added to the hyphae in a final volume of 100 μl pDC media. Control wells contained hyphae but no pDCs. Following 2 hr incubation, the pDCs were subjected to hypotonic lysis by three gentle washes with distilled water followed by a 30 min incubation with distilled water at 37°C. Supernatants then were removed, with great care taken not to remove the hyphae. pDC media containing 400 μg/ml of XTT and 50 μg/ml of Coenzyme Q, were added, and the wells were incubated for 2 hr at 37°C. The OD450 and OD650 were then measured, and data were expressed as the percent of antifungal activity according to the published formula [19].
A. fumigatus conidia (5 × 104) were plated in 96-well plates and grown in pDC media to hyphae of 10–20 μm average length. pDCs (5 × 104) were left untreated or incubated with anti-Dectin-2 or anti-Dectin-1 antibodies for 30 minutes at 4°C. pDCs were then added to the hyphae in a final volume of 200 μl pDC media containing voriconazole (0.5 μg/mL) to inhibit fungal overgrowth. Control wells contained pDCs only, pDCs and antibodies, or pDC and CpG. After 6 hr of incubation at 37°C, the supernatants were removed and TNF-α and IFN-α levels were measured by ELISA according to the manufacturers’ protocols (eBioscience for TNF-α; PBL Assay Science for IFN-α).
Transgenic cell lines were used to assess the involvement of Dectin-2 and FcRγ in the recognition of A. fumigatus hyphae. The hybridoma T cell line, B3Z has a reporter for nuclear factor of activated T-cells (NFAT) driven activation of β-galactosidase [64]. B3Z cells were retrovirally transduced with murine Dectin-2, wild-type FcRγ chain, Dectin-2 and wild-type FcRγ chain, or Dectin-2 and a signaling-deficient mutant of FcRγ chain as described by Robinson et al., 2009 [26]. These cell lines were a gift from Caetano Reis e Souza (Immunobiology Laboratory, Cancer Research UK, London Research Institute, England, UK) and obtained from Marcel Wϋethrich (University of Wisconsin, Madison). A. fumigatus hyphae (1 × 105) or conidia (1 × 105) were incubated in 48 well plates with each of the BZ3-derived cells (2 × 105) in RPMI media for 2 hr at 37°C. Control wells contained BZ3-derived cells only and were left unstimulated or were stimulated with zymosan (100 μg/mL). NF-AT activation was measured using a β-galactosidase assay. Media were removed from each well and replaced with 100 μl buffer (PBS, 0.05% Triton X-100, 2 mM magnesium sulfate) followed by incubation for 30 min at 4°C. 50 μl of each lysate were transferred to a well of a 96 well black plate and mixed with 1 ul of 10 mM 4-Methylumbelliferyl β-D-galactoside. Relative fluorescence intensities (RFU’s) were measured using a fluorescence microplate reader (Tecan GENios) at 5 min intervals for 1 hr at 37°C. β-galactosidase activity was calculated at its maximum rate as RFU/min.
Circular tissue culture slides (13 mm diameter) were pretreated with 1% Poly-L-lysine solution (Sigma-Aldrich) and placed in 24-well plates. A. fumigatus conidia (2 × 105) were then added to the wells and germinated in Aspergillus media to hyphae of 10–20 μm average length. The wells were gently washed with PBS and the pDCs (2 × 105) were then added to the hyphae in a final volume of 500 μl pDC media and incubated for 4 or 6 hr at 37°C. The samples were fixed with 2% buffered paraformaldehyde and washed three times with PBS. For immunostaining, specimens were treated as described previously [49]. Briefly, specimens were washed 3 times with PBS, permeabilized for 10 min using 0.5% Triton X-100 in PBS and washed again 3 times with PBS. Subsequently, the samples were blocked with 3% cold water fish gelatin, 5% donkey serum, 1% BSA (w/V), 0.25% Tween 20 in PBS (blocking solution) for 30 min at room temperature, and incubated with primary antibodies directed against histone H3 (citrulline R2 + R8 + R17; ab5103, Abcam) and human CD123 (clone 6H6, eBioscience) diluted in blocking solution overnight at 4°C. After 3 washing steps with PBS, primary antibodies were detected with species-specific secondary antibodies coupled to Alexa Fluor 488- and 568-conjugated secondary antibodies (Life Technologies) diluted in blocking solution, respectively. DNA was visualized with 4′,6-diamidino-2-phenylindole (DAPI; Life Technologies) and slides were mounted with fluorescence mounting medium (Dako). Images were captured with a C1 plus confocal microscope (Nikon Instruments) and a 60x oil immersion objective using the operating software EZ-C1 3.91 (Nikon Instruments). Wavelengths of 405 nm (diode), 488 nm (Argon), and 543 nm (HeNe) were used to excite DAPI, Alexa Fluor 488 (and transmission images), and Alexa Fluor 568, respectively. Images were captured in separate passes to avoid cross talk and are presented as maximum intensity projections from Z-stacks. All images were slightly adjusted for background fluorescence and signal intensity in NIS elements software AR 3.2 (Nikon Instruments).
A. fumigatus conidia (4 × 105) were plated in 18 mm cover slips in 12-well plates and grown in Aspergillus media to hyphae of 10–20 μm average length. pDCs (2 × 105) were then added to the hyphae in a final volume of 1 mL pDC media for 2 and 4 hr at 37°C. After fixation with 2.5% (v/v) glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.2 for 1 hr at room temperature, specimens were contrasted using repeated changes of 0.5% OsO4 in dH2O and 0.05% tannic acid. Specimens were then rinsed in dH2O and dehydrated through a graded series to 100% ethanol and then critical point dried in liquid CO2. The cover slips with the specimens were affixed with carbon tape to the surface of SEM aluminum stubs and first coated with 30 nm of carbon, and further sputter coated with Au/Pd (80/20). The specimens were examined using a FEI Quanta 200FEG MK II scanning electron microscope at 10Kv accelerating voltage. Areas containing pET-like structures were recorded at high magnification.
A. fumigatus conidia (2 × 105) were plated in 48-well plates and grown in pDC media to hyphae of 10–20 μm average length. pDCs (2 × 105) were then added to the hyphae in a final volume of 300 μl pDC media for 2 and 4 hr at 37°C. Unstimulated and CpG-stimulated pDCs were incubated for 2 hr and 4 hr, respectively. The total RNA was extracted with the RNeasy Mini Kit (Qiagen, Hilden, Germany). The quantity of total RNA was measured with a spectrophotometer at 260 nanometers, and the RNA integrity was assessed using an RNA 6000 Nano LabChip Kit on an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, U.S.A.). Total RNA (80–500 ng) was reverse transcribed and the single stranded cDNA was amplified using the Ambion WT Expression Kit (Life Technologies, Inc.). The purified cDNA (5.5 μg) was subsequently fragmented and labeled using the GeneChip WT Terminal Labeling Kit (Affymetrix Inc., Santa Clara, CA, U.S.A.). Labeled cDNA (3.5 ug) was then hybridized to the GeneChip Human Gene 2.0 ST array (Affymetrix, Inc.) using the GeneChip Hybridization Oven 640 (Affymetrix, Inc.) at 60 rotations per minute at 45°C for 16–18 hrs. After hybridization, the arrays were washed and stained according to the Affymetrix protocol using a GeneChip Fluidics Station 450 (Affymetrix). The arrays were scanned using the GeneChip Scanner 3000 (Affymetrix). The data were analyzed using Express Console and Transcriptome Analysis Console (TAC) software (Affymetrix, Inc.). The regulated genes were calculated by dividing the linear intensity value found for each probe from experimental groups (2 or 4 hr of pDC-A. fumigatus interaction and CpG-stimulated pDCs) by the linear intensity value found for each probe from the control group (unstimulated pDCs). We considered 1.7 linear fold changes the cut-off to classify the down-regulated and up-regulated genes [65]. Microarray data were deposited in NCBI under GEO accession number GSE55467.
For comparisons of two groups, means ± standard errors were analyzed by the two-tailed unpaired Student t-test with the Bonferroni correction applied when making multiple comparisons. For comparisons of greater than two groups, significance was determined using the one- or two-way analysis of variance (ANOVA) with Tukey’s multiple correction. Calculations were performed using statistical software (GraphPad Prism 5). Statistical significance was defined as P<0.05 following corrections. For the microarray analysis, Transcriptome Analysis Console (TAC) software (Affymetrix, Inc.) was used.
|
10.1371/journal.pgen.1000442 | Harmonics of Circadian Gene Transcription in Mammals | The circadian clock is a molecular and cellular oscillator found in most mammalian tissues that regulates rhythmic physiology and behavior. Numerous investigations have addressed the contribution of circadian rhythmicity to cellular, organ, and organismal physiology. We recently developed a method to look at transcriptional oscillations with unprecedented precision and accuracy using high-density time sampling. Here, we report a comparison of oscillating transcription from mouse liver, NIH3T3, and U2OS cells. Several surprising observations resulted from this study, including a 100-fold difference in the number of cycling transcripts in autonomous cellular models of the oscillator versus tissues harvested from intact mice. Strikingly, we found two clusters of genes that cycle at the second and third harmonic of circadian rhythmicity in liver, but not cultured cells. Validation experiments show that 12-hour oscillatory transcripts occur in several other peripheral tissues as well including heart, kidney, and lungs. These harmonics are lost ex vivo, as well as under restricted feeding conditions. Taken in sum, these studies illustrate the importance of time sampling with respect to multiple testing, suggest caution in use of autonomous cellular models to study clock output, and demonstrate the existence of harmonics of circadian gene expression in the mouse.
| Circadian rhythms confer adaptive advantages by allowing organisms to anticipate daily changes in their environment. Over the last few years, many groups have used microarray technology to systematically identify genes under circadian regulation. We have extended on these studies by profiling the circadian transcriptome from the mouse liver and two immortalized cell lines at an unprecedentedly high temporal resolution. We identified over 3,000 different transcripts in the mouse liver that cycle with a period length of approximately 24 hours. To our surprise, we also identified two classes of genes which cycle with period lengths of 12 and 8 hours; i.e., harmonics of the circadian clock. Importantly, we were able to identify harmonics in five other tissue types; however, these rhythms were undetectable in disassociated cells. Moreover, harmonics were lost in the liver when mice are subjected to restricted feeding, suggesting that at least one component of circadian harmonics is driven by feeding.
| Circadian rhythms are daily, 24-hour (h) oscillations in physiology and behavior such as food consumption, blood pressure, metabolism, body temperature, and locomotor activity [1],[2]. These rhythms are thought to give an adaptive advantage by allowing an organism to anticipate changes in the environment and regulate physiology accordingly. Moreover, disruptions of circadian rhythms contribute to numerous pathologies including metabolic and cardiovascular disorders, cancer, and aging [3]–[5]. A molecular and cellular clock composed of transcriptional feedback loops generates these oscillations [6]. The central loci of the mammalian clock are two small clusters of hypothalamic neurons called the suprachiasmatic nuclei (SCN), which constitute the master pacemaker that orchestrates rhythmic patterns of behavior and physiology throughout the organism [7]. Remarkably, most tissues in the body also contain autonomous circadian clocks that are necessary for the rhythmic expression of clock output genes [8] and capable of sustained oscillations outside of the body (e.g. [9]). These peripheral clocks are principally regulated by stimuli downstream from the SCN, and are entrained by the SCN via a number of different physiological signals such as glucocorticoid production, core body temperature, or cAMP input (e.g. [7],[10]).
Rhythmic physiology is thought to manifest from the transcriptional output of core oscillator components. Consequently, studies have been performed in several model systems to identify rhythmically expressed genes in both central and peripheral tissues [8], [11]–[21]. One consistent observation is that the vast majority of circadian transcriptional output is tissue-, and not locus-, specific, implying that both local and systemic cues heavily influence circadian output. In order to more fully understand the mechanism by which local and systemic signals translate into rhythms of physiology and behavior, a detailed understanding of the circadian transcriptome is necessary. To address this question, we have developed a high resolution temporal profiling experimental design in which samples are taken every hour for 48 hours and subjected to rigorous statistical analysis. This approach has the capacity to identify rhythmic output genes with precision and accuracy. We applied this method to the study of gene transcription in the liver , an organ system that receives and integrates systemic cues, as well as synchronized NIH3T3 and U2OS cells, conventional models of the autonomous cellular oscillator [22],[23].
Here we report the identification of thousands of circadian transcripts in the mouse liver. Surprisingly, using identical statistical methods dramatically fewer cycling transcripts were identified from two models of the autonomous circadian clock, NIH3T3 and U2OS cells. In addition, we found hundreds of transcripts in the liver that cycle at the second and third harmonic of circadian oscillations. Like circadian genes, these ultradian rhythms are severely dampened in ex vivo hepatocytes. Moreover, these rhythms are shifted in a restricted feeding paradigm, demonstrating their responsiveness to systemic cues.
Wildtype C57BL/6J mice were entrained to a 12 h light, 12 h dark (LD 12∶12) environment before being released into constant darkness. Starting 18 h after the first subjective day (CT18), liver samples from 3–5 mice per time point were collected every hour for 48 h. In parallel, we collected a 48 h time course from two different cellular models of the circadian clock in order to study circadian output in the absence of systemic, circadian cues. After synchronization by forskolin shock, NIH3T3 cells were sampled every hour for 48 h, starting 20 h after synchronization. Likewise, a human osteosarcoma cell line, U2OS, was synchronized with dexamethasone and samples were collected every hour for 48 h, starting 24 h after shock. To confirm that these cells were properly synchronized, parallel cell cultures were transfected either transiently (NIH3T3) or stably (U2OS) with a circadian reporter gene, Bmal1:luciferase (Bmal1:Luc) and imaged every 10 minutes for several days to validate synchronization and rhythmicity (Figure S1). Total RNA was purified from these samples and Affymetrix arrays were used to assess global gene expression.
To account for mode failure, two different statistical algorithms were then used to identify rhythmically expressed transcripts as previously described [24]. The first algorithm, COSOPT [25], measures the goodness-of-fit between experimental data and a series of cosine curves with varying phases and period lengths. p-values are then calculated by scrambling the experimental data and re-fitting it to cosine curves in order to determine the probability that the observed data matches a cosine curve by chance alone. The second algorithm, Fisher's G-test [26], uses Fourier transforms to systematically screen experimental data for sinusoidal components. The probability (and thus, the significance) of any observed periodicity can then be tested using Fisher's g-statistic. Importantly, neither algorithm is sensitive to amplitude nor are they intrinsically biased towards any single period length, and they work with different underlying principles minimizing the risk of mode failure.
These tests were corrected for multiple comparisons post hoc using the method described by Storey and colleagues[27],[28]. Briefly, by examining the distribution of p-values from a given data set, an estimate of the proportion that are truly non-rhythmic can be derived. Using this approach to model the rate of false-discoveries, the p-value for each transcript, which estimates the frequency that a truly null observation will be labeled as significant, can be converted to a more stringent q-value which instead estimates the frequency that significant observations are truly non-rhythmic. At a false discovery rate [27] of <0.05, over 3000 transcripts were found to oscillate by both statistical tests in liver, while fewer than a dozen were found in NIH3T3 and U2OS cells. As expected, the majority of cycling transcripts from liver (and all from NIH3T3 and U2OS cells) had period lengths of approximately 24 h (Figure 1A, Figure 2A and B, Table 1).
Strikingly, there were two additional clusters of genes in liver cycling with a frequency two or three times faster than the circadian clock, a second and third harmonic of circadian gene expression (Figure 1A). We identified 260 transcripts that oscillate with a period length of approximately 12 h, and 63 transcripts with a period of approximately 8 h at a false-discovery rate of <0.05. Traces from the microarray expression data show examples of 24, 12 and 8 h cycling genes (Figure 1B–D). Table 1 summarizes the results of this statistical analysis, while the complete list of all liver cycling genes can be found in Tables S1, S2, S3.
Although 24 h transcriptional rhythms are well characterized, to our knowledge there has been no previous studies that either observe or predict the presence of circadian harmonics. Therefore, we took several steps to validate the results of our microarray studies. First, we tested the possibility that these ultradian rhythms may be variants of a 24 h rhythm. To this end, we re-ran COSOPT on both the 12 h genes (n = 260) and the 8 h genes (n = 63) while restricting the possible period lengths to either circadian or ultradian rhythms (Figure S2). We found that in practically every case ultradian period lengths more successfully fit these data than conventional circadian rhythms (median p-values of 0.001 and 0.002 for 12 h and 8 h datasets, respectively). In contrast, circadian period lengths (i.e. >20 and <28 h) dramatically failed to detect rhythms in these data (median p-values of 0.4 and 1.0 for 12 h and 8 h datasets, respectively).
Second, to verify experimentally the presence of sub-circadian transcriptional rhythms, an independent time course of mouse liver samples was collected and analyzed using quantitative PCR (qPCR). In this experiment, both core clock and sub-circadian genes oscillated with period lengths in agreement with the original microarray study (Figure S3). Typically, 12 h rhythms showed closer agreement between independent experiments than 8 h rhythms, however, in both cases, there is evidence of agreement between microarray and qPCR profiles.
Third, 48 h collections were made from a number of different tissues, and qPCR was used to examine the gene expression of a handful of known 12-h cyclers from the liver. One transcript, Hspa1b, showed clear 12 h transcriptional rhythms in every tissue tested (Figure 3), indicating that the presence of circadian harmonics is not restricted to the liver. Strikingly, the phase of Hspa1b rhythms was nearly identical between tissues, suggesting a common underlying mechanism. At the same time, gene expression analysis of additional 12 h genes shows that many transcripts revert to 24 h periodicity in tissues outside the liver (Figure S4), suggesting that 12 h rhythms are driven by both systemic circadian cues and local, tissue-specific factors. Interestingy, the expression patterns of Hspa5 and Armet show considerable similarity across multiple tissue types, reinforcing the possibility that these genes (and thus their rhythms) are driven by systemic cues.
These 12 h rhythms were not seen in NIH3T3 or U2OS cells (Figure 2, Tables S4 and S5), nor a second tissue, the pituitary gland, analyzed in the same fashion [24]. The novelty of this observation can be explained in part by the statistical power of the current study. Simulations reveal that both 12 and 8 h cycling genes are undetectable by conventional 4 h sampling densities (Figure 4A–B). Moreover, both Fisher's G-test and COSOPT were found to be dramatically underpowered when used at sampling densities less than every 2 h (Figure 4C–D). In contrast, by increasing the frequency of time points to every 2 or 1 h, substantial numbers of additional cycling genes can be detected at low false-discovery rates (Figure 4C–D).
In addition to improving the confidence by which both circadian and sub-circadian genes are identified, a 1 h sampling density increases the precision of phase estimates. At a 4 h resolution, only six different phases can be confidently assigned to circadian genes; in contrast, the current study allows the discrimination of phase differences of as little as 1 h. Consequently, subtle but nonetheless consistent phase differences have been identified between core components of the circadian clock (Figure S5). To extend this result, the expression of all cycling genes was median-normalized and plotted as a heat map (Figure 5). Conventional circadian genes show peak expression levels throughout the day with little bias in their phase (Figure 5A). In contrast, the majority of 12 h genes cluster into a single group with similar phases (Figure 5B). Interestingly, the peak of most 12 h genes coincides with dusk and dawn, suggesting that these genes may anticipate the stress of these daily transitions in light and darkness.
Similarly to transcripts with the circadian rhythm, 12 h rhythmic transcripts are involved in a number of different pathways and processes (Table S6). Ingenuity pathway analysis reveals that a number of 12 h rhythmic genes are integrally involved in regulating cell division and protein processing while 8 h rhythms may be involved in NF-kB signaling and lipid metabolism (Figure S6). Taken as a whole, however, available annotations suggest that sub-circadian rhythms regulate a broad spectrum of cellular physiologies.
Like circadian transcripts, the majority of 12 and 8 h genes in the liver oscillate with between 1.5 and 4-fold amplitude (Figure S7). However, core components of the circadian clock (e.g. Bmal1 and Per2) cycle with exceptionally strong amplitudes, while most 12 and 8 h oscillatory transcripts rarely demonstrate greater than 10-fold amplitudes. In addition, unlike circadian transcripts, many 12 h genes show differences in amplitude between their morning and evening peaks, which suggests the possibility that different physiological signals are responsible for driving the twice-daily peaks of 12 h rhythms (Figure S8).
To test whether 12 h rhythms persist in the cultured cells, we compared their expression profiles in commonly used models of the autonomous circadian clock, NIH3T3 and U2OS cells. Using the same statistical analysis as above, neither 12 nor 8 h rhythms were detected in either cell line and fewer than a dozen (mostly core clock components and first order clock controlled genes) showed clear circadian oscillations (Figure 2, Table S4, S5). This paucity of cycling output genes was not due to poor oscillator function, as clear rhythms in reporter gene expression could be seen in parallel experiments (Figure S1). Most importantly, the RNA expression profiles of core circadian genes showed amplitudes of oscillation in agreement with previous studies of circadian cell lines (Figure 2, online supplemental data) [17],[29],[30]. Although core-clock components oscillated as expected, the amplitude of individual components was dampened relative to their profiles in liver and there was no evidence for sub-circadian rhythms (Figure S9, Tables S4, S5). This result was validated using an independent sample collection and qPCR (Figure S10). When compared to high-resolution circadian profiling of the liver and pituitary [24], these data indicate that NIH3T3 and U2OS cells recapitulate the oscillations of core clock components, but fail to adequately maintain robust circadian transcriptional output seen in vivo (Figure S11).
An obvious caveat to this observation is the possibility that tissue-specific cues may drive sub-circadian oscillations in hepatocytes, but not in fibroblasts or osteosarcoma cells. To examine this, disassociated cultures of primary hepatocytes were prepared from Per2:Luc [9] mice and synchronized with dexamethasone. Real-time imaging of luciferase (Figure 6A) as well as qPCR of core clock genes (Figure 6B–E) demonstrated their oscillations with a period of approximately 24 h. The gene expression pattern of Per2 may reflect its role as an immediate early gene; however, the expression patterns of Bmal1, Dbp and Nr1d1 all suggest the presence of an oscillating 24 h clock. Similar to NIH3T3 and U2OS cells, the amplitude of the core clock genes in this system is dampened relative to samples taken from intact liver in vivo. However, 12 h oscillations were either severely dampened (Figure 6F) or entirely absent (Figure 6G–M, statistical analysis: Table S7). Combined with the results from NIH3T3 and U2OS cells, these data show dampening of both circadian rhythms and their harmonics in three different isolated cellular models. Given the sensitivity of gene expression assays, it is impossible to distinguish between loss of harmonic oscillations and extremely low amplitude cycling, but for practical purposes, these cellular models are not useful for the study of ultradian rhythms.
Food metabolism represents a candidate driver of these cues. To address this, we examined 12 h transcripts in a restricted feeding paradigm. Under normal circumstances, mice feed almost exclusively during the night and generally have a larger meal shortly after lights out [31]. In this restricted feeding design, the availability of food is restricted to an 8 h time window during the subjective day, when mice are normally asleep and not eating. Previous experiments have shown that core clock components in the liver invert their phase by 12 h during restricted feeding [32]. We tested the expression pattern of 12 h genes using quantitative PCR and found that seven of the eight genes dramatically changed their expression patterns in response to restricted feeding, while one transcript became entirely arrhythmic (Figure 7 and data not shown). These genes maintained peak expression at approximately CT26, coinciding with feeding; however, the subjective evening peak was largely absent. Taken as a whole, these data support the hypothesis that at least one component of 12 h rhythms are driven by feeding.
Here we have used genome-scale RNA profiling to identify and compare rhythmic transcripts from mouse liver and two models of the autonomous circadian clock, NIH3T3 cells and U2OS cells, at a 1 h time resolution. To detect rhythmic genes, we have employed a pair of statistical algorithms with different underlying principles to score every transcript for evidence of rhythmicity without bias to period length or amplitude. Our simulations indicate that increasing the sampling resolution of circadian profiling studies dramatically increases the confidence with which cycling genes can be detected and minimizes both false positive and false negative observations (Figure 4). To stimulate use by biologists, these data have been made available to the public by depositing raw data in GEO and using a web-based interface http://bioinf.itmat.upenn.edu/circa/mouse.
It is our hope that this resource will fuel additional investigations into mechanisms of physiological rhythms. For example, these data may be used to identify candidate rhythmic genes which may govern behavioral or physiological rhythms. Alternatively, these data may suggest that a given gene or pathway has a previously unsuspected circadian component to its transcription or mRNA abundance. In either case, the cost of false-positives in our dataset would be considerable in terms of time and resources spent following bad leads.
Therefore, to be most useful to future studies, we have employed the q-value statistic based on the concept of false-discovery rate [27],[28] to estimate the likelihood that a given transcript identified as cycling is in actuality non-rhythmic. q-values for every transcript in this study are available on the web-based interface described above. For the purposes of this manuscript, we have chosen to estimate the total number of cycling genes in each dataset using a q-value threshold of <0.05. In the liver, this confidence level allowed the detection of over 3000 cycling transcripts. Unexpectedly, fewer than a dozen cycling genes were detected using the same statistical paradigm in NIH3T3 and U2OS cells, in contrast with previously published work [17],[29],[30]. We suggest that the increased statistical rigor enabled by higher density profiling has led to both fewer false positives and negatives in detection of oscillating genes. Consistent with this notion, we sampled our data at a 4 h resolution, did not account for multiple testing, and found similar levels of oscillating transcription as reported in previous studies (Table S8). However, when corrected for multiple testing, most of these transcripts are not considered significant at a false-discovery rate <0.05. In other words, they may truly be cycling, but not at that false discovery rate, which allows for only one false positive picked amongst 20 truly cycling transcripts. As the majority of detected cycling genes in these cells are either core clock components or 1st order output genes, we are convinced that these cells will continue to be a fruitful model for studies of circadian clockwork. However, the relative paucity of rhythmic genes in cultured cells is cause for caution regarding studies of circadian output in these systems.
Genetic and epigenetic variations accumulated over many years in vitro may account for the loss of robust circadian output in these cell lines. Additionally, the isolation of these cells from circulating cues normally found in vivo may contribute to this phenotype. Based on the loss of amplitude of clock oscillations we observed in disassociated hepatocyte cultures (Figure 6), we speculate that in vitro techniques to synchronize cultured cells may insufficiently reproduce systemic cues that synchronize and drive rhythmic gene expression in vivo. Recently, Schibler and colleagues have shown that the peripheral clock oscillations are necessary for most circadian output [8]. This elegant study, however, does not address the sufficiency of these autonomous cellular models to generate robust rhythmic transcripts. In combination with the results of the Schibler group, we suggest the possibility that robust circadian output in the liver may depend on the combination of an intact peripheral clock as well as circulating, rhythmic cues found in intact animals.
Surprisingly, during the course of this investigation, we discovered second and third harmonics of circadian gene expression in liver and using qPCR subsequently validated 12 h rhythmic transcription in liver as well as in several other peripheral tissues. Several lines of evidence suggest that these rhythms are driven by systemic, circulating cues rather than distinct self-sustained molecular clocks. First, similar to circadian output in NIH3T3 and U2OS cells (Figure 2), 12 h oscillations are dramatically dampened in ex vivo hepatocytes (Figure 6), consistent with the possibility that external signals synchronize and/or reinforce these rhythms in vivo. Furthermore, systemic cues triggered by restricted feeding substantially change the expression pattern of a subset of these genes by eliminating the evening peak of expression (Figure 7). We speculate that 12 h transcriptional rhythms may be generated by changes in behavior and stress-levels coincident with phase-transitions, and may thus provide an advantage to organisms that need to anticipate dusk and dawn. In this model, two or more physiological rhythms with a 24 h period (e.g. feeding behavior) may integrate to generate ultradian rhythms in peripheral tissues. These cues need not be transcriptional, one could envision a transcriptional rhythm of 24 hours intersecting with an out of phase 24 h RNA degradation rhythm producing apparent 12 h rhythms in transcript levels. Interestingly, a number of proteins involved in mediating endoplasmic reticulum (ER) stress, including hsp70 and the transcription factor XBP1, have been independently shown to oscillate at the protein level with 12 h period lengths (F. Gachon, personal communication). Taken together, these data suggest an attractive hypothesis that feeding behavior and food metabolism may regulate 12 h rhythms via the ER stress machinery.
Our investigations have also shown that at least one gene, Hspa1b, also known as HSP 70-2, a heat shock factor that also regulates many processes including immune system function and metabolism, cycles with a 12 h period in at least six different tissues (Figure 3). These data strongly suggest that ultradian transcriptional rhythms have importance beyond the liver. However, the prevalence of non-24 h rhythms in additional peripheral tissues as well as the extent to which they depend on the tissue-autonomous circadian clock remain open questions and the subject of further investigation. Importantly, these data demonstrate the existence of non-24 h biological rhythms and a screening methodology by which to discover them. Finally, these data emphasize the idea that robust rhythms in vivo are a product of interactions between autonomous circadian clocks and systemic cues that are difficult to replicate in vitro.
Collection of liver time points was performed as previously described [14]. Briefly, 6-week-old male C57BL/6J mice (Jackson) were housed in light-tight boxes and entrained to a 12 h light, 12-h dark schedule for one week before being switched to complete darkness. Starting at CT18, 3–5 mice were sacrificed in the dark per time point. Liver samples were quickly excised and snap-frozen in liquid nitrogen. Mice under a restricted feeding regiment were allowed access to food between ZT1 and ZT9. To prevent hoarding of food, the mice were subject to cage changes twice a day, alternating between feeding and fasting cages. Control animals were similarly handled, with the exception that food was present in both cages. All animal experiments were performed with the approval of the Institutional Animal Care and Use Committee.
Liver and cell samples were homogenized in Trizol (Invitrogen) and RNA was extracted with RNeasy columns using the manufacturer's protocol (Qiagen). RNA expression for the liver and NIH3T3 cells was assayed using Affymetrix Mouse Genome 430 2.0 array and data were extracted using GCRMA implemented in ‘R’. Present/absent calls were made using MAS5 in Expression Console (Affymetrix) for Mouse Genome 430 2.0 arrays; liver arrays had an average of 18,581 present transcripts (41.2%), NIH3T3 arrays had an average of 29,220 present transcripts (64.8%). Samples from U2OS cells were analyzed on Affymetrix Human Gene 1.0 ST arrays and the data was extracted using Expression Console (Affytmetrix). Present/absent calls were made using RMA in Expression Console (Affymetrix) at an exon level; U2OS arrays had an average of 68,169 present transcripts (26.5%). COSOPT and Fisher's G-test were performed as described [24], and the raw data and statistics were complied into an Access database (Microsoft). All .cel files are available from GEO (liver accession = GSE11923, NIH3T3 accession = GSE11922, U2OS accession = GSE13949) and microarray data are available in a web-based interface at http://bioinf.itmat.upenn.edu/circa/mouse/.
1 µg total RNA was used to generate cDNA with the High Capacity cDNA Archive Kit using the manufacturer's protocol (Applied Biosystems). Quantitative PCR reactions were performed using iTaq PCR mastermix (BioRad) in combination with gene expression assays (Applied Biosystems) on a 7800HT Taqman machine (Applied Biosystems). Importin 8 (Mm01255158_m1) was used as an endogeneous control for all experiments. Primer and probe information is available from the manufacturer's webpage: Bmal:Mm00500226_m1, Dbp:Mm00497539_m1, Gramd3: Mm00509320_m1, Gmppb:Mm00626032_g1, Gosr2:Mm00444711_m1, Hsap5:Mm00517691_m1, Hspa1b: Mm03038954_s1, Sec23b:Mm00444887_m1, Ints2:Mm00660825_m1, Yipf5:Mm00834912_g1, Creld2:Mm00513021_m1 (Applied Biosystems). All data were analyzed using RQ manager v1.2 (Applied Biosystems).
NIH3T3 cells (ATCC) were grown to confluence and synchronized with 10 µM forskolin (Sigma). U2OS cells were grown to confluence and schocked with 0.1 µM dexamethasone (Sigma). Transfections were performed with Fugene HD using the manufacturer's protocol (Roche). Primary hepatocytes were extracted from Per2-luciferase mice [9] as previously described [33], cultured on collagen coated plates (BD Biosciences) in DMEM (Gibco) supplemented with 10% FBS (Hyclone) and synchronized with 0.1 µM dexamethasone (Sigma) after 48 h in vitro. Cells were homogenized in Trizol (Invitrogen) and snap frozen in liquid nitrogen at the indicated time points. For lumicycle analysis, cells were cultured in DMEM supplemented with 10% FBS (Hyclone), 10 mM HEPES (Gibco), and 0.1 mM Luciferin, sealed in 35 mm tissue culture dishes and analyzed using a Lumicycle (Actimetrics).
|
10.1371/journal.ppat.1003486 | IL-22 and IDO1 Affect Immunity and Tolerance to Murine and Human Vaginal Candidiasis | The ability to tolerate Candida albicans, a human commensal of the gastrointestinal tract and vagina, implicates that host defense mechanisms of resistance and tolerance cooperate to limit fungal burden and inflammation at the different body sites. We evaluated resistance and tolerance to the fungus in experimental and human vulvovaginal candidiasis (VVC) as well as in recurrent VVC (RVVC). Resistance and tolerance mechanisms were both activated in murine VVC, involving IL-22 and IL-10-producing regulatory T cells, respectively, with a major contribution by the enzyme indoleamine 2,3-dioxygenase 1 (IDO1). IDO1 was responsible for the production of tolerogenic kynurenines, such that replacement therapy with kynurenines restored immunoprotection to VVC. In humans, two functional genetic variants in IL22 and IDO1 genes were found to be associated with heightened resistance to RVVC, and they correlated with increased local expression of IL-22, IDO1 and kynurenines. Thus, IL-22 and IDO1 are crucial in balancing resistance with tolerance to Candida, their deficiencies are risk factors for RVVC, and targeting tolerance via therapeutic kynurenines may benefit patients with RVVC.
| This study disentangles resistance and tolerance components of murine and human C. albicans vaginal infection and introduces the challenging notion of a disease due to a defective tolerance mechanism. Vulvovaginal candidiasis (VVC) and recurrent VVC (RVVC) are two forms of disease that affect a large number of otherwise healthy women. Uncomplicated VVC is associated with several predisposing factors, whereas RVVC, marked by idiopathic recurrent episodes, may be virtually untreatable. Despite a growing list of recognized risk factors, further understanding of anti-Candida host defense mechanisms in the vagina is needed to optimize vaccine development and immune interventions to integrate with, or even replace, antifungal therapy. Indeed, medical treatments that increase host resistance, such as antifungals, are highly effective for individual symptomatic attacks but do not prevent recurrences and there is concern that repeated treatments might induce drug resistance. As tolerance mechanisms are not expected to have the same selective pressure on pathogens, new drugs that target tolerance will provide therapies to which low-virulence pathogens, such as C. albicans, will not develop resistance. This study provides a proof-of-concept that targeting tolerance via therapeutic kynurenines may benefit patients with RVVC.
| Candida species are the causative agents of vulvovaginal candidiasis (VVC) and recurrent VVC (RVVC), two forms of disease that affect a large number of otherwise healthy women [1], [2]. Uncomplicated VVC is associated with several predisposing factors, including antibiotic and oral contraceptive usage, hormone replacement therapy, pregnancy and uncontrolled diabetes mellitus, and it usually responds to treatment. In contrast, RVVC, marked by idiopathic recurrent episodes, may be virtually untreatable. Despite a growing list of recognized risk factors, further understanding of anti-Candida host defense mechanisms in the vagina is needed to optimize vaccine development [3], [4] and immune interventions to integrate with, or even replace, antifungal therapy.
Colonization of the vaginal mucosa by the fungus induces both humoral and Th immunity [5]–[7], with the contribution of epithelial [8] and dendritic cells [6]. Acquired Th1 [9]–[11] and Th17 [12] immunity have been described in murine and human VVC. Thus, multiple effector mechanisms of resistance to the fungus are apparently present in VVC. As IL-22 is known to contribute to antifungal resistance at mucosal surfaces by assuring epithelial integrity [13]–[15], and low levels of IL-22 are associated with chronic and recurrent mucosal candidiasis [16]–[19], a role for this cytokine in vaginal immune resistance, beyond the polymorphonuclear neutrophil's (PMNs) response and alarmins production [20], is likely.
In addition to resistance mechanisms that reduce pathogen burden during infection, tolerance mechanisms that protect the host from immune- or pathogen-induced damage have recently emerged in the area of animal immunity [21], [22]. It has been argued that a high rate of infection, but low virulence, should select for host tolerance, whereas the opposite condition should favor resistance [23]. Therefore, it is not surprising that tolerance is a complementary host defense trait that increases fitness in response to low-virulence C. albicans in the host-Candida symbiosis [24]. Considerable evidence for an association of recurrent episodes of symptomatic infection with immune hyper-reactivity to the fungus [25]–[27] point to the contribution of a deregulated immune reactivity to the pathogenesis of VVC and support a role for immunoregulation in this disease. As a matter of fact, protection from VVC is associated with limited or absent inflammatory responses that will not necessarily cause the elimination of the fungus, whereas symptomatic infection is associated with a heavy vaginal cellular infiltrate of PMNs and a variable degree of fungal presence [8], [28]. A plethora of tolerance mechanisms, despite less clarified than resistance mechanisms, have been described [29], [30]. In murine VVC, CD4+ CD25+ regulatory T (Treg) cells [31], γ/δ T cells [32] and immunoregulatory cytokines, such as IL-10 and transforming growth factor β, have all been demonstrated. Thus, resistance and tolerance are complementary host antifungal defense mechanisms that likely operate in the vaginal mucosa, where the ability to tolerate the fungus implies immune strategies that favor the induction of non-sterilizing protective immunity in an environment permissive for fungal persistence.
Indoleamine 2,3-dioxygenase 1 (IDO1), the rate-limiting enzyme in tryptophan degradation along the kynurenine pathway [33], is a master regulator of antifungal tolerance at mucosal surfaces [34]. Pathogenic inflammatory responses due to IDO1 deficiency account for the inherent susceptibility of mice to aspergillosis [35] and mucosal candidiasis [36], owing to the unopposed inflammatory responses that compromise the host's ability to efficiently oppose fungal infectivity. By regulating the balance between Th17 and Treg cells, IDO1 may not only contribute to local immune homeostasis but also limit the pro-survival and virulence-promoting activity of IL-17A on fungal cells [37]. A role for IDO1 in the genitourinary system seems likely, because of the intense IDO1-specific staining in a number of tissues from the genitourinary system [38]. There is also evidence for IDO1 involvement in persistent genitourinary Chlamydia trachomatis infection [39]. However, whether IDO1 contributes to protective tolerance to C. albicans in the vagina is presently unknown.
In the current study, we evaluated the role of IL-22 and IDO1 in murine and human VVC. We used mice with selective deficiency of IL-22 or IDO1 to explore innate and acquired Th/Treg mechanisms of antifungal protection and patients with VVC and RVVC in which common genetic variants in the IL22 and IDO1 genes were analyzed and correlated with local cytokine production. We found that genetic deficiencies of IL-22 or IDO1 were associated with VVC in mice, due to impaired resistance and tolerance mechanisms to the fungus. Two functional genetic variants in human IL22 and IDO1 were associated with a decreased risk for RVVC and correlated with increased local expression of IL-22, IDO1 and kynurenines. This study demonstrates that IL-22 and IDO1 mediate antifungal resistance and tolerance to C. albicans in the vagina and that their deficiencies are risk factors for RVVC.
IL-22 and IDO1 are key mediators of resistance and tolerance to Candida [13], [36] and other fungi [14], [35] at mucosal surfaces. To assess the role of IL-22 and IDO1 in murine VVC, we intravaginally infected C57BL/6, IL-22- or IDO1-deficient mice with Candida blastospores and evaluated resistance to infection in terms of vaginal histopathology, PMN recruitment in vaginal lavages, expression of chemotactic S100A8 and S100A9 proteins, known to mediate PMN migration in murine VVC [8] and local fungal growth. In C57BL/6 mice, histological analysis revealed the presence of fungal and inflammatory cells infiltrating the vaginal parenchyma with signs of epithelial damage at the early stages (Figure 1A). Robust PMN recruitment (Figure 1B and insets in Figure 1A, dpi 3), significant S100a8 and S100a9 gene expression (Figure 1C) and calprotectin production (Figure 1D) were also observed. Mice eventually controlled the infection, as indicated by a reduction in fungal burden (Figure 1E), PMN number, S100a8 and S100a9 expression, calprotectin level and amelioration of inflammatory pathology at 21 and 42 dpi (Figure 1A–D). The course of the infection was different in IL-22- vs. IDO1-deficient mice and in those mice vs. their respective wild-type counterparts. IL-22-deficient mice were susceptible to C. albicans in the early but not late stages of infection, as indicated by signs of vaginal epithelial damage and inflammation, robust PMN recruitment, high S100a8 and S100a9 gene expression, calprotectin levels and fungal growth at 3 dpi (Figure 1A–E). An opposite pattern of resistance was observed in IDO1-deficient mice, in which resistance to infection was increased early but not late in infection. Indeed, at days 21 and 42 after the infection, mice were unable to restrict fungal growth and inflammation (Figure 1A–E). Wild-type and IL-22-deficient, but not IDO1-deficient mice, showed resistance to re-infection (Figure S1A in Text S1), a memory response requiring an intact T cell compartment (Figure S1B–E in Text S1). These findings suggest that IL-22 mediates early antifungal resistance, whereas IDO1 is required to restrain inflammation during ongoing infection and to provide antifungal memory.
To directly prove the role of IL-22, we evaluated levels of IL-22, as well as of companion cytokines, such as IL-17A and IL-17F, in infection and the consequences of IL-22 inhibition or administration. We found elevated levels of IL-22 through the infection in IDO1-deficient mice in which high levels of IL-17F, but not IL-17A, were also higher as compared to wild-type mice (Figure 1F). Blocking IL-22 (Figure 1G) greatly increased fungal growth in IDO1-deficient mice (Figure 1H), whereas exogenous IL-22 decreased fungal growth in wild- type mice (Figure 1I), a finding confirming that IL-22 mediates antifungal resistance in VVC particularly under conditions of IDO1 deficiency. Experiments in IL-17A- or IL-17F-deficient mice confirmed the superior activity of IL-22 vs. IL-17A in early anticandidal resistance in VVC. Despite both types of mice show a higher fungal burden early in infection, as compared to wild-type mice (Figure 2A), IL-17F-deficient, more than IL-17A-deficient, mice showed signs of inflammation (Figure 2B) and PMN recruitment (Figure 2C and inset in Figure 2B) associated with high levels of IL-17A (Figure 2D). Thus, confirming recent findings [20], the neutrophil response in vaginal candidiasis occurs independently of IL-22 and IL-17F. Interestingly, the fact that IL-17A binds fungal cells in the vagina and affects fungal growth and morphology [37], may account for the numerous hyphae observed in vaginal fluids from IL-17F-deficient mice (inset in Figure 2B).
In gastrointestinal candidiasis, IL-22 is mainly produced by NKp46+ NK1.1low cells, an innate lymphoid cell (ILC) subset expressing the aryl hydrocarbon receptor (AhR) [13]. We searched for IL-22-producing NKp46+ cells in the vagina of C56BL/6 mice along with γδ T cells, also known to produce IL-22 at mucosal surfaces [40]. Both types of cells, with the predominance of γδ T cells, were present in the vagina, as revealed by FACS analysis (Figure 3A). However, NKp46+cells mainly produced IL-22, as shown by intracellular cytokine staining in vitro (Figure 3B) and vaginal immunostaining in vivo (Figure 3C). Indeed, IL-22-producing NKp46+ cells expanded in infection in wild-type but not IL-22-deficient mice (Figure 3C). In contrast, γδ T cells produced IL-17A more than IL-22 (about 15% IL-17A+ vs. 5% IL-22+ cells) (Figure 3B). IL-22, but not IL-17A, production was significantly decreased in AhR-deficient mice (Figure 3D), a finding suggesting that IL-22 is produced by vaginal NKp46+ via AhR. As a matter of fact, not only was AhR expression increased during VVC, particularly in IDO1-deficient mice (Figure 3E) but AhR-deficient mice were also more susceptible to VVC (Figure 3F and G). Thus, much like in the gastrointestinal tract [41], IL-22 produced via AhR may serve as a first-line resistance mechanism against yeast infection at the vaginal level.
IDO1 is known to contribute to Treg cell function in mucosal candidiasis [13] and Treg cells are essential components of immune memory to the fungus [42]. We looked, therefore, for IDO1 protein and gene expressions, kynurenine-to-tryptophan ratios, a valid indicator of IDO1 activity [43], and IL-10-producing T cells in mice with VVC. We found that IDO1 was promptly induced in infection at the protein (Figure 4A) and gene (Figure 4B) expression levels, maintained elevated thereafter and was associated with increased levels of kynurenines, downstream products of IDO1 with immunoregulatory functions [44], [45] (Figure 4C) and increased kynurenine to tryptophan ratio [43] (Figure 4D). Both the kynurenine levels and the kynurenine-to-tryptophan ratio were lower in IDO1-deficient mice (Figure 4C and D). The kynurenines were functionally active as replacement therapy with a mixture of l-kynurenine, 3-hydroxykynurenine and 3-hydroxyanthranilic acid, all molecules downstream of the IDO1 pathway [35], restored immunoprotection to VVC, as indicated by restriction of fungal growth (Figure 4E), amelioration of tissue inflammation (Figure 4F), decreased IL-17A and increased IL-10 production at 21 dpi (Figure 4G). These data suggested that IDO1 mediates the production of tolerogenic kynurenines in VVC. To define the effector mechanism of tolerance in VVC, we evaluated IL-10 and FoxP3 expression in the vaginal parenchyma of re-infected mice. Immunostaining revealed the presence of cells expressing both IL-10 and FoxP3 in wild-type but not IDO1-deficient (Figure 4H), a finding consistent with the levels of IL-10 production (Figure 4G). Interestingly, the kynurenine levels were also lower in IDO1-deficient than wild-type mice after re-challenge (2.2±0.7 vs. 0.5±0.3, wild-type vs. IDO1-deficient mice, 3 days after re-challenge). Thus, IDO1, promptly induced in infection, is apparently required for local immunoregulation via IL-10+ Treg cells.
The increased resistance seen later in infection in IL-22-deficient but not IDO1-deficient mice, prompted us to define mechanisms of resistance that are independent from IL-22 but dependent on IDO1. We evaluated the production and expression of IFN-γ and IL-17A in the vagina and the expression of Th-specific transcription factor genes in purified CD4+ T cells from the draining lymph nodes of re-infected mice. We found that resistance to re-challenge in C57BL/6 or IL-22-deficient mice correlated with high-level production (Figure 5A) and expression (Figure 5B) of IFN-γ and IL-17A. IFN-γ but not IL-17A, production (Figure 5A) and expression (Figure 5B) were instead reduced in IDO1-deficient mice in which Tbet expression was also lower and Rorc expression higher as compared to wild-type mice (Figure 5C). Importantly, the production of IFN-γ and IL-17A by CD4+ T cells isolated from mice during the primary infection (3 dpi) was significantly lower than that observed upon re-challenge (Figure 5A). Therefore, these data indicated that an appropriate Th1/Th17 cell balance is required for the expression of acquired antifungal immunity. Studies in IL-10-deficient mice confirmed that IL-10+ Treg cells essentially control this balance. IL-10-deficient mice, although capable of restraining the fungal growth during the primary infection (Figure 5D) and after re-challenge (Figure S1F in Text S1), were unable to control tissue inflammation (Figure S1G in Text S1) and PMN recruitment (Figure 5E) during the infection, and this was associated with high-level production (Figure 5A) and expression (Figure 5B) of IFN-γ and IL-17A, and with high Tbet and Rorc expressions in CD4+ T cells (Figure 5C). Thus IL-10 is required to restrain immunopathology, to which Th17, more than Th1, cells contribute as revealed by subsequent studies in IFN-γ- or IL-17RA-deficient mice. While more resistant to the infection in the early stage– likely due to the high levels of IL-22 (Figure S2A in Text S1)–both IFN-γ- and IL-17RA-deficient mice progressively become susceptible to infection, as indicated by the failure to restrain fungal growth during the primary infection (Figure S2B in Text S1) or after re-challenge (Figure S2C in Text S1) and to limit inflammation (Figure S2D-F in Text S1). In contrast to IL-17A, the levels of IFN-γ were either absent or greatly reduced (Figure S2G in Text S1), a finding indicating that IFN-γ is a key mediator of acquired resistance to the fungus, to which IL-17RA signaling contributes, as already suggested [13]. Studies in IL-12p40-deficient mice confirmed the combined requirement of Th1 and Th17 cells for optimal antifungal memory resistance, as compared to IL-12p25-deficient or IL-23p19-deficient mice (data not shown).
We know that functional yet balanced reactivity to C. albicans at mucosal surfaces requires both the myeloid differentiation primary response gene 88 (MyD88) and the TIR-domain-containing adapter-inducing interferon-β (TRIF) as well as different upstream innate receptors [13]. We evaluated resistance to the primary infection and re-challenge in MyD88-or TRIF-deficient mice intravaginally infected with the fungus. Early in infection, fungal burden was higher in MyD88-deficient mice and declined thereafter (Figure 6A). The inflammatory response with PMN recruitment observed early in infection resolved later in infection (Figure 6B), at a time when those mice developed resistance to re-infection (Figure 6C). Growth was instead restrained in TRIF-deficient mice early in infection, but fungal growth was observed in the vagina at a later stage (Figure 6A), when mice effectively controlled fungal growth if re-challenged (Figure 6C), but not the associated inflammatory response (Figure 6B). Consistent with these findings, IL-22 and IL-17A were particularly lower in MyD88-deficient mice as compared to TRIF-deficient mice (Figure 6D), whereas IL-17A more than IFN-γ/IL-10 were present in TRIF-deficient mice (Figure 6E). To identify the upstream innate receptors involved in the response, we infected mice deficient in receptors known to recruit the MyD88 (TLR2/TLR6/TLR9), TRIF (TLR3), or both (TLR4) pathways and whose expression was observed in the vagina (Figure S3 in Text S1). Compared to wild-type mice, resistance to primary infection (Figure 6F) and to re-challenge (Figure 6G) was similar in TLR9-deficient or TLR6-deficient mice; it was greatly increased in TLR2-deficient mice but impaired in TLR3- or TLR4-deficient mice. Thus, the MyD88 pathway mainly contributes to antifungal mucosal resistance, through the involvement of TLR4, TLR3 (this study) and the beta-glucan receptor Dectin-1 [46], while the TRIF pathway contributes to tolerance via TLR3 and TLR4.
Because of the above results, we investigated whether genetic variants possibly affecting the functions of IL-22 and IDO1 might influence susceptibility to human VVC and RVVC (Table S1 in Text S1). Within the set of single nucleotide polymorphisms (SNPs) tested (Table S2 in Text S1), we found that the genotype frequencies for rs2227485 in IL22 and rs3808606 in IDO1 were significantly different between controls and RVVC, but not VVC, patients (Figure 7A and Table S3 in Text S1). Specifically, the TT genotype at rs2227485 in IL22 was significantly associated with a decreased risk for RVVC [12.4% in RVVC vs. 22.8% in controls; odds ratio (OR), 0.48; 95% confidence interval (CI), 0.27–0.85; P = 0.01]. Likewise, the TT genotype at rs3808606 in IDO1 also correlated with a minor susceptibility to RVVC (13.8% in RVVC vs. 24.0% of controls; OR, 0.51; 95% CI, 0.29–0.88; P = 0.02).
As dectin-1 mediates IL-22 production in mucosal candidiasis [46], and the early stop-codon mutation Y238X in DECTIN1 (rs16910526) has been shown to predispose to familial RVVC in homozygous individuals [47], we also assessed the distribution of this substitution within our study subject group. As expected, we found that the TG genotype conferred an increased risk for RVVC (24.1% in RVVC vs. 11.8% in controls; OR, 2.38; 95% CI, 1.40–4.06, P = 0.001) (Figure 7A and Table S3 in Text S1). Functional analyses confirmed that protection afforded by carriage of the TT genotype at rs2227485 in IL22 correlated with high levels of vaginal IL-22 and decreased levels of pro-inflammatory IL-17A, TNF-α and calprotectin (Figure 7B and C). Likewise, high levels of IL-22 and decreased levels of IL-17A and TNF-α were observed in women carrying the protective TT genotype at rs3808606 in IDO1 (Figure 7D), and they were associated with enhanced IDO1 expression in vaginal cells (Figure 7E) and increased kynurenine-to-tryptophan ratio (Figure 7F). Altogether, these data provide evidence that common genetic variants leading to enhanced expression phenotypes of IL22 and IDO1 may contribute to protection in RVVC.
This study disentangles resistance and tolerance components of murine and human C. albicans vaginal infection and introduces the challenging notion of a disease due to a defective tolerance mechanism. While some degree of inflammation is required for protection, particularly at mucosal tissues during the transitional response occurring between the rapid innate and slower adaptive responses, progressive inflammation worsens disease and ultimately prevents pathogen eradication [24], [35]. Much like in gastrointestinal candidiasis [13], resistance and tolerance mechanisms in VVC are activated through the contribution of innate and adaptive immune responses, involving distinct modules of immunity, i.e., IL-22 and Th1/Th17 cells for resistance and IL-10+ Treg cells for tolerance. It has already been shown that IL-22 provides antifungal resistance through disparate mechanisms, including i) growth control of the early infecting morphotype, the yeasts, via the induction of antimicrobial peptides with anticandidal activity and ii)epithelial integrity control, via phosphorylation of STAT3 [13], known to be required for limiting damage and inflammation in mucosal candidiasis [48]. Thus IL-22 variations may explain why some patients are at high risk of vaginal yeast infection.
Although many different cell types produce IL-22 [41], intestinal ILCs expressing AhR, now termed ILC3 [49], are known to produce, in addition to other cytokines, IL-22 [50]. ILCs reflect many functions of CD4+ T helper cells but expand and act shortly after stimulation. They play fundamental roles early in response to infection and injury, in the maintenance of homeostasis, and possibly in the regulation of adaptive immunity [51]. We found here that NKp46+ cells expanded in the vagina in infection and produced IL-22, more than IL-17A, likely via AhR. As expression of cytokines by ILCs is regulated by signals provided by epithelial cells in response to microbiota, our finding may provide a mechanistic explanation for the link between microbial dysbiosis and vaginal candidiasis. Indeed, the fact that IL-22 production is driven by commensals may explain how antibiotic therapy and iatrogenic immunosuppression are major predisposing factors in candidiasis and, more generally, how the bacterial-fungal population dynamics impact on vaginal homeostasis and inflammatory diseases.
Mutations in IL17F, IL17R and DECTIN1 in patients with chronic mucocutaneous candidiasis, as well as neutralizing autoantibodies against IL-17 and IL-22 in patients with autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy, directly impair IL-17 and IL-22 immunity [47], [52]–[54]. We found here that IL-22, more than IL-17A, contributes to resistance to VVC. In mice, resistance was abrogated under conditions of IL-22 or IL-17F, more than IL-17A, deficiency, whereas a common genetic variant in human IL22 leading to enhanced production of IL-22, but not IL-17A and TNF-α, conferred protection against RVVC. Thus, in addition to functional polymorphisms in genes coding for mannose-binding lectin [55]–[57], IL-4 [58], and the inflammasome component NALP3 [59] in predisposing to RVVC, our study identified an IL22 variant that is associated with a decreased risk for RVVC and is consistent with the findings obtained relative to DECTIN1 deficiency [47].
To our knowledge, this is the first genetic variant in IL22 found to be associated with a human disease [41], and our finding confirms the important functions of the IL-22–IL-22R pathway in regulating immunity, inflammation and tissue homeostasis, and the therapeutic potential of targeting this pathway in human disease [41].
Considered a master regulator of antifungal resistance and tolerance at mucosal surfaces [34], IDO1 has gained reputation in the field of immune mycology owing to its ability to generate immunomodulatory kynurenines that induce Treg cells suppressing local antifungal T-cell responses, thus favoring fungal persistence [34]. It is known that IDO1 activity exhibits relatively large interindividual variability, in particular under pathological conditions [60], [61] and that there are naturally occurring polymorphisms that transcriptionally regulate the human IDO1 gene [62], [63]. Thus, genetic factors are involved in interindividual variability of IDO expression and/or activity. We found that an IDO1 variant leading to enhanced IDO1 expression and concomitant production of kynurenines was associated with decreased risk for RVVC. IDO1 regulates tolerance to the fungus at the vaginal level, as it ameliorates immunopathology that positively correlated with the magnitude of the immune response. It was indeed required for the generation of IL-10+ Treg cells negatively affecting Th1/Th17 cells. Moreover, IDO1 also favored the induction of optimal antifungal memory resistance, an activity likely due to its ability to limit tissue damage, thus allowing for a higher magnitude and duration of the immune response than would have been otherwise possible. Finally, the finding that IL-22 was up-regulated in condition of IDO1 deficiency indicates that the resistance and tolerance mechanisms are to some extent reciprocally regulated.
Understanding the mechanisms that are critical for host survival is important for the choice of therapeutic approaches. Antifungal therapy is highly effective for individual symptomatic attacks but does not prevent recurrences. In addition to the associated costs that are very high [64], there is concern that repeated treatments might induce drug resistance. Thus medical treatments that increase host resistance, such as antibiotics, place selective pressures on pathogens. As tolerance mechanisms are not expected to have the same selective pressure on pathogens, new drugs that target tolerance will provide therapies to which low-virulence fungi, such as C. albicans, will not develop resistance. Kynurenines appear to fulfill this requirement.
Murine experiments were performed according to the Italian Approved Animal Welfare Assurance A-3143-01 and Legislative decree 245/2011-B regarding the animal license obtained by the Italian Ministry of Health lasting for three years (2011–2014). Infections were performed under avertin anesthesia and all efforts were made to minimize suffering. The experimental protocol was designed in conformity with the recommendations of the European Economic Community (86/609/CEE) for the care and the use of laboratory animals, was in agreement with the Good Laboratory Practices and was approved by the animal care Committee of the University of Perugia (Perugia, Italy). All patients and control subjects were observed at the Microbiology Department, S. Maria della Misericordia Medical Center (Perugia, Italy) and answered a detailed questionnaire reporting social and demographic information, medical history, subjective symptoms for gynecological infections and sexual behavior. The study approval was provided by the University ethics committee (Prot. 2012-028) and informed written consent was obtained from all participants. Enrollment took place between January 2009 and June 2012.
Female C57BL/6 and NOD.SCID (NOD.CB17-Prkdcscid/NCrCrl) mice, 8–10 wk old, were purchased from Charles River (Calco, Italy). Homozygous Il22−/−, Ido1−/−, Ahr−/−, Ifng−/−, Il17ra−/−, Il17a−/−, Il17f−/−, Il10−/−, MyD88−/−, Trif−/−, Tlr2−/−, Tlr3−/−, Tlr4−/, Tlr6−/− and Tlr9−/− mice on the C57BL/6 background were bred under specific pathogen-free conditions at the Animal Facility of the University of Perugia, Perugia, Italy.
Mice were treated subcutaneously with 0.1 mg of β-estradiol 17–valerate (Sigma Chemical Co.) dissolved in 100 µl of sesame oil (Sigma) 48 h before vaginal infection. Estrogen administration continued weekly until completion of the study to maintain pseudoestrus. The estrogen-treated mice were inoculated intravaginally with 20 µl of phosphate-buffered saline (PBS) suspension of 5×106 viable C. albicans 3153A blastospores from early–stationary-phase cultures (i.e., 18 h of culture at 36°C in Sabouraud-dextrose agar with chloramphenicol plates, BD Diagnostics). Re-challenge was performed by intravaginal inoculation of 5×106 blastospores, 3 or 5 weeks after the primary infection. In the vaccine-induced resistance experiments, 5×106 heat-killed C. albicans (HCA), obtained by exposing 1×108 cells/ml at 56°C for 30 min, or live low-virulence mutant cells obtained by mutagenesis [65] were intravaginally injected into estradiol-treated mice, 3 weeks before re-challenge. Control mice received PBS. The time course of infection was monitored in individual mice by culturing 100 µl of serially diluted (1∶10) vaginal lavages on Sabouraud-dextrose agar with chloramphenicol plates. Vaginal lavages were conducted using 100 µl of sterile PBS with repeated aspiration and agitation. CFUs were enumerated after incubation at 36°C for 24 h and expressed as log10 CFU/100 µl of lavage fluid. Quantitative counts of CFU in lavage fluids were evaluated successively in mice anesthetized for each lavage. Cytospin preparations of the lavage fluids were stained with May-Grünwald-Giemsa and observed with a BX51 microscope equipped with a high-resolution DP71 camera (Olympus). IL-22 blockade was achieved as described [13] by intraperitoneally injecting mice with a total of 300 µg of mAb neutralizing IL-22 (clone AM22.3, mouse IgG2a) or isotype control mAb (Sigma-Aldrich) the day of and 1 day after the infection. Recombinant murine IL-22 (BioLegend) was given intravaginally (1 µg/mouse) daily, the day of the infection and 1 and 2 days after. Control mice received PBS. Mice were treated intraperitoneally daily, starting 3 days before and up to 3 days after the infection with PBS or 20 µg/kg of a mixture of l-kynurenine, 3-hydorxykynurenine, and 3-hydroxyanthranilic acid (Sigma-Aldrich).
For isolation of CD4+ cells, iliac and inguinal lymph nodes were aseptically removed and cut into small pieces in cold medium. The dissected tissue was then incubated in medium containing collagenase XI (0.7 mg/ml; Sigma-Aldrich) and type IV bovine pancreatic DNase (30 mg/ml; Sigma-Aldrich) for 30–45 min at 37°C. A single cell suspension was obtained and incubated with CD4 MicroBeads (Miltenyi Biotech) before magnetic cell sorting. Vaginal tissues were chopped into fragments and incubated in 1.3 mM EDTA for 30 min at 37°C, as described [66], followed by digestion for 90 min in collagenase type XI (1 mg/ml). These digested pieces were minced and filtered through a nylon mesh, and the resulting cells were filtered through a 70-µm filter. NKp46+ and γδ T cells were purified as per the manufacturer's instruction, with the mouse anti-NKp46 MicroBead Kit and the TCR γδ T cell isolation kit (Miltenyi Biotec).
Antibodies were as follows: anti-CD3ε (145-2C11), anti-γδ (GL3) (BD PharMingen) and anti-NKp46 (CD335) (eBioscience). All staining reactions were performed at 4°C on cells first incubated with an Fc receptor mAb (2.4G2) to reduce non-specific binding. For intracellular staining, cells were stimulated with PMA/ionomycin, added of brefeldin and then permeabilized with the CytoFix/CytoPerm kit (BD Biosciences) for intra-cytoplasmic detection of IL-17A (clone: eBio17B7, eBioscience), and IL-22 (clone 14.03.01, R&D System). Cells were analyzed with a FACScan flow cytofluorometer (Becton Dickinson) equipped with CELLQuest software.
For histology, the vaginas were removed and immediately fixed in 10% neutral buffered formalin (Bio-optica, Milan) for 24 h. The vaginas were dehydrated, embedded in paraffin, sectioned into 3–4 µm and stained with periodic acid-Schiff reagent. Histology sections were observed using a BX51 microscope equipped with a high-resolution DP71 camera (Olympus).
The level of murine and human cytokines in the lavage fluids were determined by Kit ELISA (R&D Systems). The detection limits of the assays were <10 pg/ml for IL-22, <3 pg/ml for IL-10, <30 pg/ml for IL-17F, <10 pg/ml for IL-17A and <1.6 pg/ml for IFN-γ for murine cytokines and <2.7 pg/ml for IL-22, <15 pg/ml for IL-17A, <15 pg/ml for TNF-α and <4 pg/ml for IL-10 for human cytokines. Data were normalized to total protein levels for each sample as determined using the Bio-Rad Protein assay (Life Science, Bio-Rad Laboratories S.r.l. Milan, Italy) and expressed as pg cytokine/mg total protein. Results represent mean cytokine levels (± s.e.m.) from samples pooled from two similar experiments (n = 3–4 total samples per group). Human and murine calprotectin were determined by ELISA (Hycult biotech, Uden, The Netherlands and Immundiagnostik AG, Bensheim, Germany, for human and murine detection, respectively). Real-time PCR was performed using the iCycler iQ detection system (Bio-Rad) and SYBR Green chemistry (Finnzymes Oy). Vaginas and purified cells from lymph nodes were lysed and total RNA was extracted using RNeasy Mini Kit (QIAGEN, Milan, Italy) and was reverse transcribed with Sensiscript Reverse Transcriptase (QIAGEN) according to the manufacturer's directions. The PCR primers were as described [13]. Amplification efficiencies were validated and normalized against Gapdh. The thermal profile for SYBR Green real-time PCR was at 95°C for 3 min, followed by 40 cycles of denaturation for 30 s at 95°C and an annealing/extension step of 30 sec at 60°C. Each data point was examined for integrity by analysis of the amplification plot. The mRNA-normalized data were expressed as relative mRNA in knockout vs. wild-type mice and infected vs. naïve mice. TLR expression was evaluated in vaginal lysates using TLR specific primers and conditions as described [67]. The normalized CT value for the target amplification (ΔCT, Tlr) was determined by subtracting the average Gapdh CT value from the average Tlr CT value.
The vagina was removed and fixed in 10% phosphate-buffered formalin, embedded in paraffin and sectioned at 5 mm. Sections were then rehydrated and after antigen retrieval in Citrate Buffer (10 mM, pH 6), sections were blocked with 5% BSA in PBS and stained with PE anti-IL-17A,-IFN-γ (XMG1.2),-IL-22, FITC-anti-NKp46 (eBioscience) followed by anti-rabbit TRITC (Sigma). Double staining with FITC anti-IL-10 (JES5-16E3) and rabbit polyclonal to FOXP3 (abcam) was followed by anti-rabbit TRITC. All mAbs were incubated overnight at 4°C. Images were acquired using a fluorescence microscope (BX51 Olympus) with a 20× objective and the analySIS image processing software (Olympus). 4′-6-Diamino-2-phenylindole (DAPI, Molecular Probes) was used to counterstain tissues and to detect nuclei.
Cells from murine vaginas or from human vaginal washes were lysed in 2× Laemmli buffer (Sigma-Aldrich) and the lysates were separated in 10% Tris/glycine SDS gel and transferred to a nitrocellulose membrane. Blots of cell lysates were incubated with mouse anti-human IDO1 antibody clone 10.1 (Millipore Billerica) or rabbit polyclonal anti-murine IDO1 antibodies [61]. Normalization was performed by probing the membrane with mouse-anti-β-tubulin antibody (Sigma-Aldrich). Images were acquired with LiteAblotPlus chemiluminescence substrate (Euroclone S.p.A.) using ChemiDoc XRS and Imaging system (Bio-Rad Laboratories) and quantification was obtained by densitometry image analysis using Image Lab 3.1.1 software (Bio-Rad).
The kynurenine to tryptophan ratio was calculated by relating concentrations of kynurenine and tryptophan determined by HPLC with the internal calibrator 3-nitro-L-tyrosine, as described [43].Chromatography was performed on reversed-phase cartridges LiChroCART RP18 columns, tryptophan was monitored via its fluorescence at 285 nm excitation and 365 nm emission wavelengths, kynurenine was measured by its UV absorption at 360 nm wavelength. The kynurenine to tryptophan ratio (Kyn/Trp) was calculated and expressed as µmol kynurenine/mmol tryptophan.
The study population included Caucasian women who had ≥4 (n = 145) or 1–3 (n = 293) culture-verified symptomatic episodes of a vulvovaginal Candida infection during a 12-month period (Table S1 in Text S1). Control subjects consisted of 263 age-matched healthy Caucasian women with no gynecologic complaints, no history of vaginal Candida infection, and who were currently culture-negative for vaginal pathogens. Exclusion criteria were pregnancy, diabetes mellitus, endocrine or immune deficiency disorders, use of immunosuppressive medications, antibiotics or high estrogen content contraceptives, chemotherapy or prior hysterectomy.
Cervicovaginal samples were obtained from all participants by instilling 3 ml of sterile saline into the posterior vagina, mixing the saline with secretions and withdrawing the solution with a syringe. All vaginal washes were centrifuged at 12,000× g for 10 min to separate the mucus from the PBS wash solution shortly after collection and pellet fractions and immediately frozen at −20°C. Genomic DNA was isolated from the pellet fraction using the QIAamp DNA Mini kit (Qiagen). SNPs were selected either from the literature or based on their ability to tag surrounding variants in the HapMap-CEU population of the International HapMap project, NCBI build B36 assembly HapMap phase III (http://www.hapmap.org). The Haploview 4.2 software was used to select haplotype-based tagging SNPs by assessing linkage disequilibrium blocks from the genes of interest with a pairwise correlation coefficient r2 of at least 0.80 and a minor allele frequency of ≥5% in the HapMap-CEU population. SNPs evaluated are indicated in Table S2 in Text S1. Genotyping was performed as previously described [68], [69]. Primer sequences are available upon request. Each genotyping set comprised randomly selected replicates of sequenced samples and negative controls. Concordant genotyping was obtained for >99% assays.
For the functional assays, and unless stated otherwise, measurements were performed in vaginal washes obtained from at least 10 different women without ongoing symptoms and that had negative culture results for each genotype under study, assessed in quadruplicates.
Student's T-test or analysis of variance (ANOVA) with Bonferroni's adjustment were used to determine statistical significance (P<0.05). The data reported are either from one representative experiment out of three to five independent experiments (western blotting and RT–PCR) or pooled from three to five experiments, otherwise. The in vivo groups consisted of 6–8 mice/group. Data were analyzed by GraphPad Prism 4.03 program (GraphPad Software). Genotype distributions among controls and VVC and RVVC patients were analyzed by Fisher's exact test and P values less than 0.05 were considered as significant. Genotype frequencies were distributed according to the Hardy-Weinberg equilibrium for all SNPs (P>0.05).
|
10.1371/journal.pgen.1007245 | Bombyx mori histone methyltransferase BmAsh2 is essential for silkworm piRNA-mediated sex determination | Sex determination is a hierarchically-regulated process with high diversity in different organisms including insects. The W chromosome-derived Fem piRNA has been identified as the primary sex determination factor in the lepidopteran insect, Bombyx mori, revealing a distinctive piRNA-mediated sex determination pathway. However, the comprehensive mechanism of silkworm sex determination is still poorly understood. We show here that the silkworm PIWI protein BmSiwi, but not BmAgo3, is essential for silkworm sex determination. CRISPR/Cas9-mediated depletion of BmSiwi results in developmental arrest in oogenesis and partial female sexual reversal, while BmAgo3 depletion only affects oogenesis. We identify three histone methyltransferases (HMTs) that are significantly down-regulated in BmSiwi mutant moths. Disruption one of these, BmAsh2, causes dysregulation of piRNAs and transposable elements (TEs), supporting a role for it in the piRNA signaling pathway. More importantly, we find that BmAsh2 mutagenesis results in oogenesis arrest and partial female-to-male sexual reversal as well as dysregulation of the sex determination genes, Bmdsx and BmMasc. Mutagenesis of other two HMTs, BmSETD2 and BmEggless, does not affect piRNA-mediated sex determination. Histological analysis and immunoprecipitation results support a functional interaction between the BmAsh2 and BmSiwi proteins. Our data provide the first evidence that the HMT, BmAsh2, plays key roles in silkworm piRNA-mediated sex determination.
| Sex determination is an essential and universal process for metazoan reproduction and development. Insect sex determination is highly diverse, especially for the primary signal and transductory genes. Mechanism of sex determination in the model lepidopteran insect, Bombyx mori, is largely unknown, although a piRNA, named Fem, has been identified recently as the initial factor. In the current report, we generate somatic mutants for the two silkworm piRNA-bound proteins, BmSiwi and BmAgo3, and identify that the histone methyltransferase BmAsh2 is involved in silkworm sex determination. Loss of BmAsh2 function produces a phenocopy of BmSiwi mutation and induces partial female-to-male sexual reversal. Importantly, we find the co-localization and protein interaction between BmAsh2 and BmSiwi, further supporting critical roles of BmAsh2 in the piRNA-mediated sex determination in B. mori.
| Insect sex determination is highly diverse in different species [1,2]. Destiny of the zygote in Drosophila melanogaster depends on the number of X chromosome [3–5]. Female flies carry two X chromosomes which activate the transcription of Sex-lethal (Sxl) and lead to female sexual development, while a single copy of X chromosome in male flies suppresses Sxl expression to determine male sexual fate [6,7]. Subsequently, the female-specific Sxl protein regulates splicing of transformer (tra), which cooperates with the product of the non-sex-specific transformer 2 (tra2) gene to regulate the alternative splicing of doublesex (dsx) [8,9]. In contrast, the insect WZ sex determination system is found in most lepidopteran insects. For example, in the lepidopteran model insect Bombyx mori, females are heterogametic (WZ), while males are homogametic (ZZ) [10,11]. The B. mori W chromosome exerts a dominant control over sex determination since its presence is sufficient for feminization, and the W chromosome-derived PIWI-interacting RNA (piRNA), named Feminizer (Fem), has been identified as the primary factor for silkworm sex determination [12]. The Fem piRNA is arranged tandemly in the sex determination region of the W chromosome and binds to the PIWI protein BmSiwi to exert its functions [12]. In female silkworms, the Masculinization (BmMasc) gene is transcribed from the Z chromosome and responsible for both sex determination and dosage compensation. The Fem piRNA cleaves the BmMasc mRNA in a ping-pong cycler manner to promote the female-specific transcription of Bmdsx, resulting in the female fate of animals [10]. Inhibition of Fem leads to the production of the male-specific transcript of Bmdsx and up-regulates BmMasc in female embryos, revealing the critical roles of both Fem and BmMasc in the silkworm sex determination process, which is distinct from any other species reported [13–15].
The high diversity of sex determination mechanisms indicates that multiple factors may participate in this pathway. Epigenetic modifications are trans-regulators of gene expression that control germline cell imprinting, X chromosome gene inactivation, and gonadogenesis [16]. The histone 3 lysine 9 (H3K9) demethylase, Jmjd1a, positively regulates the sex determination gene Sry in mice [17]. A lack of Jmjd1a causes the H3K9me2 mark to be retained on the Sry gene and dysregulation of Sox9 and Fox12, resulting in male-to-female sexual reversal, as demonstrated by the appearance of a uterus in the testis [17–20]. In B. mori, siRNA-mediated knockdown of the histone methyltransferase (HMT) DOT1L (H3K79 methyltransferase) abolishes male-specific expression of Imp, an insulin-like growth factor II mRNA-binding protein thought to be a potential regulator of male-specific dsx splicing [21]. More recent researches reveal that the prevalent messenger RNA epigenetic modification, N6-methyladenosine RNA (m6A), controls the alternative splicing of Sxl in Drosophila, thus functions in the sex determination process [22,23]. These cases indicate that epigenetic modifications, including histone methylation, are involved in sex determination. However, whether histone methylation participates in B. mori piRNA-mediated sex determination was previously unknown.
The mechanism of silkworm sex determination has long been in mystery until recent identification of the W-derived Fem piRNA which functions as the initial signal for silkworm sex determination [12]. Multiply genes that potentially function in the silkworm sex determination pathway have been functional investigated since then [24,25]. However, how does piRNA regulate the downstream sex determination genes remain largely unknown. Here we describe that depletion of the piRNA-bound protein BmSiwi causes partial female-to-male sexual reversal, revealing its critical role in silkworm piRNA-mediated sex determination. Furthermore, we find significant down-regulation of three HMTs in BmSiwi mutant. Depletion of BmAsh2, one of the HMTs, causes partial sexual reversal as well as dysregulation of piRNAs, TEs, Bmdsx and BmMasc. We further demonstrate that there is a functional interaction between the BmSiwi and BmAsh2 proteins. In conclusion, our data provides the first evidence that the HMT BmAsh2 plays key roles in the silkworm piRNA-mediated sex determination.
Gonad-specific expression of PIWI subfamily proteins (PIWIs) has been identified in the silkworm as well as other organisms [15,26]. In this study, we used qRT-PCR to confirm the predominant expression of two silkworm PIWIs, BmSiwi and BmAgo3, in gonads at the larval wandering stage (S1A and S1B Fig). The transcript abundance of these two PIWIs was low during the larval stages, increased more than 10-fold after pupation and peaked at the pupal and adult stages in gonads (S1C and S1D Fig). Furthermore, we used immunostaining to investigate the localization of silkworm PIWIs in the gonads at the translational level. Similar to D. melanogaster, B. mori ovary possesses several ovarioles which are composed by sequentially developed egg chambers, and serve as an assembly line for oogenesis [27,28]. In order to distinguish the germline and somatic cells in silkworm ovary, we used a primary antibody recognizing BmVasa, which gene has been described as a conserved molecular marker for germline cells in insects, to perform the immunostaining analysis. As the results, distribution of BmVasa and BmAgo3 presented a circular pattern, surrounding the nucleus of germline cells (Fig 1A). In comparison, BmSiwi localized in both the germline cells and the somatic supporting cells which were not stained by the BmVasa antibody (Fig 1A). Localization of silkworm PIWIs was similar to the products of the orthologous genes in D. melanogaster, suggesting that they may participate in B. mori piRNA regulation (Fig 1A and 1B) [29,30]. In testis, both BmSiwi and BmAgo3 were detected in the spermatogonium and their distribution completely overlapped with BmVasa (S2 Fig). These results indicated that BmPIWIs may function in gonadogenesis.
Using the binary CRISPR/Cas9 system, we established somatic mutant lines for BmPIWIs to explore their comprehensive physiological functions (S3A and S3B Fig) [26,31]. Different types of deletions were detected around the target sites in the F1 progeny obtained when the IE1-Cas9 and U6-sgRNA transgenic lines were crossed, demonstrating efficient mutagenesis of both genes (S3C and S3D Fig). In addition, the depletion efficiency was further confirmed by histological analysis using corresponding antibodies (Fig 2A).
Compared with wild-type (WT) animals, the larval ovaries from Δsiwi and Δago3 animals were oval-shaped, which was resemble to the WT testis. In details, we observed the development arrested ovarioles were shorter and vacuole filled in both mutants (Fig 2B). As the result, the mature female adults produced few eggs and decreased in fecundity significantly (Fig 2B and 2C). In addition, no clear individual egg chamber was observed in Δsiwi and Δago3 ovarioles since the germline cells divided excessively but differentiated defectively (Fig 2A). However, the testes developed normally in both Δsiwi and Δago3 males, revealing the female-specific function of BmPIWIs (S4A Fig). In conclusion, depletion of silkworm PIWIs perturbed germline cell development and arrested oogenesis specifically in females.
Female Δsiwi moths developed a male-specific eighth abdominal segment and asymmetrical clasper-like structures on the genital papilla, leading to failure in mating with normal male animals (Fig 3 and S4B Fig). However, neither Δsiwi males nor Δago3 females and males showed developmental defect in abdominal segmentation or the structure of the externalia (Fig 3 and S4C Fig). These partial sexual reversal phenotypes indicated that BmSiwi regulates silkworm female sexual dimorphism but BmAgo3 does not.
Since the alternative splicing of Bmdsx and expression amount of BmMasc were the two reporters for masculinization, hence we detected the bands of Bmdsx and expression of BmMasc in the mutants [12,25,32]. Male-specific splicing production of Bmdsx (BmdsxM) and an increase in BmMasc transcript abundance (2.01-fold higher than WT) were detected in Δsiwi but not Δago3 female animals (Fig 4A and 4B), indicating that BmSiwi controlled silkworm female sexual dimorphism by regulating Bmdsx and BmMasc. In addition, no significant change on Bmdsx splicing form or BmMasc expression was detected in the males of either mutant (Fig 4A and 4B).
RNA-seq analysis was performed using the mixed ovary samples from three individual mutants at the larval wandering stage. In Δsiwi females, we identified 1460 differentially-expressed genes (DEGs) in which 1325 genes were down-regulated and 135 genes were up-regulated when compared to WT. In addition, the DEGs were enriched in 268 KEGG terms and 45 GO terms (S5A and S5B Fig). Only 198 DEGs (114 up-regulated and 84 down-regulated) were identified in the Δago3 females, and these were enriched in 127 KEGG and 36 GO terms (S5A and S5B Fig). Interestingly, the Δago3 enriched terms completely included in those of Δsiwi (S5A and S5B Fig). Two GO items, “reproduction” and “reproduction process”, were identified from both mutants, confirming that BmPIWIs involve in the oogenesis (S5C Fig).
We also detected significant decrease of piRNA abundance in ovaries of PIWIs female mutants. Comparing to WT females, piRNA abundance decreased to 89.6%, 74.5% and 36.5% in Δsiwi females and 95.5%, 85.2% and 66.7% in Δago3 females for 28-nt, 29-nt and 30-nt piRNAs respectively (S5D Fig). The relative abundance of six known piRNAs, Fem (BmSiwi-specific binding piRNA), Masc (BmAgo3-specific binding piRNA), Judo1, Judo2, Inoki and Suzuka (the latter four of which have no previously-identified binding specificity), were further examined in the two mutants using qRT-PCR. Consistent with previous reports [12], the Fem and Masc piRNAs were down-regulated in Δsiwi and Δago3 respectively (Fig 5A). Three piRNAs, Judo1, Judo2 and Inoki, were down-regulated in Δsiwi, but not Δago3, supporting the hypothesis that they may be able to bind BmSiwi (Fig 5A). However, the Suzuka was down-regulated in both mutants, likely due to a lack of binding specificity between BmSiwi and BmAgo3 (Fig 5A). In addition, qRT-PCR analysis revealed that seven TEs were up-regulated in the Δsiwi female silkworms but down-regulated in the Δago3 female animals (Fig 5B). The up-regulation of TEs in Δsiwi females was expected due to the decrease of its repressor, while this was the first report indicating that disruption of BmAgo3 induced TEs down-regulation. We proposed that this was caused by compensation between the primary and secondary piRNA biosynthesis pathways, although more evidences were needed [10,33]. In conclusion, dysregulation of piRNAs and TEs in Δsiwi and Δago3 female animals indicated a conserved function of PIWIs in insects.
Epigenetic modifications were shown to affect gonadogenesis in M. musculus and D. melanogaster, raising the possibility that B. mori HMTs may participate in piRNA-mediated sex determination [16,34]. Based on the RNA-seq data, qRT-PCR analysis revealed that the transcripts of three HMTs, BmAsh2, BmSETD2 and BmEggless decreased in abundance to 67%, 35% and 32% respectively in Δsiwi females comparing with WT ones, while no significant difference was found in Δago3 females (S5E Fig). These three genes showed tissue-specific expression in the gonads and predominantly in the ovaries (S6A–S6C Fig).
We established somatic mutant lines for each HMT using the transgenic CRISPR/Cas9 system to further investigate their physiological roles (S3E–S3G Fig). Δeggless animals showed no deleterious phenotype in physiology or sexual development (Figs 2B, 2C and 3). In contrast, Δash2 and Δsetd2 animals showed abnormal wing development from pupal stage, resulting in small and curly wings in adults (S7A and S7B Fig). This deleterious phenotype was similar to knock-out phenotypes in D. melanogaster, in which Δash2 flies developed absent, small and homeotic wings and Δsetd2 flies showed blistered wings, indicating a conserved function of Ash2 and SETD2 in insect wing morphogenesis [35–37].
Δash2 and Δsetd2 females showed defective oogenesis phenotype similar to Δsiwi and Δago3 female moths. Histological analysis revealed that Δash2 and Δsetd2 ovaries contained shorter and vacuolated ovarioles (Fig 2B and 2C). However, no defects were observed in the Δash2 and Δsetd2 male animals (S4A and S4C Fig). Interestingly, only Δash2 females showed partial sexual reversal characteristics, such as the appearance of eight abdominal segments and asymmetrically differentiated genital papilla (Fig 3). Furthermore, the BmdsxM splicing form and increased BmMasc expression were detected in Δash2 females (Fig 4A and 4B). These results demonstrated that BmAsh2, but not BmSETD2, was involved in silkworm sex determination.
We further investigated the relationship between HMTs and BmPIWIs because of their similar effects on silkworm female sex determination. We found that piRNAs expressions of Fem, Judo1, Inoki and Suzuka were down-regulated in Δash2 ovaries, consistent with the results found in Δsiwi female animals (Fig 5A and 5C). However, in Δsetd2 ovaries, Fem, Judo2 and Inoki levels were comparable to those observed in WT, while Suzuka was down-regulated, and this trend was consistent with the results from Δago3 females (Fig 5A and 5C). The expression of seven TEs was up-regulated in Δash2 females, while all of them, except TE1, were down-regulated in Δsetd2 animals, supporting the hypothesis that the regulation of BmAsh2 and BmSETD2 was piRNA-dependent (Fig 5D).
Since BmAsh2 phenocopied BmSiwi both at the female sexual reversal phenotype and piRNA regulation, we further investigated its localization in silkworm ovary by using immunostaining. BmAsh2 distributed in both the germline and somatic cells in the ovary and accumulated in the spermatogonium of the testis, similar to the localization of BmSiwi (Fig 1A and S2 Fig). Only weak signal of BmAsh2 could be detected in the Δash2 females, demonstrating that Cas9/sgRNA-mediated mutagenesis was highly efficient (Fig 6A). Since Ash2 is responsible for H3K4me3 modification [38,39], we next examined histone methylation using an anti-H3K4me3 antibody in Δash2 ovaries and found that the signal decreased significantly, suggesting that H3K4me3-mediated histone methylation was disrupted in Δash2 animals (Fig 6A). In addition, significant decrease of BmAsh2 protein accumulation was detected in Δsiwi ovaries, being consistent to qRT-PCR results (Fig 6B and 6C and S5E Fig). However, both relative mRNA and protein expressions of BmSiwi were comparable between Δash2 and WT ovaries, indicating that BmAsh2 did not function upstream of BmSiwi in silkworm sex determination pathway (Fig 6B and 6C and S5F Fig).
To elucidate the molecular basis of BmAsh2 involvement in sex determination, we expressed epitope-tagged BmAsh2 and BmSiwi and performed immunoprecipitation in BmN cells, which were derived from silkworm ovaries and exhibit both the primary and secondary piRNA biosynthesis processes. Successful ectopic expression for both proteins were detected in the input samples using anti-His or anti-Flag primary antibodies (Fig 6D). Furthermore, the BmSiwi protein was detected in the Flag immunoprecipitation products, revealing a protein interaction between BmSiwi and BmAsh2. In conclusion, the molecular evidence revealed that BmAsh2 plays critical roles in BmSiwi- and piRNA-mediated sex determination in B. mori (Fig 7).
PIWIs belong to the clade of gonadal Argonaute family proteins and silence TEs to maintain genomic integrity [15,40,41]. PIWI involvement in gonadal development has been demonstrated by studies showing that depletion of it caused sterility in Mus musculus, D. melanogaster and Danio rerio [13,15,42,43]. Absence of the piRNA-bound protein, Miwi, Mili and Miwi2, arrested spermatogenesis at different meiosis stages in mice [13,44,45]. Drosophila Piwi depletion caused the accumulation of germline stem cell-like tumors, leading to female infertility [43,46]. Gonadogenesis defect was attributed to DNA damage caused by random TE insertion, which disrupted the integrity of the germline stem cell (GSC) genome and homeostasis between GSC self-renewal and differentiation [47,48]. We showed here that a deficiency in BmSiwi and BmAgo3 in the silkworm results in degenerated ovarioles with fused egg chambers and germline cell hyperplasia, revealing the conserved function of PIWIs in gonadogenesis. Since no phenotypic defect was observed in testis development, we concluded that the effect of BmPIWIs on gonadogenesis was female-specific, although high expression of BmSiwi and BmAgo3 was detected in testes.
In addition to its function on oogenesis, BmSiwi, but not BmAgo3, also was involved in female sex determination. Although BmSiwi was reported to function in Bmdsx splicing in silkworm embryos [12], there was no previous physiological evidence reported. Here, we found that depletion of BmSiwi caused oogenesis arrestment and partial female sexual reversal, including the appearance of additional abdominal segments, asymmetrically differentiated genital papilla and a male-like clasper structure. Furthermore, dysregulation of BmMasc expression and splicing of Bmdsx further confirmed the function of BmSiwi on silkworm sex determination from molecular level. In comparison, no similar phenotype was observed in Δago3 females, supporting the conclusion that BmAgo3 does not function in silkworm sex determination. We speculated that the oogenesis arrestment observed in Δago3 females may be caused by a deficiency in a dsx-independent pathway, such as the bone morphogenetic protein (BMP) or epidermal growth factor receptor (EGFR) signaling pathway [49,50]. Thus, the current report provides genetic evidence for the involvement of BmSiwi in silkworm sex determination.
Ash2 is part of the SET1/MLL histone methyltransferase complex and is responsible for histone 3 lysine 4 (H3K4) methylation [51–55]. Drosophila spermatogenesis is controlled by multiple mechanisms, including epigenetic modifications [56]. In mouse, TE expression was repressed by CpG DNA methylation in a Mili-piRNA-dependent manner during sperm development. The repressive histone methylation at H3K9, which was responsible for heterochromatin formation, was active on retrotransposons at the meiotic pachytene stage when DNA methylation was inactive [57]. Expression of a breast tumorigenesis key factor, piRNA-021285, altered the methylation status of a number of related genes [58]. Drosophila TEs were silenced by PIWI-piRNA complex-dependent heterochromatin formation along with the silencing signal that spread to its adjacent genes [59]. Furthermore, PIWI-piRNA could recruit an epigenetic factor complex including the heterochromatin protein HP1a and the Su(var)3-9 histone methyltransferase to the target DNA [60]. These data support the conclusion that methylation is critical for gonadogenesis.
We show that the H3K4 HMT BmAsh2 was functional in piRNA-mediated sex determination in B. mori. Firstly, loss of BmAsh2 resulted in phenocopies of the BmSiwi mutant in females, which we interpreted to indicate that they function similarly in regulating silkworm sex determination. Furthermore, we detected colocalization of BmSiwi and BmAsh2 in both the germline and somatic cells in silkworm ovary. These two proteins also showed the similar localization at perinucleus in the germline cells, further confirming their important functions in piRNA regulation. More directly, we proved the direct interaction between BmSiwi and BmAsh2 proteins by immunoprecipitation assay. In conclusion, these results support the hypothesis that BmAsh2 regulates silkworm female sex determination through a piRNA-dependent pathway. Our report provides the first genetic evidence that BmAsh2 plays critical roles in BmSiwi- and piRNA-mediated silkworm sex determination.
A multivoltine, nondiapausing silkworm strain, Nistari, was used in these experiments. Larvae were reared on fresh mulberry leaves under standard conditions at 25°C [61]. The silkworm ovary-derived cell line BmN used for transfection was cultured at 25°C in TC100 insect medium [31].
Total RNA was extracted from silkworm ovaries, testes, and other tissues using TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. The isolated RNA was purified with phenol:chloroform and subjected to first-strand cDNA synthesis using the ReverAid First Strand cDNA Synthesis Kit (Vazyme). Relative mRNA amounts were measured using SYBR Green Real-time PCR Master Mix (Toyobo) according to a previously described method [31]. The qRT-PCR primers used here were as following: BmSiwiRTF: 5’-ATCACCCCAGAAAGACAACG-3’, BmSiwiRTR: 5’-GCACAGTATCAGGGCAGGAT-3’, BmAgo3RTF: 5’-GAGCAGTGCACAAAGCGATA-3’, and BmAgo3RTR: 5’-GGCACACCTGTTTCACCTTT-3’. As an internal control for qRT-PCR, we used a primer set that amplified a 136-bp PCR product of B. mori ribosomal protein 49 (Bmrp49) [31]. Three independent biological replicates were used for qRT-PCR, and other primers are listed in S1 Table. PiRNA sequences were found by referring to Kawaoka et al. [11], and the relative expression was measured using the stem-loop method [62].
A binary transgenic CRISPR/Cas9 system was used to construct silkworm mutants as described in Li et al. [31]. Six plasmids were constructed: the first, pBac[IE1-DsRed-IE1-Cas9] (IE1-Cas9), expresses the Cas9 protein constitutively driven by the baculovirus immediate-early gene IE1 promoter; and the other five, U6-BmSiwi sgRNA (pBac[A3-EGFP-U6-BmSiwi sgRNA]), U6-BmAgo3 sgRNA, U6-BmSETD2 sgRNA, U6-BmAsh2 sgRNA and U6-BmEggless sgRNA, express small guide RNAs (sgRNAs) targeted to BmSiwi (5’- CCTGAGTTGATATATCTAGTGCC-3’), BmAgo3 (5’-GGAGTGAGTATAGGCGGTAGAGG-3’), BmSETD2 (5’- CCATTAGCTAGTCCAGGTCTGCC-3’), BmAsh2 (5’-GGCAACGTGAAGGGCAGGCAAGG-3’) and BmEggless (5’- GGAGGCGGCGCAGCTCCGCGCGG-5’), respectively, under the control of the silkworm U6 small nuclear RNA promoter.
The plasmids were injected into preblastoderm embryos with a mixture of helper plasmids, piggyBac transposon mRNA and transgenic vectors. G0 animals were incubated at 25°C for 10–12 d until hatching, fed with fresh mulberry leaves, sib-mated or back-crossed with WT moths, and screened at late G1 embryos under a fluorescence microscope (Nikon, AZ100). Crossing the IE1-Cas9 and U6-sgRNA transgenic silkworms generates the gene-specific mutants used for the following experiments.
Total RNA from the ovary of wandering stage (when the ovary undergoes maturation) animals was extracted from three individual animals of Δsiwi, Δago3 and WT and mixed together. For mRNA sequencing, mRNA was enriched with Sera-mag Magnetic Oligo(dT) Beads (Illumina), fragmented to 200 nt in average, and used for cDNA synthesis. After that, the cDNA was sent to purification, end repair, nucleotide A and adapters addition (Illumina). Subsequently, the modified RNA was amplified with PE 1.0 and PE 2.0 PCR primers for 15 rounds and sequenced on an Illumina HiSeq 2000 platform (Shanghai OE BIOTECH CO., LTD). Sequenced raw data was qualified, filtered, and mapped to the reference silkworm genome database (http://silkworm.genomics.org.cn/) using tophat/bowtie2. Unigene abundance was measured by fragment per kilobase of exon per million fragments mapped (FPKM) and used for subsequent annotation.
RNA samples extracted from the ovary were also used for piRNA sequencing. Ten micrograms RNA was separated using 15% denaturing polyacrylamide gels and the small RNAs in length from 18 to 30 nt were used to construct library. Subsequently, small RNAs were sent to adaptors ligation at both the ends, cDNA synthesis and amplification were performed by using small RNA Cloning Kit (Takara). After sequencing with illumine HiSeq 2500 platform, the generated reads were filtered and small RNA reads from 24 to 30 nt in length were selected for mapping to the silkworm genome (http://silkworm.genomics.org.cn/silkdb/#), 121 annotated transposons and 1668 ReAS clones to identify the piRNAs as reported previously [63].
Silkworm ovaries and testes dissected from WT, Δsiwi, Δago3, Δash2, Δsetd2 and Δeggless animals at larval wandering stage were prefixed with Qurnah’s fixative [31]. Cross sections of 5 μm were cut with a Leica RM2235 microtome and used for staining. Sections were hydrated and stained with hematoxylin solution for 2 min, washed with running tap water for 5 min, stained with eosin solution for 2 min and dehydrated with 95% and 100% ethanol for 2 min each [64]. The stained tissues were analyzed and photographed under a microscope (Olympus BX51, Japan).
Paraffin-embedded sections were rehydrated and subjected to antigen retrieval with 0.1% trisodium citrate containing 0.1% Triton X-100 for 10 min at room temperature. The samples were washed with phosphate buffered saline (PBS) once and blocked with 1% bovine serum albumin (BSA) for 1 hour at room temperature. The silkworm gonads were incubated with Rabbit anti-BmVasa (1:200, Youke Biotech, indicating the germline lineage cells) [65], anti-BmSiwi (1:200, Youke Biotech), anti-BmAgo3 (1:200, Youke Biotech), anti-BmAsh2 (1:200, Youke Biotech) and anti-H3K4me3 (1:200, ABclonal) primary antibodies for 48 hours at 4°C. Samples were washed with PBS twice and treated with a FITC-conjugated Goat-anti-Rabbit secondary antibody (diluted 1:100 with 1% BSA, YEASEN) for 2 hours. Nuclei were stained with Hoechst (Beyotime) for 10 min at room temperature. After staining, samples were washed three times with PBS and analyzed with a fluorescence microscope (Olympus, BX53).
Flag-tagged BmAsh2 and His-tagged BmSiwi coding sequences were cloned into the pIZT/V5-His A insect expression plasmid under the control of an optimized baculovirus immediate-early gene promoter IE2 (OpIE2). The plasmids were transfected into the silkworm ovary-derived cell line BmN using Effectene transfection reagent (Qiagen) according to the manufacturer’s instructions. Three days after transfection, crude proteins were extracted and used for immunoprecipitation with a mouse monoclonal anti-Flag M2 antibody (1:1000, Sigma) according to Song et al. [66]. BmSiwi was detected using a Mouse anti-His (1:1000, Youke Biotech) primary antibody.
All data were analyzed using GraphPad Prism (version 5.01) with two-way ANOVA and the Dunnett’s tests. All error bars were the means ± S.E.M. p<0.05 was used to determine significance in all cases.
|
10.1371/journal.ppat.1006740 | Novel mechanisms to inhibit HIV reservoir seeding using Jak inhibitors | Despite advances in the treatment of HIV infection with ART, elucidating strategies to overcome HIV persistence, including blockade of viral reservoir establishment, maintenance, and expansion, remains a challenge. T cell homeostasis is a major driver of HIV persistence. Cytokines involved in regulating homeostasis of memory T cells, the major hub of the HIV reservoir, trigger the Jak-STAT pathway. We evaluated the ability of tofacitinib and ruxolitinib, two FDA-approved Jak inhibitors, to block seeding and maintenance of the HIV reservoir in vitro. We provide direct demonstration for involvement of the Jak-STAT pathway in HIV persistence in vivo, ex vivo, and in vitro; pSTAT5 strongly correlates with increased levels of integrated viral DNA in vivo, and in vitro Jak inhibitors reduce the frequency of CD4+ T cells harboring integrated HIV DNA. We show that Jak inhibitors block viral production from infected cells, inhibit γ-C receptor cytokine (IL-15)-induced viral reactivation from latent stores thereby preventing transmission of infectious particles to bystander activated T cells. These results show that dysregulation of the Jak-STAT pathway is associated with viral persistence in vivo, and that Jak inhibitors target key events downstream of γ-C cytokine (IL-2, IL-7 and IL-15) ligation to their receptors, impacting the magnitude of the HIV reservoir in all memory CD4 T cell subsets in vitro and ex vivo. Jak inhibitors represent a therapeutic modality to prevent key events of T cell activation that regulate HIV persistence and together, specific, potent blockade of these events may be integrated to future curative strategies.
| HIV persists in infected hosts in a small number of CD4 + memory T cells as latently infected proviruses. Homeostatic cytokines, which play a major role in the maintenance of T cell memory, also enable the persistence of latently infected cells. The pathways downstream of these homeostatic cytokines are well known and drugs that target these pathways have been developed and have been safely used in inflammatory diseases and in myelofibrosis. We have used these drugs to inhibit the maintenance and the spread of HIV infected cells carrying latent forms of the virus. We show that ruxolitinib and tofacitinib will inhibit the expansion of these cells and their capacity to infect other cells upon reactivation. This class of drugs is currently being tested in clinical trials.
| Current antiretroviral therapy (ART) has yielded significant success in achieving long-term suppression of viral load and in improving survival of HIV infected subjects [1–4]. Even so, ART fails to eliminate a small number of cells harboring integrated, replication competent viral DNA. This HIV reservoir has represented a major limitation in eradicating HIV. The HIV reservoir has been shown to persist in central memory TCM, transitional memory TTM and effector memory TEM CD4 T cells which require exposure to γ-C receptor cytokines for their long-term persistence [5, 6]. IL-2, IL-7 and IL-15 are γ-C receptor, homeostatic cytokines involved in the maintenance of T cell memory and which activate STAT5 mediated signaling [7]. In addition, the Jak-STAT pathway is also triggered by type I and type II Interferons, two important mediators of inflammation in viral infections including HIV [8–11]. Initial attempts to purge HIV involved the use of IL-2; results of these studies while promising, since virus was undetectable, did not reach their objective as viral load rebounded upon cessation of therapy [12]. We have previously shown that IL-7 driven homeostatic proliferation contributes to HIV persistence by promoting the survival and proliferation of latently infected cells [13, 14]. Further highlighting the role of IL-7 in the expansion and maintenance of the viral reservoir, an ACTG sponsored trial (ACTG protocol number 5214; www.clinicaltrials.gov as # NCT00099671) demonstrated that IL-7, also leads to a 70% increase in the absolute numbers of CD4 T cells harboring integrated viral DNA [14], suggesting that this intervention would not be compatible with an HIV eradication strategy. IL-15, which also signals through STAT5, has also been demonstrated to induce homeostatic proliferation of CD4 T cell subsets [15–18]. Furthermore, recent in vitro and ex vivo studies with IL-15, the IL-15 superagonist (ALT-803) and IL-2 illustrated that not only did these γ-C cytokines increase viral reactivation, but they also primed the reservoir within CD4 T cells for recognition by autologous HIV-specific CD8 T cells [19].
Phosphorylation of STAT-5 (pSTAT5) is triggered following the engagement of IL-2, IL-7 or IL-15 cytokines with their receptors leading to pro-survival signaling and increased proliferation [15, 20, 21]. Given the presence of multiple binding sites for pSTAT5 within the HIV long terminal repeat (LTR) [22], IL-2, IL-7 and IL-15 enhanced viral expression from productively infected cells [14, 15, 20, 21, 23]. Interestingly, a dominant negative STAT5 inhibited Jak-induced HIV LTR activity and decreased productive HIV infection while overexpression of STAT5 enhanced virus production in unstimulated primary T cells [22]. Together, these events underscore the relationship between activation of the STAT5 pathway and production of HIV, including events that impact the establishment of latency, its maintenance, and /or expansion of the HIV viral reservoir [24–27].
Tofacitinib and ruxolitinib are two FDA-approved Jak inhibitors for long-term use for the treatment of rheumatoid arthritis, myelofibrosis, or polycythemia vera [28–30]. Ruxolitinib demonstrates potent inhibition of pro-inflammatory cytokines in vivo, including IL-6, IL-1α/β, and TNF-α [29–32], all of which have been shown to enhance HIV replication in vitro [33, 34]. Inhibition of Jak-STAT signaling by ruxolitinib was shown to significantly impede T cell homeostasis, reducing CD4 T cell numbers as well as decreasing numbers of T regulatory cells and activated (HLA-DR+) CD4 T cell populations after a few weeks of treatment [28]. Tofacitinib showed only small changes in CD3+, CD4+ and CD8+ counts and an increase in B cell counts after 24 weeks of treatment [29, 30, 35]. Attenuated activation and proliferation were not specific to CD4 T cells but were also reported for Natural Killer (NK) cells treated with ruxolitinib in vitro and also in ruxolitinib or tofacitinib treated patients where the number of mature NK cells was reduced [28, 35]. Ruxolitinib treatment was further shown to block monocyte–derived DC differentiation, DC-derived IL-12 production and activation marker expression triggered by exposure to lipopolysaccharide (LPS) [28], demonstrating the impact of Jak inhibitors on innate and adaptive immune responses.
We previously reported that ruxolitinib and tofacitinib blocked reactivation of HIV in a primary T cell latency model at physiologic concentrations, underscoring its potential to block HIV reservoir expansion and viral dissemination from latent stores [34]. Herein, we monitored the impact of these clinically approved and extensively evaluated Jak-STAT inhibitors on several stages of HIV persistence including seeding of bystander cells and HIV reactivation from latency. Overall, we show ex vivo and in vitro that these Jak inhibitors use several mechanisms to impede the seeding and maintenance of the HIV reservoir.
We investigated the association between the Jak-STAT pathway and HIV persistence in a cohort (n = 37) of aviremic (<50 RNA copies/ml) ART-treated participants (S1 Table) to assess the in vivo relevance of Jak-STAT signaling in the establishment and maintenance of the HIV reservoir as monitored by frequencies of cells with integrated HIV DNA. Decreased CD4 numbers and immune activation are features of aberrant T cell homeostasis [13] and increased HIV reservoir size [36, 37], which was confirmed in the cohort studied here (S1 Table). We fit a linear model to identify univariate markers (Panel 1—activation markers, Panel 2—PD-1/IL-7R and Panel 3—STAT phosphorylation; S2 Table) associated with integrated HIV DNA (see methods). As previously described, the percent CD4+ T cells was significantly higher in immune responders (IR; median 34.7%) compared with non immune responders (NIR; median 22.4%) or recently treated (RT; median 24.7%) subjects (S3 Table) and increased integrated HIV DNA was associated with decreased frequencies of CD4+ T-cells (p = 0.019 [S2 Table]) [13, 14]. Indeed, we showed that increased levels of pSTAT5 expression in CD4 central memory (CM), transitional memory (TM) and in effector memory (EM) cells, all known to harbor HIV DNA, were associated with increased integrated HIV DNA (p = 0.043, 0.008, 0.001 respectively [S2 Table]). After adjusting for CD4 counts, CD4/CD8 ratio, nadir CD4 and pre-ART HIV-1 plasma RNA, CD4 CM, TM and EM pSTAT5 expression remained significantly associated with integrated HIV DNA (p = 0.031, 0.048 and 0.0048, respectively). As our cohort included immune responders (> 500 CD4 cells /mm3) and immune non responders (< 350 CD4 cells /mm3), successfully treated non-classified (SC NC) and recently treated subjects (< 1 year of treatment) (S1 Table), we determined the levels of pSTAT5 expression in each class. pSTAT5 in naïve CD4 cells was significantly lower in immune responders (median 1160 MFI) compared to immune non responders (median MFI 1500) or recently treated subjects (median 1523 MFI; p = 0.037 and 0.018 Wilcoxon rank test, respectively) (S17 Fig and S3 Table). There was no difference in STAT5 phosphorylation when comparing the different classes of subjects in CM, TM or EM CD4 cells, the subsets that encompass the bulk of HIV integrated DNA. The relationship between pSTAT5 and integrated viral DNA remains across immune responders and non-immune responders.
IL-7R engagement by its ligand leads to its internalization and to phosphorylation/activation of STAT5. Higher levels of surface IL-7R could reflect the lack of chronic receptor engagement, which would result in lower numbers of homeostatically proliferating cells and lower levels of the HIV reservoir [13, 14]. In line with this observation, we observed that increased surface expression of IL-7R on total CD4+ T-cells was associated with decreased integrated HIV DNA (p = 0.046, S2 Table), which was also significant in CD4 memory T cells (CD45RA-) (p = 0.032) and in CD4 CM cells (p = 0.032). For total CD4 T cells and CD4 memory T cells, IRs had significantly increased IL-7R levels (median 6916 MFI and 7163 MFI, respectively) compared with NIR (median 5314 MFI and 5288 MFI, respectively), RT (median 5536 MFI and 6091, respectively) or successfully treated non classified (median 6193 for CD4 memory T cells) classes of subjects; for the CD4 CM subset IL-7R levels in IRs (median 8647 MFI) was only significantly greater than ST NCs (median 7018 MFI, S3 Table). As cells from IRs are known to express lower levels of T cell activation and proliferation markers compared with NIRs [36, 37], increased expression of IL7R in IRs further supports our observation of decreased homeostatic proliferation and lower numbers of HIV infected cells in these subjects. We also monitored PD-1 expression, as it is up-regulated by γ-C receptor cytokines [38] and is a marker of cells undergoing homeostatic proliferation and the size of the HIV reservoir [37]. PD1 expression (MFI) on total CD4 (p = 0.027, S2 Table), naïve CD4 (p = 0.023), CD45RA- memory CD4 (p = 0.023), CD4 CM (p = 0.015) and CD4 TM (p = 0.037) cells was positively associated with increased integrated HIV DNA confirming previous reports [13, 14]. There was no difference in PD-1 levels (MFI) between IR and NIRs within this group of subjects however there was greater PD-1 levels on ST NC (median 9538 MFI) compared with IRs (median 7218 MFI, p = 0.048, S3 Table).
We also showed that an increase in frequencies of naïve CD4 T cells expressing Ki67 (p = 0.012, S2 Table), and cells expressing HLA-DR/CD38 (in total CD4 (p = 0.024), naïve CD4 (p = 0.008), CD45RA- memory CD4 (p = 0.03) and CD4 CM cells (p = 0.029)) were associated with increased frequencies of cells harboring integrated HIV DNA. The cells expressing HLA-DR/CD38 in naïve CD4 were correlated as well with increased frequencies of pSTAT5 positive CM, TM and EM subsets (p = 0.005, 0.027, 0.009 respectively). As observed in previous studies [36, 37], IRs had significantly lower levels of HLA-DR/CD38 in total CD4 T cells (median values for: IR—1.21%, NIR—2.36% and RT—2.22%), CD4 memory cells (CD45RA-) [median values for: IR– 1.83%, NIR—3.12%, RT– 3.55%] or CD4 CM cells (median values for: IR– 0.78%, RT– 1.38%) compared with NIRs and/or RT subjects; IRs also had lower levels of Ki67 in naive CD4 T cells (median 0.7%) than NIR (median 0.18%) (S3 Table). Heightened frequencies of cells with integrated HIV DNA were also associated with a decrease in frequencies of CD25+ cells in total CD4 (p = 0.017, S2 Table), memory (CD45RA-) CD4 (p = 0.002), as well as in CM (p = 0.0002) and TM (p = 0.009). In contrast to HLA-DR/CD38 and Ki67 expression, IRs had significantly greater levels of CD25 on CD4 T cells (median values for: IR– 40%, RT– 32%, ST NC– 29%), CD4 memory cells (CD45RA-) [median values for: IR– 56%, NIR– 42%, RT– 44%, ST NC– 42%], and CD4 CM cells (median values for: IR– 57%, NIR– 50%, ST NC– 46%) compared with NIR, RT and/or ST NC (S3 Table). These results demonstrated that up-regulated levels of pSTAT5 were associated with deregulated homeostatic proliferation and the size of the HIV reservoir and immune activation.
Within each of the panels (panel 1 –activation markers, panel 2 –PD-1/IL-7R or panel 3 –STAT phosphorylation), we next identified the combination of the univariate markers that could best predict the size of the HIV reservoir by building a multivariate regression model using feature selection [39] (see methods; Table 1). The activation markers panel showed that frequencies of total CD4+ T-cells, of naïve CD4+ cells expressing HLA-DR/CD38 (markers of immune activation) and Ki67, and of CD4+ CM cells expressing CD25 best predicted the size of the HIV reservoir (F-statistic p-value = 0.004 –Table 1). Within the PD-1/IL-7R panel, expression (MFI) of PD-1 by CM and TM cells, and of IL7R by CM cells best predicted the size of the HIV reservoir (F-statistic p-value = 0.04), while the STAT phosphorylation panel, showed only STAT5 phosphorylation on CD4+ EM cells as a predictor of the size of the HIV reservoir (F-statistic p-value = 0.001).
These multivariate models further confirm results of the univariate analyses showing the significant association between immune activation, T cell homeostasis and the HIV reservoir. These results provide further rationale for the use of interventions that target the Jak-STAT pathway i.e inhibitors of the Jak-STAT pathway that are known to inhibit immune activation and γ-C cytokine induced proliferation which could impact the size of the HIV reservoir.
We monitored, in vitro, STAT5 phosphorylation following γ-C receptor cytokine stimulation in total CD4 T cells and in the different memory subsets known to harbor the HIV reservoir in the presence or absence of the Jak inhibitors ruxolitinib and tofacitinib (Fig 1, S1 Fig and S4 Table). As expected, ruxolitinib significantly (p < 0.001) reduced IL-2, IL-7, and IL-15 induced pSTAT5 expression (Fig 1A) in a dose dependent fashion. Activity of ruxolitinib on STAT5 phosphorylation triggered by all three cytokines was completely abrogated at concentrations ≥ 0.1 μM (Fig 1A) although this inhibitor had a significantly stronger impact on IL-2 induced CD4+ STAT5 phosphorylation (10- fold reduction) when compared to IL-7 and IL-15 at 0.01 μM (1.9- and 2.7- fold reduction, respectively) (Fig 1A). These results were confirmed with tofacitinib where we also observed a similar dose dependent inhibition of the frequencies of pSTAT5+ cells (Fig 1C; p < 0.001). Importantly, ruxolitinib and tofacitinib completely abrogated frequencies of pSTAT5+ cells exposed to IL-2, IL-7 and IL-15, compared with DMSO for concentrations > 0.1 μM (p < 0.01) in all CD4 memory subsets (TCM,TTM and TEM) all known to harbor HIV integrated DNA (S1 Fig and S4 Table) [13]. Both Jak-STAT inhibitors showed a stronger impact on IL-2 induced STAT5 phosphorylation in all memory subsets as we had observed in total CD4+ T cells (S1 Fig). A linear regression model to study the association between the frequencies of pSTAT5+ cells within each CD4+ memory subset with a range of Jak-STAT inhibitor concentrations showed a negative correlation between these two variables, i.e. percent pSTAT5+ cells significantly decreased with increasing concentrations of ruxolitinib, specifically in CM and TM cells when exposed to IL-2 (p-values < 0.05) and with increasing concentrations of tofacitinib, in CM and TM cells when exposed to IL-15 and IL-2, while in the EM cells when exposed to IL-7 (at p-values < 0.05) (S4 Table and S1 Fig). These inhibitors also significantly (p < 0.01) decreased IFN-α stimulated STAT5 phosphorylation in CD4+ T-cells, IFN-α induced STAT1 phosphorylation in CD14+ monocytes and CD4+ T-cells and IL-10 induced STAT3 phosphorylation in CD14+ monocytes and CD4+ T-cells (S2 Fig) in a dose dependent fashion highlighting a possible role for these inhibitors in lowering the levels of hyper immune activation.
We also monitored for the upregulation of Bcl-2 expression triggered by these homeostatic cytokines; Bcl-2 is a transcriptional target of STAT5, which enhances the survival of cells exposed to IL-2, IL-7 and IL-15 [20, 40]. Both ruxolitinib (Fig 1B) and tofacitinib (Fig 1D) significantly (p < 0.01) reduced IL-2, IL-7, or IL-15-induced Bcl-2 levels (MFI) of expression in a dose dependent fashion to levels similar to the unstimulated control. Significant reduction of Bcl-2 expression (p < 0.01) was achieved at concentrations of the Jak-STAT inhibitors as low as 0.01 μM for IL-2 and IL-15 induced Bcl-2 expression (1.3- and 1.5- fold reduction, respectively). IL-7 induced Bcl-2 regulation required higher concentrations of these inhibitors as inhibition of IL-7 induced Bcl-2 levels (MFI) was only observed at concentrations ≥ 0.1 μM of ruxolitinib and tofacitinib (Fig 1B and 1D). Using a linear regression model that included Bcl-2 levels (MFI) within each CD4+ memory subset as the dependent variable and Jak-STAT inhibitor concentration as the independent variable, we showed that Bcl-2 expression (MFI) also significantly (p < 0.05) decreased with increasing concentrations of ruxolitinib or tofacitinib in all memory subsets (CM, TM and EM) upon exposure to IL-2, IL-15 and IL-7 (S4 Table and S1 Fig).
Altogether these results confirmed the inhibitory impact of ruxolitinib or tofacitinib on the activation of the Jak-STAT pathway in all CD4+ memory T cell subsets known to harbor the HIV reservoir.
Our group previously demonstrated that ruxolitinib and tofacitinib confer submicromolar anti-HIV-1/2 activity in human PBMCs and in macrophages without demonstrable toxicity [34], however further mechanistic information about how these drugs modulate key events involved in the expansion and maintenance of the HIV viral reservoir have not been explored. Herein we show that both inhibitors significantly (p < 0.0001) reduced p24 production by CD4+ T cells isolated from viremic donors and stimulated with CD3/CD28 when compared to DMSO controls (Fig 2A and 2B and S3 and S4 Figs for data on individual donors). This inhibition was observed when tested in the absence of ART (to observe the effect of the Jak inhibitors on viral replication and de novo infection of cells) and in the presence of ART (to observe the effect of the Jak inhibitors when spreading of HIV infection is inhibited, hence on viral production). These results showed that Jak inhibitors prevented uninfected primary CD4 T cells from HIV infection (described in detail under bystander infection assay) and inhibited the production of HIV from infected cells (ruxolitinib antiviral potency with ART; EC50 0.17 μM; EC90 6.2 μM; ruxolitinib antiviral potency without ART; 0.007 μM; EC90 0.26 μM).
Since ruxolitinib and tofacitinib decreased γ-C-cytokine induced Bcl-2 expression (Fig 1), we monitored cell viability of CD4+ T cells isolated from viremic donors and cultured for 6 days with CD3/CD28 and increasing concentrations of Jak inhibitors. Decreased viability was observed in 3 of the 5 donors at 0.1 μM or greater ruxolitinib or tofacitinib (S15 Fig, Panel A); using a linear regression model, addition of either inhibitor to anti-CD3/CD28 stimulated cell cultures led to significantly lower frequency of viable cells (%AnnexinV-LIVE/DEAD-) compared to DMSO taking into account inhibitor concentration and adjusting for donor. In contrast, we did not observe a loss in viability of in vitro infected cells after 3 day culture with Jak inhibitors plus ART (S15 Fig, Panel B); and no loss in viability was observed after 6 day culture of uninfected cells treated with IL-2 or IL-7 (n = 3) and in 2 of 3 uninfected donors treated with IL-15 in the presence of Jak inhibitors (S15 Fig, Panel C). As HIV-1 infected subjects are known to express lower levels of Bcl-2 [41], decreased viability as a result of inhibition of Bcl-2 expression may be a potential mechanism of Jak inhibitors on blocking viral persistence along with HIV infection in CD4 cells from viremic donors.
The inhibition of viral replication and production mediated by ruxolitinib and tofacitinib ex vivo was confirmed in an in vitro HIV infection model using a CXCR4-tropic, GFP tagged virus (eGFP NL4-3 replication competent HIV-1 reporter virus) that was used to identify productively infected cells by flow cytometry. At day 6 post infection, ruxolitinib and tofacitinib (0.1, 1.0, and 10 μM) significantly (p < 0.05) reduced extracellular p24 production (Fig 2C) as well as the frequency of HIV-GFP+ CD4+ cells (Fig 2D). Both inhibitors significantly (p < 0.01) decreased CCR5 surface expression on CD4 T cells that was upregulated as a result of anti-CD3/CD28 stimulation in viremic subjects (n = 5) whereas the levels of CXCR4 were not impacted by the addition of these inhibitors (S5 and S6 Figs). Since CXCR4 expression remained unchanged, decreased viral production and viral spread may be mediated through a mechanism beyond entry such as reduced T-cell activation (S16 Fig). As CCR5 dependent viral strains are mostly prevalent in vivo [42–44], these results indicate that inhibition of anti-CD3/28 induced CCR5 expression by ruxolitinib could have an impact on viral spread and dissemination in an infected untreated host [42, 43] and specifically in tissues from ART treated subjects where HIV replication still prevails in spite of the presence of ART [33, 45].
Since γ-C cytokines promote the activation of the Jak-STAT pathway (Fig 1) [8, 9, 11], paracrine or autocrine inhibition of IL-2, IL-7, and IL-15 induced signaling by ruxolitinib could be in part responsible for the inhibitory effects exerted by this compound on p24 production. Addition of an exogenous source of IL-7 (30 ng/ml) in ex vivo CD4 T cell cultures from viremic donors reversed the block by ruxolitinib at an antiviral EC50 concentration of 33 nM conferred on extracellular HIV production (p < 0.01), measured by p24 after 6 days in culture (S7 Fig; n = 4). More specifically, exogenous IL-7, which signals largely through STAT5, reverses latency, underscoring the link between IL-7, the ability to control latency, and STAT5-mediated signaling.
We monitored the frequency of cells harboring integrated viral DNA in cultures of T cells obtained from viremic donors that were activated by T cell receptor (TCR) engagement. These experiments aimed at measuring the impact of ruxolitinib on the maintenance of the existing reservoir as cultures were generated in the presence of ART. Cultures were also performed in the absence of ART to measure the seeding of the viral reservoir, as under these conditions infected cells that can produce virions will infect new T cells. Ruxolitinib and tofacitinib significantly (p < 0.05) decreased the frequency of cells with integrated DNA in cultures of T cells activated by TCR in the presence or absence of ART (Fig 3A and 3B) with doses of ruxolitinib as low as 0.01 μM.
IL-15, which signals through STAT5, also regulates memory T cell homeostasis [17, 18, 21], and could be involved in modulating the HIV reservoir size by promoting the persistence of cells with integrated DNA or by enhancing HIV reactivation and dissemination. Since ruxolitinib and tofacitinib inhibited STAT5-mediated signaling triggered by IL-7 and IL-15 (Fig 1) we sought to define the impact of ruxolitinib on IL-15 induced reactivation of latent HIV in CD4 T cells from aviremic subjects (Fig 3C–3E). Memory CD4+ T cells were isolated from ART treated aviremic donors (n = 3), activated with a concentration of IL-15 that is known to activate Jak-STAT signaling (10 ng/ml) (Fig 3D) or CD3/CD28 (Fig 3E) and maintained with or without 1 μM ruxolitinib in the presence of ART. We showed in preliminary experiments that 10 ng/ml of IL-15 triggered optimal STAT5 phosphorylation and viral reactivation. Extracellular viral RNA copies were quantified by qRT-PCR six days post IL-15 activation of purified memory CD4 T cells. Indeed, IL-15 increased reactivation of latent HIV (200 to > 1,000-fold versus non stimulated controls), although to a lesser extent when compared to the CD3/28 control (> 30,000 fold versus non stimulated controls; Fig 3C–3E). However, ruxolitinib was found to significantly reduce (p < 0.01) IL-15 and CD3/28 induced reactivation of HIV from latently infected cells, resulting in values similar to unstimulated controls. Additionally, our results demonstrated that ruxolitinib and tofacitinib decreased anti-CD3/28 induced T cell proliferation (dilution of Cell Trace Violet; S8 and S9 Figs) and activation (CD25, CD38/HLA-DR and PD-1 expression; S8 Fig and S10–S12 Figs) and more importantly, led to a decrease in the frequency of p24+ cells (Fig 2) as well as a reduction in cells harboring integrated HIV provirus (Fig 3) and decreased IL-15 induced HIV reactivation (Fig 3). Each of these events mechanistically signals through STAT5, further highlighting the role of STAT5, and subsequent block by Jak inhibitors, in controlling these key events that drive viral persistence. Altogether these results indicate that Jak-STAT inhibitors can negatively impact de novo seeding and the maintenance of the HIV reservoir.
The impact of ruxolitinib on the magnitude of the HIV reservoir could result from the inhibition of TCR triggered CD4 T cell activation and proliferation (S8–S12 Figs). It could also result from the direct antiviral activity of the Jak-STAT inhibitors (Fig 2A and 2B). Finally, decreased frequencies of cells harboring HIV integrated DNA could result from the diminished infection of bystander activated T cells due to the effects of these compounds on T cell activation and viral replication. We sought to determine if the presence of ruxolitinib could prevent the transfer of infectious viral particles to uninfected bystander cells or upon the formation of a virological synapse [44]. We developed an in vitro model to assess the impact of ruxolitinib on infection of bystander cells (schematic: Fig 4A; representative dot plots: Fig 4B) where bystander cells were labeled with cell trace violet (CTV+) and stimulated with CD3/28 in the presence of increasing concentrations of ruxolitinib to inhibit their activation. Unlabeled cells, (CTV-), were stimulated with CD3/28 in the absence of ruxolitinib, prior to infection with eGFP NL4-3. Infected CTV-negative cells and bystander CTV-positive cells were then co-cultured for two days to determine the number of ruxolitinib treated bystander cells susceptible to infection by eGFP virus (CTV+GFP+). As expected Ruxolitinib blocked proliferation (CTV-lo) of bystander cells (range 23% - 94%; p < 0.01 compared to no drug controls; paired t test) at all concentrations tested (Fig 4D). Cultures that did not include ruxolitinib led to the infection of fifty percent of bystander CTV+ cells (CTV+GFP+ cells) while addition of ruxolitinib led to a 75–80 percent reduction (p < 0.001) in CTV+GFP+ infected bystander cells. Our results show that ruxolitinib inhibition of Jak-STAT signaling impacts the susceptibility of uninfected cells to infection by HIV-1 at concentrations ≥ 0.01 μM (Fig 4B: representative dot plots, and Fig 4C: graphical representation) which are comparable to the ex vivo EC50 of 0.007 μM for ruxolitinib in CD4 T cells from viremic donors and well below the EC90 of 0.26 μM (Fig 2A). These findings suggest that Jak-STAT inhibitors could synergize with ART and decrease seeding of the HIV reservoir [34] in uninfected bystander cells.
An HIV cure will involve the elimination of residual persistently infected cells by immune effector mechanisms that include HIV specific T cells. Therefore therapeutic strategies such as Jak-STAT inhibition should not have an impact on T cell effector functions. We monitored the impact of ruxolitinib on the early and late signaling events downstream of TCR activation (n = 3). Fig 5A evaluated the effect of ruxolitinib on phosphorylation of SLP76 and CD3 zeta, two early events of the T cell receptor-signaling cascade [16, 46]. Fig 5A shows that ruxolitinib did not significantly alter CD3 zeta and SLP76 phosphorylation at all physiological concentrations (steady-state plasma concentrations found in vivo for all doses of ruxolitinib) tested (0.01, 0.1, and 1.0 μM) (S13 Fig- gating strategy and Fig 5A).
We next monitored the impact of ruxolitinib on late TCR signaling events such as the capacity of CD4+ T cell to produce effector cytokines (TNF-α, IL-2 and IFN-γ) when stimulated with CD3/CD28 (Fig 5B and S14 Fig) Ruxolitinib concentrations of 0.01, 0.1 and 1.0 μM did not alter cytokine production of IL-2, TNF-α, or IFN-γ single positive cells or IL2+TNFα+IFNγ+ polyfunctional CD4+ and CD8+ T-cells, as quantified by intracellular flow cytometry (Fig 5B and S14 Fig). Decreased cytokine production was observed only at 10 μM, which is above the Cmax and steady-state concentrations of ruxolitinib found in vivo [31]. Similarly, upstream events of TCR signaling and production of TNF-α, IL-2 and IFN-γ were inhibited only upon addition of a supra-physiological concentration (10 μM) of ruxolitinib, [32]. When we monitored cytokine production after gag peptide stimulation of PBMCs from stably treated, HIV-infected subjects, statistically significant differences in gag-specific antigenic responses were observed only at 10 μM ruxolitinib for TNF-α and at 0.1 μM for TNFα+IFNγ+ in CD4+ cells (Fig 5C and S18 Fig). There were no significant differences in gag-specific responses in CD8+ T-cells at the different concentration of ruxolitinib, however only 3 of the 6 subjects tested responded to gag-peptide stimulation and two subjects demonstrated decreased cytokine production at ruxolitinib concentrations ≥ 0.1 μM (Fig 5C and S18 Fig). These results demonstrate that ruxolitinib does not inhibit initial TCR function in both CD4 and CD8 T-cells nor HIV specific responses in CD4 T-cells with some donor-specific responses in CD8 T-cells, highlighting the ability of ruxolitinib to specifically inhibit signal transduction pathways that alter T cell proliferation and expansion of the HIV reservoir without interfering with the development of effector antiviral T cell functions.
Pharmacokinetic simulations were performed using reported plasma drug levels [31, 32] to determine if the plasma concentrations of drug during either 10 mg (Fig 6A), or 20 mg (Fig 6B) bid oral ruxolitinib treatment, could be correlated with in vitro efficacy. To assess PK/PD (pharmacokinetics/pharmacodynamics) relationship relative to the in vitro and in vivo efficacy reported herein, baseline population pharmacokinetics (PPK) model parameters without explicit patient covariates were used to construct a PPK model. Our model used δ2 as the ↔ subject variance (IIV) for that PK parameter, and ⌠2 = residual variance (within subjects). Log-normal error structure was used and 1,000 theoretical participants administered 10 mg ruxolitinib twice per day were modeled. Resulting computed percentile (P10, P25, P50 P75, P90) plasma concentrations versus time were used to mimic in vivo doses of 10 and 20 mg bid, which represent the low and high FDA-approved doses.
The model presented in Fig 6 shows that the pharmacodynamic effects (anti-HIV effects on viral reservoirs, antiviral, and HIV-induced activation/proliferation) of ruxolitinib falls within the Cmax and steady-state range for all FDA approved doses of the drug (dotted lines; 0.01–1.0 μM: Fig 6), highlighting that the concentrations required to inhibit these pro-HIV events are equivalent in vivo for individuals taking ruxolitinib, even for the 10 mg bid dosing, which represents the lowest effective dose in humans.
The IL-7 and IL-15 γ-C cytokines are essential for long term maintenance of memory T cells by regulating homeostatic proliferation and enhancing cell survival which paradoxically leads to HIV persistence (Fig 7A and 7B) [13–15, 17, 18, 20]. Herein, we demonstrate that engagement of this pathway as shown by increased STAT5 phosphorylation is positively associated to the frequencies of cells with integrated HIV DNA and to immune activation which has also been shown to be positively associated with the size of the HIV reservoir. Conversely, we show an inverse relationship between IL-7R levels and integrated HIV DNA in CD4 T cells from ART-treated HIV-infected donors (Table 1; Fig 7A and 7B), confirming the role attributed to STAT5 signaling, in increasing the size of the HIV reservoir by promoting the survival of infected cells expressing receptors for these homeostatic cytokines and by triggering the homeostatic maintenance of these cells (Fig 7A and 7B). Importantly, we show that ruxolitinib or tofacitinib inhibit signaling of the Jak-STAT pathway in all memory T cell subsets known to harbor the HIV reservoir. Our results show that Jak inhibitors impede T cell activation and proliferation both critical for HIV replication and seeding of the reservoir without impeding effector CD4 and CD8 HIV specific responses (Fig 7B–7D). Indeed we show that ex vivo and in vitro, exposure of T cells to Jak-STAT inhibitors prevents CD3/28-induced up-regulation of several cell surface markers (PD-1, HLADR/CD38 and CCR5) that are expressed by cells that harbor integrated DNA (Table 1, S8–S12 Figs; Fig 7C) [36, 37]. Although Jak inhibitors are not specific to the STAT5 pathway (S2 Fig), careful design of these experiments in the context of cytokines that signal almost exclusively through STAT5, allowed us to assess how blockade of STAT5 contributes to sentinel events driving reservoir establishment, maintenance, and expansion. We also demonstrated that ruxolitinib blocks IFN-α, induced pSTAT1 (CD4+ T cells and CD14+ monocytes) and IL-10 induced pSTAT3 (CD4+ T cells and CD14+ monocytes), underscoring that phosphorylation of other STATs may also confer the observed outcome, in combination with blockade of STAT5 (S2 Fig). The involvement of the STAT pathway was confirmed by the findings that exogenous IL-7 could overcome the Jak-inhibitor mediated block on latency reversal, and that Jak inhibitors block IL-15 induced viral reactivation. Importantly, these ex vivo and in vitro effects were obtained at concentrations (Jakafi.com; [31, 32]) that are achieved during steady-state of this drug in humans for all FDA approved doses. Of note, we previously reported that no apparent toxicity was observed for ruxolitinib or tofacitinib up to 50 μM in T cells and macrophages [34]. Furthermore, decreased viability was primarily observed after the cells from HIV infected subjects were treated with anti-CD3/28 which may be a consequence of HIV infected subjects having lower levels of Bcl-2 and consequently impaired cell survival (20, 44–45).
Our results show that ruxolitinib and tofacitinib can exert direct antiviral potency in infected cells isolated from viremic individuals and in primary CD4 T cells infected in vitro (Fig 2). These Jak inhibitors could confer a direct antiviral activity by blocking the phosphorylation of STATs and subsequent binding of this transcription factor to several STAT5 binding sites within the HIV-1 LTR (Fig 7E) [7, 22, 27]. Decreased virus production (summarized in Fig 7C–7F) could also be a consequence of the Jak-STAT induced down-regulation of T cell proliferaton. CD4 T cells exposed to these inhibitors should show decreased expression of transcription factors (NFAT-c, NF-κ-B) expressed by activated T cells [47] and which are required for HIV LTR mediated transcription [48]. Herein, we report that Jak inhibitors demonstrate nanomolar inhibition of viral replication in ex vivo CD4 T cells from viremic individuals, even in the presence of an EC99 of ART (Fig 2). Therefore, Jak inhibitors may block viral replication in ART treated subjects [33] which highlights their potential as a therapeutic modality, since they could block potential residual viral replication in pharmacological sanctuary sites in vivo where antiviral agents do not reach optimal concentrations [45]. Additionally, our results suggest that preferential cell death or shortened lifespan (down regulation of Bcl-2 across T cell subsets; S1 Fig) conferred by Jak inhibitor exposure lead to a reduction in the overall size of the viral reservoir. Dissemination of HIV to bystander cells is also inhibited by addition of Jak inhibitors further confirming the capacity of this class of molecules to prevent further seeding of the reservoir.
CD4 T regulatory cells (Tregs) are the major rheostat of immune homeostasis [49], in addition to suppressing cell mediated immunity to viral infections including HIV infection [50–52]. These cells are dependent on IL-2 and Jak-STAT signaling for their persistence [53]. STAT5-driven homeostatic proliferation is critical for the maintenance of absolute numbers of Tregs [53, 54] where increased number of Tregs could lead to further dysregulation of immune homeostasis (Fig 7, panel G). Jak inhibitors would potentially prevent the increase in Treg homeostatic proliferation in HIV infected subjects which was demonstrated in recent studies of subjects treated with these Jak Inhibitors for myeloproliferative disorders [55]. Inhibition of Treg function would trigger potent cell mediated immune responses that could control residual viral replication [5, 56, 57]. Indeed, ruxolitinib inhibits the production of IL-6, TNF-α IFN-α/β, D-dimer, IL-10, and IL-1α/β in vivo [32], markers associated with immune senescence and maintenance of the HIV reservoir. Together, these findings demonstrate that the anti-inflammatory properties of ruxolitinib that are associated with reduced activation and proliferation of T cells in an HIV-infected host could result in reduced bystander cell infection (Fig 7F), diminished levels of homeostatic proliferation (Fig 7B), and decreased permissiveness of uninfected cells by down-regulating the CCR5 co-receptor (Fig 7C).
Lastly, our results also show that this class of compounds can impact HIV reactivation without affecting TCR signaling and function. Indeed, we show that the most proximal events of TCR signaling as well as TCR induced effector function (i.e. cytokine production) are not affected by Jak inhibitors, although these molecules inhibit TCR induced proliferation. This is in line with previous observations showing that the threshold of MHC/peptide complexes required to trigger antigen specific effector functions is different than that required to induce T cell proliferation. Huang et. al. showed that a single peptide-major histocompatibility complex ligand interaction with a CD4+ T cell is enough to stimulate cytokine production in CD4+ T cells [58]. Weak signaling of the TCR still allows for phosphorylation of the TCR signaling pathway and production of cytokines, however only optimal TCR signaling may allow for T cell differentiation and proliferation [46, 58, 59]. This is consistent with our data, wherein steady-state plasma concentrations found in vivo (defined as physiological concentrations) did not impair TCR signaling or cytokine production while at the same time decreased CD4 T cell proliferation. Importantly, ruxolitinib did not block HIV gag-peptide T cell responses in CD4 T cells and only decreased CD8 response in 2 of 3 subjects at concentrations ≥ 0.1 μM, indicating that ruxolitinib will not negatively impact HIV specific responses in CD4 T-cells.
Jak inhibitors represent a potential therapeutic modality that addresses a clinical need which traditional direct-acting antiviral agents that interfere with steps in the viral replication cycle have not been successful. We demonstrate that Jak inhibitors confer STAT5-mediated block in homeostatic proliferation, and inflammation-driven HIV reservoir expansion. The effects of Jak inhibitors, described herein on mechanisms that promote HIV dissemination and persistence may be advantageous as an alternative therapeutic intervention that could be implemented in future curative strategies. The impact of ruxolitinib on inflammation associated with HIV infection, which plays an important role in seeding viral reservoirs, viral persistence and ongoing low-level replication in sanctuary sites is currently being evaluated in an NIH-ACTG sponsored 21-site Phase 2a study in humans that is currently enrolling.
Written informed consent was provided to study participants and approved by the Vaccine and Gene Therapy Institute of Florida ethics review board. Research conformed to ethical guidelines established by the ethics committee of the Vaccine and Gene Therapy Institute of Florida. IRB consent is encompassed under IRB # IRB00006031, obtained on April 5, 2010 for all subjects included in this study.
All multivariate (Table 1) and univariate analyses (S2 Table) were performed with ART treated aviremic donors (S1 Table). To confirm aviremic status, plasma viral load was measured with the Amplicor HIV Monitor Ultrasensitive Method (Roche, Basel, Switzerland), wherein donors were considered aviremic when viral load was below 50 copies/ml.
Human peripheral blood mononuclear cells (PBMCs) were isolated from leukapheresis by density gradient centrifugation as previously described [13]. Total and memory CD4+ T cells were isolated from PBMC of viremic, aviremic or HIV-negative individuals using magnetic bead–based negative selection (Stemcell Technologies, Vancouver, British Columbia, Canada). Cells were used for antiviral potency, signaling, or reservoir studies as described below.
A previously described, in-house p24 ELISA [60] was used to quantify viral production by CD4+ cells isolated from viremic donors ex vivo and in vitro infected CD4+ cells cultured in the presence of Jak inhibitors. Briefly, freshly enriched CD4+ T cells from viremic donors (1 to 1.5 x 106 cells) were stimulated for 3–6 days with 1 μg/mL of anti-CD3 (OKT3 hybridoma cell line; NIH AIDS and reference reagent program) and 1 μg/mL of anti-CD28 (BD Biosciences, Franklin Lakes, New Jersey) in the presence of increasing concentrations of Jak inhibitors either with or without ART (180 nM zidovudine, 100 nM efavirenz, 200 nM raltegravir (Cat # 11680) from Merck & Company, Inc.; NIH AIDS and reference reagent program, Division of AIDS, NIAID, NIH). Enriched CD4+ T cells isolated from HIV negative donors were pre-activated for 3 days with anti-CD3/CD28, infected with an eGFP NL4-3 replication competent HIV-1 reporter virus (pBR43IeG-nef+ (Cat #11349) from Dr. Frank Kirchhoff, NIH AIDS and reference reagent program, Division of AIDS, NIAID, NIH) [61–63] and cultured for 3–6 days in the presence of increasing concentrations of Jak inhibitors. Serial dilutions of cell-free supernatant from the ACH2 cell line with a known p24 concentration were used for the standard curve.
To determine if Jak inhibitors block IL-15-induced reactivation, 5 x 106 CD4+ T cells from aviremic donors were isolated and cultured with 1.0 μM ruxolitinib (Selleck Chemicals, Boston, MA, USA) or DMSO and either 10 ng/ml IL-15 or CD3/CD28 for 6 days in the presence of ART (180 nM zidovudine, 100 nM efavirenz, 200 nM raltegravir). Viral quantification (HIV particles containing HIV RNA) in cell-free supernatants was measured by qRT-PCR as previously described [14] and compared to an unstimulation control. Experiments were conducted in three independent donors.
To establish IL-7 rescue of virus reactivation from Jak inhibition, freshly isolated CD4+ T cells from viremic donors (1.5 x 106 cells) were pre-incubated with anti-CD3/CD28 and 0.033 μM ruxolitinib for 30 min prior to addition of 30 ng/mL IL-7. Quantification of virus in supernatants from these cultures was measured by p24 ELISA after a 6-day culture. Experiments were conducted in four independent donors.
To quantify the frequency of cells harboring integrated viral DNA, fresh CD4+ T cells isolated from viremic donors were cultured in the presence of Jak inhibitors and stimulated with anti-CD3/CD8 with or without ART (180 nM zidovudine, 100 nM efavirenz, 200 nM raltegravir) and lysed at the termination of the experiment on Day 6. Cell lysates were directly used in a nested Alu PCR. This method provides an accurate quantification of the absolute numbers of cells per million that carry integrated HIV DNA, as previously described [13, 14].
The expression of proliferation/activation markers and HIV co-receptors were monitored by flow cytometry using the following antibody panel: CD3-A700 (clone UCHT1, BD Biosciences, Franklin Lakes, New Jersey), CD4-Qdot605 (clone S3.5, Invitrogen, Carlsbad, California), CD8–PerCP Cy5.5 (clone RPA-T8, eBiosciences, San Diego, CA), HLA-DR–PerCP (clone L243, BD Biosciences), CD38-PE (clone HIT2, BD Biosciences), CD25-PE-Cy7 (clone M-A251, BD Biosciences) or PD-1-PE-Cy7 (clone EH12.2H7, BioLegend, San Diego, CA), AnnexinV-APC (BD Biosciences) and Cell Trace Violet (CTV; which was applied at the beginning of culture to allow to track proliferation due to dilution of the dye; Invitrogen). CCR5 and CXCR4 expression was evaluated using the panel described above replacing AnnexinV-APC and CD38-PE with CXCR4-APC (clone, 12G5, eBioscience) and CCR5-PE (clone 2D7, BD Biosciences). Bcl-2 was quantified using anti-Bcl-2-Pacific Blue (clone Bcl-2/100, BD Biosciences) in combination with CD3-A700, CD4-Qdot605, CD8-PerCP (clone SK1, BD Biosciences), CD45RA-BV650 (clone HI100, BioLegend), CCR7-FITC (clone 150503, R&D Systems) and CD27-APCeFluor780 (clone O323, eBiosciences). For all stains, dead cells were excluded with the LIVE/DEAD Aqua marker (Invitrogen). For all flow cytometry panels, cells were acquired on an LSRII flow cytometer using the FACSDiva software (Becton Dickinson, Franklin Lakes, New Jersey) and analyzed using FlowJo v9.9.6 and v10.2 (TreeStar Inc., Ashland, Oregon).
To measure the impact of Jak inhibitors on γ-C-cytokine-mediated STAT5 phosphorylation, PBMCs were first pre-incubated for 60 minutes with ruxolitinib (Selleck Chemicals, Boston, MA, USA) or tofacitinib (Selleck Chemicals, Boston, MA, USA) followed by a LIVE/DEAD stain. Stained cells were washed and incubated at 37°C with 50 ng/mL IL-2 (R&D Systems, Minneapolis, MN), 2 ng/mL IL-7 (R&D Systems), 5 ng/mL IL-15 (R&D Systems), 10 ng/mL IL-10 (R&D Systems) or 10,000 Units IFN-α A (R&D Systems) in the presence of compound. After 15 minutes, cells were fixed with BD Cytofix Fixation Buffer (BD Biosciences) and permeabilized with Perm Buffer III (BD Biosciences). Permeabilized cells were stained with CD3-A700, CD4-Qdot605, CD8-Pacific Blue, CD45RA-BV650, CCR7-FITC, CD27-APCeFluor780, STAT1 (pY701)-A647 (clone 4a, BD Biosciences), STAT3 (pY705)-PE (clone 4/P-STAT3, BD Biosciences) and STAT5 (pY694)-PE-CF594 (clone 47/STAT5 (pY694), BD Biosciences).
To measure the impact of Jak inhibitors on T cell receptor (TCR) signaling, bead enriched CD4+ T cells were pre-incubated with ruxolitinib for 30 minutes followed by a cell surface stain with LIVE/DEAD dye in the presence of compound to exclude dead cells. Stained cells were washed and cross-linked in the presence of compound for 10 minutes on ice with 0.5 μg/mL anti-CD3 (OKT3), 5 μg/mL anti-CD28 (BioLegend) and 10 μg/mL goat anti-mouse IgG (BioLegend). After 10 min, cells were fixed with BD Cytofix Fixation Buffer (BD Biosciences) and permeabilized with Perm Buffer III (BD Biosciences). Permeabilized cells were stained with CD3-A700, CD4-Qdot605, CD3 Zeta (pY142)-A647 (clone K25-407.69, BD Biosciences) and SLP76 (pY128)-PE (clone J141-668.36.58, BD Biosciences).
To measure the impact of Jak inhibitors on TCR-induced cytokine production in the absence of HIV infection, PBMCs were isolated from HIV negative donors, pre-treated for 1 hr with 0.01–10 μM ruxolitinib and stimulated for 6 hours with anti-CD3/CD28 in the presence of Brefeldin A (5μg/ml) and increasing concentrations of ruxolitinib versus DMSO. Following stimulation, cells were stained with CD3-PB, CD4-APC, CD8-PerCP and permeabilized with saponin. Permeabilized cells were then stained with fluorochrome-conjugated mAbs to the following cytokines: IL-2 FITC (clone MQ1-17H12, BD Biosciences), TNF-α A700 (clone MAb11, BD Biosciences), and IFN-γ PE-Cy7 (clone 4S.B3, BD Biosciences).
To measure the impact of ruxolitinib on CD4 and CD8 T cell responses to HIV gag peptide, total PBMCs from ART-suppressed HIV+ donors were thawed and rested overnight in media containing antiretrovirals to a final concentration of 200 nM raltegravir, 100 nM efavirenz and 180 nM AZT. The following day, 1.5 million cells per condition were pre-treated as follows: DMSO alone, or with Ruxolitinib at: 0.01 μM, 0.1 μM, 1.0 μM, or 10 μM for one hour at 37°C. Following the one hour preincubation, cells were either not stimulated (DMSO alone at a quantity equivalent to that used for gag peptide pool) or stimulated with: 1.0 μg/ml pooled Gag consensus A peptides (Cat #8116 HIV-1 Consensus A Gag Peptide Set, NIH AIDS Reagent Program) or platebound 1.0 μg/ml OKT3 anti-CD3 (hybridoma clone). All stimulations included brefeldin A (Sigma, Cat# B5936) at 10 μg/ml, soluble 1.0 μg/ml anti-CD28 (BD Biosciences), and were performed for 6 hours. Ruxolitinib was included at the previously noted concentrations for the full 6 hours. The intracellular cytokine staining assay was performed as follows. Stimulated samples were stained with live/dead fixable violet viability dye (Invitrogen) and the following surface antibody panel: anti-CD19 BV510 (clone HIB19, Biolegend), anti-CD4 Qdot605 (Invitrogen), anti-CD14 V500 (clone M5E2, BD Biosciences), anti-CD8 BV711 (clone RPA-T8, BD Biosciences) and anti-CD3 Alexa700 (BD Biosciences). After a single wash cells were fixed in 2% formaldehyde (paraformaledhyde) for 10 min at room temperature, then permeabilized with 0.05% Saponin. The following antibodies were added for intracellular staining: anti-TNF-α Alexa Fluor 488 (clone MAb11, BD Biosciences) and anti-IFN-γAPC (clone B27, Biolegend). Cells were washed, fixed in 2% PFA, and data was collected on a Fortessa flow cytometer (BD Biosciences) and analyzed using FlowJo software (TreeStar Ashland, OR, USA).
To quantify the impact of ruxolitinib on bystander infection of CD4+ T cells by HIV, CD4+ T cells isolated from HIV-negative individuals were stained with 0.1 μg/ml Cell Trace Violet (CTV) dye (Invitrogen) to track and quantify bystander cells that become infected (CTV+/GFP+) or left unstained. Cells with CTV dye were stimulated for 3 days with anti-CD3/CD28 in the presence of various concentrations of ruxolitinib or DMSO (negative control). Cells without CTV were stimulated with anti-CD3/28 for 3 days and spinoculated for two hours at RT (at 2000g) with NL4-3 GFP reporter virus. After two additional hours incubation at 37°C, infected CTV negative cells were washed in RPMI plus 2% Fetal Bovine Serum and co-cultured with uninfected CTV positive cells (ratio 1:1) for an additional two days. Bystander infected cells were identified by the expression of GFP in CTV positive cells.
To confirm that the observed in vitro and ex vivo anti-HIV effects mediated by ruxolitinib occur within the steady-state concentration range observed in humans, a pharmacokinetic simulation was performed. Ruxolitinib plasma concentrations were simulated for 10 and 20 mg bid dose regimens, using parameters of the basic 2-compartmental population pharmacokinetics model with first-order oral absorption (not including patient covariates), fitted to data from male patients (163 of 272 patients) from Phase 1 and 2 trials undergoing treatment for myelofibrosis. Males are the primary subjects of an ACTG trial (NCT02475655) of ruxolitinib in HIV infected subjects; hence males were utilized for this model [31]. Monte-Carlo simulations were performed to predict concentration versus time profiles of 1,000 theoretical subjects, and used to compute percentile ranges (P10, P25, P50, P75, and P90) of plasma concentrations versus time over the first 36 hours after start of ruxolitinib, which were then plotted. Since ruxolitinib is about 90% bound to serum in humans (ruxolitinib package insert), and cells were incubated in media with 10% fetal bovine serum (FBS), the in vitro EC50 were multiplied by 10 before plotting. The median and % CV of pharmacokinetic parameters used for the simulation were: first order absorption rate constant (Ka = 3.43 hr-1, 75%), oral systemic clearance (CL/F = 20.2 L/hr., 37.9%, inter-compartmental clearance (Q/F = 2.6), central and peripheral volumes of distribution (Vc/F = 57.7 L, 30.9%; Vp = 11.8 L, 85.7), absorption lag-time (p.052 hr.) and residual variance (⌠2 = 35.8%). Inter-individual variances were modeled as proportional (log-normally distributed), noting that the formula for % CV for a log-normal distributed parameter = ℘(e^⎤2–1) x 100, where ⎤2 = variance of a log-normally distributed parameter. Simulations were run using the NONMEM program (7.3 ICON Development Solutions, Ellicott City, MD), and statistics and graphical analyses were performed using R 3.0.1 (R Statistical Foundation, Vienna Austria, https://cran.r-project.org/web/packages/RODBC/index.html).
For identifying markers of the Jak-STAT pathway that are associated with the size of the HIV reservoir, we fit a linear model with frequencies of cells harboring integrated HIV DNA as a dependent variable and the different markers as independent variables (S2 Table). The obtained p-values were then corrected for multiple comparisons using the Benjamini and Hochberg (BH) method. Feature selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO) technique [39], in order to determine the combination of the significant univariate markers that best predict the size of the HIV reservoir (Table 1). The model was optimized using leave-one-out cross validation and the least cross-validated mean square error (MSE) was determined.
For STAT5 phosphorylation or Bcl2 expression across CD4 memory subsets, a linear regression model was applied with Bcl2 MFI (or %pStat5) as the dependent variable and the drug as the independent variable taking concentration into consideration (S4 Table). For HIV gag-peptide responses in CD4 or CD8 Tcells, a paired Wilcoxon rank sum test was used to determine significance (Fig 5 and S18 Fig).
For quantification of HIV production and reactivation; quantification of integrated HIV DNA; measurement of cell activation, proliferation, HIV co-receptors and Bcl-2 expression; measurement of Jak-STAT and T cell receptor signaling; and measurement of cytokine production following TCR signaling, significance was determined by two-way ANOVA followed by Sidak’s multiple comparison. Statistics were performed using GraphPad Prism 6.0 software. A p-value less than 0.05 was considered statistically significant.
|
10.1371/journal.pgen.1004285 | Genetic Predisposition to In Situ and Invasive Lobular Carcinoma of the Breast | Invasive lobular breast cancer (ILC) accounts for 10–15% of all invasive breast carcinomas. It is generally ER positive (ER+) and often associated with lobular carcinoma in situ (LCIS). Genome-wide association studies have identified more than 70 common polymorphisms that predispose to breast cancer, but these studies included predominantly ductal (IDC) carcinomas. To identify novel common polymorphisms that predispose to ILC and LCIS, we pooled data from 6,023 cases (5,622 ILC, 401 pure LCIS) and 34,271 controls from 36 studies genotyped using the iCOGS chip. Six novel SNPs most strongly associated with ILC/LCIS in the pooled analysis were genotyped in a further 516 lobular cases (482 ILC, 36 LCIS) and 1,467 controls. These analyses identified a lobular-specific SNP at 7q34 (rs11977670, OR (95%CI) for ILC = 1.13 (1.09–1.18), P = 6.0×10−10; P-het for ILC vs IDC ER+ tumors = 1.8×10−4). Of the 75 known breast cancer polymorphisms that were genotyped, 56 were associated with ILC and 15 with LCIS at P<0.05. Two SNPs showed significantly stronger associations for ILC than LCIS (rs2981579/10q26/FGFR2, P-het = 0.04 and rs889312/5q11/MAP3K1, P-het = 0.03); and two showed stronger associations for LCIS than ILC (rs6678914/1q32/LGR6, P-het = 0.001 and rs1752911/6q14, P-het = 0.04). In addition, seven of the 75 known loci showed significant differences between ER+ tumors with IDC and ILC histology, three of these showing stronger associations for ILC (rs11249433/1p11, rs2981579/10q26/FGFR2 and rs10995190/10q21/ZNF365) and four associated only with IDC (5p12/rs10941679; rs2588809/14q24/RAD51L1, rs6472903/8q21 and rs1550623/2q31/CDCA7). In conclusion, we have identified one novel lobular breast cancer specific predisposition polymorphism at 7q34, and shown for the first time that common breast cancer polymorphisms predispose to LCIS. We have shown that many of the ER+ breast cancer predisposition loci also predispose to ILC, although there is some heterogeneity between ER+ lobular and ER+ IDC tumors. These data provide evidence for overlapping, but distinct etiological pathways within ER+ breast cancer between morphological subtypes.
| Invasive lobular breast cancer (ILC) accounts for 10–15% of invasive breast cancer and is generally ER positive (ER+). To date, none of the genome-wide association studies that have identified loci that predispose to breast cancer in general or to ER+ or ER-negative breast cancer have focused on lobular breast cancer. In this lobular breast cancer study we identified a new variant that appears to be specific to this morphological subtype. We also ascertained which of the known variants predisposes specifically to lobular breast cancer and show for the first time that some of these loci are also associated with lobular carcinoma in situ, a non-obligate precursor of breast cancer and also a risk factor for contralateral breast cancer. Our study shows that the genetic pathways of invasive lobular cancer and ER+ ductal carcinoma mostly overlap, but there are important differences that are likely to provide insights into the biology of lobular breast tumors.
| Invasive lobular breast cancer (ILC) accounts for 10–15% of all invasive breast carcinomas and it has distinct etiological, clinical and biological characteristics compared with the more common invasive ductal/no special type carcinoma (IDC) [1]. Lobular cancers show stronger associations with the use of hormone replacement therapy (HRT) than IDC, [2] and its incidence follows a similar temporal pattern as the use of combined HRT [3]. ILC is characterized by E-cadherin loss and the malignant cells therefore infiltrate the breast stroma in single files with little associated stromal reaction. This makes it difficult to detect these tumors by palpation or mammography, and they are often larger at presentation than IDCs [4]. ILCs are generally of histological grade 2 and estrogen receptor positive (ER+), with the exception of the pleomorphic subgroup. They typically have a different pattern of metastatic spread to IDCs, tending to infiltrate the peritoneum, ovary and gastrointestinal system. There is some evidence that they are less chemo-sensitive than IDC and that the 10-year survival rate of women with ILC is lower than that of ER+ IDCs [5], [6].
ILC is often associated with lobular carcinoma in situ (LCIS), a form of non-invasive breast cancer that is difficult to detect clinically and typically found incidentally on biopsy. The increased breast biopsy rate associated with screening mammography has led to an increase in the diagnosis of LCIS. LCIS shares many of the same genetic aberrations as ILC, suggesting that it is a precursor lesion in an analogous manner to ductal carcinoma in situ (DCIS) and IDC [7]. Women who have had LCIS are 2.4 times more likely to develop invasive breast cancer compared to the general population, with an excess of ILC (23–80% of cases) [8], [9]. However only 50–70% of invasive cancers associated with LCIS have lobular morphology [10, unpublished data from GLACIER study]. The remaining cancers have a IDC or mixed ductal-lobular appearance, but again are generally ER+ (95% of IDC and mixed ductal-lobular cancers associated with LCIS in the GLACIER study were ER+). Unlike DCIS, LCIS is also a risk factor for developing invasive cancer in the contralateral breast [8].
Genome-wide association studies (GWAS) in breast cancer have identified loci that predispose to invasive breast cancer in general, or specifically to ER+ or ER-negative disease [11]–[25]. However, no previous study has focused specifically on lobular carcinomas. Only one common single nucleotide polymorphism (SNP; rs11249433 at 1p11.2) has been shown to be more strongly associated with lobular than ductal histology [26]. For the remaining SNPs predisposing to ER+ tumors, it is unclear whether the studies have lacked statistical power to identify differential associations by histology, or whether associations tend to be non-differential by morphology after accounting for ER status.
The aim of this study was to identify new breast cancer susceptibility loci specific to lobular carcinoma, and to evaluate the heterogeneity of associations of known loci by morphology. This involved pooling genotyping data from over 6,000 cases of lobular carcinoma (ILC and/or LCIS) and over 34,000 controls genotyped using the iCOGS chip, a custom SNP array that comprises 211,155 SNPs enriched at predisposition loci for breast and other cancers [24].
In a phase I analysis, we evaluated risk associations between SNPs on the iCOGS chip and risk of ILC and LCIS using 1,782 lobular cases (1,470 ILC with or without LCIS, 312 pure LCIS) from GLACIER, a UK study of lobular breast cancer, and 4,755 UK controls from the Breast Cancer Association Consortium, BCAC (Figure 1). There was little evidence for systematic inflation of the test statistics, based on 37,544 uncorrelated SNPs that had not been selected on the basis of breast cancer risk (λ = 1.04; Figure S1). Data were combined by meta-analysis with a further 4,241 cases (4,152 ILC, 89 LCIS) and 29,519 controls of European ancestry, derived from 34 studies in BCAC, and previously typed on the iCOGS chip (Tables S1 and S2). This resulted in a total of 6,023 cases (5,622 ILC, 401 LCIS) and 34,271 controls with data on 199,961 iCOGS SNPs (after quality control exclusions and with minor allele frequency (MAF) >0.01) included in the meta-analysis.
All SNPs reaching genome-wide significance (P<5×10−8) in the meta-analysis were correlated with one of the known breast cancer predisposition loci. In order to identify new loci that predispose to lobular carcinoma, we selected six uncorrelated SNPs (rs11977670, rs2121783, rs2747652, rs3909680, rs9948182, rs7034265) that were only weakly correlated (r2<0.25) with known loci and that showed the best evidence of association (P between 5×10−8 and 5×10−5) in the overall lobular case-control analysis (ILC and LCIS). These SNPs were genotyped in a Phase II including 516 cases (481 ILC, 35 LCIS) and 1,467 controls, all from white European donors (Figure 1).
One of the six SNPs, rs11977670 at 7q34, reached genome-wide significance in a pooled analysis of phase I and II ILC cases and controls (OR = 1.13, 95%CI = 1.09–1.18, P = 6.0×10−10, Table 1, Figure 2). rs11977670 showed a similar association with LCIS (P-het for ILC vs LCIS = 0.198), and a very weak or no association with IDC (OR = 1.02, 95%CI = 1.00–1.05, P = 0.070; P-het for ILC vs IDC = 1.3×10−5), indicating that this is a lobular specific predisposition locus (Table 2). The risk allele appeared to act in a dominant rather than additive manner: ORAG = 1.21, 95%CI = 1.14–1.30; ORAA = 1.27, 95%CI = 1.17–1.38; P for departure from log-additivity = 0.009; Table S3. rs11977670 was not significantly associated with age at onset of ILC (Ptrend = 0.16) and risk alleles were not significantly over-represented in cases with a positive family history (FH) (P = 0.90, FH+ vs FH−). None of the other 5 SNPs genotyped were associated with lobular breast cancer at a genome-wide significance level, with the strongest association being for rs2121783 at 3p13 (OR = 1.11, 95%CI = 1.07–1.15, P = 4.5×10−7; Table S4).
rs11977670 at 7q34 (position:139942304, GRCh Build 37) is intergenic, 65 kb from the nearest gene, JHDM1D, a histone demethylase and 500 kb from BRAF, a gene frequently mutated in melanoma. It is also in close proximity to a predicted novel U1 spliceosomal RNA that contains two U1 specific promoter motifs (Figure S2). ENCODE data on normal human mammary epithelial cells (HMEC), and breast carcinoma (MCF-7), were used to establish chromatin states in the region and showed that rs11977670 lies in region marked by H3K27 acetylation, Figure S3.
Using expression data from the Cancer Genome Atlas Network (TCGA database) [27], we assessed expression of the nine genes within 0.5 Mb of rs11977670 by breast cancer subtype (ER+ ILC, 40 cases; ER+ IDC, 341 cases; and ER-negative IDC, 108 cases; Figure S4). Three genes showed differential expression in ER+ ILC compared to ER+ IDC (BRAF, P = 0.006; NDUFB2, P = 0.02, SLC37A3, P = 0.05), however none reached statistical significance when correcting for multiple testing. Another two genes, JHDM1D and ADCK2, showed a difference in expression between ER-negative and ER+ cancers, but this was not lobular-specific. To further investigate which genes may be influenced by SNPs tagged by rs11977670, germline genotype data for rs13225058 (A/G), a surrogate for rs11977670 (G/A) (r2 = 0.79) was taken from the TCGA database (SNP6.0 Affymetrix array) and compared to expression of these genes, correcting for copy number variation, in 335 ER+ primary breast cancers where both genotype and expression data was available. A significant difference, after correcting for multiple testing, was found in expression between the AA and GG genotype for two genes JHDM1D (P = 0.0005) and SLC37A3 (P = 0.004), Figure S5a. Confining the analysis to the 36 ILC cases with data in TCGA showed no significant genotype specific expression due small numbers although there was the suggestion of a trend towards overexpression with the GG genotype (2 cases), Figure S5b. 48 of the cases also had expression data on adjacent normal breast tissue, but due to the small numbers no significant genotype specific expression changes were detected, Figure S6. There was no evidence of copy number variation around rs11977670 and no evidence of an excess of somatic mutations in JHDM1D, SLC37A3 or BRAF in ILC.
Most (56 of 75) known common breast cancer susceptibility loci were associated with ILC at P<0.05 with the effect in the same direction as previously reported (Table S5), and 13 of these reached genome-wide significance (P<5×10−8, Table 3). The strongest associations were with SNPs close to FGFR2 (rs2981579, OR = 1.38, P = 5.1×10−52), TOX3 (rs3803662, OR = 1.33, P = 1.1×10−35), at 1p11.2 (rs11249433, OR = 1.25, P = 2.7×10−25) and 11q13.3 (rs554219, OR = 1.33, P = 1.6×10−22). All 13 loci had previously been shown to be associated with ER+ breast cancer and one locus, rs11249433 (1p11.2), with lobular histology in subgroup analysis. Of the remaining 19 SNPs with P≥0.05, 18 had ORs in the same direction as previously reported for overall breast cancer (Sign test P = 0.0001), suggesting that these SNPs are also likely to predispose to LCIS. Only one of the seven ER-negative specific loci on the iCOGS array showed a significant association with ILC (rs12710696, P = 0.037). In case-only analyses, no SNP showed an association with family history of breast cancer or young age at onset of ILC.
For the 75 known breast cancer susceptibility loci, case-control analysis for the 401 cases of pure LCIS (without invasion) and 24,045 controls, revealed 15 out of 75 SNPs associated with LCIS at P<0.05 (Table 3). The strongest associations were for rs865686 (9q31.2, P = 2.2×10−5); rs3803662 (TOX3, P = 1.2×10−4), c11_pos69088342/rs75915166 (11q13.3, P = 7.8×10−4) and rs1243482 (MLLT10, 10p12.31, P = 7.8×10−4) that is partially correlated (r2 = 0.69) with rs7072776, a recently identified ER+ breast cancer predisposition locus that showed a weaker association with LCIS (OR = 1.17, 95%CI = 1.00–1.36, P = 0.05; Table S5). Forty-seven of the remaining 60 SNPs at P>0.05 had ORs in the same direction as for ILC. This is greater than one would expect by chance (Sign Test P = 1.2×10−5) suggesting many of these SNPs predispose to LCIS, but the study did not have enough power to detect these associations with the small sample size.
A global test in case-only analysis (ILC vs LCIS) indicated no significant differences in associations of the 75 SNPs between LCIS and ILC (likelihood ratio test (75 df) = 0.438). However, individual SNP analyses suggested some differences. Two loci showed stronger associations with ILC than pure LCIS: rs2981579, FGFR2 (P-het = 0.02); and rs889312, 5q11.2 (P-het = 0.03). Case-only analysis also suggested that two ER-negative specific SNPs [23], [25] were more strongly associated with LCIS than ILC: rs6678914, 1q32.1 (P-het = 0.0007) and rs17529111, 6q14.1 (P-het = 0.04) Table 3. The remaining SNPs showed no significant heterogeneity between ILC and LCIS.
In order to identify lobular specific SNPs, we performed a case-only analysis of 3,201 ER+ ILC cases and 15,024 ER+ IDC cases from BCAC. Analysis was confined to ER+ cases since 94% of ILC cases were ER+ (compared to 78% of IDC in BCAC). A global test indicated significant differences in SNP associations between ILC and IDC (likelihood ratio test (75 df) P = 5.9×10−6). The SNP showing the largest difference between ILC and IDC was rs11249433 at chr 1p11.2 (P-het = 2.7×10−8; Table 4), a SNP previously associated with lobular histology. At P<0.05, a further two loci were associated more strongly with ILC than IDC: rs2981579, FGFR2 (P-het = 5.3×10−3) and rs10995190, 10q21.2 (P-het = 0.002). This analysis also identified four IDC-specific SNPs at P<0.05: rs10941679, 5p12 (P-het = 1.5×10−4); rs2588809, RAD51L1 (P-het = 0.001); rs6472903, 8q21.11 (P-het = 0.004); rs1550623, CDCA7 (P-het = 0.031) Table S6.
Case-control analysis of 690 mixed ductal–lobular carcinomas revealed 25 loci that showed an association with these mixed cancers at P<0.05. The top hits were at FGFR2 (rs2981579, OR = 1.37, P = 1.6×10−7), rs941764 (CCDC88C, OR = 1.25, P = 3.6×10−4) and rs10995190 (ZNF365, OR = 0.74, P = 3.9×10−4). The case-only analysis above showed that two of these SNPs are more strongly associated with ILC than IDC (rs2981579, rs10995190). rs941764 showed no association with ILC and only weak association with ER+ IDC, Table S6.
Our analyses of a total of 6,539 lobular cancers (including 436 cases of pure LCIS) and 35,710 controls has identified for the first time a lobular-specific SNP, rs11977670 (JHDM1D; OR = 1.13 P = 4.2×10−10, that showed little evidence of association with IDC (P = 0.07) or DCIS (P = 0.23). Identification of the target of this association will require fine mapping of the region, followed by functional assays to determine which gene(s) the key SNPs regulate. The preliminary in silico functional analysis suggests that SNPs in this region may be influencing expression of JHDM1D (a histone demethylase) and SLC37A3 (a sugar-phosphate exchanger). For JHDM1D this appears to be a recessive effect, in contrast to the susceptibility data, which suggests a dominant effect. There are little data on the role of these genes in cancer. There is some evidence that increased expression of JHDM1D can suppress tumor growth by regulating angiogenesis [28] and decreased expression promotes invasiveness, which is contrary to what one would expect from the risk data [29]. This inconsistency does shed some doubt on these results and further analysis of the region is required before any firm conclusion can be made. Studies of syndecan-1-deficient breast cancer cells, which show increased cell motility and invasiveness, demonstrate decreased expression of both JHDM1D and E-cadherin [29], suggesting the two genes may interact. Somatic mutations in CDH1 (E-Cadherin) are frequent in ILC and rare germline frameshift mutations in CDH1 have been described in ILC, particularly in families with hereditary diffuse gastric cancer (HDGC), but also in cases of familial ILC with no HDGC [30], [31]. However, none of the 56 SNPs in CDH1 that were typed on the iCOGS chip showed any association with lobular cancer at P<0.05.
It should also be noted that this study is not a true genome wide association study for lobular breast cancer as the SNPs on the iCOGS chips were chosen on the basis of some prior evidence of association with breast cancer as a whole. Although ILC would have been a small proportion of the samples in the discovery sets for these SNPs it is possible that other lobular specific loci exist that have not been included on the iCOGS chip. This is particularly true for LCIS, which would only have been included in the discovery set as a parallel phenotype when associated with invasive disease.
75 of the known common breast cancer susceptibility loci were assessed for association with ILC and LCIS. As cases of ILC were included in the discovery sets that generated these susceptibility loci and lobular breast cancer is generally ER+ (94% of the ILC cases in this study were ER+) with the majority of ILCs classified as luminal tumors [32], it is not surprising that the majority of SNPs that we found to be associated with ILC were known to also predispose to ER+ breast cancer. However, some loci were only associated with ER+ IDC and not with ILC, particularly rs10941679 at 5p12, previously shown to predispose more strongly to ER-positive, lower-grade cancers [33], P-het = 2.7×10−8. Others showed a much stronger association with ILC than IDC, particularly rs11249433 at 1p11.2, as previously described [26]. These data suggest specific etiological pathways for the development of different histological subtypes of breast cancer, in addition to common pathways that predispose to multiple tumor subtypes.
Despite the small number of pure LCIS cases without invasive disease, our analyses have shown for the first time that many of the SNPs that predispose to ILC also predispose to LCIS. Although only 15 of the known breast cancer SNPs were associated with LCIS risk at P<0.05, 47 of the remaining 60 SNPs at P>0.05 had ORs in the same direction as for ILC (Sign Test P = 1.2×10−5) suggesting that many more SNPs are likely to be associated with pure LCIS but did not reach statistical significance individually because of the relatively few LCIS cases without associated ILC in our sample set. This is not unexpected if LCIS is an intermediate phenotype for ILC. However, a small number of SNPs had differential effects on LCIS or ILC risk. Specifically, rs6678914 at 1q32.1 (LGR6), known to be an ER-negative specific SNP [25], that appeared to be associated with LCIS but not ILC (P-het = 0.0007), and rs17529111 at 6q14 preferentially associated with ER-negative tumors [23] that had a stronger association with LCIS than ILC (P-het = 0.04). We also identified SNPs in FGFR2 and at 5q11.2 (MAP3K1) that appear only to predispose to ILC, but have little effect on LCIS suggesting that SNPs affect different parts of the lobular carcinoma pathway. These findings are surprising and as based on small numbers need confirmation in future studies.
Some of the SNPs associated with both ILC and LCIS showed a stronger effect size in LCIS compared to ILC (for example SNPs at TOX3, 9q31.2, 11q13.3, ZNF365 and MLLT10). It is possible that the SNPs that showed an association with both LCIS and ILC predispose to the development of LCIS rather than ILC, and that the effect size is smaller in ILC as not all cases of LCIS will become invasive cancer. SNPs that predispose strongly to LCIS were also associated with ER+ IDCs but again with stronger effect sizes in LCIS, consistent with the fact that 30–40% of invasive tumors associated with LCIS will not be ILC but will be IDC, mixed ductal-lobular or other morphology.
One SNP, rs1243182 (MLLT10), that showed a strong association with LCIS (LCIS: P = 7.8×10−4, OR = 1.29; ILC: P = 6.1×10−9,OR = 1.14; ILC+LCIS: P = 3×10−10,OR = 1.15, IDC: P = 1.4×10−5,OR = 1.07, is partially correlated (r2 = 0.69) with rs7072776, a recently identified ER+ breast cancer predisposition locus, which showed no association with LCIS in this study. It is also strongly correlated with rs1243180 (r2 = 0.80), an ovarian cancer predisposition variant [34] and rs11012732 (r2 = 0.57), which predisposes to meningioma [35]. The ovarian SNP, rs1243180, also showed a strong association with lobular cancer (ILC+LCIS: P = 5.54×10−10; OR = 1.13). Conditional analysis confirmed that this was not independent of rs1243182. rs11012732 was not genotyped on the iCOGS chip. The increased risk of ovarian carcinoma after breast cancer is well documented in epidemiological studies [36]. Of note, there are also reports suggesting an association between breast cancer and meningioma [37].
In conclusion, we have identified a novel lobular-specific predisposition SNP at 7q34 close to JHDM1D that does not appear to be associated with IDC. Most known breast cancer predisposition SNPs also predispose to ILC, with some differential effects between ILC and IDC. In addition, many SNPs predisposing to invasive cancer are also likely to increase the risk for LCIS. Overall, our analyses show that genetic predisposition to IDC and lobular lesions (both ILC and LCIS) overlap to a large extent, but there are important differences that are likely to provide insights into the biology of lobular breast tumors.
All studies were performed with ethical committee approval, Table S7, and subjects participated in the studies after providing informed consent.
In order to establish the SNP's functional role, a window of 10 kb both up and downstream was formed around the marker and pairwise r2 values calculated using 1000 genome CEU population data. Three SNPs were identified as being in LD (r2>0.5) with rs11977670 and were compared to next generation sequence technologies to elucidate the overlap between chromatin states (ENCODE Project). Two cell lines, normal human mammary epithelial (HMEC), and breast carcinoma (MCF-7), were used to establish these chromatin states, i.e. active or engaged enhancers (H3K27ac), nucleosome-depleted regions (DNase I and FAIRE), and RNA polymerase linked regions (Pol II). Expression data from the Cancer Genome Atlas Network for each gene within a 1 Mb window of rs11977670 was analyzed looking for differential expression in each breast cancer subtype (ER+ ILC, 40 cases; ER+ IDC, 341 cases; and ER-negative IDC, 108 cases). Allele data for surrogate SNP rs13225058 was obtained for all ER+ cases from TCGA. These 335 cases were used to produce genotype specific gene expression data in R. Differences in gene expression between the three genotypes were tested for using one-way-anova, verified by t-test and visually by boxplot. Linear regression was performed across all three genotypes using copy number variation as a co-variate. Level 3 copy number variation data (hg19 build) was obtained from the TCGA data portal.
|
10.1371/journal.pgen.1002712 | Neurospora COP9 Signalosome Integrity Plays Major Roles for Hyphal Growth, Conidial Development, and Circadian Function | The COP9 signalosome (CSN) is a highly conserved multifunctional complex that has two major biochemical roles: cleaving NEDD8 from cullin proteins and maintaining the stability of CRL components. We used mutation analysis to confirm that the JAMM domain of the CSN-5 subunit is responsible for NEDD8 cleavage from cullin proteins in Neurospora crassa. Point mutations of key residues in the metal-binding motif (EXnHXHX10D) of the CSN-5 JAMM domain disrupted CSN deneddylation activity without interfering with assembly of the CSN complex or interactions between CSN and cullin proteins. Surprisingly, CSN-5 with a mutated JAMM domain partially rescued the phenotypic defects observed in a csn-5 mutant. We found that, even without its deneddylation activity, the CSN can partially maintain the stability of the SCFFWD-1 complex and partially restore the degradation of the circadian clock protein FREQUENCY (FRQ) in vivo. Furthermore, we showed that CSN containing mutant CSN-5 efficiently prevents degradation of the substrate receptors of CRLs. Finally, we found that deletion of the CAND1 ortholog in N. crassa had little effect on the conidiation circadian rhythm. Our results suggest that CSN integrity plays major roles in hyphal growth, conidial development, and circadian function in N. crassa.
| Cullin-RING E3 ubiquitin ligases (CRLs) play important roles in regulating a wide range of processes, such as signal transduction, transcription, cell cycle progression, circadian rhythm, and development, via the ubiquitin-proteasome pathway. The activity and stability of CRLs is precisely controlled by the COP9 signalosome (CSN), an evolutionarily conserved multisubunit protein complex. Under the control of the CSN, CRL activity can be either downregulated via cleavage of NEDD8 (an ubiquitin-like protein) from cullin proteins (deneddylation) or preserved by maintaining the stability of CRL components. We generated point mutations of key residues in the JAMM domain of the CSN-5 subunit to disrupt CSN deneddylation activity, thereby creating a series of mutants containing the intact CSN complex but lacking deneddylation activity. Surprisingly, hyphal growth, conidial development, circadian rhythm, and stability of the SCFFWD-1 complex in these CSN-5 point mutants were comparable to that observed in wild-type N. crassa. Furthermore, we showed that CSN containing mutant CSN-5 efficiently prevents degradation of the substrate receptors of CRLs. Finally, deletion of the N. crassa ortholog of CAND1 (cullin-associated NEDD8-dissociated protein 1) had little effect on conidial development and the circadian clock. Our results suggest that the integrity of the CSN is important for growth and development in N. crassa.
| The COP9 signalosome (CSN) is an evolutionarily conserved multifunctional complex in eukaryotes; it is composed of eight subunits (CSN1–CSN8) in plants and mammals [1]. The CSN was initially discovered to be an important regulator of photomorphogenesis in Arabidopsis thaliana [2] and was later found to participate in a wide range of processes [1], [3]. The CSN potentially influences these cellular pathways by regulating the activity of cullin-RING ubiquitin ligases (CRLs, e.g., CRL1, CRL3, and CRL4 complexes in most eukaryotes) [3], [4], [5]. CRLs are a big family of ubiquitin ligases that share a common cullin/RING-E2 module [6], [7], [8]. They are necessary for substrate ubiquitination in a cascade of enzymatic reactions involving E1, E2, and E3 [9]. Under the control of the CSN-regulated ubiquitin-proteasome pathway, cells coordinate the expression of an array of genes involved in the regulation of growth and development in order to respond to environmental signals, such as light, temperature, and changes in nutrient conditions [1], [10]. Loss-of-function mutations in CSN subunits result in dysfunction of hundreds of CRLs [3], which explains the pleiotropic phenotypes observed in CSN mutants [1], [3], [7].
In 2002, Deshaies and his colleagues first described that the CSN-5 metalloprotease (JAMM) motif is required for removing the ubiquitin-like protein NEDD8 from Cul1 [11]. Later studies confirmed that the isopeptidase activity of the CSN complex is responsible for cullin deneddylation in eukaryotes [1], [3], [4], [5], [12]. In this process, the CSN binds to CRL E3 ligase and cleaves NEDD8 from cullins via the catalytic activity of its CSN-5 subunit, and then inhibits CRL activity [5], [13], [14], [15]. Thus, the deneddylation activity of CSN requires the metalloprotease motif located in the CSN-5 subunit and the functional core subunits of the CSN [11], [16]. However, CSN-5–dependent metalloprotease activity is not essential in Schizosaccharomyces pombe, as no obvious phenotype was detected in csn-5 deletion strains [17], [18].
The physiological importance of CSN deneddylation activity in development and cell differentiation was examined in Drosophila melanogaster, in which the lethality of csn-5Δ/Δ animals was rescued by expression of a CSN-5 transgene but no adult flies were recovered upon equivalent expression of CSN-5 (D148N) (loss of deneddylation activity) [11], [19]. In CSN-5-downregulated HeLa cells, however, the accelerated degradation of c-Jun was rescued equally by over-expression of either the JAMM domain mutant CSN-5D151N or wild-type CSN-5 [20]. These results suggest that the requirement for neddylation/deneddylation cycle of cullins is not absolutely necessary during normal growth and certain developmental stages. In plants, genetic studies suggest that although neddylation/deneddylation cycle is not absolutely necessary during early embryonic development and germination, it is required during seedling establishment and the later developmental stages [12], [21]. In Aspergillus nidulans, deletion of csnE/csn-5 or mutation in JAMM domain results in a block in fruiting body formation at the primordial stage, with a few other observed phenotypic changes, such as light-dependent signaling [22], [23]. Although deneddylation is a major activity of the CSN, it alone cannot explain all of the phenomena described above. These observations raise the possibility that the CSN may have other functional activities in addition to its deneddylation activity.
Recent genetic evidence suggest that the CSN has one additional major function: it controls the stability of CRL ubiquitin ligases in vivo by mediating assembly/disassembly of CRL complexes and by protecting substrate receptors in CRLs from degradation [3], [24], [25], [26], [27], [28]. A recent structural and biochemical study showed that the protective effect of the CSN on DDB2 and CSA autoubiquitination in CRL4 complexes does not require CSN-5–mediated deneddylation activity [29]. However, both of the CSN activities occur when CSN associates with cullins in CRL E3 complexes. Furthermore, there is also a tight correlation between CSN deneddylation activity and the ability of the complex to modulate the stability of CRLs [3]. Thus, it is difficult to determine which function is more important for growth and development through regulation of CRL activity, or how these two functions cooperate with each other in regulating CRL dynamicity in eukaryotes. In A. thaliana, the MPN (Mpr-Pad1-N-terminal domain) subunits CSN-5 and CSN-6 are essential for the structural integrity of the CSN holocomplex [12]. Several studies have shown that point mutations in the JAMM metal-binding site of CSN-5 do not interfere with the proper assembly of CSN complexes in S. pombe, A. thaliana, and A. nidulans [11], [21], [23]. In N. crassa, CSN-5 is not an essential gene; the deletion mutant can survive, and displays obvious growth and developmental defects, making it an excellent model system for investigating the distinctions between the deneddylation and CRL complex assembly/disassembly functions of the CSN [16].
The CSN takes part in a wide range of cellular and developmental processes in N. crassa, including hyphal growth, conidial formation, light and temperature responses, and circadian clock function [16], [24]. To further investigate its biological function in vivo, we created a series of point mutations in the JAMM metal-binding motif of the CSN-5 subunit to disrupt the deneddylation activity of the CSN complex. In those mutant strains, the integrity of the CSN and its interactions with Cul1 and Cul4 were not affected. Surprisingly, mutated CSN-5 almost retained the ability to restore the phenotypic defects of a csn-5KO strain and partially maintained the stability of the SCFFWD-1 complex, which was able to carry out degradation of the clock protein FREQUENCY (FRQ) in vivo. Moreover, the stability of four other substrate receptors of CRLs can be efficiently restored by the CSN containing mutant CSN-5. However, deletion of the CAND1ortholog in N. crassa had little effect on conidiation circadian rhythm and the degradation of FRQ. Our results suggest that the integrity of CSN plays major roles in hyphal growth, conidial development, and circadian function in N. crassa.
The N. crassa genome encodes seven COP9 signalosome subunits (CSN-1–CSN-7) [16], [24]. Several studies have shown that the JAMM metal-binding sites in the MPN domain of CSN-5 are required for metalloprotease activity in the CSN [11], [21], [23]. When the CSN-5 protein sequence was used in a BLAST search against protein databases, a highly conserved MPN domain in the N. crassa CSN-5 subunit was identified. As shown in Figure 1A, three conserved residues corresponding to His127, His129, and Asp140 lie within the putative metal-binding motif (EXnHXHX10D) of the N. crassa CSN-5 JAMM domain. To determine whether these conserved residues form the metalloprotease-like active site of JAMM, we used the JAMM domain of Archaeoglobus fulgidus as a template to generate the tertiary structure of N. crassa CSN-5 [30]. Because of the low similarity between these two JAMM domains, the generated structure was poor. Thus, we instructed SWISS-MODEL to automatically select a template protein for generating the structure of N. crassa CSN-5 [31]. SWISS-MODEL selected the pre-mRNA splicing factor Prp8 as template (Protein Data Bank [PDB] accession number 2P8R) for N. crassa CSN-5. The functional sites were mapped into predicted structure according to the structural alignment with AfJAMM (PDB accession number 1R5X). As shown in Figure 1B, His127, His129, and Asp140 within EXnHXHX10D of the N. crassa CSN-5 JAMM corresponded to the putative metal-binding motif as metalloprotease-like active site in AfJAMM [30], [32].
To confirm the contribution of CSN-5 to CSN-mediated deneddylation of cullins, we mutated these three highly conserved amino acids (H127A, H129A or D140N) using site-directed mutagenesis. We then introduced quinic acid (QA)–inducible Myc-tagged wild-type CSN-5 or one of the three mutant CSN-5 constructs into a csn-5KO strain expressing Myc-Cul1 protein. As shown in Figure 1C, Myc-CSN-5, Myc-CSN-5H127A, Myc-CSN-5H129A, and Myc-CSN-5D140N were expressed in the csn-5KO strains in the presence of QA. Expression of Myc-tagged wild-type CSN-5 in the csn-5KO strain resulted in a decrease in hyperneddylated Cul1 to the level of the wild-type strain (Figure 1C), indicating that the Myc-tagged CSN-5 protein was functional for CSN deneddylation activity. In contrast, expression of mutant CSN-5 (H127A, H129A, or D140N) failed to decrease the hyperneddylated Cul1 in the csn-5KO strain (Figure 1C). Similarly, hyperneddylation of Cul3 (Figure 1D) and Cul4 (Figure 1E) in the csn-5KO strain was rescued by expressing the Myc-tagged wild-type CSN-5, but not by any of the mutated Myc-CSN-5s. This indicates that the metal-binding motif of JAMM is essential for CSN-mediated deneddylation of cullins.
Because all of the Cul3 and Cul4 was neddylated while not all of the Cul1 was neddylated in the csn-5KO strain and csn-5KO strains complemented by JAMM-domain mutant CSN-5, we rechecked Cul1 modification in the csn mutants. As shown in Figure S1, c-Myc antibody detected three specific protein bands in first generation of csn-5KO or csn-6KO transformants and two specific bands in the csn-1KO transformants. In most positive transformants, there was slightly less unneddylated Cul1 than neddylated Cul1, but the signal remained strong. This is different from the studies in yeast, plants, and fruit fly in which deletion of csn-5 results in hyperneddylation of Cul1 [11], [21], [33]. Possible explanations are that N. crassa genome codes for another deneddylase in addition to CSN complex or there is large amount of newly synthesized Cul1 proteins. We next examined the neddylation of Cul4 using a polyclonal antibody that recognizes the N terminus of N. crassa Cul4. As shown in Figure 1F, only the neddylated Cul4 was detected in csn-5KO strain, while in the wild-type strain, most of the detected Cul4 was unneddylated. Next, we transferred endogenous csn-5 promoter-driven constructs of either wild-type CSN-5 or CSN-5 with JAMM triple point mutations (H127A, H129A, and D140N) (hereafter referred to as CSN-5tri) into a csn-5KO strain expressing Myc-Cul1 protein. Myc-CSN-5 and Myc-CSN-5tri were expressed in the csn-5KO strains (Figure 1G). Similar to what we observed in csn-5KO expressing CSN-5 with a single point mutation (Figure 1C), expression of the CSN-5tri failed to decrease the hyperneddylation of Cul1 in the csn-5KO strain (Figure 1G) as well. Interestingly, the amount of unneddylated Cul1 in csn-5KO strains expressing either single (Figure 1C) or triple (Figure 1G) point mutant CSN-5 was less than that in a csn-5KO strain. Furthermore, expression of CSN-5tri in the csn-5KO strain also failed to decrease hyperneddylated Cul4 to the levels observed in the wild-type or csn-5KO strain complemented with wild-type CSN-5 (Figure 1H). Taken together, these data confirm that the JAMM domain metal-binding motif of N. crassa CSN-5 is essential for the deneddylation activity of the CSN.
To examine whether the JAMM metal-binding site of CSN-5 functions in growth and development, we analyzed the phenotypes of the csn-5KO strain expressing either Myc-tagged wild-type or mutant CSN-5. On minimal slants with QA, the csn-5KO strain produced fewer aerial hyphae and conidia than the wild-type strain (Figure 2A). Expression of wild-type CSN-5 in the csn-5KO strain restored aerial hyphal growth and conidial formation to levels similar to those in the wild-type strain (Figure 2A). Surprisingly, when csn-5KO, Myc-CSN-5H127A; csn-5KO, Myc-CSN-5H129A; and csn-5KO, Myc-CSN-5D140N strains (hereafter referred to as csn-5H127A, csn-5H129A, and csn-5D140N, respectively) were grown in minimal slants containing QA, the transformants exhibited hyphal formation and conidiation that were the same as the wild-type strain and the csn-5KO, Myc-CSN-5 strain (Figure 2A). We next measured the growth rates of the wild-type strain, the csn-5KO strain, and the transformants by race tube assay in constant darkness. Interestingly, the growth of csn-5H127A, csn-5H129A, and csn-5D140N strains was slightly faster than that of the wild-type strain (approximately 4.2 cm per day vs. 3.7 cm per day, respectively) and the csn-5KO, Myc-CSN-5 strain (Figure 2B). These results suggest that these CSN-5s with a point mutation within the JAMM metal-binding motif function similarly as the wild-type CSN-5 on N. crassa growth and conidiation. In QA-containing race tubes, the conidiation rhythms of the csn-5H127A, csn-5H129A, and csn-5D140N strains (a period of about 22.5 h) were pretty much (only slightly longer) to those of the wild-type and csn-5KO, Myc-CSN-5 strains (a period about 22.2 h) (Figure 2C) in constant darkness after light entrainment. To characterize the effect on light response of each CSN-5 point mutation, we further examined the light-entrained conidiation rhythm of each csn-5KO transformant during light–dark (LD) cycles (12 h light/12 h dark). As shown in Figure 2D, although the LD cycles entrained the conidiation rhythm of the csn-5KO strains expressing wild-type CSN-5 or mutant CSN-5, however, the conidiation bands of the csn-5H127A, csn-5H129A, and csn-5D140N strains were broader than those of the wild-type and csn-5KO, Myc-CSN-5 strains. Similarly, 12 h 27°C/12 h 22°C temperature cycles entrained the conidiation rhythm of the csn-5H127A, csn-5H129A, and csn-5D140N strains, but not the patterns of conidiation bands (Figure 2E). Taken together, these results suggest that point mutations within CSN-5 are functional in growth and conidiation, and partially functional in circadian rhythm, light response, and temperature-entrained clock process.
The loss of deneddylation activity of the JAMM domain mutations may be due to the disruption of the CSN complex. To examine this, we tested the interactions between the CSN-6 subunit and wild-type or mutant CSN-5s. Myc-tagged CSN-6 was co-expressed with Flag-tagged CSN-5 or mutant CSN-5 proteins in csn-5KO strains. As shown in Figure 3A, the Flag-tagged CSN-5 strongly interacted with Myc-tagged CSN-6 in an immunoprecipitation reaction, suggesting that they were both in the intact CSN complexes. As expected, the Flag antibody pulled down the Myc-tagged CSN-6 protein in the csn-5KO strain co-expressing Myc-CSN-6 and each of the mutant Flag-CSN-5 proteins (Figure 3A), similar to what was observed in the csn-5KO strain co-expressing CSN-6 and wild-type CSN-5. This result indicates that the point mutations within the CSN-5 JAMM metal-binding motif did not affect the interactions between the CSN-5 and CSN-6 subunits and those two MPN proteins within PCI (Proteasome, COP9, eukaryotic Initiation factor 3) complexes may form dimers. To further examine whether Myc-His-tagged CSN-5 point mutants are incorporated into a larger molecular mass CSN complex, we performed gel filtration and followed by Western blot analysis. As shown in Figure 3B, like wild-type CSN-5, CSN-5H127A, CSN-5H129A, and CSN-5D140N fusion proteins were eluted in larger molecular mass fractions, suggesting that each of the Myc-tagged CSN-5 point mutants can be incorporated into the intact CSN complex.
Using protein affinity purification followed by Mass Spectrometry analysis, we further examined whether the CSN complex is properly assembled with CSN-5 point mutants. Myc-His-tagged CSN-5H127A, CSN-5H129A, CSN-5D140N, or wild-type CSN-5 was purified on a nickel column followed by immunoprecipitation with a c-Myc monoclonal antibody. As shown in Figure 3C, similar immunoprecipitated protein profiles were detected in the Myc-His-CSN-5H127A, Myc-His-CSN-5H129A, Myc-His-CSN-5D140N, and Myc-His-CSN-5 (wild-type CSN-5) samples, but not in the wild-type strain (negative control). Liquid chromatography–mass spectrometry/mass spectrometry (LC-MS/MS) analysis of excised gel bands led to the identification of seven subunits, from CSN-1 to CSN-7a, in the Myc-His-CSN-5 purified products and in the Myc-His-CSN-5H127A purified products (Figure 3C). Taken together, these results confirm that the integrity of the CSN complex is not affected by mutations within the JAMM motif of CSN-5 in N. crassa.
Next, we examined whether CSN complexes with mutant CSN-5 subunits can still interact with Cul1 protein. As shown in Figure 3D, both wild-type CSN-5 and each of the mutant CSN-5 proteins co-immunoprecipitated with Cul1 protein. We further examined whether CSN complexes with mutant CSN-5 subunits can also interact with Cul4 protein in vivo by IP/western blotting experiments. As shown in Figure 3E, the Myc-tagged wild-type CSN-5 co-immunoprecipitated with the neddylated and unneddylated Cul4, indicating that the N. crassa CSN complex can interact with all species of Cul4 in vivo. Similarly, the Myc-tagged mutant CSN-5s also co-precipitated with Cul4 (Figure 3E), further confirming that mutations within the JAMM metal-binding motif of CSN-5 do not interfere with interaction between CSN and cullins. These results strongly suggest that the point mutations within the JAMM metal-binding motif abolish NEDD8 isopeptidase activity but have no effect on the integrity of the CSN or on its interactions with cullins.
In N. crassa, the clock protein FREQUENCY (FRQ) is a negative regulator in the negative feedback loop that controls the circadian clock under constant conditions [34], [35]. Impaired FRQ degradation in csn-2 mutants results in the loss of circadian rhythm [24]. To investigate whether the mutant CSN-5s can rescue circadian rhythm defects in the csn-5KO strain, we examined the degradation of FRQ protein in the wild-type and csn-5KO strains expressing wild-type CSN-5 or mutant CSN-5s after addition of the protein synthesis inhibitor cycloheximide (CHX). In the wild-type strain, the FRQ was gradually degraded after CHX treatment, with a half-life of about 2.5 h (Figure 4A and 4B). However, in the csn-5KO strain, the degradation of FRQ was mostly blocked (Figure 4A and 4B), similar to what was observed in the csn-2KO strain and the fwd1RIP mutant [24], [36]. As shown in Figure 4A and 4B, the expression of Myc-tagged wild-type CSN-5 in the csn-5KO strain restored the degradation of FRQ to wild-type levels, so that the conidiation period on race tubes was similar to that of the wild-type strain (Figure 2C). We next checked FRQ degradation in the csn-5KO strain expressing CSN-5 proteins with mutations in the JAMM metal-binding site. As shown in Figure 4A and 4B, the expression of Myc-tagged CSN-5H127A, CSN-5H129A, or CSN-5D140N in the csn-5KO strain partially rescued the degradation of FRQ in the csn-5KO strain. FRQ was degraded slightly slower in the mutants than the wild-type strain or the csn-5KO strain complemented by wild-type CSN-5, with a half-life of ∼5 h, consistent with the prolonged period of the conidiation rhythms in the csn-5KO strains expressing the mutant CSN-5, indicating that both deneddylation activity and integrity of CSN are needed in this process. Taken together, these results demonstrate that CSN-5 with point mutations in the JAMM metal-binding site partially restore the SCF-mediated FRQ degradation.
Previous studies showed that FRQ ubiquitination and degradation is mediated by the SCFFWD-1 E3 ligase complex [24], [36], and that the stability of E3 ligase components is controlled by CSN in vivo [3], [7], [16], [24]. Because the ectopic expression of mutated CSN-5 partially rescued both the circadian conidiation rhythm and FRQ degradation in the csn-5KO strain, we decided to check whether CSN with mutant CSN-5 can prevent the degradation of components of the SCFFWD-1complex. As shown in Figure 5A, Myc-Cul1 was stable after induced expression of Myc-CSN-5 in the csn-5KO strain, with a half-life of >9 h in the presence of CHX, similar to that of the wild-type strain. In the csn-5 mutant, however, both the neddylated and unneddylated Myc-Cul1 became very unstable, with a half-life about 1.5 h (Figure 5A and 5D) [16]. Interestingly, the expression of JAMM mutant CSN-5 had a differential effect on the neddylated and unneddylated Cul1. In mutant CSN-5 transformants, the stability of neddylated Cul1 was only partially rescued, with a half-life of >3 h in the presence of CHX (Figure 5A and 5D), whereas the stability of unneddylated Cul1 was almost rescued, with a half-life of >12 h (Figure 5A and 5D). These data indicate that although CSN containing JAMM mutated CSN-5 fails to cleave NEDD8 from neddylated Cul1, it still functions to protect hyperneddylated and unneddylated Cul1 from degradation to a certain extent.
In N. crassa, deletion of csn-5 or csn-3 has no effect on the stability of SKP-1 protein in the SCFFWD-1 complex [16]. As expected, Myc-SKP-1 were very stable in the wild-type strain and the csn-5KO strain and in the complementation strains with mutant CSN-5, with a half-life of >12 h (Figure 5B and 5E).
FWD-1, the substrate-recruiting subunit of the SCFFWD-1 complex, was quite stable in the wild-type strain, whereas it became undetectable after only 3 h of CHX treatment in csn-5KO strain (Figure 5C and 5F). In the csn-5H127A, csn-5H129A, and csn-5D140N strains, however, FWD-1 signals could still be detected after 6 h of CHX treatment (Figure 5C and 5F), indicating that CSN with mutated CSN-5 partially functions to protect F-box proteins from degradation. This finding further confirms that regulation of SCF-mediated FRQ degradation by the CSN is a key step in the N. crassa circadian clock. Therefore, both the deneddylation activity and the integrity of the CSN are important for preventing the degradation of components of the SCFFWD-1 complex.
We next asked whether CSN with mutated CSN-5 still functions to protect other CRL substrate receptors from degradation. N. crassa SCON-2, an F-box protein involved in regulating sulfur metabolism, was previously shown to interact with SKP-1 and is very unstable in a csn-2KO strain [24], [37]. We compared the stability of Myc-SCON-2 in wild-type, csn-5KO and csn-5KO expressing wild-type CSN-5 or mutant CSN-5H127A strains. The half-life of Myc-SCON-2 was approximately 12 h in the wild-type and csn-5KO expressing wild-type CSN-5 strains in the presence of CHX. Myc-SCON-2 was very unstable in the csn-5 mutant and became undetectable after 3 h of CHX treatment (Figure 6A and 6B). In the csn-5H127A strain, the detection of Myc-SCON-2 signal extended to 6 h after CHX treatment (Figure 6A and 6B). FBP94 (NCU04785), another F-box–containing protein in N. crassa, can also interact with SKP-1 (data not shown). As shown in Figure 6C and 6D, FBP94 was quite stable in the wild-type strain and csn-5KO strain complemented with Myc-CSN-5, whereas in the csn-5KO strain it became undetectable after only 6 h of CHX treatment. In the csn-5H127A strain, detection of FBP94 signal extended to 12 h after CHX treatment (Figure 6C and 6D). Therefore, CSN complex with mutated JAMM domain can partially function in maintaining the stability of other F-box–containing adaptor proteins.
In a previous study, we determined that the N. crassa Cul3 protein interacts with BTB1 protein, and both proteins become unstable in the csn-5KO strain [16]. The half-life of Myc-BTB1 was >12 h in the wild-type and csn-5KO expressing wild-type CSN-5 strains in the presence of CHX, whereas in the csn-5KO strain it became undetectable after 6 h of CHX treatment (Figure 6E and 6F). As expected, in the csn-5H127A strain, BTB1 signals were detectable at 12 h after CHX treatment (Figure 6E and 6F), indicating that CSN with the JAMM mutated CSN-5 still partially functions to protect the substrate adaptor proteins of CRL3 from degradation. We also investigated whether CSN with the JAMM mutated CSN-5 regulates the substrate receptor protein of CRL4 in a similar manner. N. crassa Cul4 was previously shown to interact with DCAF11, a putative substrate receptor of CRL4DCAF11 [16], [38]. As shown in Figure 6G and 6H, the half-life of Myc-DCAF11 was >12 h in the wild-type and csn-5KO expressing wild-type CSN-5 strains in the presence of CHX, whereas in the csn-5KO strain it became undetectable after 6 h of CHX treatment [16]. As expected, in the csn-5H127A strain, the detection of DCAF11 signal was extended to 9 h after CHX treatment (Figure 6G and 6H). Taken together, these in vivo results indicate that the CSN complex containing mutant CSN-5 efficiently prevents degradation of substrate receptors of CRLs.
Current models suggest that the activity and assembly of CRLs are controlled by cycles of CRL deneddylation and CAND1 binding of deneddylated cullins [39], [40], [41], [42], [43], [44], [45], [46]. In plants and worms, CAND1 mutants exhibit defects consistent with a positive role in regulating the function of a subset of CRLs [40], [47], [48], [49]. However, in yeast and human cells, loss of CAND1 has little effect on the abundance of neddylated cullins, suggesting that the neddylation/deneddylation cycle may function independently of CAND1 [50], [51]. To test whether CAND1 is involved in maintaining the function of CRLs in N. crassa, we examined the role of CAND1 in the regulation of circadian conidiation rhythm and proper functioning of the SCFFWD-1 complex. We first measured the growth rates of the wild-type and cand1KO strains by race tube assay under constant darkness. The growth of the cand1KO strain (about 3.0 cm per day) was slightly slower than that of the wild-type strain (about 3.7 cm per day), suggesting that CAND1 is involved in regulating hyphal growth. After entrainment by light, like the wild-type strain, the cand1KO strain exhibited a robust circadian conidiation rhythm with a period of about 22 h at 25°C in constant darkness (Figure 7A), suggesting that CAND1 is not required for circadian rhythms in N. crassa. To test whether CAND1 functions in a manner similar to the CSN, we examined the conidiation rhythms of the cand1 mutant in LD cycles (12 h light/12 h dark). As shown in Figure 7B, the conidiation rhythms of the cand1KO strain were entrained by LD cycles, indicating that unlike CSN, CAND1 is not required in light regulation of the circadian clock. We also examined the responses of the cand1 mutant to temperature entrainment using race tube assays. As expected, in 12 h 27°C/12 h 22°C temperature cycles, as shown in Figure 7C, like the wild-type strain, the conidiation rhythm of the cand1KO strain was synchronized by the temperature cycles, indicating that CAND1 is not required for the temperature-entrained conidiation process. These results suggest that CAND1 does not play a significant role in the regulation of circadian rhythm in N. crassa.
Deletion of cand1 also had no effect on degradation of the clock protein FRQ, which is the substrate of the SCFFWD-1 ubiquitin ligase complex in N. crassa (Figure 8A and 8B). We also examined the stability of FWD-1 of the SCFFWD-1 complex in the cand1KO strain. As shown in Figure 8C and 8D, FWD-1 was very stable in the cand1 mutant, as in the wild-type strain, with a half-life of >12 h. Together, these results suggest that CAND1 is not required for regulation of the circadian rhythm and for maintaining the proper function of the SCFFWD-1 complex in N. crassa.
In eukaryotes, the COP9 signalosome (CSN) is a highly conserved multifunctional complex that has two major biochemical roles: cleaving NEDD8 from cullin proteins [1], [3], [7], [11] and maintaining the stability of the CRL components [7], [24]. In this study, we used mutation analysis to confirm that the JAMM metal-binding motif of the CSN-5 subunit is responsible for NEDD8 cleavage from cullin proteins in N. crassa. Point mutations of the key residues in the metal-binding motif (EXnHXHX10D) of the CSN-5 disrupted CSN deneddylation activity without interfering with the CSN assembly. We demonstrated that those mutant CSN-5s could almost restore the growth and conidiation defects of the csn-5KO strain. Furthermore, even without the deneddylation activity, the CSN partially maintained the stability of the SCFFWD-1 complex and partially restored the degradation of clock protein FRQ in vivo. Finally, we also showed that CSN containing mutant CSN-5 could efficiently prevent the degradation of the substrate receptors of CRLs. In addition, deletion of the CAND1 ortholog in N. crassa had little effect on the circadian rhythm of conidiation. Thus, our results suggest that maintenance of CRL stability by the CSN integrity is even more crucial in hyphal growth, conidial development, and circadian function in N. crassa.
As the key regulator of CRLs, both deneddylation and maintenance of CRL stability by the CSN occurs when the CSN binds to CRLs. Thus, it is difficult to distinguish which function is more important for maintaining the proper function of CRLs in eukaryotes. To precisely determine the function of the CSN in maintaining the stability of CRLs, we sought to separate the two functional aspects of CSN from one other in N. crassa. In those csn-5KO strains expressing CSN-5 proteins with different point mutations in the JAMM metal-binding motif, the deneddylation activity was disrupted, while the assembly of the CSN complex and interactions between CSN and cullin proteins were not affected. Therefore, this system has great potential as a model for distinguishing between the two activities of the CSN.
A recent study suggests that neddylated Cul1 and Cul3 are unstable in D. melanogaster csn mutant cells due to a defect in CSN deneddylation activity, whereas unneddylated cullins are stable in csn-5 mutant cells [33]. The results presented here show that unstable forms of Cul1 in the csn-5 mutant were partially restored by expression of mutant CSN-5 protein without deneddylation activity. Like the stability of unneddylated Cul1 and Cul3 in D. melanogaster CSN-5–defective cells [33], unneddylated Cul1 remained stable in the csn-5H127A, csn-5H129A, and csn-5D140N strains, similar to that in the wild-type strain (Figure 5A), indicating that CSN integrity with catalytically dead CSN-5 effectively maintains the stability of cullins. Studies in D. melanogaster and A. nidulans CSN-5 mutants indicated that the CSN deneddylation activity is essential for cell differentiation and developmental initiation [11], [19], [22], [33]. However, in the A. thaliana fus6/C231 mutant (a CSN1 N-terminal deletion mutant), although the Cul1 neddylation still works in a wild-type pattern, it was lethal and exhibited severe gene expression defects [52]. This genetic evidence raises questions concerning whether the CSN has other important functions aside from its deneddylation activity. The accelerated degradation of c-Jun in HeLa cells in which CSN-5 is downregulated is rescued equally by over-expression of the deneddylation mutant CSN-5D151N or wild-type CSN-5 [20]. These data suggest that two activities of CSN may function parallel for regulating the activity of CRLs.
Bennett et al. found that Cul1K720R (a constitutively unneddylated Cul1 mutant) assembles with CSN, SKP-1, and most F-box proteins to the same extent as wild-type Cul1 [51]. Our IP experiments also show that wild-type CSN interacts with both neddylated and unneddylated Cul4. These findings suggest that the CSN can interact with CRLs independent of the prior neddylation of cullins. In plants, genetic results also suggest that during early embryo development and germination, neddylation/deneddylation cycling is not absolutely required, although it becomes more important during seedling establishment and later in development [21], suggesting that the CSN has distinct biochemical functions that orchestrate development in the appropriate spatial and temporal setting.
Protection of substrate receptors by the CSN has been described for the CRLs in vivo [24], [25], [26], [27], [28]. We found that CSN with mutated CSN-5 had a contribution to the stabilities of five receptor proteins of CRLs in vivo. These results provide evidence for the idea that the abundance of adaptor modules (rather than cycles of neddylation/deneddylation and CAND1 binding) drives CRL network organization [51]. This possibility is supported by our genetic observations that the csn-5H127A, csn-5H129A, and csn-5D140N strains exhibited normal growth and conidiation phenotypes. The integrity of CSN was maintained in these JAMM mutation complementation strains, thus it can serve as a platform to recruit other proteins for regulating the activities of CRLs, such as the recruitment of UBP12 in yeast, as well as USP15 in human [14], [53]. In addition, a non-catalytic CSN itself may stabilize the substrate receptors of CRLs. A very recent study has shown that the protective effect of the CSN on DDB2 and CSA autoubiquitination is independent of CSN-5 mediated deneddylation in vitro [29]. These results suggest that the partial rescue of stability of substrate receptors by the catalytically dead CSN is mainly dependent on its protective effect. Therefore, the stability of cullins and some substrate receptors of CRLs are dependent on both deneddylation activity and integrity of the CSN in N. crassa.
The csn-5KO strain exhibits abnormal conidiation rhythms in DD, which cannot be entrained by either LD or temperature cycles, indicating that light and temperature regulation of the conidiation process is impaired in this mutant [16]. We found that degradation of the clock protein FRQ is impaired in the csn-5KO strain, especially when protein synthesis is completely blocked. To further characterize the molecular mechanism of how the CSN regulates the conidiation rhythm, we focused on the SCFFWD-1 ubiquitin ligase, which controls the N. crassa circadian rhythm by ubiquitinating FRQ [36]. Our results demonstrated that defective FRQ degradation in the csn-5KO strain is due to the drastically reduced stability and levels of FWD-1 and Cul1 proteins in the SCFFWD-1 complex. Ectopic expression of mutant CSN-5 without deneddylation activity restored the defects of growth and conidiation in the csn-5KO strain, and almost restored the defects of the circadian conidiation rhythm in DD and FRQ degradation in the csn-5KO strain. Our data further showed that the low levels of FWD-1 in the csn-5KO strain were dramatically increased after expression of each of the CSN-5 proteins with point mutations in the JAMM metal-binding site, however, the increased stability and levels of the components in the SCFFWD-1 ubiquitin ligase are not enough to fully restore the degradation of FRQ to wild-type level, indicating that regulation of FRQ degradation plays a key role in maintaining the precise period length of conidiation rhythm in N. crassa. This is further supported by the finding that accelerated degradation of c-Jun in HeLa cells in which CSN-5 is downregulated can be rescued equally by over-expression of the deneddylation mutant CSN-5D151N or wild-type CSN-5; however, accelerated c-Jun degradation is not rescued in CSN-1– or CSN-3–downregulated cells by over-expression of wild-type CSN-5 [20]. Furthermore, the degradation of EB1 (microtubule-end-binding protein 1) is accelerated by over-expression of wild-type CSN-5 or CSN-5D151N in HeLa cells [20]. These results suggest that the integrity of CSN might contribute more to regulating the stability of some substrates of CRLs. Current models suggest that the CRL complex is controlled by cycles of CRL deneddylation and CAND1 binding [7]. Our experiments further suggested that CAND1, a putative regulator of CRLs, is not required for maintenance of SCFFWD-1 ubiquitin ligase activity and circadian rhythm in N. crassa. These data provide additional evidence that the CSN is an important regulator of the circadian clock in N. crassa through maintenance of SCFFWD-1 ubiquitin ligase stability.
In conclusion, the results of our experiments indicate that even without deneddylation activity, the N. crassa CSN can still regulate hyphal growth, conidial development, and circadian function by regulating the activities of E3 ubiquitin ligases. Because the function of the CSN in the regulation of CRL activities is conserved in higher eukaryotes, we propose that the CSN may have a similar role in plants and animals.
The N. crassa strain 87-3 (bd, a) was used as the wild-type strain in this study. The bd ku70RIP strain, which was generated previously [54], was used as the host strain for creating the cand1 knockout mutants. We also used csn-5KO, csn-2KO and csn-5KO, his-3 strains that were generated previously [16]. The 301-6 (bd, his-3, A) strain and the csn-5KO, his-3 strain were used as the host strains for the his-3 targeting construct transformation [24]. Liquid culture conditions were the same as described previously [34]. For QA-induced protein expression, 0.01 M QA (pH 5.8) was added to liquid medium containing 1× Vogel's medium, 0.1% glucose, and 0.17% arginine. The medium for the race tube assay contained 1× Vogel's, 0.1% glucose, 0.17% arginine, 50 ng/mL biotin, and 1.5% agar [55]. For race tubes containing QA (10−3 M), glucose was excluded from the medium.
All three JAMM point mutations of CSN-5 were generated using the Quikchange Site-Directed Mutagenesis Kit (Stratagene). pUC19-CSN-5 was used as the template for mutagenesis. Afterwards, the mutated CSN-5 DNA fragments were subcloned into either the pqa-5Myc-6His or pqa-3Flag vectors. The triple point mutant CSN-5 (H127A, H129A and D140N) generated from pUC19-CSN-5 was subcloned into the endogenous csn-5 promoter-driven vector pcsn-5-Myc-His-CSN-5, resulting in pcsn-5-Myc-His-CSN-5tri. The previously constructed plasmids pqa-Myc-Cul1, pqa-Myc-His-Cul3, pqa-Myc-His-Cul4, pqa-Myc-His-CSN-6, pqa-Myc-His-SCON-2, pqa-Myc-His-FBP94, pqa-Myc-His-BTB1, and pqa-Myc-His-DCAF11 were also used for his-3 targeting transformation in the csn-5KO, his-3 and 301-6 (bd, his-3, A) strains [16] and cotransformation in the csn-5H127A, csn-5H129A, and csn-5D140N strains.
GST-Cul4 (containing Cul4 amino acids 1–113) fusion protein was expressed in RIL cells and the recombinant protein was purified and used as the antigen to generate rabbit polyclonal antiserum, as described previously [56].
The csn-5KO, Myc-His-CSN-5H127A, csn-5KO, Myc-His-CSN-5H129A, or csn-5KO, Myc-His-CSN-5D140N strain, wild-type strain (negative control), and csn-5KO, Myc-His-CSN-5 strain (positive control) were cultured for approximately 24 h in constant light (LL) in liquid medium containing QA (0.01 M QA, 1× Vogel's medium, 0.1% glucose, and 0.17% arginine). Approximately 10 g of tissue from each strain grown in LL was harvested. The purification procedure was the same as described previously [16]. Fractions containing purified Myc-tagged CSN proteins were immunoprecipitated by adding 25 µL of c-Myc monoclonal antibody-coupled agarose beads (9E10AC, Santa Cruz Biotechnology). The precipitates of each sample were analyzed by SDS-PAGE (4%–20% acrylamide), which was subsequently silver stained following the manufacturer's instructions (ProteoSilver Plus, Sigma). Specific bands in the Myc-His-CSN-5 purified products or in the Myc-His-CSN-5H127A purified products were excised and subjected to tryptic digestion and LC-MS/MS.
The protocol of gel filtration chromatography was the same as described previously [16], [21]. Briefly, purified proteins (400 µg) were loaded onto a Superdex™ 200 (GE) gel filtration column that was equilibrated with 25 mL (150 mM NaCl, 20 mM Tris Cl pH 7.4). The proteins were eluted in the same buffer at a flow rate of 0.3 mL/min. Fractions of 0.4 mL were collected starting from the onset of the column void volume (8.0 mL) and finishing at 18 mL (25 fractions). 20 µL of each fraction were prepared in 20 µL of 2× SDS loading buffer, separated by 7.5% SDS-PAGE, and then examined by Western blot analysis using c-Myc antibody (9E10, Santa Cruz Biotechnology).
Protein extraction, quantification, western blot analysis, protein degradation assays, and immunoprecipitation assays were performed as described previously [24], [56]. Western blot analyses using a monoclonal c-Myc antibody (9E10, Santa Cruz Biotechnology) or Flag antibody (F3165-5MG, Sigma) were performed to identify the positive transformants. Immunoprecipitates or equal amounts of total protein (40 µg) were loaded into each protein lane for SDS-PAGE. After electrophoresis, proteins were transferred onto a PVDF membrane, and western blot analysis was performed using c-Myc antibody, Flag antibody, FWD-1 antiserum, FRQ antiserum, or Cul4 antiserum.
|
10.1371/journal.pmed.1002487 | Immune-related genetic enrichment in frontotemporal dementia: An analysis of genome-wide association studies | Converging evidence suggests that immune-mediated dysfunction plays an important role in the pathogenesis of frontotemporal dementia (FTD). Although genetic studies have shown that immune-associated loci are associated with increased FTD risk, a systematic investigation of genetic overlap between immune-mediated diseases and the spectrum of FTD-related disorders has not been performed.
Using large genome-wide association studies (GWASs) (total n = 192,886 cases and controls) and recently developed tools to quantify genetic overlap/pleiotropy, we systematically identified single nucleotide polymorphisms (SNPs) jointly associated with FTD-related disorders—namely, FTD, corticobasal degeneration (CBD), progressive supranuclear palsy (PSP), and amyotrophic lateral sclerosis (ALS)—and 1 or more immune-mediated diseases including Crohn disease, ulcerative colitis (UC), rheumatoid arthritis (RA), type 1 diabetes (T1D), celiac disease (CeD), and psoriasis. We found up to 270-fold genetic enrichment between FTD and RA, up to 160-fold genetic enrichment between FTD and UC, up to 180-fold genetic enrichment between FTD and T1D, and up to 175-fold genetic enrichment between FTD and CeD. In contrast, for CBD and PSP, only 1 of the 6 immune-mediated diseases produced genetic enrichment comparable to that seen for FTD, with up to 150-fold genetic enrichment between CBD and CeD and up to 180-fold enrichment between PSP and RA. Further, we found minimal enrichment between ALS and the immune-mediated diseases tested, with the highest levels of enrichment between ALS and RA (up to 20-fold). For FTD, at a conjunction false discovery rate < 0.05 and after excluding SNPs in linkage disequilibrium, we found that 8 of the 15 identified loci mapped to the human leukocyte antigen (HLA) region on Chromosome (Chr) 6. We also found novel candidate FTD susceptibility loci within LRRK2 (leucine rich repeat kinase 2), TBKBP1 (TBK1 binding protein 1), and PGBD5 (piggyBac transposable element derived 5). Functionally, we found that the expression of FTD–immune pleiotropic genes (particularly within the HLA region) is altered in postmortem brain tissue from patients with FTD and is enriched in microglia/macrophages compared to other central nervous system cell types. The main study limitation is that the results represent only clinically diagnosed individuals. Also, given the complex interconnectedness of the HLA region, we were not able to define the specific gene or genes on Chr 6 responsible for our pleiotropic signal.
We show immune-mediated genetic enrichment specifically in FTD, particularly within the HLA region. Our genetic results suggest that for a subset of patients, immune dysfunction may contribute to FTD risk. These findings have potential implications for clinical trials targeting immune dysfunction in patients with FTD.
| Frontotemporal dementia (FTD) is the leading cause of dementia in individuals less than 65 years old.
Currently, there is no approved treatment of FTD and no diagnostic tests for predicting disease onset or measuring progression.
Increasing evidence suggests that inflammation and immune system dysfunction play an important role in the pathogenesis of FTD.
We used summary data from genome-wide association studies to investigate genetic overlap, or “pleiotropy,” between FTD and a variety of immune-mediated diseases.
Through this approach, we found extensive FTD–immune genetic overlap within the HLA region on Chromosome 6, an area rich in genes related to microglial function, as well as in 3 genes not previously identified as contributing to the pathophysiology of FTD.
Pointing to the functional relevance of these genetic results, we found that these candidate FTD–immune genes are differentially expressed in postmortem brains from patients with FTD compared to controls, and in microglia/macrophages compared with other central nervous system cells.
Using bioinformatics tools, we explored protein and genetic interactions among our candidate FTD–immune genes. These results suggest that rather than a few individual loci, large portions of the HLA region may be associated with increased FTD risk.
Immune dysfunction may play a role in the pathophysiology of a subset of FTD cases.
For a subset of patients in whom immune dysfunction in general—and microglial activation in particular—is central to disease pathophysiology, anti-inflammatory treatment is an important area for further investigation.
| Frontotemporal dementia (FTD) is a fatal neurodegenerative disorder and the leading cause of dementia among individuals younger than 65 years of age [1]. Given rapid disease progression and the absence of disease-modifying therapies, there is an urgent need to better understand FTD pathobiology to accelerate development of novel preventive and therapeutic strategies.
Converging molecular, cellular, genetic, and clinical evidence suggests that neuroinflammation plays a role in FTD pathogenesis. Complement factors and activated microglia, key components of inflammation, have been established as histopathologic features in brains of patients [2] and in mouse models of FTD [3,4]. Genome-wide association studies (GWASs) have shown that single nucleotide polymorphisms (SNPs) within the immune-regulating human leukocyte antigen (HLA) region on Chromosome (Chr) 6 are associated with elevated FTD risk [5]. Importantly, there is increased prevalence of immune-mediated diseases among patients with FTD [6,7]. Together, these findings suggest that immune-related mechanisms may contribute to and potentially drive FTD pathology.
Recent work in human molecular genetics has emphasized “pleiotropy,” where variations in a single gene can affect multiple, seemingly unrelated phenotypes [8]. In the present study, we systematically evaluated genetic pleiotropy between FTD and immune-mediated diseases. Using large neurodegenerative GWASs and recently developed tools to estimate polygenic pleiotropy, we sought to identify SNPs jointly associated with FTD-related disorders [9,10]—namely, FTD, corticobasal degeneration (CBD), progressive supranuclear palsy (PSP), and amyotrophic lateral sclerosis (ALS)—and 1 or more immune-mediated diseases including Crohn disease (CD), ulcerative colitis (UC), rheumatoid arthritis (RA), type 1 diabetes (T1D), celiac disease (CeD), and psoriasis (PSOR).
We conducted a meta-analysis of summary data obtained from published data. More specifically, we evaluated complete GWAS results in the form of summary statistics (p-values and odds ratios) for FTD, CBD, PSP, and ALS and 6 immune-mediated diseases, including CD [11], UC [12], RA [13], T1D [14], CeD [15], and PSOR [16] (see Table 1). We obtained FTD GWAS summary statistic data from phase I of the International FTD-Genomics Consortium (IFGC), which consisted of 2,154 clinical FTD cases and 4,308 controls with genotyped and imputed data at 6,026,384 SNPs (Table 1; for additional details, see [5]). The FTD dataset included multiple clinically diagnosed FTD subtypes: behavioral variant (bvFTD), semantic dementia (sdFTD), primary nonfluent progressive aphasia (pnfaFTD), and FTD overlapping with motor neuron disease (mndFTD). These FTD cases and controls were recruited from 44 international research groups and diagnosed according to the Neary criteria [17]. The institutional review boards of all participating institutions approved the procedures for all IFGC sub-studies. Written informed consent was obtained from all participants or surrogates. We obtained CBD GWAS summary statistic data from 152 CBD cases and 3,311 controls at 533,898 SNPs (Table 1; for additional details, see [18]). The CBD cases were collected from 8 institutions, and controls were recruited from the Children’s Hospital of Philadelphia. CBD was neuropathologically diagnosed using the National Institutes of Health Office of Rare Diseases Research criteria [19]. The institutional review boards of all participating institutions approved the procedures for CBD GWAS data. Written informed consent was obtained from all participants or surrogates. We obtained PSP GWAS summary statistic data (stage 1) from the National Institute on Aging Genetics of Alzheimer’s Disease Data Storage Site (NIAGADS, https://www.niagads.org) for 1,114 individuals with autopsy-confirmed PSP and 3,247 controls at 531,451 SNPs (Table 1; for additional details, see [20]). The institutional review boards of all participating institutions approved the procedures for all NIAGADS sub-studies. Written informed consent was obtained from all participants or surrogates. We obtained ALS GWAS summary statistic data from 12,577 ALS cases and 23,475 controls at 18,741,501 SNPs (Table 1; for additional details, see [21]). The ALS GWAS summary statistics and sequenced variants are publicly available through the Project MinE Data Browser (http://databrowser.projectmine.com). The institutional review boards of all participating institutions approved the procedures for all ALS GWAS sub-studies. Written informed consent was obtained from all participants or surrogates.
The pleiotropic enrichment strategies implemented here were derived from previously published stratified false discovery rate (FDR) methods [22,23]. For given phenotypes A and B, pleiotropic “enrichment” between phenotype A and phenotype B exists if the proportion of SNPs or genes associated with phenotype A increases as a function of increased association with phenotype B. To assess for enrichment, we constructed fold enrichment plots of nominal −log10(p)-values for all FTD-related-disorder SNPs and for subsets of SNPs determined by the significance of their association with the 6 immune-mediated diseases. In fold enrichment plots, the presence of enrichment is reflected as an upward deflection of the curve for phenotype A with increasing strength of association with phenotype B. To assess for polygenic effects below the standard GWAS significance threshold, we focused the fold enrichment plots on SNPs with nominal −log10(p) < 7.3 (corresponding to p > 5 × 10−8). The enrichment seen can be directly interpreted in terms of the true discovery rate (1 − FDR).
To identify specific loci jointly involved with each of the 4 FTD-related disorders and the 6 immune-mediated diseases, we computed conjunction FDRs. The conjunction FDR is a test of association between 2 traits [22]. Briefly, the conjunction FDR, denoted by FDRtrait1& trait2, is defined as the posterior probability that a SNP is null for either trait or for both simultaneously, given that the p-values for both traits are as small, or smaller, than the observed p-values. Unlike the conditional FDR, which ranks disease/primary-phenotype-associated SNPs based on genetic “relatedness” with secondary phenotypes [24], the conjunction FDR minimizes the possibility/likelihood of a single phenotype driving the common association signal. The conjunction FDR therefore tends to be more conservative and specifically pinpoints pleiotropic loci shared between the traits/diseases of interest. We used an overall FDR threshold of <0.05, which means 5 expected false discoveries per 100 reported. To visualize the results of our conjunction FDR analysis, we constructed Manhattan plots to illustrate the genomic location of the pleiotropic loci. We ranked all SNPs based on the conjunction FDR and removed SNPs in linkage disequilibrium (r2 > 0.2) with any higher ranked SNP. Key aspects and detailed information on fold enrichment plots, Manhattan plots, and conjunction FDRs can be found in prior reports [22,23,25,26].
To assess whether SNPs that are shared between FTD and immune-mediated disease modify gene expression, we identified cis-expression quantitative trait loci (cis-eQTLs, defined as variants within 1 Mb of a gene’s transcription start site) associated with shared FTD–immune SNPs and measured their regional brain expression in a publicly available dataset of normal control brains (UK Brain Expression Consortium; http://braineac.org/) [27]. We also evaluated cis-eQTLs using a blood-based dataset [28]. We applied an analysis of covariance (ANCOVA) to test for associations between genotypes and gene expression. We tested SNPs using an additive model.
To evaluate potential protein and genetic interactions, co-expression, co-localization, and protein domain similarity for the functionally expressed (i.e., with significant cis-eQTLs) pleiotropic genes, we used GeneMANIA (http://genemania.org), an online web portal for bioinformatic assessment of gene networks [29]. In addition to visualizing the composite gene network, we also assessed the weights of individual components within the network [30].
To determine whether functionally expressed (i.e., with significant cis-eQTLs) pleiotropic genes are differentially expressed in the brains of FTD patients, we analyzed the gene expression of pleiotropic genes. Postmortem expression data from the brains of 17 patients with frontotemporal lobar degeneration with ubiquitinated inclusions (FTD-U) (with and without progranulin [GRN] mutations) and 11 controls were obtained from a publically available dataset (Gene Expression Omnibus [GEO] dataset GSE13162; for additional details, see [31]). These data consist of global gene expression profiles from all histopathologically available regions from human FTD-U and control brains (frontal cortex, hippocampus, and cerebellum) analyzed on the Affymetrix U133A microarray platform. Given the small sample size of each individual region, we combined all 3 regions to maximize statistical power. Details about this dataset and analysis—including the human brain samples used, RNA extraction and hybridization methods, microarray quality control, and microarray data analysis—are provided in the original report [31].
Using a publicly available RNA-sequencing transcriptome and splicing database [32], we ascertained whether the functionally expressed (i.e., with significant cis-eQTLs) pleiotropic genes were expressed by specific cell classes within the brain. The 8 cell types surveyed were neurons, fetal and mature astrocytes, oligodendrocyte precursor cells, newly formed oligodendrocytes, myelinating oligodendrocytes, microglia/macrophages (henceforth “microglia”), endothelial cells, and pericytes (for additional details, see [32]).
Using progressively stringent p-value thresholds for FTD SNPs (i.e., increasing values of nominal −log10[p]), we observed genetic enrichment for FTD as a function of several immune-mediated diseases (Fig 1). More specifically, we found strong (up to 270-fold) genetic enrichment between FTD and RA, and comparable enrichment between FTD and UC, T1D, and CeD, with weaker enrichment between FTD and PSOR and CD.
At a conjunction FDR < 0.05, we identified 21 SNPs that were associated with both FTD and immune-mediated diseases (Fig 2; Table 2). Five of these SNPs demonstrated the opposite direction of allelic effect between FTD and the immune-mediated diseases (Table 2): (1) rs9261536, nearest gene = TRIM15; (2) rs3094138, nearest gene = TRIM26; (3) rs9268877, nearest gene = HLA-DRA; (4) rs10484561, nearest gene = HLA-DQB1; and (5) rs2269423, nearest gene = AGPAT1. Of the remaining 16, 2 SNPs showed strong linkage disequilibrium (LD), suggesting that they reflected the same signal: rs204991 and rs204989 (nearest gene: GPSM3; pairwise D′ = 1, r2 = 1). After excluding SNPs that demonstrated the opposite direction of allelic effect and SNPs that were in LD, we found that 8 of the remaining 15 identified loci mapped to the HLA region, suggesting that HLA markers were critical in driving our results. To test this hypothesis, we repeated our enrichment analysis after removing all SNPs in LD with r2 > 0.2 within 1 Mb of HLA variants (based on 1000 Genomes Project LD structure). After removing HLA SNPs, we saw considerable attenuation of genetic enrichment in FTD as a function of immune-mediated disease (Fig 3), suggesting that the observed overlap between immune-related diseases and FTD was largely driven by the HLA region. Further, to determine causal associations for FTD and the 6 immune-mediated diseases, we applied the recently developed summary-data-based Mendelian randomization (SMR; http://cnsgenomics.com/software/smr/) method. This approach is described in detail within the original report [33]. As shown in S1 Table, results from the SMR analysis have identified significant loci that are consistent with the main findings, which suggest that HLA markers on Chr 6 are critical in driving our pleiotropic results.
Outside the HLA region, we found 7 other FTD- and immune-associated SNPs (Fig 2; Table 2), including 2 in strong LD that mapped to the H1 haplotype of microtubule associated protein tau (MAPT) (LD: rs199533 and rs17572851; nearest genes: NSF and MAPT, pairwise D′ = 1, r2 = 0.94). Beyond MAPT, we found 5 additional novel loci associated with increased FTD risk, namely, (1) rs2192493 (Chr 7, nearest gene = TWISTNB), (2) rs7778450 (Chr 7, nearest gene = TNS3), (3) rs10216900 (Chr 8, nearest gene = CR590356), (4) rs10784359 (Chr 12, nearest gene = SLC2A13), and (5) rs2134297 (Chr 18, nearest gene = DCC) (see Table 2 for additional details).
To evaluate the specificity of the shared genetic overlap between FTD and immune-mediated disease, we also evaluated overlap between the 6 immune-mediated diseases and CBD, PSP, and ALS. For CBD and PSP, a few of the immune-mediated diseases produced genetic enrichment comparable to that seen for FTD (S1–S3 Figs; S2–S4 Tables). For example, we found 150-fold genetic enrichment between CBD and CeD and 180-fold enrichment between PSP and RA. In contrast, we found minimal enrichment between ALS and the immune-mediated diseases tested, with the highest levels of enrichment between ALS and RA (up to 20-fold) and between ALS and CeD (up to 15-fold).
At a conjunction FDR < 0.05, we identified several SNPs associated with both immune-mediated disease and CBD, PSP, or ALS (S4–S6 Figs; S2–S4 Tables). Few of the SNPs shared between CBD, PSP, or ALS and immune-mediated disease mapped to the HLA region. Only 2 PSP–immune SNPs mapped to the region of MLN and IRF4 on Chr 6, and no CBD–immune or ALS–immune SNPs mapped to the HLA region (S4–S6 Figs; S2–S4 Tables).
Beyond the HLA region, we found several overlapping loci between the immune- mediated diseases and CBD, PSP, and ALS (S4–S6 Figs; S2–S4 Tables). For PSP, these were (1) rs7642229 with CeD (Chr 3, nearest gene = XCR1, FDR = 1.74 × 10−2); (2) rs11718668 with CeD (Chr 3, nearest gene = TERC, FDR = 3.00 × 10−2); (3) rs12203592 with CeD (Chr 6, nearest gene = IRF4, FDR = 4.17 × 10−2); (4) rs1122554 with RA (Chr 6, nearest gene = MLN, FDR = 2.09 × 10−2); and (5) rs3748256 with RA (Chr 11, nearest gene = FAM76B, FDR = 2.09 × 10−2). For ALS, these were (1) rs3828599 with CeD (Chr 5, nearest gene = GPX3, FDR = 2.27 × 10−2) and (2) rs10488631 with RA (Chr 7, nearest gene = TNPO3, FDR = 3.42 × 10−2).
To investigate whether shared FTD–immune SNPs modify gene expression, we evaluated cis-eQTLs in both brain and blood tissue types. At a previously established conservative Bonferroni-corrected p-value < 3.9 × 10−5 [34], we found significant cis-associations between shared SNPs and genes in the HLA region on Chr 6 in peripheral blood mononuclear cells, lymphoblasts, and the human brain (see S5 Table for gene expression associated with each SNP). We also found that rs199533 and rs17572851 on Chr 17 were significantly associated with MAPT (p = 2 × 10−12) expression in the brain. Beyond the HLA and MAPT regions, notable cis-eQTLs included rs10784359 and LRRK2 (p = 1.40 × 10− 7) and rs2192493 and TBKBP1 (p = 1.29 × 10−6) (see S5 Table).
We found physical interaction and gene co-expression networks for the FTD–immune pleiotropic genes with significant cis-eQTLs (at a Bonferroni-corrected p-value < 3.9 × 10−5). We found robust co-expression between various HLA classes, further suggesting that large portions of the HLA region, rather than a few individual loci, may be involved with FTD (Fig 4; S6 Table). Interestingly, we found that several non-HLA functionally expressed FTD–immune genes, namely, LRRK2, PGBD5, and TBKBP1, showed co-expression with HLA-associated genes (Fig 4).
To investigate whether the FTD–immune pleiotropic genes with significant cis-eQTLs are differentially expressed in FTD brains, we compared gene expression in FTD-U brains to that in brains from neurologically healthy controls. We found significantly different levels of HLA gene expression in FTD-U brains compared to control brains (Table 3). This was true of FTD-U brains regardless of progranulin gene (GRN) mutation status. In spite of the fact that the FTD GWAS used to identify these genes was based on patients with sporadic FTD (without GRN mutations), GRN mutation carriers showed the greatest differences in HLA gene expression (Fig 5; Table 3). These findings are compatible with prior work showing microglial-mediated immune dysfunction in GRN mutation carriers [3].
For the FTD–immune pleiotropic genes with significant cis-eQTLs, across different central nervous system (CNS) cell types, we found significantly greater expression within microglia compared to neurons or fetal astrocytes (Fig 6A; Table 4). Interestingly, HLA genes showed the greatest degree of differential expression. Four of the FTD–immune HLA-associated genes, namely HLA-DRA, AOAH, HLA-A, and HLA-C, showed highest expression in microglia (ranging from 10 to 60 fragments per kilobase of transcript per million; see Fig 6B). In addition, MAPT was predominantly expressed in neurons, LRRK2 in microglia, PGBD5 in neurons, and TBKBP1 in fetal astrocytes (Figs 6B and S7–S9).
We systematically assessed genetic overlap between 4 FTD-related disorders and several immune-mediated diseases. We found extensive genetic overlap between FTD and immune-mediated disease particularly within the HLA region on Chr 6, a region rich in genes associated with microglial function. This genetic enrichment was specific to FTD and did not extend to CBD, PSP, or ALS. Further, we found that shared FTD–immune gene variants were differentially expressed in FTD patients compared with controls, and in microglia compared with other CNS cells. Beyond the HLA region, by leveraging statistical power from large immune-mediated GWASs, we detected novel candidate FTD associations requiring validation within LRRK2, TBKBP1, and PGBD5. Considered together, these findings suggest that various microglia and inflammation-associated genes, particularly within the HLA region, may play a critical and selective role in FTD pathogenesis.
By combining GWASs from multiple studies and applying a pleiotropic approach, we identified genetic variants jointly associated with FTD-related disorders and immune-mediated disease. We found that the strength of genetic overlap with immune-mediated disease varies markedly across FTD-related disorders, with the strongest pleiotropic effects associated with FTD, followed by CBD and PSP, and the weakest pleiotropic effects associated with ALS. We identified 8 FTD- and immune-associated loci that mapped to the HLA region, a concentration of loci that was not observed for the other disorders. Indeed, only 2 PSP–immune pleiotropic SNPs and no CBD–immune or ALS–immune pleiotropic SNPs mapped to the HLA region. Previous work has identified particular HLA genes associated with CBD, PSP, and ALS [35,36]. In contrast, our current findings implicate large portions of the HLA region in the pathogenesis of FTD. Together, these results suggest that each disorder across the larger FTD spectrum has a unique relationship with the HLA region.
Our results also indicate that functionally expressed FTD–immune shared genetic variants are differentially expressed in FTD brains compared to controls and in microglia compared to other CNS cell types (Fig 6). Microglia play a role in the pathophysiology of GRN+ FTD. Progranulin is expressed in microglia [37], and GRN haploinsufficiency is associated with abnormal microglial activation and neurodegeneration [3]. It is perhaps expected, therefore, that GRN+ brains show differential expression of FTD–immune genes, even though these genetic variants were derived from a GWAS of patients with sporadic FTD (who lack GRN or other established FTD mutations). More surprising is the presence of similar enrichment in GRN− brains, suggesting that dysfunction of microglial-centered immune networks may be a common feature of FTD pathogenesis.
By leveraging statistical power from the large immune-mediated GWASs, we identified novel candidate FTD associations requiring validation within LRRK2, TBKBP1, and PGBD5 and confirmed previously shown FTD-associated signal within the MAPT region. LRRK2 mutations are a cause of Parkinson disease [38] and CD [39]. We previously described a potential link between FTD and the LRRK2 locus [40], and another study using a small sample showed that LRRK2 mutations may increase FTD risk [41]. Together, these results suggest that the extended LRRK2 locus might influence FTD despite common genetic variants within LRRK2 not reaching genome-wide significance in a large FTD GWAS [5]. We suggest that increased expression of LRRK2 in microglia results in proinflammatory responses, possibly by modulating TNF-alpha secretion [42]. TBKBP1 also modulates TNF-alpha signaling by binding to TBK1 (TANK binding kinase 1) [43]; rare pathogenic variants in TBK1 cause FTD-ALS [44–46]. Importantly, elevated CSF levels of TNF-alpha are a core feature of FTD [6,47]. Building on these findings, in our bioinformatics “network”-based analysis, we found that both LRRK2 and TBKBP1 interact with genes within the HLA region (Fig 4). Further, physical interactions between MAPT and the HLA network are compatible with research suggesting that under different conditions reactive microglia can either drive or mitigate tau pathology [48,49]. MAPT mutations, which disrupt the normal binding of tau protein to tubulin, account for a large proportion of familial FTD cases [50]. Together, these findings suggest that LRRK2, TBKBP1, and MAPT may, at least in part, influence FTD pathogenesis via HLA-related mechanisms.
This study has limitations. Particularly, in the original datasets that form the basis of our analysis, diagnosis of FTD was established clinically. Given the clinical overlap among neurodegenerative diseases, we cannot exclude the potential influence of clinical misdiagnosis. As such, our findings would benefit from confirmation in large pathologically confirmed cohorts. In addition, given the complex interconnectedness of the HLA region (see Fig 4), we also were not able to define the specific gene or genes on Chr 6 responsible for our pleiotropic signal. However, given the number of HLA loci associated with increased FTD risk, it may be the case that no single HLA variant will be clinically informative; rather, an additive combination of these microglia-associated genetic variants may better inform FTD risk. This possibility is supported by our observation that the expression levels of FTD-immune shared genetic variants differ on average between FTD brains and controls, but with considerable overlap between the two groups, again suggesting that no single variant is likely to be the determinant of FTD risk (Fig 5). Further, we acknowledge the lack of transcriptomic and epigenetic data that would help to identify possible causal associations and mechanisms driving our pleotropic signal.
In conclusion, across a large cohort (total n = 192,886 cases and controls), we used pleiotropy between FTD-related disorders and immune-mediated disease to identify several genes within the HLA region that are expressed within microglia and differentially expressed in the brains of patients with FTD. Building on prior work [6,7], our results support a disease model in which immune dysfunction is central to the pathophysiology of a subset of FTD cases. These findings have important implications for future work in FTD focused on monitoring microglial activation as a marker of disease progression and investigating anti-inflammatory treatments as modifiers of disease outcome.
|
10.1371/journal.pcbi.1000576 | Attention Increases the Temporal Precision of Conscious Perception: Verifying the Neural-ST2 Model | What role does attention play in ensuring the temporal precision of visual perception? Behavioural studies have investigated feature selection and binding in time using fleeting sequences of stimuli in the Rapid Serial Visual Presentation (RSVP) paradigm, and found that temporal accuracy is reduced when attentional control is diminished. To reduce the efficacy of attentional deployment, these studies have employed the Attentional Blink (AB) phenomenon. In this article, we use electroencephalography (EEG) to directly investigate the temporal dynamics of conscious perception. Specifically, employing a combination of experimental analysis and neural network modelling, we test the hypothesis that the availability of attention reduces temporal jitter in the latency between a target's visual onset and its consolidation into working memory. We perform time-frequency analysis on data from an AB study to compare the EEG trials underlying the P3 ERPs (Event-related Potential) evoked by targets seen outside vs. inside the AB time window. We find visual differences in phase-sorted ERPimages and statistical differences in the variance of the P3 phase distributions. These results argue for increased variation in the latency of conscious perception during the AB. This experimental analysis is complemented by a theoretical exploration of temporal attention and target processing. Using activation traces from the Neural-ST2 model, we generate virtual ERPs and virtual ERPimages. These are compared to their human counterparts to propose an explanation of how target consolidation in the context of the AB influences the temporal variability of selective attention. The AB provides us with a suitable phenomenon with which to investigate the interplay between attention and perception. The combination of experimental and theoretical elucidation in this article contributes to converging evidence for the notion that the AB reflects a reduction in the temporal acuity of selective attention and the timeliness of perception.
| Our visual system keeps pace with a rapidly changing stream of information as we view the natural world. To do so, it uses a strongly regulated system of attentional filters to constrain which visual stimuli are permitted to be fully processed to the level of conscious awareness. This article explores what happens when these filters are opened and closed in response to important visual stimuli. To understand these dynamics, our neural network model provides simulations of the role played by attention. These simulations can be tested by recording neural data in the form of ‘brain waves’ (EEG) and comparing the resultant signals to the output of the model. The data discussed here confirm a prediction of the model, which suggests that after the attentional filter has opened to allow one visual stimulus in, there is increased temporal variability or ‘jitter’ in the subsequent opening of the filter within an interval of about one-half of a second. These results have implications for the way our brains process multiple important stimuli perceived in rapid succession, such as the sequence of events that might occur at a critical moment in an airline cockpit or during an automobile accident.
| During ongoing perception of the world, humans are constantly faced with an abundance of visual sensory information. As this information feeds through the various layers of visual cortex, it is progressively integrated by a sequence of cortical areas that gradually generalise over spatial information to extract complex structural detail [1]. Whereas early visual areas extract orientations, textures and borders, brain areas situated higher in the visual processing pathway can detect complex objects [2]. Bottom-up input flowing through this feedforward hierarchical pathway is constantly monitored for salience (e.g. task relevant or intrinsically prominent features like luminance or orientation pop-outs).
Within this general description of the visual system, attention is considered to play a key role, filtering out irrelevant information and selectively enhancing salient input for further processing. Here we investigate the temporal dynamics of visual attention with regard to its role in conscious perception, which becomes apparent when stimuli are presented in rapid succession [3],[4]. Such circumstances occur in rapid serial visual presentation (RSVP), in which stimuli are presented at a rate of approximately 10 items per second in the same spatial location. As each stimulus replaces its predecessor, its featural representation becomes fleeting due to masking effects, and a transient enhancement by attention is thought to be crucial in ensuring that salient items can be successfully encoded into working memory [5].
An apparent temporal limitation of visual perception is illustrated by the attentional blink (AB; [6]). The AB describes a finding that observers often fail to detect a second target stimulus (T2) presented in short succession (between 100 and 600 ms) after an identified first target stimulus (T1). If T2 is presented in immediate succession to T1, however, detection accuracy is typically excellent (‘lag 1 sparing’; [7]). Behaviourally, the AB has been replicated numerous times [8],[9]. It has also been investigated electrophysiologically [10], where researchers have compared grand average Event-related Potentials (ERPs) evoked by targets outside and inside the AB, to investigate how target processing differs during the AB.
Despite extensive study of the AB, its effect on the underlying temporal mechanisms of target identification remains to be fully explored. Evidence from ERP [10],[11] and priming [12],[13] studies suggest that targets, rather than being completely lost during the AB, are processed quite extensively, but fail to enter the final stage of conscious perception. Furthermore, researchers have found that when targets in RSVP consist of multiple features, observers often report features from items neighboring the target in the RSVP stream and make binding errors referred to as illusory conjunctions [14]. Behavioural analysis of the changes in the patterns of such binding errors provides strong support for the claim that the AB reveals a reduction in the temporal precision of the deployment of transient attention and target processing [15],[16].
In this article, we use the dynamics of temporal visual processing as embodied in the (Simultaneous-Type-Serial-Token) model, a connectionist model of temporal attention and working memory [5], to propose an explanation for the observed effect of the AB on the temporal precision of transient attention. The model explains a broad set of experimental findings relating to the AB, Repetition Blindness and RSVP in general. Before elaborating on our central hypothesis, we explain the fundamental principles of how the model describes temporal attention and working memory. For a more detailed description please refer to [5]. It should be emphasised that throughout this article, we retain the model's parameters as published in [17], and use it to generate predictions and virtual EEG traces comparable to human EEG data.
The model suggests that working memory encoding involves creating a binding between the type of a stimulus (which can include its visual features and semantic attributes) and a token (an episodic representation specific to a particular occurrence of an item) [18],. In the model, Transient Attentional Enhancement (TAE) from the blaster amplifies the type representation of a salient (i.e., task relevant) stimulus to assist in its binding to a token, in a process referred to as tokenization. This TAE can serve as an attentional gate, which can be temporarily deactivated to allow one target's encoding to be completed before a second is begun.
From the perspective of the model, the AB is an artifact of the visual system attempting to assign unique tokens to targets [22]. More specifically, the process of encoding T1 into working memory is triggered by TAE, and TAE itself is subsequently suppressed until T1 encoding has completed. The period of TAE unavailability varies from trial to trial depending on how long it takes to tokenise T1, depending on its bottom-up strength. In an RSVP stream, if a T2 is presented 100–600 ms after a perceived T1 (as is the case during the AB), its processing outcome depends on multiple factors. T2's own strength determines its dependence on TAE, since highly salient T2s can ‘break-through’ the AB [31] and get encoded relatively early. T2s with strength values slightly lower in the range ‘outlive’ the AB (and thus the unavailability of TAE), and hence are indirectly influenced by T1 strength. Overall, the variability in the temporal dynamics of T2's encoding process is influenced both by T1 and T2 strengths. Hence, over all possible strengths, the model proposes that there should be increased variance in processing latency for targets seen during the AB.
This article investigates the hypothesis that diminished attentional control increases the temporal jitter in the latency of a target's working memory consolidation. The AB provides us with a suitable phenomenon with which to test our hypothesis: we propose that the reduced availability of attention during the AB increases the temporal noise in visual attention. To answer this question, we compare the EEG signatures evoked by targets seen outside vs. inside the AB, and determine whether there is a comparative increase in the variability of the latency of working memory encoding of targets presented inside the AB. EEG has the advantage of excellent temporal resolution, allowing us to study short-lived cognitive events that evoke changes in ongoing EEG activity. If one averages over multiple segments of such EEG activity time-locked to the event, the resulting averaged ERP waveform contains a number of positive and negative deflections, referred to as ERP components. To test for increased temporal jitter, we analyse the P3 ERP component, commonly associated with encoding items into working memory [10],[23].
However, analysis of averaged ERP components cannot directly inform our hypothesis. This is because the averaging collapses across and hence discards information about temporal fluctuations in the individual EEG trials contributing to the ERP. Given a set of trials that are averaged together, both decreases in amplitude and increases in latency variation within that set will attenuate the mean amplitude of the ERP. Hence, examining the average does not directly provide the necessary information to decide which of the two sources of variation in the individual trials (amplitude or latency) caused the reduction in ERP amplitude. Further, measures like 50% area latency analysis [24] cannot be used to measure latencies in single trials, due to the levels of irrelevant noise activity. Consequently, we employ time-frequency analysis techniques that provide alternative measures to investigate single trial dynamics underlying grand average ERPs. These methods enable us to perform a more fine-grained analysis of EEG data, and test our hypothesis using both qualitative and quantitative means.
In addition to presenting and analyzing human EEG data, we use the model's neural network implementation to generate virtual P3 ERP components [17], which are hypothesised to correspond to the human P3 ERP component. For each of the experimental conditions, the virtual P3 is contrasted with the human P3, both at grand average and single trial level. This comparative evaluation allows us to validate the model and propose explanations for the human ERP effects.
The following section describes the human EEG activity evoked by targets outside and inside the AB. The data set used in the following analysis was the same as that contributing to the analyses presented in [17]. In the final part of the section, we use the model to generate virtual ERPs, which we compare to human ERPs, and discuss the implications of this comparison for the theory underlying the model. Please refer to the Materials and Methods section for more details on the experimental design and computational modelling.
The experiment consisted of RSVP trials presented at a rate of 105.9 ms per item, with two letter targets, T1 and T2, embedded among digit distractors. T2 was presented at lags 1, 3 and 8 following the T1. The P3 EEG data analysed in this section was recorded at the Pz electrode. Please refer to the Materials and Methods section for further information.
Mean human accuracy for T1 identification was 82%. The accuracy of T2 identification (conditional on correct report of T1) was 83% at lag 1, 54% at lag 3, and 74% at lag 8. There was a significant effect of lag on accuracy (F(1.48,12.58) = 15.58, MSE = 0.03, p0.001, after applying a Greenhouse-Geisser correction on the degrees of freedom). Additionally, in pairwise comparisons, T2 accuracy was significantly lower at lag 3 compared to lag 8 (F(1,17) = 11.66, MSE = .03, p = .003) and lag 1 (F(1,17) = 60.88, MSE = 0.01, p0.001). Consequently, the paradigm employed in this study evoked a reliable AB effect.
The ERPimages [25] in figure 2 compare the P3 evoked by targets seen outside the AB (seen T2s at lag 8 following a seen T1) with targets seen inside the AB (seen T2s at lag 3 following a seen T1). They allowed us to visualise the EEG trials underlying the grand average P3 ERPs (plotted below them) for targets seen outside and inside the AB. These ERPimages represent time with respect to target onset along the X-axis (Note that trials are time-locked to T2 onset), individual trials along the Y-axis, and the single-trial EEG amplitude using a colour scale. The trials comprising these images were sorted from bottom to top by descending order of the phase angle of the single-trial P3 at the time point indicated by the dashed line, which was set to the peak latency of the corresponding grand average P3. This phase angle was estimated at the frequency at which the power of the P3 was maximal. This sorting method effectively ordered the trials according to the approximate latency of the single-trial P3 for a target, as estimated by a wavelet-based time-frequency analysis (see the Materials and Methods section for more details). The ERPimages were then plotted for each condition, with trials having longer latency P3s being placed at the bottom, and trials with shorter latency P3s at the top. Following from our hypothesis, for targets inside the AB, we expected to observe an increased “slope” in the red “smear” representing the P3. This would indicate that these targets suffer greater temporal variance compared to targets outside the AB.
A visual comparison of the ERPimages clearly suggested that the P3 for targets outside the AB (figure 2A) had relatively little variation in the phase angle across most trials. In other words, the P3 onset occurred at approximately the same time in these trials. In contrast, the P3 evoked by targets inside the AB (figure 2B) appeared to exhibit an increased temporal fluctuation, as reflected by the increased variance of the phase angle of the P3 across all trials. A natural consequence of this jitter in the temporal onset of the P3 was a ‘smearing out’ of the grand average ERP.
In summary, if there was indeed a reduction in the precision of the deployment of attention in response to targets during the AB, we expected this to indirectly affect the working memory encoding of targets as reflected by the P3. The ERPimages in figure 2 provided qualitative support for our hypothesis. We then extended this investigation by analysing the distribution of phase angles corresponding to the P3, to generate numerical evidence that could be verified statistically.
To back up the qualitative comparisons of the previous section, we statistically analysed the time-frequency data obtained therein. We used an approach similar to inter-trial phase coherence analysis [25], but adapted the idea to directly examine the subject-wise P3 phase distributions and quantitatively compare temporal jitter. The phase angles used to sort the individual trials comprising the P3 ERPimages in the previous section formed a circular distribution [26] of angular data values that effectively represented the temporal latency between the onset of the target and its P3. By statistically comparing the variance in the distribution of phase angles across targets outside and inside the AB, we tested whether the visual differences observed were consistent across subjects.
To do so, we performed a subject-wise grouping of the P3 phase angles calculated at the peak latency of the grand average P3 for each condition (the same phase angles that were used to sort the ERPimages presented earlier). This generated multiple smaller distributions of P3 phase angles, one per condition and subject. These distributions were then modelled as von Mises distributions [26] for which the concentration parameter was calculated using maximum likelihood estimation. The parameter of a distribution is a measure of its density around its mean value, and is an analogue of the inverse of its variance. The larger the value of a circular distribution, the more concentrated it is around the mean. Importantly, is a linear parameter, and can be compared using conventional statistical tools. Hence, in order to test whether targets inside the AB suffered from increased temporal jitter, we compared values of the subject-wise P3 phase distributions evoked by targets outside and inside the AB, using a standard one-way repeated-measures ANOVA. The results of the ANOVA validated what the visual differences observed in the ERPimages clearly indicated: The of the phase distribution for the P3 for targets outside the AB was statistically greater than that for targets inside the AB: Mean for targets outside the AB was 0.95, whereas mean for targets inside the AB was 0.52 (F(1,17) = 15.21, MSE = 0.11, p = 0.001).
In order to validate the model, we used it to generate ‘artificial electrophysiological’ traces, so-called virtual ERPs [17]. In analogy to human ERP components, we generated virtual ERP components for targets outside and inside the AB. This approach, in addition to allowing us to validate the internal dynamics of the model, provided theoretical explanations for the human EEG effects observed in the previous section. Please refer to the Materials and Methods section for more details on how virtual ERPs and ERPimages were generated.
Our qualitative and quantitative comparisons of human ERPimages support the notion of increased temporal variance in target processing during the AB. Further, we have shown that the observed differences in the phase distributions of targets seen outside and inside the AB are indeed real, and cannot be explained by differences in amplitude or any methodological limitations. Finally, our analysis also suggests that T1 processing significantly influences the variance in T2 processing during the AB window, though this could not be confirmed by a trial-by-trial correlation of T1 and T2 phases. At the end of this section, we interpret this finding in relation to predictions from the model.
The virtual ERPs and ERPimages have provided a means for visualising the theory underlying the model, at a fine-grained level of detail. Using this novel methodology of comparing model and data both at the level of averages and single-trials, we have shown that, in line with the model's hypothesis, activation traces of attentional response and consequent working memory encoding show decreased temporal precision for targets inside the AB compared to targets outside it. However, it is clear from the virtual ERPimages that the virtual P3 for targets inside the AB is exaggerated in terms of its delay and duration. This is a consequence of the strong suppression of TAE in the model during target consolidation. But it does not affect the qualitative comparisons with the human ERPimages, or the conclusions we have drawn therefrom.
To further clarify the causes of temporal variability in the model, we now summarise the underlying mechanisms that produce it. In the model, transient attentional enhancement (TAE) is evoked by detection of a target, and this attention triggers the encoding of that target into working memory by binding its type representation to a working memory token, which results in this target being correctly reported at the end of the trial. For targets presented outside the AB, such as a T2 at lag 8, the TAE mechanism (i.e. the blaster circuit) is readily available. It fires as soon as an item is classified as a target, and encoding is thus tightly timelocked to the target onset. Thus, there is little variation in the tokenization delay and consequently the latency of the virtual P3. Also, because attention is immediately deployed, the model's behavioural accuracy at detecting targets outside the AB is high.
However, as described previously, the processing of a target presented during the AB (a T2) is complicated by multiple factors. Firstly, T1's strength determines the period of unavailability of the blaster. In addition, T2's own strength determines its dependence on the blaster, as highly salient T2s (at upper end of the range of target strength) can break-through the AB [31] and get encoded early. T2s with slightly lower strength values can outlive the AB, but require the blaster's enhancement. Quite a few T2s, however, have insufficient strength to survive the delay in the blaster response and are missed, producing an AB. This complex relationship between T1 and T2 at lag 3 increases temporal variability in the latency of T2's virtual P3, but implies that the model does not predict a strong, direct correlation between T1 and T2 P3 latencies. A possible reason for the lack of any such correlation between the corresponding human P3 phase distributions could be insufficient variation in T1 strength in our experiment, combined with noise obscuring a weak effect. With sufficient variation in T1 strength (for example, when comparing across T1 masked vs. unmasked) the dynamics of the model propose a stronger relationship between the duration of the T1 P3 and the latency of the T2 P3 during the AB. Indeed, the model suggests that there should be a reciprocal influence of T1 strength on its encoding duration [29], which would in turn have implications for T2 P3 latency. Testing for such a relationship would be informative, but a detailed investigation of this topic is beyond the scope of this article.
Our experimental results and theoretical explorations complement and inform previous research on temporal selection and the AB. We now discuss these findings and propose interpretations in terms of the model.
In this article, we have presented human ERP evidence in favour of a reduction in the temporal precision of transient attention during the AB. The AB provides us with a suitable phenomenon with which to investigate the interplay between attention and perception. The interplay between these tightly linked cognitive processes is adversely affected during the AB, producing the reduction in precision observed in behavioural and EEG data.
Using ERPimages, we have provided qualitative evidence arguing for an increase in temporal variation in the dynamics of P3s evoked by targets seen outside vs. inside the AB window. This evidence is supported quantitatively, by statistical comparison of the phase distributions corresponding to the P3. This analysis suggests that there is significantly increased temporal jitter in the ERP activity evoked by targets inside the AB. This notion of a decrease in the temporal precision of attention is inherent in the theoretical framework of the model. Specifically, we have used the model’s neural implementation to generate both virtual ERPs and ERPimages, which we have then compared to their human counterparts. We believe that correlating model and electrophysiological data in this way provides a two-fold benefit. Firstly, it has provided a sufficient explanation for the modulatory effects of the AB on the temporal precision of visual processing. Secondly, it has allowed us to instantiate and test the model at the level of single-trial dynamics, and show that the theoretical assumptions about the nature of temporal visual processing embodied by it can be validated using EEG data, in addition to traditional behavioural verification. We believe that the combination of experimental and theoretical analysis presented in this article contributes to converging evidence for the notion that the AB results in a reduction in the temporal acuity of selective attention, which is an important mechanism for ensuring the timeliness of conscious perception.
This section describes the experiment (the same as Experiment 2 from [17]) used to collect the human EEG data analysed in this article.
|
10.1371/journal.pgen.1005010 | Rapid Evolution of Recombinant Saccharomyces cerevisiae for Xylose Fermentation through Formation of Extra-chromosomal Circular DNA | Circular DNA elements are involved in genome plasticity, particularly of tandem repeats. However, amplifications of DNA segments in Saccharomyces cerevisiae reported so far involve pre-existing repetitive sequences such as ribosomal DNA, Ty elements and Long Terminal Repeats (LTRs). Here, we report the generation of an eccDNA, (extrachromosomal circular DNA element) in a region without any repetitive sequences during an adaptive evolution experiment. We performed whole genome sequence comparison between an efficient D-xylose fermenting yeast strain developed by metabolic and evolutionary engineering, and its parent industrial strain. We found that the heterologous gene XylA that had been inserted close to an ARS sequence in the parent strain has been amplified about 9 fold in both alleles of the chromosomal locus of the evolved strain compared to its parent. Analysis of the amplification process during the adaptive evolution revealed formation of a XylA-carrying eccDNA, pXI2-6, followed by chromosomal integration in tandem arrays over the course of the evolutionary adaptation. Formation of the eccDNA occurred in the absence of any repetitive DNA elements, probably using a micro-homology sequence of 8 nucleotides flanking the amplified sequence. We isolated the pXI2-6 eccDNA from an intermediate strain of the evolutionary adaptation process, sequenced it completely and showed that it confers high xylose fermentation capacity when it is transferred to a new strain. In this way, we have provided clear evidence that gene amplification can occur through generation of eccDNA without the presence of flanking repetitive sequences and can serve as a rapid means of adaptation to selection pressure.
| Xylose is an important component of lignocellulose hydrolysates used for the production of bioethanol, but the yeast Saccharomyces cerevisiae is unable to utilize xylose. Insertion of a bacterial xylose isomerase gene and improvement of growth on xylose by evolutionary adaptation resulted in amplification of this gene and efficient xylose fermentation capacity. Further analysis of the final and intermediate strains from the evolutionary adaptation process revealed interesting features about the mechanisms involved in gene amplification events, which have occurred frequently in natural evolution. We now show that a circular DNA element was spontaneously created by the yeast, encompassing the xylose isomerase gene and an ARS element, present by coincidence adjacent of the inserted xylose isomerase gene. ARS elements are the sites where DNA polymerase initiates duplication of DNA. Interestingly, this has revealed for the first time in yeast that circular DNA plasmids can be created from genomic DNA in the absence of flanking repetitive sequences.
| Microbial evolutionary experiments have received considerable attention in recent years for various reasons. First they allow in depth understanding of the fundamental process of evolution in a rapid and rigorously controlled way [1]. Second, microbial evolution raises great interest in various fields such as in medicine and industrial applications [2–4]. Using nature’s evolutionary principle of variation and selection, microbial evolution has been used for development and optimization of several production host organisms in industrial applications. The speed of fitness gain in a new environment depends on the rate of genetic changes as well as their advantage [5]. Genetic changes that occur during evolution include point mutations, gene deletions or amplifications, and often gene rearrangements involving transposable elements, which in turn might generate deletions or amplifications.
In a broader context, gene duplications and amplifications have played a crucial role in the evolution and genetic diversity of species, in particular for adaptation to restrictive environmental conditions [6,7]. Segmental duplications and amplifications are common in eukaryotes. In the yeast Saccharomyces cerevisiae genome, about 1 out of 5 genes have been identified as duplicates [8]. Moreover, nearly 2% of the coding sequences in S. cerevisiae are tandem gene arrays [9]. Tandem repetitive DNA sequences that include ribosomal DNA (rDNA) and the telomeric loci are very prone to copy number alterations as a consequence of homologous recombination (HR). Such regions play a significant role in the plasticity of the genome. Other repetitive elements like Ty elements and solo Long Terminal Repeats (LTRs) that are widely dispersed in the yeast genome are potential substrates for HR between the short repeats flanking a DNA segment. In spite of the major contribution of repetitive DNA sequences in elevated rates of genome plasticity, segmental amplifications are not restricted to regions with repetitive sequences. However, the generation of tandem gene amplifications from originally single copy sequences is not well understood. The creation of extrachromosomal circular DNA (eccDNA) has been proposed as a possible mechanism for the origin and plasticity of tandem gene repeats [10]. The formation of eccDNA has been attributed to the circularization of a DNA segment from a chromosome during HR between preexisting closely located homologous sequences such as LTRs, resulting in the excision of the DNA segment [11]. There has only been little experimental evidence for the formation of eccDNA in the absence of repeat sequences [12].
The yeast S. cerevisiae has a very long proven record of industrial application, due to its efficient conversion of glucose into ethanol with high productivity, and its substantial tolerance to various inhibitory compounds, including ethanol [13,14]. However, it is unable to efficiently metabolize D-xylose into ethanol. Typically, D-xylose accounts for about one-third of the sugars in lignocellulosic biomass [15,16]. Due to the recent interest in biofuel production with biomass from waste streams and bioenergy crops, engineering S. cerevisiae for efficient D-xylose to ethanol conversion has become an important research focus [17]. Expression of the heterologous structural genes responsible for D-xylose to ethanol conversion in S. cerevisiae did not lead by itself to sufficient productivity for industrial scale application [18].
Recently, using a combination of metabolic and evolutionary engineering strategies, we have developed a robust industrial yeast strain that displayed the highest D-xylose to ethanol conversion rate and yield compared to any other recombinant yeast strain reported previously [19]. Here, we report the elucidation of one of the crucial genetic changes responsible for the rapid D-xylose utilization rate in this strain. Using whole genome sequence comparison of the evolved strain with that of the parent strain, we identified a major copy number variation in the heterologous gene XylA, encoding Clostridium phytofermentans XI (xylose isomerase)), that correlated with the high enzymatic activity measured in crude cell extracts. In addition, we investigated the evolutionary process that resulted in stable integration of this gene in tandem high copy number into the genome using several intermediate strains with varying xylose fermentation rate. We confirmed the formation of self-replicating eccDNA carrying the gene XylA during adaptive evolution in the absence of any homologous sequence flanking the repeat DNA segment. During later stages of the adaptive evolution, the generated eccDNA had integrated into the locus of origin in the genome, generating increasing numbers of tandem repeats first in one of the chromosomes and later in the second chromosome in the diploid strain. We propose that formation of eccDNA can occur in the absence of HR and can serve as a rapid means of adjustment to selection pressure during evolutionary adaptation, especially when higher gene dosage serves as a selective advantage for proliferation or survival.
Recently, we reported the development of an industrial D-xylose utilizing strain of S. cerevisiae, GS1.11–26, using a combination of metabolic engineering, genome shuffling and evolutionary adaptation [19]. Briefly, we integrated the gene XylA coding for Xylose Isomerase from the bacterium Clostridium phytofermentans together with all the known genes important for D-xylose and L-arabinose metabolism into an industrial bioethanol production strain, ER, (Ethanol Red). However, the recombinant strain named HDY.GUF5 was unable to utilize any D-xylose. We then mutagenized this strain by EMS and selected a mutant isolate M315 that was able to grow slowly on D-xylose but had poor D-xylose fermentation capacity. Genome shuffling of this mutant with its parent HDY.GUF5, followed by selection for faster D-xylose growth resulted in little improvement. Subsequently, we performed evolutionary adaptation in serial batch cultures containing D-xylose as main carbon source. A striking observation during this adaptation process was a drastic gain in D-xylose fermentation capacity at the second serial batch culture. We proposed that a crucial genetic change had happened at that step, which resulted in a rapid gain in performance. To elucidate this change, we sequenced the genome of the original parent strain HDY.GUF5 and the final evolved strain GS1.11–26 and performed a global genome sequence comparison.
Since the coverage depth of the sequence reads reveals CNVs among genomes of different strains [20], the sequence coverage of all 16 nuclear chromosomes was analyzed at the nucleotide level with an average window of 500 bp. The log2 ratio of the read depth between the evolved and the parent strains was then calculated and plotted over the length of the chromosomes (Fig. 1). Accordingly, chromosome IX and chromosome XVI showed a 50% higher coverage in the evolved strain compared to the parent strain, indicating duplication of one of the two homologues for each chromosome in the evolved strain.
In chromosome XII, the coverage at the ribosomal DNA locus was reduced in the evolved strain by about 50% relative to the parent strain, indicating the loss of several copies of the ribosomal DNA (rDNA) genes. rDNA genes are present in about 150 to 200 tandem copies in the S. cerevisiae genome [21]. The possible effect of the reduction of the rDNA copies in our evolved strain was not investigated further. However, large deletions of multiple rDNA copies are common spontaneous phenomena in the yeast genome. Strains with up to 50% reduction in the number of rDNA copies compared to wild type laboratory strains did not show any noticeable defect in mitotic growth and meiotic reproduction [22].
The most prominent CNV occurred in a region at the right arm of chromosome XV. This was the region where the D-xylose and L-arabinose metabolism gene cassette had been integrated in the genome of the parent strain HDY.GUF5 [19]. Part of the integrated gene cassette, containing the XylA gene, and a sequence upstream of the integrated cassette, that includes the gene REV1, the tRNA gene tP(UGG)O3 and the autonomously replicating sequence ARS1529, were amplified about 9 times (estimated from the log2 ratio) in the evolved strain compared to the parent strain (Fig. 2). XylA encodes xylose isomerase that converts D-xylose to D-xylulose, the rate-limiting step in D-xylose fermentation. We previously showed that the evolved strain GS1.11–26 displayed significantly higher (about 17 times) XI activity than the parent strain, which displayed only moderate activity [19]. The high copy number of XylA in the evolved strain is therefore consistent with its high XI activity, though the fold increase in the XI activity was higher than that of the copy number.
We sought to understand the structural arrangement of the amplification of the XylA-locus. The presence of the autonomous replication sequence ARS1529 in the amplified XylA-locus made us consider the possibility that this region got amplified as a self-replicating eccDNA. This idea was supported by the observation of break points on either end of the amplified region when the Illumina sequence reads were mapped to the reference genome (Fig. 3A). The sequence reads at one end of the break point contained partially unmatched sequences that matched with the sequence of the opposite end. This condition implies either circular DNA or tandem repeat formation.
To validate this assumption, we performed PCR, (polymerase chain reaction) using a primer set P1, which consisted of a pair of primers inside the amplified region that project outwards in opposite direction (Fig. 3C). A PCR product of 1.7kb was expected only if a circular or tandem repeat sequence had been generated. The evolved strain GS1.11–26 gave a PCR product of the expected size, while no PCR product was obtained from the parent strain HDY.GUF5, the mutant M315 and the original industrial strain Ethanol Red that does not have the cassette (Fig. 3D). The PCR product obtained was then sequenced using conventional Sanger sequencing. The resulting contigs were shown to read through the break point that was obtained from the alignment of Illumina sequence reads at either end of the amplified region (Fig. 3B), indicating a continuous DNA sequence, which in turn points to either a circular form or a tandem repeat.
To differentiate between circular DNA and tandem repeat, we first evaluated if the chromosomal copy of the amplified region had been deleted in the evolved strain. Deletion of the locus would contradict the possibility of tandem amplification of the locus. For this purpose, two sets of primers were used (Fig. 3C). The primer set P2 contained a forward primer annealing upstream of the amplified locus and a reverse primer annealing inside the amplified locus, and detects the presence of the XylA-locus at the right position in the chromosome. The primer set P3, contained a forward primer upstream of the amplified locus and a reverse primer downstream of the amplified locus, and was expected to give a PCR product only upon deletion of the XylA locus since the locus is too large to be amplified with the PCR conditions used (2 min extension time). A band of the expected 1.1 kb size with the PCR set P2 (Fig. 3D) and a negative result with the PCR set P3 with the 2 min extension time was obtained for both the parent HDY.GUF5, the M315 mutant and the evolved strain GS1.11–26. The positive band with PCR set P2 was expected since the whole genome sequencing data indicated that at least one of the alleles was present in the locus. However, the absence of a PCR product using primer set P3 indicated that neither of the two alleles of the chromosomal XylA-locus was deleted. Therefore, neither tandem amplification nor eccDNA could be excluded on the basis of this PCR analysis.
We then performed PCR using the primer set P3 under conditions that allow amplification of the whole amplified XylA-locus (long extension time). A single copy of the XylA-locus in the genome was expected to produce a 9.4 kb PCR product while chromosomal duplication or amplification of the locus should not produce any PCR product since it would be too large to be amplified. The parent HDY.GUF5 and the mutant M315 strains gave rise to a PCR product with the correct size of 9.4 kb but the evolved strain GS1.11–26 did not give rise to any band after several attempts (Fig. 3D). The HDY.GUF5 positive control gave rise to the expected band in all repetitions. This indicates that only one copy of the XylA locus is present in each of the two alleles in the parent strain and the M315 mutant, but that the evolved strain might have multiple copies in both alleles. To confirm this assumption, Southern blot analysis was performed with genomic DNA digested with two different restriction enzymes. First, the DNA was digested with HindIII that cuts only once inside the amplified XylA-locus. A unique probe that hybridizes in the XylA sequence was used to visualize the band. In the presence of only a single copy of the XylA-locus, a single band of 4.3 kb was expected while a circular or tandem repeat sequence should give two bands of 4.3 kb and 7.4 kb (Fig. 4A). Two bands of the expected size were detected for the evolved strain, GS1.11–26. The intensity of the 7.4 kb band was estimated to be about 8 fold higher than the intensity of the lower band, which closely correlated to the amplification level deduced from the whole genome sequence analysis. In the parent strain HDY.GUF5 and the mutant M315, only the 4.3 kb band was detected indicating a single copy of each allele (Fig. 4B). No band was detected in the control strain Ethanol Red, which does not contain the gene cassette in the genome.
We then digested the genomic DNA with SacII, which cuts only outside the amplified XylA-locus, and hybridized with two different probes annealing either inside (same as the previous probe, Fig. 4C), or outside the amplified locus (between the left SacII restriction site and the amplified locus). An 11 kb band was expected if a single copy of the XylA-locus was present in the chromosomal locus. Accordingly, the presence of the expected 11 kb fragment in the strains HDY.GUF5 and M315 using both probes hybridizing inside (Fig. 4D) or outside the amplified XylA-locus confirmed the existence of a single copy of the XylA-locus in both alleles. On the other hand, the evolved strain showed only a higher molecular weight band, both with the inside (Fig. 4D) and outside probes confirming the presence of multiple copies of the XylA-locus in both chromosomal alleles. This result together with the PCR amplification using PCR set P1 (primers directed outwards on either side of the amplified locus) clearly indicates that the amplification of the locus in GS1.11–26 had occurred in the form of a tandem repeat.
As described in our previous report [19], the evolutionary adaptation step used to obtain the strain GS1.11–26 involved a series of 11 sequential batch cultures in D-xylose medium. To verify at which stage of the evolutionary adaptation process the amplification of the XylA-locus had occurred, a sample from the culture before the evolutionary adaption (GS1.0), and samples at the end of the first 5 serial transfers during the evolutionary adaptation (GS1.1, GS1.2, GS1.3, GS1.4 and GS1.5) were tested by PCR for the presence of the tandem amplification or circular DNA formation using PCR primer set P1. Interestingly, a positive PCR result was obtained in all the samples derived from the second culture (GS1.2) onwards, whereas isolates from GS1.0, GS1.1, as well as the original strains used for the genome shuffling step (HDY.GUF5 and M315) did not give rise to the PCR product (Fig. 5). Southern blot analysis of the same samples after HindIII digestion also confirmed the presence of either circular or multiple copies of the locus in the samples obtained from GS1.2 onwards (Fig. 4B). This strongly suggests that amplification of the XylA-locus had occurred at the second step of the evolutionary adaptation process (GS1.2). As we anticipated, the sharp rise in the rate of D-xylose fermentation in the second culture [19], correlated with amplification of the XylA-locus. Although XylA was expressed from a strong promoter in the parent strain HDY.GUF5, the level of expression was not high enough to confer strong D-xylose fermentation capacity. Amplification of the gene likely increased the expression of XI, which in turn alleviated the rate limiting bottleneck for fermentation of D-xylose.
Remarkably, the chromosomal tandem amplification of the XylA locus in GS1.11–26 was not detected in two D-xylose fermenting single cell clones obtained from GS1.2 (second culture) that showed a positive PCR using the primer set P1. When the Southern blot was performed after SacII digestion on these single cell isolates, called GS1.2–2 and GS1.2–6, only the 11kb band was obtained for both strains, which excludes the possibility of chromosomal amplification (Fig. 4D). This was supported by the result of the PCR amplification of the whole amplified XylA-locus using primer set P3, which gave only the expected 9.4 kb band (Fig. 5). Since no smaller PCR band was obtained when this primer set was used (indicating that the XylA-locus was not deleted), and only a single band was obtained with the Southern blot assay (Fig. 4D), both chromosomes should have a single copy of the XylA-locus in these two strains. Given the positive PCR result obtained using primer set P1 (Fig. 5) that indicates either circular or tandem copies of XylA in all the cultures obtained from GS1.2 onwards, and the absence of chromosomal tandem repeats in the culture of GS1.2 clearly indicate that a circular DNA was generated at the second stage (GS1.2) of the evolutionary adaptation process.
We also performed the Southern blot assay with SacII digested DNA using genomic DNA of two other single cell isolates from GS1.4 (4th culture) to test for the presence of the eccDNA. The first isolate GS1.4–14 had the highest D-xylose fermentation rate among all the isolates obtained from the culture GS1.4. Another isolate GS1.4–17 (with only moderate D-xylose fermentation capacity) was also used for comparison. Accordingly, GS1.4–14 showed both the 11kb and a higher molecular weight band (Fig. 4D). Together with the result of the PCR assay shown in Fig. 5, this result clearly indicates the presence of multiple copies of the locus in one of the alleles and a single copy in the other allele in strain GS1.4–14. The strain GS1.4–17 showed only the 11kb band in the Southern blot assay, which was also consistent with the PCR amplification of the whole XylA-locus using primer set P3 (Fig. 5), indicating the presence of a single copy of the XylA-locus in both alleles. Similar to GS1.2–6, strain GS1.4–17 gave a positive PCR result using primer set P1, indicating the presence of a circular DNA in this strain. This result also suggested a correlation between the multiple integration of the XylA-locus in the genome and the faster D-xylose fermentation.
Eventually, we had obtained clear indications that amplification of the XylA-locus had arisen through a circular intermediate in an early stage of the evolutionary adaptation process, and subsequently recombined in tandem array at the same locus in one of the chromosomes. Later, unequal crossover or other mechanisms might have led the tandem array to be copied into the second chromosome, since GS1.11–26 carried the amplified locus in both alleles.
Next, we evaluated the stability of the high xylose fermentation capacity phenotype in GS1.2–6. If the strain GS1.2–6 carried only the circular plasmid and not the genomic XylA amplification, the loss of the plasmid should cause loss of its high D-xylose growth capacity. To allow for loss of the plasmid, GS1.2–6 was grown in rich medium with glucose (YPD, Yeast extract peptone dextrose) for about 25 generations. The culture was spread for single colonies and 43 single cell isolates were tested for growth in liquid YPX (Yeast extract peptone D-xylose) medium. All isolates except one colony lost the ability to efficiently grow in D-xylose medium, consistent with loss of the XylA carrying circular DNA from GS1.2–6 (Fig. 6A). All the 43 colonies were further tested by PCR for the presence or absence of the eccDNA using primer set P1. Accordingly, the eccDNA could be detected in none of the colonies that lost the D-xylose growth capacity except in the one colony that kept the high growth efficiency in D-xylose (Fig. 6B). This indicates that the GS1.2–6 carried only the circular DNA and not the chromosomal amplification of XylA. The rapid D-xylose fermentation capacity by the final strain GS1.11–26 was previously shown to be stable for more than 50 generations [19]. Consequently, we concluded that the stability of the phenotype in GS1.11–26 is due to the integration of the circular DNA into the genome.
To further confirm the presence of the eccDNA, plasmid DNA isolation was performed from the strain GS1.2–6, GS1.4–14 and GS1.11–26, using a protocol modified from Singh and Weil [23]. Cells were pre-grown in 100 ml YPX medium for 24 h to enrich the pXI2–6 plasmid content in the cells. The whole 100 ml culture was used for plasmid isolation (see material and methods). As a result, a substantial amount of pXI2–6 plasmid DNA (more than 1 μg) was obtained from GS1.2–6 (Fig. 7A). On the other hand, the amount of pXI2–6 plasmid DNA obtained from GS1.11–26 and GS1.4–14 was too low to be conclusive. This is probably due to the loss of the pXI2–6 plasmid in the later steps of the evolutionary adaptation process, since there was no longer a need for the strain to maintain the plasmid when enough copies of the essential gene XylA had been integrated in the genome sustaining rapid D-xylose utilization. When the pXI2–6 plasmid isolated from GS1.2–6 was sequenced, a 7483 bp circular sequence was obtained, matching the size of the amplified XylA-locus. The complete sequence of the pXI2–6 plasmid has been provided as supplementary information (S1 Dataset). Though there were several polymorphisms compared to the corresponding sequences in the reference S288c genome, the pXI2–6 plasmid sequence was identical to that of the original parent strain obtained by Illumina sequencing. Restriction analysis using two different enzymes also confirmed the correct size of the isolated pXI2–6 plasmid (Fig. 7A).
We further tested if the isolated pXI2–6 plasmid could be transferred to a new strain and confer the strain with the ability to ferment xylose. For that purpose, strain M315 was chosen since this strain was shown to be able to ferment xylose when XylA was over-expressed [19]. We first deleted both chromosomal copies of XylA from strain M315. Deletion of XylA in M315 completely abolished its growth ability on xylose. Subsequently the isolated pXI2–6 plasmid was transformed into the M315 deletion mutant. We were able to select transformants based on growth on plates containing xylose as a sole carbon source. No colonies were obtained with a control plasmid devoid of the XylA gene. We then evaluated three independent transformants for xylose fermentation capacity. Interestingly, all three transformants carrying the isolated pXI2–6 plasmid were able to efficiently ferment xylose (Fig. 7B) indicating that the isolated pXI2–6 plasmid can be transferred to a different strain and is sufficient to confer the ability to ferment xylose efficiently. To rule out the possibility of integration of the pXI2–6 plasmid in the genome, the three transformants were then grown in YPD medium for about 20 generations and 10 independent colonies from each strain were subsequently checked for growth on xylose as sole carbon source. All of the colonies lost the pXI2–6 plasmid and the ability to grow on xylose, indicating that the XylA gene had not been integrated into the genome and could easily be lost in the absence of selection pressure.
ARS1529 was previously shown to be a functional replication site in yeast [24]. However, compared to the reference S288c genomic sequence, the ARS1529 sequence in the industrial parent as well as in the evolved strains contained a 5 bp deletion just 13 bp downstream of the ARS consensus sequence (ACS) and also 6 SNPs in an AT-rich region downstream of the ACS (Fig. 8). In order to validate the functionality of the modified ARS1529 version in the eccDNA intermediate, we assessed the ability of this sequence to confer self-replication. We first amplified the region containing ARS1529 together with the tRNA coding sequence and the XylA gene from the genomic DNA of the evolved strain GS1.11–26. The PCR product was then cloned into a yeast integrative vector containing kanMx as a selection marker. After transformation of this construct (pXI-ARS) into the mutant yeast strain M315 and selection for growth in the presence of geneticin, we obtained several transformants, with similar transformation efficiency as that of the 2μ-based plasmid, showing that the plasmid was able to propagate using ARS1529. To confirm that ARS1529 alone is sufficient to render replication capacity, we deleted all the sequences originating from the yeast genome, including the tRNA(UGG) and the XylA sequence, except the 236 bp ARS1529 sequence. Transformation of this plasmid into M315 and selection in the presence of geneticin, resulted in several transformants, which were confirmed by PCR for the presence of the plasmid. No transformants could be obtained when the same plasmid devoid of the ARS1529 sequence was transformed into M315. This resulted confirmed that the ARS1529 sequence is sufficient for plasmid replication in yeast.
Copy number variations are a major driving force for rapid genome evolution. The frequency of CNVs in eukaryotes is remarkably high. While many CNVs are detrimental or have no discernible effect, certain gene duplications or amplifications offer adaptive advantage under specific environmental conditions [6]. Typical examples include the copy number increase of the human salivary amylase gene AMY1, that is advantageous in a high starch diet [25] and duplication of genes coding for pepsin in Antarctic fish that allows the fish to efficiently digest at low temperature [26]. In the yeast S. cerevisiae, adaptation to new ecological niches has also been associated with several gene copy number variations and other chromosomal rearrangements, both in natural conditions [27,28] and in artificial evolution experiments [20,29].
In our previous report, we presented a strikingly rapid gain in D-xylose fermentation capacity observed during evolutionary adaptation of a recombinant industrial yeast strain for D-xylose utilization [19]. Here, we showed that amplification of a heterologous gene XylA in the form of an eccDNA, followed by reintegration in multiple tandem repeats in the genome has acted as a mechanism for rapid gain of function.
We compared the whole genome sequence of the original parent strain HDY.GUF5 that was unable to ferment D-xylose with its derivative strain GS1.11–26 that efficiently ferments D-xylose to ethanol. The strain GS1.11–26 was developed by mutagenesis, genome shuffling and evolutionary adaptation from the parent HDY.GUF5. Analysis of CNVs between the two strains revealed amplification of a chromosomal segment where the XylA gene had been integrated. The amplification of XylA correlated with elevated activity of XI. The inherently low activity of XI in the recombinant strains developed previously was a limiting factor for efficient D-xylose utilization [18]. Therefore, the high D-xylose utilization rate of the evolved strain GS1.11–26 is due, at least in part, to the high copy number of XylA, which resulted in high XI activity. A similar, elevated D-xylose utilization rate due to multi-copy integration of XylA, has been reported recently in a strain developed by expression of the Piromyces sp. XI and further evolutionary adaptation [30]. In that report, the original recombinant strain was constructed through expression of XI and xylulose kinase (XK) from a multi-copy plasmid. Since both genes coding for XI and XK present in the plasmid were under the same (but separate) promoter and terminator sequences (TDH3p and CYC1t), it was suggested that duplication and further copy number increase of the XylA gene might have been initiated though unequal crossover of homologous regions in the plasmid. However, the recombined plasmid carrying multiple copies of the XylA gene could not be isolated from the evolved strains due to integration of the plasmid into the genome. In our study, however, the original recombinant strain HDY.GUF5 had been constructed through direct integration of the gene cassette into the genome using an integrative vector. Moreover, we have confirmed the presence of a single copy of the XylA gene in both alleles of the target chromosomal locus in the strain HDY.GUF5 by whole genome sequencing and Southern blot assays. Therefore the amplification of the XylA locus did not start from the original plasmid that was used to construct the recombinant strain.
On the other hand, we have clearly demonstrated the occurrence of self-replicating eccDNA carrying the XylA gene in the course of the evolutionary adaptation process. The eccDNA carried not only the heterologous XylA but also a sequence upstream of the genomic locus in which it had been integrated. It included the gene REV1 that encodes for a Deoxycytidyl transferase, an autonomous replication sequence ARS1529, and a tRNA gene tP(UGG)O3. The presence of these genes in the eccDNA indicates that the eccDNA was generated from the chromosomal locus. During adaptation in a serial batch culture containing D-xylose as main carbon source, a dramatic increase occurred in the rate of D-xylose fermentation just after the second transfer. The eccDNA carrying XylA was detected at this step in which the rapid gain of function was observed, while it was not present in the cultures of the preceding steps, indicating that the eccDNA was apparently formed in that second culture. The dramatic increase in the rate of D-xylose fermentation can now be explained by the gain of this crucial genetic change, an increased copy number of the gene XylA sustaining higher XI activity. We believe that formation of self-replicating eccDNA intermediates that can be maintained in the cell for several generations allows enough time (and therefore, a higher chance) for recombination to happen into the genome in multiple copies. Our results provide convincing evidence for the role of self-replicating eccDNA in the generation of tandem repeat sequences originating from a previous single copy sequence.
Selection pressure that causes a high demand for the product of a specific gene apparently results in a high chance of amplification of the locus in which the gene is located. Examples of this phenomenon have been reported previously. They include amplification of the genes HXT6, SUL1 and GAP1 in response to glucose, sulfur and nitrogen limitation, respectively [11,31–34]. In those and other studies, the explanation for the mechanism of this gene amplification has commonly been attributed to recombination between repetitive sequences flanking the gene. However, such repetitive sequences were not always present, as in the case of SUL1 amplification under sulfur starvation [33,34]. Evidence for amplification of a DNA segment without involvement of repetitive flanking sequences was also reported in natural yeast populations and suggested to occur in the course of natural evolution [35]. The high frequency of such amplification events that do not involve repetitive sequences is not well understood.
A possible mechanism for formation of eccDNA is the involvement of transposons. Comparing the genome sequence of both the parent and evolved strains with the sequence of reference strain S288c, we noted that the reference sequence has a transposon sequence flanked by two LTRs just 3322 bp upstream of the amplified locus. However, this sequence was not present in our strain HDY.GUF5 nor in the evolved strain GS1.11–26. We also confirmed the absence of this sequence by PCR. A second solo LTR sequence inside the amplified XylA-locus is also present in the S288c genome but not in the strain background we used. Hence, we found no evidence for transposon mediated gene rearrangements.
Another possible mechanism involved is initial duplication by unequal sister chromatid exchange or microhomology mediated break induced repair mechanism (MMBIR). In yeast, microhomology mediated segmental duplications have been suggested to commonly occur by BIR mechanisms [36]. MMBIR occurs during DNA repair using single-stranded DNA template carrying a microhomology sequences as short as 5 to 25bp [37]. The single stranded DNA templates might be generated on several occasions: during replication, from stalled transcription complexes, or in promoter regions [36]. Analysis of the sequence of the amplified XylA-locus in our strain background revealed two potential sequences with microhomologies. The first 8 bp sequence homology (GGAAAGGG) is located just at the junction of the 3’ end of the amplified locus and 21 bp downstream of the 5’ end of the amplified locus (Fig. 3A). A second 7 bp homology (TATGATG) flanking the amplified locus is located 14 bp upstream and 15 bp downstream of the amplified locus. The former sequence homology is more likely to be the recombination point as it is located exactly at the break point where the DNA segment has been amplified. Since MMBIR mechanisms are known to induce duplications [36], this region might have first been duplicated and subsequently, one of the duplicated copies might be excised as a circular DNA element.
Though formation of eccDNA frequently involves terminal repeat sequences flanking a mobile DNA region [35], there was no evidence for such repeat sequences in the parent strain HDY.GUF5 nor in the evolved strain GS1.11–26 at the XylA-locus, that might have resulted in HR and excision. In addition, lack of evidence for deletion of the locus in any of the cultures or several single cell isolates indicated that the circular DNA was not formed by initial excision from the genome.
A common feature of previously identified gene amplifications is the presence of an ARS element in the amplified fragment. The genes HXT6, SUL1, GAP1 and CUP1, of which amplification under the appropriate selective conditions has been documented [31,33,34,38], all have an ARS element adjacent to the gene. For instance, in separate evolutionary engineering experiments under sulfur limitation [33,34] all clones characterized contained an amplification of SUL1 and none of SUL2 (which lacks an adjacent ARS element), while both genes encode a high-affinity sulfate permease [39]. This is generally explained by the ARS element being required for eccDNA maintenance. Our work here presented clear evidence for the formation and maintenance of ARS based circular DNA during adaptive evolution.
Chromosomal integration of the eccDNA in the form of a tandem repeat occurred in our work during further evolutionary adaptation. As a consequence, the high D-xylose fermentation phenotype was completely stable for several generations without selection pressure. Multiple integration of the XylA-locus as tandem array in the genome seemed to further improve the D-xylose fermentation rate compared to presence of the eccDNA form. This was evident from the fact that GS1.2–6 containing only the eccDNA still fermented much slower than GS1.4–14, which has the tandem amplification in one of the alleles of the chromosomal locus. In addition, the strain GS1.11–26 that contains the amplified XylA-locus in both alleles, showed still faster D-xylose fermentation than GS1.4–14. This suggests that higher copy numbers of XylA improve D-xylose fermentation capacity. On the other hand, the possibility that other genetic changes might have arisen during the subsequent evolutionary adaptation process that contribute to the higher D-xylose fermentation rate cannot be excluded.
Gene amplification has profound effects on rapid adaptation to new ecological niches. Several of the gene amplifications documented to date are associated with preexisting duplication that involves HR. Gene amplifications arising from single copy genes without flanking homologous sequences have been proposed to be generated by MMBIR mechanisms. Given the high frequency of gene amplification events arising from initially single copy sequences, the presence of an ARS element in many of the amplified sequences reported so far, and the possibility of recombination events with only little sequence homology, self-replicating eccDNA formation followed by tandem gene amplification probably serves as a general, rapid means of adaptation to novel environments that require high expression of a specific gene.
The S. cerevisiae strains used in this study are listed in Table 1. Yeast cells were propagated in yeast extract peptone (YP) medium (10 g/L yeast extract, 20 g/L bacteriological peptone) supplemented with either 20 g/L D-xylose (YPX) or 20 g/L D-glucose (YPD). For solid plates, 15 g/L Bacto agar was added. For batch fermentation, synthetic complete (SC) medium (1.7 g/L Difco yeast nitrogen base without amino acids and without ammonium sulfate, 5 g/L ammonium sulfate, 740 mg/L CSM-Trp and 100 mg/L L-tryptophan) supplemented with 40 g/L D-xylose was used. For selection of strains expressing the KanMX resistance marker, 200 mg/L geneticin was added to the medium. Yeast strains were maintained at -80°C in stock medium composed of YP and 26% glycerol.
Semi-anaerobic batch fermentations were performed in 100 mL SC medium containing 40 g/L D-xylose as carbon source, in cylindrical tubes with cotton plugged rubber stopper. The strains were pre-grown for 24 hours in 5ml YPD medium. For strains carrying plasmid pXI-ARS, geneticin (200 mg/L) was added to the YPD to maintain the plasmid in the strains. The pre-culture was transferred to 50 ml YPD (+ geneticin) and grown to early stationary phase. Cells were harvested and fermentation was started by inoculating the pellet to an initial OD600 value of 5 into 100 ml SC + 4% xylose. The fermentation cultures were continuously stirred magnetically at 120 rpm and incubated at 35°C. The profile of the fermentation was followed based on the rate of CO2 production at different time intervals during the fermentation period. The CO2 production rate was estimated by measuring the weight loss of the fermentation tubes due to CO2 release.
Yeast cells were transformed with the LiAc/SS-DNA/PEG method [40] or by electroporation [41]. Genomic DNA from yeast for PCR amplification was extracted using the PCI [phenol/chloroform/isoamyl-alcohol (25:24:1)] method [42]. PCR was performed with Phusion DNA polymerase (New England Biolabs) for construction of the vectors and sequencing purposes, and ExTaq polymerase (Takara) for diagnostic purposes. Sanger sequencing was performed by the Genetic Service Facility of the VIB, Belgium.
The genomic DNA was extracted using a standard protocol [43]. About 6 μg high quality DNA samples were sent for sequencing to BGI (Hong Kong). The sequencing was conducted by the facility using high-throughput Illumina sequencing technology. A paired end sequence library of 500 bp was constructed and sequence reads of 90 bp were generated. The sequencing reads provided from BGI were aligned to the reference S288c genome sequence using CLC Genomics Workbench5. Out of the 6 million reads with average length of 89.2 bp, 99% matched to the reference sequence when a 93% sequence similarity parameter was used. Additionally, 98% of the reference sequence has been covered with an average coverage depth of 44. The coverage depth per nucleotide position was extracted from the alignment and plotted using GraphPad prism software.
Genomic DNA digested with the appropriate restriction enzyme was run on 0.8% agarose gel overnight at 50 V. A specific probe was prepared by PCR amplification from genomic DNA. The probe was labeled using Amersham Gene Images AlkPhos Direct labeling and detection system (GE Healthcare). The labeled probe was immediately used to hybridize the DNA that was blotted on a nylon membrane. Chemifluorescent signal was generated and detected using CDP-Star as a substrate in conjugation with LAS-4000 luminescent image analyzer.
Plasmid DNA isolation was performed from the strain GS1.2–6, GS1.4–14 and GS1.11–26, using a modified protocol from [23]. Cells were pre-grown in 100 ml YPX medium for 24 h to enrich for the plasmid. The pellet from the 100 ml culture was divided into two and each pellet was resuspended in 5 ml buffer P1 from the QIAGEN plasmid purification kit. Freshly prepared lyticase solution (1.2 M sorbitol, 0.1 M Na3PO4 buffer pH 7.4 and 1 mg/ml lyticase) was added to the mixture and incubated for 45 min at 37°C. Once the cell lysate was obtained at this step, the protocol from the QIAGEN plasmid Maxi kit was followed.
All sequence data have been deposited in the Sequence Read Archive (SRA) at the National Center for Biotechnology Information (NCBI) and can be accessed with references SRX651886 for Samplen56 (HDY-GUF5) and SRX647780 for Samplen57 (GS1.11–26).
|
10.1371/journal.pntd.0000975 | Determinants of Inapparent and Symptomatic Dengue Infection in a Prospective Study of Primary School Children in Kamphaeng Phet, Thailand | Dengue viruses are a major cause of morbidity in tropical and subtropical regions of the world. Inapparent dengue is an important component of the overall burden of dengue infection. It provides a source of infection for mosquito transmission during the course of an epidemic, yet by definition is undetected by health care providers. Previous studies of inapparent or subclinical infection have reported varying ratios of symptomatic to inapparent dengue infection.
In a prospective study of school children in Northern Thailand, we describe the spatial and temporal variation of the symptomatic to inapparent (S:I) dengue illness ratio. Our findings indicate that there is a wide fluctuation in this ratio between and among schools in a given year and within schools over several dengue seasons. The most important determinants of this S:I ratio for a given school were the incidence of dengue infection in a given year and the incidence of infection in the preceding year. We found no association between the S:I ratio and age in our population.
Our findings point to an important aspect of virus-host interactions at either a population or individual level possibly due to an effect of heterotypic cross-reactive immunity to reduce dengue disease severity. These findings have important implications for future dengue vaccines.
| Dengue viruses are a major cause of illness and hospitalizations in tropical and subtropical regions of the world. Severe dengue illness can cause prolonged hospitalization and in some cases death in both children and adults. The majority of dengue infections however are inapparent, producing little clinical illness. Little is known about the epidemiology or factors that determine the incidence of inapparent infection. We describe in a study of school children in Northern Thailand the changing nature of symptomatic and inapparent dengue infection. We demonstrate that the proportion of inapparent dengue infection varies widely among schools during a year and within schools during subsequent years. Important factors that determine this variation are the amount of dengue infection in a given and previous year. Our findings provide an important insight in the virus-host interaction that determines dengue severity, how severe a dengue epidemic may be in a given year, and important clues on how a dengue vaccine may be effective.
| Dengue virus infection can manifest as a clinically inapparent infection, an undifferentiated febrile illness, classic dengue fever (DF), or its more severe form, dengue hemorrhagic fever (DHF) [1]. Inapparent dengue infection is defined as a dengue virus infection that results in no clinical manifestations or an illness that is mild and is not associated with a visit to a health care provider or an absence from school or work due to illness. Inapparent dengue infection because of its nature is not detected by surveillance programs as most programs use visits to a health care provider or hospitalization as an indicator of dengue illness. Thus, inapparent dengue infection represents a burden of dengue infection that goes undetected and hence “inapparent”. The majority of dengue infections in children is thought to be inapparent or present as an undifferentiated febrile illnesses [2], [3]. Determining the incidence of inapparent dengue infection requires detailed prospective cohort studies of populations in dengue endemic areas that can detect dengue infection by paired dengue antibody serology without an associated clinical illness during the time of seroconversion. Few studies of this scope have been performed and thus our knowledge on the full burden of dengue infection is limited. In the first study of this type, a 2-year (1980–1981) school-based study involving 1,757 children, ages 4–16 years, was conducted in Bangkok, Thailand [4]. In this study a symptomatic-to-inapparent (S:I) ratio of 1∶8 was found. A 4-year study was conducted in Managua, Nicaragua in approximately 3,800 schoolchildren, ages 2–9 years, during 2004–2008 [5]. The ratio of S:I infections was 1∶18 during the first year of the study and 1∶3 during the fourth year of the study. A 3-year study, 2000 to 2002, was conducted in 2,536 adults, ages 18–66 years, in West Java, Indonesia [6]. The S:I ratio in this study was 1∶3. Two studies were performed in Kamphaeng Phet, Thailand, Kamphaeng Phet Study I (KPSI) in 1998–2002 and KPSII in 2004–2008. The overall S:I ratio in KPS I was 1∶0.9 and in KPS II an S:I ratio of 1∶3 [7], [8]. We previously reported that there was variation in the S:I ratio during KPS I that varied by school and year with some schools experiencing an S:I ratio of 2∶1 in one year and another school experiencing an S:I ratio of 1∶9 in another year [9].
The pathogenesis of severe dengue disease, DHF, is thought to be a consequence of a heightened immune response due to cross-reactive T-cell responses and/or enhancing dengue antibody during secondary dengue virus infection [10], [11]. It is recognized that dengue infection occurs across the clinical spectrum from subclinical to dengue fever to severe shock and hemorrhage. Understanding the pathogenesis and epidemiology of inapparent dengue infections is important to understand the full burden of dengue infection in a population, its role in underestimating the degree of dengue transmission in a community, and the host and viral factors responsible for influencing subclinical infections such as sequence of serotype infection, heterotypic protective antibody, and host genetic factors. Here we report the spatial and temporal trends of the S:I ratio in dengue infection in a five-year prospective cohort study conducted in Northern Thailand and the epidemiologic factors associated with this variation.
This protocol was reviewed and approved by the Human Use Review and Regulatory Agency of the Office of the Army Surgeon General, the Institutional Review Board of the University of Massachusetts Medical School, and the Thai Ethical Review Board of the Ministry of Public Health, Thailand. All parents/guardians of all children who participated in the study provided informed consent.
The study methods have been previously described [7], [12]. Briefly, this study was conducted in the subdistrict Muang, Kamphaeng Phet Province, Thailand during 1998 to 2002. At the time of the study, subdistrict Muang had 92 public schools of which 12 primary schools were selected to participate in this study based on reliable road access, a desire to participate in the study, and a location within a 3-hour driving radius from the field station laboratory. Children were recruited during January 1998 from grades 1 through 5 and eligible to remain in the study until graduation from 6th grade. During each subsequent year, new 1st grade students were enrolled into the cohort each January. Exclusion criteria included intent to move outside of the study area during the 12 months following enrollment and a history of thalassemia requiring blood transfusion. Baseline demographic information, height and weight, and a blood sample were obtained every January. Evaluations of the entire study population (height, weight, blood sample for dengue serology) occurred three times during the surveillance period (June 1st, August 15th and November 15th) of each year. Active acute illness case surveillance of the study participants occurred during the dengue season from June 1st to November 15th. Students who were absent from school or visited the school nurse with fever were evaluated the same day with a symptom questionnaire and an oral temperature was obtained. Students with a history of fever in the previous 7 days or an oral temperature ≥38°C were evaluated with a physical exam and an acute illness blood sample was obtained. A convalescent blood sample was obtained 14 days later. Throughout June to November, including weekends and holidays, research nurses tracked children who reported to the public health clinic with an illness or were admitted to the hospital. Students were evaluated using the same methods as above.
The laboratory assays used to detect acute dengue infection were polymerase chain reaction [12], hemagglutination-inhibition assays (HAI) [13], and anti-dengue Immunoglobulin (IgM/IgG enzyme immunoassay (EIA) [14]. Dengue virus isolation in Toxorhynchites splendens mosquitoes and enzyme immunoassay were performed for identification of dengue virus serotypes [15], [16], [17], [18].
Clinical definitions of serologically or virologically confirmed dengue virus infection were based on evidence of acute dengue infection and a school-absence associated with fever. By EIA, acute dengue virus infection was defined as a dengue virus-specific IgM level of 40 units or more. Symptomatic dengue virus infection was further classified as symptomatic non-hospitalized or symptomatic hospitalized DF based on their admission into the hospital as decided by the treating physician. Hospitalized symptomatic DF was further defined as either DF or dengue hemorrhagic fever (DHF) using World Health Organization criteria as previously described [7].
Inapparent (subclinical) dengue virus infection was defined as a four-fold rise in HAI antibody against any dengue virus serotype between two sequential sera obtained during the surveillance months (June, August or November) without a febrile illness identified during active surveillance during the time period that seroconversion occurred. For example in sera obtained from a volunteer in June and August where an HAI antibody rise of four-fold was detected to dengue virus with no period of school-absence with fever in the same period was classified as an inapparent dengue infection. Sera were tested concurrently for JEV specific HAI antibody to exclude JEV infection and antibody cross-reactivity as a cause for a fold-four rise in dengue antibody. HAI assays were performed per referenced assays according to standard protocols and interpreted using World Health Organization recognized standards of HAI antibody in acute dengue virus infection [19].
Statistical analyses were performed using SPSS for Windows version 12.0 (SPSS Inc.) and SAS analytic software, version 9.1 (Cary, NC: SAS Institute, Inc.). Only symptomatic and inapparent dengue infections that were detected between June and November each year were included in the analysis. This was done to avoid misclassification of mild, non-hospitalized cases as inapparent infections and altering the SI ratio as an artifact of surveillance. To eliminate bias and due to the small number of primary dengue infections noted in this population, only secondary dengue infections were included for analysis.
Incidence rates were determined using the total study population at the time of surveillance as the denominator. Pearson's chi-square and Spearman correlational analysis were used to identify factors that were significantly associated with the proportional occurrence of symptomatic infection by school and by year. These proportions were then translated into SI ratios using the conversion that the ratio of symptomatic to inapparent (S:I) infections is equal to the probability of being symptomatic over the probability of being inapparent for a given subgroup, i.e.
RatioSI = (where PS = the proportion of infections that were symptomatic). 95% confidence intervals were calculated for the upper and lower limit of the probability of symptomatic infection as and then translated to ratios as above. Where the lower confidence interval for the probability of symptomatic infection was negative or the upper limit was greater than or equal to 1, these limits were reset to be 0 and 0.9999 to restrict the range to legitimate values for a ratio.
Logistic regression models were constructed to evaluate the probability a child's infection was inapparent, given school-level characteristics of past (the year prior to a child's infection) and present (the year of the child's infection) dengue epidemics at that school. School-level characteristics that were considered included the incidence of dengue infection, the proportion of infections that were inapparent, the proportion of polymerase chain reaction (PCR)-positive infections that were DENV-1 – DENV-4, and the number of serotypes in circulation (as detected by PCR) at that school. The incidence, proportion inapparent, and the proportions of PCR-positive infections by serotype were categorized into quartiles with the upper limit based upon the observed range for each variable. Statistical analysis accounted for both clustering of children by school and the longitudinal observation of children who may have experienced multiple infections, using SAS PROC GLIMMIX procedure and two levels of random effects. Multiple individual models were constructed, controlling for each of the exposure variables singly and incorporating the random effects.
The study population characteristics and incidence of acute dengue infection was previously reported [7], [12]. In January 1998, 2,214 students were initially enrolled in the study with 2,044 remaining for the start of the surveillance. The overall flow of study enrollment and case identification is demonstrated in the flow diagram, figure 1. There was a gradual decline of the study population at the start of active surveillance: 2,044 in 1998; 1,915 in 1999; 2,203 in 2000; 2,011 in 2001; and 1,759 in 2002. The gradual decline of the study population over time is a reflection of the changing demographics of the surveillance schools with smaller 1st grade classes. The mean dropout rate over the study period was 5% primarily due to movement of families out of the surveillance schools. No differences in sex distribution were noted from year to year and between schools (data not shown).
1,024 dengue virus infections were detected in total over the 5 years of the study; 909 children experienced at least one dengue infection during the study period and 115 experienced a second infection. Restricting analyses to the 615 infections detected during the active surveillance period (June 1– November 1), 66% (406) were inapparent and 34% (209) were symptomatic. The median duration of enrollment in the cohort study was two years (25th percentile: 1 year and 75th percentile: 4 years). 3331 children were enrolled for at least one full year. 27.3% of children experienced at least one dengue infection during their period of enrollment. The vast majority of acute infections were secondary by EIA (98%); only 4 infections detected during the active surveillance period were primary.
The one-year incidence of total dengue infection and its constituents, symptomatic and inapparent infection was: for 1998 16.2% (of which 68.1% of those infection detected during the active surveillance period were inapparent); for 1999 13.6% (66.7% inapparent); for 2000 5.2% (81.3% inapparent) for 2001 17.5 (58.0% inapparent); and for 2002 7.1% (66.2% inapparent). There were no fatal cases of dengue. DHF represented 14.8% of symptomatic infections, with the proportion varying yearly from a minimum of 9.6% in 1998 to a maximum of18.2% in 2000. There were no significant differences in the S:I ratio between males and females (p = 0.372 by Pearson chi-square).
Figure 2 demonstrates the annual variation of the S:I ratio for the study population. The S:I ratio in this figure and all figures is transformed into a single, scaled integer which represents the number of symptomatic infections per 1 case of inapparent infection. A value of 1 denotes equal numbers of symptomatic and inapparent cases, an integer greater than 1 indicative of increasing numbers of symptomatic infections, and an integer less than 1 indicative of increasing numbers of inapparent infections. As shown, there is a significant annual variation of this ratio (p<.0001 by Pearson chi-square) with a relatively severe year noted in 2001.
As demonstrated in figure 3, there was no correlation between age and the S:I ratio (p>.30 by Pearson chi square, after categorizing into three-year intervals). Further, there was no difference in the proportion of dengue infections that were hospitalized by age (not shown). These findings are contrary to the current belief that more severe dengue is seen among older children and young adults. There were however, four cases detected in the 15 and 16 years age group over the 5 years of the study: all were symptomatic dengue infections.
The spatial diversity of the S:I ratio was examined by schools and by year. Children attending a given school tended to come from the same village, therefore stratifying on school can provide insight into smaller-scale spatial trends in disease and transmission. The total distance across the study site was 30 miles and therefore schools (and neighborhoods) tended to be separated by several miles.
As demonstrated in figure 4, there was considerable diversity in the S:I ratio amongst schools. In 1998, school 7 had much more severe disease with an S:I ratio of 5.0 as compared to other schools during that year. In 1999, schools 6, 7 and 9 were more severe than other schools; in 2000 school 3, 6, and 9 were more severe; in 2001 schools 2, 10 and 12 were more severe and in 2002, schools 4 and 10.
Temporal diversity of the S:I ratio was also examined for a given school over time as demonstrated in figure 5. In general each school had a unique experience with regard to changes in the S:I ratio over time. For example, school 8 had relatively mild epidemics every year from 1998–2002, but small-scale annual oscillations in severity were observed. This was different than school 10, which had relatively mild epidemic years from 1998 to 2000, followed by relatively severe years in 2001 and 2002. School 7 had a clinically-severe epidemic in 1998, a less a severe epidemic in 1999, and a mild epidemic in 2001. Though each school had unique characteristics regarding timing and the period of the changing S:I ratios in their population, it is striking that in general, there were cyclical shifts in the S:I ratio with 1 or 2 years of relative inapparent dengue seasons were followed by more severe dengue years.
All four dengue virus serotypes were detected over the study period, though the number of serotypes in circulation varied each year. In 2000, for example, only DENV-2 was detected in PCR+ symptomatic individuals while in 2002, all four DENV serotypes were detected. One predominant dengue serotype was usually the cause of an outbreak in any given year for a specific school as previously described [12].
A higher incidence of infection at a given school, for a given year, was associated with a lower proportion of inapparent infections (i.e., a higher S:I ratio) during that epidemic season (OR = 0.62, 95% CI = 0.53–0.74 for a 25% increase in incidence) (Table 1). Similarly, having a higher number of serotypes in circulation for a given year was associated with a decreased likelihood of inapparent infections (OR = 0.78, 95% CI: 0.62–0.99). Serotype-specific effects were observed, with a higher proportion of DENV-3 circulating at a given school being associated with a lower likelihood of inapparent infections (OR = 0.82, 95% CI: 0.68–0.99) and a higher proportion of DENV-2 being weakly associated with a higher likelihood of inapparent infections (OR = 1.12, 95% CI: (0.98–1.28).
When considering the influence of a previous year's epidemic characteristics on the subsequent epidemic's S:I ratio at a given school, the prior incidence of infection had a strongly positive association with the subsequent S:I ratio. That is, the higher the incidence at a given school the year previous, the greater the probability a child experienced an inapparent infection the following year (OR = 1.34 for a 25% increase in incidence, p<0.01). A higher proportion of symptomatic infections at a given school for the previous epidemic season was weakly associated with an increased likelihood of inapparent infection the following year (OR = 0.85, 95% CI: 0.70–1.03).
In 1998, 2001, and 2002, there was a significant overall increase in the proportion of inapparent dengue infections during the second half of the active surveillance period for the entire study population (Aug 16-Nov 1) compared to the first half (Jun 1-Aug 15) (p<0.01 for all, figure 6). For 1999 and 2000, the proportion inapparent remained relatively constant between surveillance periods.
Understanding the full burden of dengue infection and in particular the burden of inapparent dengue infection has important public health implications in understanding virus transmission during an outbreak, the degree of under-reporting during an outbreak year and our understanding on the pathogenesis of dengue illness. Our study is unique from other cohort studies on dengue infection for (1) its long-term follow-up of a well-defined school cohort population; (2) its ability to distinguish the entire spectrum of dengue virus infections and (3) determining spatial and temporal patterns associated with dengue virus transmission. In this report we examined the symptomatic to inapparent ratio as an endpoint and as a correlate for disease severity in a population spatially and over time.
Our results demonstrate that the S:I ratio, unlike in previous reported studies, is not fixed and has wide variation in a population in one specific geographic area and over time. Our data demonstrates that dengue disease severity taken as a whole in a population is an aggregate of multiple experiences of disease severity in sub-populations, with some sub-populations experiencing more severe illness and others less severe dengue infections. In particular, we demonstrated that the previous year's incidence of dengue virus infection influences the subsequent year's dengue disease severity with higher incidence years followed by more clinically-inapparent years. A similar trend has been observed at the national level in Thailand- a cyclical nature in dengue outbreaks and DHF with more severe years followed by milder years [20].
Disease severity observed in our study could be related to a number of factors such as the dengue serotype circulating in the population, viral genetic factors associated with severe disease, and the host's preexisting immunity from a prior dengue virus infection to another serotype leading to antibody enhancement and cross-reactive memory T-cells or conversely cross-protective heterotypic antibody. This would be consistent with previous observations on the pathogenesis of severe dengue illness. Our findings did not demonstrate an association between age and the S:I ratio in our population. This might be due to the narrow age range of our study population, ages 7 to 16 years, but more likely due to the competing nature of other more important factors that influence disease severity, such as the previous year's severity. Our study did not show significance for the proportion of a specific dengue virus serotype and the number of circulating serotypes occurring during an epidemic year or the previous year on the S:I ratio though important factors as demonstrated in previous studies. This may be due to the overwhelming influence of dengue infection in general on disease severity rather than specific dengue serotypes.
Our finding that the greater the incidence of dengue infection of the previous year's epidemic, the milder the subsequent year's disease severity, raises the interesting possibility that herd immunity is an important contributor to disease severity. This could occur either by changing the subsequent year's predominant circulating dengue serotype(s) to a new circulating dengue virus serotype because of protective antibody or by changing the host immune response to experience milder illness. We have previously demonstrated that cross-reactive heterotypic dengue neutralizing antibody does not induce sterile immunity but may influence disease severity on an individual level, at least for some viral serotypes [21]. We believe that what we are observing in the fluctuating nature of the S:I ratio in our population is the influence of this heterotypic cross-reactive antibody at a population level. Based on these findings we postulate that, as a dengue outbreak with a high incidence and disease severity occurs in a given year in a population, there is continued heterotypic herd antibody in a population the following year that doesn't provide protective immunity from infection, but enough cross-protective activity to lower disease severity. This could occur either at the individual or population level. Thus the prior year's S:I ratio and incidence has an inverse relationship on the subsequent year's S:I ratio.
Dengue is a global health concern with no effective vaccine to prevent infection. Understanding the nature of inapparent dengue infection has important public health implications in understanding virus transmission, and determining disease burden during an outbreak. Our findings increase the understanding of this disease and the full burden of infection and point to a role of cross-reactive heterotypic antibody in influencing disease severity the year after an outbreak year in a population. Understanding the factors associated with the development of inapparent dengue infection will increase our understanding on the pathogenesis of this disease and have important implications on how a tetravalent dengue vaccine might influence disease severity in a population.
This project and publication was made possible by NIH Grant P01 AI34533 and the United States Army Medical Research and Materiel Command, Ft Detrick MD. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Institutes of Health or the Department of Defense. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
|
10.1371/journal.pcbi.1002564 | Coding Conspecific Identity and Motion in the Electric Sense | Interactions among animals can result in complex sensory signals containing a variety of socially relevant information, including the number, identity, and relative motion of conspecifics. How the spatiotemporal properties of such evolving naturalistic signals are encoded is a key question in sensory neuroscience. Here, we present results from experiments and modeling that address this issue in the context of the electric sense, which combines the spatial aspects of vision and touch, with the temporal aspects of audition. Wave-type electric fish, such as the brown ghost knifefish, Apteronotus leptorhynchus, used in this study, are uniquely identified by the frequency of their electric organ discharge (EOD). Multiple beat frequencies arise from the superposition of the EODs of each fish. We record the natural electrical signals near the skin of a “receiving” fish that are produced by stationary and freely swimming conspecifics. Using spectral analysis, we find that the primary beats, and the secondary beats between them (“beats of beats”), can be greatly influenced by fish swimming; the resulting motion produces low-frequency envelopes that broaden all the beat peaks and reshape the “noise floor”. We assess the consequences of this motion on sensory coding using a model electroreceptor. We show that the primary and secondary beats are encoded in the afferent spike train, but that motion acts to degrade this encoding. We also simulate the response of a realistic population of receptors, and find that it can encode the motion envelope well, primarily due to the receptors with lower firing rates. We discuss the implications of our results for the identification of conspecifics through specific beat frequencies and its possible hindrance by active swimming.
| Effectively processing information from a sensory scene is essential for animal survival. Motion in a sensory scene complicates this task by dynamically modifying signal properties. To address this general issue, we focus on weakly electric fish. Each fish produces a weak electrical carrier signal with a characteristic frequency. Electroreceptors on its skin encode the modulations of this carrier caused by nearby objects and other animals, enabling this fish to thrive in its nocturnal environment. Little is known about how swimming movements influence natural electrosensory scenes, specifically in the context of detection and identification of, and communication with conspecifics. Using recordings involving free-swimming fish, we characterize the amplitude modulations of the carrier signal arising from small groups of fish. The differences between individual frequencies (beats) are prominent features of these signals, with the number of beats reflecting the number of neighbours. We also find that the distance and motion of a free-swimming fish are represented in a slow modulation of the beat at the receiving fish. Modeling shows that these stimulus features can be effectively encoded in the activity of the electroreceptors, but that encoding quality of some features can be degraded by motion, suggesting that active swimming could hinder conspecific identification.
| Sensory systems must effectively extract relevant information from an animal's environment. Their ability to encode natural scenes and tease out salient sensory features relies on a range of neural mechanisms, e.g. [1], [2]. In social contexts, individuals generate signals with characteristic temporal and spatial frequencies, and time-varying amplitudes [3]. From these signals, an individual can reconstruct the sensory “social” scene [4] by sorting out the identities, locations and behaviours of its neighbors [5].
Narrowband signals with slow amplitude modulations, known as envelopes, are a nonlinear signal feature of particular importance for scene analysis in the auditory system [6]–[8], human speech recognition [9], [10], and coding of textures in visual cortex [11]. Envelopes have also been studied in the electric sense [12]–[15]. Weakly electric fish have a submicrosecond-precision neural pacemaker, under behavioural control [16], that produces a weak quasi-sinusoidal dipolar electric organ discharge (EOD). Each animal has its own EOD frequency (EODf) [17]. For example, the species studied here, Apteronotus leptorhynchus, has EODfs in the 700–1100 Hz range, with males generally having higher EODfs than females (see Figure 1A for example EOD recordings). These fish sense prey, navigation cues and other animals including conspecifics by encoding amplitude modulations (AMs) of the EOD carrier with the quasi-linear modulation of the mean firing rates of cutaneous electroreceptor afferents [18]–[20].
Two fish in close proximity sense the sum of their electric fields as a time-varying beating AM [17]; the beat frequency is a basic component of electrocommunication [17] (see Figure 1B for example compound EOD with beating AM). In groups of fish, multiple beat frequencies result in “beats of beats” and slow envelopes with narrowband AMs [14]. The spatial aspects of these EOD interactions are less well-understood, though for stati'c fish, the complex electric images of conspecifics have been recently predicted under some conditions [21]. Sinusoidal and narrowband random AMs (SAMs and RAMs, respectively) are typically generated through electrodes to mimic social interactions under experimental conditions (see Figure 1C), and have subsequently led to much insight into the underlying electrosensory processing (e.g. [12], [13], [15], [17]). However, little is known about how movement, resulting in relative changes in distance and orientation, influences the processing of complex AMs in carrier-based senses, such as auditory and electrosensory systems. Signals with SAMs and narrowband RAMs do not have explicit low-frequency power associated with motion. The more natural scenario involves EOD AMs resulting from the motion of a small number of conspecifics [22], which spectrally contain a small discrete set of narrow peaks. A thorough characterization of these natural dynamic signals is necessary to better understand the neural mechanisms required for effective electrosensory processing.
In this study, we first describe the naturalistic AMs and slow envelopes resulting from the relative motion of interacting fish. We contrast the properties of the EOD modulations for static and swimming fish, providing a mathematical model for the associated motion in terms of band-limited random AMs. We then determine the consequences of motion on the neural encoding of number, identity, and movement characteristics of socially interacting conspecifics, by computing what information about the sensory scene is represented in electroreceptor spike trains.
Since groups of Apteronotus in the wild rarely contain more than a few individuals [14], we recorded the signals during interactions of pairs ( = 8) and triplets ( = 4) of fish. The EODfs ranged from to Hz; the beat frequencies (), equal to the differences between the EODfs of the interacting fish, ranged from to Hz. In each experiment, only one of the fish, named the “receiving fish” and denoted “fish 1”, was restrained. We denote the neighbouring fish as “fish 2” and “fish 3”. The compound EOD signal due to all fish (including fish 1) was recorded through two electrodes locally on one side of fish 1, very close to its head, approximating the signal received by fish 1 near its receptive surface. The other one or two fish swam around freely in the same tank (see Figure 2A and Materials and Methods). The amplitude modulation (AM) resulting from the proximity of neighbouring fish is referred to hereafter as the first envelope, E1. The slow envelope of E1 (modulation of the AM) is referred to as the second envelope, E2.
Each individual fish senses its own EOD highly reliably. When a pair of fish are in a “static” state where both are stationary, fish 1 receives a constant stimulus from fish 2 in addition to its own EOD, resulting in a stable periodic E1 at the beat frequency (see Figure 2B). In these conditions, E2, as the envelope of E1, is nearly a constant. In contrast, when fish 2 is allowed to swim freely, both E1 and E2 at fish 1 vary in time (Figure 2C,D). This is a consequence of Coulomb's law, albeit in a complex geometry: shorter distances between the two fish, each of which acts as an oscillating electric dipole, lead to stronger electric current flow caused by the neighbouring fish; larger distances result in a weaker or even undetectable input from the neighbour. Thus, the mean of E2 reflects the average distance between two fish, while the variance of E2 is associated with the pattern of swimming of fish 2 including its bending and turning.
The three-dimensional spectrogram in Figure 2E allows a visualization of the time-varying amplitudes and frequencies of each fish's EOD as experienced by fish 1 for the same segment of experimental data used in Figure 2D. The amplitude of the spectral density (ASD) of fish 1 (around Hz) is very stable in time, but the ASD of fish 2 (around Hz) varies in time as it swims around in the tank, in a manner very similar to the E2 (c.f. Figure 2D). It is worth noting that the occurrence of chirps can also be indicated in E2. A chirp is a communication signal commonly produced during social interactions, and is characterized by a 20 msec modulation of the EOD frequency [23]. For example, the fast dip of E2 (“” in Figure 2D) indicates a chirp, which is also seen as a cleft in the ASD (“” in Figure 2E) at a time of 65.5 sec. Apart from these brief EODf shifts during a chirp, the EODf did not change over the course of our recordings, and can be assumed constant.
A. leptorhynchus generates a quasi-sinusoidal EOD, so the superposition of EODs of multiple fish can be well-approximated by a sum of sinusoidal waves at the EOD fundamental frequencies. Since the ASD of the stationary fish 1 is highly reliable, the amplitude of its EOD can be taken as one, without loss of generality; on the other hand, the time-varying ASD from the free-swimming fish 2 is better represented by a stochastic process. Given constant EODfs, the composite EOD signal can be modeled as(1)where is the group size, is the EODf of the n-th fish ( for the stationary fish), and is the stochastic amplitude for the n-th free-swimming fish, with a mean of and standard deviation (STD) . is a stochastic variable with zero mean and unit variance, mimicking the amplitude variations due to movement of the free-swimming fish, and is modeled here for simplicity as an Ornstein-Uhlenbeck process (OUP, or lowpass-filtered Gaussian white noise); the spectral power of an OUP is concentrated in the low-frequency range like the movement itself (see Materials and Methods). The phase difference does not affect the spectral components of E1 and E2 and is set to zero; phase may however play an important role in other computations, for example those involved in the jamming avoidance response (JAR) [17].
The OUP is characterized by an exponential autocorrelation function, with a decay time constant that defines its correlation time, (see Materials and Methods). To estimate this correlation time, we compared the autocorrelations of E2 obtained from the experimental trials and the artificial signal above, for the case of two fish. All autocorrelations of E2 extracted from the natural electric signals recorded separately from pairs of fish exhibit a decaying behaviour (coloured curves in Figure 3A). These experimental curves can be fit very well by E2 of when (dotted curve in Figure 3A). This agreement also confirms that the OUP is an appropriate stochastic model for . Other parameters , and have negligible influence on the autocorrelation as expected from the properties of the OUP and verified by numerical simulation (not shown).
The relative motion of the fish results in a time-varying contrast. The mean and STD of “instantaneous”contrast obtained from the raw data (see Materials and Methods for detailed definition) are used to estimate and , respectively, of the simulation signal, , in the case of two fish. Both numerical simulation and theoretical analysis for (see Equation (6) in Materials and Methods) show that the mean and STD of the contrast of are approximately equal to and , respectively. This is illustrated in Figure 3B which compares the mean contrast from both theory (filled circles) and simulation (solid black line), along with the STD of the contrast (theory: open circles; simulation: dashed black line). Also shown in Figure 3B are the mean contrasts (STD) calculated from the recordings of different fish pairs; these values are used as estimates of the corresponding (and ) in . For the five-minute recording with the highest mean contrast of and relatively high STD of (the fish pair with red data point in Figure 3B), fish 2 was fairly aggressive, making fast approaches sometimes resulting in physical contact. In contrast, for the pair with the lowest mean contrast of and lowest STD of (the fish pair with dark green data point in Figure 3B), fish 2 stayed mainly in the corner of the tank and rarely moved. A similar situation occurred for the pair marked by the purple data point (), but fish 2 in this trial was slightly more active. This behaviour is also reflected in the autocorrelation times, which are relatively long in these two trials (Figure 3A, dark green and purple curves). The other five pairs of fish exhibited intermediate levels of swimming and approach behaviours (Figure 3B), with mean contrast ranging from to and STD varying from to . Thus, for the model, we chose the parameter range for as 0.07 to 0.20, and for as 0.5 to 0.9.
We can now construct an artificial signal to simulate the signal arising from a real interaction. For instance, Figure 3C shows a realization of that mimics (statistically) the interaction indicated by the blue fish pair in Figure 3B ( = 0.143 and = 0.08), along with its E1 and E2. We also checked the similarity between the calculated E2 and the stochastic amplitude (Figure 3D). The sum of amplitudes and E2 exhibit very good agreement, which confirms again that motion of the fish produces the second envelope E2. The same parameter values will be used in our simulation work below for two fish, unless otherwise stated.
To quantitatively compare the power spectral densities (PSD) of the simulated signal, , with the raw recordings, was rescaled to make its total energy equal to that of the experimental data. We consider 12 seconds for two different experimental trials, one from a fish pair and another from a three-fish group (EODfs = [827,763]Hz and [831,740,889]Hz, respectively). The PSDs for the raw signal, as well as for E1 and E2, are plotted side-by-side in Figure 4 (green curves), along with the PSDs for the corresponding simulated signals (swimming state with 0: black solid curves and static state with = 0: black dashed curves). The stationary fish 1's EOD is significantly stronger than that of any neighbour, and produces the largest peak in Figure 4A. Thus, the dominant frequencies of E1 are located at the beat frequencies , , while other beats at , contribute less. Similarly, differences between any two 's, referred hereafter as secondary beats, are the prominent spectral components of E2 in groups of three fish (or more). This is evident in the spectral peak of E2 at 33 Hz (– = 91-58) in Figure 4C (right). This clearly describes quantitatively the common notion that electric field modulations at the skin contain spectral information about the number of neighbouring fish as well as their identities (EODf).
Further, at low frequencies (0–20 Hz), the comparison between swimming fish and “static” fish indicates that E1 contains power related to motion (Figure 4B). This power is two or more orders of magnitude less than the power at the beat frequencies. Nevertheless, this contrasts with the narrowband RAMs used in experimental studies [12], [15] to mimic the E1 resulting from the interaction of many static conspecifics; these RAMs have no power at these low frequencies. The E1 motion power (Figure 4B) can also be larger than that of the secondary beats in E1.
Interestingly, the power associated with motion is clearly highlighted by E2 (Figure 4C). This motion produces a decaying spectral floor mainly in the range 0–20 Hz, but extending out beyond 100 Hz over five orders of magnitude or so. The peaks associated with the secondary beats ride on top of this floor, with very low power for the chosen EOD parameters. Neural circuitry specialized in extracting information from slow E2 envelopes [7], [11], [12] could do so using the lower frequency structure; in this context, the E2 floor would be considered a signal. Alternatively, this E2 power could obscure other potentially significant envelope signals in the same range, and this motion-induced spectral floor would act as a noise floor. Finally, Figure 4 shows that there is very good agreement, both qualitative and quantitative, between the spectral features of the experimental and simulated signals, providing further support for our model.
Electric fish recognize the EODf of conspecifics through the beat frequencies [17]. Therefore, higher spectral peaks or narrower peak width (PSD of E1) at beat frequencies should improve the ability of fish 1 to encode the beat frequencies. To investigate how inter-fish distance and motion can influence this encoding, we define the spectral resolution of beat frequencies in E1 as the ratio between the height and width of the corresponding spectral peaks (see Figure 4B). and , which relate respectively to the inverse of the distance between two fish and the movement variation of neighboring fish, are important factors for quantifying E1. A group of two fish, the simplest and most common group for A. leptorhynchus, with the same EODfs as in Figure 4 was taken as an example. We computed the average height and width of the PSD peaks centered at beat frequencies for different combinations of and and plotted the simulation results in Figure 5. The peak width is measured at a power of (slightly above the “noise floor”), because the increment of spectral peak width is more sensitive to at this value. Figure 5A clearly demonstrates that when is fixed at 0.1, a larger results in an increased peak height (solid line), while the peak width (dotted line) barely changes. Therefore, shorter distances between two fish increase the resolution of the beat frequency (see Figure 5C). Figure 5B shows that, for fixed and increasing , both width and height increase [24], but at different rates. The result is that the ratio of height to width decreases with increasing (by about over the range tested, see Figure 5D), indicating that a larger swim variance of fish 2 reduces the spectral resolution of the beat frequency. A comparison of the rates of changes in Figure 5C and 5D indicates that the resolution of the beat frequency is more sensitive to than .
The influences of and on E1 can also be observed in the time domain, via the behavior of successive periods of the E1 waveform. For the current example, the probability density function (PDF) of the E1 period shows a peak at 15.6 msec (i.e. beat period of = 1/64 Hz) when = 0.2 and (Figure 6A; red line); a larger introduces more jitter around this beat period. On the other hand, a larger reduces fluctuations of the beat period (Figure 6B). We quantify these effects using the coefficient of variation (CV), defined as STD divided by the mean of the periods of the E1 waveform (Figure 6C). Over the parameter range shown, increasing leads to a larger CV, whereas increasing decreases CV. Interestingly, combinations of and corresponding to the experimental trials (plotted as filled circles in Figure 6C, colored as in Figs. 2 and 3) show a systematic relationship, suggesting that the fish do not vary distance and motion independently under the conditions tested (whether or not this is a general feature of social interactions will be determined in future studies). In summary, both types of analysis suggest that increasing inter-fish distance (decreased ) and increased motion (increased ) lead to a degradation in the quality of the E1 signal with respect to the beat frequencies. In the next section, we assess the impact of and at the level of sensory encoding by considering the responses of model electroreceptors (P-units) to these same signals.
P-unit electroreceptors are the first processing site in the electrosensory pathway, encoding information contained in the transdermal voltage fluctuations. Using artificial SAM and RAM-type signals, P-units have been shown to encode the time-varying raw electrical signal into instantaneous changes in their stochastic firing rate [25], [26]. These changes track (almost linearly) the AM represented by E1 in those studies (except at higher stimulus contrasts where nonlinear effects are involved [13], [15], [27]). The leaky integrate-and-fire model with dynamical threshold (LIFDT, see Material and Methods) has been shown to capture most essential features of the spiking dynamics of P-unit afferents [13], [28]. Therefore, to provide insight into electrosensory coding during natural interactions, we describe the response of this P-unit model to the composite EOD signals described in the previous section.
Figure 7A shows three example spike trains from the model P-units with different P-values (see Material and Methods) in response to the recording from Figure 2C (with E2, spectrogram and PSD shown in Figure 2D, E and Figure 4A, respectively). These spike trains clearly show that the instantaneous firing rate increases with increasing E1. To investigate the envelope-output transfer function of a P-unit, we use the simplest signal ( is constant) as the input instead of in Equation (1). According to Equation (6), the motion of fish 2 can be seen as fluctuations in the envelope mainly in the range of . Previous studies have shown that P-units can exhibit firing rate saturation with time-varying E1 [15], [29]. Here Figure 7B demonstrates that, within the range of interest, the output firing rate is basically proportional to ; P-units with larger P-values simply encode the EOD fluctuations into modulations of a higher baseline firing rate. Figure 7C demonstrates the E2-output transfer function, where the spike counts within 0.1 second increase with increasing E2 in the same time window. This suggests that the motion of neighboring fish varies the firing rate of P-units of fish 1.
This can be clearly demonstrated by looking at the time-varying firing rate calculated from a heterogeneous population of P-units in Figure 7D. Each electroreceptor has its characteristic P-value, and across receptors, the P-values form a log-normal distribution with a mean value at 0.26 (see [29] and Figure 7E). We calculated the mean firing rate with a time window longer than (e.g. 0.1 second) using models of 200 P-units with such distributed P-values. We also computed the time-varying firing rate of a single P-unit with a P-value equal to 0.23 for comparison. By comparing E2 with the firing rate curves obtained from the P-unit population and single P-unit in Figure 7D, we can conclude that the population encodes the motion of the neighboring fish better than individual P-units with average to large P-values. Note however that the single unit already encodes it quite well on its own, at least over this frequency range. Also note that the raw input signals here do not have direct power at the beat frequencies, and so the extraction of the beat frequencies must involve a nonlinear operation. This nonlinearity is implemented in our analysis by the Hilbert transform (HT, see Materials and Methods), allowing us to obtain the E1 of the raw signal (Figure 4B). However, implementation by the P-unit model involves the spike threshold (and possibly other) nonlinearities [12], [13], [15]. The P-unit plays a role similar to that of the HT to extract E1 and eliminate the EODfs. The power spectrum of P-unit spike train in Figure 8A indicates peaks at beat frequencies () which are not presented in the PSD of input signal but are in the PSD of E1. Further, the peaks at in E1 are strongly correlated to the spike train (see the cross spectral density between E1 and the P-unit response in Figure 8B). However, reveals that these peaks are much less correlated to E2 (Figure 8B).
The coherence was then used to estimate the linearity of the encoding of E1 and E2 by the output of the P-unit model (see Materials and Methods). For P-units with a P-value of 0.26, the E1-R coherence, , has a peak at (Figure 8C), suggesting that P-units can efficiently encode this beat frequency. However, the very low E2-R coherence, , implies that most individual P-units do not linearly represent information about slow envelopes associated with natural motion (Figure 8C), except perhaps at the very low frequencies where a slight rise is seen. However, the coding of motion-related information can be improved for P-units with low P-values (e.g. 0.12 marked by green dashed curve in Figure 8C). Further, as numerous P-units participate in the processing of sensory information, a population code could relay motion-related information embedded in E2 to downstream electrosensory neurons (see insets of Figure 8C). For the case of three fish (Figure 8, right panels), the same features hold qualitatively (even the low frequency motion - not shown). In addition, the slower secondary beat frequency is clearly revealed by the response function (Figure 8B), as was seen for the raw signal in Figure 4C. now has a large peak at both main beat frequencies, and a very small peak at the secondary beat frequency. again emphasizes the slower secondary beat.
We now describe the influence of and on E1-R coherence at the beat frequency for two interacting fish. Similar to our evaluation of the PSD peaks, we measure the height and width of the coherence peaks to quantify coding quality. Figure 9A shows that the maximum height of the peak at the beat frequency increases with , and the width (measured at a coherence of 0.15, slightly above “noise floor”) slowly decreases with . This leads to an increasing height-to-width ratio with A2, and thus could improve the accuracy of beat frequency estimation in the hindbrain (Figure 9C). Since varies inversely with inter-fish distance, this confirms that at the receptor level, shorter distances between two fish enhance their ability to detect each other via the beat. On the contrary, increasing (akin to increasing the strength of swimming variation) enlarges the peak width and decreases the height (Figure 9B); consequently the height-to-width ratio drops with , i.e. when swimming is more erratic or less confined (Figure 9D). This implies that rapid changes in distance between two fish could blur the sensing of the other through the beat frequency. The same conclusion can be obtained using the width measurements at half max (not shown). With multiple fish, this blurring would have even more impact if beat frequencies were close. Thus motion, through degradation of the P-unit encoding of the beat frequencies, could be actively used as a form of crypsis, decreasing identification by conspecifics.
Considering that most of the motion power is concentrated over the frequency range of 0–20 Hz in E2 (also see [30], [31]), the mean peak coherence of over 0–20 Hz was plotted to examine the information encoded from E2. In Figure 10A and B, the peak height and width of increase with both and over 0–20 Hz. Similarly, the mutual information rate over 0–20 Hz also increases with both parameters (Figure 10C,D). These results show that electroreceptors encode motion of conspecifics increasingly well for smaller inter-fish distances and increased relative movement. Thus, it appears that when a fish increases movements towards a conspecific, there may be a trade-off between improved encoding of motion (as a signal or a noise, as mentioned above) and degraded identification, which are coded by E2 and E1, respectively.
Our study describes the naturalistic signals generated by relative motion among small groups of weakly electric fish. The analysis of the raw signals and the simulated responses of primary electroreceptor afferents show that these signals contain important cues for the identification of individuals and their behaviour. This information is available from the spectral properties of the first and second envelopes (E1 and E2) of the composite electrical signal, which relate to the beat frequencies (E1) and the secondary beats and relative motion patterns (E2).
The phenomenological model for motion fitted our data very well, and its parameters and are directly related to the contrast mean and contrast STD of the experimental recordings (Figure 3B). Further, our experiments revealed a proportional relationship between the STD and mean of the contrast (and thus between and ; see Figure 6). However, we can not infer that this relationship is universal across all experimental and social contexts. The possible context-dependence and behavioural significance of the relationship will be explored in future studies. In addition, the relationship between model parameters and and behavioural measures is not entirely clear. While the mean contrast is inversely related to the mean distance separating the fish, it is also influenced by the complex interactions between fish bodies [21]. We also note that is related to motion (variations in swimming). This relationship is complex and is influenced not only by changes in inter-fish distance, but also by turning and bending. A thorough characterization of the physical bases of and is beyond the scope of this study and will be pursued in subsequent work.
Our experimental and modeling work shows that movements of neighbouring fish generate power in the first envelope E1 that is small relative to the power in the beat (AM of the sum of EODs; Figure 4B). But the movements produce relatively more power in the second envelope E2 (envelope of the AM; Figure 4C), especially below 10 Hz. Our model reveals that the peak resolution for the beats in E1 increases slightly over a range of amplitudes , i.e. of contrasts (Figure 9C), but decreases strongly with motion stochasticity (Figure 9D). It also reveals that the encoding of E2, while no longer representing beats, is proportional to the mean amplitude of the neighbouring EOD (Figure 10A,C), which is inversely proportional to inter-fish distance. The encoding of E2 is also proportional to the variance of the motion (Figure 10B,D). E2 also highlights the secondary beats between the primary beats. As the electrosensory system can extract the envelope post-synaptically to the electroreceptors [12], such envelope information about motion and secondary beats can be readily relayed to midbrain electrosensory regions. Our results imply that this structure has access to both the beats, the secondary beats and motion information, which can in principle feed the directional selectivity circuitry [32].
Further, our analysis suggests that P-units effectively encode beat frequencies, but single P-units with normal-to-large P-values can not encode motion information well over a range of normal contrasts (Figure 8C). This is consistent with previous modeling [13] and experimental work [15] where E2 obtained from narrowband RAMs could be represented by P-unit activity only for large contrasts, and otherwise the transmission of E2 to higher-order cells relies on a parallel pathway via interneurons [12]. Another experimental study reported a tracking between mean firing rate of P-units and a low-frequency 0–4 Hz RAM [29]; taken together with our observation that P-unit mean firing rate varies with E2 (Figure 7), this suggests that a population code might instead be involved in encoding E2.
These fish can transform spatial information about the motion of other fish into a temporal signal with a second envelope. The amplitudes of the EODs reflect the distances of fish 1 to its neighbours, and are clearly reflected in the height of the beat peaks in E1, as well as the mean of E2. Therefore E2 may play an important role in electrolocating conspecifics. It remains to be seen whether E2 improves stimulus localization, as can occur for static auditory sources [6].
The identity of conspecifics, given by their individual EODfs, is well represented by beat peaks in E1, especially at short distances (large ). However, for dynamic swimming (larger ), these peaks broaden, and the sensory system may no longer be able to differentiate different beats that are close in frequency. Animals use various forms of camouflage and other behaviours to avoid predators. Non-visual crypsis has been reported in auditory, olfactory, and electrosensory systems in recent years [33]. Electric fish have a high risk of being detected by electroreceptive predators, and therefore may have to take extraordinary measures to protect themselves. The pulse-type fish Brachyhypopomus may use “signal cloaking” by shifting the spectrum of its EOD pulse to a less detectable high-frequency range [34]. Other species of electric fish must use other strategies to avoid detection. Figure 9 predicts that identification (via EODf) declines with increasing , suggesting that fast motion (e.g. back-and-forth swimming, as well as rapid bending, turning or spinning) could be another implementation of non-visual crypsis. The well-described behaviours requiring EODf estimation (such as the jamming-avoidance response, JAR [17]) make wave-type electric fish an attractive model in which to test this intriguing hypothesis.
Our study also points to a novel method of synthesizing more natural mimics of other fish in the laboratory. The established approach uses a SAM modulation of a restrained fish's EOD, and thus mimics a static conspecific. This actually leads to additional frequency components of the EOD that are not present naturally. The model signal presented here could be used to mimic swimming conspecifics, applied either locally, or globally using the usual configuration of two electrodes straddling the animal, or with a method that better preserves ipsilateral and contralateral contrasts and polarities [21].
The results demonstrated in this study involved one artificially restrained fish. The more natural situation is of course that in which all fish are free to swim. The question then arises as to what influence self-motion has on the results of our analysis. This means that the amplitude in our model would not be fixed but would vary with self-motion, and the P-units would encode the associated potential excursions. This is known from e.g. tail movements [35], [36]. While a full analysis of this problem must rely on actual measurements and involve field simulations, we can speculate that self-movement will likely have an impact on any identification and crypsis strategy. We note however that some body movements (such as tail bending) are known to be cancelled by plasticity at the pyramidal cell level [36]. This may mitigate self-motion signals, and emphasize beats and motion signals due to the relative motion of other conspecifics.
The proximity of the fish in our experiments resulted in chirping behaviour, and the features of E1 and E2 that triggered such communication will be explored elsewhere. This nonetheless raises the possibility that certain patterns of E1 and E2 lead to changes in the EODfs of the interacting fish over longer time courses, as they may avoid certain low frequencies that interfere with e.g. prey stimuli [14], [37]. Likewise the fish may engage in the JAR, which are predictable in static SAM-type mimics of three interacting fish in a related species [38]. The interplay of jamming, chirping and movement can be used in experiments to understand more properties of the primary afferents, which exhibit non-trivial responses to AMs and their slow envelopes. It would also be interesting to eventually relate these findings to those on hydrodynamic cues for the lateral line detection system [39]. Future work should also consider the phase variations of E1 and E2 across fish 1 as neighbours move, and on which the JAR relies. Our study highlights important information available for the analysis of such complex social sensory scenes.
All experimental protocols were approved by the University of Ottawa Animal Care Committee (BL-229).
A. leptorhynchus were obtained from a tropcial fish supplier and housed in 115 L flow-though community tanks maintained at 26–29 with a conductivity of 200–250 . Fish were kept on a 12h∶12h light∶dark cycle, and fed frozen bloodworms 3 days per week. Five fish were chosen randomly for our analyses and isolated in 20 L tanks. Recordings were performed in an experimental tank measuring in length-width-depth. During the trials, the restrained fish (fish 1) was placed in a hand-sewn tulle hammock, closed along the top with a strip of Velcro and suspended in the middle of the experimental tank. The top of the hammock was positioned about 1 cm below the water surface. Fish 1 was unable to turn or swim, and tended to remain quite still while in the hammock. Depending on the trial, one (two fish experiment) or two (three-fish experiment) other fish were added to the tank, and allowed to swim freely around the centrally positioned fish 1.
To record the natural inputs resulting from interacting conspecifics, a pair of Teflon-coated silver recording electrodes (diameter: 0.38 mm; WPI, Inc., Sarasota, FL, USA) were positioned adjacent to the head of fish 1 just anterior to the operculum. The exposed electrode tips were 1 cm apart, and oriented perpendicular to the axis of the restrained fish (Figure 2A) to measure the component of the electric field normal to the skin. A grounded Teflon-coated electrode (insulated to the tip) was attached to one corner of the test tank. The electrical signals were amplified using an AM Systems model 1700 (Carlsborg, WA, USA) differential amplifier (100 amplification, low-frequency cut-off of 10 Hz, high frequency cut-off of 5 kHz) and sampled at 100 kHz using a dSpace Inc (Wicom, MI, USA) 1011 data acquisition board and dSpace Control Desk software. During the dark stage of the daily cycle, a randomly chosen fish was isolated, and its EOD was recorded in isolation. A second, free-swimming fish was added to the experimental tank, and a five-minute recording began. After five minutes, the free-swimming fish was removed and returned to its home tank. The water heater was removed during the 5-minute interactions and replaced afterwards to maintain temperature at 26–27C. Recordings from eight pairs and four triads of interacting fish were obtained, without controlling for sex. Fish were identified based upon their anatomical differences and EODfs.
All data analyses and numerical simulations were carried out in MATLAB (The MathWorks, Inc., Natick, MA, USA).
The envelope of a real-valued oscillatory process is defined by where is in quadrature with . is commonly obtained using the Hilbert transform (HT) [40] defined by(2)with P denotes the Cauchy principal value and * denotes convolution. Thus one can create an analytic signal , and obtain an instantaneous amplitude and instantaneous phase of the raw signal. and have a clear physical meaning when the amplitude evolves on a slow time scale compared to the fast phase [40]. and extracted from an EOD recording that consists of multiple harmonic frequency components of the EOD share a remnant spectral peak at the EODf. A low pass filter (LPF) was used following the HT operation to eliminate the EODf and obtain the correct first envelope E1. Its cut-off frequency was set to 200 Hz. A HT was implemented to extract the second slower E2 envelope segments from E1 over every 0.1-second time window; these segments were assembled end to end to produce E2. E1 and E2 obtained above were also compared with the direct envelope extraction (i.e. connecting the successive peak points of EOD cycles or E1 oscillatory curves [41]). The PSD profiles and time series of E1 and E2 calculated from both ways are very similar.
The OUP, , a simple form of lowpass-filtered Gaussian white noise, is used here to model the stochastic EOD amplitude caused by a free-swimming fish at fish 1. It is the solution of(3)where = is Gaussian white noise of zero mean and autocorrelation = , and is the Dirac delta function. Brackets denote average over the Gaussian ensemble. is Brownian motion, whose increments are taken from a zero-mean Gaussian density with a variance of . The OUP has zero mean and is exponentially correlated: = 0 and = exp; it has unit variance. The correlation time is ; the larger it is, the slower the exponential decay of the autocorrelation, and the slower and smoother the fluctuations in time are.
The autocorrelation function quantifies the average linear correlation between successive points in a time series , as a function of the temporal lag between these points:(4)where is the mean of and E denotes the expected value. For the envelopes, E2, was calculated from a 5-minute uninterrupted recording for one pair of fish by dividing the record into 15-second segments. The autocorrelations over each segment were averaged and plotted as one coloured curve in Figure 3A.
An EOD carrier with frequency and a random amplitude modulation (RAM) is denoted as where is a random process. Contrast is defined as (STD of AM)(mean of AM) = 1 = , i.e. it is the coefficient of variation (CV) of the AM. An EOD carrier with sinusoidal amplitude modulation (SAM) is denoted as . Its contrast is (STD of AM)(mean AM) = M/1 = M.
According to the HT defined above or the method in [42], the AM (or E1) of our model signal in Equation (1) with = 2 can be expressed asIt can also be rewritten as
We can use the Taylor expansion with to get an approximation for the AM:(5)where higher order terms with frequencies at harmonics of are neglected, a procedure similar to LPF used in the numerical approach described in the section “Envelope extraction”. Because and (as seen in Figure 3B), is a small perturbation and can be approximated by 0. Thus we have(6)The AM of thus combines the RAM and SAM. If we regard this AM as a special version of a SAM, its contrast is approximately equal to because higher order terms are neglected. contains fluctuations introduced by , therefore we have a mean contrast and STD . For the contrast calculation of the experimental data, we extracted the AM, then collected all its highest and lowest points of AM and calculated an “instantaneous” contrast = , i.e. the half-difference between a highest point and the lowest point closest on its right , divided by the average of and . Mean contrast and STD derived from Equation (6) are consistent with this definition by taking and . The mean and STD of these instantaneous contrasts for each pair of fish are plotted in Figure 3B.
The linear integrate-and-fire model with dynamic threshold (LIFDT) used to simulate P-unit afferents is written as(7) represents the transmembrane potential measured from its resting level; is a dynamical threshold incremented by a fixed amount every time P-unit fires; is the input to P-units, here the experimental recording or the simulation signal used as the input. is the Heaviside function that accounts for the fact that many receptors rectify a periodic forcing [43]. mimics intrinsic noise, where is the noise intensity and is Gaussian white noise with zero mean (different from that used to generate the OUP). Parameter values are , , , , , and the time step is 0.01 ms. A P-unit fires when after which the voltage is reset to zero. The firing times generate a binary spike train .
The P-value is the characteristic parameter of a given receptor [29], [44]–[46]. It is calculated as the baseline firing rate (i.e. with the EOD of fish 1 alone and no AM) of P-type afferents, divided by the EODf. Because P-values follow a log-normal distribution and ranges from 0.1 to 0.6 with a mean value at 0.26 [29], we vary to obtain different P-values as shown in the densities of Figure 7E.
This statistic is used to measure the linear relationship between the frequency components of two signals, and . It can be seen as a signal-to-noise ratio, and reflects a lower bound on the mutual information (see below) between input and output. It is defined as(8)where , are, respectively, the auto-spectral densities of and , and is the cross-spectral density of and . As is the Fourier transform of the cross-correlation between and , coherence can be regarded as a correlation coefficient in the frequency domain, ranging at each frequency between 0 (no linear correlation) and 1 (perfect linear correlation).
Mutual information (MI) quantifies the mutual dependence of the two random variables. It has been widely used in computational neuroscience to analyze spiking neural systems, for example, characterizing the amount of information that the output spike trains carry about input signals. In the frequency domain, if the stimulus possesses Gaussian statistics, the estimate of MI rate can be expressed explicitly via the coherence function [43], [47], [48](9)where and indicate the lower and upper cutoff frequencies of the stimulus, respectively.
|
10.1371/journal.pntd.0002442 | Comparison of Phylogeny, Venom Composition and Neutralization by Antivenom in Diverse Species of Bothrops Complex | In Latin America, Bothrops snakes account for most snake bites in humans, and the recommended treatment is administration of multispecific Bothrops antivenom (SAB – soro antibotrópico). However, Bothrops snakes are very diverse with regard to their venom composition, which raises the issue of which venoms should be used as immunizing antigens for the production of pan-specific Bothrops antivenoms. In this study, we simultaneously compared the composition and reactivity with SAB of venoms collected from six species of snakes, distributed in pairs from three distinct phylogenetic clades: Bothrops, Bothropoides and Rhinocerophis. We also evaluated the neutralization of Bothrops atrox venom, which is the species responsible for most snake bites in the Amazon region, but not included in the immunization antigen mixture used to produce SAB. Using mass spectrometric and chromatographic approaches, we observed a lack of similarity in protein composition between the venoms from closely related snakes and a high similarity between the venoms of phylogenetically more distant snakes, suggesting little connection between taxonomic position and venom composition. P-III snake venom metalloproteinases (SVMPs) are the most antigenic toxins in the venoms of snakes from the Bothrops complex, whereas class P-I SVMPs, snake venom serine proteinases and phospholipases A2 reacted with antibodies in lower levels. Low molecular size toxins, such as disintegrins and bradykinin-potentiating peptides, were poorly antigenic. Toxins from the same protein family showed antigenic cross-reactivity among venoms from different species; SAB was efficient in neutralizing the B. atrox venom major toxins. Thus, we suggest that it is possible to obtain pan-specific effective antivenoms for Bothrops envenomations through immunization with venoms from only a few species of snakes, if these venoms contain protein classes that are representative of all species to which the antivenom is targeted.
| Snakebite envenomation is a serious health issue in Latin America, particularly in the Amazon, where antivenom administration may be delayed due to logistic constraints. Bothrops snakes are involved in most of the snakebite-related accidents in Brazil. This work reports a comparative study of the toxin composition and antigenicity of the Bothrops venoms used to prepare the commercial antivenom and its effectiveness against the venom from Bothrops atrox, a prevalent Amazon species that is not included in the pool. Our data show a lack of connection between Bothrops taxonomic identity and venom composition. We also show that different toxins display distinct reactivity with the tested antivenom. However, the antivenom reacted similarly with each class of toxin present in the venoms of the different snakes studied. Important evidence was the neutralization of the major toxic effects of B. atrox venom, not included in the mixture of antigens used to produce the antivenom. Based on the observed antigenicity of the distinct protein classes of toxins, we suggest that it is possible to obtain pan-specific and efficient Bothrops antivenoms via immunization with venoms from a few species of snakes that are representative of the protein composition of a large number of targeted species.
| Envenomation by snakebites, which is incorporated by the World Health Organization (WHO) in its list of neglected tropical diseases, constitutes an important worldwide public health concern, particularly in the rural areas of tropical countries as Africa, Asia and Latin America, affecting mostly agricultural workers and children [1]. The estimated number of global envenoming events exceed 400,000, with more than 20,000 fatalities [2]. In Brazil, the incidence is above 25,000 accidents/year, and the incidence in the northern region was 52.6 accidents/100,000 inhabitants in 2008 [3]. Most of the Brazilian accidents with species notification are due to vipers of the genera Bothrops (83.8%), Crotalus (8.5%) and Lachesis (3.4%), with only 3.4% of accidents related to the Elapidae snakes of the genus Micrurus [3]. Antivenoms raised in horses are the recommended treatment in Brazil.
Based on early reports [4], it was accepted that the efficacy of a specific antivenom covers bites by those snake groups with venom represented in the pool of antigens used for horse immunization for the production of that specific antivenom. Recently, the knowledge of venom toxins has increased considerably, especially due to the characterization of detailed composition of venom proteomes based on mass spectrometry. In 2007, the concept of ‘venomics’ was introduced by Calvete et al. [5] and the method was important to describe the venom composition from a great number of snake species, as revised recently [6], [7]. Then, it was possible to characterize the families of venom toxins represented in the venoms of different species of snakes [6], [7]. The implications of venomics in the rational necessary for the development of antivenoms was further supported by the ‘antivenomics’ [8], [9], that allowed the identification of venom proteins bearing epitopes recognized by one antivenom and the toxins not covered by the immune response of the hyperimmunized animal. The importance of venomics and antivenomics was readily incorporated in antivenom development, indicating the possibility of a rational design of pan-specific antivenoms combining distinct protein families in immunization pools [10]–[12].
The venom composition of many species of Bothrops complex is already known by venomics [13]–[27] or indirectly by transcriptomics [28]–[32]. From these studies, it has become clear that a limited number of protein families compose the venoms of Bothrops snakes, with snake venom metalloproteinases (SVMPs), snake venom serine proteinases (SVSPs) and phospholipases A2 (PLA2s) being the most abundant and most frequently correlated with the clinical symptoms of envenoming. SVSPs are generally thrombin-like enzymes that are involved in the coagulation disturbances observed in most patients [33]. PLA2s are involved in local effects and the myotoxicity observed in bites with some species [34]. SVMPs are multifunctional enzymes involved in the local and systemic symptoms of bites, such as the induction of local hemorrhage, inflammatory reaction, activation of coagulation factors and inhibition of platelet aggregation [35]. The variability in venom composition is notable and can be correlated with phylogeny [36], [37], age [38], [39], sex [40], geographical distribution [13], [40], [41] and diet [42]–[44] of the snake. However, venom variability is mostly related to the expression level of each group of toxin rather than to the presence or absence of major families of venom proteins. Moreover, within the same protein family, variability in the toxic properties may also occur when distinct functional motifs are introduced in structurally related toxins, increasing the diversity of targets that can be affected by venom toxins [45], [46]. Thus, the relevance of variability in venom composition should also be reflected in the reactivity with antivenom and its efficacy.
This problem particularly affects Bothrops snakes, which are diverse in their morphological and ecological traits and are distributed in different habitats throughout Latin America [47]. Due to the great diversity of Bothrops snakes, the systematics and phylogenetic relationships of this group are not completely resolved, and the distinction of snakes in different genera is often suggested. Based on morphology and mtDNA sequences, a broad classification of the Bothrops complex by Wüster et al. recognized Bothrops and Bothrocophias as independent genera [48]; furthermore, Castoe and co-authors [49] have proposed the classification of Bothrops, Bothrocophias and also Bothriopsis as independent genera. More recently, the Bothrops genus was further divided into three independent genera by Fenwick et al. [50]: Bothropoides, Rhinocerophis and Bothrops, representing the groups of “jararaca/neuwiedi”, “alternatus” and “jararacussu/atrox” snakes, respectively, previously recognized by Wüster et al. [48]. This classification was further questioned by Carrasco and collaborators [47], and the maintenance of Bothrocophias as an independent genus and synonymizing Bothriopsis, Bothropoides and Rhinocerophis within the Bothrops genus was suggested. However, according to the emerging methodology of DNA sequencing for cladistic analyses, it is reasonable to expect that further revisions of Bothrops systematics will be offered in the near future.
Following the classification of Fenwick and coworkers [50], several species of Bothropoides, Rhinocerophis and Bothrops groups are involved in snakebite envenomings, contributing to the high number of reported incidents in Brazil [3]. Antibothropic antivenoms are used in the treatment of these patients and are produced in Brazil by horse immunization with the venoms of five species of these snakes: Bothropoides jararaca, Bothropoides neuwiedi, Rhinocerophis alternatus, Bothrops moojeni and Bothrops jararacussu. In spite of venomics evidences showing the venom composition of several species, there are still concerns about the efficacy of Bothrops antivenoms in the treatment of envenomings inflicted by species whose venom is not used for animal immunization. These objections include mostly the accidents by Bothrops atrox, which is the snake responsible for the majority of snake bites in the Amazon, whose venom is not included in the immunization mixture. Most of these concerns arise because, in previous studies, the venoms were independently analyzed and, also, by the lack of comparative neutralization assays in the few papers showing antivenomics data for Brazilian Bothrops [16], [19], [26]. Thus, the complexity of the Bothrops group and the relevance of these species from a public health viewpoint justify the need for a multifaceted study comparing the venoms of the most relevant species and their reactivity with antivenoms in the light of recent proteomics studies.
In this study, we used a shotgun approach that allowed a simultaneous comparison of the composition of venoms collected from six species of snakes from the Bothrops complex, distributed in pairs from three distinct genera [50]. Fractionated venom components were tested for reactivity with the widely-used antivenom (SAB). The efficacy of the antivenom was then assessed for the neutralization of relevant symptoms of experimental envenomings by (a) B. jararaca, which accounts for 50% of venom composition in the immunization pool and is prevalent in the southeastern Brazil, and (b) B. atrox, which is not present in the immunization pool although representing a common cause of snakebite in the Amazon. The venom analysis showed that phylogenetic classification per se is not directly linked to venom composition. Furthermore, the antivenoms reacted equally with the toxins from the same protein family, regardless of snake phylogeny or the presence of the venom in the immunization pool used for antivenom production, highlighting new priorities when considering the selection of venoms to be used in the production of antivenoms.
The venoms of Bothropoides jararaca, Bothropoides neuwiedi (B. n. pauloensis, B. n. matogrossensis, B. n. marmoratus, B. n. neuwiedi and B. n. diporus subspecies), Rhinocerophis alternatus, Rhinocerophis cotiara, Bothrops jararacussu and Bothrops atrox were obtained from adult snakes of both sexes kept in captivity at the Laboratório de Herpetologia, Instituto Butantan, Brazil. The venoms from more than 10 specimens of each species were pooled, freeze-dried and stored at −20°C until use. Venoms from snakes kept under captivity represented as close as possible the same pools of venoms used for antivenom production and were used for proteomics and immunoreactivity assays. For experiments involving the neutralization of B. atrox venom toxic activities, we used venoms from wild B. atrox snakes collected at the Amazonian Floresta Nacional (FLONA) do Tapajós, Pará, Brazil, under SISBio license 32098-1, aiming to get venom samples as close as possible to the ones responsible for human accidents. Eight snakes were collected in pitfalls or by active search (five males and three females, with sizes ranging from 82 to 110 cm). The snakes were extracted in the herpetarium of Faculdades Integradas do Tapajós, Santarém, Pará, Brazil, and the venom from each snake was individually lyophilized and stored frozen until use, for which a pool was generated with equal proportions of venom from each snake. The chromatographic profile of the pool of venoms from snakes collected at Floresta Nacional do Tapajós was similar to that described below for the B. atrox venom pooled from snakes kept under captivity (data not shown).
The antibothropic serum (SAB) was produced at the Instituto Butantan, São Paulo, Brazil in horses immunized with a mixture of the following venoms: B. jararaca (50%), B. neuwiedi (12.5%), R. alternatus (12.5%), B. moojeni (12.5%) and B. jararacussu (12.5%). The final preparation consists of soluble IgG F(ab′)2 fragments: 1 mL neutralizes the lethality of 5 mg standard B. jararaca venom (according to the manufacturer). Anti-jararhagin monoclonal antibodies (MAJar-3) were produced in hybridomas previously selected and maintained in our laboratory, as previously described [51]. The MAJar-3 antibodies are IgG1 isotypes and recognize conformational epitopes located on the jararhagin disintegrin-like domain. MAJar-3 neutralizes jararhagin collagen binding and hemorrhagic activity and cross-reacts with hemorrhagins from venoms of different species of viper snakes [52].
Fifty micrograms of each venom were subjected to trypsin digestion, as previously described [53]. The tryptic digests were desalted with in-lab-generated columns packed with Poros R2 resin (Life Technologies, USA). Each of the 12 venom digests generated (6 venoms in duplicate) were analyzed in triplicate by nanoLC-MS/MS. The separation was performed on a 75 µm×30 cm column packed with a 5-µm, 200 A Magic C-18 AQ matrix (Michrom Bioresources, USA). The eluted peptides were directly injected into an LTQ/Orbitrap XL mass spectrometer (Thermo, USA) for analysis. The MS1 spectra were acquired using the orbitrap analyzer (300 to 1,700 m/z) at a 60,000 resolution (for m/z 445.1200). For each spectrum, the 10 most intense ions were subjected to CID fragmentation, followed by MS2 acquisition on a linear trap analyzer. The tandem mass spectra were extracted by RAW Xtractor (version 1.9.9.2) [54]. All of the MS/MS samples were analyzed using ProLuCID (version 1.3.1) [55]. ProLuCID was set up to search a database (forward + reverse decoy) that was built from the protein entries contained in the NCBI non-redundant database from April 29, 2012 that satisfied the following search terms criteria: “serpentes OR snakes OR snake OR venom OR venoms OR bothrops OR bothriopsis OR bothrocophias OR rhinocerophis OR bothropoides”. The database was comprised of 87,384 entries (43,692 “forward” and 43,692 “reverse decoy”). The ProLuCID search was performed with a fragment ion mass tolerance of 600 ppm and a parent ion tolerance of 70 ppm. Cysteine carbamidomethylation was specified as a fixed modification. Scaffold version 4.0.4 (Proteome Software Inc., USA) was used to validate the MS/MS-based peptide and protein identifications. The peptide identifications were accepted if they could be established at greater than 99.0% probability by the Peptide Prophet algorithm [56], with Scaffold delta-mass correction, and the protein identifications were accepted if they could be established at greater than 99.0% probability and contained at least 2 identified peptides. The protein probabilities were assigned by the Protein Prophet algorithm [57]. The acceptable false discovery rates, at the peptide and protein levels, were less than or equal to 1%.
The venoms were fractionated by reverse-phase high-performance liquid chromatography (HPLC) according to previously described reports [16], with some modifications. Samples of 5 mg of crude lyophilized venom were dissolved in 250 µL 0.1% trifluoroacetic acid (TFA), and the insoluble material was removed by centrifugation at 18,400×g for 10 min at room temperature. The proteins in the soluble material were applied to a Vydac C-18 column (4.6×250 mm, 10-µm particle size) coupled to an Agilent 1100 HPLC system. The column was eluted at 1 mL/min with a gradient of 0.1% TFA in water (solution A) and 0.1% TFA in acetonitrile (solution B) (5% B for 10 min, followed by 5–15% B over 20 min, 15–45% B over 120 min, 45–70% B over 20 min and 70–100% B over 10 min). The separations were monitored at 214 nm, and the peaks were collected manually and dried in a Speed-Vac (Savant). The fractions were resuspended in PBS, and the protein concentration was estimated by OD at 280 nm in a NanoVue plus spectrophotometer (GE Healthcare).
The venoms were classified according to their toxin composition by hierarchical clustering of observations constructed using nearest neighbor linkage method (minimum Euclidean distance between items in different clusters), considering initially each observation as an individual cluster. The degrees of similarity between observations were expressed in terms of a cluster tree (dendrogram). We performed also a Principal Component Analysis (PCA) in order to understand the key toxins responsible for the venom clustering. The principal components 1 (PC1) and 2 (PC2), which were responsible for explaining more than 70% of the total variability, were calculated using the covariance matrix. The toxin composition loadings and venom scores were expressed in terms of loading and score plots. These procedures were performed in Minitab 16 software.
The variables used for clustering and PCA were the relative concentrations of each toxin family, accessed by shotgun mass spectrometry. The mean of each protein family spectral counts was normalized by the total venom counting [1,891 (B. atrox); 1,727 (B. jararacussu); 2,719 (B. jararaca); 2,287 (B. neuwiedi); 1,252 (R. alternatus) and 1,767 (R. cotiara)], distributed within the identified protein families: SVMP-I, -II and –III (snake venom metalloproteinase - classes P-I, P-II and P-III); PLA2 (phospholipase A2); SVSP (snake venom serine proteinase); CLEC (C-type lectin); CLECL (C-type lectin-like); LAAO (L-amino acid oxidase); NGF (nerve growth factor); HYALU (hyaluronidase); VEGF (vascular endothelial growth factor); CRISP (cysteine-rich secretory protein); PDIEST (phosphodiesterase 1); ECTONT (ecto-5′-nucleotidase); PLB (phospholipase B); GLUTCYC (glutaminyl cyclase) and ACTIN (actin).
The venoms were also analyzed by the relative mAU of the highest peaks collected in C-18 reverse-phase chromatography in the elution time intervals of 56–57, 57–58, 58–60, 67–71, 108–112, 113–116, 121–123, 124–127, 128–129, 130–132, 134–136, 136–138, 139–140, 140–150, 150–152, 153–155, 157–159, 160–162, 163–164, 164–166, 166–168, 169–170, and 171–172 minutes. The mAU values of the peaks were normalized in % by the mAU of the highest peak eluted in the chromatography, taken as 100%.
Samples containing 100 µL whole venom (10 µg/mL) or isolated fractions (1 µg/mL), in carbonate buffer (pH 9.6), were used to coat maxisorb microplates (Nunc). To determine the antibody titers, plates coated with whole venom were incubated with serial dilutions of SAB (from 1∶10,000), followed by incubation with anti-horse IgG labeled with peroxidase (1∶2,000). For assessing the antigenicity of the fractions, the plates were incubated with a fixed dilution of SAB (1∶1,000) or MAJar-3 (1∶50), followed by incubation with anti-horse IgG (1∶1,000) or anti-mouse IgG (1∶1,000) labeled with peroxidase. The reactions were developed with ortho-phenylenediamine/H2O2 as the enzyme substrate, and the products were detected at 490 nm. The reactions were performed in duplicates in three independent experiments. The results of antivenom titration are expressed as mean ± sd of the six OD values. The results of fraction antigenicity were calculated as mean of the six OD values after normalization using as 100% the maximal OD value obtained in each of the independent experiments [(Fraction OD/maximal OD of the test)×100].
Samples of crude venom (10 µg) were subjected to 12.5% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) under non-reducing conditions. After SDS-PAGE, the separated proteins were transferred to nitrocellulose membranes, which were then immersed in a blocking solution (5% non-fat milk in Tris-saline). The membranes were incubated with SAB (1∶1,000) as the primary antibody and then with peroxidase-labeled goat anti-horse IgG (1∶1,000). The reactive bands were detected by incubation with 4-chloro-α-naphthol and H2O2. The results shown represent three independent experiments.
For accessing the neutralization of the lethal and hemorrhagic venom activities, Swiss mice bred and maintained at the Instituto Butantan (Brazil) animal house were used as an animal model.
For the neutralization of hemorrhagic activity, doses of 10 µg B. jararaca or B. atrox venom were incubated with SAB at ratios of 1, 2 or 4 times the SAB volume required to neutralize 10 µg of reference venom, according to the manufacturer. The mixtures were incubated at 37°C for 30 min, and a 50-µL aliquot of each mixture was injected intradermically in the dorsa of a group of 5 mice. The control groups included mice injected with PBS or with venom incubated with PBS. At three hours after the injection, the mice were sacrificed by CO2 inhalation; the skin of the dorsa was removed, and the hemorrhagic spots were measured (longest diameter multiplied by the diameter perpendicular to it). The results represent the values obtained for 5 different mice and are expressed as the % neutralization using as 100% activity the value obtained after an injection with venom incubated with PBS.
For the neutralization of lethal activity, the LD50 values of B. jararaca and B. atrox venoms were estimated according to previous studies [58] to avoid unnecessary animal sacrifice. In all experiments, 3 LD50 doses of B. jararaca (105 µg) or B. atrox (225 µg) venom were incubated with SAB at ratios of 1, 2 and 4 times the potency reference value (1 mL/5 mg venom). The mixtures were incubated at 37°C for 30 min, and 500-µL aliquots were injected intraperitoneally in groups of 5 mice. Control groups included mice injected with PBS or with venom incubated with PBS. Lethality was recorded over a period of 48 hours. The results shown represent the values obtained in 3 independent experiments and are expressed as the % neutralization considering the number of dead/live mice after 48 hours.
The neutralization of the coagulant activity was determined as previously described [59], with some modifications. Samples containing 2 minimum coagulant doses of B. jararaca (71.3 µg/mL) or B. atrox (21.7 µg/mL) venom were incubated with several dilutions of SAB for 30 min at 37°C. Each mixture was added to 100 µL bovine plasma, and the clotting times were recorded using a model ST4 mechanical coagulometer (Diagnostica Stago). Neutralization was expressed as the effective dose (ED), defined as the antivenom/venom ratio at which the clotting time was increased threefold when compared to the clotting time of plasma incubated with venom alone.
All experiments involving mice were approved by the Ethical Committee for Animal Research of the Instituto Butantan (CEUAIB), São Paulo, Brazil, (application approval number 752/10), who certified its agreement with the Ethical Principles in Animal Research adopted bt the Brazilian College of Animal Experimentation (COBEA).
To evaluate the relationship between venom composition and phylogenetic position of the species, we analyzed the proteome of the venoms from the six selected species using shotgun nanoESI-LTQ/Orbitrap. The distribution of the protein families in selected venoms was calculated according to the normalized total spectral counts. As shown in Figure 1, the data analysis revealed 15 different protein groups in different proportions: SVMP-I, -II and –III (snake venom metalloproteinase - classes P-I, P-II and P-III); PLA2 (phospholipase A2); SVSP (snake venom serine proteinase); CLEC (C-type lectin); CLECL (C-type lectin-like); LAAO (L-amino acid oxidase); NGF (nerve growth factor); HYALU (hyaluronidase); VEGF (vascular endothelial growth factor); CRISP (cysteine-rich secretory protein); PDIEST (phosphodiesterase 1); ECTONT (ecto-5′-nucleotidase); PLB (phospholipase B); GLUTCYC (glutaminyl cyclase) and ACTIN (actin). The SVMPs were the most abundant toxins in all of the venoms, particularly in the B. atrox, R. alternatus, R. cotiara and B. jararaca venoms, in which class P-III was notably the predominant toxin. PLA2s predominated in the B. jararacussu venom and was found in significant amounts in the B. neuwiedi venom. A significant contribution of C-type lectin-like proteins was also detected in the B. jararaca, R. alternatus and B. atrox venoms, whereas the SVSPs and LAAOs were almost equally distributed in all of the venoms. One interesting fact was the significant contribution of C-type (true) lectins in the B. jararacussu (8.8%) and B. neuwiedi (3.5%) venoms, in parallel with its absence (<1%) in the other venoms (Figure 1). Comparing these data with previous venomics studies [16], [18]–[20], [23], the major venom protein families as SVMPs, PLA2s and SVSPs were detected in our study in equivalent proportions. However, shotgun nanoESI-LTQ/Orbitrap allowed the detection in all venoms tested of some proteins not yet described as PDIEST, ECTONT, PLB and GLUTCYC. Also, NGF, detected here in all venoms, and HYALU, present in B. atrox, B. jararaca, R. alternatus and R. cotiara venoms, were previously detected in transcriptomes of B. jararacussu and Bothropoides pauloensis, respectively [19], [30], but not in their venomes. Five spectra identified as actin were detected in R. cotiara venom shotgun and due to the high sensitivity of the method, may derive from a minor contamination of the venom with venom gland cells. The most striking difference was the presence of significant amounts of LAAO, CLECL and CLEC spectra detected in our samples, compared to the previous venomics studies. Proteomics by shotgun nanoESI-LTQ/Orbitrap is based on a whole venom digestion by trypsin and the peptide mixture is then fractionated and analyzed in a high sensitive detection system. This approach may bias peptides with higher ionizable efficiency, but all protein families will be represented in the original mixture at the same proportions as they are present on venoms and the bias due to ionization efficiency will be the same for similar peptides present on venoms from different species. Thus, this method is appropriate for comparative studies, allowing the simultaneous analysis of different venoms, under exactly the same conditions. On the other hand, the traditional venomics [5] includes one step in which proteins are quantified and selected after SDS-PAGE separation, according to their staining by Coomassie blue. After trypsinization of selected bands, peptide detection and protein identification will also depend on peptide ionizable efficiency. It is well known that proteins present in venom mixtures in low proportions are hardly detectable by SDS-PAGE as some other venom proteins may be weakly stained. These proteins would be neglected in total protein detection and also when calculating their proportional participation in venom composition. The differences in protein separation methods and sensitivity of detection systems could explain the higher participation of some protein families described in our study when compared to the traditional venomics.
The venoms were also compared according to the elution profile from reverse-phase C-18 columns. To compare our findings with the previous data from B. atrox, B. cotiara and B. neuwiedi venomics studies [16], [19], [20], C-18 reverse-phase chromatography protocols using similar columns, buffer systems and elution conditions were used to fractionate the venoms. Figure 2 shows the chromatographic profile of the venoms from the six species selected for this study. As expected, the venoms presented comparable chromatographic profiles to those reported in the referenced studies. According to these previous studies, the major protein families were eluted as follows: disintegrins at approximately 50–60 min [19], [20]; basic PLA2s at approximately 110–120 min [19]; P-I SVMPs, some D-49 PLA2s and SVSPs between 120 and 160 min [18]–[20] and P-III SVMPs predominating after 160 min [18]–[20]. Using these data as references, P-III SVMPs appeared to be the most abundant antigens in the chromatograms of the B. atrox, R. alternatus, R. cotiara, B. jararaca and B. neuwiedi venoms, whereas several different peaks in the region corresponding to P-I SVMPs and SVSPs were detected. These observations are consistent with our venomic analysis results shown in Figure 1 and with previous proteomic studies in which P-III SVMPs comprised more than 50% of B. atrox venom [16], [18], approximately 50% of R. alternatus venom [23], approximately 70% of R. cotiara venom [20] and approximately 25.9% of B. neuwiedi venom [19]. SVMPs were also reported to comprise 53.1% of B. jararaca venom gland toxin transcripts [31]. The B. jararacussu venom was the most distinct venom in this group, showing a predominant peak in the PLA2 region and a low abundance of SVMPs, which is consistent with the literature showing a high expression of PLA2 in B. jararacussu venom glands and representing 35% of the total transcripts, followed by only 16% SVMPs and 2% SVSPs [30]. The marked difference in B. jararacussu venom compared to the other Bothrops species was previously reported [60], and a K-49 myotoxin yield of 25% from the crude venom was purified and considered to be the predominant antigen of the B. jararacussu venom [61].
According to the independent parameters used to compare the venoms, in Bothrops, the B. jararacussu profile was very different from that of B. atrox, showing a higher content of phospholipase A2 and a smaller amount of the class P-III metalloproteinase (SVMP) group, as detected either by proteomics or by the elution profile of the native proteins. Within the Bothropoides genus, major differences were observed by proteomics, such as the higher content of CLECL and P-III SVMP in B. jararaca and PI and PII SVMPs, PLA2 and CLEC in B. neuwiedi. The venoms were more similar within the Rhinocerophis genus, particularly when comparing the elution profile of the native proteins, though a higher contribution of CLECL was found in R. alternatus, and higher contents of L-amino acid oxidase and serine proteinase were detected in the R. cotiara venom using the proteomics approach. However, the distribution of B. atrox venom components was very similar to that of R. alternatus by both methods. Furthermore, the pattern observed for B. neuwiedi was closer to that of B. jararacussu venom due to the presence of higher levels of PLA2 and CLEC (Figures 1 and 2). Thus, apparently, venom composition was not related to the phylogenetic position of the snakes.
In order to statistically demonstrate these differences, the normalized values of the venom composition obtained by the total spectrum counts of each protein family, and the mAU values of the major peaks eluted in different volumes during the C-18 chromatography, were used as variables to cluster the venoms of snake species. A Principal Component Analysis (PCA) was also carried out in order to understand the key toxins responsible for the venom clustering. The resulting dendrograms and loading and score plots of the PCA are shown in Figures 3 and 4, respectively. Clustering according to the C-18 elution profile shows a strong similarity between R. alternatus and B. jararaca venoms. B. atrox and R. cotiara venoms also show similar elution profile, but different than R. alternatus and B. jararaca venoms, forming, therefore, two different clusters. On the other hand, B. neuwiedi and B. jararacussu venoms reveal lower similarity with the two former clusters, with B. jararacussu having the most distinct features (Figure 3). In the PCA, shown in Figure 4A, components with most prominent loadings that contributed to venom clusterization are the fractions eluted after 160 min with the highest negative values of PC1 (Fraction 164–166: PC1 = −0.365, PC2 = 0.209; Fraction 166–168: PC1 = −0.175, PC2 = −0.128; Fraction 169–170: PC1 = −0.461, PC2 = 0.481; Fraction 171–172: PC1 = −0.311, PC2 = −0.783). These fractions were characterized mostly as class P-III SVMPs in other studies [18]–[20] and reacted with MAJar-3 monoclonal antibodies in this study (see below). Fractions with the highest PC1 positive values were eluted between 108–112 min (PC1 = 0.630, PC2 = 0.029), recognized as PLA2s in previous studies [19], and fractions between 130–132 min (PC1 = 0.330, PC2 = −0.018), characterized as class P-I SVMP in the venom of adult B. atrox from El Paují (Orinoquia, Venezuela) that underwent ontogenetic variation [16]. With respect to proteomic data, B. atrox and R. alternatus venoms were the most closely related, and distances to this group increased gradually for R. cotiara, B. jararaca, B. neuwiedi and B. jararacussu venoms. The clustering of B. atrox and R. alternatus venoms is related to high values of CLECL and P-III SVMPs, which are the proteins with most prominent loadings (CLECL: PC1 = −0.431, PC2 = 0.789, P-III SVMPs: PC1 = −0.592, PC2 = −0.472), and low values of PLA2 and CLEC, also with significant loadings (PLA2: PC1 = 0.245, CLEC : PC1 = 0.245). R. cotiara venom shows similar pattern with respect to P-III SVMP, PLA2 and CLEC, but low values of CLECL and high values of LAAO (PC1 = 0.037, PC2 = −0.339). On the other hand, B. jararaca venom reveals low values of LAAO and large values of CLECL. B. neuwiedi and B. jararacussu venoms present an opposite pattern, with high values of PLA2 and CLEC and low values of PIII-SVMP (Figure 4 B).
The dendrograms and PCAs obtained using the distinct sets of variables do not coincide, as they were based in distinct parameters. The number of total spectral counts of a given protein is not necessarily related to its mAU 214; moreover, chromatographic fractions represent mixtures of protein families treated as independent variables in the cluster corresponding to the proteomic data. In spite of these differences, both sets of variables indicate that the distribution of venoms is not related to the phylogenetic position of the snakes. It is important to note that a more comprehensive study using venoms from a larger number of species, quantitative assays for isolated components and also complete sequences of venom proteins would be essential to a definitive support of the lack of connection referred to above. Nevertheless, our data are supported by the literature. Taken together, the clusterization and PCA analysis indicate a polarization among the venoms. According to significant PC1 loadings, B. atrox, R. alternatus, R. cotiara and B. jararaca venoms are clearly opposite to B. jararacussu venom, the former group with prominent negative PC1 values of class P-III SVMPs, while B. jararacussu venom shows a polarization towards the presence of PLA2s and class P-I SVMPs. The same toxin polarization has been indicated to venoms from snakes that conserved the paedomorphic characteristics in their venoms (first group) and venoms of snakes whose venom underwent ontogenetic variation (in our study, B. jararacussu venom) [13], [16], [18], [38], [39]. Interestingly, B. neuwiedi venom was grouped closer to B. jararacussu in the cluster analysis, but showed smaller negative PC1 scores, in opposition to B. jararacussu venom. According to the distances, B. neuwiedi venom apparently conserved the paedomorphic phenotype, but may be suffering a transition to the ontogenetic changes observed in B. jararacussu or B. atrox from Colombia.
Correlations between phylogeny and venom composition have been appointed in the literature [36], [37]. Nevertheless, differences in composition of venoms from snakes belonging to the same genera are also present in the literature [62]–[64]. In a recent study, Gibbs et al. [65] found no evidence for significant phylogenetic signal in venom variation of Sistrurus spp, suggesting that diet variation may play a more important role in molding the venom composition. A remarkable variation in venom composition and toxicity was reported for rattlesnakes from Crotalus viridis/oreganus complex [66] and Crotalus durissus and Crotalus simius in Central and South American species [67]. In the latter, differences were related to the conservation of the newborn characteristics of Central American rattlesnake, C. simus, in the South American species and sub-species of C. durissus, a typical example of paedomorphism [67]. These examples are also found in snakes of the Bothrops complex. Tashima et al. [20] reported significant differences in venom composition between two species closely related, R. cotiara and R. fonsecai. A paedomorphic characteristic was also conserved along the dispersion of B. atrox from Central America to the Brazilian Amazon [16], including in the population used in this study. The conservation of the paedomorphic characteristics in B. atrox accounted for the concentration of class P-III SVMPs, which greatly contributes to the overall toxicity of Bothrops venoms [35]. Paedomorphic characteristics were not conserved in B. jararacussu venom, which has predominance of enzymatically inactive myotoxic PLA2s [60] and therefore, presents lower toxicity compared to B. atrox venom. The difference in composition and toxicity of B. atrox and B. jararacussu venoms argues in favor that the gain in toxicity was favorable in B. atrox due to its smaller size. According to this hypothesis, paedomorphic characteristic would not be essential to B. jararacussu snake that is very large and capable of inoculating large amount of venoms in mammalian preys.
Our next approach was to evaluate the reactivity of the whole venoms and their isolated fractions with antivenoms. Figure 5 shows the titration curves of the antibothropic serum (SAB) in ELISA plates coated with equal amounts of each venom. The SAB antibody titers were the same, regardless of the antigen used, and they corresponded to a dilution of 640,000. The only differences among the venoms were the values obtained for the 10,000 and 20,000 dilutions of SAB against the B. jararacussu venom, which were significantly lower than comparing with other venoms. These dilutions reflect the zone at which the antigen concentration is the limiting factor, and differences in antibody binding may reflect the lower amount of reactive antigens in B. jararacussu venom, highlighting the antigenic relevance of P-III SVMPs. Indeed, the region correspondent to bands of approximately 50 kDa, which is the approximate molecular mass of P-III SVMPs, were less intense in the B. jararacussu venom electrophoresis than others (Figure 6A). SAB preferentially recognized bands of approximately 50 kDa by western blotting (Figure 6B), confirming the higher immunogenicity of SVMPs class P-III. Bands between 20 and 30 kDa, with masses corresponding to SVSPs and P-I SVMPs, were also recognized by SAB (Figure 6B). The SAB reactivity with each fraction from reverse-phase chromatography was also assessed and compared to the reactivity of a monoclonal antibody, MAJar-3, which recognizes the disintegrin domain of P-III SVMPs [51]. In Figure 7, we demonstrate the strong reactivity of the monoclonal antibody with the fractions eluted after 160 minutes (in all chromatograms), confirming that these fractions correspond to P-III SVMPs. The same fractions were the most SAB-reactive antigens in all venoms, regardless of whether these venoms were included in the immunization pool used to prepare the SAB antivenom. Even for the B. jararacussu venom, with a low abundance of SVMPs, the fractions eluted after 160 minutes were the most reactive. Intermediate levels of reactivity were detected with the fractions eluted between 120 and 160 minutes, with very limited reactivity for some, particularly the venoms of B. atrox and B. alternatus, suggesting a lower antigenicity of P-I SVMPs and SVSPs in relation to the SAB antivenom. Interestingly, three small peaks collected from the R. cotiara venom at approximately 140 minutes were strongly reactive with SAB and also with MAJar-3, suggesting the presence of P-III SVMPs in this venom, with distinct structural features and elution profiles. Despite the inclusion of B. jararacussu and B. neuwiedi venoms in the immunization pool, the reactivity of SAB with their fractions (showing PLA2 elution characteristics) from 100 to 110 minutes was moderate. The fractions eluted prior to 100 minutes in all of the chromatograms were poorly recognized by SAB. In other publications, fractions that eluted before 100 min under similar chromatographic conditions corresponded to disintegrins [19], [20], vasoactive peptides [19] or DC fragments of SVMPs [20].
Interestingly, despite the different methods used in this study, our results are comparable to those of Núñez et al. [18] and Calvete et al. [16], who showed the complete immunoprecipitation of PIII-SVMPs, to a minor extent of SVSPs and DC-fragments, and limited immunoreactivity towards PLA2 molecules and PI-SVMPs by antivenomics of B. atrox venom with commercial antivenoms. Using antivenomics of B. asper venom and commercial antivenoms, Gutiérrez et al. [9] also showed complete immunodepletion of P-III SVMPs and partial depletion of PLA2s, some serine proteinases, and P-I SVMPs. Correa-Neto et al. [26] approached the same issue by immunomics where the western blots of 2D-gel electrophoresed venoms revealed that antiserum against B. jararacussu venom showed higher reactivity to SVMPs and weaker reactivity towards SVSPs and PLA2s, and anti-jararaca serum preferentially recognized SVMPs and SVSPs among other antigens. Both of these sera failed to recognize low-molecular weight proteins [26]. Comparing the different methods, antivenomics is the best choice for a detailed study, since identifications of non-depleted proteins will show exactly the antigens that are partially immunodepleted or non-reactive with the antivenom. However, the method used here has the advantage to allow simultaneous tests of different venoms, at exactly the same conditions, and gives comparable results to antivenomics, thus is appropriate for comparative studies.
Important conclusions arise from these results. It becomes clear that P-III SVMPs are the predominant antigens in the venom of snakes from the Bothrops complex. Moreover, at least among the Bothrops, SVMPs are cross-reactive antigens that are equally recognized in venoms, regardless of their inclusion in the immunization pool. This is a good indication for antivenom efficacy, as P-III SVMPs are also abundant in most of these venoms and are related to the important symptoms of local and systemic envenomings, such as hemorrhage, the activation of coagulation factors, the inhibition of platelet aggregation and the activation of several factors that lead to local symptoms [35]. Interestingly, P-III and P-I SVMPs share similar catalytic domains and catalytic properties [68], which are involved in most of the toxic activities of SVMPs. Therefore, it is very intriguing that P-I SVMPs are less recognized by the antivenoms than are P-III SVMPs and raises some concerns about the neutralization efficacy of those activities related to the catalytic domain of these molecules. This observation suggests different interpretations: the most immunogenic epitopes of SVMPs may be located within the disintegrin-like or cysteine-rich domains; or catalytic domains of P-III SVMPs are more immunogenic than catalytic domains of P-I SVMPs. For instance, high hemorrhagic activity and the inhibition of platelet aggregation are typical for P-III SVMPs and depend upon disintegrin-like/cysteine-rich domains [69], [70], yet P-I SVMPs are able to induce local reactions [71] and activate coagulation factors [72], which are important effects of snake bites.
SVSPs and PLA2s are important toxins involved in the coagulopathy and local effects, respectively, of patients bitten by snakes of the Bothrops complex. Thus, the limited reactivity of SAB with these fractions should be addressed. Most SVSPs are thrombin-like enzymes involved in the blood coagulation disturbances induced by venom [33], and this symptom is easily controlled in patients after antivenom administration [73], suggesting that the presence of anti-SVSP antibodies in SAB is appropriate to neutralize the activity. However, PLA2s are generally myotoxic or pro-inflammatory [34], and these symptoms are not well neutralized by antivenoms. In the case of SVSPs, it appears that the low levels of antibodies present in SAB are sufficient to neutralize the systemic effects of SVSPs after intravenous administration. In contrast, this does not appear to be the case for the neutralization of the local effects of envenomings induced by PLA2s or P-I SVMPs. This lack of efficacy could most likely be dependent upon antivenom biodisponibility at the site of the lesion rather than on the potency of an antivenom against the myotoxic or dermonecrotic components of the venom [74] or the antibody titer against the toxins inducing the local effects.
Another important point observed in this study was the limited reactivity of antivenom with disintegrins and the DC fragments of SVMPs, which are recognized as inhibitors of platelet aggregation [69], [75], and its reactivity with vasoactive peptides. Although they are not presently considered major toxins correlated with the symptoms of envenomings, the additive or synergistic role of these small toxins in snake bite disorders cannot be ruled out. These low molecular mass peptides are known to be weakly immunogenic; however, in antivenomics studies, at least DC fragments and disintegrins were depleted from B. atrox [18] and B. asper [9] venoms by commercial antivenoms. Nevertheless, the presence of antibodies against such classes of low molecular size toxins in antivenoms should be regarded with more attention.
The next step of this study was to evaluate the SAB neutralization efficacy of the lethality, hemorrhagic and coagulant activities of B. atrox venom in comparison to B. jararaca venom. For these experiments, we used venoms from snakes collected in a region where many accidents are reported. The accepted potency of SAB efficacy, calculated as the volume necessary to neutralize the lethality of standard B. jararaca venom, is 1 mL antivenom/5 mg venom. This value was used as a reference to design the neutralization protocols used in this study, whereby this proportion was sufficient to protect more than 50% of mice from the challenge with 3 LD50 doses of B. jararaca venom (105 µg). However, neutralization of the 3 LD50 doses of B. atrox venom (225 µg) was achieved only when the proportion of 2 mL antivenom/5 mg venom was used (Figure 8). Most of the standard protocols to assess antivenom potency use a fixed LD50 value to challenge experimental mice. Therefore, this is also the reference assay used to compare the antivenom efficacy against different venoms. However, it is important to consider that LD50 values are variable among venoms and reflect the toxic activity of each toxin and their synergistic effect to induce death. Additionally, in most tests, the mice are challenged with pre-incubated mixtures of venoms and antivenoms, and, in these reactions, toxins are neutralized or cleared from the solution on a molar concentration basis rather than according to the neutralization of activity. This fact may explain why several previous studies reported that some venoms with higher LD50 values, such as B. atrox and B. jararacussu, are neutralized with higher concentrations of commercial antivenoms.
Similar findings were observed in our study regarding the neutralization of the coagulant activity of B. atrox and B. jararaca venoms. In this case, the B. atrox venom was more coagulant (minimal coagulant concentration in plasma: 10.8 µg/mL) than the B. jararaca venom (minimal coagulant concentration in plasma: 35.6 µg/mL), and higher concentrations of B. jararaca venom were used in the assays. The SAB neutralized the coagulating activity of both venoms; in this case, however, lower amounts of antivenoms were needed to neutralize the B. atrox activity (ED = 627 µL antivenom/mg venom), whereas B. jararaca venom neutralization required a higher antivenom concentration (ED = 1400 µL antivenom/mg venom), as shown in Figure 8.
The hemorrhagic activity was comparable between the venoms, and the ratio of 1 mL antivenom/5 mg venom neutralized more than 50% of the hemorrhage induced by both venoms (Figure 8). Taken together, these data suggest that SAB is efficient in neutralizing the most important effects of B. atrox venom despite the phylogenetic distance of the snake and the fact that the venom is not included in the immunization pool used to produce SAB. There are previously published papers in the literature suggesting the need to include B. atrox venom for horse immunization [76]–[78]. However, our data showing the opposite are supported by a previous study in which SAB immunodepleted the venom proteins from B. atrox populations exhibiting the paedomorphic venom phenotype, the same pattern found in specimens collected in Pará State, Brazil [16]. Moreover, our present data are supported by other pre-clinical assessments showing neutralization of the toxic activities of venoms not included in immunization protocols [79], [80] and by a clinical trial for the treatment of snake bite patients clinically classified as mild and moderate in Pará State (Brazil) demonstrated that the efficacy of a conventional antivenom (SAB) was comparable to the efficacy of an experimental antivenom prepared through horse immunization with B. atrox venom [81].
Recently, the understanding of venom composition by venomics [5], [6] and tests of the efficacy of antivenoms by antivenomics [8], [9], [11] have been extremely important approaches in order to achieve efficient antivenoms [82]–[84]. In this work, we approached this issue by a multifaceted comparative study of venoms from six species of snakes of distinct phylogenetic clades of Bothrops complex. Important differences were observed in venom composition of the snakes from Bothrops complex, mainly for B. jararacussu venom. However, these differences showed no apparent relationship with the phylogeny of the snakes. In this regard, although the taxonomy of this group is still under revision, the toxins present in the venoms are similar, in agreement with previous molecular data showing that the ancestral genes encoding Bothrops major toxin families were already present before the differentiation of the Bothrops species [85], [86]. As a result, the antivenom reacted similarly with toxins from the same protein family, as SVMPs, SVSPs or PLA2s, regardless of the snake phylogeny or the presence of the venom in the immunization pool used for antivenom production. Thus, we confirm previous data of antivenomics and suggest that it is possible to obtain pan-specific and efficient antivenoms to Bothrops snakes through immunization with venoms from a few species of snakes, if immunogenicity and antigenicity of the distinct protein classes of toxins are considered.
|
10.1371/journal.pntd.0002156 | Aerosol Exposure to Rift Valley Fever Virus Causes Earlier and More Severe Neuropathology in the Murine Model, which Has Important Implications for Therapeutic Development | Rift Valley fever virus (RVFV) is an important mosquito-borne veterinary and human pathogen that can cause severe disease including acute-onset hepatitis, delayed-onset encephalitis, retinitis and blindness, or a hemorrhagic syndrome. Currently, no licensed vaccine or therapeutics exist to treat this potentially deadly disease. Detailed studies describing the pathogenesis of RVFV following aerosol exposure have not been completed and candidate therapeutics have not been evaluated following an aerosol exposure. These studies are important because while mosquito transmission is the primary means for human infection, it can also be transmitted by aerosol or through mucosal contact. Therefore, we directly compared the pathogenesis of RVFV following aerosol exposure to a subcutaneous (SC) exposure in the murine model by analyzing survival, clinical observations, blood chemistry, hematology, immunohistochemistry, and virus titration of tissues. Additionally, we evaluated the effectiveness of the nucleoside analog ribavirin administered prophylactically to treat mice exposed by aerosol and SC. The route of exposure did not significantly affect the survival, chemistry or hematology results of the mice. Acute hepatitis occurred despite the route of exposure. However, the development of neuropathology occurred much earlier and was more severe in mice exposed by aerosol compared to SC exposed mice. Mice treated with ribavirin and exposed SC were partially protected, whereas treated mice exposed by aerosol were not protected. Early and aggressive viral invasion of brain tissues following aerosol exposure likely played an important role in ribavirin's failure to prevent mortality among these animals. Our results highlight the need for more candidate antivirals to treat RVFV infection, especially in the case of a potential aerosol exposure. Additionally, our study provides an account of the key pathogenetic differences in RVF disease following two potential exposure routes and provides important insights into the development and evaluation of potential vaccines and therapeutics to treat RVFV infection.
| Rift Valley fever (RVF) is an important neglected tropical disease that has caused severe epidemics and epizootics throughout Africa and the Arabian Peninsula. Severe outbreaks have involved tens of thousands of both human and livestock cases for which no effective, commercially available human vaccines or antiviral drugs are available. Mosquito transmission is the primary means for human infection although it can also be transmitted by aerosol or through mucosal contact. In this study, we directly compared the pathogenesis of RVF virus (RVFV) following aerosol exposure to a peripheral exposure in the mouse model. Additionally, we evaluated the effectiveness of ribavirin, a known antiviral compound, administered prior to virus exposure. The virus affected the liver despite the route of exposure. However, viral invasion of brain tissues occurred much earlier and was more severe in mice exposed by aerosol compared to peripherally exposed mice. Mice exposed peripherally and treated with ribavirin were partially protected whereas treated mice exposed by aerosol were not protected. Our results have important implications for understanding the pathogenesis of RVFV and for the development and evaluation of potential vaccines and therapeutics.
| Rift Valley fever virus (RVFV) is a negative sense, single-stranded RNA virus of the Bunyaviridae family, genus Phlebovirus. An important pathogen of humans and domesticated animals, the virus may be transmitted when an infected mosquito feeds on a host, by contact with tissues, blood, or fluids from infected animals, or by exposure to infectious aerosols, such as those generated in abattoirs during handling of infected animals [1]. RVFV was first isolated during a 1930 epizootic in East Africa [2]. Subsequent outbreaks ranged from South Africa to Egypt, and west to Senegal [3]. Until the 21st century, it was assumed that RVFV was confined primarily to continental Africa; however, epidemics in Saudi Arabia and Yemen after unusually heavy rainfall in 2000 indicate the possibility of a RVFV introduction occurring wherever a competent vector exists [4]. Additionally, productive experimental infection of mosquitoes from multiple distinct geographical regions (including the most widespread vector, Culex pipiens) reinforces the feasibility of accidental or intentional import of RVFV from endemic regions with subsequent maintenance in nascent vector and host populations [5]–[8].
In livestock the susceptibility and virulence differ by host species and age [3]. Spontaneous abortion is such a common consequence of infection among domesticated animals that the resulting abortion “storms” arising during epizootics are often the first recognized sign of a RVFV outbreak [9]. Necrotizing hepatitis, encephalitis, and death commonly occur in infected livestock. The disease in humans is generally a self-limiting, mild influenza-like condition with fever. However, a smaller but still poorly defined percentage of individuals develop a severe illness characterized by acute-onset hepatitis, delayed-onset encephalitis, retinitis, blindness, or a hemorrhagic syndrome [3]. While it has been traditionally reported that 1%–2% of human infections result in severe disease, case-fatality rates have varied between epidemics and are generally estimated to range from 10–20% for hospitalized individuals [10]–[12].
In conjunction with its infectiousness and pathogenicity in humans, the range and prolificacy of its host species highlight the importance of RVFV as a potential agent of bioterrorism. Furthermore, there are currently no licensed vaccines or approved antivirals for preventing or treating RVF in humans, and its infectiousness by aerosol exposure coupled with ease of proliferation adds to the threat posed by intentional dissemination of RVFV. As such, RVFV is listed as a select agent by the Centers for Disease Control and Prevention (CDC), and is classified as a Department of Health and Human Services (HHS) and United States Department of Agriculture (USDA) overlap select agent.
Several animal models of RVFV infection have been described [13]. Recently, we completed a detailed study to characterize the pathogenesis of RVFV in the BALB/c mouse model [14], [15], which showed important similarities to severe human infections. Infection of BALB/c mice with RVFV subcutaneously (SC) resulted in high-titer viremia and demonstrated RVFV tropism for a variety of tissue and individual cell types on the basis of histopathology, immunohistochemistry, and electron microscopy. A major consequence of infection was overwhelming infection of hepatocytes that subsequently underwent apoptosis. Most mice succumbed to RVFV between days 3 and 6 post-infection (PI) which we attributed primarily to severe hepatitis as indicated by the overwhelming infection of hepatocytes and increase in high levels of hepatic enzymes in the blood. The remaining mice were able to effectively clear virus from the liver and blood, but exhibited neuroinvasion and developed lethal panencephalitis [14].
The route of neuroinvasion in our mouse model was unclear and we postulated one potential route may have occurred by way of the olfactory nerves leading to infection of the olfactory bulbs, which was based on our observation that olfactory neurons lining the nasal tract are a target of RVFV. We therefore hypothesized that the incidence and severity of central nervous system infection subsequent to aerosol infection may be significantly higher than is seen with a peripheral (SC) infection. RVFV is transmitted easily by aerosol to humans, which is evident by the number of laboratory workers who have become infected [16] and the potential for infection of veterinarians and abattoir workers who handle infected animals. Mice are known to be highly susceptible to RVFV aerosol exposure [17]. However, detailed studies describing the pathogenesis of RVFV after aerosol exposure have never been completed. Additionally, potential antivirals have not been evaluated for effectiveness following an aerosol exposure to RVFV.
Due to the possibility of infection by aerosol exposure (such as during a bioterrorism event), we have expanded our studies of pathogenesis to include mice exposed to a lethal dose of virus by aerosolization. Additionally, we evaluated the effectiveness of ribavirin, an antiviral compound with known activity against RVFV both in vitro and in vivo [18]–[21]. By comparing the course of disease after aerosol and SC RVFV exposures in the presence and absence of antiviral treatment, we determined the suitability of the mouse model in antiviral therapeutics trials and provided further insight into the efficacy of ribavirin to treat RVFV infection.
Research was conducted under an IACUC approved protocol in compliance with the Animal Welfare Act, PHS Policy, and other Federal statutes and regulations relating to animals and experiments involving animals. The facility where this research was conducted is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care, International and adheres to principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council, 2011.
Female BALB/c mice were obtained from the National Cancer Institute, Frederick Cancer Research and Development Center (Frederick, MD), and were used at 6–8 weeks old. Mice were housed in microisolator cages and were provided water and chow ad libitum.
Recombinant viral strain ZH501 was rescued as previously described [22], and the exact complete genome sequence was confirmed by techniques described previously by Bird et al. [23]. Strain ZH501 was originally isolated from a fatal human case during the 1977 epidemic in Egypt. In this study, the lethal dose at 50% (LD50) of this strain was determined in BALB/c mice infected SC or by aerosol. To determine the LD50, five cohorts of mice comprised of 10 mice each were infected with a range of doses (1000-0.1 PFU) and monitored for survival.
Prior to animal exposure studies, a sham aerosol spray using only the virus was performed in order to calculate a spray factor. The resulting spray factor was used to calculate the starting concentration of the virus necessary to obtain the target dose. Groups of ten unanesthetized mice were exposed to aerosolized RVFV created by a three-jet collision nebulizer (BGI Inc., Waltham, MA) for 10 min at a constant flow rate of 19 L/min in a whole-body exposure chamber housed within Class III biological safety cabinets in a biosafety level-3 suite by using automated aerosol exposure system. The particle sizes generated with this system has an average of 1–2 µm mass median aerodynamic diameter. Relative humidity was a steady state of 43–65% and temperature was ambient (approximately 20–22°C). Actual exposures received by each group of animals were determined by performing standard plaque assays on samples collected from an all-glass impinger (AGI; Ace Glass, Vineland, NJ). Complete medium with antifoam A 0.001% w/v (Sigma, St. Louis, MO) was used as collection medium in the impinger for titration by standard plaque assay. The dose was calculated using the following formulas: Dose = [Aerosol] (µg/mL)×minute volume (mL)×exposure time (min); minute volume = 2.1(weight (g))0.75.
Mice were exposed to a target dose of 1000 PFU of RVFV either by aerosolization as described above or by SC injection in a total volume of 100 µL. Two cohorts of mice were used for the pathogenesis study design. One cohort of mice (n = 100) were used to monitor survival and collect blood/tissue samples when the mice exhibited terminal (end-stage) signs of disease (these mice received an actual dose of 1600 PFU). The second cohort of mice (n = 200) was used to serially sample mice (randomly selected) on days 1–8 PI (these mice received an actual dose of 646 PFU). All mice were implanted with IPTT-300 temperature chips (Biomedic Data Systems, Seaford, DE) to identify individual animals and monitor body temperature throughout the studies. Temperature and signs of clinical disease were observed daily, and one group of mice from both the SC and aerosol exposure groups (n = 4/day PI) was selected for retroorbital blood sampling to analyze viremia, blood chemistry, and hematology. These mice were then perfused with PBS (to remove virus contaminated blood from tissues) and tissues were collected for virus titration by standard plaque assay and histology. Tissues collected for titration were weighed and homogenized in Eagle's minimal essential medium (EMEM) containing 5% fetal bovine serum and gentamicin. Tissues were homogenized using the Qiagen Tissue Lyser II (Qiagen, Valencia, CA) followed by centrifugation at 9,000× g for 10 min and the supernatant stored at −70°C until further evaluation by standard plaque assay to determine titers.
Mice were treated by intraperitoneal (IP) injection with PBS (n = 30; drug delivery vehicle control) or with ribavirin (n = 30; Sigma; 100 mg/kg) beginning two hours prior to virus exposure and continuing once a day thereafter for 9 days (10 days total). This dose was selected based on a previous study which determined that 100 mg/kg of ribavirin administered prophylactically was protective in mice and higher doses of ribavirin were found to be toxic [21]. Mice were infected either by aerosol (n = 10/treatment group) or SC exposure (n = 10/treatment group) with a target dose of 1000 PFU (actual dose was 1951 PFU) or remained as uninfected controls (n = 10/treatment group). All mice were monitored for changes in weight, temperature, and survival.
An additional cohort of mice was treated with PBS or ribavirin and infected either by aerosol (n = 15/treatment group) or SC (same as described above; n = 15/treatment group) for a comparative pathology study on days 3, 5, and 8 PI. These time-points were chosen based on previous pathogenesis studies that indicate that these are key days for RVFV infection in this model; and that they reflect early, middle, and late disease states [14]. Mice were bled retroorbitally for virology, hematology, and blood chemistry analyses (n = 5/day PI). Immediately following blood sampling and under deep anesthesia, mice were perfused with PBS to remove residual blood in tissues. Tissues were then collected for titration by standard plaque assay as described above.
Plaque assays for RVFV used 90–100% confluent Vero cells (American Type Culture Collection, Manassas, VA) in 12-well plates. Samples for titration were serially diluted 10-fold in culture medium, and then 100 µL of each dilution was added to each of two wells. Plates were rocked every 15 minutes during 1 hour of incubation at 37°C with 80% relative humidity and 5% CO2. After incubation, a primary overlay of 0.5% agarose and 5% fetal bovine serum in 1× Earle's minimum essential medium (EMEM) was added to each well. Plates were then incubated for an additional 3 days before addition of a secondary overlay, which contained 4% neutral red in addition to the formula of the primary overlay. After addition of neutral red stain, plates were incubated for an additional 24 hours, and plaques were counted 4 days PI.
Whole blood was added either to an EDTA tube (Safe-T-Fill, RAM Scientific Inc., Yonkers, NY) for complete blood count (CBC) determination using a Hemavet (Drew Scientific, Dallas, TX) or whole blood was added to a lithium heparin tube (Safe-T-Fill) for clinical chemistry analysis using the comprehensive diagnostic panel analyzed on a Vetscan (Abaxis, Union City, CA). The Hemavet measured the following parameters: white blood cells (WBC), total number and percent of neutrophils, lymphocytes, macrophages, eosinophils, and basophils, red blood cells (RBC), hemoglobin (Hb), hematocrit (HCT), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), platelets (PLT), and mean platelet volume (PLT). The vetscan determined levels of alanine aminotransferase (ALT), albumin (ALB), alkaline phosphatase (ALP), amylase (AMY), total calcium (CA++), creatinine (CRE), globulin (GLOB), glucose (GLU), phosphorus (PHOS), potassium (K+), sodium (NA+), total bilirubin (TBIL), total protein (TP), and urea nitrogen (BUN).
Mice were euthanized by perfusion under deep anesthesia and a complete necropsy was performed to collect a full complement of tissues for histopathology. Tissues were fixed in 10% neutral-buffered formalin for a minimum of 21 days, and then removed from biocontainment and processed for routine histopathology. Paraffin-embedded tissues were sectioned and then stained with hematoxylin and eosin (HE). Duplicate tissue sections were stained by immunohistochemistry for RVFV antigen using a kit (EnVision System, Dako Corp., Carpinteria, CA). Sections were deparaffinized and pretreated with proteinase K for 6 minutes before incubation with primary anti-RVFV antibody (R1547, rabbit polyclonal), diluted 1∶800 and incubated at room temperature for 30 minutes. This was followed by incubation with a peroxidase-conjugated, polymer-based secondary antibody. Tissues stained by IHC were examined microscopically to determine the types of cells labeled for viral antigen.
SAS version 9.1.3 (SAS Institute Inc., Cary, NC) was used to determine differences in the mean blood chemistry and hematology results between day −1 or 0 (uninfected controls) and subsequent time points using T-tests with step-down bootstrap adjustment. Signed rank tests with step-down Sidak adjustment was used for comparisons of temperature and weight from day 0 to subsequent time points within each study and Wilcoxon-Mann-Whitney tests for comparisons of temperature and weights between exposure routes (aerosol and SC) at each time point. A two-way ANOVA was used to compare viral titers between exposure groups.
Exposure to the ZH501 strain of RVFV was uniformly lethal by both SC and aerosol challenge routes. The LD50 was determined to be 0.27 PFU and 0.75 PFU for mice exposed SC and by aerosol, respectively (data not shown). When mice were exposed to 1600 PFU, signs of clinical illness become apparent on day 3 PI at which time the mice presented with ruffled fur and a hunched posture. The majority of mice in both groups were euthanized or succumbed to infection between days 3 and 4 PI, with 70% overall mortality from systemic disease by day 5 PI by both routes (Figure 1). The mean time-to-death (MTD) for mice exposed SC was 4.87 days and for mice exposed by aerosol 4.94 days, which was not significantly different.
Mice exposed by both challenge routes began to lose weight and have changes in their body temperature as early as day 3 PI (Figure 2). Mice exposed by aerosol began to lose significantly more weight (p = 0.0144) and have significantly lower body temperatures (p<0.0001) compared to SC exposed mice on day 4 PI. This trend continued until day 7 PI when mice exposed SC began to gain weight, which was followed by another decrease.
SC exposed mice that succumbed or were euthanized around day 7–9 PI exhibited signs of neurological disease such as hind limb paralysis. Aerosol exposed mice exhibited signs of neurological disease between days 6–8 PI. Therefore, based on disease signs, it appears that the percentage of mice exposed SC and by aerosol that succumbed from systemic disease was 74% and 70% respectively vs. 26% (SC exposure) and 30% (aerosol exposure) that succumbed from neurological disease. For mice exposed by aerosol, 100% mortality from neurological disease was observed 1 day earlier compared to SC exposed mice (day 8.5 PI vs. day 9.5 PI, respectively).
The blood chemistry and CBC values were determined in the uninfected controls (day 0 PI) and the infected mice on days 1–8 PI (Figure 3). Generally, both aerosol- and SC-infected mice exhibited marked changes in multiple analytes as early as 3 days PI. There were few significant differences in these results between exposure groups, and blood chemistry portrayed a similar picture of the disease courses overall. The liver enzymes ALT (p = 0.0102) and ALP (p = 0.2579) peaked on day 4 PI in SC exposed mice when compared to uninfected controls. In aerosol exposed mice the ALT (p = 0.0198) and ALP (p = 0.0009) levels peaked on day 5 PI (Figure 3A–B), which is one day later compared to SC exposed mice. However, the only statistically significant difference when directly comparing the two exposure routes was higher ALP levels on day 5 PI (p = 0.0102) for mice exposed by aerosol compared to SC exposed mice. A significant amount of variability was observed in mice from both exposure groups. This is most likely due to the study design where mice were randomly sampled and at different stages of RVF disease. Chemistry analyses from mice sampled only when they exhibited signs of end-stage disease were more consistent among animals (data not shown). Other analytes found to be significantly different when comparing mice exposed by aerosol vs. SC included TBIL, CA, PHOS, GLU, and K+ on various days PI (data not shown).
CBC results also showed markers of hematological dysfunction that were mostly conserved across both exposure groups. The most drastic was a drop in percent circulating lymphocytes relative to baseline values (paero = 0.0003, psc = 0.0541) with a concurrent spike in percent circulating neutrophils (paero = 0.0004, psc = 0.0368) 3 days PI in both exposure groups (Figure 3C–D). The most pronounced difference between the aerosol and SC exposure groups was a trend toward recovery of normal values after 3 days PI among SC-infected animals; while the aerosol exposed cohort maintained the same aberrant lymphocytopenia and neutrophilia established 3 days PI through the end of the study. However, the only statistically significant difference (for percent lymphocytes and neutrophils) when directly comparing the results from both exposure groups resulted on day 2 PI where mice exposed by aerosol had slightly higher levels of percent circulating neutrophils (p = 0.0135). This is most likely because significant variability was observed in each cohort of mice, which is similar to the chemistry results. Other hematology values found to be significantly different when comparing mice exposed by aerosol vs. SC included %MO, %BA, RBC, Hb, %HCT, MCV, MCH, MCHC, %RDW, and PLT on various days PI (data not shown).
The viral titers in the plasma, liver, lung, and brain were determined by standard plaque assay on days 1–5 and days 7–8 PI (Figure 4). Both SC and aerosol exposure routes resulted in viremia on day 2 PI, which peaked on day 3 PI for both exposure groups (Figure 4A). Virus was first detected in the liver (Figure 4B) and lung (Figure 4C) from aerosol exposed mice on day 1 PI whereas no virus was detected in the tissues of SC exposed mice on the same day. Virus was detected in all tissues acquired 3 days PI from mice in both exposure groups. Viral titers in the liver and lung were highest 3 days PI in both exposure groups, and then gradually decreased throughout the course of the study in both cases. Viral titers peaked in the brain on day 7 PI for mice exposed SC and on day 8 for mice exposed by aerosol (Figure 4D). Overall, no significant differences resulted in the viral titers in the plasma, liver, and lung from mice exposed either by aerosol or SC. However, mice exposed by aerosol did have significantly higher titers of virus in the brain compared to SC exposed mice on day 8 PI (p<0.001).
Histological examination of tissues from infected mice exposed by aerosol and SC showed tropism of RVFV for a variety of tissues and cell types throughout the course of infection similar to the results we previously described [14]. Viral antigen was present at various times in epithelial cells (hepatocytes, adrenocortical cells, odontogenic epithelium, olfactory neuroepithelial cells), mesenchymal cells (perineural cells, periosteal and endosteal cells, perivascular cells, bone marrow stromal cells, fibroblastic reticular cells, myocardial cells, vascular smooth muscle cells, adipocytes), neural cells (olfactory neurons and multiple neuronal types in the brain), hematopoietic cells (macrophages and other unidentified cells) and endocrine cells (pancreatic islet cells and pituicytes).
Regardless of the exposure route, only minimal viral antigen was detected in the lungs of mice exposed by aerosol and SC (Figure 5A–B, Table 1). In mice that were exposed by aerosol, scant, weak antigen localization was observed in the lungs of 2 of 4 mice on day 1 PI and in one of 4 mice on day 2 PI. The antigen observed may have been residual antigen deposited in the lungs as respirable sized particles from the aerosol. The mediastinal lymph nodes, which drain the lungs were available for microscopic examination in 2 of 4 mice on day 1 PI, but these regional lymph nodes did not have imunohistochemical evidence of viral infection. In both models, viral antigen localization was observed within intraalveolar cells, alveolar septal cells, and intravascular cells of the lungs. No histological evidence of pneumonia was observed in the lungs of mice exposed by aerosol or SC.
The liver was a clear early and dominant target in mice exposed by aerosol and SC. A few infected hepatocytes were evident as early as day 2 PI (3 of 4 mice for aerosol exposure and 2 of 4 for SC exposure; Table 1). Strong and widespread viral antigen staining of hepatocytes was apparent in livers of both aerosol- and SC-infected mice by day 4 PI (Figure 5C–D, Table 1), which is similar to our previous study results [14]. Also evident on days 3, 4, and 5 PI was damage to hepatocytes ranging from a few to numerous dead cells. Despite extensive infection of, and damage to hepatocytes, clearance of virus from the liver occurred to a great extent in mice that survived to later time points, as reduced amounts of viral antigen were evident by day 8 PI in mice exposed by aerosol and SC.
An early target of aerosolized RVFV was the olfactory neuroepithelium lining nasal turbinates, with a few immunolabeled cells present in 4 of 4 mice at day 2 PI and in 3 of 4 mice on day 3 PI (Figure 6, Table 1). Infected cell types included cells morphologically compatible with neuroepithelial cells and cells of the subjacent fila olfactoria. Similar to the liver, the olfactory neuroepithelium exhibited progressive infection in some mice at later time points. In contrast, the same tissues from animals exposed SC were uniformly immunonegative until 7 days PI at which point the nasal turbinates and olfactory neuroepithelium were positive for viral antigen.
The earliest evidence of neuroinvasion in mice exposed by aerosol was antigen localization limited to the olfactory bulbs, in one mouse on day 2 PI, in one mouse on day 4 PI, and in 3 of 4 mice on day 5 PI. In contrast, viral antigen was not detected in the same region until day 8 PI for mice exposed SC (Figure 7, Table 1). Histological evidence of infection in the brain proper was first apparent in neurons of the brainstem in 1 of 4 mice euthanized on day 5 PI that were exposed by aerosol. For mice exposed SC, histological evidence of infection of the brain occurred on day 6 PI in 1 of 4 mice. Neurons were the main target cell in the brain for both exposure routes. By day 7 PI, 4 of 4 mice had evidence of neuronal infection in the brainstem and other areas of the brain for mice exposed by aerosol. Less viral antigen was detected in the brain of mice exposed SC with 1 of 4 mice exhibiting neuronal infection on day 8 PI (Figure 8, Table 1). Morphologic changes in the brain developed concurrently with the detection of viral antigen in mice exposed by aerosol and SC. The brain lesions were generally characterized as meningoencephalitis with neuronal necrosis.
Cohorts of mice were treated with PBS as a control or ribavirin prophylactically and exposed to 1951 PFU of RVFV by aerosol or SC. As expected, all uninfected controls survived (Figure 9) and did not exhibit significant decreases in weight or temperature (Figure 10). All mice that were treated with the PBS control and exposed by aerosol or SC were euthanized or succumbed to infection and did have significant decreases in weight and temperature. MTD for PBS injection treatment control aerosol- and SC-exposed groups were 6 and 4.5 days, respectively. Mice that were treated with ribavirin and exposed SC to RVFV showed 70% survival with a MTD of 15 days and did not experience significant changes in weight or temperature. In contrast, ribavirin treated, aerosol-exposed mice showed 0% survival with a MTD of 9.5 days and had significant decreases in weight and temperature (Figure 9 and 10). Non-surviving treated animals exposed SC and by aerosol exhibited signs of neurological disease at the time of death or euthanasia.
Despite the variances in MTD and survival, there were no significant differences between SC- and aerosol-exposed mice in clinical blood chemistry and hematology values from blood samples acquired 3, 5, and 8 days PI (data not shown). Viral titers from tissues sampled at these same time points demonstrated more virus in tissues from ribavirin treated, aerosol exposed animals compared to SC exposed animals (Figure 11). Virus was only detected in the plasma, liver, and spleen of ribavirin treated, SC exposed mice. When tissues were assessed for viral antigen by IHC, ribavirin was partially effective at reducing the amount of viral antigen in the lungs and liver, but had only a delayed effect on the amount of viral antigen detected in the olfactory epithelium and olfactory bulb of aerosol exposed mice (Table 2). Our histology results also indicate that ribavirin treatment had little or no effect on CNS pathology where ribavirin treated vs. untreated aerosol exposed animals had similar degrees of pathological changes in the olfactory bulb and brain. This is in contrast to ribavirin treated, SC exposed mice, which had no evidence of pathological changes on day 8 PI (Table 3).
Because of the possibility of RVFV exposure by multiple routes, it is necessary to discern the differences in pathogenesis and treatment accordingly. In the current study, we first performed detailed comparative pathological analyses on mice infected by aerosol and SC exposure routes in order to determine the pathogenetic events unique to each. Additionally, we evaluated the efficacy of the nucleoside analog ribavirin by examining its effects on survival, weight, temperature, and viral load after either aerosol or SC exposure.
RVF disease in humans and livestock is often characterized by early-onset hepatitis and delayed-onset encephalitis, manifestations which are reproduced in the BALB/c RVFV mouse model exposed SC and by aerosol. Despite the route of exposure, no significant differences in overall survival were observed. For mice exposed by aerosol and SC, the liver was a clear early and dominant target of RVFV, with specific targeting of hepatocytes. Hepatic enzymes in the blood peaked one day earlier (day 4 PI) in SC exposed mice compared to aerosol exposed mice. This small difference may be due to the route of infection of the liver. However, this remains unclear because the primary site of replication for RVFV following SC and aerosol exposure has yet to be determined in the BALB/c mouse model. We speculate that mononuclear phagocytic cells and dendritic cells at the SC injection site would first become infected and migrate to the nearest draining lymph node. The virus would then replicate in the lymph node, resulting in primary viremia and spread to other target organs such as the liver via the bloodstream. This is supported in part by our earlier work that demonstrated viral antigen in mononuclear phagocytic cells and dendritic cells in the lymph nodes from infected mice [14]. Additionally, a study utilizing bioluminescent and fluorescent RVFV in immunodeficient mice demonstrated the importance phagocytic cells as targets for viral infection and following intradermal inoculation of virus, the nearest draining lymph node became the main site of early replication [24].
For mice exposed by aerosol, we observed weak immunohistochemical evidence of viral infection in the lungs, which was not a major target for infection despite the route of exposure. Similar to a previous study of aerosolized RVFV in Swiss Webster mice, pneumonia was not a dominant factor in the pathogenesis of RVFV in mice following aerosol exposure [17]. We speculate that alveolar macrophages might be an important early target cell for infection in the lungs and these cells would then migrate to the mediastinal lymph nodes prior to 1 day PI, which could possibly explain the weak immunohistochemical evidence of viral infection in the lungs. The mediastinal lymph nodes is most likely an important primary site for replication followed by spread to other target organs such as the liver, even though we failed to detect viral antigen by immunohistochemistry. Viral antigen could have been present at a level below the detection level for our assay. Other possible routes for infection of the liver include ingestion of virus with seeding of the liver by the enterohepatic circulation or by vascular spread from the lungs.
Viral titers and immunohistochemistry from brain tissues indicated that mice exposed to RVFV by aerosolization developed neuropathology more rapidly and to a greater effect than SC exposed mice. Early, intense immunohistochemical staining of the nasal turbinates and especially the olfactory bulb among mice in the aerosol exposure group may explain the significant development of neuropathology associated with aerosol exposure. Exposure by aerosolization results in direct exposure of the lung and the nasal turbinate epithelia. Therefore, RVFV viral neuroinvasion likely occurs by viral propagation in nasal turbinate epithelial cells; followed by infection of olfactory nerves, the olfactory bulb, and eventually the higher structures of the brain. This is in contrast to the SC exposure group, in which the less severe evidence of neuropathology and rarity of immunopositive cells in the nasal turbinates and olfactory bulb suggests another mechanism for viral entry into the brain. In a previous study of mice following SC exposure to RVFV, we showed that tissues of the spinal cord and then brainstem become progressively infected prior to onset of fatal pan-encephalitis [15], suggesting that entry to the brain may occur by cephalic progression from the neurons and neuroglia of the spinal cord into the basal structures of the brain. These differences in the neuropathogenesis of RVFV should be taken into account when developing medical countermeasures to protect against peripheral and aerosol exposures. Despite these important differences in the development of neuropathology, no significant difference exists in the overall survival of mice exposed by both exposure routes. This is most likely because mice are highly susceptible to RVFV regardless of the route of exposure. The overall disease outcome is similar with most mice developing acute-onset hepatitis and the remaining mice developing encephalitis. However, with such a highly susceptible model it is important to take into account the progression of disease (not just the disease outcome) when designing and evaluating therapeutics. This key point is highlighted by our evaluation of ribavirin in SC and aerosol exposed mice (discussed below) where we observed significant differences in the outcome.
A recent study by Bales et al. assessed the susceptibility of different inbred rat strains to aerosolized RVFV and determined that Wistar-Furth rats developed a similar disease course and outcome when exposed SC and by aerosol (although not directly compared in the same study), which is similar to our results in the BALB/c mouse model. This is most likely because these two models are highly susceptible to RVFV and it is hard to distinguish differences in survival when high mortality results in a short time frame. In contrast, Bales et al. report that Lewis rats developed fatal encephalitis after aerosol infection, but only mild disease following SC exposure. No pathology data was presented, but Bales et al. hypothesized that aerosolized RVFV may gain access to the central nervous system through the olfactory bulb, which is similar to our results in the mouse model. Interestingly, Lewis rats that survived SC exposure were not protected against subsequent re-challenge by aerosol exposure to the same virus [25]. This observation, along with our findings, implicate the importance for further studies to evaluate the route of neuroinvasion by RVFV, which has important implications for the development of countermeasures to protect against multiple exposure routes.
We observed aberrations in certain hematologic parameters, which were conserved across both exposure route groups. First, early and severe neutrophilia occurred in all mice in both exposure groups, with an 80% average increase in percent circulating neutrophils above baseline values observed 3 days PI. This sharp increase reflects the systemic nature of granulocytic inflammation during acute RVFV infection, and correlates well with histological observation of polymorphonuclear white blood cell infiltration into multiple organ systems; particularly secondary lymphoid and liver tissues 3 days PI. Neutrophilia and polymorphonuclear white blood cell inflammation are often associated with acute tissue damage during infectious diseases; such as viral pneumonia [26], [27] and septic shock [28]. The findings of severe neutrophilia and polymorphonuclear leukocyte tissue inflammation in the present study further implicate the importance of the host innate immune response in RVFV pathogenesis.
Ribavirin is a nucleoside analogue and its mechanism of antiviral action is not clear. Previous studies have demonstrated the protective efficacy of ribavirin (administered either prophylactically or post exposure) in mice for preventing the hepatic disease, but some mice were euthanized or succumbed from late-developing encephalitis [18], [20], [21], [29]. Similar to these studies, we observed that ribavirin was partially effective in preventing mortality among mice in the SC exposure group. The ribavirin treated mice lived an average of 10.5 days longer than the untreated control mice that were exposed SC and the mice that were euthanized or succumbed developed late-onset encephalitis. To the best of our knowledge, ribavirin has never been evaluated following an aerosol RVFV exposure. We have demonstrated that ribavirin was ineffective among the aerosol group and all mice succumbed from encephalitis. The ribavirin treated mice that were exposed by aerosol lived an average of 3.5 days longer than the untreated control mice. Our results suggest that ribavirin would be completely ineffective in cases where aerosol exposure occurs. This is most likely because of the limited ability of ribavirin to penetrate the blood-brain barrier. Intranasal or aerosol administration of ribavirin does increase the bioavailability of the drug to the brain [30], [31], which could prove more effective to increase survival.
The pathogenetic differences in RVF disease following exposure by either route offer important insights into the relative efficacy of ribavirin against RVFV infection. In particular, early and aggressive viral invasion of brain tissues following aerosol exposure likely played an important role in ribavirin's failure to prevent mortality among these animals. Tissue titers from animals exposed via aerosol showed early infection and rapid propagation in the brain throughout the study, indicating that virus was able to enter and multiply in cells of the brain unimpeded by ribavirin treatment.
Also of key importance was the lack of infectious virus in the liver of any ribavirin-treated mice in the aerosol exposure group at any time-point, and in the SC group after 3 days PI. This rapid elimination of virus from the liver suggests that early ribavirin treatment is efficacious in preventing the primary hepatic syndrome that correlates with early mortality in the mouse model. Serum viral titers also diminished rapidly and then disappeared in both groups after 5 days PI. This lends further support to the hypothesis that ribavirin only fails to protect mice from RVF viral panencephalitis, which was the cause of morbidity and mortality among mice succumbing to RVF viral challenge in spite of treatment in our current study.
Finally, the significant delay in MTD among mice succumbing to RVFV infection after SC exposure and ribavirin treatment suggests that infection may have resulted in lethal panencephalitis after viral neuroinvasion in nonsurviving mice. As in untreated animals, the spleen appeared to serve as a chronically infected viral reservoir, which may have eventually resulted in viral seeding of the basal brain structures via the spinal cord as previously described [14]. While we failed to recover infectious virus from brain tissue in the SC exposure group days 3, 5, or 8 PI; it is likely that direct infection of the brain occurred closer to the MTD, which was on 15 days PI. Viral titers in tissues of treated mice suggest the importance of early viral neuroinvasion in aerosol-exposed mice as a contributor to mortality. Despite a failure to detect infectious virus in plasma at the same time-point, brain tissues collected 8 days PI yielded average viral titers of approximately 8 log10 PFU/g. This is equivalent to the titer observed in brain tissue from untreated, terminally sampled animals; and therefore indicates that RVFV proliferation in the brain may be unaffected by ribavirin treatment.
Despite the proven activity of ribavirin against RVFV in rodent and non-human primate models [20], it has not been thoroughly evaluated for the treatment of RVFV infection in humans. During the Saudi Arabia outbreak in 2000, a small-scale, randomized, placebo-controlled clinical trial evaluating ribavirin to treat severe RVFV infections was conducted [32]. Although the study was inconclusive, results suggested that an increased incidence of the encephalitic form of the disease was observed in cases treated with ribavirin (P. Rollin, presented at the Treatment of Viral Hemorrhagic Fever Workshop, Bethesda, MD, 24–27 February 2007). This study and our results highlight the need for more candidate antivirals to treat RVFV infection, especially in the case of a potential aerosol exposure.
In conclusion, our results indicate the utility of the mouse model of RVFV infection as a test system for antiviral therapeutics efficacy. They also highlight the differences in pathogenesis following exposure to either aerosolized or SC-injected RVFV; as well as the effect of exposure route on outcome of disease. Therefore, future evaluation of vaccines and antivirals should be evaluated following multiple exposure routes.
|
10.1371/journal.pntd.0005755 | "We need people to collaborate together against this disease": A qualitative exploration of perceptions of dengue fever control in caregivers' of children under 5 years, in the Peruvian Amazon | Dengue Fever presents a significant and growing burden of disease to endemic countries, where children are at particular risk. Worldwide, no effective anti-viral treatment has been identified, thus vector control is key for disease prevention, particularly in Peru where no vaccine is currently available. This qualitative study aimed to explore the perceptions of dengue control in caregivers’ of children under 5 years in Peru, to help direct future mosquito control programmes and strategy.
Eighteen semi-structured interviews were conducted in one health centre in Iquitos, Peru. Interviews were audio-recorded, transcribed and translated by an independent translator. Data were analysed using an inductive thematic approach.
Three core analytic themes were interpreted: (1) awareness of dengue and its control, (2) perceived susceptibility of children, rural riverside communities and city inhabitants, and (3) perceived responsibility of vector control. Participants were aware of dengue symptoms, transmission and larvae eradication strategies. Misconceptions about the day-time biting behaviour of the Aedes aegypti mosquito and confusion with other mosquito-borne diseases influenced preventative practice. Community-wide lack of cooperation was recognised as a key barrier. This was strengthened by attitudes that the government or health centre were responsible for dengue control and a belief that the disease cannot be prevented through individual actions. Participants felt powerless to prevent dengue due to assumed inevitability of infection and lack of faith in preventative practices. However, children and rural communities were believed to be most vulnerable.
Perceptions of dengue control amongst caregivers to under 5’s were important in shaping their likelihood to participate in preventative practices. There is a need to address the perceived lack of community cooperation through strategies creating a sense of ownership of community control and enhancing social responsibility. The belief that dengue cannot be prevented by individual actions in a community also warrants attention. Specific misconceptions about dengue should be addressed through the community health worker system and further research directed to identify the needs of certain vulnerable groups.
| Dengue fever is the most rapidly spreading mosquito-borne viral disease in the world and is a significant international health problem. It is endemic in the Peruvian Amazon and since there is no available vaccine or effective treatment, mosquito control is key. This novel study used qualitative interviews to explore the perceptions and experiences of dengue control in caregivers’ to children under 5 years, to help inform future Peruvian mosquito control programmes. Findings suggested that caregivers had basic knowledge about dengue, but that misconceptions around the mosquito itself and confusion with other diseases (like malaria) influenced people’s protective practice. Furthermore, participants felt powerless to prevent dengue since the invisible mosquito, perceived ineffective prevention methods, and rural riverside environment made the disease seem inevitable. Despite this, children and rural riverside communities were identified as most vulnerable groups. Key findings suggested a lack of community cooperation in mosquito prevention due to an attitude that the government and health centre were responsible for disease control, and the belief that dengue cannot be prevented through individual actions alone. These were important barriers to community mosquito prevention efforts. The findings from this study provide novel insights into how perceptions of dengue control can impact prevention in an endemic country.
| Dengue fever (DF) remains the most rapidly spreading mosquito-borne viral disease in the world and it has caused more human morbidity and mortality than any other arbo-viral infection [1,2]. It currently infects 390 million people each year [3]. Dengue fever has presented an acutely difficult and progressive public health challenge in Latin America, as it has increased in prevalence, severity and spread throughout the region [4,5]. In Peru, over half the population live in areas at risk of the disease since it re-emerged in 1984 after 30 years of successful vector eradication [6]. This re-emergence has been attributed to: rapid growth and migration of urban populations, and limited resources directed towards dengue control [6]. Climate change, extreme poverty and poor water supply leading to household water storage have further compounded these effects by increasing the habitation sites of the mosquito, which now populates 17 of the 25 departments in Peru [6].
Dengue viruses are transmitted by female mosquitoes of the Aedes genus which are wide-spread in tropics and sub-tropics [7, 8, 9]. Dengue has been a particular health threat to children living in endemic areas [10] since young children are particularly susceptible to dengue haemorrhagic fever [1,11]. Ninety percent of cases of DHF occur in infants under 5 years of age, where there is a 2.5% risk of fatality [12]. The situation regarding a vaccine against dengue remains complex. Although a vaccine has undergone Phase III clinical trials and is recommended for use in endemic areas, it is not registered for use in young children [13]. Since no effective anti-viral treatment exists globally, and no vaccine is currently available in Peru, mosquito control is the foundation of dengue prevention nationwide [4,14]. Control strategies against the Aedes aegypti vector may also have the combined effect of enhancing protection from other diseases transmitted by the mosquitoes, such as zika and chikungunya. In dengue-endemic countries, control focuses on community and household mosquito eradication strategies. In the Peruvian Amazon, the city of Iquitos is endemic for DF and experiences frequent outbreaks [15]. The inaccessibility of drinking water in homes leads to the use of water storage containers [16], which provide a breeding site for the Aedes aegypti mosquito. In Iquitos, health promotion messages are broadcasted via radio, television and roadside billboards. These predominantly focus on symptom recognition and household eradication of breeding sites through cleaning the house and care or removal of containers that collect water (buckets, tubs, tyres for example) [17]. The local health authority further streamlines control strategies through case-by-case fumigation, larvicide water treatments, home inspections and community education [6].
According to the World Health Organisation, dengue control needs improvement to alleviate its significant burden of disease to endemic countries [18]. Further research is required to target dengue control in areas at risk. To date, few qualitative studies have explored perceptions and experiences around dengue control worldwide, but these have identified certain issues such as misconceptions, confusion with other febrile diseases, the “invisibility” of dengue and lack of responsibility in control methods [19–25]. In Iquitos, dengue-related research has concentrated on studies of epidemiology and entomology [26–28]. A recent survey evaluated community knowledge and practices [29], highlighting the misconception that Aedes aegypti bite during the night-time and the subsequent incorrect use of mosquito bed nets. Although informative, the nature of the survey did not allow researchers to explore in depth why misunderstandings exist or how preventative practices are influenced by perceptions. A recent qualitative study performed in Northern Peru and one Peruvian focus group study included within a WHO report found that people may consider vector control to be a lot of work, with reservations about the efficacy of interventions. [17, 30] However, as far as the authors are aware, there have been no other published qualitative studies in Peru which have explored this topic [31], highlighting an important gap in the knowledge base, particularly in this region [32,33].
This novel qualitative study sought to better understand the interrelated social and individual factors that determine perceptions, experiences and practices of dengue control, in caregivers of children under 5 years (a term which describes biological parents, step/foster/adoptive parents, wider family members and will be used from this point forward) to an at-risk group. Further understanding of caregivers’ thoughts and actions around DF can help to inform and direct future mosquito control programmes and strategy in the Peruvian Amazon.
Ethical approval was gained from the University of Birmingham Internal Ethics Review Committee (Ref: 2015-6/C1/DK/11) and the Institutional Ethics Committee of Research at the Department of Health, Loreto (Constancia No: 001-CIEI-DRSL-2016).
This pragmatic exploratory study was undertaken in the San Juan health centre in Iquitos, north-eastern Peru. The health centre serves 32,848 people in the southern district of the city and offers free state-funded health consultations under the Seguro Integral de Salud (SIS) system for those who qualify as either ‘poor’ or ‘very poor’. Residents who do not qualify for support are still able to access the health centre but a fee is incurred.
The power of a purposive sample lies in the intentional selection of participants defined as ‘information-rich’ [34]. This embraces the importance of individual views to best answer a research question concerning personal perceptions and practices [34]. Participants were purposively selected based on the following eligibility criteria: (i) caregiver to a child under 5 years of age, (ii) aged over 18 years and (iii) living in Iquitos. Following each interview, participant demographics were entered into a database to facilitate exploration of which characteristics (differing in age, gender, level of education and dengue experience) were missing from the sample. Cases with these characteristics were then purposively sought to ensure a maximum variation sample [35].
Recruitment was undertaken during the ‘wet season’ in February 2016, where the Aedes aegypti vector may be more abundant due to higher levels of rainfall [36]. Eligible participants were identified by clinicians and they were then approached by lead author (AF) with support from an interpreter who explained the study and provided participant information sheets. The reason for participant attendance at the health centre was not recorded. It was made clear that participation was voluntary and would not affect the care participants received at the health centre.
Semi-structured interviews followed a topic guide developed using existing, albeit limited, literature [19–25] and discussion within the research team (S1 Topic Guide). The first two interviews were deemed pilots, due to difficulties with the topic guide and translational discrepancies. After this, the topic guide was revised and translation difficulties addressed before the final eighteen interviews. An iterative process allowed inclusion of novel ideas, until no new concepts were presented and data saturation was achieved [37].
Interviews were conducted in English and Spanish using real-time interpretation [38] between the researcher, interpreter and participant. Two interpreters were used throughout the data collection; the first interpreter was present for five interviews and the second present for 15. Both interpreters were English language students at the university in Iquitos, with experience interpreting in clinical environments. Although acknowledged that use of one interpreter is preferable, efforts were made to ensure the similarity between interviews by briefing the second interpreter on the issues arisen from the pilot interviews and thoroughly practising interview questions.
Informed written consent was obtained before each interview which took place in a private room in the health centre. Interviews were audio-recorded and field notes taken. The spoken Spanish was transcribed into English by the researcher and a translator who had experience translating medical projects. Cultural context and the meaning behind colloquialisms were discussed with the researcher to support analysis. A translation lexicon was developed to establish conceptual equivalence [39]. An independent translator reviewed a random sample of two transcripts to assess the quality of translation. Minimal discrepancies between translations were noted, thus no changes were made.
Data were analysed using Braun and Clarke’s [40] inductive thematic analysis method to identify and interpret themes. An iterative approach was employed, where a constant comparison method ensured data analysis was undertaken in parallel with data collection [41]. Immersion in the data was ensured through repeated reading and familiarisation of the transcripts. Initial codes and themes were generated and a coding rationale produced, with support of NVivo software. Thematic mind maps were developed and themes reviewed using quotes from the text and interviews as a whole.
To improve the credibility of analysis, five of the most data rich transcripts were independently coded by a second analyst (EB). Differences and similarities were evaluated and additional interpretations incorporated into the coding framework. Themes were discussed between the analysts and further reviewed by the research team (LJ, GW) which sought to enrich interpretations and reduce bias presented by a single analyst [42]. A reflexive approach to analysis was employed, where notes were made throughout the analytical process to acknowledge the potential impact of research bias [43]. Continued communication between the lead author (AF) and Peruvian clinicians (GMS) clarified cultural misunderstandings and minimised the risk of cross-cultural bias in interpretation and analysis [39].
Eighteen interviews formed the final analysis dataset, of which equal numbers were male and female. The average age of the sample was 32 years (range 21 to 56 years). Seven participants reported having suffered DF themselves. A demographic summary of the participants is presented in Table 1. The average length of interviews was 36 minutes (range 24–51 minutes).
Three core analytic themes and eight subthemes (Fig 1) were interpreted within the data: (1) awareness of dengue and its control, (2) perceived susceptibility of children, rural riverside communities and city inhabitants and (3) perceived responsibility of vector control.
The overarching theme of responsibility runs through each narrative and exists in three interlinked but separate dimensions: (i) external responsibility, (ii) community responsibility, (iii) personal responsibility.
This pragmatic qualitative study explored the perceptions of the control of dengue fever with 18 Peruvian Amazon caregivers to children under 5 years. Findings highlighted the importance of these perceptions in shaping individual preventative practice in a community. Key findings include the impact of misconceptions around dengue transmission and the importance of attitude within a community for effective preventative practice.
This study found that participants had good basic knowledge of DF symptoms, transmission and eradication of mosquito larvae as preventative practices. In Iquitos, a recent survey found that 65% of people knew someone who had experienced dengue [29]. The current study builds upon this evidence by suggesting explanations for this basic knowledge acquisition such that experience of the disease, information distributed by media campaigns and the existing community health worker system may have influenced and improved knowledge. However, most participants were unaware that dengue mosquitoes bit during the day, which agrees with research by Paz-Soldan et al. that identified only 18.6% of respondents recognised the Aedes aegypti as a day-biting mosquito and most chose mosquito nets use at night as their main protective practice [29]. The present study further explored the link between these misunderstandings and seeks to partly explain this through a conflation of the information regarding dengue and other mosquito-borne diseases such as malaria, whereby people seem to see these diseases as one combined arbo-viral infection. Perhaps, previous dengue control messages in Iquitos have not provided sufficient information differentiating the two mosquitoes and subsequent diseases. Thus many participants may have perceived malaria and dengue as indifferent, which may have effected their preventative practices. For instance, although effective against malaria prevention, the WHO does not advocate bed nets to prevent dengue fever unless individuals sleep during the day [44]. Thus the specific misconception that mosquitoes transmitting dengue are night-biters directly hindered peoples’ abilities to protect themselves at the correct time.
This study can help to inform health education strategies to be directed towards individuals and communities to build upon basic knowledge, to target misconceptions and make recommendations for successful day-time protection from mosquitoes (such as using household fumigation and insect repellent). Perhaps, health promotion materials need to place more emphasis on distinguishing between dengue and malaria and the biting habits of their associated vectors to better educate and ultimately help the population to protect the themselves. Regular appraisal of the success of these strategies is required to identify and address future misunderstandings. Findings also recommend an evaluation to appraise how the community health worker system influences knowledge and preventative practice.
Furthermore, study findings implied that a hierarchy of risk was perceived by participants. Children and rural riverside communities were felt to be the most vulnerable to DF. A perception and subsequent desire to protect children was echoed in the participants’ responses. It should be acknowledged that these views may be specific to caregivers of children under 5 years, the target group of the study, as reported beliefs could have been affected by social desirability [45]. Dengue control campaigns may wish to target this group by focusing on child protection to incentivise families to participate in preventative strategies. Likewise, rural riverside communities were highlighted as vulnerable groups. In the Loreto region, evidence suggests significantly poorer health outcomes and access to health services in rural communities [46]. Participants proposed differences in mentality between people in rural communities and city as a consequence. A study in the Peruvian Amazon identified a reduced dependence on western medicine in rural communities [47], therefore it may be that these rural communities are forced to take more responsibility for their own health with less access to state provided services. This suggests an area warranting future qualitative research since the health beliefs of these people may require independent dengue control strategies.
Participants viewed themselves as susceptible to dengue since there was a belief that they were unable to protect themselves or their families from this disease. The assumed inevitability of infection, lack of faith in preventative practices and a powerless attitude towards protection compounded this belief. These results are supported by an existing health promotion theory: the Health Belief Model [48]. This model explains how health behaviour is determined by perceptions of a disease and the strategies used to reduce it and has been used an effective tool in previous published research surrounding dengue control [49–52]. In particular, it suggests the impact of perceived susceptibility and self-efficacy as factors influencing health behaviour [48]. In Iquitos, it may be appropriate to suggest that the high perceived susceptibility of dengue positively impacted participants’ preventative practice, whereas a low sense of self efficacy in response to disease prevention may have had the opposite effect. Research in Peru has suggested that a barrier to successful preventative practices may be a perception that control methods like household cleaning are seen as a lot of work. [17] Therefore, acknowledgement that these perceptions shape health behaviour is important for targeting effective health promotion materials.
A heavy reliance on the government or health centre for health protection and education was further illustrated in the study findings and showed parallels with a common concept in healthcare—paternalism [53]. This describes the act of seeking to promote someone’s wellbeing by interfering with their freedom of choice [54]. In some parts of the world, dengue control programmes have been described as “too paternalistic”, focusing on an attitude where the ‘government controls everything’ [55]. In Iquitos, the community health worker “charlas”, mandatory fumigation and household inspections could be viewed in a similar way. Although these services have been linked to a reduction in Aedes aegypti indices [56], paternalistic approaches may have the effect of disempowering communities and reinforcing feelings of powerlessness [57]. In turn, disempowered communities may be more susceptible to disease [57].
In this way, the combination of reliance on the government and health centre and a belief that dengue cannot be prevented through individual actions highlights a particular problem with the attitudes of the community in Iquitos, in relation to dengue control. These attitudes highlight a challenge to dengue prevention on an individual level, as well as encouraging a lack of participation and cooperation on a community level and vice versa.
In endemic countries, community participation is vital to ensure the success, suitability and sustainability of dengue control programmes [57]. In this study, a key finding illustrated the frustration that participants felt towards a perceived lack of community participation. Similar findings have been shown in other studies in Latin America where the concept has been described as “desunión” or “mala unión”. In English this translates to ‘disunion’ and represents a lack of community cooperation in dengue control practices [22,58]. In Ecuador, apathy and social irresponsibility were defined as barriers to community participation [22], the results of which are consistent with the study findings. Although the current study offers explanations behind this social phenomenon, additional qualitative research could add insights and clarity to build upon these findings, perhaps by using a more heterogeneous sample or a focus group design.
Evidence suggests that social mobilization and community efforts are crucial for maintaining dengue prevention [59]. In Iquitos, a lack of community empowerment has been suggested as a barrier to vector control strategies [17]. Thus, there is a need to foster strategies to promote community organisation and empowerment to engage residents in protecting the health of their families and the community as a whole. Successful community-based strategies have been used to reduce the abundance of Aedes aegypti in countries such as Cuba, Colombia, Honduras and Mexico [60–63]. For instance in Cuba, ‘community working groups’ were created using existing formal and informal community leaders and took responsibility for the management of vector control strategies in their own constituencies [60]. Other novel approaches have been used such as involving high school students as health educators in Colombia [62] and school-based education in Honduras [63]. Similar strategies could be utilised in Iquitos to encourage a sense of empowerment with campaigns promoting united participation in communities and emphasising the importance of each individual in working towards control of dengue fever. In this way, recommendations may focus on community led groups organising control efforts, perhaps through dissemination of educational materials or promotion of larvae eradication strategies (the Untadita method of cleaning wash basins, for example) [64]. This may prove pertinent since evidence suggests that people in Peru may not be well informed about effective cleaning methods to be able to effectively eliminate the Aedes aegypti vector from their homes. [17] Such recommendations seek to improve control of the Aedes aegypti vector to not only enhance dengue prevention but protection from other mosquito borne diseases such as zika and chikungunya. Thus, the results from this study may have a wider importance for the public health of this community.
It is important to acknowledge the strengths and limitations of the study. The strength of this study lies with the important feedback and insight it provides in the ongoing control of dengue fever and the influence of current health campaigns to community and individual perceptions in an endemic region. However, the qualitative nature requires findings to be interpreted with caution, particularly if generalising the conclusions beyond the sample explored. Qualitative studies are limited by the use of a relatively small number of participants and cannot be considered representative [34]. It is important to consider this when applying the findings to caregivers to children under 5 years who live in similar endemic situations.
Furthermore, in cross-language qualitative research, the use of translators and interpreters may impact the findings [39]. Braun and Clarke emphasise the importance of deriving meaning from words and data [65], therefore use of an interpreter and translator may have impacted analysis by increasing the risk that meaning was lost in translation. Steps were taken to avoid this potential bias of misinterpretation [39]. Interviews were piloted, a translation lexicon was created and two transcripts were externally validated. Moreover, the benefits of a local interpreter included useful insights into the meaning behind responses and increased cooperation from participants who seemed more relaxed and willing to participate in in depth discussion with a member of their community.
In retrospect, a further limitation to the study may be the absence of deeper exploration into the conflation presented in the interviews between dengue and other arbo-viral diseases. Further questioning about the different types of mosquitoes known by participants, as well as distinct control methods, may have proved useful when discussing appropriate recommendations to improve vector control. Likewise, the findings from this study would have been further enriched by closer examination of the individual motivators or barriers to performing protective household practices. Further research may aim to better explore these two limitations.
Finally, the impact of the lead author (British, female medical student) conducting cross-cultural research must be acknowledged. Culture is defined as a set of distinctive features of a social group [66] thus, as an outsider to the community, preconceptions and values of the principal researcher may have influenced data interpretation and analysis. The potential effect of this was minimised through support with interpretation from the wider research team, to ensure cultural understanding and the use of multiple coders and analysts to reduce the risk of researcher bias.
This study contributes to the understanding of perceptions and experiences that influence dengue control amongst caregivers of children under 5 years in an endemic setting. Specific misconceptions around dengue transmission and confusion with other diseases, such as malaria, need to be addressed. In terms of disease control, the lack of community cooperation in mosquito prevention was perceived as a key barrier. This social phenomenon was intensified by attitudes that the government and health centre were responsible for dengue control and the belief that the disease cannot be prevented through individual actions. There is a need to address these perceptions through targeted campaigns encouraging individual and community empowerment in dengue control.
|
10.1371/journal.pbio.1000560 | Connecting Variability in Global Transcription Rate to Mitochondrial Variability | Populations of genetically identical eukaryotic cells show significant cell-to-cell variability in gene expression. However, we lack a good understanding of the origins of this variation. We have found marked cell-to-cell variability in average cellular rates of transcription. We also found marked cell-to-cell variability in the amount of cellular mitochondrial mass. We undertook fusion studies that suggested that variability in transcription rate depends on small diffusible factors. Following this, in vitro studies showed that transcription rate has a sensitive dependence on [ATP] but not on the concentration of other nucleotide triphosphates (NTPs). Further experiments that perturbed populations by changing nutrient levels and available [ATP] suggested this connection holds in vivo. We found evidence that cells with higher mitochondrial mass, or higher total membrane potential, have a faster rate of transcription per unit volume of nuclear material. We also found evidence that transcription rate variability is substantially modulated by the presence of anti- or prooxidants. Daughter studies showed that a cause of variability in mitochondrial content is apparently stochastic segregation of mitochondria at division. We conclude by noting that daughters that stochastically inherit a lower mitochondrial mass than their sisters have relatively longer cell cycles. Our findings reveal a link between variability in energy metabolism and variability in transcription rate.
| Though pairs of cells may have identical genes, they still show behavioural differences. These cell-to-cell differences may arise from variations in how genes are transcribed and translated by the cellular machinery. Identifying the origins of this variation is important as it helps us understand why genetically identical cells can show a range of responses to the environment. In this work, we measured the rate at which the genes yield transcripts in cultured human cells. We found marked cell-to-cell variability in average rates of transcription. This variability is related to mitochondrial content. Cells with a higher mitochondrial mass have a faster rate of transcription, and we show that part of this variability is due to the unequal distribution of mitochondria to daughter cells when cells divide. Additionally, we find that cells that inherit more mitochondria divide earlier. These findings make a connection between variability in transcript production and variability in cellular mitochondrial content.
| Genetically identical populations of cells can exhibit cell-to-cell variations in the amount of individual gene products; this can result in phenotypic diversity [1],[2]. The study of cellular variability was pioneered by Delbrück in the mid-forties, who measured differences in the number of phages produced by individual Escherichia coli [3]. Fluctuations in the small numbers of molecules involved in gene expression have been indicated as a source of this variation, and current experimental and theoretical approaches seek to anatomize the potential sources of variability, or “noise”. Variation between cells could be due to global factors such as cell cycle position or differences in numbers of transcription factors. Such changes can affect all genes and so constitute “extrinsic” sources of variability. In contrast, “intrinsic” noise is identified as molecular variation that occurs at the level of single genes and their products [4]. Cell-to-cell variability could be mainly the combined effect of large amounts of intrinsic variation or might be attributable to more system-wide extrinsic variation. In the following we investigate how global factors can influence transcription rate across the eukaryotic cell.
Experiments investigating gene expression noise suggest that gene expression variability has a mix of intrinsic and extrinsic sources [5],[6]. Intrinsic noise has been modelled extensively, and we have a relatively refined idea of its origin in the molecular machinery of transcription, translation, and degradation [1],[2],[7],[8]. The magnitude of extrinsic noise is largest at intermediate levels of gene expression and dominates when gene expression is high [6],[7],[9]. However, the sources of extrinsic noise are not as well characterised as those of intrinsic noise [7],[10]. Studies carried out in yeast have, for example, suggested cell size, cell shape, cell cycle stage, and fluctuations in an as yet unidentified upstream regulator as potential sources of extrinsic noise [9],[11]–[13]. While there has been discussion of variability in the process of transcription both in polymerase binding and in transcription elongation, e.g., [14]–[16], this is often with the principal aim of understanding intrinsic noise: in the following we will investigate how extrinsic factors might modulate transcription rate.
To investigate the origins of global variability in eukaryotic gene expression we undertook a study of global transcription rate. We define global transcription rate as the average rate of production of transcripts within the nucleus of a single cell. Our results, obtained using direct measurement approaches, demonstrate that there is marked cell-to-cell variability in global transcription rate. The elongation rate of RNA polymerase II (RNA pol II) is a likely determinant of transcriptional rate, and we demonstrate that RNA pol II elongation is very sensitive to ATP concentration. We find that differences in [ATP] between cells relate to the transcription rate variability observed. We further find that the amount of mitochondrial mass and total membrane potential (indicated by total cellular luminescence of CMXRos dye) both correlate with transcription rate. Finally, we find that there is pronounced variability in mitochondrial mass in cellular populations and that a source of this variability is asymmetric segregation of mitochondria during mitosis. In combination these findings suggest that variability in mitochondrial content represents a likely source of global variability in transcription rate in eukaryotic cells.
We directly measured transcription rate by recording levels of bromouridine (BrU) incorporated into nascent RNA [17],[18]. The intensity of the BrU signal in RNA containing BrU (Br-RNA) rises with time, reaching a plateau after 1 h of incubation (Figure S1), when equilibrium between synthesis and degradation is reached. In these experiments BrU levels were analysed on confocal sections, providing a measure of the transcription rate per unit of nuclear volume. After a short pulse of BrU (30 min) the amount of Br-RNA produced by different cells (Hela) varied dramatically across the cell population (Figures 1A and S1A). This variation in BrU incorporation per unit of nuclear volume was not limited to Hela cells, but was observed in other established mammalian cell lines (murine erythroleukemia cells and Chinese hamster ovary cells), immortalised cultures (EBV-transformed lymphoblasts and mouse embryonic stem cells), and, importantly, in primary cells (lymphocytes and primary human fibroblasts) with coefficients of variation (CVs) ranging from 0.3 to 0.6 (data not shown; CV is the standard deviation of the data points divided by their mean).
Many factors could possibly contribute to variability in BrU incorporation. One source of variation could be differences in staging between cells in a population [9],[12]. We observed that the variability in total nuclear BrU incorporation remains substantial throughout the cell cycle, from a CV of 0.36 in G1 to 0.35 in G2 (Figure S1D and S1E), and thus, in agreement with previous studies performed in yeast [11], the cell cycle was ruled out as a principal source of variability in BrU incorporation.
Another source of variability in BrU incorporation could be differences in the number of active molecules of RNA pol II between individual cells. Therefore, we estimated the number of active RNA pol II molecules in different cell types, using run-on experiments [18]–[20] (see Figure S2A and S2B). The results suggest that the amount of active RNA pol II molecules was approximately constant per unit of nuclear volume in a given population and across the different cell types analysed. This suggests that the variation observed in BrU incorporation is not due to differences in the number of active RNA pol II molecules between individual cells.
We next asked whether variability in transcription rate by RNA pol II could account for the differences in BrU incorporation observed. The transcription cycle by RNA pol II can be understood as follows: free RNA pol II molecules interact with DNA, making a complex that can either be abortive (binding to DNA and not transcribing, or transcribing a very short transcript) or that can proceed into elongation mode after being modified. Once RNA pol II elongating molecules finish the transcription cycle, they become free and diffuse throughout the nucleoplasm. This simple model thus involves steps with different kinetic properties, which we exploited to gain insight into the rate of transcription of RNA pol II in single cells. We generated a cell line (C23) in which a GFP-tagged version of wild-type RNA pol II was introduced into Chinese hamster ovary cells containing a temperature-sensitive mutation in the largest catalytic subunit of RNA pol II (tsTM4). At the restrictive temperature, only the wild-type GFP–RNA pol II was functional [21], complementing the endogenous RNA pol II mutant (tsTM4) and thereby enabling the mutant Chinese hamster ovary cells to grow normally [22]. We performed fluorescence loss in photobleaching (FLIP) analysis of the wild-type GFP–RNA pol II (Figure 1B–1D) and obtained Koff values for RNA pol II consistent with the presence of at least two populations of RNA pol II molecules (Figures 1D and S3), as has been previously suggested: one freely diffusible (short half-life), and another associated with the DNA (long half-life) [21],[23]. When we analysed the Koff data from individual cells of the DNA-associated population (long half-life), we found a huge variation between cells (Figure 1E), with a half-life (t1/2) ranging from 2.5 to 30 min with a mean of 10.1 min and a standard deviation of 4.9 min, suggesting the existence of significant variation in rates of transcription elongation. The FLIP analysis exhibited a CV of 0.49, comparable with the variation in BrU incorporation we observed in this cell type (CV = 0.46). To assess whether the differences in Br-RNA between cells correspond to variation in transcription rate, we performed FLIP analysis in a group of C23 cells, followed by BrU incorporation. This experiment showed a strong relationship between DNA-associated RNA pol II t1/2 and Br-RNA production (Figure 1F). The faster the RNA pol II was dissociating from the DNA, the more Br-RNA was produced, supporting the suggestion that variability in the rate of DNA dissociation was coupled to variability in the rate of transcript elongation.
Transcription through intact chromatin involves the removal of histone H2B in order to destabilise the nucleosome [24], and consequently the dynamic properties of histone H2B reflect transcription elongation rate [25]. We therefore analysed the rate of exchange of fluorescently tagged histone H2B as a complementary approach to assess RNA pol II elongation rates in individual cells. Half of the nuclei of Hela cells expressing histone H2B–GFP were photobleached, and the decay of the signal in the unbleached halves was analysed. H2B–GFP showed a bi-exponential decay with a short t1/2 population that exchanges in a transcription-dependent manner (Figure S4) (∼7% of the histone H2B–GFP). The t1/2 of the fast-turnover histone H2B–GFP showed a wide range of values, consistent with different cells transcribing at different speeds (Figure 1G and 1H). This was again corroborated by experiments where cells were incubated with BrU after photobleaching, showing a good relationship between H2B (t1/2) and Br-RNA production (Figure 1I). As in the case of RNA pol II, the more dynamic the exchange of H2B, the more Br-RNA was produced, and vice versa. Taken together, these results provide good evidence that transcription elongation varies significantly between different individual cells within an otherwise homogenous population.
Next, we asked whether all the elongating RNA pol II molecules in a given cell were transcribing at a similar speed. In order to analyse only the nascent transcripts we limited the BrU pulse to 15 min and immediately “froze” cells with sarkosyl [18]. We measured the intensity of multiple individual Br-RNA foci within each nucleus (Figure 1J and 1K). We plotted the CV of the intensity of these nascent transcripts (Br-RNA foci) versus the mean intensity of these foci in the same cell, and carried out the analysis in cells exhibiting different amounts of transcription (Figure 1L). The data show scant change in the CV, consistent with all the polymerases that share the same nucleus transcribing at similar speed. There, thus, appears to be a global factor coupling the transcription rates of all foci across the nucleus (the variability in the rate of expression between different foci in the same nucleus is independent of the average rate of expression in the nucleus).
To summarise, we found a marked variability in the levels of steady state incorporation of BrU in genetically identical populations (Figures 1A and S1A) (and this appears to be independent of cell cycle position; Figure S1E). We then investigated the connection with transcription rate. The RNA pol II experiments suggested that there was marked cell-to-cell variability in the rate of dissociation of RNA pol II from the DNA (even though run-on data suggested that the amount of associated RNA pol II is relatively constant between cells; Figure S2A and S2B). The H2B–GFP experiments (Figure 1H) suggested this was related to cell-to-cell variation in transcription elongation rate. Both of the bleaching experiments suggested a correlation between DNA dissociation rates, RNA elongation rates, and the levels of BrU incorporation (Figure 1). This leaves the factor responsible for this cell-to-cell variation in global transcription rate unexplored, but, as the experiments in Figure 1L show, the factor appears to be affecting all transcription foci equally in the nucleus.
Next, we investigated whether the global factor responsible for the variation in transcription rate was soluble. In a first approach we incubated cells with BrU for 30 min and analysed the intensity in the nucleus and mitochondria in individual cells. This experiment showed a strong correlation between BrU incorporation in these two compartments (Figure 2A and 2B), suggesting the factor is not nucleus specific. In a second experiment, we fused Hela cells with polyethylene glycol and after 2.5 h we carried out BrU incorporation for 30 min. This experiment showed that nuclei sharing the same cytoplasm have almost identical levels of BrU incorporation per unit of nuclear volume, showing an average CV of 0.04 (in contrast, the average CV for randomly selected pairs of nuclei from different cells was 0.50). The same was observed when the dynamic properties of RNA pol II–GFP or histone H2B–GFP were analysed in fused cells (Figure 2C and 2D) (note that in Figures 2D and S12D the CV is the average of the CVs calculated for pairs of nuclei). Both sets of experiments suggested the existence of a small soluble factor responsible for the variation. An obvious candidate is differences in substrate content (nucleotides) available to RNA pol II in each cell. This was supported by the observation that “in vitro” transcription using a fixed concentration of bromouridine triphosphate (BrUTP) as a tracer showed a much lower degree of variability than BrU incorporation “in vivo” (CV<0.10) (Figure S5). BrUTP incorporation “in vitro” was performed in permeabilized cells, guaranteeing an even concentration of precursors to all cells.
Based on these results, we sought to analyse the relationship between nucleotide precursors and BrU incorporation. However, even though we have a good knowledge of NTP concentrations in cell populations [26], there are no methods available to measure the nucleotide content in individual cells that are appropriate in this case. Instead, we studied the behaviour of RNA pol II with respect to [NTPs]. We used a nonradioactive method to measure the kinetic properties of RNA pol II, attached to the appropriate template, in the natural environment of the cell nucleus. Our method is based on measurements of the amount of incorporated BrUTP in nascent RNA, detected by immunofluorescence. We checked the dependence of the speed of transcription on different substrate concentrations. Cells were incubated with a cocktail containing different concentrations of all the NTPs except for ATP, which was fixed at a cell physiological level of 1 mM (henceforth NTP will refer to UTP, CTP, and GTP only). Plotting transcription rate, V, versus [NTP] yields a hyperbolic curve (Figure 2E), consistent with Michaelis-Menten kinetics with a Km of 80±10 µM (R2 = 0.996) (Figure 2E and 2F). This suggests that RNA pol II activity depends on the nucleotide content of the cell. However, the concentration of NTPs inside the cell is believed to be in the millimolar range [26]. From Figure 2E, this means that RNA pol II is effectively working at full speed with respect to NTPs (even if NTP concentration falls from 1 mM to 250 µM). Therefore, [NTP] is unlikely to be the factor responsible for the observed variation.
Some models for transcription in the literature have explicit and implicit energy dependences (see [27] for an example). Given this energy dependence, we also studied the RNA pol II activity with respect to [ATP] (this time fixing NTP concentration at 100 µM). The plot of V versus [ATP] showed a sigmoidal curve (Figure 2G). A plot of [ATP]/V versus [ATP] (Figure 2H) [28] emphasises this. It is thus possible that RNA pol II behaves as an allosteric enzyme (Hill coefficient of 1.5±0.34; R2 = 0.99; Km 870±450 µM) with respect to ATP.
An allosteric behaviour of RNA pol II has not to our knowledge been previously reported, possibly because all other studies have been performed either in vitro with purified enzymes or without the near-physiological conditions necessary to minimise the perturbation of essential macromolecular complexes. Our transcription system uses physiological salt concentrations and macromolecular crowding agents that keep the molecular complexes as close as possible to “in vivo” conditions.
The apparent allosteric behaviour of RNA pol II is consistent with evidence that active RNA pol II forms structures containing several molecules [18],[20],[29]. There are also reports of more simple viral RNA polymerase molecules that oligomerize and show cooperativity [30]. Another explanation for this allosteric behaviour could be an effect of ATP on other proteins that influence the catalytic activity of RNA pol II. Obvious candidates are remodelling factors and/or DNA helicases that are generating template for RNA pol II in an ATP-dependent manner. In this category we can find the ATPase CHD1 (chromo-ATPase/helicase–DNA-binding domain), which remodels nucleosomes in vitro and appears to function in both elongation and termination [31]. Another example is the remodelling complex SWI/SNF, which is also ATP dependent and associates with the RNA pol II holoenzyme [32]. Therefore, the activity of all these factors should affect the apparent activity of RNA pol II. To study if this was the case we decided to uncouple transcription from remodelling. We reasoned that by decondensing chromatin, remodelling factors would not limit the availability of DNA, and therefore these factors would contribute very little, if at all, to the kinetics of RNA production. We explored such a possibility by repeating the study of the relation between RNA pol II kinetics and [ATP] in swollen cells. Incubation of cells in hypotonic buffer for 10 min induced chromatin decondensation (Figure S6), and in these swollen nuclei the kinetic behaviour of RNA pol II with respect to [ATP] was hyperbolic (Figure 2I), in contrast to the sigmoidal kinetics observed in unswollen native cells. This hyperbolic behaviour with respect to [ATP] has also been reported for remodelling factor(s) [33]; the sigmoidal kinetics of RNA pol II with respect to [ATP] may be the result of two consecutive sub-processes (elongation and remodelling) with hyperbolic kinetics.
Chromatin remodelling effects have been suggested as a cause of intrinsic noise [2], so it is interesting to note their possible role in global variability. Whatever its origin, sigmoidicity seems to be dependent on the native status of these molecules on the natural template, which means that it probably reflects an in vivo scenario. As the intracellular [ATP] is believed to be ∼1 mM [26] (close to the RNA pol II Km of ∼870 µM, found in our conditions), small fluctuations in [ATP] are likely to affect transcription elongation in vivo. (This paper is concerned with the connection between transcription rate and mitochondrial function, but we also investigated the connection between mitochondrial mass, ATP, and protein synthesis; more details can be found in Figure S13.)
We presented evidence that the global factor modulating transcription rate does so for both nuclear and mitochondrial genes (and so is not nuclear specific; Figure 2B). Fusion studies suggested this factor is small and rapidly diffusing (Figure 2D). In vitro studies indicate a sensitive dependence of transcription rate on [ATP] (at around cellular concentrations), while this is not the case for other NTPs (Figure 2E and 2G). Decondensing the chromatin eliminates this sensitivity (Figure 2H).
The experiments described above suggest that the differences seen in BrU incorporation could be a reflection of cellular heterogeneity in ATP content. Indeed, in population studies where we perturbed intracellular [ATP], we observed a direct relationship between BrU incorporation and [ATP] (Figure 2J). A similar effect was observed in the rate of dissociation (t1/2) of RNA pol II (Figure S7). By sorting cells according to their mitochondrial content (using MitoTracker Green FM dye), we found that cells with a higher transcription rate per unit volume of nuclear material also have more mitochondrial mass (Figures 3A, 3B, and S8). Using similar sorting experiments, we found evidence that a crude measure of cellular [ATP] covaried with mitochondrial content (Figure S9A and S9B). We explored this correlation further using another indicator of [ATP]. ATP is a product of mitochondrial function so we assessed the mitochondrial membrane potential (Δψ), which is the driving force for ATP production [34]. Cells were sorted according to tetramethyl rhodamine methyl ester (TMRM) levels (an indicator of Δψ), and we found evidence for a correlation with an approximate measure of cellular [ATP] (Figure S9C). Single-cell studies showed that both total membrane potential and also transcription rate are slowly varying (Figure S10; Video S1). For further discussion, see Text S1.
To study the relationship between Δψ and the rate of BrU incorporation per unit of nuclear volume we used MitoTracker Red (CMXRos), a fixable probe (TMRM is not fixable) sequestered inside the mitochondria that depends on Δψ. After incubation with both reagents we quantified the signals in individual cells. The two parameters showed a strong relationship, suggesting that total membrane potential relates to the BrU incorporation (Figure 3C and 3D). To give further support we used phosphorylated ribosomal protein S6 (P-S6) as a reporter of the energy state of the cell. P-S6 is located downstream in the mTOR pathway. mTOR is a homeostatic [ATP] sensor, and phosphorylation of its targets is dependent on ATP concentration [35]. One target of mTOR is the ribosomal S6 kinase (S6K1) that phosphorylates the ribosomal protein S6 [36]. The use of P-S6 as a reporter for energy status was validated by induction of energy stress after deprivation of glucose and incubation with deoxyglucose (DG) for 12 h, which resulted in depletion of cellular ATP ([36] and this study, data not shown). As predicted, P-S6 decreased in response to energy depletion (), working as a surrogate reporter of the energy status of the cell. Next, we incubated cells for 30 min with BrU, and after immunolabelling with BrU and P-S6 antibodies, we observed a correlation between both signals (Figure 3E), and both decreased in a manner proportional to the concentration of DG (Figure 3F). We also increased the intracellular concentration of ATP by incubation with succinate at 5 and 10 mM, which increased [ATP] to 135% of normal levels, resulting in an increase in transcription rate and reduction in transcription rate variability (assessed by measuring total nuclear BrU incorporation and H2B t1/2 exchange (Figure 3G and 3H).
If mitochondrial activity is coupled to variability in transcription rate, then changing mitochondrial function by altering the presence of anti- or prooxidants might affect this rate and its variability. We undertook studies using the antioxidants dithiothreitol (DTT) and MnTMPyP and prooxidants diamide and N-ethylmaleimide (NEM) (Figure S12). These studies suggested that the presence of antioxidants increases transcription rate and reduces rate variability, with the opposite holding for prooxidants (Figure S12B and S12C). For further discussion, see Text S2.
In summary, we find evidence suggesting that transcription rate per unit volume of nuclear material covaries with the mitochondrial mass of cells (Figures 3A, 3B, and S8). We also found that a measure of membrane potential (integrated over the cell) correlated with transcription rate per unit volume (Figure 3D). By modulating intercellular nutrients we modulated intracellular [ATP] and found that this also correlated with the degree of BrU incorporation (Figure 2J). Further indirect studies gave support to this connection between ATP levels, mitochondrial mass, and transcription rate (Figures 3F–3H, S9, and S12).
In order to understand the origin of cell-to-cell differences in transcription rate, and given the observed connection between transcription rate and mitochondrial mass, we measured the mitochondrial content in Hela cells using MitoTracker Green FM dye. This staining demonstrates that Hela cells are heterogeneous in terms of mitochondrial content (Figure 4A and 4B). We investigated the asymmetric segregation of mitochondria between daughter cells as a possible source of this heterogeneity. We used for this analysis a stable cell line containing mitochondria tagged with yellow fluorescent protein (YFP). A plasmid encoding subunit VIII of cytochrome c oxidase fused with YFP was transfected into an epithelial-like human cell line derived from a bladder carcinoma (ECV304). The tagged subunit is incorporated into mitochondria, diffusing rapidly throughout the interior of the mitochondrion [37]. This makes this chimeric protein an ideal reporter to study the behaviour of mitochondrial mass at mitosis. We focused on cells in telophase or late mitosis (Figure 4C), where we measured the mitochondrial mass for each daughter cell as the integrated intensity of the mitochondrial signal [38]. This analysis showed that cells generically segregate mitochondria in an uneven manner (Figure 4C and 4D).
Given the observation that mitochondrial mass segregates asymmetrically, one can ask whether this is relevant to cell physiology. Cell tracking experiments showed that mitochondrial content at mitosis correlates with cell cycle length. The daughter cells with more mitochondria progressed through the cell cycle proportionately faster than their sisters (Figure 4E). To rule out the trivial explanation that asymmetry in mitochondrial content was an effect of asymmetry in the volume of daughter cells, we used ECV304 cells expressing DsRed, which is a soluble protein and distributes evenly throughout the cell. The analysis of the ratio of DsRed between daughters (ratio of cell volumes) versus the ratio of time to complete a cell cycle did not show a clear relationship (Figure 4F), making it unlikely that asymmetries in the volume of daughter cells are the principle cause. We also found that daughters that inherit more mass than their sisters also have a higher rate of translation of some proteins (Figure S13A); this further suggests that the uneven inheritance of mitochondria has an effect through the cell cycle. For further discussion, see Text S3.
Since mitochondrial segregation is correlated with variation in cell cycle length, one might want to understand how mitochondrial partition at birth is controlled. We analysed whether the process of mitochondrial segregation had a memory. We measured the mitochondrial content ratio between daughters at birth (F1), followed each daughter for one cell cycle, and measured the mitochondrial content ratio of daughters in generation F2. When we compared F1 versus F2 in terms of asymmetry, there was no clear relationship (Figure 4G). The time to division of generation F1 cells was largely independent of the interdivision times of respective F2 cells (Figure 4H). This gives us a more refined view of the stochastic character of mitochondrial segregation.
We have found evidence for variability in mitochondrial mass within a population (Figure 4A). A possible cause is asymmetric segregation of mitochondrial mass at division (Figure 4C and 4D). Daughter cells that inherit more mitochondrial mass progress through their cell cycles faster and can show a faster rate of protein synthesis (Figures 4 and S13A–S13D). We found no strong evidence for a dependence between one cell cycle duration and the next (Figure 4G and 4H).
This paper investigated the connection between two forms of cellular variability: variation in mitochondrial mass and variation in global transcription rate. We found marked heterogeneity in the amount of mitochondrial mass present in cells (Figure 4B) and evidence for an origin of this variability in the stochastic partition of mass at point of division (Figure 4D). We further found that this variation has a cell physiological correlate: daughters that inherit relatively smaller amounts of mitochondrial mass than their sisters have longer cell cycles (Figure 4E). We also presented evidence for global (Figure 1L) variability in transcription rate (Figures 1E, 1H, and S1A). While our experiments suggest that the numbers of bound RNA polymerases are constant and transcription rate variability is independent of cell cycle stage (Figures S1E, S2, and S5), they also suggest that a small diffusing factor may be responsible for this global transcription rate modulation (Figure 1D).
Given the above, we hypothesized, first, that there was aconnection between global transcription rate variability and variability in cellular mitochondrial content and, second, that this was mediated by variation in the fast-diffusing factor ATP. Studies in permeabilized cells (Figure 2G) found a sensitive dependence of transcription rate on [ATP]. In vivo perturbation studies also found a correlation between cellular [ATP] and transcription rate (Figure 2J). We found that cells with more mitochondrial mass transcribe faster per unit volume of nuclear material (Figure 3B). We further found that those with higher total membrane potential (as indicated by CMXRos) transcribed faster (Figure 3D). We also found evidence correlating levels of ATP with mitochondrial mass and total membrane potential (Figure S9B and S9C). Finally, we found that perturbing mitochondrial function with anti- or prooxidants perturbed transcription rate variability (Figure S12).
Studies thus far have left our understanding of the origins of global variability in gene expression in higher eukaryotes unclear [2]. This paper suggests that cell-to-cell variability in mitochondria is coupled to cell-to-cell variability in global transcription rate.
For in vivo transcription, cells were incubated in the presence of different concentrations of BrU (Sigma) for different times (stated in figure legends). Incubation for 1 h with 100 µM DRB or for 1 h with 1 µg/ml actinomycin D prior to BrU incubation abolished BrU incorporation completely (data not shown).
For individual transcript analysis, Hela cells were grown on coverslips at low density then incubated for 15 min with 5 mM BrU, washed with PBS, and treated with 0.375% sarkosyl, 25 U/ml ribonuclease inhibitor, 10 mM EDTA, and 100 mM Tris-HCl (pH 7.4) for 10 min at 20°C. Next, coverslips were tilted to allow the cell content to run out for 5 min. Samples were air-dried and fixed with 4% paraformaldehyde for 10 min and processed for Br-RNA detection.
For transcription in vitro we used the conditions described in [18] plus 5% Ficoll 400.
For detection of primary transcripts, we used mouse anti-IdU/BrdU (5 mg/ml; Caltag Laboratories). Secondary antibodies were donkey anti-mouse IgG tagged with Cy3 (1/200 dilution; Jackson ImmunoResearch). The immunodetection procedure was performed as described in [18],[19]. DNA was stained with 200 nM TO-PRO-3 (Molecular Probes) for 5 min, then slides were mounted in Vectashield (Vector Laboratories), and images were collected using a Radiance 2000 confocal microscope (BioRad Laboratories). Intensities in the nucleoplasm were measured using EasiVision software (Soft Imaging Systems) and data exported to Excel (Microsoft) for analysis.
For cell fusion experiments, Hela cells were grown on coverslips to 80% confluence. Cells were fused using polyethylene glycol as described by Schmidt-Zachmann et al. [39]. After 2.5 h cells were incubated with 2.5 mM BrU for 30 min and then immunolabelled as described above.
A clone stably expressing GFP–RNA pol II (C23) [22] was cultured at 39°C, and images were collected with the microscope stage heated to 39°C. Fluorescence images were collected using a confocal microscope (Zeiss LSM 510 META), with an EC PlnN 40×/1.3 oil objective, with the pinhole completely open. We selected a rectangle at the bottom half of each nuclei where we applied 100% laser power, in order to bleach all the fluorescent molecules in the rectangle. This operation was repeated every 5 s for a period of 1,200 s, and we analysed the decay of the fluorescence in the unbleached top half. Fluorescence intensity was analysed in MetaMorph 6.1 (Universal Imaging). Curves were analysed using Sigma Plot 8.0 for Windows. For the analysis we assumed that there were two populations, freely diffusible, bound to DNA and fully engaged in transcription. For the fitting we allowed the two components to optimise with no restriction. Data were fitted to two populations with exponential decay (always R2>0.99). Fixing the slow population to an average speed rendered unacceptable fittings with the second population. We were concerned with the possible artefacts induced by FLIP. Therefore, transcription “run on” experiments were performed on photobleached cells, which demonstrated no alteration in the transcription pattern or intensity in the bleached area (data not shown).
Hela cells expressing histone H2B–GFP [25] were used to study the dynamics of histone H2B. FLIP was performed as for C23 cells, but the time was reduced to 10 min of photobleaching and the temperature was set at 37°C. The decay curves can be fitted to a bi-exponential decay. The two initial points were discarded because they correspond with the free population of histone H2B. One possible problem with the use of the exchange of histone H2B–GFP as a transcription reporter is the impact of its overexpression. However, in the cell line used, histone H2B–GFP represents 10% of all cellular histone H2B [25]. The production of natural histone H2B is reduced in preserving the normal amount of histones, which means that no overexpression occurs in this cell line [25]. The fraction of free histone in the cell line used is around 1% of the total H2B–GFP, which corresponds with the fraction bleached in the first two cycles of bleaching. This population was not considered for the analysis. Even if any hyperexpression occurs in this cells, only the fraction of molecules bound to DNA and not the t1/2 will be affected, which is the parameter studied. In agreement with this interpretation we did not observed a correlation between the initial fluorescence before bleaching of histone H2B–GFP and t1/2 (Figure S4C).
Tripsinized Hela cells were stained with MitoTracker Green FM dye (Molecular Probes) for 15 min in DMEM or TMRM (Molecular Probes) for 30 min, following manufacturer guidelines. Then, cells were sorted on a fluorescence-activated cell sorter (MoFlo; DakoCytomation) to purify populations of cells with different mitochondrial content or membrane potential.
MitoTracker Red (CMXRos) was used following the manufacturer guidelines (Molecular Probes). Cells were stained for 10 min in vivo after being grown in BrU for 30 min. Br-RNA was detected as previously described. For superoxide detection cells were incubated with 20 nM for 12 h with MitoSox (Molecular Probes) and then grown in BrU for 30 min. Cells were analysed using wide confocal cytometry [38].
ATP depletion experiments were carried out by incubation of cells for 12 h with different concentrations of DG (Sigma). In another set of experiments ATP was depleted by incubation with 10 mM sodium azide (Sigma) and 6 mM DG in HBSS for 1 h (BioWhittaker). ATP concentration was determined using the kit ATP Bioluminescence Assay Kit HS II (Roche) following manufacturer instructions.
For antioxidant treatments, cells were incubated for 18 h with MnTMPyP (CalBiochem) or DTT (Sigma). MnTMPyP was used at 50, 25, and 12.5 µM. DTT was used at 1,000, 500, 250, and 125 µM. GSH was depleted by incubation with 200, 100, or 50 µM diamide (Sigma) for 2 h.
CE–mitoRFP-W vector was generated from the pHR-SIN-CSGW vector [40] by exchanging the SFFV promoter for a human EF1a promoter and the GFP reporter for mitochondrial DsRed2 isolated from pDsRed2-Mito (Clontech). Lenti lox vector expressing GFP, Emerald, or Cherry was generated as described in [41].
Lentiviruses were pseudotyped with the vesicular stomatitis virus G (VSVG) protein by transient transfection of 293T cells [41]. Viral stocks were prepared by ultracentrifugation, and viral particles were used for Hela H2B–GFP infection; 2 wk after infection a clone was selected.
For in vivo analysis, cells were plated in a 48-well plate at low cell density, and left for 12 h to attach. Then the plates were transferred to the Cell IQ platform (Chip-Man Technologies). Images were recorded every 30 min, for at least 6 d. Images were analysed using MetaMorph 6.1. After completion of mitosis, the ratio of the integrated intensity of the fluorescent signal between daughter cells was measured as described in [38].
|
10.1371/journal.pbio.1001891 | Targeting Global Protected Area Expansion for Imperiled Biodiversity | Governments have agreed to expand the global protected area network from 13% to 17% of the world's land surface by 2020 (Aichi target 11) and to prevent the further loss of known threatened species (Aichi target 12). These targets are interdependent, as protected areas can stem biodiversity loss when strategically located and effectively managed. However, the global protected area estate is currently biased toward locations that are cheap to protect and away from important areas for biodiversity. Here we use data on the distribution of protected areas and threatened terrestrial birds, mammals, and amphibians to assess current and possible future coverage of these species under the convention. We discover that 17% of the 4,118 threatened vertebrates are not found in a single protected area and that fully 85% are not adequately covered (i.e., to a level consistent with their likely persistence). Using systematic conservation planning, we show that expanding protected areas to reach 17% coverage by protecting the cheapest land, even if ecoregionally representative, would increase the number of threatened vertebrates covered by only 6%. However, the nonlinear relationship between the cost of acquiring land and species coverage means that fivefold more threatened vertebrates could be adequately covered for only 1.5 times the cost of the cheapest solution, if cost efficiency and threatened vertebrates are both incorporated into protected area decision making. These results are robust to known errors in the vertebrate range maps. The Convention on Biological Diversity targets may stimulate major expansion of the global protected area estate. If this expansion is to secure a future for imperiled species, new protected areas must be sited more strategically than is presently the case.
| Under the Convention on Biological Diversity (CBD), governments have agreed to ambitious targets for expanding the global protected area network that could drive the greatest surge in new protected areas in history. They have also agreed to arrest the decline of known threatened species. However, existing protected areas perform poorly for coverage of threatened species, with only 15% of threatened vertebrates being adequately represented. Moreover, we find that if future protected area expansion continues in a business-as-usual fashion, threatened species coverage will increase only marginally. This is because low-cost priorities for meeting the CBD targets have little overlap with priorities for threatened species coverage. Here we propose a method for averting this outcome, by linking threatened species coverage to protected area expansion. Our analyses clearly demonstrate that considerable increases in protected area coverage of species could be achieved at minimal additional cost. Exploiting this opportunity will require directly linking the CBD targets on protected areas and threatened species, thereby formalizing the interdependence of these key commitments.
| In 2010 the 193 parties to the Convention of Biological Diversity (CBD) adopted a new strategic plan and set of targets to tackle the continuing decline in biodiversity [1],[2]. A key element of this plan is Aichi target 11, which includes a commitment to expand the global coverage of terrestrial protected areas from the current 13% to 17% by 2020 [1]. This could drive the most rapid expansion of the global protected area network in history [3], but corresponding biodiversity benefits are far from guaranteed. This is because protected areas are often preferentially established in locations that are remote or have little agricultural value [4], failing to protect the imperiled biodiversity found on more valuable land.
Recognizing the failures of past protected area expansion, the current CBD text directs that protected areas should target places of “importance for biodiversity” that are “ecologically representative” [1]. However, these locations can be expensive to protect. For instance, the cost of expanding protected areas to cover all “important bird areas” (IBAs) has been estimated at US$58 billion annually (although these sums are still small compared to government budgets) [5]. Moreover, the majority of terrestrial regions have been identified as important for biodiversity by one or more global prioritization schemes [6], which provides myriad alternatives for meeting protected area targets in locations that are cheap. Given this, where should new protected areas be located to deliver on the Aichi biodiversity targets? One option could be based on Aichi target 12, which aims to “prevent the extinction of all known threatened species and improve and sustain their conservation status.” In situ conservation of viable populations in natural ecosystems has long been recognized as the fundamental requirement for the maintenance of biodiversity [7]. Hence measuring “biodiversity importance” in terms of protected area coverage of threatened species would help countries to simultaneously meet these two CBD targets.
Using new data from the World Database on Protected Areas [3] and distribution maps for 4,118 globally threatened birds [8], mammals [9],[10], and amphibians [10],[11], as well as ecoregions [12], we first perform a gap analysis to determine the representation of these species in the current global protected area network. We then use a systematic conservation planning framework [13] to build scenarios for cost-efficiently expanding the global protected area network to contribute to meeting the protected area and threatened species Aichi targets. Recent works have investigated strategies for achieving Aichi Target 11 by protecting IBAs [5],[14] or meeting the Global Strategy for Plant Conservation [15]. Our study is the first, to our knowledge, to use an optimization approach to develop scenarios for meeting the Aichi targets in a cost-efficient manner. Incorporating cost efficiency allows the identification of options for meeting Aichi target 11 that contribute optimally to target 12 while minimizing conflict with agricultural production.
All spatial overlays were performed at a spatial resolution of 500 m and then aggregated into 30 km×30 km pixels to identify candidate land for protection. By processing data at the finer resolution, we are able to account for protected areas at the subpixel level, thereby minimizing omission of small-sized protected areas. This resolution of ∼⅓ degree (at the Equator) falls in the midrange between scales of ½ degree [16] and of ⅛ degree [17] typically used in such analyses.
To determine the extent of current protected areas, we extracted data on International Union for Conservation of Nature (IUCN) category I–VI protected areas from the 2012 World Database on Protected Areas [3], excluding all proposed protected areas and those lacking “national” designation. For terrestrial protected areas with a known areal extent but lacking polygonal representation, we created a circular buffer of the appropriate area around its centroid. To prevent overestimation of the areal coverage of protected areas caused by overlapping designations, we merged buffered points and polygons into a single layer. Our final protected area layer contained 135,062 protected areas covering a total of 17,026,214 km2, or 12.9% of the Earth's non-Antarctic land surface (Figure 1A).
We used distribution maps for birds [8], mammals [10], and amphibians [10]. We focused on these taxa as they are the only major terrestrial taxonomic groups that have been comprehensively assessed for their distribution and extinction risk [10]. We excluded marine species and areas, noting that there are specific coverage targets for protecting the marine realm. For all three taxonomic groups, we focused on those species that are listed by the IUCN Red List as Critically Endangered, Endangered, or Vulnerable, hereafter referred to as “threatened,” resulting in 4,118 species in total (birds = 1,135, mammals = 1,107, amphibians = 1,876; Figure 1B). We focus only on threatened species as these are by definition the most likely species to go extinct, and therefore are most important for slowing biodiversity loss and contributing to CBD Aichi target 12. We excluded all portions of species ranges where the species was identified as extinct, introduced, or of uncertain origin. In addition to these data, we used data on the distribution of ecoregions as defined by the World Wildlife Fund [12].
To account for the spatial variation in the cost of protected area expansion, we used a dataset on agricultural opportunity cost [18], converted to 2012 US$ and with no data values filled using regularized spline interpolation with tension (Figure 1C). The dataset provides the estimated gross agricultural rents for terrestrial areas mapped at approximately the 5 km resolution. We use these data as our surrogate for the opportunity costs of establishing new protected areas, as agricultural expansion is the greatest single cause of habitat loss, as well as the one most commonly associated with habitat loss driven by multiple factors [19],[20]. Agricultural opportunity costs also reflect the reduction in food security and tax revenue that national governments face when implementing protected areas. We applied a fixed cost of US$100 per km2 to reflect the transaction costs of acquiring new protected areas [21], although we recognize there is likely to be considerable spatial variation in these costs. We did not attempt to estimate the ongoing management costs of protected areas following establishment, as this metric needs to account for a number of difficult-to-measure social and socioeconomic factors [22], but a recent analysis estimated that these equate to ∼14% of the agricultural opportunity costs of protection [5].
We assessed the occurrence of threatened vertebrates within protected areas using a representation target and an adequacy target. The representation target was achieved if any portion of the species' distribution overlapped with the protected area network. To set adequacy targets we followed the method of Rodrigues et al. [23] to scale the target to the species' overall geographic range size. Complete (i.e., 100%) coverage by protected areas was required for species with a geographic range of <1,000 km2. For wide-ranging species (>250,000 km2), the target was reduced to 10% coverage, and where geographic range size was intermediate between these extremes, the target was log-linearly interpolated.
To explore future scenarios for the growth of the global protected area network we used the systematic conservation planning software Marxan [24]. Marxan uses a simulated annealing algorithm to select multiple alternative sets of areas that meet prespecified conservation targets (described in the following section) while trying to minimize overall cost. All spatial data on the distribution of conservation features and conservation costs were summarized into a “planning unit” layer consisting of 30 km×30 km square pixels comprising the world's non-Antarctic terrestrial areas. We intersected this planning unit layer with the protected areas and agricultural opportunity layers and the geographic distribution of each of the 4,118 threatened species and ecoregions at a 500 m resolution. This allowed us to determine the agricultural opportunity cost of the unprotected portion of each planning unit and the protected and unprotected extent of each biodiversity feature within each planning unit.
To explore the costs and benefits of alternate scenarios for achieving 17% protection of terrestrial areas, we developed four separate spatial scenarios using contrasting conservation targets. We accounted for the existing protected area network's contribution to the targets in each scenario, and then added additional protected areas to ensure all targets are met. In each scenario, the aim is to minimize the costs of meeting the conservation targets. However, to avoid the global protected area target being met only through increased protection in low-cost countries, which would reduce the total cost of the target, in all scenarios we maintain the constraint that each country must meet its national protected area target. Moreover, it is at the national level that the target is being interpreted and implemented. For each scenario, we used Marxan to perform 10 runs of 1 billion iterations each, each of which represents an alternate near optimal reserve network for meeting the relevant conservation targets at the lowest overall cost. From these 10 runs, we select and report on the results from the lowest cost solution.
The IUCN [10] and Birdlife International and NatureServe [8] range maps used in this study comprise polygons showing distribution of 4,118 globally threatened birds, mammals, and amphibians. These maps may be subject to commission errors [26]–[29], where the species is mapped as present in locations where it is in fact not present. As they affect range-based species conservation targets and lead to an overestimation of occurrence in existing or prioritized areas, commission errors could influence our study's main conclusions. We performed two analyses to determine the sensitivity of our primary results to commission errors (Text S1). First we created 100 range maps for each of the 4,118 species of birds, mammals, and amphibians that simulated commission error rates [25] by deleting 50% of the range of narrow-ranged species (range<1,000 km2), by deleting 25% of the range of wide-ranging species (range>250,000 km2), and by linearly extrapolating the deletion rate for species of intermediate ranges. Second, we identified the “Extent of Suitable Habitat” (ESH) using high-resolution species distribution models for 1,063 mammals [30]. The ESH maps were used to identify locations in the original maps for mammals that are likely to be commission errors. We then reran our analyses using (a) the maps with simulated commission errors and (b) the ESH maps, to quantify the effects of the simulated and mapped commission errors on our estimated biodiversity value of meeting the 17% protected area target, and the shape of the efficiency frontier between cost and threatened vertebrate coverage.
We find that 17% of threatened vertebrates are not found in a single protected area and 85% are not covered to the level of our adequacy targets (Figure S1A). A decade ago, 20% of globally threatened terrestrial birds, mammals, and amphibians were not found in a single protected area and 89% were inadequately protected [15]. Our analysis using updated datasets indicates that the global protected area network has made little progress since then toward securing a future for the world's threatened biodiversity.
We discover that if countries choose to expand their protected areas in a manner that minimizes agricultural opportunity cost, meeting their national-level targets for 17% coverage would entail a once-off transaction cost of US$0.9 billion and an annual agricultural opportunity cost of $4.9 billion (Table 1). As this option aligns with the previous pattern of protected area establishment, we view it as a likely business-as-usual scenario for meeting the terrestrial coverage aspect of Aichi target 11. We find that this would result in only 852 (21%) threatened vertebrates reaching targets for adequate coverage (Figure S1B), an increase of only 249 species over existing protection (Table 1) and arguably a failure to meet Aichi target 12. Moreover, even if highly ambitious areal targets were to drive further growth of the global protected area network beyond 2020, the costs of expansion would rise steeply without providing cost-effective coverage for threatened species (Figure 2).
An alternative is to ensure a representative sample of major vegetation communities is protected, as this would protect a broader range of habitats and could lead to improved conservation outcomes. Target 11 calls for ecologically representative protected area coverage. We find that if countries meet their 17% coverage targets in a way that distributes protection across ecoregions equally, the opportunity cost of establishing the additional protected areas would be 4.5 times higher than the business-as-usual scenario ($24.8 billion annually; Table 1), but that coverage of threatened species would increase only marginally (Figure S1C). Moreover, the majority of species that reach their adequacy targets are those with a geographic range size ≥250,000 km2 (Figure S1C), as their wide distribution renders them more easily captured when distributing protected areas equitably across ecoregions. The species most likely to be left unprotected are narrowly distributed species, which often are those in greatest need of protection [31],[32].
These results indicate that protected area expansion targeting either the cheapest land or representation of ecoregions is not an efficient approach for covering threatened species. Alternatively, we find that locating protected areas to ensure they meet targets for adequate coverage of all 4,118 threatened species would cost about $42.5 billion annually (Table 1), which is about 7.5 times more than the cheapest option for meeting the 17% target. This difference in cost is driven by low concordance between areas that are cheap to protect and those that capture the distributions of threatened species (Figure 1D). Land selected for threatened species tends to align with tropical forest hotspots (Figure 1B), such as the tropical Andes and eastern Madagascar, whereas the cheapest land to protect is remote and often in more arid zones (Figure 1D). This lack of overlap helps explain why the existing protected area network, which has favored low-cost areas in each country [4], represents threatened species rather poorly.
How can countries reconcile the attraction of low-cost conservation with the benefits of protecting places that contribute to threatened species conservation? By varying the importance placed on meeting targets for adequate coverage of threatened species, we discover a nonlinear tradeoff between the cost of establishing additional protected areas and the proportion of threatened vertebrates covered by these areas (Figure 3). The shape of the curve illustrates that large gains in the number of species potentially protected could be achieved for relatively small increases in cost. For instance, increasing by 5-fold the number of species protected relative to the low-cost, business-us-usual scenario would increase opportunity costs to only $7.4 billion annually (1.5 times as much; Table 1).
We find that our primary results are robust to randomly simulated commission errors in the range maps. Although the number of species meeting range-based coverage targets generally decreases once commission errors are simulated (Text S1), this drop averages only 5% across the tradeoff curve (Figure S2). Moreover, both a visual interpretation and a quantitative measure of the shape of the tradeoff curve reveals that the original and commission error updated curves are similarly nonlinear. Moreover, using high-resolution expert-based habitat suitability models for 1,063 threatened mammals, we again find that commission errors are unlikely to alter our primary findings (Figure S3).
A small minority (15%) of threatened vertebrates are adequately covered by existing protected areas. However, the adoption of the Aichi targets marks an historic opportunity for achieving conservation of the world's biodiversity. If countries are to meet the protected area Aichi target, at least 5.8 million km2 of new protected areas will need to be created by 2020. Although this is a significant opportunity for biodiversity conservation, we have shown that protected area expansion that targets low-cost areas in each country and ignores threatened species is unlikely to protect such species incidentally. This remains the case even if protected areas are further expanded to cover 30% of land areas, or if they are located to cover a representative sample of Earth's terrestrial ecoregions. On the other hand, we find that if protected areas are directed in a cost-efficient manner to protect threatened vertebrates, these species could be protected for an estimated agricultural opportunity cost of about $42.5 billion annually. We also find that there is a nonlinear relationship between cost and species protection, indicating that options exist for increasing threatened species protection above the business-as-usual level at little additional cost.
Our estimate of the cost of reaching adequacy targets for all threatened birds, mammals, and amphibians is lower than the $58 billion annually estimated for protecting the world's IBAs [5], though each option comprises a similar land area. There are three primary reasons for this. First, the estimated costs of protecting IBAs include management costs, which are estimated at ∼$7 billion annually [5]. Second, IBAs are identified for their contribution to global bird conservation, without consideration of the cost of protecting these areas, whereas we used an optimization approach to identify low-cost options for meeting conservation targets [33],[34]. Third, IBAs are identified based on the presence of both threatened and nonthreatened species (e.g., congregatory species), while we focused on threatened species alone.
Our analyses are subject to a number of caveats. First, we considered relative cost based on gross agricultural rents, not management costs or the opportunity costs for other land uses [33], nor the practicalities of establishing reserves among these competing land uses. Second, overlay of coarse scale maps of species distributions onto fine-scale protected area maps generates commission errors [26],[35], though these are unlikely to qualitatively change our results. Still, as commission errors mean that species distributions overlap less than these coarse-scale maps suggest, our estimate of the area needed to protect all threatened species is a minimum [30]. Locations identified here should therefore be considered as broad indications of where specific areas for protection might be located, and our estimates of cost and the area requiring protection will be minima. Third, although we recognize that our analyses have limited taxonomic breadth, no other taxonomic groups (e.g., plants) have undergone comprehensive assessment of both extinction risk and distribution at a sufficiently fine scale for a comparable analysis [10]. Yet good indications exist from the literature that protected areas identified for broad taxonomic groups cover the majority of species in other, nontarget groups [36],[37]. Finally, our species-specific targets for protection do not account for minimum viable protected areas or connectivity and do not guarantee the long-term survival of all species. Moreover, many species are threatened by processes other than habitat loss and therefore require additional conservation actions both inside and outside protected areas [38].
For the global protected area network to fulfill its potential role as the cornerstone of biodiversity conservation [39], and for governments to meet their commitments on protected areas and species extinctions, the distribution of threatened species must inform future protected area establishment. Preventing the further loss of all threatened species is a lofty goal and will require substantial efforts. But expanding protected areas requires managing tradeoffs among societal objectives [40], and here we have shown that considerable increases in protected area coverage of species could be achieved at modest additional cost. Exploiting the nonlinearity of this tradeoff will require directly linking the Aichi targets on protected areas and threatened species (as well as other targets, including target 5 on slowing habitat loss), thereby formalizing the interdependence of these key commitments.
|
10.1371/journal.pntd.0006391 | A holistic approach to the mycetoma management | Mycetoma, one of the badly neglected tropical diseases, it is a localised chronic granulomatous inflammatory disease characterised by painless subcutaneous mass and formation of multiple sinuses that produce purulent discharge and grains. If untreated early and appropriately, it usually spread to affect the deep structures and bone resulting in massive damage, deformities and disabilities. It can also spread via the lymphatics and blood leading to distant secondary satellites associated with high morbidity and mortality. To date and despite progress in mycetoma research, a huge knowledge gap remains in mycetoma pathogenesis and epidemiology resulting in the lack of objective and effective control programmes. Currently, the available disease control method is early case detection and proper management. However, the majority of patients present late with immense disease and for many of them, heroic substantial deforming surgical excisions or amputation are the only prevailing treatment options. In this communication, the Mycetoma Research Center (MRC), Sudan shares its experience in implementing a new holistic approach to manage mycetoma patients locally at the village level. The MRC in collaboration with Sennar State Ministry of Health, Sudan had established a region mycetoma centre in one of the endemic mycetoma villages in the state. The patients were treated locally in that centre, the local medical and health personals were trained on early case detection and management, the local community was trained on mycetoma advocacy, and environmental conditions improvement. This comprehensive approach had also addressed the patients’ socioeconomic constraints that hinder early presentation and treatment. This approach has also included the active local health authorities, community and civil society participation and contributions to deliver the best management. This holistic approach for mycetoma patients’ management proved to be effective for early case detection and management, optimal treatment and treatment outcome and favourable disease prognosis. During the study period, the number of patients with massive lesions and the amputation rate had dropped and that had reduced the disease medical and socioeconomic burdens on patients and families.
| Mycetoma enjoys all the neglected tropical diseases (NTDs) characteristics. It frequently affects the poorest of the poor in poor communities in remote regions. The affected population are of low socio-economic status, of low visibility and low political and social voice and hence they are neglected. The disease is considered as a social stigma in particularly among females and children thus they tend to hide it for prolong period and when they are compelled to seek medical care the condition is then at a late stage. The mycetoma patients have many financial constraints that hinder them from seeking medical and health care. In the remote mycetoma endemic areas, the health and medical facilities are meagre, and it is difficult for the patients to reach the regional health centres and thus the majority of patients present with late advanced disease. To overcome these treatment difficulties, the MRC had adopted this holistic management approach to decentralised the patient's care, improve the disease awareness and advocacy, provide free medical and surgical treatment locally at the village level, and to improve the affected villages environmental and hygienic conditions. In this communication, the MRC is reporting on this unique experience, discussing the advantages and difficulties faced it and suggesting recommendations to improve it to be adopted worldwide. Reviewing the medical literature revealed, no report on such management approached for mycetoma patients and thus it worth reporting it.
| Mycetoma is a common neglected tropical disease, reported worldwide but endemic in many tropical and subtropical regions in what is known as the mycetoma belt and Sudan seems to have the highest endemicity [1, 2]. It is a chronic granulomatous inflammatory disease caused by several true fungi or certain actinomycetes, and hence it is classified as eumycetoma and actinomycetoma respectively [3, 4]. More than 70 organisms are incriminated in causing mycetoma [5, 6]. It is believed that these causative organisms, which are soil inhabitants, are implanted in the subcutaneous tissue via traumatic inoculation [7, 8]. Mycetoma usually spread to involve the skin, deep structures and bones leading to devastating destruction, deformities and disability [9, 10]. Early localised disease is amenable to cure and good prognosis, however, the late advanced disease is characterised by high morbidity and can be fatal [11, 12, 13]. Mycetoma has serious medical, health and socioeconomic bearings on patients, families, and communities particularly in endemic areas [14, 15, 16].
To date, its global incidence and prevalence are not well documented as mycetoma is a neglected disease, not a notified or a reportable one. Furthermore, in most of the endemic regions, there is no proper disease surveillance system, especially in Sudan. Thus most of the reported cases are limited to anecdotal case reports and passive case detection [17, 18]. Moreover, the disease susceptibility, resistance and route of infection are not well characterised [19, 20]. Nevertheless, mycetoma is commonly seen in communities of poor hygiene and environmental conditions where population live in proximity to animals and their dungs. It is believed that thorn pricks and minor injuries are important routes of mycetoma infection. This is supported by the facts that mycetoma is seen more frequently in the feet of patients of low socioeconomic status, with poor hygiene and in villages with animals enclosures made of thorny trees [21, 22].
Clinically, mycetoma starts as a small painless subcutaneous mass that gradually increases in size, then multiple sinuses with seropurulent discharge that contained grains of different colour and sizes develop [23, 24, 25, 26]. Most patients present late with advanced disease and serious complications due to the painless nature of the disease, patients’ low health education level and lack of health facilities in endemic areas [27, 28, 29]. It affects all age groups, but children and young adults of low socio-economic status are affected most, leading to serious economic and social consequences [30, 31, 32].
The proper treatment of mycetoma depends on mycetoma type and disease extent. Numerous mycological and molecular tests are required to identify the causative organisms, and that include grain microscopy and culture, cyto-histopathological examinations and PCR identification [33, 34, 35, 36, 37, 38]. Various imaging techniques such as X-ray, ultrasound, MRI, CT scans are required to determine the disease spread along the various body planes [39, 40, 41, 42]. However, most of these tests and techniques are invasive, of low specificity and sensitivity, expensive for patients and health providers in endemic areas [37]. Currently, there is no point of care diagnostic test for mycetoma. Patients need to travel for long distances to regional centres to establish the diagnosis, and that is not always feasible due to their low socio-economic status, low health education, and roadblocks in particularly during the raining season.
Patients need prolonged periods of management involving diagnosis, treatment both medical and surgical and regular follow up. Treatment may last at least one year for the minor lesions to resolve and several years for large lesions. Even after full recovery, patients need to be followed up closely for evidence of recurrence, which is not uncommon [43, 44, 45, 46].
The currently available treatments for mycetoma are suboptimal and disappointing, characterised by low cure rate (28%) and high patients’ follow up dropout (54%) rate [47, 48]. In general, actinomycetoma is treated by a combination of antibiotics, and for eumycetoma, a combination of antifungals and wide local surgical excision is needed. The available medication is not very effective, expensive, with many side effects and hence the high patients’ dropout rate [47, 48].
The diagnostic tests and treatment are expensive that amount up to $2500 per year. Whereas to the annual income in Sudan is less than $400 per capita (according to the UNDP, 2006), that creates an enormous economic burden on the patients, their families, community and eventually the whole health system in the country [49].
It is interesting to note, worldwide, there are neither preventive or control measures nor programs for mycetoma [20]. The disease surveillance, especially in Sudan, is limited to targeted prevalence studies, case reports, and passive case detection. The absence of a standardised, centralised mycetoma surveillance system has far-reaching effects on how the existing interventions are delivered in a cost-effective and evidence-based manner.
In summary, mycetoma is a very devastating endemic disease of the most underprivileged population whether socially, economically or in terms of development. Mycetoma patients tend to travel from distant remote parts of the country to central centres for the treatment. This causes high financial burden and delay in treatment initiation. Furthermore, the disabling nature of the disease hinders access to healthcare service for the majority of the patients. Thus health services decentralisation will improve the accessibility and equity of health services to patients and will directly drop the huge financial burden.
With this background, this community-based study was conducted with the objective of applying a new holistic approach to the management of mycetoma patients at the village level. That included setting up a regional mycetoma centre with a telecommunication network, offering free of charge both medical and surgical treatment at the centre, training of medical and health staff on early case detection and management, community health education, improvement villages’ hygiene all these were based on available health system structure and minimum requirement.
This is a community-based cross-sectional study which was conducted at Eastern Sennar Governate, Sennar State, Sudan in the period 2015–2017. The Governate is 400 km south of Sudan capital Khartoum. It has 292 villages with a total population of 219,800 inhabitants. In this study, 19 villages in the Governate were surveyed for early mycetoma patients’ detection and management. One village; Wad EL Nimear, the highest endemic village, was studied in depth and compared to another village Wad Onsa which is two kilometres apart with less disease prevalence, Table 1, S1 Map. During the study period, four medical and health mobile missions were organised and conducted by the MRC to the studied villages. The mission team consisted of three consultants surgeons, one physician, two consultants radiologists, four surgical registrars, two radiology registrars, one anaesthesiology registrar, one molecular biologist, one epidemiologist, one environmental consultant, one pharmacist, five medical officers, three surgical scrub nurses, two anaesthetic assistants, four laboratory technologists, two nurses, 10 medical students, one photographer, two fine artists and one musician. The study was implemented in partnership with Federal Ministry of Health, Sennar State Government, the local administrative authority and local community leaders and activists to assure the services sustainability and the study outcome execution.
Early case detection and management required house to house total coverage survey. That was conducted in 19 villages in the Eastern Sennar Governate. The data were collected by well-trained teams of medical officers, house officers, medical students, health care providers and community activists using a digital pre-designed validated closed-ended questionnaire in smart tablets. Computer Assisted Patients Identifier (CAPI) a computer application which was designed for this study was used. To validate the study questionnaire and the CAPI, a small pilot study was conducted before the data collection in a nearby village.
The CAPI is a computer application predesigned for this study to collect data from the study villages and suspected patients. It was designed by an information and technology expert from the Faculty of Mathematical Sciences, University of Khartoum. It was used in computer tablets or smartphones. It can be used offline and online. CAPI was connected to the MRC, Data Centre system and the data analysis was performed spontaneously, the results can be displayed on Google maps, Fig 1.
The data collection questionnaire had included the suspected patient’s demographic characteristics; name, age, state and village localities, lesion site, the presence of mass, sinuses, grain colour, contact address, lesions photographs, the suspected patient’s locality geographic coordinates (latitude, longitude, altitude) and the neighbourhood photographs.
All suspected mycetoma patients from the Governate were referred to Wad Onsa Mycetoma Regional Center (WOMRC). The WOMRC was established in 2015 as a partnership between the Federal Ministry of Health, Ministry of Health, Sennar State, MRC and the local community to manage mycetoma patients locally in their region. The centre consists of small surgical operation complex, two wards, pharmacy, laboratory, ultrasound and out-patient suites and telemedicine facility connecting the WOMRC and the MRC, Fig 2.
At WOMRC centre, the patients were managed by the MRC mobile mission team with continuity of care provided by a surgical team from the regional Sennar Teaching Hospital and the resident doctor at the WOMRC in direct contact with the MRC team via the telemedicine facility.
The diagnosis of mycetoma was established by careful clinical examination and lesion ultrasound examination by mobile ultrasound machine (Paolus–UF-760AG) conducted by the consultant radiologists. All mycetoma suspected patients underwent surgical excisions under general or spinal anesthesia by the consultants surgeons and surgical registrars. The histological examination of the surgical biopsies and grains culture were performed at the MRC in Khartoum as described previously [33, 35]. Some patients with massive lesions were referred to the MRC for further assessment and management. All the investigations and treatment were provided free of charge. The Sennar Ministry of Health had provided free meals and transportation for the patients and their families
The confirmed mycetoma patient’s information was entered into a predesigned patient management record. This included patient’s demographic characteristics, diagnostic tests results, management decisions, treatment received, follow-up and final patient treatment results. This information was regularly checked and updated throughout the patient management and follow-up period.
A system for medicines and consumables procurement, delivery and storage was designed. The medicines included antifungals, antibiotics, analgesics, intravenous fluids and anesthetic medicines as well as the surgical and anesthetic consumables. The medicines were procured from the Central Medical Supply Corporate in Khartoum, shipped and stored at the WOMRC pharmacy at optimum conditions. They were dispensed by the local assistant pharmacist. Patients living in remote areas, of low socio-economic status and unable to attend the outpatient's clinic at WOMRC usually receive their medicines by a community activist who dispensed them use a toktoko (large motorcycle). The patients’ information, the medicines doses and quantities were regularly registered. All these information and procedures were documented and regularly reported to the MRC.
More than 300 care providers; medical assistants, nurses and public health officers were trained on different aspects of mycetoma, which included the disease causation, presentation, diagnosis and treatment, patients’ care, referral indications and system, community health education and disease advocacy. The instructional training methods included presentations, group discussions, clinical sessions and ultrasound diagnosis demonstration. Suspected patients’ referral card was designed and distributed to the trainee, Fig 3. The improvement in knowledge, attitude and practice (KAP) of the trainees was assessed by pre and post-training tests S1 File. The training sessions were conducted at Singa Town, Sennar State capital, WOMRC and in various East Sennar Governate villages.
Medical students from the local university; Sennar University, as well as the University of Khartoum and other health institutes were training at WOMRC, on early patients’ detection, referral and management and the CAPI use was conducted
To gain the Sennar State political involvement and support, the training sessions were addressed by the Sennar State Governor and the Minister of Health, Eastern Sennar locality Governor and community leaders.
Several meetings with the local villages’ leaders, villagers and community activists were conducted at the Wad Onsa village leader’s home, the village’s mosque and WOMRC to explain the study objectives and to gain their confidence and support. They were actively involved in the mycetoma advocacy and awareness activities. The local Red Crescent volunteers were trained in mycetoma advocacy and took an important role in improving the local environment and hygiene in the affected villages, Fig 4.
A toktok was donated by the MRC to a community activist at Wad EL Nimear village for transporting patients and their medicines between the different villages and for mycetoma advocacy. In appreciation of excellent, active and energetic involvement in mycetoma advocacy and awareness, three Mycetoma Ambassadors from the Sennar State were selected.
The study Health Education Team was led by social workers from the Association for Aid and Relief, Japan, Khartoum Office, an active NGO in Sudan with several fine artists, musicians and community volunteers. The health care providers, community leaders and activists, school teachers and medical students from the local university were trained to conduct health education and advocacy sessions. Several health education sessions and activities were carried out for early active case detection were conducted. The sessions included small group discussion, school visits sessions, video films watching, and interactive open theatre drama, Fig 5.
“Mesaket Story”, a drama film documented a mycetoma patient journey from minor infection which was neglected till limb amputation was produced and was shown to more than 2000 individuals at WOMRC and other villages.
Several campaigns to improve Wad El Nimear village environment, sanitation, and hygiene to reduce the mycetoma transmission risk factors such as thorns, sharp objects, animals dungs, were organised by the State Government, official local authorities, community leaders and activists, and Red Crescent volunteers in collaboration with the study team. The thorny trees and bushes, thorny animals enclosure, animals dungs, dirt and rubbish, were removed and burnt, Fig 6.
To improve the village hygiene, reduce the contact with the animals and their excreta and to eradicate the thorny cages, 72 modern animal enclosures were constructed outside Wad EL Nimear village. This project was conducted by a kind donation from an engineering company as its social reasonability activity. These new animals cages were distributed free of charge to the villagers, Fig 6.
In mycetoma, the foot is affected most, and traumatic inoculation of the causative organisms which are present in the soil is believed to be the route of infection. The habit of going barefooted in the villages and the minor trauma are considered the risk factors for mycetoma. To reduce these risk factors, the study team has distributed around 800 new shoes to the school pupils at Wad EL Nimear village to improve the personal hygiene and to reduce the risk of developing mycetoma.
Forty students from the Department of Social Sciences, at the University of Khartoum, spent two weeks at Wad Onsa village studying the social background of the population in the affected villages in the study area and assessed their KAP to mycetoma and its socioeconomic impacts. They surveyed in depth ten villages in the locality. Opened ended questionnaire and focus group discussions were used to obtain the data.
A Project Management Board was established headed by the Minister of Health and the senior health officials, Sennar State, the Sennar State Mycetoma control programme officer, local villages’ leaders and activists, local health care providers and MRC representative. The Board oversees the project implementation and update, problem sharing, analysis and solving. The Board has regular meetings to review the quarterly reports to provide advice and recommendations for improvement.
The study ethical clearance was obtained from Soba University Hospital Ethical Committee to conduct the study. Informed consents were obtained from the leaders of the villages, informed written consents were obtained from State Ministry of Health and every suspected and confirmed patients. All medical records were anonymised
During the study period, 758 mycetoma suspected patients from the surveyed villages and other villages in Eastern Sennar Governate were seen at WOMRC. All of them had an ultrasound examination of the suspected lesions. Of them, 220 patients had ultrasonic evidence of mycetoma, and they underwent wide local surgical excisions (218 patients), and two patients had amputations. They were 134 males (60.9%) and 86 females (39.1%). Their ages ranged between 2 and 70 years and age group 15–30 years was the most affected one. Most of them were students 68 (30.9%), housewives 46 (20.9%), farmers 35 (15.9%), (Table 2). The geographical distribution was uneven, but Wad El Nimear village had the highest prevalence, (Table 2).
Most of the pateints (72.2%) had short disease duration. Pain at the mycetoma site was not a common symptom in these patients; seen in only 39 patients (17.7%). Local trauma at the mycetoma site was reported in only 38 patients (17.3%). Most of the patients had no sinuses (early lesion) 142 (64.5%), and 72 patients (32.7%) had black grains discharge from their sinuses, (Table 2).
The foot 159 (72.2%) and hand 59(26.8%) were affected the most. Less common sites were the back and gluteal one each, (Table 1). The majority of patients 139 (63.2%) had small lesions less than 5 cm in diameter, 51 patients (23.2%) had lesion between 5–10 cm in diameter, and only two patients (0.9%) had lesions more than 10 cm in diameter, (Table 2).
The lesions ultrasound examination findings were mycetoma in 202 patients (91.8%) and foreign body granuloma in 18 patients (8.2%). The surgical procedures performed ranged from wide local excision 218 (99%) to amputation 2 (1%). All patients had an uneventful postoperative recovery. The operatives findings included mycetoma lesions 192 (87.3%), foreign body granulomas with thorns 18 (8.2%), fibroma 2 (1%) and others soft tissue masses. The diagnosis was confirmed by surgical biopsies histopathological examinations, and that showed evidence of eumycetoma in 189 patients (85.9%), foreign body granuloma 17(7.7%), actinomycetoma 3 (1.4%) and others 11(5%). The latter included no-specific granuloma, neuromas and fibromas, (Table 3).
Most of the patients were followed up at the WOMRC. Thirty-seven patients (16.8%) developed recurrence, due to multifactorial factors which included massive lesion, patients’ non-compliance with treatment or other factors. Twenty-five patients (11.4%) were lost to follow-up.
Confirmed Mycetoma patients’ information was entered into the pre-designed patient’s management records. These records included full details of the patient’s demographic characteristics, diagnostic tests results, the management offered, follow-up and final patient treatment result. This information was regularly monitored and updated throughout the patient journey.
All these data were systematically reported to the MRC in quarterly basis through two types of reporting format; hard copy and a digital one, the latter one was transmitted through a telemedicine facility at the WOMRC and MRC. The reported information was systematically entered in the pre-designed data analysis software for further analysis and systematically checked for information accuracy. Data from management teams, diagnostic services and inventory was crosschecked and discussed regularly to improve recording and reporting process.
The current treatment of choice for eumycetoma is itraconazole in a dose of 400mg /day. It costs around 26 US$/day, that is not affordable by neither patients nor local health authorities, and hence the MRC managed to raise funds to procure and dispense itraconazole free of charge to patients at the WOMRC.
A system for medicines procurement, delivery, storage and dispensing at the WOMRC was designed and tested during the study.
A random sample of 218 individuals were tested before and after showing them “Mesaket Story” a drama film. The results showed improvement in their knowledge, attitude and practice and towards mycetoma, (Table 4).
Several small group sessions were organised at different villages, schools, mosques and community clubs. 200 community activists, 50 Red Crescent volunteers and 500 school teachers were trained on mycetoma advocacy and awareness.
The fine artists and musician had organised several interactive open theatre dramas. Different health education materials in different forms were used.
Experts in watercolouring, oil painting and photography have greatly contributed to mycetoma advocacy and awareness through their production of high-quality paintings, photographs and videos captured from the studied Governate, Fig 7.
During the study, six patients with amputations received limb prosthesis donated by the Agent of Aid and Relief, Japan. This had remarkably improved the life quality.
The MRC records showed that the EL Gazeria, White Nile and Sennar States are the top endemic states in the country [28]. Despite been the third, Sennar State has been chosen for the present study due to the strong commitment of the political leaders, civil societies and communities leaders to support and implement the study and its research outcome. The local communities’ leaders were aware of the negative impacts of mycetoma on health and its socioeconomic bearings, and hence their response to the study team requests was swift and extremely positive.
The concept of the village specialised mycetoma centre reported in this communication is a unique one. The WOMRC had delivered integrated medical and social services at the heart of an endemic area. The centre was established as a joint project between the Federal, Sennar State Ministries of Health and the local community, which by itself an exceptional initiative. This study demonstrated numerous positive impacts of the centre on the local communities. It provided local, decentralised mycetoma services in a location with bare minimal health service provision and has improved the local population health education and disease awareness.
The telemedicine which links the MRC in Khartoum with the centre has facilitated the management and follow-up of patients, thus reducing the financial and geographic burdens on the patients and families, and also reduced the patients’ follow-up dropout rate. The dropout reported in this study (11.8%) is less than that reported at the MRC (54%). Although the capital cost of the telemedicine setup is high in developing countries, however, in the long run, it is cost-effective for the patients, families and health authorities in mycetoma endemic regions. This is a unique experience that has not previously been reported for mycetoma or other neglected tropical diseases (NTD) and can be replicated for other endemic NTDs.
The global lack of disease control or elimination programmes due to the unavailability of basic disease epidemiological characteristics has resulted in early case detection and management as the only available method to reduce the disease incidence and prevalence and its community impacts [28]. It is now evident that WOMRC has tremendous bearings on disease management by offering early case detection facilities and free, decentralised medical, health and advocacy services. That is evidenced the fact that many patients (63.2%) with small lesions and patients (72.2%) with disease duration of less than five years were seen at the centre and only two patients underwent amputation during the study period compared patients seen at the MRC [28]. Furthermore, that is supported by the improvement in disease awareness as evidenced by the KAP study results.
The mycetoma onset and progress are usually slow and painless, affecting patients of low socio-economic and health education levels. Hence these patients are different from patients with other deadly infectious diseases, e.g. malaria, cholera, leishmaniasis, where patients have no other choice but to report early and follow medical instructions [50, 51]. Moreover, mycetoma patients with early lesions differ from patients with large disabling mycetoma lesions. Early lesions are usually tiny and painless thus not interfering with their normal daily activities. Some patients consider it at this stage as a trivial or even normal event. These patients usually have many other more pressing social and economic problems than these tiny lesions, e.g. the short busy seasonal farming session, raising children in poor conditions and others. Most of the patients consider these early lesions are not a priority, and in fact, they believe that treatment will delay undertaking other urgent duties, and this explains the late presentation with massive lesions [28,31].
It is evident from this study that our holistic management has addressed many issues. The community engagement activities have led to active early case detection which is supported by the high number of patients with early disease seen at the WOMRC reported in this study. Such patients have a high chance of cure and were amenable to treatment with a good outcome [47,52].
The immediate access to free treatment at the village level has reduced patients’ delay in starting treatment, eliminated patients’ geographical and financial burdens, treatment interruption, and reduced the high follow up dropout rate. Treatment interruption can induce drug resistance. It is therefore vital to ensure sustainability and availability of the free mycetoma treatment services.
The health system in Sudan consists of the three levels; the rural, regional and central levels. The medical assistants, the health care providers, are the backbone for the management of mycetoma patients in Sudan at the rural level. Most of them have poor surgical experience and used to operate on the mycetoma patients under local anesthesia and suboptimal conditions. This practice has led to the high recurrence rate which was documented in many reports [1,2,53]. Recurrent disease is usually associated with wide local disease spread. Hence it is difficult to cure and necessitate repeated surgical excisions, numerous deformities and disabilities [1,2]. At Sennar State, most of the medical assistants were successfully trained on the different aspects of mycetoma care, management and referral.
A well-trained multidisciplinary team on mycetoma care was developed in Eastern Sennar region. It consists of a trained surgeon, surgical theatre attendants, anesthetic assistants, pharmacy assistant, ultrasound technicians, nurses, information technology expert, statistical clerks, community leaders and activists. This is an essential step in providing comprehensive and holistic management for the mycetoma patients.
To date, the definitive route of infection in mycetoma is an enigma. However, it is clear that mycetoma incidence is high in areas of poor environmental conditions, among people with poor personal hygiene and people living in proximity to animals and their dungs and where thorns, dirt and mud prevail. Hence this study aimed to improve the living and hygienic standards of Wad EL Nimear village, one of the highly endemic villages in the locality. The local villagers were encouraged to improve their living conditions. To achieve this goal, many advocacy and awareness campaigns were conducted, and new modern hygienic animals’ cages were constructed and donated to them free of charge. In support of these measures, the local Governate authorities issued a law banning the presence of animals inside the village. The kind donation of the animals’ cages by the engineering company as a response to the intensive mycetoma awareness and advocacy in Sudan.
In this study, the community leaders and activists were actively involved in conveying messages to their community in their own culture and traditions. This was important to accept these holistic disease management procedures. Likewise, the local villagers have actively engaged in promoting their health and improvement of local environmental conditions that believed to be the main source of transmitting mycetoma.
In conclusion, the holistic and comprehensive management approach implemented in this study has improved the mycetoma patients’ quality of care in the studied endemic area. More early disease was detected and treated. The treatment interruption rate was reduced thus increasing the cure rate and decreasing the recurrence and hospitalisation rates. This will eventually lead to decrease in the amputation and disability rates. The results obtained from this study suggest that such a study can be expanded to other endemic areas in the country. The MRC, as a WHO Collaborating centre on Mycetoma, will communicate this experience to the WHO to share it with other mycetoma endemic countries and assist in better management, prevention and control of the disease.
|
10.1371/journal.pbio.1002293 | CRY Drives Cyclic CK2-Mediated BMAL1 Phosphorylation to Control the Mammalian Circadian Clock | Intracellular circadian clocks, composed of clock genes that act in transcription-translation feedback loops, drive global rhythmic expression of the mammalian transcriptome and allow an organism to anticipate to the momentum of the day. Using a novel clock-perturbing peptide, we established a pivotal role for casein kinase (CK)-2-mediated circadian BMAL1-Ser90 phosphorylation (BMAL1-P) in regulating central and peripheral core clocks. Subsequent analysis of the underlying mechanism showed a novel role of CRY as a repressor for protein kinase. Co-immunoprecipitation experiments and real-time monitoring of protein–protein interactions revealed that CRY-mediated periodic binding of CK2β to BMAL1 inhibits BMAL1-Ser90 phosphorylation by CK2α. The FAD binding domain of CRY1, two C-terminal BMAL1 domains, and particularly BMAL1-Lys537 acetylation/deacetylation by CLOCK/SIRT1, were shown to be critical for CRY-mediated BMAL1–CK2β binding. Reciprocally, BMAL1-Ser90 phosphorylation is prerequisite for BMAL1-Lys537 acetylation. We propose a dual negative-feedback model in which a CRY-dependent CK2-driven posttranslational BMAL1–P-BMAL1 loop is an integral part of the core clock oscillator.
| The circadian system imposes daily rhythms on behavior, physiology, and metabolism and allows organisms to anticipate daily recurring changes in the environment. Circadian clocks are composed of clock genes, acting in transcription–translation feedback loops. At the heart of this molecular oscillator are the BMAL1–CLOCK transcription factors that drive expression of Cryptochrome (Cry) and Period genes, which in turn encode inhibitors of BMAL1–CLOCK–driven transcription. Phosphorylation and other posttranslational modifications also specify the activity and stability of clock proteins. Here, we unveiled the key mechanism underlying the rhythmic phosphorylation of BMAL1 at Ser90 by Casein Kinase-2 alpha (CK2α). Performing live monitoring of protein–protein interactions, we show that CRY proteins facilitate cyclic BMAL1–CK2β binding and subsequent cyclic inactivation of CK2α-mediated BMAL1-S90 phosphorylation; lack of this cyclic event abolishes circadian rhythmicity. We propose a dual negative-feedback model in which CRY proteins not only act as inhibitors of BMAL1–CLOCK transcription but also of BMAL1-S90 phosphorylation.
| The mammalian circadian system orchestrates a wide variety of metabolic, physiological, and behavioral rhythms through intracellular clockworks, present in the neurons of the suprachiasmatic nuclei (SCN) and in virtually all other cells and tissues [1]. At the heart of circadian time keeping is a molecular core oscillator consisting of a set of transcription factors (clock proteins) that operate in transcriptional–translational feedback loops [1–5] and drive rhythmic expression of approximately 10% of the mammalian transcriptome [6]. The BMAL1–CLOCK heterodimer is the primary genome-wide driver for transcription of clock genes (including three period [Per1, 2, and 3] and two cryptochrome [Cry1 and 2] clock genes) and clock-controlled output genes (CCGs) via binding to E-box containing promoters [7,8]. Periodic association of Cryptochrome (CRY)–PER complexes with BMAL1–CLOCK after the nuclear translocation of CRY–PER is facilitated by CLOCK-mediated BMAL1 acetylation, which represses the expression of E-box clock (controlled) genes and accounts for the negative limb of the circadian feedback loop [4]. The BMAL1–CLOCK complex also drives transcription of the retinoic acid receptor related orphan receptor α (Rorα) and nuclear receptor subfamily 1, group D, member 1 (NR1D1 or Rev-erbα) orphan nuclear receptor genes. RORα and REV-ERBα, after binding to RORE elements in the Bmal1 promoter, activate and repress Bmal1 transcription and stabilize the circadian oscillator by driving cyclic expression of Bmal1 [9]. Importantly, to date, only BMAL1- and CRY1/2-deficient mice exhibit immediate and complete loss in circadian rhythms [7,10].
Posttranslational modification represents an essential control feature of molecular oscillators in both prokaryotic and eukaryotic organisms [11] can specify longevity, activity, stability, and subcellular localization of core clock proteins. Indeed, mammalian clock proteins are important targets of posttranslational modification events [3–5,12], and their rhythmic phosphorylation appears to be a critical step for clock function [8,13–15]. For instance, Rev-erbα knockout mice (in which cyclic Bmal1 transcription is blunted, but rhythmic BMAL1 modification probably remains intact) still exhibit circadian rhythms [16], suggesting that cyclic BMAL1 modification is more critical in clock machinery than cyclic transcription.
Casein kinase (CK)-2α [17] rhythmically phosphorylates BMAL1 and is pivotal for regulating the mammalian circadian clock [12]. Depletion of CK2α and/or mutation of the CK2-phosphorylation site in BMAL1 (Ser90), results in impaired circadian nuclear BMAL1 accumulation and impairment of BMAL1–CLOCK–driven rhythmic CCG expression [12]. CK2 is shown to be an indispensable component of the mammalian molecular clock [12,18,19]. CK2 also phosphorylates PER2 to control PER2 degradation, and depletion of CK2 results in impaired clock gene oscillation because of lower amplitude of the expression rhythm and an extended period.
To date, the molecular mechanism underlying rhythmic mammalian clock protein phosphorylation remains elusive. Previously, we have reported circadian BMAL1 phosphorylation at Ser90 by CK2 [12], which is generally thought to be a constitutively active kinase [20]. Interestingly, the BMAL1 protein is hyperphosphorylated in CRY1/2-deficient mice [8,14], leading us to hypothesize that CRY proteins are involved in rhythmic BMAL1 modification. Here, we investigated the universal role and oscillatory mechanism of the circadian CK2-mediated BMAL1 phosphorylation. Accordingly, a novel role of CRY as a repressor for protein phosphorylation was found. We propose a model that explains how CRY proteins produce circadian oscillations and integrate posttranslational modification events (i.e., BMAL1 phosphorylation) in the negative limb of the core transcription–translation feedback loop.
To investigate the critical role of CK2-mediated circadian phosphorylation of BMAL1 at Ser90 (referred to as BMAL1-S90) in regulating the central clock in the SCN and peripheral clocks in the liver, and in cultured fibroblasts, we designed a competitive inhibitor of BMAL1-S90 phosphorylation, consisting of a 14 amino acid BMAL1 peptide (BMs90p), centered around Ser90 (Fig 1Aa). As expected, and in line with S90A mutagenesis data [12], BMs90p dose-dependently (optimum at ~6 μM) suppressed both the formation of BMAL1 phosphorylated at Ser90 (hereafter referred to as P-BMAL1-S90), and mPer2 promoter-driven luciferase (Per2L) bioluminescence rhythms in dexamethasone (Dex) clock-synchronized NIH-3T3 fibroblasts (Fig 1Aa, 1Ab and 1Ac). P-BMAL1-S90 was recovered only partially around 2-6h post-treatment (S1A Fig). In contrast, a control peptide with Ser90 replaced by Ala (BMa90p) did not inhibit Per2L rhythms and P-BMAL1-S90 phosphorylation, demonstrating the specificity of BMs90p (S1B Fig). Thus, BMs90p perturbs the circadian core oscillator (as evident from the suppressed Per2L rhythms) by inhibiting BMAL1-S90 phosphorylation. To test the effect on the central and peripheral clocks, we applied BMs90p to SCN and liver organotypic slices from mPER2Luc mice [21]. BMs90p-treatment provoked an evident reduction of the amplitude (liver; 0.255, SCN; 0.322; average value pre-treatment set as 1) and peak intensity (liver; 0.420, SCN; 0.525; average value pre-treatment set as 1) of Per2L (PER2::LUC) Per2L rhythms, without any evident phase-shifting effect (Fig 1B). Notably, this effect was only observed when BMs90p was administered at the trough of Per2L reporter gene activity (liver; ~CT5, SCN; ~CT2) (S2 and S3 Figs). Similar to the whole slice data, BMs90p suppressed Per2L rhythms in the SCN, as determined by imaging multiple (n = 24) small SCN cell clusters (Fig 1C, S4 Fig and S1 Movie). Taken together, these data strongly suggest a pivotal role of cyclic CK2-mediated BMAL1-S90 phosphorylation in the circadian oscillator of SCN neurons (central clock) and liver cells (peripheral clocks).
How are circadian oscillations in P-BMAL1-S90 levels generated? A clue may be found in previous observations that BMAL1 is constitutively hyperphosphorylated in CRY1/2-deficient (CRYdKO) cells with a dysfunctional clock [8,14].
We therefore first investigated whether hyperphosphorylation of BMAL1 in CRYdKO cells includes Ser90. As shown in Fig 2A and 2B, P-BMAL1-S90 level was significantly higher in CRYdKO cells than in wild-type (WT) cells. Importantly, expression of Cry1 promoter-driven Myc-CRY1 in CRYdKO cells (S5 Fig) caused the P-BMAL1-S90 level to return to WT levels (Fig 2A and 2B), suggesting that the CRY proteins act as suppressors of BMAL1-S90 phosphorylation.
Based on these results, we next focused on CK2-mediated BMAL1-S90 phosphorylation and the role of CRY proteins therein. Whereas the majority of CK2α catalytic subunits are likely recruited to a CK2β (regulatory unit) dimer to form a constitutively active CK2α2β2 tetramer that can phosphorylate a wide range of substrates [20], CK2β behaves as a strong inhibitor of CK2α-mediated BMAL1 phosphorylation in a dose-dependent manner [12]. Notably, CK2β does not directly inactivate CK2α [20]. Rather, CK2α activity is thought to fluctuate by the influence of yet unidentified cellular molecules.
As reported previously [12], we demonstrated that the CK2α monomer, but not CK2α2β2, phosphorylates BMAL1 at Ser90, and that CK2 kinase activity dramatically declined at a ratio of CK2β/CK2α ≥1 (Fig 2C). Similarly, we confirmed that CK2β inhibits CK2α-mediated BMAL1-S90 in vitro kinase activity as a function of the CK2β/BMAL1 ratio (Fig 2C). Kinase activity dramatically declined at a ratio of CK2β/BMAL1 ≥1, suggesting that CK2β interferes with CK2α monomer-mediated BMAL1-S90 phosphorylation by direct interaction with BMAL1. Indeed, in WT cell homogenates, CK2β is shown to co-precipitate with BMAL1 (Fig 2A). In marked contrast, however, BMAL1–CK2β interactions were significantly reduced in hyperphosphorylated P-BMAL1-S90 containing CRYdKO cells, while the amount of CK2α bound to BMAL1–CK2β complexes was comparable to that observed in WT cells (Fig 2A and 2B). Taken together, these data suggest that the amount of CK2α interacting with BMAL1 itself does not reflect the BMAL1 phosphorylation status. Rather, it points to a model in which CK2β is recruited to BMAL1 to inhibit CK2α activity. In the absence of BMAL1–CK2β interactions in CRYdKO cells, CRY proteins are likely candidates for such recruiting function.
We therefore assessed the binding ability of CK2 subunits to mammalian CRY1/2. In vitro, recombinant GST-tagged CK2α, α′ and β subunits could pull down CRY1 and CRY2 (S6A Fig). Thus, both CK2α and β subunits can bind to BMAL1, as well as to CRY1/2. Nonetheless, consistent with the data shown in Fig 2A, CRY proteins preferentially bind to CK2β. Notably, CRY1/2 can still interact with CK2β in BMAL1-deficient cells, demonstrating that BMAL1 is not required for CRY-CK2β interactions (S6B Fig). These results indicate that CRY proteins mediate BMAL1–CK2β binding by sequentially interacting with CK2β. Moreover, expression of Myc-CRY1 (resembling native CRY1 in that it also preferentially binds to CK2β; see Fig 2A) resets the level of BMAL1-bound CK2β to WT levels (Fig 2A and 2B). As the level of BMAL1-bound CK2α remained unchanged by Myc-CRY1 expression, we propose a model in which CK2α-mediated BMAL1-S90 phosphorylation is cyclically inhibited by CRY-dependent binding of CK2β to BMAL1, resulting in rhythmic P-BMAL1-S90 levels.
To test our hypothesis that the CRY-dependent periodic binding of CK2β to BMAL1 results in circadian P-BMAL1-S90 oscillation, we first examined the effects of CRY1/2 deficiency on the circadian pattern of P-BMAL1-S90 levels. As expected, and in contrast to the robust circadian oscillations in Dex-synchronized WT cells, P-BMAL1-S90 levels were constitutively expressed at high levels in CRYdKO cells (Fig 3 and S7 Fig). Moreover, periodic Myc-CRY1 expression (CRYdKO+CRY1) restored the circadian P-BMAL1-S90 oscillation, which peaked at a similar time (18–24 h after Dex treatment) as in WT cells (Fig 3B). The levels of BMAL1-bound CK2β in WT and Myc-CRY1 clock-rescued (CRYdKO+CRY1) cells exhibited robust circadian oscillation with peaks at 6–12 h and approximately 36 h with a phase nearly inverse to that of the P-BMAL1-S90 oscillation, whereas CRYdKO (with or without GFP) exhibited constitutively high P-BMAL1-S90 levels (Fig 3A and 3B). Reciprocally, CK2β-IP uncovered a similar temporal interaction pattern of BMAL1 with CK2β (S7 Fig). Moreover, CRY1 and CRY2 were shown to co-precipitate with BMAL1 or CK2β in a circadian manner and in phase with BMAL1–CK2β interactions (Fig 3A and S7 Fig). Notably, the circadian oscillation pattern of CK2β–CRY1/2 closely matched BMAL1–CK2β, rather than BMAL1–CRY1/2 rhythms (S7 Fig). As CK2β–CRY1/2 complexes can be formed in the absence of BMAL1 (S6B Fig), these data suggest that CRY proteins periodically facilitate BMAL1–CK2β interaction by first associating with CK2β. Moreover, circadian changes in posttranslational modification events may affect circadian patterns of BMAL1–CRY1/2–CK2β complexes. CK2α/β levels remained constant over time (S7 Fig), indicating that CK2α/β levels do not determine circadian P-BMAL1-S90 oscillation. These data lead us to conclude that the cyclic phosphorylation of BMAL1-S90 originates from periodic suppression of CK2α-mediated BMAL-S90 phosphorylation through CRY-mediated BMAL1–CK2β association.
To rule out the possibility of nonspecific associations in pull-down experiments, we next used a Split Luc complementation assay [22,23] for real-time monitoring of CK2β–BMAL1 interactions in living cells. In such an assay, bioluminescence can be detected only when N-(ELucN)- and C-(McLuc1 or ELucC)- tagged proteins associate and allow Luc moieties to complement each other and form active luciferase (Fig 4A). To this end, we ectopically expressed ELucN-CK2β and McLuc1/ELucC-BMAL1 in Cos7 and U-2OS cells at a level comparable to that of the native proteins (Fig 4A and 4C).
In Cos7 cells over-expressing Myc-tagged CRY1 or CRY2, bioluminescence due to BMAL1–CK2β interactions was significantly enhanced (approximately 6-fold; p < 0.001) (Fig 4B). This demonstrates that, consistent with the co-immunoprecipitation data obtained with CRYdKO-transfected MEFs (see Figs 2 and 3), BMAL1–CK2β binding in living cells requires CRY1/2. The in vitro CK2β-mediated inhibition of BMAL1-S90 phoshorylation (Fig 2C) and the BMAL1–CK2β binding in living cells (Fig 4B) strongly indicate that CRY proteins facilitate BMAL1–CK2β binding and subsequent suppression of CK2α-mediated BMAL1-S90 phosphorylation.
Dex-synchronized U-2OS cells exhibit robust circadian rhythmicity and can express high levels of ectopic proteins [24]. To investigate temporal changes in BMAL1–CK2β interactions, we monitored Split Luc activity in real-time mode in U-2OS cells and observed a robust circadian oscillation of BMAL1–CK2β binding, peaking approximately 15 h and 40 h after Dex-synchronization, and as such, inversely phased to P-BMAL1-S90 oscillations (Fig 4C and S8A and S8B Fig). These P-BMAL1-S90 patterns are consistent with those of asynchronous WT and CRY1-rescued CRYdKO MEFs (Fig 2A and 2B). As the RREx3/CMV promoter is constitutively active (see S8C Fig), the observed circadian BMAL1–CK2β Split Luc activity originates from rhythmic BMAL1-CK2β interaction rather than rhythmic BMAL1 expression. Ectopic expression of BMAL1 and CK2β in the Split Luc assay did not affect endogenous circadian phase or amplitude of the circadian core oscillator, as monitored through Bmal1-promoter driven luciferase activity (S9 Fig). Accordingly, the circadian patterns of BMAL1–CK2β association monitored by the Split Luc assay represent a nearly endogenous circadian pattern. Taken together with the result of Figs 3 and 4B, the antiphase circadian oscillations of BMAL1–CK2β interactions (as revealed by the binding Split Luc assay) and P-BMAL1-S90 rhythms strongly suggests that under physiological conditions (i.e., in the living cell) cyclic CRY-mediated BMAL1–CK2β association drives circadian BMAL1-S90 phosphorylation.
Next, we generated a panel of ELucC-mBMAL1 deletion constructs (Fig 5A) to determine the region of BMAL1 [25] critical for BMAL1–CK2β interaction. BMAL1–CK2β binding was detected at comparable levels in Cos7 cells expressing ELucN-CK2β with either ELucC-BMAL1-WT (full length) or deletion mutants in the N-terminal half of BMAL1 (Bd1-3), and was stimulated by co-expression of CRY1 or CRY2 (Fig 5B). However, irrespective of the presence of CRY1/2, expression of a BMAL1 mutant protein lacking the CRY-binding domain (Bd5) [26,27] resulted in significantly lower levels of bioluminescence (approximately 30%; p < 0.001) as compared to BMAL1-WT (Fig 5B), indicating that the C-terminal region of BMAL1 is critical for BMAL1–CK2β binding. Yet, BMAL1-Bd5:CK2β Split-Luc activities were significantly enhanced by co-expression of CRY1 (p < 0.001) or CRY2 (p < 0.001), suggesting that CRY binding to CK2β can also enhance CK2β binding to BMAL1 without direct physical interaction between CRY and BMAL1. Moreover, independent of the presence of CRY1/2, expression of the Bd4 mutant lacking the BMAL1 PAC (PAS-associated C-terminal) domain resulted in significantly higher levels of bioluminescence (approximately 2-fold; p < 0.001) as compared to BMAL1-WT (Fig 5B), implicating this region as a potential regulatory site for BMAL1–CK2β binding. CK2β binding with WT BMAL1, Bd4, and Bd5 was largely enhanced by ectopic CRY1/2 expression. These findings identified critical regions in BMAL1 for CRY-enhanced binding to CKβ.
To identify CRY1 protein regions critical for facilitating BMAL1–CK2β interaction, we generated a panel of mCRY1 deletion constructs (Fig 5C) for co-expression with ELucN-CK2β and ELucC-BMAL1 in Cos7 cells. CRY-facilitated BMAL1–CK2β binding, as detected by the Split-Luc assay, was not significantly altered by deletion of either the N-terminal DNA photolyase domain (Cd1) or the C-terminal region (Cd4) of CRY1 (Fig 5D). However, co-expression of a CRY1 mutant protein lacking the FAD-binding domain (Cd2) [28] resulted in significantly lower levels of bioluminescence (approximately 20%; p < 0.001) as compared to CRY1-WT (Fig 5D), indicating that the FAD-binding domain of CRY1 is critical in enhancing BMAL1–CK2β binding. Notably, co-expression of a mutant CRY1 protein (Cd3), lacking the pivotal region for inhibition of BMAL1–CLOCK-mediated transcription in the FAD-binding domain [28,29] also resulted in significantly lower levels of bioluminescence (approximately 40%; p < 0.001) as compared to CRY1-WT (Fig 5D). This finding indicates that the FAD-binding region is not only critical for suppression of CLOCK–BMAL1 -mediated transcription activation but also for CK2α-mediated BMAL1-S90 phosphorylation by facilitating BMAL1–CK2β binding.
BMAL1-K537 acetylation has been shown to facilitate the recruitment of CRY1/2 to BMAL1 [4]. To examine whether BMAL1-K537 acetylation affects S90 phosphorylation and CRY-mediated BMAL1–CK2β binding, we first stably transfected Bmal1-/- knockout MEFs with a mBmal1 promoter-driven wild type (Myc-BMAL1-WT) or acetylation mutant BMAL1 (Myc-BMAL1-K537R) expression construct. BMAL1-K537R retains the ability for S90 phosphorylation (S10 Fig), indicating that K537 acetylation is not prerequisite for S90 phosphorylation. S90 phosphorylation levels were even higher for BMAL1-K537R than for BMAL1-WT. Interestingly, CK2β and CRY1/2 binding to BMAL1-K537R was reduced as compared to BMAL1-WT (S10 Fig), suggesting that K537 acetylation is required for CRY-mediated recruitment of CK2β to BMAL1.
Next, we performed a Split-Luc live cell assay experiment using an ElucC-mBMAL1-K537R mutant protein. In the absence of CRY1/2, the BMAL1-K537R–CK2β binding was not significantly different from BMAL1–CK2β binding. However, in the presence of CRY proteins BMAL1–CK2β interactions increased by approximately 4-fold, while BMAL1-K537R–CK2β binding did not significantly increase (p < 0.001 in comparison with WT) (Fig 6A). Thus, CLOCK-mediated acetylation of K537 in the C-terminal region of BMAL1 appears critical for CRY-enhanced BMAL1–CK2β binding.
To further investigate whether BMAL1-K537 acetylation up-regulates BMAL1–CK2β binding and subsequently represses BMAL1-S90 phosphorylation, we utilized a mouse fibroblast cell line deficient in silent information regulator 1 (SIRT1), a member of the sirtuin family of NAD+-dependent histone deacetylases (HDACs) that also targets acetyl BMAL1-K537 [30]. As previously reported [30], the acetyl-BMAL1-K537 level was substantially higher in SIRT1KO (Sirt1-/-) MEFs than in WT MEFs (Fig 6Ba). Importantly, as predicted, P-BMAL1-S90 levels were significantly reduced in SIRT1KO MEFs (approximately 50% of control; p < 0.001) as compared to WT cells (Fig 6Bb). Consistently, BMAL1–CK2β binding was significantly increased (approximately 330% of control; p < 0.01) in SIRT1KO MEFs (Fig 6Bb). Consistent with our previous report [4,30], BMAL1–CRY1/2 binding was increased in SIRT1KO MEFs (Fig 6Bb).
Taken together, these data demonstrate that acetylation of BMAL1 at Lys537 facilitates BMAL1–CK2β association, and as such represses BMAL1-S90 phosphorylation.
For a variety of proteins phosphorylation has been shown to trigger subsequent acetylation events [31,32]. Accordingly, S90 phosphorylation is potentially involved in the regulation of BMAL1 acetylation. To assess the link between BMAL1-S90 phosphorylation and BMAL1-K537 acetylation and to further establish the integral role of the CK2-mediated BMAL1-S90 phosphorylation in the circadian core oscillator, we examined the effect of BMAL1-S90 mutation on BMAL1-K537 acetylation, which, as shown above, is, critical for CRY-mediated BMAL1–CK2β binding and P-BMAL1-S90 rhythms.
To this end, we stably expressed wild-type BMAL1 (BMAL1-WT), mutant BMAL1-S90A (with Ser90, located in the basic helix-loop-helix DNA binding motif, substituted for Ala) or GFP (negative control) in MEFs derived from clock-deficient Bmal1-null mice [7]. As shown in Fig 7A, Myc-BMAL1-WT and Myc-BMAL1-S90A were expressed at equal level. The acetylation level of BMAL1-S90A was dramatically reduced as compared to BMAL1-WT (approximately 20% of control; p < 0.001; Fig 7A and 7B). Thus, CK2α-mediated BMAL1-S90 phosphorylation might be a prerequisite for CLOCK-mediated BMAL1 K537 acetylation. Consistent with this notion, BMAL1–CLOCK levels were decreased (approximately 50% of control; p < 0.002) by the S90A mutation (Fig 7A and 7B), suggesting that reduction in BMAL1–CLOCK association is the cause of the reduced BMAL1-K537 acetylation level.
Furthermore, the BMAL1–CRY1/2 interactions were also significantly decreased by the S90A mutation (Fig 7A and 7B), indicating that S90 phosphorylation is prerequisite for K537-facilitated CRY recruitment to BMAL1. As previously reported, reduction of BMAL1–CRY interactions by K537R mutation demonstrates K537-acetylation-dependent recruitment of CRY1/2 to BMAL1 [4]. Thus, CK2α-mediated BMAL1-S90 phosphorylation is a prerequisite for CLOCK-mediated BMAL1-K537 acetylation, subsequent CRY1/2 recruitment to BMAL1 and CRY1/2-facilitated BMAL1–CK2β binding to regulate negative feedback suppression of clock gene expression and CK2α-mediated BMAL1-S90 phosphorylation, respectively. Our results do not exclude the existence of other BMAL1 phosphorylations, which directly trigger BMAL1-acetylation and other modifications, located downstream of CK2-mediated BMAL1-S90 phosphorylation.
Circadian BMAL1-S90 phosphorylation has been shown to be an important regulatory step in the mammalian core clock oscillator [12]. In the present study, we addressed the underlying mechanism and uncovered a vital interplay between CRY proteins and circadian BMAL1 phosphorylation. First, by applying a novel clock-perturbing peptide (BMs90p) to SCN and liver organotypic slices from mPER2Luc mice and subsequent live monitoring of circadian clock performance, we further highlighted a universal critical role of BMAL1-S90 phosphorylation in central and peripheral clocks. BMs90p (a small 14 amino acid peptide containing the BMAL1-Ser90 phosphorylation site targeted by CK2) behaves as a competitive inhibitor of BMAL1-S90 phosphorylation and was shown to reversibly blunt Per2L bioluminescence rhythms in a dose- and circadian time-dependent manner
Next, triggered by the observation that a CRY1/2-deficiency causes hyper-phosphorylation of BMAL1 [8,14], we focused on the molecular mechanism underlying circadian BMAL1-S90 phosphorylation and showed that in wild-type cells, circadian phosphorylation of BMAL1-S90 is accompanied by inverse phase cyclic association of BMAL1 with CK2β, a known inhibitor of CK2α-mediated BMAL1 phosphorylation [12]. Notably, a CRY1/2-deficiency abolishes BMAL1-CK2β interactions, and as such prevents cyclic inhibition of BMAL1-S90 phosphorylation, resulting in constitutively hyperphosphorylated BMAL1. P-BMAL1-S90 in CRY1/2 deficient cells could be rescued by rhythmic Cry1 expression, which points to a model in which CRY proteins cyclically recruit CK2β to BMAL1 to inhibit CK2α activity.
To provide further evidence for this model, we developed a Split-Luc–based assay system for real-time monitoring of clock protein–protein interactions in living cells. Using this assay, we have shown that BMAL1 cyclically binds to CK2β and that circadian BMAL1–CK2β binding is enhanced by CRY proteins. Moreover, using the same Split-Luc approach in combination with mutant versions of the BMAL1 protein, we have shown that the PAC and CRY-binding domains in the C-terminal region of BMAL1, as well as BMAL1-K537 acetylation (known to enhance CRY-recruitment to BMAL1 [4]) are important in regulating BMAL1–CK2βbinding. Indeed, using SIRT1KO cells, we demonstrated that BMAL1-K537 hyper-acetylation reduces BMAL1-S90 phosphorylation through enhanced CRY-driven BMAL1–CK2β association. As BMAL1-S90 phosphorylation is prerequisite for BMAL1-K537 acetylation (see below), the low but significant P-BMAL1-S90 level in SIRT1KO MEFs is apparently sufficient to trigger BMAL1-K537 acetylation (Fig 6B). Reciprocally, BMAL1-S90A expressing MEFs, lacking BMAL1-S90 phosphorylation, cannot trigger significant BMAL1-K537 acetylation (Fig 7A and 7B).
By BMAL1-S90A mutagenesis, we showed that BMAL1-S90 phosphorylation is prerequisite for BMAL1-K537 acetylation. The S90A mutation significantly reduces the nuclear BMAL1–CLOCK levels [12,33] and S90-phosphorylated BMAL1 is mostly detected in the nuclear fraction (S11 Fig), strongly suggesting BMAL1 enters the nucleus promptly after S90 phosphorylation. Taken together with the lower K537 acetylation and CLOCK binding capacity of BMAL1-S90A, as compared to BMAL1-WT, it is assumed that K537 acetylation mainly occurs after BMAL1–CLOCK nuclear entry. The enhanced K537 acetylated/S90-phosphorylated BMAL1 level in Sirt1 knockout cells suggests a mutual regulatory loop between K537 acetylation and S90 phosphorylation and supports the notion that S90 phosphorylation is prerequisite for K537 phosphorylation, while K537 acetylation represses S90 phosphorylation.
In conclusion, we established a circadian clock-controlling role of CK2 kinase, formerly thought to be a constitutively active kinase [20] in BMAL1 phosphorylation and uncovered a novel role of CRY as a regulator of cyclic CK2-mediated BMAL1 phosphorylation. Fig 8 illustrates our model for the molecular mechanism of the CK2-mediated posttranslational loop and its role in regulating the intracellular circadian core oscillator. In this model, cyclic CK2α-mediated BMAL1-S90 phosphorylation serves as the periodic gateway that controls BMAL1–CLOCK heterodimerization (step I) and time-delayed nuclear accumulation of BMAL1–CLOCK (step II) [12]. Step I and II may play a critical role in the events described below and serve as a time-delay factor that fine-tunes the circadian periodicity [14]. Therefore, we refer to CK2-mediated BMAL1-S90 phosphorylation as the first gate, probably located at the boundary between the cytoplasm and nucleus. Consistently, constitutive nuclear predominant BMAL1 localization in CRYdKO MEFs through the circadian cycle might be largely due to constitutive active BMAL1-S90 phosphorylation [14]. After BMAL1–CLOCK accumulates in the nucleus, E-box promoter containing clock genes, including CRY1/2, are temporally transactivated (step III). This is followed by negative feedback suppression of BMAL1–CLOCK transcription of E-box genes by the recruitment of CRY1/2 to BMAL1 (step IV), which is regulated by CLOCK-mediated BMAL1-K537 acetylation [4] and requires phosphorylated BMAL1-S90. In the next step (step V), because of the delayed surge in CRY1/2–CK2β binding, the BMAL1–CLOCK–CRY complex is released from the E-box. Thereafter, we hypothesize that CRY proteins are released from the complex to make way for newly incoming CRY1/2–CK2β complexes that bind to BMAL1–CLOCK via direct CRY–BMAL1 interaction. Deletion of the CRY-binding domain in BMAL1 does not completely abolish CRY-mediated enhancement of BMAL1–CK2β binding in the Split Luc assay (Fig 5), suggesting direct BMAL1–CRY interaction is not absolutely necessary for the enhancement of BMAL1–CK2β binding. Given that CRY has also been shown to bind to CLOCK [29], docking of CRY–CK2β to BMAL1–CLOCK may also involve CRY–CLOCK interactions. Through formation of CRY1/2–CK2β intermediates, CRY1/2 facilitates BMAL1–CK2β association. Notably, the release of CRY from and re-entry of CRY–CK2β in the BMAL1–CLOCK complex (instigated by the observation that CRY can bind CK2β in the absence of BMAL1) is the most speculative step in the model. In the absence of experimental/mechanistic evidence we cannot fully exclude that CRY proteins enhance BMAL1–CK2β binding while still bound to the BMAL1–CLOCK–CRY complex.
In vitro, CK2β can bind to BMAL1 in the absence of CRY and inhibit BMAL1-S90 phosphorylation by CK2α (Fig 2C and S6A Fig). Under these non-physiological conditions, inhibition of S90 phosphorylation may occur through (simultaneous) formation of CK2β-BMAL1 complexes that prevent CK2α from binding to BMAL1, and/or formation CK2α2β2 tetramers [20], which are probably incapable of phosphorylating BMAL1 [12]. However, as formation of a CK2α2β2 tetramer requires formation of a CK2β dimer that subsequently binds two CK2α monomers, and as in vitro BMAL1-S90 phosphorylation by CK2α is maximally inhibited at a CK2β:BMAL1 ratio of 1 (Fig 2C), interaction of 1 CK2β monomer rather than a tetramer with 1 BMAL1 molecule appears sufficient to inhibit S90 phosphorylation. In vivo, BMAL-CK2β association appears to block (BMAL1-bound) CK2α activity, rather than BMAL1–CK2α association, and requires the help of CRY proteins. Although we do not exclude a model in which CK2β is included in the BMAL1–CLOCK–CRY–CK2β–CK2α as a dimer or α2β2 tetramer complex, we consider inhibition of CK2α-mediated BMAL1-S90 phosphorylation by a CK2β monomer the most plausible option (step V). Taken together, inverse-phased circadian BMAL1–CLOCK–CRY–CK2β−CK2α complex formation might be the primary determinant for circadian CK2α-mediated BMAL1-S90 kinase activity. Next, S90 phosphorylated BMAL1 undergoes a SIRT1-mediated deacetylation step (step VI) [30] that likely liberates BMAL1 from the complex. Subsequent SUMOylation [3] and ubiquitination [34] of BMAL1 may target the protein for proteasomal degradation. In addition, non-degraded BMAL-S90 needs to be dephosphorylated by yet unknown phosphatases to initiate a new cycle through CK2α-mediated phosphorylation of BMAL1.
BMAL1-S90 phosphorylation by the CK2α monomer most likely occurs at step I (Fig 8). S90-phosphorylation of BMAL1 takes place in the cytoplasm and triggers CLOCK binding and subsequent BMAL1–CLOCK nuclear accumulation [12]. We have shown that BMAL1–CK2α complexes exist throughout the circadian cycle (S7 Fig). This suggests that the CK2α monomer remains bound to the BMAL1–CLOCK complex up to step IV. Likely, CK2α remains catalytically active (though its substrate is no longer available) and only gets inactivated after CRY-mediated binding of CK2β (step V).
In this study, we have unveiled the underlying mechanism for the cyclic CK2-mediated BMAL1 phosphorylation as a critical event in the mammalian circadian core clock machinery. BMAL1–P-BMAL1 loop forms a distinct interlocked loop in the clock machinery (step I, V, and VI) and have integral roles in the core circadian oscillator through periodic CRY-mediated negative feedback suppression. In this scenario, CRY proteins have a dual function. Strikingly, in addition to their known function as repressors of BMAL1–CLOCK-driven transcription, we found a novel role of CRY proteins as a repressor of CK2 protein kinase activity toward BMAL1-S90. Notably, we observed that the FAD binding region of CRY1, known to be essential for repression of BMAL1–CLOCK-driven transcription [28], is also critical for inhibition of CK2α-mediated BMAL1-S90 phosphorylation.
Interestingly, CLOCK-mediated BMAL1-K537 acetylation [4], through sequential recruitment of CRYs and then CRY1/2–CK2β to the BMAL1–CLOCK complex, acts as a common molecular key for evoking CRY-mediated feedback inhibition of BMAL1–CLOCK transcription activity and CRY-dependent suppression of BMAL1 phosphorylation. Ultimate verification of our model ideally requires in vitro reconstitution of the CRY-driven circadian BMAL1–P-BMAL1 loop, as shown for cyanobacterial KaiC phosphorylation [35]. In a first experiment in which purified recombinant CRY1 was added to the in vitro BMAL1-S90 phosphorylation assay (as performed in Fig 2C), we observed that despite its ability to bind BMAL1 (S12A Fig), CRY1 could not inhibit CK2α-mediated BMAL1-S90 phosphorylation (S12B Fig). This apparent difference markedly contrasts with the in vivo data, where CRY has been shown pivotal for CK2β-mediated inhibition of BMAL1-S90 phosphorylation by CK2α. Clearly, in vitro assays differ from the in vivo situation in that they do not take into account the effect of subcellular localization of the proteins studied, their interaction with DNA or chromatin, or the involvement of other protein partners. Moreover, in vitro synthesized proteins probably do not undergo posttranslational modification, leading us to hypothesize that CRY can only recruit CK2β to BMAL1 after acetylation of BMAL1-K537.
CK2 phosphorylates a large array of cellular proteins and is widely involved in regulating mammalian physiology [17]. However, temporal aspects of CK2 function are still elusive. Therefore, in addition to its role in the core clock, future investigations should focus on CK2-mediated circadian signaling as a regulator of various physiological and pathological pathways. A genome-wide phospho-proteomics study of periodic signaling systems focused on CK2 may help elucidate the chronobiological attributes of diverse physiological events and facilitate the development of therapies for circadian-system–related disorders [36], such as metabolic syndromes, cancer, and neuropsychiatric diseases. Recently, we demonstrated that CK2-BMAL1 kinase plays a critical role in controlling protective pathways evoked by reactive oxygen species and is crucial for preventing oxidative-stress–related diseases [37].
Immunoprecipitation and immunoblotting were performed using sample solutions as previously described [4,12,14]. Antibodies utilized in the immuno-detection assays included anti-BMAL1 [14], PER1, BMAL1-phospho-Ser90 (P-BMAL1-S90) and BMAL1-acetyl-Lys537 (Acetyl-BMAL1) previously generated in our laboratories [12,30], CK2β (Calbiochem, San Diego, CA, United States), CK2α, actin, RNA polymerase II (Santa Cruz. Biotechnology Inc., Santa Cruz, CA, US), Myc-tag (Upstate Biotechnology Inc., Lake Placid, NY, US), CLOCK (Affinity BioReagents, Golden, CO, US), CRY1 (kindly donated by Dr. Todo), CRY2 [14], His-tag, FLAF-tag (MBL Co. Ltd., Nagoya, Japan), and HRP-conjugated anti-rabbit/goat/mouse IgG (Zymed, South San Francisco, CA, US). Immunoblot data were quantified by computerized densitometry and statistical analysis were performed as described previously [12,14,38]. The density of the protein bands was normalized to actin levels and the value of the control sample was set as 1. Statistical analysis was performed using the Student’s t test. Calculated error bars indicate standard deviation (SD).
In vitro kinase assays were performed as described previously [12,13], using CK2 subunits, 1 mM ATP, and GST-BMAL1, with/without His-CRY1 (see below). CK2α, CK2β, and GST-BMAL1 were prepared as described previously [12]. Kinase activities were measured by immunoblot using an anti-P-BMAL1-S90 antibody, and quantified as described above. Immunoblot and kinase assay data were normalized to the control values.
GST-CK2α, α’, and β subunits were expressed in bacteria, purified and analyzed by CBB staining. Mammalian clock proteins labeled with 35S-methionine were produced using the TNT Quick Coupled Transcription/Translation system (Promega) with expression vectors for BMAL1 (donated by Dr. Ikeda) and V5-CRY1/2 (donated by Dr. Reppert). CK2 subunits and clock proteins were mixed (combinations as indicated in the text), incubated, and affinity-precipitated with glutathione Sepharose beads. Recombinant His (and V5)-tagged CRY1 protein was expressed in High Five insect cells using pIB/V5-His vector and InsectSelect System (Invitrogen, CA, US). His-CRY1 protein was purified using Talon-resin (Clontech, CA, US).
The plasmid vector pGL4-Per2, constructed to express the ~1.7 kb mPer2 promoter-driven luciferase construct (Per2-Luc), has been described previously [12]. The vector pGL4-Bmal1 was constructed to express the ~1 kb mBmal1 promoter-driven luciferase. The pcDNA (Invitrogen, Carlsbad, CA, US)-based vectors Myc-HA-mCRY1, Myc-mCRY2, FLAG-mCRY1 and the deletion mutants (Cd1–4; Fig 5C) were constructed to express CMV promoter-driven CRY1/2.
Retroviral expression vectors (pCLNCX) for the ~1.6 kb mPer3 promoter [39]-driven luciferase (Per3-Luc) and the ~1 kb mCry1 promoter-driven Myc-CRY1 and CMV-promoter-driven GFP constructs have been described previously [12]. In addition, these experiments utilized previously prepared expression constructs for mBmal1-promoter-driven BMAL1-WT/mutants and CMV-promoter-driven GFP [4,12].
For the Split Luc complementation assay [23], the cDNA of Emerald Luciferase (ELuc) was obtained from Toyobo Co. Ltd. (Osaka, Japan). The multiple complement luciferase fragment cDNA construct (McLuc1) was generated as described previously [22]. The full-length mouse BMAL1 and human CK2β cDNAs were ligated downstream of the C-terminal (McLuc1 or ELucC) and N-terminal (ELucN) luciferase fragments, respectively (Fig 4A). Full-length BMAL1, BMAL1 deletion mutants (Bd1–5; Fig 5A), and CK2β cDNAs, were ligated downstream of ELucC and ELucN, respectively. Each cDNA fragment was amplified by polymerase chain reaction (PCR) and inserted into pcDNA4/V5-His (B) or pcDNA3.1 (Invitrogen, Carlsbad, CA, US) vector backbones using multiple cloning sites. In the ELucC/McLuc1-BMAL1 expression vector, three repeats of the BMAL1 Rev-erbA/ROR binding element (RRE) were added. To test promoter performance, the McLuc1-BMAL1 and ELucN-CK2β coding sequences were replaced by Luc (from pGL4), giving rise to the RREx3-CMV-Luc and CMV-Luc expression vectors. The sequences of all expression vectors were checked using a genetic analyzer (ABI310; Applied Biosystems, Foster City, CA, US). For long-lasting (stable) expression of the plasmids in human cells, episomal-type vectors were constructed by replacement of the CAG-promoter-MCS in pEBMulti-Hyg (Wako, Osaka, Japan) with CMV-ELucN-CK2β and RREx3-CMV-McLuc1-BMAL1 constructs. The expression vectors were purified with the PureLink HiPure Plasmid Filter Midiprep Kit (Invitrogen, Carlsbad, CA, US) prior to transfection of mammalian cells.
Established (NIH-3T3) and primary (MEF) mouse embryonic fibroblasts, and human osteosarcoma (U-2OS) cells were cultured as previously described [12,24,30,37]. BMAL1-deficient (KO) MEFs [7] were kindly donated by Dr. Bradfield. For clock synchronization, cells were treated with 0.1 μM dexamethasone (Dex) for 2 h. DNA transfection was performed using FuGENE HD according to manufacturer’s protocol (Roche Diagnostics Basel, Switzerland). The peptides BMs90p (RRDKMNSFIDELAS, a 14 amino acid BMAL1 peptide centered around BMAL1 Serine 90) and BMa90p (RRDKMNAFIDELAS, a negative control peptide with serine 90 replaced by alanine) were custom-made by Gen Script (Piscataway, NJ, US). Peptides were dissolved in water and applied to actively growing cultured cells at ≤60% confluence. The RetroMax expression system (Imgenex, San Diego, CA, US) was used to produce retrovirus for the rescue experiment. Retroviral infection was performed as previously described [12]. Real time bioluminescence activities were monitored using the Kronos Dio system (ATTO, Tokyo, Japan) as previously described [37].
All animal experiments were approved by the Toho University Animal Committee, and carried out under the control with Guidelines for Proper Conduct of Animal Experiments by Science Council of Japan. mPER2Luc mice (B6.129S6-Per2tm1Jt/J)21 were purchased from Jackson Laboratories (Bar Harbor, ME, US) and maintained at 25°C on a 12 h light/dark (LD) cycle (light: zeitgeber time [ZT] 0–12; dark: ZT 12–24). Preparation of organotypic slices from 4–8 week old mice, real-time bioluminescence assay, and microscopic imaging analysis (using the LV200 Bioluminescence Imaging System; Olympus, Tokyo, Japan) and MetaMorph analysis (MetaMorph, Nashville, TN, US) were performed using previously published procedures [40], briefly described below. The reduction of the peak (Fig 1Bb) was quantified by evaluating the difference in peak values in Fig 1Ba over 2 d before and after BMs90p treatment. The reduction of rhythm amplitude (Fig 1Bb) was quantified by comparing peak and trough values in the detrended data (S3 Fig) over 2 d before and after BMs90p treatment.
Values obtained from bioluminescence analyses were normalized by the maximum peak intensities over time and further normalized by the averaged intensity over time, as described previously [37,41]. Real-time bioluminescence in cell cultures and organotypic slices treated with 0.2 mM Luciferin (Toyobo) were monitored using the Kronos Dio system with acquisition times of 2 (promoter-Luc assays) or 3 min (Split-Luc assays), according to the manufacturer’s protocol. Values were obtained from each sample in a given experiment using the same detectors. The n-values indicated for each experiment refer to the number of samples analyzed with the same detectors in the same experiments. The y-axis label “Bioluminescence” indicates that the relative photo-counting values reflect arbitrary units (a.u.) from raw data; “RLU” (Relative Light Units) indicates that the relative photo-counting values were normalized by averaging intensity over time. The y-axis label “Deviation from the moving average” indicates that the values were detrended according to the Kronos Dio instrument protocol (ATTO). In many cases, as indicated in the figure legends, detrended values were further normalized by averaging intensity over time. The data in these graph labeled "deviation from the moving average” were further normalized using maximum circadian peak intensities over time. Real-time bioluminescence for microscopic imaging was monitored using the LV200 microscope with acquisition times of 48 min (EM-gain = 400) according to the manufacturer’s protocol (Olympus). Values obtained from each tracked region of interest (ROI) surrounding neighboring small clusters of cell-areas were processed similarly.
We used factorial design analysis of t test to analyze data as appropriate. The data presented in this study represent the average of multiple experiments, as specified in the figure legends.
|
10.1371/journal.pcbi.1000041 | A Hidden Feedback in Signaling Cascades Is Revealed | Cycles involving covalent modification of proteins are key components of the intracellular signaling machinery. Each cycle is comprised of two interconvertable forms of a particular protein. A classic signaling pathway is structured by a chain or cascade of basic cycle units in such a way that the activated protein in one cycle promotes the activation of the next protein in the chain, and so on. Starting from a mechanistic kinetic description and using a careful perturbation analysis, we have derived, to our knowledge for the first time, a consistent approximation of the chain with one variable per cycle. The model we derive is distinct from the one that has been in use in the literature for several years, which is a phenomenological extension of the Goldbeter-Koshland biochemical switch. Even though much has been done regarding the mathematical modeling of these systems, our contribution fills a gap between existing models and, in doing so, we have unveiled critical new properties of this type of signaling cascades. A key feature of our new model is that a negative feedback emerges naturally, exerted between each cycle and its predecessor. Due to this negative feedback, the system displays damped temporal oscillations under constant stimulation and, most important, propagates perturbations both forwards and backwards. This last attribute challenges the widespread notion of unidirectionality in signaling cascades. Concrete examples of applications to MAPK cascades are discussed. All these properties are shared by the complete mechanistic description and our simplified model, but not by previously derived phenomenological models of signaling cascades.
| Cellular signaling is carried out by a complex network of interactions. A structure that is found commonly in signaling pathways is a sequence of on-off cycles between two states of the same protein, referred to as a cascade. By analyzing and reducing the basic kinetic equations of this system, we have constructed a new mathematical model of an intracellular signaling cascade. It is widely accepted that information travels both outside-in and inside-out in signaling pathways. Conversely, cascades, even while being main components of those pathways, have been so far understood as structures where signal transmission occurs in a manner analogous to a domino effect: the information flows in only one direction. Adding explicit connections linking a particular level with an upstream location has been the way bidirectional propagation has been explained so far. In other words, up to now, unidirectional cascades would allow bidirectional propagation only when embedded in more complicated circuits. The proposed model shows that a cascade can naturally exhibit bidirectional propagation without invoking extra re-wiring. This result inspires novel interpretations of experimental data; since signaling pathways are usually reconstructed from such data, this outcome could have far-reaching implications in the understanding of cell signaling.
| Covalent modification cycles are one of the major intracellular signaling mechanisms, both in prokaryotic and eukaryotic organisms [1]. Complex signaling occurs through networks of signaling pathways made up of chains or cascades of such cycles, in which the activated protein in one cycle promotes the activation of the protein in the next link of the chain. In this way, an input signal injected at one end of the pathway can propagate traveling through its building-blocks to elicit one or more effects at a downstream location.
Examples of covalent modification are methylation-demethylation, activation-inactivation of GTP-binding proteins and, probably the most studied process, phosphorylation-dephosphorylation (PD) [1],[2]. In such cycles, a signaling protein is activated by the addition of a chemical group and inactivated by its removal. This protein is modified in turn by two opposing enzymes, such as a kinase and a phosphatase for PD cycles. In the absence of external stimulation, a cycle exists in a steady state in which the activation and inactivation reactions are balanced. External stimuli that produces a change in the activity of the converting enzymes, shifts the activation state of the target protein, creating a departure from steady state which can propagate through the cascade.
The advantages of these cascades in signal transduction are multiple and the conservation of their basic structure throughout evolution, suggests their usefulness. A reaction cascade may amplify a weak signal, it may accelerate the speed of signaling, can steepen the profile of a graded input as it is propagated, filter out noise in signal reception, introduce time delay, and allow alternative entry points for differential regulation [3]–[5].
Intracellular signaling through cascades of biochemical reactions has been the subject of a great number of studies (e.g., [2],[6] for reviews). Theoretical investigations have been motivated by the increased need for developing an abstract framework to understand the vast amounts of experimental data now available. This whole field of research is further motivated by the hope of characterizing pathways that are deregulated in diseases such as cancer and to define targets to combat these diseases [7].
Since the stimuli a cell receives are varied and complex, cascades do not operate in isolation, but rather the integration of stimuli depends on crosstalk between pathways. Another crucial property of signaling cascades is their ability to integrate information by transmitting the effects downstream and also feedback upstream. In spite of a few decades of intense work on signaling cascades, no models have ever been built that exhibit crosstalk with backwards and forwards transmission of a lateral input from another cascade, except when ad hoc feedback is explicitly added to the cascade model. Our model, built from first principles, naturally exhibits these characteristics and therefore inspires novel interpretations of experimental data.
A well studied example of a cascade of activation-inactivation cycles is the cascade of protein kinases. In this case, the basic signaling unit is a PD cycle, whose activating kinase is the phosphorylated protein of the previous cycle. Many proteins contain several phosphorylation sites, allowing for great versatility of regulation. Such is the case, for example, for the mitogen-activated protein kinase (MAPK) cascade, which is widely involved in eukaryotic signal transduction [3], [8]–[10]. For the sake of simplicity, in this article we will mostly consider cascades composed of simple, 2-state activation-inactivation cycles. However, the equations corresponding to the MAPK cascade are also derived and some of their properties compared with those of the simpler cascades. Even though our results are valid in general, for covalent modification cycles, we will employ the nomenclature associated with PD cycles, i.e. the converting enzymes will be referred to as kinase/phosphatase.
The focus of our study is to refine the mathematical modeling of cascades of covalent modification cycles, such us the one depicted in Figure 1. Several mathematical descriptions have been developed to describe such cascades using ordinary differential equations. Typically, those descriptions are built up starting with a model for a single cycle, which is then phenomenologically incorporated into a cascade of cycles. A well known model for describing the single cycle was introduced by the pioneering work of Goldbeter and Koshland (GK) [11]. The GK model considers the concentration of the target protein to be in large excess over those of the converting enzymes, thereby reducing the description to a single equation per cycle. The model obtained in this way was then phenomenologically extended to a cascade of individual GK cycles. Here, by the designation “phenomenological” we mean that, in the cascade, the forward coupling between the GK cycles is chosen as simply as possible, but not strictly deduced from first principles. This phenomenological framework extension of the GK model will be denoted as the GK-like model. The GK-like model has been used by several authors to describe the dynamics of signal transduction [9], [12]–[16]. For particular limiting cases, the GK-like model can be simplified further, which results in a model where the inter-converting reactions follow linear rate laws with first-order rate constants. This description was studied in several key papers [17]–[19], and we will refer to it as the linear rates model.
The concept of a “cascade” in the study of transduction pathways is appealing because of its modular structure. What is especially appealing is the possibility of defining the cascade state by only one variable per module. As mentioned above, since the building blocks of the GK-like model are the well-studied GK cycles, they involve only one equation per cycle. A different approach however, is to deal with the dynamics of the cascade of Figure 1 by considering the complete set of biochemical reactions and by writing the corresponding equations without any upfront approximations. This was accomplished, for example, for the case of the MAPK cascade [8]. We will refer to this approach as the mechanistic model. For the purposes of this paper, we will consider that the mechanistic model represents a complete description of the system under study (event though we recognize that, in reality, it is not a hypothesis-free model).
In this article, starting from the mechanistic description of a cascade composed of an arbitrary number of cycles, we derive a consistent approximation under which the cascade is described with one variable per cycle. It turns out that in this derivation, referred to as a reduced mechanistic description, the phenomenological GK-like model is not recovered. At first sight, our new approximation differs slightly from the previously derived description for signaling cascades. However, it involves qualitatively different dynamics from the GK-like model, yet it is in very good agreement with the complete mechanistic description when the approximation conditions are fulfilled.
The main difference between our simplified mechanistic description and the phenomenological one is the appearance of an intrinsic feedback from each unit to the preceding one, caused by the fact that in each cycle there is sequestration of part of the activated protein of the previous step. The new description of the cascade predicts the existence of damped oscillations along the chain, a phenomenon that cannot be observed using the previous phenomenological description. Interestingly, a corollary of our model is that if a particular unit in the middle of the chain receives an input–a common event, given the high degree of crosstalk between signaling pathways–then our reduced mechanistic description predicts that this perturbation is able to travel both forwards and backwards. This “bicistronic” propagation, which may be critical for effective eukaryotic signaling, is not possible within the GK-like description either. Our model provides a suitable framework for future experiments that investigate crosstalk and bicistronic propagation of signals.
In Text S1 we derive in detail the new class of model equations obtained as an approximation of the mechanistic model. The goal of our approach is to reduce the number of variables in the complete system by bringing into play hypothesis that allow us to use the quasi-steady state approximation. Three key dimensionless parameters are defined to facilitate the analysis:(5)εi and ηi are ratios of total amounts of proteins. εi is the ratio of total phosphatase over total targeted protein. ηi is defined as the total targeted protein in one cycle over the corresponding amount in the next cycle in the cascade, or, equivalently, the ratio of total kinase over total targeted protein. The parameter µi is the ratio of the kinetic rates of product formation in both the activation and the inactivation reactions (see reactions in Equation 1).
Using a standard singular perturbation analysis, we have found that the state of each biochemical cycle can be described by a single variable defined as xi = yi*+ci+1, which is the natural slow variable describing the total kinase i available at a given time for the phosphorylation in cycle i+1. This reduction is only valid if the total phosphatase in the cycle is much lower than the total targeted protein, i.e., in the limit εi«1. The other parameters must satisfy µi ηi∼εi. The dynamics of xi is described by the differential equation:(6)with the following conservation equation from which yi has to be extracted:(7)x0 = S is the normalized input signal and yn+1 = 0. In Equation 6, Vi = (k′i µi ηi)/(ε k′) and V′i = (εi k′i) /(ε k′), where ε k′ is a typical number representing the set of εi k′i (i = 1, … ,n), e.g. the arithmetical or the geometrical average over this set. In the conservation equation (Equation 7), the notation O(εi) is just a reminder that this equation is written in the lowest order in εi, as is also the case for the differential equation for xi. In Text S1, we discuss an improvement of this conservation equation which takes into account the first correction in εi. Although this extension does not alter the new properties discussed below, its numerical integration is easy and it increases the accuracy of the approximation.
The reduced system given by Equations 6–7 seems to be, in principle, equivalent to the GK-like model given by Equation 4. However, two main features make it significantly different. First, in our novel system, termed the reduced mechanistic model, the conservation equation depends on the variable of the previous cycle. Second and more interesting, the denominator of the negative term in Equation 6 is now a function of the next variable yi+1, in contrast to the GK-like model. This function has the appearance of an effective Michaelis-Menten coefficient K′eff,i = K′i (1+yi+1/Ki+1), which is a typical way to indicate competitive inhibition in enzyme kinetics [22]. In the context of activation-inactivation cycles, a similar type of equation was obtained by Salazar and Höfer in the systematic study of a single cycle taking into account the competition between kinase and phosphatase to bind the same target protein [23]. In that case, an effective Michaelis-Menten coefficient appears also in the negative term of Equation 4, but with the form K′eff,i = (1+yi/Ki). In our study, however, the competition is induced by the next substrate yi+1, and this precisely describes a negative feedback from cycle i+1 on cycle i: the higher the level of xi+1, the smaller yi+1 and, therefore, the larger the value of the negative term in Equation 6. This modified denominator reflects the influence of the downstream step on the state variables of one given cycle. It is not a detail of the formalism. It has consequences upon the dynamics and on the properties of the signaling pathway, as it will be demonstrated in the following sections. Moreover, we will see that, since our system arises from a controlled approximation of the mechanistic model, the dynamics of both models can be made comparable.
In the limit ηi∼εi«1, one retrieves the simple conservation law xi+yi≈1. However, we note that even in that limit and due to K′eff,i, our resulting system is not equivalent to the GK-like model. Notice that ηi∼εi«1 is the closest we can be to the hypothesis behind the GK-like model, where it is considered that the concentration of the targeted protein is in large excess over those of the converting enzymes. In our description, the converting enzymes for unit i are E′iT (phosphatase) and Yi−1,T (kinase). Taking the limit ηi«1 together with the fact that the targeted protein of each cycle is the activating protein of the next one, results in increasing protein concentrations as the cascade proceeds. Even though this is not the usual condition in signaling cascades, examples could arise where this limit is suitable. As a possible relevant example, the concentrations reported for the MAPK cascade go from nM in the first unit to µM in the second and third ones [8].
In addition to the limit ηi∼εi«1, our perturbation scheme encompasses situations where the total protein does not necessarily increase along the cascade. We then allow ηi∼1, for all or for some index i, as long as µi ηi∼εi, which results in the limit µi∼εi«1. Since µi = ki/k′i, this limit requires that the phosphatase of that cycle be much more active that the corresponding kinase. In this limit however, the conservation law remains as expressed in Equation 7 and no further simplification can be made. As a result, in this limiting case, the first term in Equation 6 depends on the variable describing the previous step in a different (and more complicated) way compared to Equation 4.
Finally, we notice that our description enables a reduction of the cascade equations with mixed hypothesis concerning the enzymatic reactions. For example, we could have µ1∼ε1 and η1∼1 for the first cycle, µ2∼1 and η2∼ε2 for the second cycle, etc. Or even µi∼εi½ and ηi∼εi½ for all or for some index i.
In Text S3, we present the extension of the reduced mechanistic model for a cascade involving double-phosphorylation. Notwithstanding that these equations are more complicated than Equations 6–7, the distinguishing feature is maintained: each level in the cascade is subject to influence from the following level which, in the appropriate xi variable, can be identified as a negative feedback. In the current study we analyze mostly static properties of these more complicated equations and compare them to those of Equations 6–7, while a more exhaustive characterization will be presented in a future article.
In this section we report on dynamic and static properties of the new chain equations (Equations 6–7), when studied by numerical simulations, and compare them to those of previous cascade models. We will consider both short (n = 3) and long chains (n = 10, or 15), respectively. In all the figures we plot yi*, the level of active protein, obtained from xi in Equations 6–7 (see Text S4 for a comparison between variables xi and yi*). In this section, each parameter in the reduced mechanistic model is considered to be homogeneous throughout the chain, i.e., the parameters do not depend on the index i characterizing the position of a particular unit in the chain.
The homogeneity assumption implies that Vi≡V = µη/ε and V′i≡V′ = 1. Parameter S indicates the level of input stimulation the chain receives. The parameter K′ is chosen by considering the relation K′ = K/µ. We have performed numerical simulations with other parameter relationships and the properties reported below are not critically dependent upon that choice. The control parameters are, then, V, K, µ, and η. Since V′ = 1, the range of V values of interest lies around 1. The initial condition for all the numerical simulations considered is, at t = 0, xi = 0 (and yi = 1) for every i.
In this section we apply the reduced mechanistic model to a well-known signaling pathway, the mitogen-activated protein kinase (MAPK) one [3], [8]–[10],[25],[26]. We first base our description on a particular published set of parameters for this pathway [8]. Importantly, the results obtained are not qualitatively modified by variations of the selected values in the ranges suggested in the literature [8]. Moreover, they are not modified by choosing different sets of parameters [9],[25],[26], as described in the Text S6.
It is well know that the MAPK cascade consists of three levels, the second and the third ones involving a double-phosphorylation mechanism. In this section we consider both the MAPK cascade and a simpler case, a 3-unit chain where each unit is a 2-state cycle.
Starting with the published set of parameters (see [8] and also [10], for a summary), we have computed the parameters involved in the reduced mechanistic description and listed them in Table 1. As described in Text S6, there are some extra parameters for the case involving double-phosphorylation, that are designated ν, K*, and K″ and take the values of 1, 0.25, and 0.25, respectively.
According to Table 1, the conditions under which the reduced model is valid are only partially satisfied, ηµ∼ε for the first unit but ηµ∼10 ε for the second and third ones. Even for these conditions and since the focus of this section is in steady states, the reduced mechanistic model provides a description that is in excellent agreement with the complete mechanistic one.
In Figure 6 we plot the normalized stimulus-response curves for a 3-unit chain, either with single-phosphorylation in all the units (A) or with single-phosphorylation in unit 1 and double-phosphorylation in units 2 and 3 (B), i.e., the case corresponding to the MAPK cascade. Both cases are characterized by the parameters in Table 1. The input stimulus was taken to be the concentration of E1T , the total amount of kinase for the first unit in the cascade (corresponding to MAPK kinase kinase in B). E1T, related to the parameter S we have used as input in the previous section, was varied over a wide range. The outcomes were obtained by both the complete mechanistic and the reduced mechanistic models and the results are indistinguishable for the scales of the figure (black, blue, and red filled lines for y1*, y2*, and y3*, respectively). For completeness, we are also including the corresponding outcomes obtained by the GK-like model (dotted lines).
In order to compare the steepness in the responses, we have computed the apparent Hill coefficient nH ([8]) for each curve, as indicated in the legend. As expected, nH increases through the chain. Moreover, nH is also considerably reduced when comparing GK-like model's predictions with the predictions of both mechanistic and reduced mechanistic models (which are, as already mentioned, undistinguishable). As explained in the section dealing with stimulus-response curves, these differences could be due to the fact that both the mechanistic and the reduced mechanistic descriptions take into account “sequestration” in the enzymatic reactions [3].
We have mentioned that the good agreement between the mechanistic and reduced mechanistic descriptions regarding the prediction of steady states is due to the conservation law, Equation 7, taking into account the first correction in εi (see Text S1). If that correction is not considered, differences could appear in the steady states predicted by the mechanistic model and the reduced mechanistic one. However, and for the parameters in Figure 6, the predicted values of nH are not modified by removing the εi correction in the conservation law or, even by, removing the ηi correction as well (i.e., using a conservation law of the form xi+yi = 1). These results strongly indicate the robustness of the new equations regarding the “ultrasensitivity” characteristics of the cascade.
In Figure 6B, the mechanistic and reduced mechanistic models' outcomes and corresponding Hill coefficients recover published results [8]. Comparing figures A and B, we also confirm that the chain involving double-phosphorylation responds in a steeper manner than the one with only single-phosphorylation, as expected from previous work [27].
In Figure 7 we show the outcome of stimulating the 3-unit chain as indicated in the schemes close to each panel: the input stimulus to the cascade was taken to be the concentration of E′3T , the total amount of phosphatase for the last unit in the cascade (corresponding to MAPK in B). E′3T was varied over its suggested range of variation [8]. Increasing the amount of phosphatase produces a decrease in the response curve y3* (red filled line), as expected. Interestingly, our new reduced model (Equations 6–7), as well as the complete mechanistic description, predict that this perturbation on the third level of the chain is propagated backwards: the variation in y2* is actually a decrease due to a higher sequestration of free y2* by the next step in the chain caused, in turn, by the increased demand of y3. This result is exhibited by both cascades in Figure 7 (the one involving only single-phosphorylation and the one with double-phosphorylation in units 2 and 3) and we call it “reverse” stimulus-response curves. As stated before, this result is obtained with both the mechanistic and the reduced mechanistic descriptions, with realistic parameters associated with a well studied signaling pathway, such as MAPK.
The insets in both figures indicate that is not necessary to vary parameter E′3T over a wide range to observe this property, rather it is clearly seen by changing it only by a factor 5 around its suggested concentration (0.12 µM), where a 20% variation in y2* is observed, a value that is high enough to be detected experimentally (meaning that it is most likely not contained within the error of the experiment). Due to the parameters characterizing this particular pathway, the effect is not propagated to y1* (black filled line), but this fact does not have to be generalized (see Text S6). The dotted horizontal lines in Figures 7A and 7B are the GK-like prediction for the response curve y2*: within that phenomenological description, a particular level in the cascade is not at all influenced by what happens in a downstream unit. However, this well known property of unidirectional influence in a signaling chain, which is embodied by the appellation of “cascade”, is shown here not to be guaranteed in general signaling cascades.
In Text S6 we extend the results in this section concerning “reverse” stimulus-response curves, for different sets of published parameters on the MAPK cascade.
A modular response analysis (MRA) [28] was applied to determine the network architecture of the cascade in the context of the new model equations (Equations 6-7). MRA has recently been proposed as a tool to characterize the interactions between “modules” in a cellular regulatory network, having the advantage of allowing direct experimental implementation.
As a matter of fact, the negative sign of the Jacobian element ∂x ˙i/∂xi+1 indicates that the (i+1)th level of the cascade exerts a negative effect in what concerns variable xi. This effect (what we have called “negative feedback”) is intrinsic, as opposed to “explicit” negative feedback which is sometimes considered in models of signaling pathways [9],[12],[13]. MRA is, then, an appropriate approach to test this bidirectional structure and to estimate the relative strength of the backward interaction, as compared with the forward coupling in a signaling cascade.
As a result of applying MRA, a matrix of local response coefficients r is obtained. An element rij in this matrix describes how the state of the variable associated with module j directly influences the state of the variable associated with module i. More precisely, a response coefficient rij lower/greater than 1 means that a relative change in module j is attenuated/amplified in module i by a factor rij (i.e., Δxi/xi = rij Δxj/xj). A zero response coefficient indicates no direct effect between the involved modules, whereas a negative response coefficient means inhibition. In this way, the matrix r provides an interaction map to characterize the type and strength of the interactions between the modules in a cellular regulatory network.
Indeed, if the rate of change of variable xi is denoted by the function fi, it easily can be shown that:(8)meaning that rij corresponds to a scaled version of the Jacobian matrix ∂fi/∂xj (evaluated in the steady state). Moreover, it was proven that the local response matrix r can be obtained from another matrix named global response matrix, Rp, that has the advantage of being accessible experimentally [28]. For example, the element (i, j) of this matrix can be obtained by perturbing a parameter pj affecting only module j and computing the relative changes induced on the steady state of xi, namely (Δxi/xi)/Δpj. For more details about the broad scope of the method, we refer the reader to the cited reference and references therein.
Using notations and concepts from the literature [28], we apply the MRA method to a 3-unit cascade involving only single-phosphorylation and characterized by the parameters in Table 1 [8]. There are three modules in this network as described by Equations 6-7, each of them corresponding to the three successive levels in the cascade and characterized by a single variable xi. Figure 8A contains the matrix of local response coefficients r. This matrix was obtained both by direct computation of the scaled Jacobian matrix (Equation 8) and by simulating experimental perturbations to the cascade, then computing the global response matrix, Rp, and finally obtaining r, as described previously [28] (details of second calculation not shown). Using MRA, the “theoretical” and “experimental” outputs were in perfect agreement and the results are displayed in Figure 8A.
The structure of matrix r is tridiagonal, meaning that the first level in the cascade does not directly influence the third one (r31 = 0), and viceversa (r13 = 0). Coefficients r21 and r32 are positive, representing the positive effect of each level in the cascade to the subsequent one. Interestingly, r12 and r23 are both negative, indicating an inhibitory effect from unit (i+1) to unit i. The resulting connections between the units in the cascade are summarized in the scheme in Figure 8A.
To understand these results in more depth, we have studied how the coefficients in the matrix in Figure 8A depend on the parameters characterizing the cascade. For example, Figure 8B shows coefficients r21, r32, r12, and r23 versus the parameter E1T. r31 and r13 are zero (data not shown), r21 and r32 are positive, and r12 and r23 are negative, throughout the range where E1T was varied. Depending on the value of E1T, each of the nonzero rij could be less or greater than 1 and the relative strength of the backward and forward couplings for a given pair of modules, e.g. |r12/r21|, could exhibit large variations. Similar curves have been reported in the literature for signaling cascades [29], but lacking the information about r12 and r23, which have always been considered to be zero in previous papers.
Studies like the one in Figure 8B help us understand and also predict the degree of backwards coupling as a function of the parameters in the model. One utility of this work is as a starting point of a more systematic study on how to enhance or attenuate that coupling in the cascade, the subject of our ongoing work.
Interestingly, the interaction map characterizing the connectivities between variables xi (matrix r(xi) in Figure 8A) shows strong differences when compared to the one computed for the “free” enzyme variables yi* (matrix r(yi*) in Figure 8C. Although an explicit set of differential equations is not written for the variables yi*, the matrix r(yi*) can be calculated using the “experimental”' method described in the literature [28]. The result in Figure 8C is the average of four outputs and the corresponding error (standard error of the mean) is lower than 4%. As indicated in the reconstructed topology close to the matrix, r12 and r23 are now positive (as are r21 and r32), r31 is zero, and r13 is negative, indicating an inhibitory coupling from variable y3* to variable y1*. The matrix r(yi*) is consistent with the results in Figure 7A (and also those in Text S6): in other words, the response in y2* goes in the same direction as the one in y3* (whereas plotting variables xi indicates a decrease in x3 and an increase in x2, data not shown).
Experimental data concerning the application of MRA to the MAPK cascade are now available in the literature [30], showing non zero r21 and r32 coefficients (and also non zero r31 and r13 coefficients). The interpretation of non zero r31 and r13 was proposed in terms of the usual “explicit” positive or negative feedbacks which are sometimes considered in models of signaling pathways [9],[12],[13]. From this perspective, the explanation for the non zero r12 and r23 coefficients was, at least regarding r23 and based on experimental evidence, that not only is y2* able to phosphorylate y3, but y3*can phosphorylate y2 as well [30],[31]. Our results however, suggest that the non vanishing backward coefficients (r12, r13, r23) can be accounted for, at least partly, by the natural “implicit” feedback which can exist in a signaling cascade. A quantitative correlation between these recent experimental results and our predictions is not possible at this time. In the published experiments, the MAPK cascade is not isolated but embedded in the complex cellular machinery; therefore, the measured connectivities could involve proteins external to the cascade itself and it would be premature to establish the connection with our simplified model. Nevertheless, the work in [30] suggests a direction for the type of experiments that could validate our results.
The main contribution of this work is to propose a new one-variable per cycle model for signaling cascades of covalent modification cycles, consistent with a mechanistic complete description. Our model reveals new and biologically relevant properties of such cascades. These properties are characterized completely for the case of single-phosphorylated cascades. Furthermore, single and doubly-phosphorylated cases are compared regarding their stimulus-response curves, while a more exhaustive characterization of the scheme involving double phosphorylation will be presented in a future article.
The scheme in Figure 1, which has been employed by many groups, is suggestive of the concept of a “cascade”. From a systemic point of view, a cascade is a system composed of units, the output of which is successively an input to the next unit. Based on this structure, powerful concepts from control theory can be applied successfully to the study of signaling cascades [14]. Although these concepts have proven its utility in many contexts, this kind of schematic representation implicitly conveys the idea that a signaling cascade is only a feed-forward chain in which signal transmission is analogous to a domino effect [32],[33]. Our study sheds a different light on this system, showing that this schematic representation can be misleading, since it turns out that each unit is actually coupled not only to the following one but also to the previous one, and interesting dynamics can arise from these interactions.
Our initial motivation for developing a new one-variable description of signaling cascades, was the following observation. The main assumption underlying the GK description of a single cycle is that the concentration of the target protein is in large excess compared to those of the converting enzymes. Holding the same assumption over a cascade of units would mean that the target proteins are in higher and higher concentration as the cascade progresses, since they act as the transforming enzyme for the following cycle. To our knowledge, this important issue has not been remarked upon in the literature, except for a brief comment in the work of Millat et. al. ([20], page 11).
In order to get more insight into this point, we have sought special limiting cases for which the mechanistic and the GK-like model are in good agreement. However, it turns out that the dynamics of the signaling cascade described by the mechanistic and the GK-like models cannot be compared consistently. The fundamental reason for this mismatch is that a careful perturbation analysis applied to the mechanistic model provides a different set of equations.
We note that in search for an adequate set of hypothesis leading from the mechanistic equations to the model given by Equations 4, we have studied an alternative scheme in which the modified protein Yi* is not directly the kinase of the next reaction. Instead, we studied the case where Yi* activates that kinase. This scheme was suggested by the work of Goldbeter [12]. The resulting equation (see Text S7) is fundamentally different from the GK-like model. In reality, no set of assumptions can give rise to the GK-like model as a limiting case of our model.
Our mathematical method relies on the standard quasi-steady state assumption (QSSA), which can be applied under well defined conditions to elicit a clear separation between the slow and fast dynamics of the mechanistic model. Under this standard QSSA framework, our analysis shows that a good slow variable for which evolution equations can be written is the sum of the free activated enzyme which is available in the ith cycle plus the amount of this protein which is captured by the next inter-converting cycle. The idea of working with a mixed variable xi can be further generalized by considering the “total” variable corresponding to the total amount of activated enzyme found not only as free molecules or bound to the next substrate, but also complexed with the reverse enzyme E′i. In fact, this choice is the key ingredient of the method called the “total” quasi-steady state approximation (tQSSA) which has been proved to be a simple but most efficient extension of the standard QSSA [34]. The application of this extended framework to the description of the signaling cascade of Figure 1 is concerned with our current research. In the same context, other authors have recently applied the tQSSA method to the study of small networks of GK cycles [35]. These systems do not form cascades, but involve a more complicated coupling between the units. Nevertheless, their results show that indeed the tQSSA method is successful in obtaining a reduced set of equations, with one variable per cycle, which faithfully reproduces the dynamics of the network for a large range of system parameters.
Even in the less extended QSSA framework, the conditions under which the model is valid are made clear. Under such conditions, our new model is indeed in perfect agreement with the complete mechanistic model (Figure 2). Those conditions are expressed in terms of three key parameters (Equation 5) we have defined to simplify the study. Even though the phenomenological equations, Equation 4, are appealing because of their simpler form and modular nature, we could not find any set of assumptions that would enable us to recover those descriptions. Our simplified model reveals properties of signaling cascades that were either hidden by the complex structure of the complete mechanistic model or lost in the simplified phenomenological descriptions.
It was stated that the reduced mechanistic model is valid whenever these two conditions are satisfied: εi«1 and µi ηi∼εi. The study of the performance of the new approximation (Figure 2 and the corresponding computed errors) makes it clear that even when those conditions are satisfied only moderately, the new model is still robust in approximating the complete description. As an example, we have computed a 5% error for ε = 0.1, η = 1, and µ = 0.5 (meaning µη∼5ε). Moreover, we have observed that the steady state predictions of the reduced model are highly accurate. Therefore the properties of signaling cascades we are unveiling thanks to the new reduced model, are not restricted by a tight relationship between concentrations and reaction rates hard to achieve in in vivo or in vitro conditions.
All the novel properties of a signaling cascade reported in this paper are linked, as previously mentioned, to the negative feedback from each unit to the previous one. This backward negative feedback can produce damped temporal oscillations in the chain, or it can create amplified “pathway” oscillations in the steady states of the cascade. Interestingly, it can also transduce a signal both forward and backwards. Given the multi-branched complex nature of many signal transduction pathways, this finding could have wide implications and can help focus further experimental investigation.
It has recently been reported that the 3-level MAPK cascade has autonomous oscillations without any kind of added explicit feedback [36]. Following a systematic numerical exploration of the corresponding mechanistic model [8], the authors provide a qualitative description of the mechanism responsible for these sustained oscillations. Their explanation strongly suggests the necessity of a bistable behavior at the second or third levels of the cascade, thus requiring double-phosphorylation at these stages [37]. Consistent with their findings, we have observed only damped oscillations in the dynamics of the single-phosphorylated cascade (Equations 6–7), which has been the main focus of the present work. Interestingly, preliminary numerical simulations of our reduced doubly-phosphorylated cascade model (Text S3), indicates that these autonomous oscillations are recovered in the simplified description.
The stimulus-response curves of the new model were also investigated (Figure 5). They have the usual sigmoidal shape characteristic of ultrasensitive responses; however, they exhibit lower steepness when compared with the output of the GK-like model. This result corroborates the conclusions stated in the work of Blüthgen et al. [3], where an analysis of the effect of sequestration was conducted. This effect is partially mitigated by double-phosphorylation (Figure 6), as expected from the literature [27].
To further characterize the new model within realistic conditions, we have studied it subject to different sets of published parameters corresponding to a well-known signaling pathway, such as the MAPK one (Figures 6 and 7, and Text S6). We have found that the ability of the model to transduce a signal both forward and backwards is widespread and that the effect is of enough magnitude to allow experimental verification.
Finally, we have applied a modular response analysis to determine the network architecture of the cascade described by the new model equations (Figure 8). This well-known approach enables not only to test the bidirectional structure of the cascade, but also to estimate the relative strength of the backward interaction.
In summary, our findings do not at all weaken the importance of previous models like the GK-like models and those with linear rates. To the contrary, the results of our model provide a different approach to deal with a simple one-variable per cycle model to describe signaling cascades. We hope that our contribution will help in the understanding of existing models for signaling cascades, will improve the description of available data, and will inspire both theoretical and experimental investigation.
All the ODEs were integrated in MATLAB 7 (Mathworks, Natick, MA). The stimulus-response curves were obtained using MATCONT, a MATLAB package for numerical bifurcation analysis of ODEs. The symbolic calculations were done using the Symbolic Math Toolbox in MATLAB. |
10.1371/journal.pcbi.1002914 | An Entropic Mechanism of Generating Selective Ion Binding in Macromolecules | Several mechanisms have been proposed to explain how ion channels and transporters distinguish between similar ions, a process crucial for maintaining proper cell function. Of these, three can be broadly classed as mechanisms involving specific positional constraints on the ion coordinating ligands which arise through: a “rigid cavity”, a ‘strained cavity’ and ‘reduced ligand fluctuations’. Each operates in subtly different ways yet can produce markedly different influences on ion selectivity. Here we expand upon preliminary investigations into the reduced ligand fluctuation mechanism of ion selectivity by simulating how a series of model systems respond to a decrease in ligand thermal fluctuations while simultaneously maintaining optimal ion-ligand binding distances. Simple abstract-ligand models, as well as simple models based upon the ion binding sites in two amino acid transporters, show that limiting ligand fluctuations can create ion selectivity between Li+, Na+ and K+ even when there is no strain associated with the molecular framework accommodating the different ions. Reducing the fluctuations in the position of the coordinating ligands contributes to selectivity toward the smaller of two ions as a consequence of entropic differences.
| Differentiating between Na+ and K+ ions is important for many cellular processes, such as nerve conduction and the regulation of membrane potentials. Different biological molecules utilise different methods to discriminate between ions. In this work, the reduced ligand fluctuation mechanism of ion selectivity is described. This entropy-driven mechanism is due to the limited thermal fluctuations of the atoms in some macromolecular ion binding sites. The elucidation of this mechanism offers a more complete picture of the ways in which the fundamental process of ion selectivity can be achieved.
| The ability of some biological molecules to discriminate between different ions is crucial for their function. This differentiation is important, for example, in the generation (or regulation) of the action potential during cellular signalling, and the maintenance of an electrochemical gradient across the cell membrane [1]. Indeed, without this ability to discriminate between ions, a cell would quickly die. Of particular interest is how such molecules are able to distinguish between the monovalent cations Na+ and K+: these ions are both spherical, they have identical charges, and they differ in atomic radius by only 0.38 Å. It is incredible that some proteins, such as potassium selective ion channels, can discriminate between these two ions at nearly diffusion limited rates [2]–[5].
Although it is generally agreed that selectivity depends on a difference in free energy relative to bulk water of one ion compared to the other at some position within the transit pathway (i.e. how well the loss of free energy from dehydration is recouped by coordination with the protein), there are several different proposals which attempt to explain how this difference in free energy occurs. These proposals fall into three broad categories related to:
In this study we focus on the last category. To date, three different cavity effects have been proposed that can lead to ion selectivity: the ‘rigid cavity’, the ‘strained cavity’, and the ‘reduced ligand fluctuation’ (RLF) mechanisms. We discuss each in turn below; Table 1 summarises their key similarities and differences.
The ‘rigid cavity’ mechanism is perhaps the easiest to understand [24]–[26]. It suggests that ion selectivity is created by the ligand framework maintaining a certain fixed position (i.e. cavity size) about an ion regardless of the type of ion that is coordinated. Specific positions will be energetically more favourable for one ion type over another, thus contributing to selectivity for that ion. For example, when the smaller ion is favoured because the binding site is too small to fit the larger ion, this is often termed ‘size selectivity’. If the favoured ion is larger and sits more favourably in the cavity, this mechanism is commonly called ‘snug fit’. In reality the positions of the ligands will never be completely fixed, and their thermal motion is often larger than the size difference between Na+ and K+. Taking these thermal fluctuations into account, it has been demonstrated, in principle, that if the ligands fluctuate about some fixed average configuration for different ions this will create ion selectivity [10]–[12]. The question of which particular ion is selected by a given cavity site depends strongly upon the actual positions to which the ligands are constrained. Even this picture of a rigid cavity is probably too simplistic as the ligands are likely to fluctuate about different average positions when coordinating different sized ions. If the difference in the average positions is less than the difference in ion radii, one may still consider this situation to be a ‘rigid’ cavity. Our studies of many proteins suggest that the difference in average ion-ligand distance when coordinating Na+ and K+ is almost always similar to the difference in ionic radii, suggesting that a true rigid cavity is uncommon in proteins [12].
Unlike a ‘rigid cavity’, a ‘strained cavity’ allows for the average ion-ligand distances to adjust according to different ion types. However, in this case the adjustment comes at an energetic cost, called ‘strain’. Strain may be realised as a deformation within the ligand itself, or as a deformation of the ligand/protein scaffold, be it local [8], [27] or non-local to the ligand site [12], [19], [28]. Non-local strain may itself precipitate a conformation change in the protein (an extreme version of the effects of strain) thus further influencing ion selectivity [28]. A rigid cavity can be considered as an extreme form of a strained cavity, wherein the coordinating ligands resist any attempt to adjust to a new ion type in the binding site, perhaps due to an even larger cost in energy of deforming the protein scaffold. A continuum exists between the two, characterised by the degree of change in the average position of the ligands upon a change in ion type. As already noted, a rigid cavity is unlikely to exist in proteins, due to the inherent flexibility of these structures.
The idea that differentiation between ions could be achieved through a rigid cavity mechanism was first suggested by Mullins [24], [29], [30] who was investigating selectivity in ion channels. It was suggested that a rigid pore of an appropriate size could allow favourable interactions with K+, but be too big for Na+, leading to unfavourable interactions. This mechanism was supported voltage clamp experiments by Bezanilla and Armstrong [25] which suggested the pore was lined with backbone carbonyl oxygens, the particular arrangement of which mimicked bulk water more closely for K+ than Na+. More recently, Doyle et al. [26] have purported to suggest that ion selectivity in KcsA resulted from a rigid cavity mechanism. Whether it be a poor choice of words by the authors or misinterpretation by others (or a combination of the two), it seems that this explanation was offered as a caricature of a strained cavity mechanism, a point that is clarified in a later study [28]. The strained cavity mechanism has also been shown to play a role in ion selectivity in some ionophores, such as valinomycin [31]–[33], where the small amount of scaffolding between ion coordinating groups can leave the molecule sensitive to subtle sizes differences between coordinating ions. We propose that the rigid and strained cavity mechanisms are two domains of the same continuum; the term ‘strained cavity’ will herein be used to encompass this, except for when contrast between the two is required. Both share the common feature of requiring a resistance to changes in the positions of the coordinating ligands in order to generate ion selectivity.
Could ion selectivity be generated in an ion binding site through a ‘cavity’ like mechanism without the need for strain? In a previous investigation using simple abstract-ligand models it was noted that ‘cavity’ based mechanisms could still create selectivity even when both the cavity size is not fixed and when there is no strain associated with adjustment of the ligand positions [12]. In this case, the only ‘cavity’ factor controlling the binding energy of the ions was the degree of thermal motion associated with the ligands as the ligands were free to adopt their optimum positions for each ion type. This is the essence of the ‘reduced ligand fluctuation’ (RLF) model for ion selectivity that is the focus of this investigation.
Reduced ligand fluctuations (i.e. small values of root-mean-square (RMS) deviations in positions of the atoms forming the ligands from their average positions) of the ligands in the ion binding site compared to the rest of the protein were noted in atomistic molecular dynamics (MD) simulations of LeuT [12], [34], a leucine transporter, and GltPh, an aspartate transporter [12]. This situation contrasts with other molecules we studied previously, in which the RMS fluctuations of the ligands in the ion binding site were not notably smaller than the average across the protein [12]. It was also demonstrated, at least in the case of GltPh, that the Na+ binding sites were able to accommodate K+ (i.e. the sites were able to change the ion-ligand distance to a more favourable one for K+), ruling out a rigid cavity mechanism creating ion selectivity in this molecule. It could be the case that there is an energetic penalty in adjusting to K+ (strain), however, this is difficult to quantify [12], [18].
In this investigation we try to better understand how reducing ligand fluctuations creates ion selectivity, improving on the simple models used in previous work [12] by using better force field parameterisations, and by investigating a greater variety of model systems. Simplified model systems have been used to study ion selectivity in a range of molecules including potassium channels [8], [9], [12], [13], [18], [19], [22], [27], a sodium channel [35], NaK channels [27] and kainate receptors [36]. In addition, the previous work never addressed exactly how reducing the thermal motion of the ligands leads to selectivity. Using the more detailed model systems, and by analyzing binding-energy components, we are able to propose an answer here. Of course, the RLF mechanism is not mutually exclusive with the other means of obtaining ion selectivity previously mentioned; it may work in concert with these other effects. However, in this investigation, we wish to study it in isolation, so as to clearly discern it from these other possibilities.
One can elucidate the effect of the RLF mechanism on the selectivity of an ion binding site by investigating how a series of ‘abstract’ model systems (pioneered by Noskov et al.) [8] respond to a reduction in the RMS fluctuation on the coordinating ligands. These model systems consist of a number, , of abstract ligands (in this case based on formaldehyde), where each oxygen atom is confined to a 3.5 Å sphere about an ion, either Li+, Na+ or K+. This first constraint is enforced by a one-sided harmonic potential with a very large force constant. This spherical constant is not meant to precisely define the coordination numbers of the ions, but rather to control the number of ligands near to the ion as a model of the composition of a biological ion binding site. The position of the central ion is fixed, while each atom in the coordinating ligands can be further constrained by placing them in an additional harmonic potential, centered at a nominated position. The amount of thermal fluctuation of the ligands can be controlled by altering the force constant, . The choice of physical location at which the harmonic restraint is placed is very important. In order to isolate RLF from strain, the harmonic restraint needs to be placed at an ‘optimal’ ion-ligand distance for each ion type. This position is defined as the maximum of the first peak in the radial distribution function from dynamics simulations of Li+, Na+ or K+ surrounded by ligands with no harmonic restraint. The systems where the position of the coordinating ligands are controlled are referred to as with ligands, where M+ = Li+, Na+ or K+, with geometries determined by the vertices of optimal coordination polyhedra for n = 4,6,8 and by packing circles on spheres for n = 5,7 [37]. is the distance between the center of the ion and the center of the coordinating atom.
In order to quantify ion selectivity, the free energy to exchange two ions between bulk water and our model binding sites is calculated. As each model system is allowed to adopt an ‘optimal’ cavity size (or ion-ligand distance) for each ion type, a series of free energy perturbation molecular dynamics (FEP MD) simulations are required in order to describe the contribution to ion selectivity from this mechanism in a meaningful way.
The overall reaction to be investigated is an exchange reaction of the ions and between an optimally sized hypothetical model system for each ion (at ion-ligand distances and ) with controllable fluctuations and bulk water:(1)To calculate and other binding energies we begin by effectively ‘morphing out’ the positional constraint on the ligands around one of the ions, . This is achieved by having two sets of ligands in the simulation: one set where each atom is subjected to a harmonic constraint () and one with no harmonic constraint ().(2)The system with no harmonic constraint, (but still with a 3.5 Å radial constraint) can now undergo an exchange reaction with another ion bound in bulk water:(3)This exchange reaction consists itself of two separate FEP MD simulations:(4)(5)The values for used here were calculated from the free energies of solvation by Joung and Cheatham [38]. Now the constraints can be morphed into the model system containing :(6)The change in exchange free energy for the overall reaction is then given by(7) will be positive if is thermodynamically preferred in the ion binding site and negative if is preferred. This quantity can also be studied as the thermal fluctuations (i.e. the value of ) is reduced without reference to the energy involved in bringing the ion into the site from the bulk. The contribution to ion selectivity due only to the RLF mechanism alone, , can be defined as:(8)
Abstract model systems consisted of abstract ligands (based on formaldehyde with partial charges of carbon +0.5, oxygen −0.5 and hydrogen 0.0) where each oxygen atom is confined to a 3.5 Å sphere by use of a spherical flat-bottomed, steep harmonic potential constraint. Reducing the ligand fluctuations was achieved by confining each atom to a harmonic potential, varying the force constant, , from to in 0.5 increments. This harmonic potential was placed at the vertices of optimal coordination polyhedra for , and at geometries governed by packing circles on spheres for [37], at a distance from the ion corresponding to the first peak in the radial distribution function of a harmonically unconstrained (but still with the 3.5 Å spherical constraint) simulation determined for each ion type. A harmonic restraint is used for this purpose as a first order approximation; the restraining potentials exhibited in nature would probably be somewhat anharmonic and anisotropic. Two sets of ligands, i.e. , are required in order to conduct FEP MD between harmonically constrained and harmonically unconstrained ligands with one set annihilated and the other set exnihilated during the simulation. However, both endpoints represent ligands coordinating to an ion. Errors in were minimised by using a large number of windows, with values of , then for then then incrementing by 0.05 to , then for then . Forward and reverse morphs were conducted for each ion/model/ combination. The maximum difference in the forward and reverse morphs for was 0.94 kcal/mol, with an average difference of 0.27 kcal/mol. The maximum error in (the summation of errors from , and ) is estimated to be 1.1 kcal/mol with an average error of 0.62 kcal/mol. Energies were averaged over 4 ns for each window. Softcore potentials were utilised using a van der Waals radius shift coefficient of 1. A cut-off distance of 12 Å and a switching distance of 10 Å is used for electrostatic and van der Waals interactions.
FEP MD simulations where the ion is being morphed used only one set of ligands, as the ligands are not bound by a harmonic constraint. was varied from 0 to 1 in 0.05 increments. Energies were averaged over 4 ns for each window. Softcore potentials were utilised using a van der Waals radius shift coefficient of 1. A cut-off distance of 12 Å and a switching distance of 10 Å is used for electrostatic and van der Waals interactions. All simulations were conducted using NAMD2 [39] with the CHARMM27 force field [40] at 310 K with 1 fs timesteps. Force field parameters for Li+, Na+ and K+ are from Joung and Cheatham [38].
The volumes occupied by the coordinating ligands were calculated using the VolMap tool in VMD [41], with a resolution of 0.1 Å, and an in-house Fortran program with an isosurface value of 1.0.
FEP MD simulations were conducted identically to the harmonically constrained abstract models and the harmonically unconstrained models discussed in the previous section. To set up these systems the positions of the atoms coordinating to Na+ in each ion binding site, along with the atoms directly bonded to these and the ions themselves were extracted from the crystal structures of GltPh, PDB accession code 2NWX [42], and LeuT, PDB accession code 2A65 [43]. These four model sites were then energy minimised with Li+, Na+ and K+ present as the central ion. These minimised structures provided the initial starting coordinates for further simulations along with the coordinates to which the harmonic constraints were placed (i.e. ).
was calculated by extracting the average total potential energy from the first and last window of each FEP MD simulation and combining them in a fashion as described for in the theory section. was calculated for each using .
The ion selectivity of the abstract models, including contributions from the RLF mechanism, is plotted versus the force constant, , for binding sites with 4–8 ligands in Fig. 1 and 2. For comparison we also plot two sets of results for the strained cavity mechanism, that is, when the same location of the restraint is used for both ions rather than using an ‘optimal’ position for each ion type.
Fig. 1 demonstrates that each abstract model system displays an inherent selectivity for K+ over Na+ when there is little or no restraint on the fluctuation of the ligands, in line with results from previous studies by ourselves [12] and others [8]. It must be stressed that this inherent selectivity is only for this particular type of ligand. Naturally, as increases the strained cavity models become more selective for the ion to which the positions of the ligands are optimised (blue, Na+, and red, K+, lines in Fig. 1). This effect can be quite large (tens of kcal/mol) and tends to plateau for the largest values of tested in this study, where the strained cavity becomes a ‘rigid’ cavity.
The selectivity of the abstract models for Na+/Li+ is a little more complicated than for Na+/K+. Each abstract model displays an inherent selectivity for Na+ when there is little or no positional constraint on the coordinating ligands, as Fig. 2 shows. The effects of introducing the strained cavity begin to show as the strength of the restraint increases; the positions to which the ligands are constrained determine the selectivity (green, Li+ and blue, Na+, lines). However, there is some anomalous behavior, especially for the six ligand case (Fig. 2 C) where strong restraints at both Li+ and Na+ optimised positions increase selectivity toward Na+.
A more subtle situation arises when the position of the restraint is different for each ion, i.e. when we consider the RLF mechanism without any strain. Although the difference looks small on the scale of Fig. 1, a 2–5 kcal/mol increase in selectivity toward Na+ occurs for the exchange reaction with K+ with ligands when the size of the thermal fluctuations is reduced ( increased). The majority of this change occurs for between and , plateauing for (see Fig. 3 to see this plotted in a different scale). This change in ion selectivity alters these models from K+ selective sites to Na+ or non-selective sites. For , the selectivity in the already K+ selective site is further enhanced by 2–3 kcal/mol.
A similar situation arises in the exchange reaction between Li+ and Na+ (Fig. 2). Again, the most drastic changes occur for the strained cavity model as increases. The changes in selectivity due the RLF mechanism are again smaller than those for the strained cavity but the trends are similar to that seen for Na+/K+ in the cases with 4 or 5 ligands, with increasing moving selectivity toward Li+. The situation with is more confusing with a reduction in fluctuations causing selectivity toward Na+ for . The model also has a much larger change in (29 kcal/mol) than the models (3–4 kcal/mol). Also, the RLF results in the models do not fall within the bounds of the strained cavity results.
Having shown that restricting the fluctuations in the positions of the ligands creates selectivity for one ion over another even in the absence of strain, the question remains, how does this occur? If we decompose into the enthalpic, , and entropic, , components the driving force behind this change in selectivity becomes apparent. In the exchange between Na+ and K+ (Fig. 3) for and Na+ and Li+ (Fig. 4) for , the contribution follows very closely with indicating this selectivity is largely due to entropy differences. Intuitively one would expect the change in the available number of states (as you decrease the allowed fluctuations) to be largest for the larger ions for the following reasons. The number of possible configurations for coordination in the (only bound by a 3.5 Å constraining sphere) is greater for the larger ion than the smaller ion because of the greater volume available at the larger ion-ligand distance, as depicted in Fig. 5. As the positional restraint is increased (), the number of states become approximately equal for different ion types. Hence the change in entropy between a non-restrained and restrained system is largest for the larger ion. This can be shown to be the case by considering the difference in volume sampled by the coordinating oxygen atoms as their fluctuations becomes more and more constrained. For instance, this change in volume for the four fold coordination state is 3640 Å3 for Li+, 5050 Å3 for Na+ and 5820 Å3 for K+ when comparing and . Reducing the thermal fluctuations on the ligands causes a greater decrease in entropy when they coordinate a larger ion compared to a small one. As a consequence, this reduction of thermal fluctuations favours small ions binding in the site.
A different situation arises with for K+/Na+ systems and for the Li+/Na+ systems. In the former both the enthalpic and entropic terms play a role, while the latter is dominated by the enthalpic contribution. Analysis of these situations shows that the reason for the different behaviour is due to the difficulty in packing a large number of ligands around the smaller ions. As the 3.5 Å constraining sphere does not precisely define the coordination numbers, it is possible for ligands to form a second coordination shell about the small ions when the number of ligands is large. Increasing the force constant, , brings all the ligands to a uniform distance, yielding enthalpic changes in addition to the entropic changes seen for the cases with fewer ligands.
The motivation for proceeding with this investigation was the realisation that in at least two amino acid transporters, the aspartate transporter, GltPh, and the leucine transporter, LeuT, the ligands forming the two Na+ binding sites display reduced RMS fluctuation in their positions compared to similar atom types elsewhere in the protein. The RMS fluctuation of the oxygen atoms forming the four sites ranged between 0.3 and 0.5 Å, whereas the other oxygens in the protein had values larger than 0.7 Å [12], [34]. This is thought to be the result of extensive hydrogen bonding networks in the vicinity of the ion binding sites, as shown to be the case with one of the LeuT sites [44]. Additional constraint may be imparted upon the coordinating ligands if they belong to an amino acid in a more rigid secondary structure, such as backbone carbonyl oxygens of -helices. The ion binding sites in both LeuT and GltPh contain many of these backbone carbonyl oxygen atoms (table 2). More generally, there may also be sterical effects that limit the motion of the coordinating ligands.
Attempts have been made to explain the Na+ selectivity in LeuT. Yamashita et al. [43] suggested that it could be a result of a more snugly fitting site for Na+ than the larger K+, which would upset hydrogen bonding or packing interactions in the protein. This is in line with the strained cavity mechanism described by Lockless et al. [28] in K+ channels. Other investigations suggest that the first binding site (Na1) achieves Na+ selectivity over both Li+ and K+ due to the strong electrostatic interactions resulting from the coordinating carboxylate ligands, while the second binding site (Na2) achieves this through a strained cavity mechanism [10], [44]. To the best of our knowledge, similar investigations have not been undertaken for GltPh. Therefore, we investigate the effect on ion selectivity of reducing the fluctuations in the ligands forming each binding site in the transporters by constructing corresponding model systems.
The model sites for GltPh were constructed from the outward facing, Na+ and aspartate bound crystal structure [42]. Only two (Na1 and Na2) of the three Na+ binding sites are considered, as the exact nature of the third is still under debate [45]–[47]. Models for the two LeuT Na+ binding sites were constructed from the Na+ and leucine bound crystal structure [43]. For each model, the coordinating atom, and atoms bonding directly to these, were used to construct simple dipolar ligands in order to model the electrostatic environment experienced by the bound ion. The composition of each model is detailed in table 2. The initial coordinates of these atoms were taken from the crystal structure and then allowed to energy minimise with Li+, Na+ and K+ independently. This gave us the final optimal coordinates for each ion type at which harmonic constraints were applied. Simulations were conducted to investigate RLF as described earlier for the abstract ligand models.
As the amount of allowed fluctuation in the ligands of the amino acid transporter models are reduced ( increased), the change in the free energy of the exchange reaction between two ions () behaves in a very similar manner to the abstract models; the decrease in fluctuation contributes selectivity to the smaller of the two competing ions (Fig. 6). If we recall that the most of the oxygen atoms in GltPh and LeuT displayed RMS fluctuations greater than 0.7 Å, we see from Fig. 6 C and D that there is little to no contribution toward ion selectivity in this region. However, this contribution becomes significant for RMS fluctuation values observed for the oxygen atoms at the ion binding sites (the grey regions in Fig. 6 C and D). In this model, the ligands are able to adopt a preferred ion-ligand distance, and at no energy cost (in contrast to the strained cavity mechanism), yet a degree of ion selectivity is still created by the reduction in ligand fluctuation. A decomposition of the free energy change in each of the sites into the enthalpic and entropic contributions clearly demonstrate that this effect is primarily a consequence of entropy differences (Fig. 7).
It is evident from the non-zero values of at large RMS fluctuations (small ) of the ligands in Fig. 6 that the chemical nature of the ligands and/or coordination numbers play a role in creating ion selectivity in the ion binding sites of LeuT and GltPh. As the RMS fluctuations decrease, the contribution from RLF merely adds to this. Nevertheless, in the absence of a strained cavity, it is crucial for enhancing selectivity for Na+ over K+ in LeuT. While a strained cavity and the chemical nature of the ligands may play a role in creating selectivity in themselves, we hope to show here that the observed selectivity could also involve a contribution from the RLF mechanism.
Note that the 8 ligand coordinated Na+/K+ abstract model is very similar to the crude model S2 K+-selective binding site in the selectivity filter of the potassium ion channel KcsA investigated previously by Thomas et al. [13] There is a very slight to no increase ( kcal/mol towards K+ selectivity) in the region corresponding the RMS fluctuations (0.75 Å) [8] of the filter, suggesting that the RLF mechanism does not play a role in KcsA. Of course this result for KcsA depends on Na+ and K+ binding at the same sites in the selectivity filter; a view which has been challenged with the proposal of distinct binding sites for the two ions [48]–[50].
It should be noted that our study of the RLF mechanism and previous work by Yu et al. [10] differ in one very significant way. In the latter, constraints are placed at the crystallographic coordinates for the Na+ model binding sites for LeuT and the K+ model binding site for KcsA. This means that there is only one set of constraint positions for both ion types (Na+ and K+) and thus their analysis includes the influence of a strained cavity, which is to say that a change in enthalpy, as well as entropy, will influence selectivity. While we do not deny that such an effect may play a role, we have isolated the RLF mechanism by allowing the ligands to freely adapt to each ion type. This means that the positions of the constraints on the ligands are optimal for each ion type, eliminating any ‘strained cavity’ effect from this analysis.
The simple models of the ion binding sites in LeuT and GltPh were not designed so as to quantitatively reproduce experimental and more detailed simulation results, only to show how the RLF mechanism may influence the overall selectivity. Other factors may become important when considering the selectivity of the protein as a whole, such as the coupling between the ion selective sites [51]. However, these simple models are able to qualitatively reproduce experimental [43] and more detailed simulation [44] results for Na+/K+ selectivity in LeuT. In fact, Na2 changes from a K+ to a Na+ selective site when the reduction in the ligand fluctuations are accounted for. When compared to experimental [42] and more detailed simulations [12], Na+/K+ selectivity in GltPh is qualitatively reproduced for Na1, while Na2 is rendered essentially non-selective with the RLF effect. A table comparing results from this study to experimental and more detailed simulations can be found in text S1.
What conclusions can we draw from this? Given that the amino acid transporter model binding sites are exceedingly simple, any conclusion drawn will be tentative. However, even though these models may be crude, they do demonstrate that reducing the fluctuation of the coordinating ligands, can affect ion selectivity even if there is no strain in the protein. Again it is shown that the RLF mechanism is primarily a consequence of entropy differences. As this mechanism relies heavily on entropic factors, experimental investigations into the temperature dependence of ion selectivity in these amino acid transporters could perhaps shed further light on its role in biological systems.
Reducing the thermal fluctuation in the positions of the coordinating ligands affects the binding of Li+, Na+ and K+ differently and is able to contribute toward ion selectivity, even when there is no strain associated with the protein adapting to different ions. This contribution to ion selectivity is due to entropic differences arising with different ions in the site, resulting from the larger difference in accessible states for the ligands surrounding the larger ions than the small ones when the thermal fluctuations are reduced. Thus, this mechanism of ion selectivity favours of small ions over larger ions.
|
10.1371/journal.pntd.0004641 | A Novel Xenomonitoring Technique Using Mosquito Excreta/Feces for the Detection of Filarial Parasites and Malaria | Given the continued successes of the world’s lymphatic filariasis (LF) elimination programs and the growing successes of many malaria elimination efforts, the necessity of low cost tools and methodologies applicable to long-term disease surveillance is greater than ever before. As many countries reach the end of their LF mass drug administration programs and a growing number of countries realize unprecedented successes in their malaria intervention efforts, the need for practical molecular xenomonitoring (MX), capable of providing surveillance for disease recrudescence in settings of decreased parasite prevalence is increasingly clear. Current protocols, however, require testing of mosquitoes in pools of 25 or fewer, making high-throughput examination a challenge. The new method we present here screens the excreta/feces from hundreds of mosquitoes per pool and provides proof-of-concept for a practical alternative to traditional methodologies resulting in significant cost and labor savings.
Excreta/feces of laboratory reared Aedes aegypti or Anopheles stephensi mosquitoes provided with a Brugia malayi microfilaria-positive or Plasmodium vivax-positive blood meal respectively were tested for the presence of parasite DNA using real-time PCR. A titration of samples containing various volumes of B. malayi-negative mosquito feces mixed with positive excreta/feces was also tested to determine sensitivity of detection. Real-time PCR amplification of B. malayi and P. vivax DNA from the excreta/feces of infected mosquitoes was demonstrated, and B. malayi DNA in excreta/feces from one to two mf-positive blood meal-receiving mosquitoes was detected when pooled with volumes of feces from as many as 500 uninfected mosquitoes.
While the operationalizing of excreta/feces testing may require the development of new strategies for sample collection, the high-throughput nature of this new methodology has the potential to greatly reduce MX costs. This will prove particularly useful in post-transmission-interruption settings, where this inexpensive approach to long-term surveillance will help to stretch the budgets of LF and malaria elimination programs. Furthermore, as this methodology is adaptable to the detection of both single celled (P. vivax) and multicellular eukaryotic pathogens (B. malayi), exploration of its use for the detection of various other mosquito-borne diseases including viruses should be considered. Additionally, integration strategies utilizing excreta/feces testing for the simultaneous surveillance of multiple diseases should be explored.
| As a non-invasive method of indirectly monitoring insect-borne disease, molecular xenomonitoring (MX), the molecular testing of insects for the presence of a pathogen, can provide important information about disease prevalence without the need for human sampling. However, given the successes of tropical disease elimination programs, including many lymphatic filariasis and malaria elimination efforts, parasite levels in many locations are declining. This decrease in prevalence requires the sampling of increased numbers of vectors for disease surveillance and recrudescence monitoring. Such increased sampling poses a challenge since it results in additional costs and labor. In light of these difficulties, high-throughput methodologies for MX are necessary to provide elimination programs with cost-reducing alternatives to long-term disease surveillance. Here we demonstrate proof-of-concept for a new method that samples large numbers of mosquitoes using PCR to screen excreta/feces for filarial or malarial parasites. If operationalized, this approach to MX will provide a practical “first-alert” system that will enable cost-minimizing surveillance in post-transmission-interruption settings. Given this potential, the applicability of this approach to the monitoring of various mosquito-borne diseases should be explored further, as this platform will prove useful for surveillance efforts for a wide variety of pathogens.
| Spanning 73 countries and territories and placing an estimated 1.39 billion individuals at risk of infection, lymphatic filariasis (LF) presents a considerable risk to global health [1]. Similarly, with an estimated 198 million malaria infections and 584,000 malaria-related deaths in 2013, the global burden of human malaria is staggering [2]. Yet despite the wide ranging impacts of these diseases, global elimination efforts have made significant strides, spearheaded by mass drug administration (MDA) programs supported by large pharmaceutical donors [3–5] and the widespread use of insecticidal bed nets [6–9]. As a result, disease prevalence in many locations has decreased dramatically, enabling a growing number of countries to discontinue their treatment efforts for LF [5, 10] and spurring the creation of an increasing number of malaria elimination programs [11–13]. However, lessons learned as a result of LF elimination efforts have shown that the cessation of MDA, recommended after the successful passing of a transmission assessment survey [14], results in an additional set of programmatic challenges. Foremost in such post-intervention settings is the issue of post-MDA surveillance, as vigilant monitoring is required to ensure that recrudescence of disease has not occurred [15]. This monitoring is costly and current efforts for LF are centered upon the periodic sampling of the human population in order to examine circulating levels of filarial antigen [16–17]. While effective, these efforts require blood sampling of the human population. The invasive nature of this practice, coupled with the requirement of informed consent, results in participation challenges [14] that logically increase as populations become further removed from the time of widespread disease transmission. While still largely of future concern, similar challenges likely await the malaria community as control efforts continue to reduce the burden of disease, making this programmatic obstacle one of utmost global importance.
Molecular xenomonitoring (MX), the testing of vectors for the presence of parasite genetic material, has been proposed as a non-invasive means of conducting post-MDA surveillance for LF [14, 17–18]. Although precise correlations between levels of parasite within the vector population and levels within the human population have not been conclusively established, parasite presence within the vector population is indicative of the potential for disease transmission. Furthermore, when monitoring for LF in locations endemic for the Wuchereria bancrofti parasite, a pathogen without a known zoonotic host [19], presence is directly indicative of active human infection. Yet despite its many advantages, MX is costly and when used for monitoring in a post-MDA setting, typically requires the collection and sampling of many thousands of mosquitoes [18, 20–21]. Therefore, as a growing number of countries continue to enter the surveillance phases of their LF eradication programs, alternative methodologies for streamlining, simplifying, and reducing the costs associated with post-MDA monitoring will be required.
As an alternative to traditional approaches to MX, excreta and feces produced by mosquitoes potentially harboring parasites can be tested for the presence of pathogen DNA. Previous work has demonstrated that vector feces-monitoring for the PCR-based detection of Trypanosoma cruzi can be used as a means of surveying insect host infection status [22]. Similarly, it has been shown that genetic material from the Brugia malayi parasite can be successfully detected in the excreta and feces collected from individual mosquitoes [23]. Building upon these findings, we describe methodological proof-of-principle for the real-time PCR-based monitoring for B. malayi parasite DNA in pools of mosquito excreta/feces as a platform for the surveillance of large numbers of insects. While unconventional, excreta/feces monitoring has the potential to provide significant time, cost, and labor savings over traditional MX methodologies due to its exceptionally high-throughput nature. Furthermore, as excreta/feces collection would likely prove readily adaptable to a variety of both passive and active trapping practices and platforms, its potential feasibility as an exceedingly low cost, long-term surveillance tool is great. Equally promising, initial experiments have demonstrated that this approach to MX can be applied to the detection of Plasmodium vivax DNA, indicating its possible usefulness for the monitoring of both unicellular and multicellular eukaryotic pathogens. Given these encouraging findings, the further exploration of mosquito excreta/feces testing as a new method for disease surveillance purposes is warranted and efforts to adapt this alternative MX approach to other mosquito-borne illnesses should be pursued.
Preliminary experiments were designed to determine the effectiveness/efficiency of extracting DNA from the excreta/feces of mosquitoes potentially infected with the B. malayi parasite. To make this determination, various extraction protocols and techniques were tested in order to evaluate their efficiency (Table 1). Because the FR3-derived mosquito cartons containing excreta/feces from potentially infected insects were non-waxed, initial samples were either scraped off of the cartons using a metal spatula, or strips of the carton material (hereafter referred to as carton strips) were directly used as the starting material for the extraction procedure. The amplification of B. malayi parasite DNA from all extracts was evaluated using the previously described real-time PCR primer-probe pairing [26]. Results demonstrated that DNA extractions performed using the QIAamp DNA Micro Kit (Qiagen, Valencia, CA) provided the most consistent and effective detection of parasite DNA. For this reason, this kit was used in all subsequent experiments.
To adapt the Qiagen protocol for use with the bulky, brittle mosquito carton material, minor modifications were made to the manufacturer’s suggested instructions for DNA extraction from bloodspots. Briefly, carton strips were soaked in 360 μl of Buffer ATL for 1 hour prior to incubation with Proteinase K at 56°C. Additionally, following incubation at 70°C, samples were centrifuged at maximum speed for 5 min and supernatants were transferred to new 1.7 ml microcentrifuge tubes. Tubes were centrifuged for an additional 5 min at maximum speed to pellet residual debris and the supernatants were transferred to QIAamp MinElute columns. Lastly, all samples were incubated in Buffer AE at room temperature for 5 min prior to the elution of samples from the columns.
Although preliminary experiments demonstrated that excreta/feces derived from vector mosquitoes fed on B. malayi microfilaria (mf)-positive blood resulted in the amplification of parasite DNA, the availability of mf-containing blood does not guarantee that all mosquitoes will feed or ingest parasites while feeding. Additionally, as the FR3’s standard operating procedure (SOP 8.3) requires that mosquitoes spend three to five days as adults prior to the time an infective blood meal is introduced, a substantial volume of parasite-negative feces was produced and deposited into mosquito cartons prior to blood feeding. Furthermore, as mosquitoes are known to excrete while taking a blood meal [27], it is likely that excreta would be deposited before parasite DNA had reached/been incorporated into the voided material. Therefore, a portion of the voided material collected from mosquitoes provided with mf-positive blood would likely not contain parasites and would therefore not result in a positive PCR. For this reason, a large panel of potentially positive excreta/feces samples was tested in order to estimate the rates of sample positivity. In total, 59 independent samples were tested, with each sample consisting of a 0.48 cm2 carton strip. Based upon observations of the volume of excreta/feces produced by single mosquitoes housed in 50 ml conical tubes, it was estimated that the volume of excreta/feces on each carton strip was equivalent to the average volume produced by one to two mosquitoes over a 24 hour period. Negative control extractions were performed on similar volumes of mosquito feces collected from uninfected C. quinquefasciatus. All samples underwent DNA extraction using the modified Qiagen procedure described above and were analyzed by 45 cycles of real-time PCR using the published reagent concentrations and cycling protocol [26]. 2 μl aliquots of each DNA extract were tested in triplicate and samples returning two or more positive results were considered positive for B. malayi parasite DNA.
In order to determine detection limits for the presence of B. malayi-infected excreta/feces in large pools of uninfected mosquito feces, a titration of samples was created, with each sample containing a 0.48 cm2 strip from a carton used to house mosquitoes provided with a B. malayi-positive blood meal mixed with various volumes of uninfected mosquito feces. Feces from uninfected C. quinquefasciatus mosquitoes were removed from cartons using a cotton swab, and the feces-covered cotton was added to each sample. As 50 uninfected mosquitoes were raised in each carton, and adult mosquitoes were observed to survive for a minimum of 10 days (with most surviving considerably longer), it was conservatively estimated that each carton contained a minimum of 500 mosquito feces/days (i.e. the amount of feces produced by 500 mosquitoes in one 24 hour period, or the amount of feces produced by a single mosquito over a 500 day period). While the distribution of feces within cartons was not precisely uniform, by sectioning cartons based upon total internal surface area (approximately 1,050 cm2), it was possible to roughly estimate the number of mosquito feces/days being added to each sample. Samples estimated to contain approximately 62.5, 125, 250, and 500 feces/days were prepared. Negative control extractions were also prepared using mosquito feces collected from uninfected C. quinquefasciatus. All samples were extracted and tested in duplicate reactions using the same extraction and detection methods as described above for the evaluation of PCR positivity testing.
To test whether the detection of mosquito-borne pathogen DNA from mosquito excreta/feces was possible for species other than the B. malayi parasite, a set of samples was created using mosquito excreta/feces produced by carton-raised A. stephensi that had been fed on P. vivax-positive blood. As was done for B. malayi detection, samples were prepared by excising 0.48 cm2 carton strips containing potentially positive excreta/feces. To establish proof-of-principle, 20 samples were prepared and DNA was extracted using the modified Qiagen protocol described above. DNA extracts from each sample were tested using a previously described primer-probe set for the universal detection of Plasmodium species [28] with reaction recipes and cycling conditions remaining consistent with the authors’ published protocol.
Carton strips were excised from containers used to house A. aegypti mosquitoes provided with B. malayi mf-containing blood and testing was conducted to determine the percentage of excreta/feces samples containing B. malayi DNA. Such testing was necessary since the production of feces can occur prior to the provision of an infective blood meal or before the ingestion of a blood meal. Furthermore, the availability of infective blood does not guarantee that each individual mosquito will feed and, dependent upon the mosquito species, localization of parasite material to voided excreta/feces may take time following blood meal ingestion. Accordingly, DNA was extracted from 59 independent samples, each consisting of a carton strip measuring 0.48 cm2 and containing excreta/feces from one to two mosquitoes over a 24 hour period (i.e. one to two mosquito feces/days). Real-time PCR testing, using 2 μl of template DNA resulted in positive detection for 21 out of 59 samples tested (35.6%). For positive samples, mean Ct values ranged from 26.62 (+/- 0.24) to 41.98 (+/- 0.03) (Table 2). Because only a fraction of the deposited mosquito excreta/feces would contain parasite DNA, 35.6% may be a true indication of the frequency of positive samples.
A titration of samples containing potentially positive 0.48 cm2 carton strips mixed with varying amounts of uninfected mosquito feces was prepared in order to estimate the limits of detection for B. malayi-based excreta/feces testing. In total, five samples containing an estimated 62.5 mosquito feces/days, six samples containing an estimated 125 mosquito feces/days, six samples containing an estimated 250 mosquito feces/days, and two samples containing an estimated 500 mosquito feces/days were assayed. As expected, due to the uncertainty of which samples actually contained B. malayi DNA, a fraction of the samples failed to give positive PCR detection of B. malayi DNA. However, detection of parasite DNA proved possible at all tested levels of sensitivity (Table 3).
To explore whether excreta/feces testing would efficiently detect pathogen DNA from species other than B. malayi, testing for the presence of the human malaria-causing parasite P. vivax was performed. To demonstrate proof-of-concept, 20 samples were prepared and tested by PCR. Each sample was comprised of a 0.48 cm2 carton strip excised from a mosquito container having housed A. stephensi female mosquitoes provided with Plasmodium-positive blood. Real-time PCR testing of DNA extracted from each sample clearly demonstrated the adaptability of excreta/feces testing to the detection of P. vivax since all samples were positive with Ct values ranging from 26.82 (+/- 0.26) to 29.21 (+/- 0.80) (Table 4).
While sensitive and less intrusive to the local population than human sampling, the number of studies implementing current MX practices for the surveillance of LF or malaria has been limited. Although such efforts provide valuable data [10, 17–18, 21] the routine use of MX for post-MDA LF surveillance or long-term recrudescence monitoring is not yet standard procedure. Despite the existence of effective molecular tools [28–29], vector monitoring for malaria is even more uncommon and World Health Organization recommendations for infection monitoring and prevalence estimation rely solely on human sampling [2]. Limited implementation has occurred for multiple reasons, including the need to process and test large numbers of mosquitoes from areas suspected of having low parasite density within the vector population [10, 18, 21]. Difficulties in establishing a concrete correlation between vector-parasite levels and human prevalence have further restricted MX implementation [21]. Yet despite these shortcomings, MX continues to receive attention as the need for post-intervention disease surveillance continues to grow and mosquito trap designs continue to improve [30–34]. Accordingly, methodologies capable of harnessing the advantageous aspects of MX while making its practice more practical and inexpensive would be of great benefit to global LF and malaria elimination efforts, as well as to monitoring efforts for other vector-borne diseases.
The work presented here provides methodological proof-of-concept for a novel approach to MX with the potential to greatly reduce the cost, time, and labor associated with large-scale surveillance efforts. The successful amplification of parasite DNA from pooled mosquito excreta/feces containing B. malayi genetic material has demonstrated that high-throughput MX for LF is feasible. In the past, real-time PCR-based MX for the presence of the filariasis-causing parasites has been restricted to the testing of pools of 25 or fewer mosquitoes. This is because the biological mass of mosquitoes and high yields of mosquito DNA associated with pools of large size results in the inability to detect the presence of small quantities of parasite DNA [35]. However, excreta/feces testing enables the sampling of material obtained from vast numbers of mosquitoes, while simultaneously limiting the biological mass associated with each sample. As we have demonstrated, it is possible to detect trace amounts of parasite DNA in pools containing the voided material from as many as 500 uninfected mosquitoes. Future studies implementing this approach will benefit from the drastic reduction in cost of DNA extractions and PCR (approximately 20-fold). Furthermore, as it has been shown that non-vector mosquitoes rid themselves of parasite material more rapidly than vector species (as indicated by a shortened period of time during which parasite detection is possible within non-vectors [23]), one would expect to find greater quantities of parasite DNA within the excreta/feces of non-vector mosquitoes. Therefore, the testing of mixed pools of vector and non-vector excreta/feces should be possible. While such testing will result in reduced ability to directly correlate the presence of parasite with individual vector species, it will likely increase the sensitivity of detection when surveying for the presence of parasite in post-transmission-interruption settings as both vector and non-vector mosquitoes potentially harboring parasite material will be screened. In addition, it is likely that excreta/feces testing will eliminate the need for the labor intensive and time consuming species-sorting efforts which are commonplace in current MX work [10, 17–18, 21, 36]. By drastically reducing the numbers of pools that must be screened and by eliminating the need for sorting mosquitoes by species, labor-related time and costs are dramatically reduced.
While operationalizing this alternative approach to MX presents some implementation hurdles, adaptation of current passive and active trapping methods to the collection of mosquito excreta/feces is possible. Such adaptation could occur by transferring live mosquitoes from a trap to a holding carton, in which they would be sugar fed using a cotton ball, thereby encouraging the voiding of waste material. Expired mosquitoes would then be removed and additional mosquitoes could be added following further collection from the trap. Periodic testing of the accumulated excreta/feces would enable the high-throughput screening of the voided material from a series of such traps. Any trap with the capacity to maintain live mosquitoes could be used for this purpose including the CDC Gravid Trap, the Ifakara tent trap and others [30, 37]. Alternatively, collection of excreta/feces could occur directly within traps of various designs. One such design proving readily adaptable to excreta/feces collection in preliminary experiments is the “Large Passive Box Trap” developed by Ritchie, et al [38]. While work aimed at evaluating the adaptability of this trap to the collection of various species of mosquitoes is currently ongoing, and further efforts to optimize this trap for the purpose of excreta/feces collection will be required, simply lining the internal surfaces of this passive trap with waxed paper provides an uncomplicated method for collecting the accumulated material voided by trapped mosquitoes (S1 Fig). Swabbing the excreta/feces from the waxed paper then enables the PCR analysis of pooled material.
Additional testing will be required to determine the stability of parasite DNA in mosquito excreta/feces over time and under field conditions. However, in the proof-of-concept experiments described in this paper, mosquito excreta/feces containing parasite DNA was allowed to accumulate for 14–16 days prior to transfer to cold storage. In this setting, parasite DNA remained stable and detectable (Table 3). While further validation under conditions mimicking tropical temperatures and humidity will be required, these results are encouraging, as DNA stability within tropical and sub-tropical climates could present another hurdle when operationalizing this method in the field.
Since production of feces can occur prior to the provision of a parasite-positive blood meal and since this provision does not ensure that all mosquitoes will ingest and/or metabolize a parasite, a percentage of the excreta/feces samples collected will likely test negative for parasite DNA. It is therefore difficult using blood-fed mosquitoes to definitively assess the consistency of detection of parasite DNA in excreta/feces. During initial testing, we demonstrated that 21 out of 59 samples comprised of 0.48 cm2 carton strips derived from containers used to rear mosquitoes with a B. malayi-positive blood source were positive (Table 2). However, although sufficient to fulfill our primary aim of providing methodological proof-of-concept, it cannot be conclusively determined whether the remaining 38 samples were all truly negative for parasite DNA. While spiking uninfected excreta/feces samples with extracted B. malayi genomic DNA would provide clear positive and negative samples, this approach is extremely artificial and has decreased biological relevance since it eliminates any possible effects of mosquito metabolism on the integrity of parasite DNA. Since the major uses of excreta/feces testing will likely center on mapping and long-term, low-cost, post-transmission-interruption recrudescence monitoring, marginally reduced consistency of detection has diminished significance as continuous, sustainable, high-throughput surveillance would enable detection of even low-levels of parasite prevalence. The high-throughput nature of this testing was clearly demonstrated by the positive detection of parasite DNA derived from pools containing various volumes of negative feces up to 500 mosquito feces/days (Table 3). Detection proved possible at all tested sensitivity levels and with overall sample positivity rates similar to those obtained when testing potentially positive excreta/feces samples without the addition of negative feces (36.8% vs. 35.6% respectively). Thus, the inclusion of large amounts of negative feces does not appear to alter detection efficiency. Given these findings, sustainable, high-throughput surveillance efforts using excreta/feces screening could serve as a “first-alert” platform, with positive detection serving as a “red flag” for recrudescence in settings of known transmission interruption. In such a scenario, detection would spur the implementation of more traditional surveillance and monitoring studies.
By successfully detecting P. vivax DNA in pools of excreta/feces produced by Plasmodium-positive-blood fed A. stephensi, we have provided proof-of-principle for the application of this platform to the detection of malaria parasites. Furthermore, the increased rates of sample positivity and decreased Ct values seen when assaying for P. vivax are not entirely surprising and indicate this system may work even better for malaria than LF. Estimates have suggested that the ratio of Plasmodium merozoites to gametocytes within the peripheral blood is as great as 156:1 [39–40]. Given this ratio, the vast number Plasmodium merozoites ingested during a blood meal (up to 32 per infected erythrocyte [41]), and knowledge that merozoites obtained during blood feeding are unable to undergo further development within the mosquito host (only gametocytes undergo further development [42]), the great majority of ingested parasites are simply metabolized and/or eliminated by the mosquito. In contrast, while mosquito hosts possess measures that provide partial protection against filarial infection [43–44], and environmental conditions are thought to impact rates of parasite survival [45], all filarial parasites taken up as part of a blood meal are of the correct lifecycle stage (mf) to potentially undergo further development within the vector host. Therefore, due to the varying natures of their lifecycles, it follows that a greater percentage of filarids ingested during a blood meal are able to successfully develop within the mosquito host as compared to Plasmodium. Since successful parasite development would likely mean the absence of parasite DNA in mosquito excreta/feces, the lower levels of sample positivity and the more modest Ct values observed during B. malayi testing compared to P. vivax testing seem logical.
With its adaptability to both B. malayi and P. vivax, MX of mosquito excreta/feces for various other mosquito-borne pathogens should be explored. Given the successes realized with the detection of these parasites, it is extremely likely that similar detection will prove possible for W. bancrofti and other malaria species. However, the applicability of this new platform to other types of pathogens should also be examined, since improved high-throughput screening for RNA viruses such as Dengue, Chikungunya, and Zika would be welcomed programmatic tools. Furthermore, since all species of biting insects draw from the same reservoir of blood within a target host, the possibility of cross-vector monitoring should also be considered. For example, excreta/feces samples collected from mosquitoes could be monitored for the presence of disease-causing agents having unrelated insect hosts (such as Leishmania ssp. or Loa loa). Adaptability to various pathogens and the possibility of cross-vector monitoring could also make excreta/feces sampling an attractive strategy for tropical disease integration efforts. In light of these factors, and the potential time, cost, and labor savings associated with such applications, we believe that this proof-of-concept study suggests that further evaluation of this new method is warranted.
|
10.1371/journal.pgen.1004885 | Interactions of Chromatin Context, Binding Site Sequence Content, and Sequence Evolution in Stress-Induced p53 Occupancy and Transactivation | Cellular stresses activate the tumor suppressor p53 protein leading to selective binding to DNA response elements (REs) and gene transactivation from a large pool of potential p53 REs (p53REs). To elucidate how p53RE sequences and local chromatin context interact to affect p53 binding and gene transactivation, we mapped genome-wide binding localizations of p53 and H3K4me3 in untreated and doxorubicin (DXR)-treated human lymphoblastoid cells. We examined the relationships among p53 occupancy, gene expression, H3K4me3, chromatin accessibility (DNase 1 hypersensitivity, DHS), ENCODE chromatin states, p53RE sequence, and evolutionary conservation. We observed that the inducible expression of p53-regulated genes was associated with the steady-state chromatin status of the cell. Most highly inducible p53-regulated genes were suppressed at baseline and marked by repressive histone modifications or displayed CTCF binding. Comparison of p53RE sequences residing in different chromatin contexts demonstrated that weaker p53REs resided in open promoters, while stronger p53REs were located within enhancers and repressed chromatin. p53 occupancy was strongly correlated with similarity of the target DNA sequences to the p53RE consensus, but surprisingly, inversely correlated with pre-existing nucleosome accessibility (DHS) and evolutionary conservation at the p53RE. Occupancy by p53 of REs that overlapped transposable element (TE) repeats was significantly higher (p<10−7) and correlated with stronger p53RE sequences (p<10−110) relative to nonTE-associated p53REs, particularly for MLT1H, LTR10B, and Mer61 TEs. However, binding at these elements was generally not associated with transactivation of adjacent genes. Occupied p53REs located in L2-like TEs were unique in displaying highly negative PhyloP scores (predicted fast-evolving) and being associated with altered H3K4me3 and DHS levels. These results underscore the systematic interaction between chromatin status and p53RE context in the induced transactivation response. This p53 regulated response appears to have been tuned via evolutionary processes that may have led to repression and/or utilization of p53REs originating from primate-specific transposon elements.
| It is well established that p53 binds DNA elements near p53 target genes to regulate the response to cellular stress. To assess factors influencing binding to response elements and subsequent gene expression, we have analyzed 2932 p53-occupied response elements (p53REs) in the context of genome-wide chromatin state, DNA accessibility and dynamics, and considered roles for binding-sequence specificity and evolutionary conservation. While p53 occupancy level shows little apparent direct relationship to gene expression change, after grouping expressed genes by their chromatin status at baseline, a relationship between occupancy of p53REs and gene expression change emerged. Analysis of p53RE sequences demonstrated that p53 occupancy was strongly correlated with sequence similarity to p53RE consensus, but surprisingly, was inversely correlated with nucleosome accessibility (DHS) and evolutionary conservation. These data revealed a systematic interaction between p53RE content and chromatin context that affects both quantitative p53 occupancy and the induced transactivation response to exposure. Moreover, this interaction appears to have been tuned via evolutionary events involving transposable elements, which strongly bind p53, but in only a few instances affect gene expression levels. Models of p53-regulated gene expression response that consider both chromatin state and sequence context may prove useful in guiding strategies for cancer prevention or therapy.
| Tumor suppressor p53 is activated in response to DNA damage and cellular stress signals and regulates the expression of target genes to elicit cell-growth arrest, DNA damage repair, or apoptosis to prevent the propagation of damaged or compromised cells [1], [2]. Understanding the regulatory logic of p53 is critical to understanding p53 biology in normal and tumor cells. The canonical p53 responsive element (p53RE) is composed of two decamers of RRRCWWGYYY, where R = purine, W = A or T and Y = pyrimidine, separated by a spacer of 0–21 nucleotides (nt), leading to millions of putative p53RE sites in the human genome [3]. About two hundred p53REs have been characterized in detail but p53 chromatin immunoprecipitation sequencing (ChIP-seq) experiments indicate there are thousands of p53 targets and numerous exposure-specific patterns of binding and transactivation [4]–[9]. These patterns have been variously attributed to sequence-specific binding [5], [6], p53 post-translational modifications [10], targeting coactivators/factors [11], post-transcriptional effects [12], as well as, chromatin status at the binding site [13]–[15]. However, the rules governing the sequence specificity and functional output of regulatory interactions between p53 and the genome are not yet fully understood.
A number of in vitro studies have clearly demonstrated that p53RE sequence variation, including polymorphisms and spacer length, affect p53 binding to DNA and subsequent transcriptional activation [16]–[20]. These studies show that p53REs with higher similarity to the p53RE consensus display stronger binding and those that contain spacer sequences larger than 1 nt between the decamers show dramatically reduced binding. These studies, however, did not consider the impact of the varied genomic chromatin contexts on p53 binding to response elements and evolutionary conservation of p53REs. In addition, p53 binding motifs occur frequently in primate-specific interspersed repeats, including retroviral long terminal repeats (LTRs [21]), short (SINE [22], [23], such as Alu), and long interspersed nuclear elements (LINEs [24]). Considering the potential role of transposition in the evolution of cis-regulatory elements [25], we ask if these recently evolved p53 binding sites have acquired sequence properties or chromatin contexts which might indicate if they are functionally suppressed or utilized.
Chromatin accessibility can be a primary determinant of transcription factor (TF) occupancy [26], [27]. For example, ligand-activated glucocorticoid receptor binding occurred with near absolute preference (∼95%) for binding sites located in accessible chromatin [26] and recent Encyclopedia of DNA Elements Project (ENCODE) reports [28] have supported this phenomenon for numerous other TFs. In contrast, Nili et. al. [14] described a different situation for the tumor suppressor protein p53, demonstrating that in the tumor cell line MCF7, p53 often binds genomic regions that have high existing nucleosome density and low chromatin accessibility. Other studies have suggested that stress-induced binding at specific CDKN1A (p21) p53REs is associated with chromatin structure at these locations [15]. These observations argue for further exploration of the role of chromatin-accessibility in regulating p53 access to DNA regulatory elements. Specific chromatin modifications at histone proteins correlate with TF binding, transcription initiation, elongation, enhancer activity and repression [29]. To elucidate how p53 RE sequences and local chromatin context interact to affect p53 binding and gene transactivation, we have taken advantage of a well-characterized lymphoblastoid cell-culture exposure model system [30]–[32]. The ENCODE project has generated Hidden Markov models (HMM) [32], [33] that annotate the human genome into distinct (15 or 7 HMM states) chromatin states. These models are based on combinatorial patterns of eight informative histone modification marks, Zinc finger CCCTC-binding factor (CTCF), the presence of DNase I hypersensitivity (DHS) and the spatial relationships with gene transcription across nine cell types. The ENCODE ChromHMM model describes the chromatin landscape in unstressed cells and highlights transcriptional regulatory elements reliably and robustly [32], [34]. We have integrated these ENCODE data sets into our experimental data with the aim of deconvoluting the roles for chromatin context/state, p53RE sequence, and p53 occupancy in the transactivation of p53 target genes.
Thus, in the present study we have examined genome-wide changes in p53 binding (ChIP-seq), H3K4me3 (ChIP-seq), gene expression, and DNase I hypersensitivity (DNase-seq) that were induced by the DNA-damaging, chemotherapeutic agent doxorubicin (DXR) in human lymphoblastoid cell lines (LCL). The goal was to assess these changes in the context of ENCODE chromatin states, and consider the role of p53RE sequence content and evolutionary conservation. We revealed several unique relationships, including inverse correlations between p53 occupancy and both chromatin accessibility and evolutionary conservation. Notably, our novel, integrated genome-wide analysis demonstrated that p53RE sequence content is highly correlated with genomic occupancy, and that chromatin state context strongly modulates the relationship between p53 occupancy level and changes in gene expression. Furthermore, the interaction between these features appears to have been shaped by evolutionary selective pressure, likely driven by transposable elements. Our study sheds new light on the complex interrelationship between chromatin state and p53RE sequence in p53 genomic occupancy, and suggests the importance of considering the interactions of sequence content and epigenetic factors in interpreting p53-mediated stress responses.
We carried out p53 activating treatments in LCLs (GM06993, GM11992, and GM12878) using a 0.3 µg/mL dose of DXR and prepared mRNA samples at 4 and 18 hrs of treatment. Chromatin for p53 and H3K4me3 ChIP-seq was prepared from samples collected at 18 hrs, a time point when both growth arrest and apoptosis-related p53 gene targets are occupied [35]. Following DXR-treated p53 ChIP-seq, we detected 6436 peaks at a 5% false discovery rate (FDR) using the QuEST Program [36]. Then, using a cut-off of 30 overlapping sequence reads, we classified 2932 p53 peaks in DXR-treated samples as high-confidence peaks, and with the same sequence read cutoff there were thirty p53 peaks evident under no treatment (NT) conditions. As in primary lymphocytes [37], we observed very limited p53 binding in the absence of stress and the DXR treatment greatly increased the occupancy of p53 at known and newly identified response elements. Among these high-confidence peaks, 91% (2664/2932) have putative p53REs, and 122 of them overlapped known p53 binding sites (Table 1, S1 Fig.). All further analyses focused on these high confidence p53 peaks and the 2415 genes that were associated with them by proximity. Fig. 1 displays a UCSC genome browser view of a sample of these data integrated with ENCODE data from GM12878 cells. Tracks A, B, F, H, I, are experimental data from LCLs generated in our lab for the present study. Tracks A–E display p53 ChIP-seq peaks in the PGPEP1 (Pyroglutamyl-Peptidase I)-GDF15 (Growth Differentiation Factor 15) region from this study and two other independent studies [7], [8] using MCF7 cells and IMR90 cells. Many p53 peaks are highly reproducible across multiple experimental conditions (orange box), while other binding locations are specific to treatment and tissue as noted by others [8]. Comparing the present data with three other published p53 ChIP-seq datasets we determined that 69% of 2932 peaks were observed in at least one other independent study. Among the high-confidence peaks unique to this study, 91% (822/903) were observed as significant peaks in an independent p53 ChIP-Seq experiment in LCLs. Histone H3 lysine 4 tri-methylation (H3K4me3, tracks H and I) is present at the PGPEP1 actively transcribed promoter (red box) but very low at the GDF15 promoter (track I, purple box), which displays insulating CTCF marks (CTCF, track J, also blue arrows, blue box) in the untreated cells.
To provide a multifaceted view of how the chromatin landscape impacts p53-occupied genes and to explore if p53-inducible genes are actively suppressed at baseline, we have used the ENCODE Hidden Markov Models (two versions, ChromHMM15 [33] and ChromHMM7 Combined [32], which are based on ChIP-seq measurements for multiple histone modifications, CTCF ChIP-seq, DNaseI-Seq, and FAIRE-seq as measured in nine cell lines, including the LCL GM12878, under basal conditions. In Fig. 1, the ChromHMM7 track (G) summarizes the chromatin characteristics of this region of the genome. For example, the red box highlights the promoter region of the p53 inducible gene, PGPEP1. Numerous histone modifications and DHS align at this position, and the ChromHMM7 track displays red indicative of an active promoter (State 1). In the same genomic region within the orange box where p53 peaks were detected, the ChromHMM track displays a strong DHS peak and multiple histone modifications (tracks J) particularly H3K27ac, H3K4me1 and H3K4me2 indicative of an active, distal enhancer. We used the ChromHMM7 Combined model (ENCODE website) to classify p53 occupied regions and genes as: 1) active promoter (e.g., PGPEP1, active histone marks present: H3K4me2, H3K4me3, H3K27ac, H3K9ac at>93% frequency); 2) promoter flanking region; 3) enhancer (H3K4me1 at>96% frequency); 4) weak enhancer/open chromatin; 5) insulator CTCF enriched; 6) transcribed; 7) repressed, including polycomb and heterochromatin. Some examples of known characterized p53 target genes and their associated HMM states are listed in Table 1.
The GDF15 gene had very low expression under no treatment conditions with very little H3K4me3 at its TSS and was strongly induced following p53 activation with the increase of H3K4me3 marks (Fig. 1, purple box). We examined these phenomena across all p53 occupied genes. Of the 2932 DXR-induced p53-occupied regions, 1697 genes had detectable gene expression levels using the Affymetrix Exon Array 1.0 either under no treatment (NT) conditions or after treatment. We grouped expression values into deciles based on baseline H3K4me3 levels at TSS (ordered from low to high, Fig. 2A), we observed a very striking, linear relationship over time (NT, 4 hr, 18 hr after DXR treatment). Expression values for untreated cells (black bars) display a strong linear trend for increasing mean gene expression values with increasing H3K4me3 decile. Following DXR treatment (open bar = 4 hr, gray bar = 18 hr), the group of genes with low initial H3K4me3 levels (deciles D1–D3, left side of graph) show the largest change in expression. Thus highly inducible p53-regulated genes typically do not display activating H3K4me3 marks in unstressed cells. We performed regression analysis on the DXR-induced change in H3K4me3 modification with the DXR-induced change in gene expression which showed a significant correlation (Fig. 2B, r2 = 0.41, P<0.0001). As mentioned, the highly-induced p53 gene GDF15 displayed both low H3K4me3 and low expression (Fig. 1, purple box) at baseline. However, GDF15 also displayed insulating CTCF binding (blue box) at TSS. We tested if all low H3K4me3 DXR-induced p53 genes were enriched with repressive chromatin marks at the TSS or gene body relative to down-regulated genes, and determined that highly-inducible p53 genes were characterized by having preexisting chromatin marks indicating insulated (CTCF), polycomb (PcG/H3K27me3), or repressive heterochromatin status (trend test, P<0.0001, Fig. 2C). Examples of known p53 genes displaying repressive chromatin marks are listed in Table 1.
Fig. 3A demonstrates the distribution of p53 peaks among ChromHMM states and the median distance of the binding locations to the nearest TSS. While the genome of GM12878 cells can be segmented by ChromHMM into promoters (∼1%), enhancers (∼15%), transcribed (∼18%) and repressed regions (∼64%) [33], [38] (S1A Fig.), it is quite clear that p53 binding was highly enriched for promoters, enhancers, and transcribed regions. Most characterized p53RE sites occur in active promoters (S1B Fig.; for examples, see Table 1). While most binding in active promoters was proximal to transcription start sites (TSS), it is notable that 42% of binding locations (those in enhancers, CTCF regions, transcribed regions) have median distances that are at least 32 Kb distal from a TSS, and many binding sites located in repressed regions are much further from a TSS (Fig. 3A). These distal p53 binding locations found in enhancers and repressed chromatin were reproducible in other published p53 ChIP-seq datasets;>76% of these peaks were observed in at least one other experiments (S1 Table).
We also used the ENCODE ChromHMM15 at the TSS to classify the state of the gene nearest to the p53 ChIP-seq peak (illustrated as in Fig. 1 lower panel and Fig. 3B) to identify “poised” promoter regions that carry both repressive (H3K27me3) and activating (H3K4me2/3) marks, and these genes on average showed>3-fold induction (S2B Fig.). Fig. 4A–B displays two genes, RRAD (Ras-Related Associated with Diabetes) and SULF2 (Extracellular Sulfatase 2) with H3K27me3 and H3K4me2 marks (purple boxes), no H3K4me3 marks were present at the TSS, and expression levels were low with no treatment. Following treatment with the DNA damaging agent DXR, H3K4me3 marks appeared (red box) and for each of these poised genes, mRNA expression increased dramatically along with increases in DHS at 4 and 18hrs (orange boxes). However, not all genes follow this pattern. While TGFA (Transforming Growth Factor, alpha) (Fig. 4C) had similar H3K4me2, H3K27me3, and CTCF marks indicative of a poised promoter at baseline and also displayed a very large, upstream p53 binding peak, only minimal H3K4me3 and gene expression changes were observed. Thus p53 occupancy had little effect on TGFA transactivation, although there was some effect on DHS at the location of CTCF binding (blue box), suggesting other factors may be important for TGFA regulation. The patterns observed for chromatin accessibility change (DHS tracks) for individual genes were notable. Peaks of DHS were co-located at sites of H3K4me3 change (red box vs orange box) and also appeared to align with pre-existing H3K4me1/2 and/or CTCF binding patterns (Figs. 4A-C, blue arrows, blue boxes). Among genes that displayed pre-existing heterochromatin marks at the location of their p53 binding peak (e.g. Fig. 4C, TGFA; green box), it appeared that p53 binding had minimal detectable effect on nucleosome accessibility (DHS) at some binding sites, even when examined at greater magnification (S3 Fig.).
Comparing the level of p53 occupancy (ChIP-seq peak density) among ChromHMM states for all peaks (n = 2932), we observed significantly higher levels of occupancy for p53 peaks located in distal enhancers relative to promoters and other states (S4A Fig.). However, the result of grouping peaks by ChromHMM at the TSSs of nearby genes indicated that the small group of genes with insulating CTCF at their TSS displayed the highest p53 occupancy (S4B Fig.). The effect of insulated CTCF binding near the TSS prior to treatment was dramatic and this group of genes showed the greatest average change in gene expression (Fig. 5A). Overall p53 peak density displayed a very modest, but significant, correlation with change in gene expression (r2 = 0.06, P <0.0001, S4C Fig.). However, the more correlated patterns observed for peak density and change in gene expression after stratifying genes by ChromHMM state at the TSS (Figs. 5B, S4D), suggested that chromatin status modifies the relationship between binding/occupancy and transactivation.
Because recent investigations have suggested that prior chromatin state (before treatment) could predict the occupancy of inducible transcription factor binding [27], [33], [39], we sought to examine the ChromHMM defined state at the location of p53 occupancy and asked the extent to which p53 occupancy level was due to specific p53RE sequence content at the binding site or due to the open-chromatin accessibility (context) at binding locations prior to exposures to DXR. S5A Fig. displays the sequence logo for 103 experimentally verified p53REs that make up the p53 consensus and it closely matches the most enriched motif among p53 ChIP-seq peaks identified by de novo motif discovery [40] in this study (Fig. 6A). The most enriched motifs were determined for peaks in each ChromHMM state (S5B–D Fig.) and as expected, each ChromHMM state contained enriched motifs similar to the p53 consensus. However, among peaks lacking full canonical p53REs, p53RE half-sites (decamers) were common (90%, 239/267, Data File S1). The CTCF motif was enriched among peaks associated with down-regulated genes at open promoter state (S5B Fig.). For individual binding sites, we identified putative response elements and considered the calculated binding strengths in relation to chromatin state and other features such as DHS or evolutionary conservation. GADD45A (Growth arrest and DNA-damage-inducible, alpha) and BCL2A1 (BCL2-related protein A1) (S5E–F Fig.) are examples of how these parameters vary considerably across p53REs and other examples are provided in Table 1. We predicted p53RE binding strength using a position weight matrix (PWM) model and a second model based on in vitro measurements of p53 binding (Binding PWM, S6A Fig. [16], [31]) and then compared p53RE strength across ChromHMM states (Figs. 6B, S6C). Of interest, mean PWM scores for the p53RE nearest the peak maximum were relatively high in enhancers, consistent with high occupancy at enhancers; the mean PWM scores in repressed chromatin regions were highest. However, the PWMs for the p53REs located in open promoters were much lower. Among p53 peaks associated with genes having open promoters, we observed that both mean PWM and mean peak density were lower among down-regulated genes relative to up-regulated genes (Fig. 6C), and that there was a distinct DNA motif difference as well. Occupancy by p53 was more frequent among induced genes (69% vs 31%; up-regulated genes vs down-regulated genes; Fig. 6C). About 85% (270/317, fold change <0.8) of down-regulated genes displayed active promoters (Fig. 6C) and were highly expressed at baseline (S2A Fig.). Using the functional annotation program DAVID we examined Gene Ontology (GO) annotations of up and down-regulated genes among all chromatin states (Data File S1). The analysis revealed that p53 was the most likely regulator of these genes and that the most significant GO term is “regulation of cell death”. The enrichment for “DNA damage response” and “induction of apoptosis” was particularly strong when we compared highly occupied p53REs vs less occupied p53REs (Data File S1, DAVID analysis).
The p53 consensus RE is also characterized by a variable spacer region of 1–21 nt between the two decamer half-sites, and since the PWM is calculated independently of the variable spacer, these represent independent features. The length of the p53RE spacer is known to reduce the affinity of p53 for binding to its consensus sequence and transactivation [41]. While spacers longer than 3 nt were relatively rare, the weak p53REs that were located in active promoters were more likely to have spacers and have larger spacers (S6B Fig.), which was associated with weaker binding sequences (Fig. 6B). The negatively correlated patterns of spacer length and PWM strength (S6B–D Fig.) were further evidence that these weak p53REs may be functional. Thus, p53REs located in active promoters were weaker by two distinct criteria, while elements in more distant locations in enhancers and in repressed chromatin states appear to be significantly stronger.
To directly visualize the relationship between p53RE binding strength score (PWM), occupancy, and chromatin accessibility, we ordered the dataset by computed PWM scores from low to high, and grouped data into deciles (Fig. 7). We first tested the relationship between PWM values and the frequency of spacers between the half-sites (Fig. 7A, spacers of various lengths shown by stacked bars) and observed spacers to be more common in p53REs with weaker PWM scores (consistent with S6B, S6D Figs.). A surprising result was that p53 peaks containing the strongest PWM scores proved to be far distant from TSSs (Fig. 7B, decile D10) while the weakest p53REs were <1 kb from the TSS. Particularly notable was the linear relationship between p53RE binding strength (PWM score) and p53 occupancy (peak density, Fig. 7C). To assess if this observation was unique only to DXR-treated LCLs, we accessed p53 ChIP-seq data from 5-fluorouracil (5FU) treatment of MCF7 breast cancer epithelial cells [8], calculated p53RE PWM scores within p53 ChIP-seq peaks (S7A Fig.), and compared PWM with ChIP-seq density (S7B Fig.). A similar linear relationship between PWM score and peak density was observed in MCF7 cells. By comparing these results with other published p53 ChIP-seq datasets [7]–[9] we determined that p53 ChIP-seq peak density was also highly significantly correlated with reproducibility of p53 peaks across multiple experimental systems (Data File S1). ChIP-seq peaks containing the weakest p53RE PWMs had lower occupancy and were associated with the highest DHS scores in untreated cells (Fig. 7D). This observation was also confirmed in MCF7 cells (S7C Fig.). Thus we observe that open promoters that were characterized by the presence of pre-existing accessible chromatin (DHS) were permissive for p53 binding to the weakest elements. Conversely, the p53 binding sites with the highest PWMs and highest occupancy tend to be in the most distal p53 binding locations (∼60 kb from a TSS and in heterochromatin), most of these sites display very low DHS, and these observations were independent of cell type.
Gene regulatory regions, including promoters and enhancers typically display high evolutionary conservation, presumably due to negative selection pressure to preserve functional DNA elements [42]–[44]. However, we have previously noted that evolutionary conservation varied greatly among well-characterized functional p53REs, with some human elements displaying little homology with other mammals [45], and we wondered if p53RE conservation varies by chromatin state. Grouping p53REs by their ChromHMM state and plotting their average PhyloP score across the 20nts of the RE dramatically demonstrates this phenomenon (Fig. 8A). Response elements located in promoters and enhancers were highly conserved relative to those in CTCF enriched, transcribed, or repressed regions. When we ordered p53REs by PWM decile we revealed an inverse correlation between p53 occupancy and evolutionary conservation of the peak region, either by mouse-human percent identity (Fig. 8B) or by PhyloP score (S8A Fig.). These analyses demonstrate that more highly conserved p53REs are relatively weak p53REs and display lower occupancy relative to poorly conserved sequences.
Some primate-specific transposable elements (TEs) are known to contain p53REs [21], [22], [24]. To evaluate their influence on p53RE conservation scores, we analyzed the overlap of occupied p53REs (full sites with spacer ≤3) with repeat elements categorized in Repeat Masker (http://www.repeatmasker.org/faq.html). We observed that 33% (970 of 2932) of p53-occupied sites were embedded in distinct families of repeat elements, with a significant enrichment for the endogenous retroviral elements MER61, LTR10B1, and MLT1H, as well as others (Data File S1). Most TE-associated p53 ChIP-seq peaks (82.7%; 803/970) were also observed in independent experimental datasets (Data File S1). The presence of TE-associated p53REs strongly affected conservation scores among HMM states (Figs. 8A, S8B–C). For example, TE frequency was inversely associated with PhyloP scores, being high in heterochromatin, intermediate in enhancers, low in promoters (S8B, C Fig.). As expected, average PhyloP values were significantly lower among all TE-associated p53REs relative to non-TE p53REs (Fig. 8C). Surprisingly, when we examined PhyloP values for individual families of TEs, we observed Alu and L2 family p53REs were dramatically negative compared to all others (Fig. 8C). We considered if chronological age of the TE might explain this pattern but this is not the case because Alu repeats are among the oldest of TE families while L2 is among the youngest [46]. Negative PhyloP values are predicted to be fast evolving and may be related to recent evolutionary selection as suggested by Pollard, et al [47] and this would be consistent with the p53 pathway acquiring new primate specific functions.
To test if TE-associated p53REs affect sequence content (PWM, Fig. 8D) and occupancy (p53 ChIP peak density, Fig. 8E), we compared TE-associated p53REs with that of non-repeat p53REs. We observed significantly higher PWM scores in all TE-associated p53REs, with several families particularly high (MER61, MLT1H, LTR10B; Fig. 8D) and the p53 ChIP-seq peak density plot displays a similar, correlated pattern (Fig. 8E). Thus the relationship between PWM and ChIP-seq peak density that we observed previously among all p53 peaks (Fig. 7C), holds among families of repeats (S8D Fig.). If these peaks were functionally associated with nearby genes we would expect treatment-induced changes in H3K4me3 peak density at the TSS. Fig. 8F shows that genes nearby TE-associated p53REs display less average H3K4me3 changes than nonTE-associated genes. Notable exceptions to this phenomenon include genes nearby L2 (66 genes) and MIR families (46 genes) of TEs which showed significantly higher levels of DXR-induced gene expression (Fig. 8F). Two known p53 genes (DDR1, p53RE in L2a; EPHA2, p53RE in MIRb) are among these groups. The p53REs that overlap L2 family repeats seemed unusual in that they show both strongly negative PhyloP values (Fig. 8C) and display changes in H3K4me3 at nearby TSS locations (Fig. 8F). We explored several of these L2-associated p53 binding sites in more detail and browser views of two strongly negative PhyloP p53REs that also show altered H3K4me3 at the TSS following p53 activation by DXR are shown in S9A–B Figs. (H3K4me3 data in Data File S2). In addition, four L2-associated p53REs with negative PhyloP values that only show DHS changes at the location of the binding site are given in S9C–F Fig.
We assessed genome-wide in vivo p53 occupancy, H3K4me3, DHS and gene expression in human lymphoblastoid cells treated with the DNA-damaging chemotherapeutic agent DXR. Previous studies of functional regulatory variation indicate that LCLs are responsive to a variety of treatments [28], including p53-activating, DNA-damaging agents like DXR [16], [31] and ionizing radiation [48]. On a genome-wide scale, we quantified the relationship between dynamic, stress-induced changes in gene expression and changes in H3K4me3 (Figs. 2A–B, 4A–B), a mark present on the nucleosomes flanking nucleosome-free regions that coincide with active TSSs [49], [50]. Notably, the group of genes with the greatest change in expression appear to be actively repressed at baseline (Figs. 2C, 4A–B), either via CTCF insulation, polycomb marks or the presence of heterochromatin. Strikingly, when we stratify the 1697 genes that display mRNA expression by the ChromHMM state at their TSS, we observe a linear trend between patterns of quantitative p53 occupancy (peak density) and fold change gene expression (Fig. 5B). Based on single-gene experimental models of p53 occupancy and transactivation, this relationship is not unexpected, but the clarity achieved by considering chromatin state on a genome-wide scale is dramatic (comparing S4C Fig. with Fig. 5B). Thus, chromatin state models created under steady-state conditions, such as the ENCODE ChromHMM, were highly predictive of genome-wide p53-mediated responses, and can help elucidate the mechanisms that enable those responses. This interesting finding should be examined in other inducible gene expression pathways and other cell types.
While greater than 50% of well-studied p53-regulated genes have binding sites proximal to a TSS, only about 25% of p53 binding was near a TSS in our study (Fig. 3B). Binding sites located in distal enhancer elements (∼34-kb from a TSS) comprised 26% of peaks, while another ∼30% were located in regions of heterochromatin even more distal to a TSS (median of 70 kb from a TSS). Approximately 80% of these distal sites in repressed chromatin were also identified in other genome-wide studies of p53 binding [6]–[9]; however, most of these sites remain functionally uncharacterized. Studies utilizing long distance chromatin capture techniques (HiC, 3C) under p53-activated conditions will be needed to confirm distal interactions between p53 binding and transactivation. Nevertheless, we did observe a clear relationship between p53 occupancy of p53REs located in repressed chromatin regions, and inducible transactivation of many genes within repressed regions (e.g., TGFA, ANK1, GDNF, CYP4F3).
In our study 91% (2664/2932) of the high confidence p53 occupied regions contained sequences similar to the motif created from 103 known p53 binding sites (S5A Fig.). We asked if p53 binding sequences located in different chromatin functional regions might display characteristic differences, and we observed that among the 592 peaks occurring in active promoters, the most prominent motif closely matched the p53 consensus (S5B Fig., p = 3.9×10−63). Interestingly, the next most common motif in peaks located in active promoters closely resembled the CTCF motif (p = 4.8×10−27) and this motif was present more commonly in down-regulated genes. However, these p53-related CTCF motifs were not associated with the CTCF insulator function as these regions displayed high nucleosome accessibility and the genes had high average expression levels. In contrast, genes classified as CTCF insulated by the ENCODE HMM model had low basal expression level but high inducibility (Fig. 5A–B). Among the most highly induced genes, we observed that prior to treatment, there was enrichment of repressive chromatin features such as CTCF, polycomb (PcG-H3K27me3) or heterochromatin (Fig. 2C). Genes with poised promoters carrying both H3K27me3 and H3K4me2 at the TSS (Figs. 4A–C, S2C) were an important group among these inducible genes and include a number of known p53 response genes (RRAD, TGFA, PERP, GPR87, RNF144B, TNFSF10A).
Espinosa and Gomes [13], [51] have hypothesized that loss of CTCF insulation activity could allow for inducible expression of some poised p53 genes. DXR induced changes in DHS that co-locate precisely with CTCF binding in untreated cells are consistent with this notion. Akdemir et. al. [52] proposed that changes in H3K27me3 status at poised genes via a mechanism of p53-JMJD3/UTX interaction could drive the conversion of poised genes to actively expressed genes. However, in the present experiment, the expression of these chromatin remodeling factors were not detectable. We did observe that changes in H3K4me3 were strongly confluent with altered DHS patterns and consistent with transcriptional activation of some poised p53 genes like SULF2 following treatment. The detailed mechanism drives the release of these specific genes from their poised state is an area of future study. Down-regulated genes had quite different characteristics; most down-regulated genes (85%) had open promoters (Fig. 6C) and displayed high expression at baseline (S2A Fig.). Interestingly, both the density of p53 binding, and the calculated p53RE PWM score associated with down-regulated genes were significantly lower compared with p53REs from induced genes (p<0.0001, Fig. 6C). Nikulenkov et. al. [8] observed a similar lower peak density among down-regulated genes. Wang et. al. [19] analyzed a small number of p53REs and suggested there was a preferred motif for down-regulated genes that is different from up-regulated genes. We also detected differences in the most enriched motif between the two groups of sequences. Specifically at nucleotides 10 and 11 in the p53RE motif, “TG” was present in the up-regulated REs and a “CA” motif occurred in the down-regulated genes (Fig. 6C). Consistent with this altered motif, the p53REs in down-regulated genes have a weaker match with the canonical sequence and lower mean PWM (Fig. 6C). However, this change is unlikely to be a general feature of p53REs because among all chromatin states many up-regulated genes contain the “CA” motif at nucleotides 10, 11. Pathway and functional analysis of each chromatin state by DAVID analysis showed that p53 is the most likely regulator of these genes and that the most significant GO terms include “regulation of cell death” and “response to DNA damage stimulus”.
The relative importance of DNA sequence content and chromatin landscape in determining a transcription factor's genomic binding has been subjected to considerable experimental exploration [28], [53]–[55], including several with p53 [14]–[16], [56]. Chromatin accessibility determines the binding of glucocorticoid hormone receptor [26] and STAT1 [57]. However, in the present study, we detected that a large fraction (29%) of p53 binding occurred at distal, repressed chromatin regions. The quantitative level of p53 occupancy was high in these regions of high intrinsic nucleosome occupancy, which is consistent with the findings of Nili et. al [14]. Thus, p53 occupancy and inducible gene expression occurred in regions of high nucleosome density indicated by existing H3K27me3 marks, heterochromatin, or low DHS. Notably, at the level of individual binding sites (e.g., upstream of TGFA), high levels of p53 occupancy can occur with minimal local displacement of nucleosomes (Figs. 4C, S3; as detected by DHS-seq). However, for poised genes marked with both H3K27me3 and H3K4me2 (Fig. 4A–C), p53 occupancy was associated with nucleosome accessibility changes at the TSS, but not necessarily at the p53 binding site. In these cases, as gene expression was induced, we observed the open DHS region spreading from the TSS, and the DHS pattern resembles the DXR-induced H3K4me3 pattern.
A particularly notable result is the relationship between quantitative occupancy and binding strength of the p53RE sequence relative to consensus. Modeling p53RE binding strength using a statistical approach (PWM), or with an experimental binding-based model (binding PWM, S6A Fig.), we observed genome-wide correlations between binding strength and occupancy (peak density). This relationship was further validated using p53 ChIP-seq data generated from a different laboratory using a different cell type and treatment (S7A–C Fig.). Thus, p53RE sequence content strongly affects occupancy which is in agreement with several in vitro studies that reported single nucleotide changes in the p53RE sequence or the spacer length alter p53 binding, transactivation of constructs or gene expression [16], [20], [41]. Recently a single nucleotide variation in a p53RE in the KITLG gene was demonstrated to alter p53 binding, transactivation of KITLG and be strongly associated with cancer [58]. However, our present study is the first genome-wide study to demonstrate the quantitative relationship between sequence content and genome-wide occupancy levels.
Surprisingly, p53RE PWM scores and occupancy were strongly negatively associated with chromatin accessibility and evolutionary conservation (see Figs. 7D, 8B, S7C) at the binding site. That is, more conserved, relatively weak p53REs were observed to be found in accessible chromatin regions. Conversely, many strong p53REs with low conservation scores were located in heterochromatic regions with very low DHS. The p53REs located within enhancers displayed relatively high levels of conservation, PWM score, occupancy and accessibility. The contrast in levels of conservation and chromatin accessibility among p53REs is driven by the presence of primate-specific repeat elements of viral transposon origin that are characterized by low conservation, high PWM, high occupancy but low accessibility. This high occupancy, low accessibility and minimal nucleosome displacement observation is consistent with the Cui et. al [22] hypothesis that p53REs are accessible on the external surface of nucleosomes. Cui et. al. described this for Alu-associated p53REs; however, our data suggests that it may be a general feature of p53-occupied REs in heterochromatin, particularly those in retroviral-associated TEs (e.g. MLT, LTR, MER). Over many millions of years of evolution transposable elements have rearranged primate genomes [46], and the presence of a multitude of p53 binding sites in various TEs [21]–[24] likely is related to an evolutionary mechanism for creating p53REs from TE sequences. Our present study uniquely reveals the interrelationships between TEs, p53RE sequence content, occupancy, and chromatin state.
There are several interesting and important questions concerning the functional nature of these TE-associated p53REs. Are these TEs actively suppressed (as a genome defense) and can they be reactivated in tumors or in normal cells following genomic stress like DNA damage? Leonova et. al. suggest that TEs are epigenetically suppressed and p53 is involved in transcriptional silencing of them [59]. We observed that TE-associated p53REs were significantly stronger REs and were more highly occupied than non-repeat p53REs; however, most reside in repressive, closed chromatin. In addition, this occupancy did not typically translate into induced H3K4me3 levels at nearby gene TSSs or changes in gene expression. This would be consistent with suppression of these TE-associated p53REs by p53 binding. Also, one may ask, what are the characteristics of p53REs located in TEs that would allow them to be co-opted (exaptation) and used to regulate the expression of nearby genes [60], [61]? It is unknown which features of chromatin or changes in chromatin structure might facilitate the utilization or exaptation of transposon-derived p53REs (e.g., domestication of transposons). While this question cannot be easily answered with the present data, it is intriguing that we observe some p53REs that overlap with TE-repeats (Alu and L2) display strongly negative PhyloP values suggesting the possibility of recent evolutionary selection [47]. In addition, if we consider that p53 occupancy of TE-associated p53REs is sometimes accompanied by altered DHS, H3K4me3 or gene expression at nearby genes (as shown in S9 Fig.), it seems likely that many of them are functionally utilized.
Thus it is possible that both suppression and exaptation have occurred with many strong, newly-evolved TE-associated p53REs being actively suppressed via chromatin state but a small number of them are exapted and brought into the p53 regulatory network for use under stress conditions. These observations imply a potential functional link between the variation in genomic p53RE sequence, chromatin state and gene regulation, which has co-evolved over time. As pointed out by Leonova et. al. [59], characterizing the p53-mediated deregulation of transposable elements in tumors could be useful in clinical cancer diagnoses. Understanding how all of these variables interact and impact the p53-regulated response to DNA damage in primary and tumor cells will be important both for developing strategies for prevention of cancer in healthy individuals and for understanding outcomes in cytotoxic therapy for cancer.
HapMap (http://hapmap.ncbi.nlm.nih.gov/) CEU human lymphoblastoid cell lines GM06993, GM11992, and GM12878 (Coriell Cell Repositories, Camden, NJ) were cultured in RPMI 1640 medium supplemented with 15% heat-inactivated fetal bovine serum (Life Technologies) and 1X antibiotics/antimycotics (Life Technologies). Cells were incubated at 37°C with 5% CO2. During experiments, cells were grown in petri dishes and treated with 0.3 µg/ml doxorubicin (Sigma, St. Louis, MO) for ChIP-seq, DNase I hypersensitivity, and gene expression.
GM12878 cells were cultured in triplicate and were either untreated or treated with 0.3 µg/ml doxorubicin for 4 and 18 hrs. Total RNAs was extracted using total RNA Miniprep kit (QIAGEN) with DNase 1 digestion. Total RNA was subjected to the standard cDNA synthesis, dye labeling and hybridization as per Affymetrix's protocols for the Human Exon ST 1.0 array (Affymetrix, Santa Clara, CA), and processed at the NIEHS Microarray Core. Exon expression data was analyzed through Affymetrix Expression Console using gene-level RMA summarization and sketch-quantile normalization methods. The t-test was used to calculate the significance of gene expression change. All the Affymetrix microarray data used here have been deposited into the GEO database (GSE51709). DXR induced gene expression array data was validated in a separate cell line using the Illumina gene expression array platform (Data File S1), as well as RT-PCR for selected genes. For replication by qRT-PCR, the cDNA was generated using the First-Strand Synthesis system (Life Technologies). For each biological replicate, target genes were amplified in triplicate using FAM probes designed to span exon junctions for each target gene and PCR Master Mix (Life Technologies). Studies investigating the relationship between gene expression in lymphoblast cell lines and in vitro p53 binding have been reported previously ([16], [31]).
Lymphoblastoid cells were seeded in 15-cm dishes and treated with 0.3 µg/mL doxorubicin or vehicle water for 18 hrs. Chromatin from 100×106 cells was used for each p53 ChIP-seq experiment. ChIP for p53 was carried out essentially as described in the Agilent Mammalian ChIP-on Chip protocol Version 20.0 (Agilent, Santa Clara, CA) using mouse monoclonal antibody DO-7 (BD Biosciences, San Jose, CA) or a non-specific rabbit IgG conjugated to secondary Dynal magnetic beads (Life Technologies). H3K4me3 ChIP was carried out with the H3K4me3 antibody (EMD Millipore #07-473, Billerica, MA). ChIP and input DNA were quantified using Qubit (Invitrogen). NIH Intramural Sequencing Center (NISC, Bethesda, MD) and the National Center for Genome Resources (NCGR, Santa Fe, NM) created the libraries using the standard Illumina ChIP-seq protocol. ChIP or input libraries were sequenced using Illumina Genome Analyzer IIx. For validation of occupancy, p53 ChIP-PCR was performed in triplicate using Sybr primers designed to span p53RE peaks in promoters of p53 target genes on a 7900 HT real-time PCR machine (Applied Biosystems).
Genome-wide mapping of DNase I-hypersensitive sites (DHS) was carried out on 2 replicates of untreated and DXR-treated GM12878 cells for 4 and 18 hrs. For DHS profiling, intact nuclei were treated with DNase I, and the DNase I-hypersensitive fraction was analyzed by sequencing as previously described [62]. Data from two biological replicates under each experimental condition are displayed as tracks in the UCSC genome browser in Figs. 4, S3, S9. For analysis of DHS relative to ENCODE genome wide chromatin state, we downloaded ENCODE/OpenChrom (Duke University, Durham, NC) GM12878 DHS data (Fig. 1, track K) [28], [33]. These data (DHS signal intensity) were averaged across p53 binding regions, transformed (log2 +1), grouped by PWM decile and plotted versus p53 peak density (Figs. 7D and S7C).
p53 ChIP-seq, H3K4me3 ChIP-seq, and input DNA sequence reads generated from the Illumina GAIIx were aligned against the human reference sequence (GRCh37p5, or hg19, June 2011) using the Burrows-Wheeler Alignment (BWA) Tool (Li et al., 2008). The uniquely mapped short reads were used to identify regions of the genome with significant enrichment in p53-associated DNA sequences. The peak detection was performed by QuEST 2.4 software [36] using the ‘Transcription factor binding site’ setting (bandwidth of 30 bp, region size of 300 bp) or the “Histone-type ChIP” setting (bandwidth of 100 bp, region size of 1000 bp) for H3K4me3 ChIP-seq, and the ‘stringent peak calling’ parameters (corresponding to 50-fold ChIP to input enrichment for seeding the regions and 3-fold ChIP enrichment for extending the regions).
We also used three published studies of p53 ChIP-seq from Nutlin-treated MCF7 cells [8], 5FU-treated MCF7 cells [8], 5FU-treated IMR90 cells [7], Nutlin and DXR-treated Human osteosarcoma U2OS cells [9], which deposited the raw reads at the NCBI SRA database. All sequence reads were aligned against the human reference sequence (GRCh37p5, or hg19, June 2011) using the Burrows-Wheeler Alignment (BWA) Tool (Li et al., 2008). For these published studies the peak detection was also performed by QuEST 2.4 software [36] using the ‘Transcription factor binding site’ setting (bandwidth of 30 bp, region size of 300 bp) for p53 ChIP-seq and the ‘stringent peak calling’ parameters. In Fig. 1 our H3K4me3 ChIP-seq tracks were prepared using the MACS program in order to display comparability with ENCODE tracks.
MEME Suite [40] was used for the de novo motif analysis of the genomic DNA sequences identified by ChIP-seq experiments. Briefly, repeat-masked DNA sequence for each peak was trimmed to 100 bp, centered to the maximal signal of original peak. The MEME algorithm was then applied to identify the top 10 most enriched motifs with a length from 6 to 25.
Position weight matrix (PWM) score was used to computationally estimate the binding strength to p53. The p53 PWM model was built on 103 published, experimentally validated p53RE sequences from the literature (NCBI PUBMED database), by converting nucleotide frequency values to a position weight matrix score as described [63]. Next the PWM score for any putative p53RE was calculated by summing the individual matrix values that correspond to the observed nucleotide at each position in that site. The potential p53REs in a genomic sequence were detected by sliding a window along the input sequence, considering the spacer in p53REs [64]. Briefly, at each chromosomal position, PWM score calculations were performed on DNA sequences with 20–23 nucleotides, corresponding to p53REs with a spacer of 0, 1, 2, and 3, respectively. For DNA sequences longer than 20, the first 10 nucleotides and the last 10 nucleotides were concatenated first to make a sequence of 20 nucleotides. For fast searching, a pattern composed of at least 3 of 4 core C and G nucleotides in the p53RE consensus was applied before PWM calculation. In addition, the same calculations were performed on the reversed complimentary sequences. In total, at each chromosomal position, eight calculations were performed. At a given location a putative p53RE was reported based on the following three criteria: (i) the sequence matches the search pattern; and (ii) the PWM score is above the PWM score threshold; and (iii) it contains the shortest spacer and has the largest PWM score. A second computational analysis based on measured binding in an in vitro binding assay, termed a binding PWM, was determined using a method described in Noureddine et. al. [31].
Datasets and sample information for histone modifications and CTCF binding for the GM12878 cell line were from the ENCODE project (http://genomes.ucsc.edu) via the laboratory of Bradley Bernstein (Broad Institute) [34], at the Massachusetts General Hospital/Harvard Medical School. The chromatin state segmentation was produced in Manolis Kellis's Computational Biology group at the Massachusetts Institute of Technology [33], [54]. We created analysis files based on 2932 p53 ChIP-seq peaks that had≥30 overlapping sequence reads (Data File S1) and 2415 genes associated with these peaks (many genes had multiple peaks, Data File S2). We merged ENCODE data (ChromHMM15 and ChromHMM7 Combined models, PhyloP, DHS) into our data analysis files that included ChIP-seq peak density (p53, H3K4me3), gene expression (intensity at each time point, fold change). Examples of these data are displayed in Fig. 1. In the various analyses we grouped p53 peaks (or genes) based on the H3K4me3 status of nearby genes, the ChromHMM7 Combined status at p53 peaks or at the TSS of nearby genes, or on the computationally calculated binding strength (PWM). ChromHMM7 Combined model failed to call states for some regions and the ChromHMM15 was used in these cases. ChromHMM15 states 14 and 15 (repetitive) were dropped from analysis as no p53 ChIPseq peaks were uniquely mapped to these regions. We also specifically looked at expression levels of genes with “poised” promoters carrying H3K4me1/2 marks and H3K27me3 marks (ChromHMM15 state 3). To visualize linear trends related to H3K4me3 or PWM scores we used decile analysis in which PWM values were ordered from low to high and grouped into deciles. We then evaluated other parameters relative to these linear trends. Regression, t-tests, and other statistics were calculated using Graphpad Prism (GraphPad Software).
To evaluate the evolution constraint on p53 binding sites, we utilized the placental mammal basewise conservation scores (PhyloP, phylogenetic p-values) downloaded from the UCSC genome browser website. These scores are computed from the PHAST package (http://compgen.bscb.cornell.edu/phast/) for multiple alignments of 45 vertebrate genomes to the human genome. The phyloP scores measure acceleration (faster evolution than expected under neutral drift) as well as conservation (slower than expected evolution). Sites predicted to be conserved are assigned positive scores, while sites predicted to be fast-evolving are assigned negative scores. The absolute values of the scores represent log p-values under a null hypothesis of neutral evolution. The sequence alignment and human-mouse conservation score were analyzed as previously described [45]. Transposon element repeats cataloged by Repeat Masker were accessed at UCSC genome browser website and aligned with p53REs to determine overlap.
Sequence reads for ChIP-seq experiments are deposited in the NCBI SRA database (p53 GSE46991, H3K4me3 GSE51713) and the NCBI Gene Expression Omnibus (GEO: http://www.ncbi.nlm.nih.gov/geo). Gene expression data are available at GEO under accession number GSE51709.
|
10.1371/journal.pgen.1002723 | EMT Inducers Catalyze Malignant Transformation of Mammary Epithelial Cells and Drive Tumorigenesis towards Claudin-Low Tumors in Transgenic Mice | The epithelial-mesenchymal transition (EMT) is an embryonic transdifferentiation process consisting of conversion of polarized epithelial cells to motile mesenchymal ones. EMT–inducing transcription factors are aberrantly expressed in multiple tumor types and are known to favor the metastatic dissemination process. Supporting oncogenic activity within primary lesions, the TWIST and ZEB proteins can prevent cells from undergoing oncogene-induced senescence and apoptosis by abolishing both p53- and RB-dependent pathways. Here we show that they also downregulate PP2A phosphatase activity and efficiently cooperate with an oncogenic version of H-RAS in malignant transformation of human mammary epithelial cells. Thus, by down-regulating crucial tumor suppressor functions, EMT inducers make cells particularly prone to malignant conversion. Importantly, by analyzing transformed cells generated in vitro and by characterizing novel transgenic mouse models, we further demonstrate that cooperation between an EMT inducer and an active form of RAS is sufficient to trigger transformation of mammary epithelial cells into malignant cells exhibiting all the characteristic features of claudin-low tumors, including low expression of tight and adherens junction genes, EMT traits, and stem cell–like characteristics. Claudin-low tumors are believed to be the most primitive breast malignancies, having arisen through transformation of an early epithelial precursor with inherent stemness properties and metaplastic features. Challenging this prevailing view, we propose that these aggressive tumors arise from cells committed to luminal differentiation, through a process driven by EMT inducers and combining malignant transformation and transdifferentiation.
| The epithelial-mesenchymal transition (EMT) is essential to germ layer formation and cell migration in the early vertebrate embryo. EMT is aberrantly reactivated under pathological conditions, including fibrotic disease and cancer progression. In the latter process, EMT is known to promote invasion and metastatic dissemination of tumor cells. EMT is orchestrated by a variety of embryonic transcription factors called EMT inducers. Among these, the TWIST and ZEB proteins are known to be frequently reactivated during tumor development. We here report in vitro and in vivo observations demonstrating that activation of these factors fosters cell transformation and primary tumor growth by alleviating key oncosuppressive mechanisms, thereby minimizing the number of events required for acquisition of malignant properties. In a model of breast cancer, cooperation between a single EMT inducer and a single mitogenic oncoprotein is sufficient to transform mammary epithelial cells into malignant cells and to drive the development of aggressive and undifferentiated tumors. Overall, these data underscore the oncogenic role of embryonic transcription factors in initiating the development of poor-prognosis neoplasms by promoting both cell transformation and dedifferentiation.
| While the disruption of embryonic processes has been acknowledged as a cause of the outgrowth of paediatric neoplasms, more recent observations suggest that the aberrant reactivation of developmental regulatory programs might also contribute to progression in the advanced stages of cancers in adults [1]. At the crux of this concept is the subversion of the epithelial-mesenchymal transition (EMT) during tumor progression. This developmental program converts epithelial cells into mesenchymal ones through profound disruption of cell-cell junctions, loss of apical-basolateral polarity and extensive reorganization of the actin cytoskeleton [2]. During embryogenesis, EMT plays critical roles in the formation of the body plan and in the differentiation of most of the tissues and organs derived from the mesoderm and the endoderm [3]. This process is tightly regulated through a delicate interplay between environmental signals from WNT, TGFβ, FGF family members, and a complex network of signaling pathways that converge on the activation of transcription factors that induce EMT through repression of CDH1 (encoding for the E-cadherin) and activation of mesenchymal genes. EMT-inducing transcription factors include several zinc finger proteins (e.g., SNAIL1, SNAIL2), basic helix-loop-helix transcription factors (e.g., TWIST1, TWIST2 and E2A) and zinc-finger and homeodomain proteins (ZEB1, ZEB2/SIP1) [4], [5]. Importantly, while EMT inducers are maintained in a silent state in adult differentiated epithelial cells, their reactivation is commonly observed in a variety of human cancers with a frequent correlation with poor clinical outcome [6]. In the course of tumor progression, the gain of cell motility and the secretion of matrix metalloproteases associated with EMT promote cancer cell migration across the basal membrane and invasion of the surrounding microenvironment, favoring metastatic dissemination. Furthermore, EMT may also facilitate second site colonization by endowing cells with stem-like features including self-renewing properties [7]–[9]. While the involvement of EMT inducers in the invasion-metastasis cascade of epithelial tumors is well delineated, their contribution to tumorigenesis remains unclear. Supporting an oncogenic activity within primary lesions, we recently demonstrated that the TWIST proteins were able to prevent cells from undergoing oncogene-induced senescence and apoptosis by abrogating both p53- and RB-dependent pathways [10], [11]. As a consequence, TWIST1 and TWIST2 can cooperate with an activated version of RAS to transform mouse embryonic fibroblasts [11]. Furthermore, the ZEB transcription factors were recently shown to overcome EGFR-induced senescence in oesophageal epithelial cells, suggesting that several EMT-inducers might share the property of inhibiting oncogene-induced failsafe programs [12]. On the basis of these findings, we sought to formally assess the oncogenic activity of these EMT-promoting factors in the model of breast tumorigenesis by generating Twist1 transgenic mouse models and by performing cooperation assays in human mammary epithelial cells (HMECs). The focus of this study was underpinned by the common reactivation of ZEB1, ZEB2 and TWIST1 in aggressive and undifferentiated human breast cancers, especially in the newly identified claudin-low intrinsic subtype [13]. Here we demonstrate that commitment to an EMT program favors breast tumor initiation by inhibiting crucial tumor suppressor functions, including PP2A (protein phosphatase 2A) activity, and thus minimizes the number of events required for neoplastic transformation. Importantly, upon aberrant activation of an EMT inducer, a single mitogenic activation is sufficient to transform mammary epithelial cells into malignant cells exhibiting all the characteristic features of claudin-low tumors. These findings extend our understanding of the role of EMT-inducing transcription factors during tumor development and highlight the claudin-low tumor subtype of breast cancers as the first example of human adult malignancies driven by aberrant reactivation of an embryonic transdifferentiation program.
To gain insight into the role of EMT commitment in tumor initiation and primary tumor growth, we used a Twist1 transgenic mouse model exhibiting a lox-STOP-lox (LSL) version of the active version of the murine TWIST1 (TWIST1-E12 tethered dimer) under a ubiquitous promoter [14]. These mice were crossed with a line expressing a lox-STOP-lox regulated knock-in of the activated K-Ras oncogene (LSL-K-rasG12D) [15], [16] for in vivo oncogenic cooperation experiments. Transgene expression was first induced using a Mouse Mammary Tumor Virus promoter driven Cre recombinase (MMTV-Cre) and thereby restricted to secretory tissues, in particular the mammary gland and skin epithelia, as well as to the hematopoietic system [17], [18]. Neither wild-type nor MMTV-Cre;Twist1 mice exhibited tumor formation by one year of age (n = 47). Expression of knock-in K-rasG12D was associated with low-grade splenic lymphomas as well as anal and oral papillomas (Figure 1). Importantly, papillomas never progressed to a malignant stage but could grow to the point to physical obstruction leading to cachexia and requiring euthanasia (n = 85, median survival 85 days). In contrast, MMTV-Cre;K-rasG12D;Twist1 mice invariably developed aggressive multifocal squamous cell carcinomas (SCC) at very young ages necessitating euthanasia at the significantly earlier median age of 35 days (n = 12, p<0.0001, Figure 1). These observations demonstrated for the first time the oncogenic properties of TWIST1 in vivo and underscored the cooperative effect between K-RAS and TWIST1 in promoting malignant conversion.
Due to the speed with which SCC developed in the MMTV-Cre;K-rasG12D;Twist1 animals, the role of TWIST1 in promoting mammary tumor formation could not be assessed. Consequently, transgene expression was next restricted to differentiated mammary epithelial cells by using mice expressing the Cre recombinase under the control of the Whey Acidic Protein promoter (WAP-Cre) [17]. WAP is a milk protein expressed late in the differentiation pathway of mammary epithelial cells [19]. The 2.6-kb fragment of the mouse WAP gene promoter used in the present study is active in the mammary alveolar epithelium during the second half of pregnancy upon the initiation of differentiation [17], [20]. In virgin animals, the promoter is only transiently activated in a few mammary alveolar and ductal cells, during estrus (average age at first estrus = 35 days), allowing mosaic transgene activation so as to better mimic the emergence of spontaneous oncogenic activations [19]. Wild-type, WAP-Cre;K-rasG12D, and WAP-Cre;Twist1 transgenic females exhibited normal mammary gland development (Figure 2B) and remained healthy for at least 240 days (n = 19). However, all virgin WAP-Cre;K-rasG12D;Twist1 females developed multifocal breast carcinomas by 140 days of age, approximately 105 days after first transgene expression (p<0.001, median survival = 125 days, n = 9) (Figure 2A and 2D). These tumors exhibited metaplastic features with a mixed morphologic aspect that included epithelial-type and spindle shaped cells (Figure 2D). In support of the relevance of EMT in vivo, the presence of the tagged-TWIST1 transgene in both the epithelial and fusiform cancer cell contingents demonstrated that mesenchymal cells arose through transdifferentiation of their epithelial counterparts (Figure 2D). Molecular characterization of human and murine breast tumors led to identifying five intrinsic subtypes (luminal A, luminal B, HER2-enriched, basal-like and claudin-low) [13], [21], [22]. Global gene expression profile analysis (Accession number GSE32905) classified tumors developed by WAP-Cre;K-rasG12D;Twist1 transgenic mice as claudin-low (Figure S1). Immunostaining of epithelial and mesenchymal markers (Figure 2D) and quantitative RT-PCR analysis (lack of E-cadherin and claudin expression, high expression of vimentin; Figure S2) were highly consistent with this classification. Of note, endogenous expression of the Zeb1, Zeb2 and Twist2 EMT inducers was also induced (Figure S2), further supporting the association of the EMT interactome with the claudin-low breast cancer subtype [23].
Both basal-like and claudin-low subgroups are aggressive, chemoresistant, triple-negative carcinomas (estrogen-receptor-, progesterone-receptor-, and HER2-negative). Yet claudin-low tumors exhibit several characteristic features, including low expression of adherens and tight junction proteins, a low level of luminal/epithelial differentiation, stem cell-like features, and a high frequency of metaplastic differentiation [24]. These tumors are believed to originate from an early epithelial precursor with inherent stemness properties and metaplastic features [24]–[26]. Nevertheless, the observation of claudin-low tumors in WAP-Cre;K-RasG12D;Twist1 transgenic mice suggested that the development of these neoplasms could rely upon an EMT-driven process affecting epithelial cells formerly engaged in differentiation. We sought to test this hypothesis by mimicking in non-stem cells events occurring commonly in this breast cancer subtype. Claudin-low tumors and cell lines frequently exhibit increased levels of the EMT-inducing transcription factors TWIST1, ZEB1, and ZEB2 [13] and show activation of RAS/MAPK pathway components [24], [27]. To functionally reproduce these two common features of claudin-low tumors, we performed oncogenic cooperation assays by transducing genes encoding a single EMT inducer (TWIST1, ZEB1 or ZEB2) and/or an active form of H-RAS (H-RASG12V) into immortalized human mammary epithelial cells (hTERT-HMECs, thereafter named HME cells). As in all experiments the infection efficiency exceeded 80%, the hypothesis of selection of a rare subpopulation of parental cells can be ruled out. Forced expression of an EMT inducer triggered acquisition of EMT features, including significant upregulation of mesenchymal markers (i.e., vimentin, fibronectin) and decreased expression of genes involved in epithelial cell-cell adhesion (i.e., E-cadherin, occludin) (Figure 3). However, the degree of EMT commitment observed after infection was highly dependent on the EMT-inducing transcription factor, both in the absence or in the presence of the active form of RAS. In the absence of the mitogenic oncoprotein, ZEB1 expression was sufficient to promote a complete transdifferentiation process, giving rise to typical spindle-like cell morphology, and a total loss of E-cadherin expression (Figure 3; HME-ZEB1 cells). In contrast, ZEB2-expressing cells (HME-ZEB2 cells) and TWIST1-expressing cells (HME-TWIST1 cells) exhibited intermediate phenotypes, maintaining significant levels of E-cadherin expression and retaining a cobblestone morphology despite increased levels of mesenchymal markers such as vimentin and fibronectin (Figure 3). Transduction of an active form of RAS further promoted EMT induction, as assessed by cell morphology (Figure 3A) and protein expression analysis (Figure 3F and 3G). Nevertheless, HME-ZEB2-RAS and HME-TWIST1-RAS cells still exhibited an intermediate phenotype. Importantly, combining the mitogenic oncoprotein with either TWIST1, ZEB1 or ZEB2 was sufficient to provide cells with transformation potential, as assessed by their ability to form colonies in an assay on semi-solid medium and by acquisition of a characteristic stellate phenotype in 3D cell culture (Figure 3C and 3D). This observation suggested that EMT inducers could exert a potent oncogenic activity in the absence of a complete EMT. Global gene expression array analysis was next performed on freshly established cell lines (Accession number GSE32905). Strikingly, HME-ZEB1 cells and HME-ZEB1-RAS cells were defined as basal-B (P = 2.7×10−20 and 2.4×10−22 respectively), reminiscent of claudin-low tumors [13], [21], HME-TWIST1, HME-TWIST1-RAS and HME-ZEB2 as basal-A (P = 3.6×10−8, P = 1.1×10−2, P = 3.8×10−9 respectively), reminiscent of basal-like tumors [13], [21], while HME-ZEB2-RAS positively correlated with basal-A/basal-like (P = 5.0×10−2) and basal-B/claudin-low (P = 2.3×10−2) (Figure 4 and Figure S3), suggesting a direct link between the extent of EMT and the intrinsic subtype. Confirming this hypothesis, exposure of HME-TWIST1-RAS and HME-ZEB2-RAS to the EMT-promoting cytokine TGFβ triggered a complete EMT with a shift to basal-B/claudin-low (P = 2.3×10−18 for TGFβ-treated HME-TWIST1-RAS cells; P = 6.8×10−22 for TGFβ-treated HME-ZEB2-RAS cells) (Figure 4 and Figure S3).
The extent of the EMT and the basal-B/claudin-low profiling were strongly associated with acquisition of stem cell-like features, as judged by the ability to form mammospheres under non-adherent culture conditions (Figure 3E) and by the fraction of cells exhibiting the CD44+/CD24−/low stem-like antigenic phenotype (respectively 84.5%, 83.2%, 20.6% and 15.3% of HME-ZEB1; HME-ZEB1-RAS, HME-ZEB2-RAS and HME-TWIST1-RAS; Figure 5). The gain of a mammary stem cell signature was also revealed by the use of the recently described Genomic Differentiation Predictor [13], following global gene expression array analysis on freshly established cell lines (Figure 6). Cells exhibiting the more pronounced mesenchymal phenotype (HME-ZEB1, HME-ZEB1-RAS; HME-ZEB2-RAS treated with TGFβ and HME-TWIST1-RAS treated with TGFβ) exhibited a mammary stem like signature, whereas cells with an epithelial or an intermediate phenotype (HME, HME-TWIST1, HME-TWIST1-RAS; HME-ZEB2; HME-ZEB2-RAS) showed a luminal progenitor signature.
As expected from earlier studies [28]–[30], immortalized HMEC cells into which only H-RASG12V had been transduced (HME-RAS) exhibited a low transformation potential (Figure 3D). Characterization of the few colonies growing on soft agar revealed a constant endogenous activation of EMT inducers, including TWIST1, ZEB1 and ZEB2 (Figure S4), and a mesenchymal phenotype, further highlighting the deleterious interplay between the mitogenic oncoprotein and EMT-promoting factors. To confirm the association between the RAS-induced transformation and the endogenous expression of EMT inducers, immortalized HMECs were transduced with an H-RASG12V and sorted by flow cytometry using the EpCAM epithelial antigen. EpCAM-positive epithelial cells were next cultured in the presence of TGFβ. As shown in Figure S5, TGFβ exposure triggered morphological and phenotypical features of EMT, associated with increased expression of TWIST1, TWIST2, ZEB1 and ZEB2 EMT inducers. Reactivation of these transcription factors was associated with the acquisition of a transformed phenotype.
Taken together, these observations showed that activation of EMT inducers, through either forced expression or endogenous induction, fosters malignant transformation of mammary epithelial cells and confers to them basal-like or claudin-low signatures, according to the extent of transdifferentiation.
Our data demonstrated that EMT inducers can promote transformation of mammary epithelial cells. We and others have previously shown that the TWIST and ZEB proteins can functionally inhibit p53- and RB-dependent pathways, preventing cells from undergoing oncogene-induced senescence and apoptosis [10]–[12]. To test whether the oncogenic properties of EMT-inducing transcription factors act only to lift these two oncosuppressive barriers or whether they might be involved in additional processes, we have generated human mammary epithelial cells deficient in both pathways. The INK4A tumor suppressor, a crucial regulator of the RB-dependent pathway, is known to be silenced by progressive promoter methylation in HMECs escaping from stasis [31]. We depleted these cells of p53 by means of RNA interference (using a shRNA TP53 thereafter named shp53 or, as a control, a scrambled shRNA). Knockdown of p53 was checked by western blotting and by demonstrating that, in response to DNA damage, p53 induction and the resulting G1 growth arrest were abolished (Figure S6). Cells were next infected with H-RASG12V and immortalized by transfection with hTERT to generate shp53/H-RASG12V/hTERT HMECs (hereafter named HME-shp53-RAS cells). Characterization of the colonies generated after growth of these cells in soft agar demonstrated that a vast majority of them expressed mesenchymal markers (Figure S7E). This observation led us to hypothesize that a subset of HME-shp53-RAS cells committed spontaneously to an EMT program and that initiation of the transdifferentiation process promoted cell transformation. In support of the first hypothesis, cells exhibiting a cobblestone phenotype and expressing epithelial markers (E-cadherin+ and EpCAM+) and cells displaying a fibroblastic morphology and exhibiting mesenchymal markers (vimentin+) were found to coexist in HME-shp53-RAS cells (Figure S7C). Epithelial and mesenchymal cell subpopulations were next sorted on the basis of their differential antigenic phenotypes (EpCAM+ and EpCAM− respectively) (Figure S8). The phenotypes of the epithelial and mesenchymal cell populations were confirmed by assessing the expression of additional epithelial markers (β-catenin, E-cadherin, ZO-1, and occludin) and mesenchymal markers (fibronectin and vimentin) by immunofluorescence staining and western blotting (Figure S9). The sorted mesenchymal-cell subpopulations specifically displayed EMT-associated features such as motility, invasiveness, and a stellate phenotype when cultured in 3D (Figure 7). Gene profile analysis classified these mesenchymal cells as claudin-low/basal B, while epithelial HME-shp53-RAS cells were classified as basal-like/basal A (Figure 4 and Figure S3). Although epithelial and mesenchymal HMEC derivatives exhibited similar H-RASG12V expression levels (Figure S10), only mesenchymal cells grew in soft agar and gave rise to tumor formation, within three months, when homotopically xenografted in nude mice (6 of 7 mice, Figure 7).
These observations demonstrated that EMT commitment fosters malignant transformation of human mammary epithelial cells deprived of functional p53- and RB-dependent pathways. To further confirm this hypothesis sorted epithelial HME-shp53-RAS cells were treated with TGFβ. Exposure to this EMT-inducing cytokine triggered a pronounced shift from epithelial to mesenchymal markers, associated with induction of ZEB1 (200-fold) and ZEB2 (10-fold) (data not shown) and with a dramatic gain in anchorage-independent growth properties (Figure S11). Importantly, forced expression of ZEB1 in sorted epithelial HME-shp53-RAS cells was sufficient to mimic TGFβ exposure, promoting both EMT and cell transformation (Figure S12).
Our observations strongly suggested that, beyond the inhibition of the p53- and RB-dependent pathways, EMT inducers display additional, as yet unidentified oncogenic activities. It has been previously shown that, in vitro, transformation of normal human epithelial cells, including mammary epithelial cells, requires disruption of the telomere maintenance system and dysregulation of at least four key signaling pathways: activation of the RAS-dependent pathway and inhibition of the p53-, RB-, and protein phosphatase 2A-dependent pathways [28]–[30]. We thus endeavored to analyze the effects of EMT commitment on protein phosphatase 2A (PP2A) activity. PP2A is a ubiquitously expressed serine/threonine phosphatase accounting, with protein phosphatase 1 (PP1), for 90% of all the serine/threonine phosphatase activity in the cell [32]. By using a peptide substrate (synthetic phosphothreonine peptide RRA(pT)VA) compatible with the phosphatase activity of PP2A but not with that of PP1 and by employing experimental conditions ensuring the specificity of PP2A activity (see Materials and Methods), we found sorted mesenchymal HME-shp53-RAS cells to exhibit lower phosphatase activity than sorted epithelial HME-shp53-RAS cells (Figure 8). More importantly, expression of either TWIST1, ZEB1, or ZEB2 in HME cells was sufficient to trigger significant (2-fold) downregulation of serine/threonine phosphatase activity, revealing a novel oncogenic feature of these proteins. This downregulation was similar to that observed in immortalized HMECs transformed with H-RASV12 and the SV40 large T and small t antigens (HMLER cells; Figure 8), the small t antigen being known to inhibit PP2A activity [33]. Notably, claudin-low HME-ZEB1-RAS cells exhibited 4-fold lower phosphatase activity than immortalized HMECs (Figure 8).
Whereas TWIST1 has been convincingly implicated in the metastatic dissemination of breast cancer cells, these data underscore the importance of EMT-inducing transcription factors in driving mammary carcinogenesis, with a dual role in cell transformation and dedifferentiation. Tumor development has been portrayed as a multistep processes with a progressive acquisition of genetic and epigenetic abnormalities providing cells with biological capabilities such as sustained proliferation, replicative immortality, survival advantages, angiogenesis and, in some cases, invasive growth and metastasis [34]. According to the Darwinian model of cancer development, each of these acquired traits confers a distinct selective advantage, originating successive waves of clonal expansion that drive tumor progression. It is well known that this complex and time consuming process requires abrogation of several oncosuppressive barriers. In epithelial cells, including mammary epithelial cells, these barriers comprise the p53-, RB- and PP2A-dependent pathways [28]–[30]. We and others have previously demonstrated that TWIST and ZEB transcriptions factors were capable to inhibit p53- and RB- dependent pathways [10]–[12]. Remarkably, our observations reveal that activation of these factors in HMECs also affects PP2A phosphatase activity. Considerable evidence highlights the tumor-suppressor functions of this serine/threonine phosphatase. For example, it has been shown in vitro that the transforming ability of the SV40 small t antigen requires interactions with PP2A and downregulation of its activity [34]–[36]. In vivo, mutations affecting different components of the PP2A holoenzyme complex have been identified in a variety of human malignancies and, in mouse models, mutation of PP2A favors tumorigenesis [37]. Loss of PP2A during cell transformation triggers multiple events, such as upregulation of kinases involved in mitogenic and survival signaling (e.g. AKT and MAPK), stabilization of protooncogenes (e.g. MYC), destabilization of tumor suppressors (e.g. p53 and RB), and loss of proapoptotic signaling pathways (e.g. BAD) [32]. Modulation of downstream components of the RAS signaling pathway by PP2A might be of particular significance in our model, as the ability of PP2A to antagonize the oncogenic properties of RAS by dephosphorylating crucial downstream effectors such as c-MYC and AKT makes its downregulation a prerequisite to RAS-induced malignant transformation. The modulation of PP2A activity by EMT inducers might thus be an important mechanism underlying the deleterious cooperation of these factors with oncogenic RAS in cell transformation. The inhibition of PP2A activity by EMT inducers might also be relevant during embryogenesis, as PP2A appears as a negative regulator of the WNT signaling cascade [38] which is required for several crucial steps in early development. Further studies are needed, however, to better characterize the mechanisms involved in this regulatory process, as PP2A represents a complex family of holoenzyme complexes known to display different activities and to exhibit diverse substrate specificities [39].
Given the importance of p53-, RB- and PP2A-dependent protective barriers against tumorigenesis and their role in regulating cell differentiation and self-renewal [40], aberrant reactivation of EMT inducers might profoundly affect the multistep nature of tumorigenesis by increasing cell plasticity and leapfrogging the mutation bottleneck toward tumor progression. This view is supported by our in vitro transformation assays demonstrating that, upon a single mitogenic activation, forced expression of either TWIST1 or ZEB1/2 is sufficient to trigger malignant conversion of immortalized human mammary epithelial cells (hTERT-HMECs). It is also consistent with the observed rapid and repeated appearance of multifocal breast carcinomas in WAP-Cre;K-rasG12D;Twist1 mice. It is further supported by the work by Phuoc T. Tran and colleagues, who demonstrated in an elegant inducible transgenic mouse model that TWIST1 overexpression accelerates K-RAS-induced lung tumorigenesis [41]. Several phenomena associated with tumor initiation, such as inflammation [42], physical constrains (including hydrostatic pressure, shear stress and tension forces) [43], abnormal activation of signaling pathways such as those controlled by WNT, NOTCH, or TGFβ [4], [44], and hyperactivation or RAS-ERK1/2 signaling [45] are known to trigger expression of EMT-promoting factors and could thus induce reactivation of these embryonic transcription factors in early stages of tumor development, as previously observed in animal models [46]. Moreover, beyond the deleterious consequences of aberrant reactivation of EMT inducers in differentiated or committed epithelial cells, the ability of EMT inducers to inhibit key oncosuppressive pathways also implies that embryonic or adult stem cells that normally express these factors are particularly vulnerable to cell transformation.
Cooperation assays demonstrate that activation of EMT-inducing transcription factors such as TWIST1 or ZEB2 is sufficient to make cells highly prone to transformation, even in the absence of a complete mesenchymal morphological shift. In line with this view, immunohistochemical analysis of TWIST1 in human non-invasive breast cancers (ductal carcinomas in situ, DCIS) has revealed frequent overexpression of this EMT inducer within the bulk of the primary lesion, while the cancer cells maintain an epithelial phenotype (Figure S13). EMT is known to be a highly dynamic process giving rise to a series of important changes in cell phenotype, including loss of cell polarity, loss of cell-cell adhesion structures, remodeling of the cytoskeleton, and promotion of cell motility. As recently highlighted by Klymkowsky and Savagner, although the term EMT is generally applied as if it were a single conserved process, EMT-related processes can in fact vary in degree from a transient loss of cell polarity to total reprogramming of the cell [47]. The existence of malignant cells with combined epithelial and mesenchymal characteristics has previously been demonstrated in vivo, in both mouse models of EMT and human tumors [48], [49]. Especially, epithelial cells coexpressing cytokeratins 5/19 and vimentin have been identified by dual immunofluorescence labeling in claudin-low and basal-like breast cancers, two breast cancer subtypes frequently exhibiting overexpression of EMT-inducing transcription factors [13]. Overall, these observations strongly suggest that EMT-promoting factors can exert oncogenic functions in cells retaining an epithelial phenotype, in the total absence of morphological features of EMT, and probably long before initiation of the invasion-metastasis cascade.
Previous in vitro studies using human mammary epithelial cells have revealed a link between EMT, malignant transformation, and acquisition of stem cell properties. For example, the transformation of HMECs by means of a combination of hTERT, SV40 large T and small t antigens, and H-RASG12V (HMLER cells) is associated with both mesenchymal and stem-like features [7], [8], [50]. In the absence of oncogenic RAS, introduction of SV40 T and small t antigens and hTERT into mammosphere-derived HMECs also generates malignant cells exhibiting EMT and stem-like properties [51]. Recent reports further demonstrate in human cancer cell lines that spontaneous EMT or TGFβ/TNFα-mediated EMT generates cells with a claudin-low phenotype [52], [53]. Yet the intrinsic role of EMT inducers was not addressed in these studies. We highlight herein a dual role of these factors in cell transformation and dedifferentiation. Remarkably, in the context of a very few genetic events, the aberrant activation of an EMT inducer can initiate mammary epithelial cell transformation in vitro and in vivo and can drive the growth of undifferentiated tumors exhibiting all the characteristic features of claudin-low tumors, including a malignant phenotype, low expression of tight and adherens junction genes, EMT traits, and stem-cell-like characteristics.
The origin of the different intrinsic subtypes of human breast cancer is a topic of contentious debate and remains ill defined. Recent in vitro observations support the view that both luminal and basal-like breast cancers derive from a common luminal progenitor cell, whereas claudin-low tumors, viewed as the most primitive malignancies, originate from a stem/progenitor cell with inherent stemness properties and metaplastic features [24]–[26]. Others suggest that basal-like and claudin-low tumors arise from transformation of a similar stem cell, but that the claudin-low tumors stay arrested in an undifferentiated state, while basal-like cancer cells divide asymmetrically and give off differentiated progeny arresting at the luminal progenitor state [54]. Our observations pave the way for an alternative model highlighting a dynamic process orchestrated by the activity of EMT-inducing transcription factors. According to this model, aberrant activation of EMT inducers in committed cells (e.g. luminal progenitors) might foster initiation of triple-negative breast tumors and confer basal-like or claudin-low signatures, according to the extent of transdifferentiation. Our model also implies that basal-like tumors might progressively evolve towards a claudin-low phenotype through completion of the EMT process. This view is supported by the histopathology of human metaplastic breast tumors: phenotypically, the acquisition of mesenchymal features can occur at various stages of the disease [55], highlighting the dynamic role of transdifferentiation during tumor development and pointing to the interaction between cancer cells and the microenvironment as a key determinant of tumor phenotype and behavior. It is noteworthy that a model of murine claudin-low tumors has recently been described, involving transplantation of p53-null mammary tissues into the cleared fat pads of wild-type recipients [56]. This observation is consistent with the role of p53 loss in EMT induction [57]–[59] and with the spontaneous generation of mesenchymal cells exhibiting a claudin-low phenotype in HME-shp53-RAS cells. Yet in this model described by Herschkowitz and colleagues, p53-null mouse mammary tumors fell into a variety of molecular groups, also including luminal and basal-like subtypes [56]. WAP-Cre;K-rasG12D;Twist1 mice thus appear as the first mouse model consistently generating claudin-low tumors. These transgenic mice might thus serve as valuable preclinical models for testing both potential therapeutic agents targeting these aggressive neoplasms and potential preventive agents.
The TWIST1-E12 tethered heterodimer was generated by PCR by fusing the human TWIST1 and E12 proteins using a G3-S2-G2-S-G3-S-G3-S2-G2-S-G3-S-G polylinker as described in [60]. The full-length murine HA-tagged ZEB1 and ZEB2 cDNAs were cloned into the pBabe retroviral construct. The TP53 shRNA (shp53) pRETRO SUPER expression construct has been described in [61].
Animal maintenance and experiments were performed in a specific pathogen free animal facility “AniCan” at the CRCL, Lyon, France in accordance with the animal care guidelines of the European Union and French laws and were validated by the local Animal Ethic Evaluation Committee. The heterozygous knock-in LSL-K-rasG12D mouse strain [16] was crossed with CAG-LSL-(Myc)-Twist1 mice (FVB background) [14]. Both TWIST1 monomer and T1-E12 dimers were used producing similar results [60]. K-rasG12D;Twist1 offspring were subsequently crossed with mice carrying the Cre recombinase under the control of the Mouse Mammary Tumor Virus or Whey Acidic Protein promoters (MMTV-Cre (B6129F1 background) or WAP-Cre (c57BL/6 background) [17]; purchased from the NCI-MMHCC. Cre (wild type), Cre;K-rasG12D, Cre;Twist1, and Cre;K-rasG12D;Twist1 virgin animals were maintained and monitored at least weekly for tumor incidence. End points were determined based on tumor diameter (>17 mm) or the sick appearance of an animal. Tissues were harvested and either snap frozen in N2(l) or immersed in formalin until pathological analysis. Genotyping of genomic DNA from tails purified using the NucleoSpin Tissue kit (Macherey-Nagel) was performed with primers described in references [14], [16], [17] using REDTaq 2× ReadyMix (Sigma).
Primary human mammary epithelial cells (HMECs) were provided by Lonza. HMEC-derivatives were cultured in 1∶1 Dulbecco's Modified Eagle's Medium (DMEM)/HAMF12 medium (Invitrogen) complemented with 10% FBS (Cambrex), 100 U/ml penicillin-streptomycin (Invitrogen), 2 mM L glutamine (Invitrogen), 10 ng/ml human epidermal growth factor (EGF) (PromoCell), 0.5 µg/ml hydrocortisone (Sigma) and 10 µg/ml insulin (Actrapid).
Three-dimensional cultures consisted in culturing 5×103 cells/well in 2% growth factor reduced Matrigel (BD Biosciences) on top of a 100% matrigel layer. 20 days after seeding, cells were fixed in 3% paraformaldehyde (Sigma), permeabilized in 0.5% Triton 100X (Sigma) in PBS buffer for 10 min. After several washes in PBS, cells were labeled with 1 µg/ml of TRITC-conjugated Phalloïdin P1951 (Sigma) for 45 min. Following washes in PBS, nuclei were stained with Hoechst 5 µg/ml for 10 min and mounted with Fluoromount-G (SouthernBiotech).
For mammosphere formation, after filtration through a 30 µm pore filter, single-cells were plated at a density of 105 cells/ml in Corning 3261 ultra-low attachment culture dishes. Primary cell spheres were enzymatically dissociated with 0.05% trypsin for 15 min at 37°C to obtain single-cell suspension. The ability to generate mammospheres was defined after three consecutive passages.
Treatment with TGFβ was performed with 2.5 ng/ml of the recombinant cytokine (Peprotech) for a three week period.
Cell distribution was performed using the FITC-EpCAM VU-1D9 (Stem Cell), the FITC-CD44 G44-26 (BD Pharmingen) and the PE-CD24 ML5 (BD Pharmingen) monoclonal antibodies, the FACScan Calibur (Becton Dickinson) and analyzed using the FlowJo software.
Matrigel (BD Biosciences) was added to the wells of an eight-well Labtek chamber in a volume of 300 µl/well. A Matrigel plug of about 1 mm diameter was removed. The hole was successively filled with 105 cells and 100 µl of Matrigel. Appropriate growth medium was added on top. Cultures were analyzed after 24 h (Figure 7) or 72 h (Figure S11). Areas of migration were visualized using an Olympus IX50 (NA 0.075). Samples were performed in duplicate.
5×104 cells were placed in the upper chamber of an 8 µM Transwells (BD Biosciences). 24 h later, chambers were washed twice with PBS. The filter side of the upper chamber was cleaned with a cotton swab. The membrane was next cut out of the insert. Cells were fixed in methanol and stained with 5% Giemsa 30 min at room temperature.
2×106 Phoenix cells were transfected by calcium-phosphate precipitation with 10 µg of retroviral expression vectors. 48 h post-transfection, the supernatant was collected, filtered, supplemented with 5 µg/ml of polybrene (Sigma) and combined with 106 targeted cells for 6 h. Cells were infected twice and selected 48 h post-second infection with puromycin (0.5 µg/ml), neomycin (100 µg/ml) or hygromycin (25 µg/ml).
To measure anchorage-independent growth, cells were detached with trypsin and resuspended in growth medium. Plates were prepared with a coating of 0.75% low-melting agarose (Lonza) in growth medium and then overlaid with a suspension of cells in 0.45% low-melting agarose (5×104 cells/well). Plates were incubated for 3 weeks at 37°C and colonies were counted under microscope. Experiments were performed in triplicate.
Eight-week old female athymic Swiss nude mice (C. River laboratories) were X-irradiated (4 Gy) prior to injection. Single cell suspensions, (5×106 HMEC derivatives resuspended in a PBS-Matrigel (1/1) mixture) were injected into the fat pad of a mammary gland. Tumor incidence was monitored up to 90 days post-injection. Animals were allowed to form tumors up to 1.5 cm in diameter, at which point animals were euthanized. Each tumor was dissected, fixed in paraformaldehyde and processed for histopathology examination.
Cells were washed twice with phosphate buffered saline (PBS) containing CaCl2 and then lysed in a 100 mM NaCl, 1% NP40, 0.1% SDS, 50 mM Tris pH 8 RIPA buffer supplemented with a complete protease inhibitor cocktail (Roche). Protein expression was examined by western blot using anti-E-cadherin clone 36 (Becton Dickinson), anti-β-catenin clone 14 (Becton Dickinson), anti-fibronectin clone 10 (Becton Dickinson), anti-vimentin clone V9 (Dako), anti-N-cadherin clone 32 (Becton Dickinson), anti-occludin clone OC-3F10 (Zymed Laboratories), anti-β-actin clone AC-15 (Sigma), anti-HA clone 11 (BabCO), anti-TWIST Twist2C1a (Abcam) monoclonal antibodies and a rabbit polyclonal anti-H-RAS clone C20 (Santa Cruz) for primary detection. Horseradish peroxidase-conjugated rabbit anti-mouse and goat anti-rabbit polyclonal antibodies (Dako) were used as secondary antibodies. Western blots were revealed using an ECL detection kit (Amersham) or a western-blotting Luminol reagent (Santa Cruz).
104 cells were seeded on 8-well Lab-TekII chamber slide, fixed in 3% paraformaldehyde (Sigma) and permeabilized in 0.1% Triton 100X (Sigma) in PBS buffer at room temperature for 10 min. The cells were then washed 3 times with PBS and incubated with a blocking solution (10% horse serum in PBS). The cells were then incubated with the anti-E-cadherin clone 36 (Becton Dickinson) or the anti-vimentin clone V9 (Dako) primary antibodies overnight at 4°C. Phalloidin labeling was performed by incubating cells with 1 µg/ml of TRITC-conjugated Phalloïdin P1951 (Sigma) for 30 min. Following extensive washes in PBS, nuclei were stained with Hoechst 5 µg/ml for 10 min and mounted with Fluoromount-G (SouthernBiotech). All matched samples were photographed (control and test) using an immunofluorescence microscope (Leica) and identical exposure times.
The immunohistochemical study was performed on three microns deparaffinized sections, using the avidin-biotin-peroxidase complex technique (LSAB universal, Dako), after 15 min heat-induced antigen retrieval in 10 mM citrate buffer, pH 6. The primary anti-E-cadherin clone 36 diluted at 1/500 (Becton Dickinson), anti-vimentin clone V9 diluted at 1/200 (Dako) and anti-c-MYC A14 at 1/100 (Santa-Cruz Biotechnology) antibodies were applied 60 min at room temperature.
Paraffin embedded tumors were serially sectioned at a thickness of 4 µm. After deparaffinisation and rehydration, endogenous peroxidases were blocked by incubating the slides in 5% hydrogen peroxide in sterile water. For heat induced antigen retrieval, tissue sections were boiled at 97°C for 40 min either in a 10 mM citrate buffer pH 6 (when anti-cytokeratin and anti-vimentin antibodies were used) or in buffer pH 7 (Dako) (for the anti-E-cadherin antibody) clone 36 (BD Biosciences). Slides were then incubated with the monoclonal pancytokeratin clone AE1/AE3, (Dako), the polyclonal anti-vimentin SC7557 (Santa Cruz) or the monoclonal anti-E-cadherin clone 36 (BD Biosciences) primary antibodies or a non-immune serum used as a negative control, for 1 h at room temperature. Slides were rinsed in phosphate buffered saline, and then incubated with a biotinylated secondary antibody bound to a streptavidin peroxidase conjugate (LSAB+ kit, Dako).
Microarray processing and data analysis as well as procedures to classify human cell lines and mammary tumors are described in detail in Text S1.
Cells were lysed in a 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 2 mM EDTA, 1 mM EGTA, 0.3% CHAPS lysis buffer supplemented with a protease inhibitor cocktail (Roche) and cleared by centrifugation. The PP2A activity was assessed using the “Serine/Threonine Phosphatase Assay System” (Promega) according to the manufacture instruction. Briefly, the cleared cell lysate was filtered through a Sephadex G25 column to remove free phosphate. Protein concentration was determined using the Bradford method. 5 µg of cell protein was incubated in presence of the RRA(pT)VA substrate in a 250 mM imidazole pH 7.2, 1 mM EGTA, 2 mM EDTA, 0.1% β-mercaptoethanol, 0.5 mg/ml BSA PP2A-specific reaction buffer at 25°C for 30 min. After incubation with 50 µl of molybdate dye/additive at 25°C for 30 min, optical density was measured at 620 nm. All determinations were performed in triplicate and the absorbance of the reactions was corrected by determining the absorbance of control reactions without phosphoprotein substrate. The PP2A activity was performed in presence of absence of 5 nM of okadoic acid to confirm the specificity of these reaction conditions. The amount of phosphate released (pmol) was calculated from a standard curve (0–2000 pmol) and was normalized with respect to HMEC-hTERT cells.
|
10.1371/journal.pgen.1007642 | Activating PAX gene family paralogs to complement PAX5 leukemia driver mutations | PAX5, one of nine members of the mammalian paired box (PAX) family of transcription factors, plays an important role in B cell development. Approximately one-third of individuals with pre-B acute lymphoblastic leukemia (ALL) acquire heterozygous inactivating mutations of PAX5 in malignant cells, and heterozygous germline loss-of-function PAX5 mutations cause autosomal dominant predisposition to ALL. At least in mice, Pax5 is required for pre-B cell maturation, and leukemic remission occurs when Pax5 expression is restored in a Pax5-deficient mouse model of ALL. Together, these observations indicate that PAX5 deficiency reversibly drives leukemogenesis. PAX5 and its two most closely related paralogs, PAX2 and PAX8, which are not mutated in ALL, exhibit overlapping expression and function redundantly during embryonic development. However, PAX5 alone is expressed in lymphocytes, while PAX2 and PAX8 are predominantly specific to kidney and thyroid, respectively. We show that forced expression of PAX2 or PAX8 complements PAX5 loss-of-function mutation in ALL cells as determined by modulation of PAX5 target genes, restoration of immunophenotypic and morphological differentiation, and, ultimately, reduction of replicative potential. Activation of PAX5 paralogs, PAX2 or PAX8, ordinarily silenced in lymphocytes, may therefore represent a novel approach for treating PAX5-deficient ALL. In pursuit of this strategy, we took advantage of the fact that, in kidney, PAX2 is upregulated by extracellular hyperosmolarity. We found that hyperosmolarity, at potentially clinically achievable levels, transcriptionally activates endogenous PAX2 in ALL cells via a mechanism dependent on NFAT5, a transcription factor coordinating response to hyperosmolarity. We also found that hyperosmolarity upregulates residual wild type PAX5 expression in ALL cells and modulates gene expression, including in PAX5-mutant primary ALL cells. These findings specifically demonstrate that osmosensing pathways may represent a new therapeutic target for ALL and more broadly point toward the possibility of using gene paralogs to rescue mutations driving cancer and other diseases.
| Mutations inactivating PAX5 disrupt B cell differentiation and occur frequently in ALL. Others have previously shown that restoring PAX5 expression normalizes B cell differentiation and leads to disease remission in a mouse model of ALL. We found that exogenous expression of PAX5’s intact and closely related gene family members, PAX2 or PAX8, which are ordinarily silent in lymphocytes but expressed in kidney and other tissues, can substitute for PAX5 and restore differentiation in ALL cells. A new approach for treating ALL might therefore be to discover ways to activate expression of PAX2 or PAX8 in leukemic cells. In the kidney, PAX2 expression is activated by changes in extracellular osmolarity. We found that PAX2 retains the capacity for osmotic activation in ALL cells and that wild type PAX5 expression also increases when ALL cells are osmotically stressed. Adjustment of serum osmolarity—or treatment with drugs targeting pathways responding to osmotic stress—may offer a potential new avenue for ALL therapy by elevating expression of PAX gene family members. More generally, our studies point toward a novel strategy of recruiting paralogs to complement mutations in genes responsible for cancer and other diseases.
| Pre-B acute lymphoblastic leukemia (ALL) is a common pediatric malignancy often successfully treated with chemotherapy [1]. Unfortunately, chemotherapy is not without side effects, including risk for secondary malignancies and other long-term complications [2]. Additionally, adolescents and adults fare less well, requiring greater reliance on allogeneic hematopoietic stem cell transplant [3]. While chimeric antigen receptor (CAR) T cell therapy for ALL [4] continues to advance, patients may benefit from additional therapeutic options.
As with other types of leukemia, pre-B ALL exhibits stage-specific hematopoietic developmental arrest, in this case, corresponding to hyperproliferation of immature B cell progenitors [5]. Treatment aimed at restoring differentiation capacity to leukemic cells has long been sought, but has proven elusive [6]. The only widely used form of differentiation therapy employs all-trans retinoic acid (ATRA), which has achieved remarkable success for the specific treatment of acute promyelocytic leukemia [7].
The transcription factor PAX5 plays a central role in the origin of pre-B ALL as the single most common somatically mutated gene observed in the disease [8–10]. About one-third of patients acquire heterozygous PAX5 mutations, with complete loss of both alleles rarely seen [9,11]. Deletions or other loss-of-function mutations are typical, but, less frequently, PAX5 rearranges to form fusion genes with ETV6 or other partners, generating dominant negative proteins [12]. Heterozygous germline PAX5 loss-of-function mutation is also a cause of inherited predisposition to ALL [13,14]. In ALL cases defined by wild type PAX5, some acquire mutations in EBF or E2A (TCF3) [9], both of which are upstream activators of PAX5 [5]. Functionally, PAX5 activates B lymphoid-specific gene expression while repressing genes specifying alternative lineages, including T lymphocyte-promoting, NOTCH1 [15]. As such, B lymphoid development in the bone marrow of Pax5-null mice arrests at the pre-B stage [16]. Pax5 loss-of-function in conjunction with Stat5 activation results in developmental blockage of the B cell transcriptional program and leukemic transformation in mice [17]. Importantly, forced re-expression of PAX5 in PAX5-deficient ALL was recently shown to normalize growth and differentiation of leukemic cells in culture and clear circulating leukemic cells in a Pax5-deficient/Stat5-activated mouse model of ALL [18,19]. While cooperating mutations in additional genes arise during leukemogenesis [20], these findings, taken together, indicate that reduced PAX5 activity reversibly drives the formation of pre-B ALL and represents an intriguing therapeutic target.
Nevertheless, modulating PAX5 activity is likely to prove challenging. Transcription factors are generally regarded as “undruggable” [21]. Gene replacement therapy or genome editing [22] may ultimately prove too inefficient when dealing with large numbers of malignant cells. Moreover, targeting or even defining ALL leukemic stem cells for correction may be problematic, if not impossible [23]. However, in the case of genes that are members of paralogous gene families, such as PAX5, genetic redundancy may offer a feasible alternative.
The mammalian PAX gene family consists of nine paralogs [24]. Divergence among its four subfamilies is largely non-coding, within cis regulatory regions, allowing for tissue specific expression among family members [25]. In particular, members of the PAX2/5/8 subfamily (Fig 1, S1 Fig) contain largely identical functional domains, share DNA binding specificity, and exhibit functional redundancy [26,27]. For example, mouse gene targeting experiments, in which PAX2 is replaced by PAX5 under control of endogenous PAX2 regulatory elements, show complementation of developmental abnormalities otherwise resulting from PAX2 deletion [28]. While there is spatiotemporal overlap of PAX2/5/8 expression, for instance in parts of the developing nervous system, less overlap occurs in adult tissues [29]. PAX8 is expressed predominantly in the adult thyroid and PAX2 in the adult kidney, where PAX2 plays a protective role in response to hyperosmolarity encountered by inner medullary cells of nephrons [30]. Only PAX5 is expressed in lymphocytes.
As neither PAX2 nor PAX8 are expressed in lymphocytes, they are unlikely to be subjected to the same selective pressures favoring PAX5 mutation during leukemogenesis, and, not surprisingly, mutations are not detected in ALL [9]. Therefore, it is not hard to imagine that PAX2 and PAX8 could represent intact yet latent functional substitutes for PAX5 in pre-B ALL. Here we demonstrate the ability of both PAX2 and PAX8 to substitute for PAX5 loss-of-function and reverse the developmental blockade in pre-B ALL cells. We show that restoration of differentiation is similar using all three PAX family members and consists of changes to downstream gene expression, cell surface marker expression, cell size, and ultimately cell growth and survival. Additionally, as the translational utility of this strategy is predicated on the ability to activate the endogenous expression of these paralogs in the B cell lineage, we evaluate the aforementioned pathway of response to hyperosmolarity, which plays a prominent role in the kidney. We show that PAX2 and PAX5 exhibit transcriptional upregulation in response to hyperosmolarity in pre-B ALL cells, that PAX2 activation in lymphocytes, as in the kidney, is mediated by the tonicity response enhancer binding protein (TonEBP/NFAT5), and, finally, that hyperosmolarity-driven PAX2/5 activation correlates with changes in B cell developmental gene expression similar to those seen with exogenous PAX2/5/8 re-expression.
PAX5 loss-of-function results in B cell developmental blockade and contributes to leukemic transformation [16,17]. As an important early B cell transcription factor, PAX5 is responsible for both positively and negatively regulating developmental genes, driving differentiation towards a B lymphoid specific fate. Transcriptional targets of PAX5 are numerous and include B cell receptor (BCR) complex protein CD79a, the B cell specific transcriptional regulator BACH2, and the canonical B cell specific surface antigen CD19.
We began by confirming recent findings that re-expression of exogenous PAX5 rescues PAX5-deficient pre-B ALL cells [18] and assessing whether exogenous expression of PAX5 paralogs, PAX2 or PAX8, could function in a similar capacity. We initially evaluated the ability of PAX5, PAX2 or PAX8 to regulate a subset of PAX5 transcriptional targets, including CD79a, BACH2, and CD19. We also included CD10, which is a marker of B cell differentiation exhibiting a bell-shaped pattern of developmental expression levels that peak at the pro to pre-B cell transition [18,31]. We tested PAX factors in Reh cells, which were derived from a primary clonal culture isolated from pre-B ALL peripheral blood [32] and contain a heterozygous p.A322fs PAX5 null mutation [33]. As a PAX5 wild type control, we compared 697 cells, which are derived from a primary clonal culture of ALL bone marrow [34] and contain an E2A(TCF3)/PBX1 fusion gene arising from a t(1;19) chromosomal translocation [35]. Cells were stably transduced with lentivirus expressing either full length human PAX5, PAX2, or PAX8, along with a fluorescent marker, ZsGreen, driven from an internal ribosomal entry site (IRES). As a functionally negative control, we used a vector expressing the clinically observed pre-B ALL PAX5 null mutation, PAX5p.V26fs [36]. At day 4 following transduction, 2×105 ZsGreen-positive cells of each transduction type were sorted by FACS (see S2 Fig for gating strategy). Using quantitative real time PCR, we found that transgene expression of PAX5, PAX2, or PAX8 in both Reh and 697 cells led to significant upregulation of PAX5 target gene expression, relative to empty vector or the negative control PAX5p.V26fs. With the exception of CD10, which is not a known PAX5 transcriptional target, this upregulation was more pronounced in PAX5-mutant Reh cells compared to PAX5-wild type 697 cells (Fig 2A and 2B, respectively).
To further evaluate the ability of PAX2 and PAX8 to rescue PAX5 loss-of-function in pre-B ALL cells, we assessed whether their transcriptional redundancy resulted in enhanced immunophenotypic progression by comparing their ability to modulate a subset of surface markers of B cell differentiation. CD10 (CALLA) and CD19 are surface markers found on normal, as well as leukemic, pre-B cells. Increases in both markers are expect to accompany B cell differentiation, whereas CD38 and CD43 are both downregulated during the large-to-small pre-B cell transition [18,37]. Reh and 697 pre-B ALL cells were transduced with lentivirus expressing PAX-IRES-ZsGreen, as before. At day 4 post-transduction, cells were stained with antibodies for cell surface markers, followed by analysis of ZsGreen-positive cells using flow cytometry. Cells expressing PAX5, PAX2, or PAX8 constructs showed significantly upregulated levels of CD10 and CD19 with downregulated levels of CD38 and CD43, relative to cells transduced with either empty vector or PAX5p.V26fs (Fig 3A and 3B). These results demonstrate a level of functional phenotypic rescue beyond simple transcriptional activation and show a shared ability within the PAX2/5/8 subfamily to promote immunophenotypic changes associated with advanced differentiation in pre-B ALL. Interestingly, PAX5-wild type 697 cells again exhibited similar results (Fig 3C, note scale of intensity).
The large-to-small pre-B cell transition occurs just prior to the emergence of the immature B cell and marks the end of the heavily proliferative large pre-B cell state, resulting instead in a population of pre-B cells which are not only smaller, as the name suggests, but also less proliferative [38]. As noted, we observed that transduction of Reh and 697 cells with PAX paralogs led to decreases in expression of CD38 and CD43, which are both downregulated during this transition [18,37]. This observation suggested that, consistent with prior observations related to PAX5 re-expression in Reh cells [18], PAX2 and PAX8 could advance differentiation in these cells, driving them through the large-to-small transition and ultimately to a normal, more quiescent state. To address this possibility, we analyzed changes in cell size as well as effects on replicative potential following transduction with PAX factors.
The flow cytometry parameter of forward scatter area (FSC-A) is a widely accepted proxy for estimating cell size [39]. Similar to PAX5, exogenous expression of PAX2 and PAX8 led to a reduction in Reh cell FSC-A ranging from 7–10.1%, relative to either empty vector or PAX5p.V26fs negative control (Fig 4A and 4B). Again, 697 cells displayed similar results (Fig 4B). However, as a negative control, the human embryonic kidney cell line, HEK293T, transduced with PAX2/5/8 or controls, did not exhibit a shift in cell size (S3C Fig).
We next evaluated the effect of exogenous PAX paralog expression on the long-term replicative potential of Reh and 697 cells. Cells were transduced with PAX5, PAX2, PAX8, empty vector, or PAX5p.V26fs negative control. At day 4 post-transduction, 2×105 cells of each group were FACS-sorted for ZsGreen (at ~98% purity, see S2 Fig for gating strategy) and returned to culture. For the following 6 days, daily measurement of culture density, performed in duplicate using a hemocytometer, allowed us to compile growth curves for all groups. While control cultures expanded normally, PAX paralog expression resulted in a complete inhibition of culture expansion in Reh cells (Fig 4C, S4A Fig). Growth inhibition was also present, but less complete, in 697 cells (Fig 4D, S4B Fig) and largely absent in HEK293T control cells (S3B Fig). From this point, it became necessary to periodically passage cultures in order to maintain viable cell densities (i.e., 2×105–2×106 cells/mL). At days 11–14, we again used flow cytometry to measure ZsGreen-expressing cell populations. Cultures transduced with PAX paralogs exhibited dramatically reduced ZsGreen expression as a percentage of total cells, ranging from 28–54% in Reh cells and 6–13% in 697 cells, whereas both the empty vector and PAX5p.V26fs control groups maintained expression in ~90% of cells (Fig 4E and S4C Fig). Growth inhibition and the reduced proportion of ZsGreen-positive cells together suggest that these cells reduce their rate of growth and are outgrown by the ~2% of ZsGreen negative cells initially harvested by mis-sorting and/or that PAX/ZsGreen-positive cells die out so that only ZsGreen-negative cells remain and continue to grow. In support of the latter interpretation, PAX gene expression led to an apparent delay in cell cycle progression and conferred a modest increase in apoptosis, as measured by flow cytometry analysis of DNA content (with DAPI staining) and Annexin V staining, respectively (S5 Fig).
We have therefore confirmed previously published literature showing that restoration of PAX5 levels rescues deficiency of PAX5 activity in pre-B ALL cells [18] and have shown for the first time that its paralogs, PAX2 and PAX8, demonstrate a high level of functional redundancy in downstream activation of B cell specific gene expression, promoting differentiation similar to that seen with PAX5.
The observation that PAX2 and PAX8 can rescue the PAX5 loss-of-function differentiation blockade in pre-B ALL cells suggests their activation in vivo could represent a potential therapeutic strategy. In such a context, the use of small molecules to induce their endogenous expression would be useful.
In attempting to identify drugs capable of activating endogenous PAX2 or PAX8 we initially surveyed a variety of agents targeting epigenetic repressive marks or that have been reported to promote lymphocyte differentiation; however, none induced detectable PAX2 or PAX8 expression. We then evaluated compounds known to induce PAX family gene expression in other model systems. Manipulation of transmembrane voltage potential in Xenopus laevis activates transcription factors, including PAX6, resulting in ectopic eye formation [40]. Based on this observation, we tested a variety of hyper- and hypo-polarizing compounds for their ability to induce PAX2 and/or PAX8 expression in Reh cells. We found that 24 hour exposure to membrane depolarizing concentrations (80mM) of C6H11KO7 (K-gluconate) in cell media led to induction of PAX2 expression to as much as 0.3 fold of baseline PAX5, as measured by qRT-PCR. Interestingly, significant upregulation of PAX5 expression was also observed (Fig 5A and S6A, S6B and S6C Fig).
While such concentrations of K-gluconate are known to induce membrane depolarization [41], treatment with monensin and other compounds that are also known to promote membrane hypopolarization did not influence expression of PAX genes. As both potassium and gluconate ions are potentially capable of independent interaction with membrane channels or other cellular machinery that could influence downstream gene expression [42], we tested a variety of salts containing these and other ions, for their ability to influence PAX expression. Surprisingly, 80mM concentrations of NaC6H11O7 (Na-gluconate), KCl, CaCl2, and NaCl all promoted detectable induction of both PAX2 at 0.08–0.5 fold and PAX5 at 3.8–6.1 fold, relative to baseline PAX5 (Fig 5A). Evaluation of downstream PAX5 target and developmental marker genes, CD19, BACH2, and CD10, demonstrated concurrent upregulation at levels similar to those seen with transgene-driven exogenous PAX expression (Fig 5B). While the ionic composition of these agents differs, a commonality is that they all increase the osmolarity of cell growth media.
We observed quantitative differences in the ability of these osmolytes to induce PAX2/5, perhaps due to their variable ability to penetrate the cell membrane, utilizing channels specific for their uptake or efflux. As such, based on their greater relative ability to upregulate both PAX2 and PAX5 in Reh cells, we selected K-gluconate and CaCl2 for further evaluation. Dose-response curves revealed that 80-100mM concentrations (corresponding to ~400-540mOsmol/kg H2O in RPMI media) were optimal for either salts’ ability to upregulate PAX2 and PAX5, with little activity occurring at lower concentrations (Fig 5C and 5D, and S7A and S7B Fig). Similar results were seen with 697 cells; however, the magnitude of induction was less than that observed in Reh cells (S7C and S7D Fig).
In studying the kinetics of this response to hyperosmolarity, 24 hour exposure to high salt concentrations, followed by sorting of live cells and return to normal media for extended incubation revealed that both PAX2 and PAX5 upregulation occurred quickly, but decreased within 24 hours post exposure to salt (Fig 5E and 5F). While CD10 followed a similar temporal pattern to PAX gene modulation, increases in direct PAX5 target genes CD19 and BACH2 were delayed and more persistent, supportive of their sequential response to PAX induction following hyperosmolarity, rather than to hyperosmolarity alone (Fig 5G). Interestingly, the RNA collection method affected the magnitude of induction for PAX2, which was as much as 10-fold greater in RNA extracted from cells immediately following treatment compared to RNA harvested from cells which were first treated, then sorted for viability (as assessed by FSC-A/SSC-A). In contrast, induction of PAX5 appeared to be similar regardless of the RNA collection method. (RNA collection methods are described in Figure Legends and Methods.) This observation suggests an interplay between cell viability and PAX2 expression (Fig 5A and 5C, compared to Fig 5E; see also S2 Fig for gating strategy).
Using cell surface markers, morphological changes, and a subset of PAX5 transcriptional targets, we have demonstrated the ability of PAX2 and PAX8 to rescue PAX5 loss-of-function in pre-B ALL cell lines. To evaluate the full extent to which PAX2 and PAX8 can substitute for PAX5, as well as to compare PAX transgene expression with the response to hyperosmolarity, we evaluated global changes in gene expression by RNA sequencing (RNA-seq) following PAX2, PAX5, or PAX8 transfection or treatment with 80mM K-gluconate or CaCl2 in Reh cells.
Gene set enrichment analysis (GSEA) revealed common enrichment pathways based on biological process and transcription factor targets (Fig 6). Gene sets previously shown to be either direct transcriptional targets of PAX5 at the pro and mature stages of B cell development or whose regulation relies on PAX5 mediation of differentiation from the pro to mature B cell stages displayed enrichment as well [13,43]. We restricted analysis to gene sets with a false discovery rate less than 0.05.
We observed enrichment of 420 gene sets in Reh cells transfected with PAX5. 35% (149) or 26% (108) of these gene sets are also enriched in PAX2 or PAX8 transfected cells, respectively, with 14% (57) common to all three samples (Fig 6A). The majority of these gene sets involve genome accessibility and protein translation (e.g., methylation, peptidyl lysine modification, translational initiation, and cytoplasmic translation), but we also see negative enrichment of known cell cycle regulation transcription factor gene sets such as those involving MYC/MAX and E2F1 (MYCMAX_01 and E2F1_Q4, respectively, S1 Table). PAX2/5/8 transfected samples also show similar enrichment patterns in the PAX5 B cell developmental gene sets (Fig 7A and 7B, S2 Table and S1 Dataset), each factor promoting the upregulation of CD72, IRF4, BACH2, CD19, EGR1, IKZF3, KLF2, and SAMHD1 as well as the suppression of CYBB and FOS.
Interestingly, K-gluconate and CaCl2 share a larger percentage of the PAX5 enriched gene sets, 66% (278) and 60% (254), respectively, than either PAX2 or PAX8 transfected samples (Fig 6B). 53% (221) of the PAX5 enriched gene sets are also enriched following both CaCl2 and K-gluconate exposure (S1 Table). In general, there is a larger, overlapping response when comparing the two salt treatments, presumably part of a general response to hyperosmolarity. Of note, gene sets related to the transport of calcium ions, chloride ions, potassium ions, and organic anions, as well as cytosolic calcium regulation, are positively enriched for all three treatments—an expected result for cells exposed to K-gluconate and CaCl2, but not for forced expression of PAX5. Again, the MYCMAX_01 and E2F1_Q4 gene sets are negatively enriched, linking increasing osmolarity with a pathway for reduced proliferation and a decrease in B cell size [44], although leading edge analysis of the gene sets suggests different genes responsible for enrichment when compared to PAX2/5/8 (S3 Table). Both CaCl2 and K-gluconate show their strongest PAX5 related response in the pro to mature B cell transition gene set (Fig 7C and S2 Table). Here overlapping clusters of upregulated genes similar to the individual pro and mature B cell gene sets (e.g., KLF2, EGR1, IKZF3, SAMHD1, and CD72) are highlighted. TNFRSF13C/BAFF-R, a regulator of peripheral B cell survival, is also upregulated, whereas downregulated genes include cell cycle initiation factors CDC6 and CDC45 and pre-replication complex components MCM3, MCM6, MCM7, and MCM10.
In total, 43 of the 57 gene sets common to PAX2/5/8 transfected samples are also enriched in the CaCl2 and K-gluconate treated samples, corresponding to 10% of the total enriched gene sets in PAX5 transfected Reh cells (Fig 6C, S1 Table). Most of the enriched sets common for both PAX2/5/8 transfectants and salt treatment again relate to genome structure and protein synthesis and also similarly include MYCMAX_01 and E2F1_Q4 transcription factor targets. Both the pro-B cell and pro to mature B cell gene sets are enriched in all samples, as well. The greatest similarity across treatment conditions is seen in the pro-B cell set of genes (Fig 6D), with the only difference being a lack of negative enrichment of genes in either CaCl2 or K-gluconate treated cells. Overall, these results suggest that B cell maturation is regulated by a set of genes and pathways commonly responsive to either PAX gene expression or hyperosmolarity.
Liu et al. [13] restored PAX5 expression in Reh cells and compared global changes in gene expression via RNA-seq to gene expression in a Pax5-deficient/Stat5-activated mouse model of ALL. They identified 31 genes in Reh cells, upregulated by greater than two-fold in response to exogenous PAX5, that are also commonly upregulated with restoration of Pax5 in the mouse model of ALL. Restoration of Pax5 in this model triggers durable disease remission. The log2 fold change values we observed for these 31 genes in PAX2/5/8 transfected and CaCl2 or K-gluconate treated cells appear in Fig 6E, charted alongside corresponding original data from Liu et al. Although treatment windows for our samples were somewhat brief compared to duration of Pax5 induction in mice, we found similar increases in relative expression across this set of 31 genes, albeit at levels roughly half of what Liu et al. reported. These data demonstrate that PAX paralog expression or hyperosmolar treatment both similarly modulate an important subset of genes associated with disease remission when PAX5 expression is restored to normal levels in cell and mouse models of PAX5-deficient ALL.
To confirm RNA-seq results, we used qRT-PCR to validate the responses of several genes where the heatmap clustering showed them to be upregulated by at least 4 of 5 treatment conditions, along with an additional gene, SNX12, which was slightly downregulated by 4 of 5 conditions. qRT-PCR analysis of all 7 of these genes accurately corroborated the trends seen in the RNA-seq data (S8A Fig). Notably, relative to RNA-seq, magnitudes of induction (if present) were almost always greater using qRT-PCR ΔΔCT values. This is likely due to the conservative estimates of differential expression from the DESeq2 normalization algorithm we employed to analyze RNA-seq data. Nevertheless, trends were consistent regardless of technique or genes referenced for comparison.
Cellular response to hypertonicity, as brought about by hyperosmolarity, is thought to be largely mediated by the tonicity-responsive enhancer binding protein, TonEBP [45]. TonEBP, also called (and referred to here as) NFAT5 (nuclear factor of activated T cells 5), is a transcription factor predominantly associated with the kidney but which is also expressed in other tissues, including B cells and, as its name suggests, T cells. Initial response to hypertonicity by NFAT5 involves post-translational modification via phosphorylation, followed by transcriptional activity, including self-induction. Interestingly, NFAT5 mediated gene regulation in the high salt environment of nephrons has been shown to include elevated PAX2 expression, seemingly as part of a survival mechanism during osmotic stress [30]. Not surprisingly, our RNA-seq data showed that hyperosmolarity in Reh cells led to induction of NFAT5, as well as several of its downstream targets (S1 Dataset), consistent with the notion that hyperosmolar concentrations of K-gluconate and CaCl2 generate a canonical response to hypertonicity (i.e., an increase in osmotic pressure gradient across the cell membrane). Subsequent evaluation by qRT-PCR confirmed that NFAT5 mRNA levels, as well as a downstream target associated with B cell maturation, B cell activating factor (BAFF), along with its receptor, TNFRSF13C (BAFF-R) [46], were upregulated in Reh cells after 24 hour treatment with 80mM K-gluconate or CaCl2 (Fig 8A). BAFF-R alone was also upregulated by PAX transgene expression.
Analysis of the 5’ enhancer/promoter regions of both PAX2 and PAX5, along with their intronic and exonic DNA, indicated numerous iterations of the consensus (TGGAAANNYNY) TonE binding site (S9A and S9B Fig) [47]. To determine whether NFAT5 was involved in hyperosmolarity-induced expression of PAX2 and PAX5 and to concurrently assess whether such PAX upregulation directly affected downstream gene modulation, we performed siRNA knockdown of these three genes (Fig 8B and 8C). We found that siRNA knockdown of NFAT5 was sufficient to abrogate PAX2 upregulation in response to 80mM K-gluconate in Reh cells (Fig 8C). Similarly, knockdown of NFAT5 quenched hyperosmolarity mediated increases in the solute carriers, SLC5A3 and SLC6A6, both of which are known targets of NFAT5 (S8B Fig) [48]. Interestingly, neither PAX5 nor PAX5 downstream genes upregulated in response to hyperosmolarity were affected by NFAT5 knockdown (Fig 8C), consistent with a separate, NFAT5 independent mechanism for induction of PAX5 or, at least, reduced sensitivity of PAX5 to changes in NFAT5 levels. Importantly, knockdown of PAX5 itself led to reductions in expression of the downstream genes we assessed, while siRNA directed against PAX2 had little effect (Fig 8C), suggesting that hypertonic induction of residual wild type PAX5 expression outweighs PAX2 with respect to regulation of their common targets. We note that PAX2 expression is detectable as a transcript, but insufficient to measure at the protein level by western blot.
The PAX5 mutation in Reh cells creates a frameshift leading to premature termination and is thus expected to be subject to nonsense-mediated decay. However, western blot indicates that, in addition to a full-length PAX5 protein corresponding to the wild type allele, a truncated polypeptide that is likely non-functional is apparently generated from the mutant allele, albeit at reduced abundance, suggesting that nonsense-mediated decay is incomplete (as evident in Fig 8B, where both products are specifically targeted by siRNA directed against PAX5). Reh cells should therefore contain mRNA from both the wild type and mutant PAX5 alleles. To determine if either salt treatment differentially activates the wild type as opposed to the mutant PAX5 allele in Reh cells, we analyzed RNA-seq data and compared the total number of reads obtained from each allele (and that also include the distinguishing mutation). In untreated cells, 27 of 105 total, non-normalized reads (26%) corresponded to transcripts from the mutant allele. In K-gluconate or CaCl2 treated cells, the equivalent proportions of mutant transcripts were 54/261 (21%) and 36/135 (27%), respectively. (Using a two-tailed test to compare two population proportions, for untreated versus K-gluconate treated cells, the Z-score is 1.05 and p-value is 0.29. The same comparison for CaCl2 treated cells yields a Z-score of -0.17 and p-value of 0.87). These differences are not significant. We conclude that neither salt treatment discriminates between wild type and mutant allele when activating PAX5 expression, as reflected in proportionately increased total read counts.
As numerous studies have shown, long term, in vitro, cell culture inherently selects for gene expression profiles differing from those seen for primary tissue samples [49,50]. To further evaluate whether the PAX2/5 response to hyperosmolarity is one that is intrinsic to ALL cells both in vitro and in vivo, we screened 10 primary pre-B ALL samples for PAX5 mutations, using Sanger DNA sequencing. Of those samples, one, from a 19 year-old male with trisomy 21 Down syndrome, possessed a heterozygous p.(K198Qfs*44) mutation, resulting in frameshift leading to early stop and protein truncation (see Methods). Pre-B ALL occurs more commonly in Down syndrome individuals and is felt to be biologically distinct from disease occurring in non-Down syndrome patients [51]; nevertheless, inactivating mutations of PAX5 are detected at similar frequency in Down syndrome-associated pre-B ALL [52]. Due to limited sample availability from this patient, we performed a single test employing primary cells alongside multiple replicates using primary cells expanded through passage in mice (see Methods). Whether direct from the patient or passaged through mice, 24 hour exposure to 80mM K-gluconate resulted in increased expression of PAX5, as well as several but not all downstream targets seen previously with Reh and 697 cells (Fig 9A and S10A Fig). PAX2 expression was not detected in this assay; however, this may be due in part to low RNA input levels, which were constrained by sample quantity.
The osmotic concentrations of K-gluconate or CaCl2 we evaluated in vitro would prove lethal if administered clinically. However, mannitol is also known to activate NFAT5 [53] and is used to adjust serum hyperosmolar concentrations to high levels in certain clinical settings [54]. To test whether mannitol could be employed to upregulate PAX2 or PAX5 in pre-B ALL, we treated Reh cells with 80mM or 160mM mannitol for 24 hours, prior to FACS sorting for live cells and harvesting of RNA. qRT-PCR demonstrated dose-dependent increases both for PAX2 and especially for PAX5, along with similar changes in downstream gene expression, albeit not to the level seen with K-gluconate (Fig 9B and 9C). Importantly, 160mM is near the range of clinically achievable therapeutic concentrations for mannitol [54]. Comparison of 160mM mannitol with 80mM K-gluconate or CaCl2, followed by FSC-A/SSC-A sorting of live cells and subsequent measurement of culture expansion demonstrated slightly reduced growth potential for K-gluconate and CaCl2 treated cells as compared to cells grown in 160mM mannitol or normal media (S10B Fig). Interpretation of long term viability in response to hyperosmolarity was complicated due to the noticeably brief induction of PAX2/5 (Fig 5E), coupled with the generally harsh nature of such treatment, even with only 24 hour exposure. The growth delay observed with K-gluconate in this case may largely be due to cell cycle arrest or other adverse effects of elevated hyperosmolarity [55], rather than the PAX dependent, developmentally programed exit from the cell cycle we appeared to observe with continuous PAX re-expression. However, even in vitro, mannitol appears to be better tolerated, and thus it or related organic osmolytes may present options for modulating tonicity that could prove tolerable in vivo.
Liu et al. recently demonstrated that restoration of PAX5 expression can reverse the developmental blockade holding PAX5-mutated pre-B ALL cells in a continuously replicating, developmentally immature state [18]. We have confirmed that result and extended it further by showing that PAX5’s closely related paralogs, PAX2 or PAX8, neither of which is mutated in ALL nor ordinarily expressed in lymphocytes, can function equivalently to normalize differentiation and growth of pre-B ALL cells. Moreover, we have shown that endogenous PAX2 expression, and unexpectedly also PAX5 itself, can be upregulated to promote similar effects on differentiation of pre-B ALL cells under hypertonic conditions.
While germline loss-of-function mutations are a cause of familial pre-B ALL [13,14], demonstrating that PAX5 deficiency can ultimately initiate leukemogenesis, loss of PAX5 activity is not by itself sufficient, and development of leukemia requires additional cooperating mutations. Cancer genome sequencing has identified a wide diversity of mutations [8,9], such that no two ALL patients are likely to share identical mutational profiles. Reh and 697 cells, tested here as well as by Liu et al. [18], are quite dissimilar, with 697 cells having only a few coding sequence alterations while Reh cells have considerably more, with very little overlap (S11 Fig). In particular, Reh cells contain a heterozygous loss-of-function PAX5 mutation [33], whereas in 697 cells, PAX5 is intact (per our sequencing, S13 Fig). However, an upstream regulator of PAX5, E2A (TCF3), is at least partially inactivated via a translocation involving PBX1 [35], suggesting that there may be similarly reduced expression of PAX5 in 697 cells. Regardless, targeting PAX2/5/8 activity may prove beneficial even in those patients lacking PAX5 mutations. Liu et al. also demonstrated that PAX5 replenishment succeeded in curing a transgenic mouse model of ALL, driven by PAX5 knockdown combined with Stat5 activation [18]. The fact that PAX5 re-expression normalizes growth and differentiation in pre-B ALL with divergent genetic backgrounds and mutational signatures, including with Down syndrome associated ALL as tested in primary cells, suggests that even after cooperating mutations have arisen, loss of PAX5 activity continues to support the leukemic state. Consistent with the concept of oncogene addiction, in which secondary mutations are dependent upon driver mutations for maintaining the cancer phenotype [56], acquisition of additional mutations may therefore possibly render ALL cells even more vulnerable following replacement of PAX5 activity.
Current approaches for treatment of pre-B ALL continue to rely on chemotherapy and, more recently, immunotherapy. Chemotherapy is often successful in pediatric settings [1] but is associated with considerable toxicity, long-term side effects [2], and substantially reduced efficacy in older children and adults, where allogenic stem cell transplant is more heavily relied upon [3]. Recent breakthroughs in CAR T cell therapy have shown great promise in treating certain disease presentations, specifically those which are highly CD19 positive [57]. However, hurdles remain, including clonal selection for PAX5 deletion with consequent downregulation of the CD19 target antigen, leading to disease resistance [58]. Since CD19 is a direct target of PAX5 and, as we have shown, can be activated equally well by PAX2 or PAX8, the therapeutic approach contemplated here may work in conjunction with CAR T cell therapy by increasing levels of the targeted CD19 B cell antigen, even after loss of PAX5. CD19 can also be targeted through other forms of therapy, such as with antibody-drug conjugates [59].
Our observations demonstrate the use of gene paralogs to resolve a human disease phenotype. A remaining challenge, however, involves approaches for activating developmentally silenced genes in vivo. We tested a variety of compounds based upon previously described properties as either generally reversing repressive chromatin modifications (zebularine, hydralazine, valproic acid, azacitidine, and vorinostat) or as mechanistically undefined inducers of lymphocyte differentiation (ATRA, methotrexate, and phorbol 12-myristate 13-acetate (PMA)). None consistently activated PAX2 or PAX8 expression or otherwise promoted pre-B ALL cell differentiation under conditions we evaluated.
Another class of compounds we tested affect cell membrane potential, the modulation of which has been shown in model systems to induce a variety of developmental transcription factors, including, for example, PAX6 [40]. We observed induction of PAX2, but not PAX8, as well as increases in downstream differentiation markers in response to K-gluconate. After testing a variety of other salts as well as several non-ionic modulators of cell membrane potential, we concluded that our observation was likely a cellular response to hypertonicity. During water diuresis, physiological concentrations of salts, mainly NaCl, in the renal inter-medullary interstitial fluid reach concentrations ranging from 600 to more than 1000mOsmol/kg H2O. Interestingly, survival mechanisms for cells in these conditions include the anti-apoptotic upregulation of PAX2, which has been shown to peak in mouse intermedullary collecting duct cells at ~500mOsmol/kg H2O [30], similar to what we observed in Reh cells (~400-540mOsmol/kg H2O in RPMI media). Unexpectedly, we observed that hypertonicity also induced expression of PAX5 in pre-B ALL.
RNA-seq performed in conjunction with GSEA highlighted similarities and differences resulting from expression of PAX2 or PAX8, compared to PAX5, in PAX5-deficient ALL cells. In a pairwise comparison of any of the three PAX factors, slightly fewer than half of all gene sets exhibiting significant expression changes were common to both, and only 13% of all gene sets (57/440, Fig 6A) enriched by PAX5 were commonly modulated by all three PAX genes. Importantly, however, the group of gene sets commonly regulated by all three PAX factors includes PAX5 targets most relevant to B cell maturation (Fig 6D), consistent with our findings that all three PAX factors similarly promote differentiation of PAX5-deficient ALL cells. We speculate that PAX5 target genes likely reside in accessible chromatin configurations in pre-B cells, such that even imperfect PAX activity from a paralog may readily induce their expression. In contrast, gene sets exhibiting significant enrichment following treatment with CaCl2 or K-gluconate exhibited much greater overlap with PAX5, and a majority of gene sets showing enrichment with PAX5 (221/420, Fig 6B) or that were commonly enriched by all three PAX factors (43/57, Fig 6C) were also enriched after treatment with CaCl2 or K-gluconate. This may not be surprising given that treatment with either salt induced expression of PAX2 and, especially, PAX5 itself. Finally, it is worth emphasizing from a translationally relevant standpoint, that a set of 31 genes found by Liu et al. to undergo significant regulation during ALL remission, as induced by Pax5 restoration in a mouse model of Pax5-deficient ALL, were similarly modulated by all tested conditions in our studies, whether it be PAX5, PAX2, PAX8, K-gluconate, or CaCl2 (Fig 6C).
We found that components of the NFAT5 pathway, including NFAT5 and TNFS13B (BAFF), along with its receptor, TNFRSF13C (BAFF-R), are upregulated in response to many or all of our treatments (i.e., PAX2/5/8 or salt treatment, Fig 8A and S1 Dataset). Named “nuclear factor of activated T cells 5,” for its role as a transcriptional coordinator of T cell immune response [60], NFAT5 is the only known osmosensing mammalian transcription factor and is active in a variety of cell types, including B cells [45,46]. Indeed, siRNA mediated knockdown of NFAT5 in Reh cells led to a reduction in PAX2 expression in response to hyperosmolarity (Fig 8B and 8C). However, the added observation that PAX5 expression was not affected by NFAT5 knockdown suggests either the presence of a separate, non NFAT5 related, osmosensing pathway upstream of PAX5, or alternatively, a substantially lower threshold for NFAT5 abundance to achieve upregulation of PAX5 under these conditions. In support of the latter, PAX5 appears to contain more potential NFAT5 binding sites than PAX2 (S9 Fig). Separately, these siRNA experiments showed that PAX5 upregulation had a greater effect on downstream gene regulation, and presumably B cell maturation, than did PAX2 (Fig 8B and 8C). Based on our observations from earlier experiments (Figs 2–4), where PAX2 effectively functionally mimicked PAX5, and the substantially lower level of PAX2 expression present relative to the induced levels of PAX5 in response to hyperosmolarity (~20 fold), we believe this most likely reflects relative levels of expression, rather than differences in functionality.
Given that components of the hypertonicity response pathway are highly conserved from single cell organisms to mammals [61], it seems reasonable to speculate that PAX genes, including PAX2 and PAX5, may play a role in osmotic adaptation across various tissue types. In fact, similar to our observations, upregulation of PAX2 occurs in mouse embryonic fibroblasts in response to hypertonicity [48]. Secondary lymphoid organs, including spleen and thymus, maintain a remarkably high osmolar environment compared to serum and other tissue [62]. It should not be overlooked that a decrease in cell size, which we observed upon expression of PAX2/5/8, normally accompanies the large-to-small pre-B cell transition as cells begin their migration from the bone marrow to secondary lymphoid organs. It is possible that exposure to differences in local osmolarity across these compartments could play a role in normal lymphocyte development. Whether upregulation of PAX2 and/or PAX5 is a normal physiologic response to osmotic stress in lymphocytes or a vestigial pathway more heavily relied on in other tissues such as the kidney, but which is capable of artefactual activation under extreme circumstances, we show here that osmotic stress exposes a potential therapeutic target for activating elements of the normal B cell differentiation program.
The osmolar concentration required for peak induction of PAX2 and PAX5 is, just barely, outside the clinically achievable range for serum based on maximum recommended dosing for mannitol [54]. It is possible that specialized delivery methods, manipulation of dosage levels, and/or exposure time may bridge this gap. It is also worth noting that serum osmolar concentrations within this range are sometimes encountered in acutely ill diabetes mellitus patients with hyperglycemic hyperosmolar syndrome [63]. However, even if the highly hyperosmolar conditions we subjected ALL cells to in vitro are not therapeutically tenable in vivo, they do suggest that the complexity of kinases and other components of the signaling pathway responding to hypertonicity, including those regulating NFAT5, at least in the case of PAX2 [64], may be ripe for investigation as drug targets.
An additional limitation relates to duration of therapy, as the replacement of PAX5 activity may only have a temporary effect on differentiation of ALL cells, though this may still be beneficial either as a form of induction therapy or as an adjuvant when combined with CAR T cell, other therapies targeting CD19, or conventional chemotherapy. Intriguingly, a relevant recent in vitro study demonstrated that hyperosmotic stress achieved with salt or mannitol treatment synergized with chemotherapeutic drugs to kill ALL cells via an NFAT5 dependent mechanism, although activation of PAX genes was not investigated [53]. It should also be emphasized that remissions achieved with differentiation therapy employing ATRA for promyelocytic leukemia can actually be enduring [7]. Finally, if differentiation of pre-B ALL cells could be pushed as far as to the plasma cell stage, where PAX5 expression is normally extinguished [65], then mutations inactivating PAX5 could become inconsequential, anyway.
Finding the right balance of PAX gene expression is another issue. PAX2, when activated, can behave as an oncogene in solid tumors [66], and PAX5 is normally down-regulated during plasma cell differentiation [65]. However, our RNA-seq data suggest that there may be an auto-regulatory ceiling for PAX gene expression, particularly for PAX5. Specifically, by examining total PAX5 transcripts and comparing differences in the read ratios of SNPs discriminating between native and exogenous PAX5, we observed an apparent suppression of endogenous PAX5 transcript by PAX5 transgene expression, and to a lesser extent, by the expression of PAX2 or PAX8 transgenes (S12 Fig). Of course, unless PAX gene activation is confined only to the leukemic population of cells, there may be undesirable effects in other tissues, although compared to oncogenic mutations, PAX gene activation by osmoresponsive mechanisms is unlikely to be permanent. Moreover, some current cancer therapies employ treatment with epigenetic modifier drugs, such as azacitidine, capable of producing genome-wide and persistent activation of many genes across multiple tissues [67].
The strategy implemented here, to activate expression of intact and functionally similar paralogs of mutated cancer-driver genes to therapeutically restore cellular differentiation, could potentially be extended to other types of cancer. For example, inactivating RUNX1 mutations frequently occur in acute myeloid leukemia, where upregulation of RUNX2 or RUNX3 exhibits anti-leukemic effects [68]. More generally, a wide variety of non-cancer illnesses possess etiologies for which complementation of inactivating mutations by activating gene paralogs may prove useful, extending the potential therapeutic application of this concept. For example, in spinal muscular atrophy, causative loss-of-function mutations in SMN1 can be rescued by a recently approved therapy which uses an antisense compound to promote exon retention in an alternatively spliced yet otherwise identical paralog, SMN2 [69]. Finally, hypertonic activation of PAX gene expression offers an example of emerging “electroceutical” approaches based on manipulation of biophysical phenomena [41].
Leukemia cells were collected, after informed consent, through the Cell Bank of the Center for Cancer and Blood Disorders at Children’s Hospital Colorado. The Cell Bank protocol is approved by the Colorado Multiple Institutional Review Board (COMIRB #00–206).
Animal use was approved by the Animal Care and Use Committee of the University of Colorado Denver (Protocol 66912(12)1E).
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Marshall Horwitz ([email protected]).
Ten B ALL samples were screened for mutation in PAX5 by Sanger DNA sequencing. One sample, CHCO-7657, was found to contain a heterozygous PAX5 mutation resulting in p.K198fs (S16 Fig). A subset of primary cells were stored in liquid nitrogen and additional leukemia cells were passaged through NOD scid gamma (NSG) mice once and amplified in NOD scid gamma Il3-GM-SF (NSGS) mice (both purchased from Jackson Labs), prior to storage in liquid nitrogen. Recipient mice were irradiated with 200cGy via X-irradiator prior to leukemia injection. Prior to experimentation, cells were thawed and resuspended in 20%FBS/MEM-alpha (Gibco,12561–056) which had been preconditioned with OP9 feeder cells (ATCC) seeded the day prior at 3×105 cells/T75 flask, in 10mL media. Primary cells were overlaid onto and co-cultured with these OP9 feeder cells for 24 hours, followed by treatment with 80mM K-gluconate for an additional 24 hours before sorting for live cells and isolation of RNA.
Cells were lysed using RIPA buffer with complete protease inhibitor (Roche), 1mM Na3VO4 and 1mM PMSF. Lysates were quantified with the Pierce BCA Protein Assay Kit (ThermoFisher) and electrophoresed on MINI-protein TGX gels (BioRad) and transferred onto PVDF membrane (BioRad). Membranes were immunoblotted with indicated antibodies. All primary antibodies were diluted in 1% milk or BSA in TBST and were incubated overnight at 4°C. Blots were incubated in secondary antibody for 1 hour at room temperature, also in 1% milk or BSA.
Flow cytometry for cell surface markers was performed on a BD LSR II flow cytometer using indicated antibodies. For antibody staining, cells were washed twice in sorting buffer (1%FBS/PBS), prior to incubation in antibody (diluted in sorting buffer) on ice and in the dark, for 30 minutes. Cells were then again washed twice in sorting buffer, and resuspended in 300–500μL sorting buffer prior to analysis. All staining and washing was done in 96 well, flat bottom plates. In between washes, cells were spun down for 3’ at 300×g. Plates were overturned and shaken to remove buffer. FACS was performed on a BD Aria II cell sorter. All raw data files were processed using FlowJo software. For experiments where RNA was harvested, cells were sorted directly into 500μL of Qiagen Buffer RLT+, prior to RNA isolation (see below). For further propagation of live cells, cells were sorted directly into complete growth media. Experiments were performed in triplicate (at minimum) unless otherwise noted in figure legends.
RNA was harvested from cells using the RNeasy Plus Mini Kit (Qiagen), following the supplied protocol, and converted to cDNA using random oligomers and either Superscript III or Superscript IV reverse transcriptase (Invitrogen). qRT-PCR was performed on cDNA using the indicated TaqMan probes and analyzed on an Applied Biosystems StepOnePlus Real-Time PCR System. Relative quantification of mRNA abundance was performed using the 2-ΔΔCT method and ACTB or GAPDH as reference genes, where ΔCT = (CTtarget-CTreference) and 2-ΔΔCT = 2-(ΔCTsample-ΔCTcontrol). Note, in the cases of PAX2 and PAX8, for which no endogenous baseline expression was detected in ALL cells, baseline PAX5 (empty vector or vehicle) expression was used in calculating ΔCTcontrol. Experiments were performed in triplicate unless otherwise noted in figure legends.
At day 4 post-lentiviral transduction, 2×104 ZsGreen-positive cells of each PAX gene or control vector type were isolated by FACS and distributed into individual single wells of a 96-well plate. Beginning with normalized concentrations of 2×105 cells/mL (i.e., 100 μL total volume/well), these cells were further propagated in culture for a time course of 15–17 days. Culture density was assessed manually every 1–2 days during this time, using a hemocytometer. Additional media was added as needed prior to each counting in order to account for evaporation and to maintain ~100 μL volume in each well. For HEK293T cells, cell viability was assessed using an MTS assay (Cell Titer 96 One Solution, Promega), which produces a formazan product in the presence of phenazine methosulfate, which is present in metabolically active cells. Soluble formazan product is detectable at a 490nm absorbance maximum in PBS. Experiments were performed in triplicate (at minimum) unless otherwise noted in figure legends.
Cells were passaged one day prior to plating at a density of 2×105 cells/mL in 5 or 10mL of regular growth media with an added 80mM K-gluconate (Sigma, P1847) or CaCl2 (Sigma, C-3306) (unless otherwise noted). After indicated incubation times and depending on the experiment, RNA was either bulk harvested from treated cells or was harvested from live cells that were first sorted and collected by flow cytometry based on FSC-A and SCC-A measurements (indicated in figure legends). For pulse/chase in Fig 5E, 5F and 5G, cells were treated as indicated for 24 hours. 3×105 live cells were then sorted and returned to culture followed by removal of aliquots at indicated time points for harvesting of RNA (0h = 24 hour pulse, 0 hour chase). Experiments for all figures were performed in triplicate (at minimum) unless otherwise noted in figure legends.
4×106 Reh cells were electroporated with SMARTpool siRNA (i.e., 3 separate target siRNAs each) for PAX2, PAX5, NFAT5, or a non-targeting control pool (Dharmacon) using a BioRad GenePulser Xcell (Square wave, 210V, 15ms, 2x pulse, 0.1sec gap). Cells were suspended in 400μL Opti-MEM buffer containing 500nM siRNA that had been previously prepared and frozen at 20μM stock concentration in siRNA resuspension buffer (GE Healthcare). Cells were allowed to recover for 24 hours prior to harvest of protein lysates or treatment with hypertonic media (80mM K-Gluconate in RPMI with 10% FBS).
Bar graphs for antibodies and cell size (FSC-A) represent mean fluorescence intensity. Values are averaged across several experimental replicates, as indicated, above. Error bars represent standard deviation. Significance was determined using one-way t-test method for deviation from a fixed value (i.e., normalized value of control sample). p-values * <0.05, ** <0.005, *** <0.0005. Bar graphs for cell cycle phase (S5 Fig) were determined from percentages of cells in G1, S, and G2 phase, based on DAPI staining, and assessed by the “Cell cycle” function in FlowJo, vX.
Normalization and differential expression calculations were performed using the R package DESeq2 [73] based on TPM data. Clustering and heatmap creation were performed using the heatmap.2 package (dist = Euclidean and method = complete).
Expressed genes in each sample were ranked based on their log2 fold change in mRNA levels when compared to the appropriate control. GSEA was conducted using GSEA Desktop 3.0 software (Broad Institute). Gene sets analyzed (Molecular Signatures Database v6.1) include the biological process group from Gene Ontology (GO:BP), the transcription factor targets (TFT) group, and a custom PAX5 related group of human genes based on genes differentially expressed at various stages of B cell development in mice that have either normal levels of PAX5 or are deficient [13,43]. Briefly, the pro and mature B cell sets are comprised of genes that are differentially expressed compared to the appropriate controls and have predicted PAX5 binding sites in their promoter region while the pro-to-mature B cell sets contain all genes differentially expressed when comparing mature B cells to pro-B cells in the presence or absence of PAX5. Analysis was conducted using the GSEAPreranked tool to calculate a classic enrichment score for each set. Gene sets with a false discovery q-value (FDR) of <0.05 were selected for further analysis.
|
10.1371/journal.pgen.1007762 | The Zn2Cys6-type transcription factor LeuB cross-links regulation of leucine biosynthesis and iron acquisition in Aspergillus fumigatus | Both branched-chain amino acids (BCAA) and iron are essential nutrients for eukaryotic cells. Previously, the Zn2Cys6-type transcription factor Leu3/LeuB was shown to play a crucial role in regulation of BCAA biosynthesis and nitrogen metabolism in Saccharomyces cerevisiae and Aspergillus nidulans. In this study, we found that the A. fumigatus homolog LeuB is involved in regulation of not only BCAA biosynthesis and nitrogen metabolism but also iron acquisition including siderophore metabolism. Lack of LeuB caused a growth defect, which was cured by supplementation with leucine or iron. Moreover, simultaneous inactivation of LeuB and HapX, a bZIP transcription factor required for adaptation to iron starvation, significantly aggravated the growth defect caused by inactivation of one of these regulators during iron starvation. In agreement with a direct role in regulation of both BCAA and iron metabolism, LeuB was found to bind to phylogenetically conserved motifs in promoters of genes involved in BCAA biosynthesis, nitrogen metabolism, and iron acquisition in vitro and in vivo, and was required for full activation of their expression. Lack of LeuB also caused activation of protease activity and autophagy via leucine depletion. Moreover, LeuB inactivation resulted in virulence attenuation of A. fumigatus in Galleria mellonella. Taken together, this study identified a previously uncharacterized direct cross-regulation of BCCA biosynthesis, nitrogen metabolism and iron homeostasis as well as proteolysis.
| Adaptation to the host niche is an essential attribute of pathogens. Here we found that the Zn2Cys6-type transcription factor LeuB cross-regulates branched-chain amino acid biosynthesis, nitrogen metabolism, iron acquisition via siderophores, and proteasome activity in the mold Aspergillus fumigatus. Lack of this regulatory circuit impaired virulence in an insect infection model. Mammals do neither express Zn2Cys6-type transcription factors nor have the capacity to produce branched-chain amino acids or siderophores. Consequently, this regulatory circuit is a paradigm for fungal pathogen-specific adaptation to the host niche.
| Aspergillus fumigatus is the most important airborne fungal pathogen, causing allergic and invasive diseases, termed aspergillosis, the latter particularly in immunocompromised patients [1,2]. A critical virulence attribute of most pathogens, including A. fumigatus, is efficient iron acquisition [3–6]. On one hand, iron is an essential cofactor required for a large number of biological processes including respiration and biosynthesis of deoxyribonucleic acid, amino acids and lipids. On the other hand, mammalian hosts represent an iron-limited niche as the iron is tightly sequestered by proteins and, moreover, the innate immune system responds with iron-withholding strategies to infections [7]. Therefore, to overcome the low bioavailability of iron during invasion, pathogens have evolved various systems to struggle for iron from host. A. fumigatus possesses two major high-affinity iron uptake systems, reductive iron assimilation (RIA) and siderophore-mediated iron acquisition (SIA), whereby the latter has been shown to be essential for virulence [8,9]. Siderophores are low-molecular mass, ferric iron specific chelators, the production and secretion of which is induced by iron starvation. The major secreted siderophore of A. fumigatus is triacetylfusarinine C (TAFC). Upon binding of environmental iron, the TAFC-iron chelate is taken up by a specific transporter, termed MirB [10,11]. Reverse genetics identified several SIA components, which proved to be crucial for virulence of A. fumigatus in animal models of aspergillosis [12–15]. As iron is not only essential but also toxic in excess, iron uptake, consumption and detoxification have to be tightly controlled. To maintain iron homeostasis, A. fumigatus has evolved two major transcription factors, SreA and HapX. The GATA-type transcription factor SreA represses RIA and SIA during sufficient iron supply [16]. In contrast, the bZIP-type transcription factor HapX represses iron-dependent pathways to spare iron and activates RIA and SIA to promote iron uptake during iron starvation [17], while it activates iron-dependent pathways and particular iron detoxification via transport into the vacuole during iron excess [18].
In A. fumigatus, one of the iron-dependent pathways comprising genes repressed during iron starvation and induced by iron excess via HapX is branched-chain amino acid (BCAA) biosynthesis [17]. Mammals do not have the capacity to produce BCAA and thus valine, leucine or isoleucine must be supplied in the diet. In contrast, fungi are BCAA autonomous. As depicted in S1 Fig (using the nomenclature for S. cerevisiae), BCAA biosynthesis consists of a common pathway that leads from pyruvate and threonine, respectively, to valine and isoleucine. Leucine biosynthesis starts with an intermediate of valine biosynthesis, 2-ketoisovalerate, which is converted to α-isopropylmalate (αIPM) by αIPM synthase (S. cerevisae contains two paralogs Leu4/9). αIPM is then further processed by the αIPM isomerase (Leu1), the β-isopropylmalate (βIPM) dehydrogenase (Leu2), and the branched-chain amino acid aminotransferase (Bat2). Leucine biosynthesis is feedback-regulated due to inhibition of Leu4/9 enzyme activity by leucine. Moreover, in S. cerevisiae the intermediate αIPM has been shown to posttranslationally activate the Zn2Cys6-type transcription factor Leu3, which activates several steps in BCAA biosynthesis (Ilv2 and Ilv5, which are involved in general BCAA biosynthesis, and the leucine biosynthetic enzymes Leu4, Leu1, Leu2 and Bat1) as well as the NADP-dependent glutamate dehydrogenase Gdh1, a crucial enzyme in nitrogen metabolism [19]. Likewise, the Leu3 homolog LeuB transcriptionally activates genes involved in leucine biosynthesis as well as the glutamate dehydrogenase-encoding gene gdhA in Aspergillus nidulans [19–21]. The BCAA biosynthetic pathway comprises two enzymes, whose activity depends on iron-sulfur clusters: mitochondrial dihydroxyacid dehydratase (termed Ilv3 in S. cerevisiae), which is required for biosynthesis of all three BCAAs, as well as cytosolic 3-isopropylmalate dehydratase (termed Leu1 in S. cerevisiae), which is required exclusively for biosynthesis of leucine [19]. In A. fumigatus, these two enzymes were found to be transcriptionally downregulated during iron starvation together with numerous other “iron-dependent genes” in order to spare iron [17].
In the current study, we identified in A. fumigatus a cross-regulatory link of BCAA biosynthesis, nitrogen metabolism and iron homeostasis via the transcription factor LeuB (AFUB_020530).
HapX is a key transcription factor coordinating the response to iron starvation. To identify potential regulators controlling expression of this transcription factor, we compared promoter regions of HapX-encoding genes. MEME analysis [22] of 1-kb 5’-upstream regions of hapX homologs from 20 Aspergillus species identified a highly conserved CCGN4CGG motif localized about 500 nt upstream of the hapX translation start point in A. fumigatus (Fig 1). This motif resembles the typical binding consensus sequence for Leu3/LeuB transcription factors [23]. In agreement, MEME analysis identified phylogenetically conserved CCGN4CGG motifs in promoters of several BCAA biosynthetic genes, i.e. α-isopropylmalate isomerase-encoding leuA (Fig 1), acetohydroxy acid reductoisomerase-encoding ilv5, α-isopropylmalate synthase-encoding leuC (homolog of S. cerevisiae leu4/9), and β-isopropylmalate dehydrogenase-encoding leu2A (homolog of S. cerevisiae leu2) as well as glutamate dehydrogenase-encoding gdhA (S1 Table). These findings suggested a role of LeuB in regulation of iron homeostasis in addition to BCAA biosynthesis in A. fumigatus.
BLAST analysis identified an A. fumigatus homolog (AFUB_020530/ AFUA_2G03460) to A. nidulans LeuB (termed AnLeuB here). To investigate whether A. fumigatus LeuB (termed LeuB here) can directly interact with the predicted CCGN4CGG motifs, we expressed the predicted DNA binding domain of LeuB in E. coli and analyzed protein-DNA interaction (Fig 2A and 2B). As shown in Fig 2C, LeuB clearly interacted with promoter fragments of hapX and gdhA genes in electrophoretic mobility shift assays (EMSA). The gdhA gene encodes NADP-dependent glutamate dehydrogenase, a key enzyme in nitrogen metabolism, which has previously been shown to be regulated by Leu3/LeuB in S. cerevisiae and A. nidulans [19–21]. Excess of unlabeled DNA and mutating the CCGN4CGG motif to CTGN4CAG in the hapX fragment blocked the interaction of LeuB with the promoter fragments, underlining the specificity of the protein/DNA interaction.
Surface plasmon resonance (SPR) analyses, shown in Fig 2D, confirmed that the DNA-binding domain of LeuB binds with high affinity to the phylogenetically conserved CCGN4CGG motifs in the promoters of hapX (KD = 159.6 nM) and leuA (KD = 95.2 nM) and that mutation of one of the CCG sequences abolishes binding, which again underlines the sequence specificity.
To further analyze the function of LeuB in A. fumigatus, we generated a leuB gene deletion mutant, ΔleuB, by replacing the leuB-coding region with the Neurospora crassa pyr4 gene. Compared to the parental wild-type (WT) strain, ΔleuB displayed dramatically reduced growth on minimal medium agar plates (Fig 3A). This growth defect was cured by re-integration of the leuB gene (strain leuBC) (Fig 3A), together with diagnostic PCR and Southern blot analyses which underline the accuracy of the genetic manipulation (S2 Fig). Moreover, leucine supplementation largely rescued the ΔleuB growth defect (Fig 3A), as previously seen in S. cerevisiae and A. nidulans mutants lacking orthologs of LeuB [19–21]. According to the hypothesis that LeuB might play a role in iron homeostasis, we analyzed the growth of ΔleuB under conditions of different iron availability (Fig 3B). Remarkably, the presence of the ferrous iron-specific chelator bathophenanthroline disulfonic acid (BPS), which inhibits RIA and creates an iron-poor environment [12], completely blocked growth of ΔleuB, while iron supplementation significantly improved the growth of ΔleuB. In agreement with LeuB being important for adaptation to iron starvation, the biomass of ΔleuB reached only 23.5% of that of the WT during iron starvation but increased to 70.8% during iron sufficiency in submersed culture conditions (S3A Fig). In contrast, other metals such as magnesium, calcium or manganese were unable to rescue the growth defects of ΔleuB (S3B Fig).
To investigate, whether LeuB is also involved in iron homeostasis in A. nidulans, we generated a leuB deletion mutant in A. nidulans, ΔAnleuB. Similar to A. fumigatus, the growth of ΔAnleuB was significantly stimulated by iron supplementation (S3C Fig). Moreover, the growth defect of the A. fumigatus ΔleuB mutant was rescued by genetic complementation with the A. nidulans leuB gene (strain ΔleuBAnleuB) (Fig 3A). These data document that the function of LeuB in iron homeostasis is conserved in A. nidulans.
The major transcription factors ensuring maintenance of iron homeostasis are SreA and HapX. SreA represses iron acquisition under sufficient iron supply and is consequently important for adaptation to iron excess [16]. HapX induces iron acquisition and represses iron consumption during iron starvation, while it induces iron detoxification during iron excess [17,18]. Consequently, HapX is important for adaptation to both iron starvation and iron excess. To further explore the role of LeuB in iron homeostasis, we generated mutants lacking either leuB and hapX (strain ΔleuBΔhapX) or leuB and sreA (strain ΔleuBΔsreA). Lack of both HapX and LeuB exacerbated the growth defect seen in mutants lacking only one of these two transcription factors, i.e. growth could not be cured by supplementation with high amounts of iron but only by leucine supplementation (Figs 3B and S3A). In comparison, ΔleuBΔsreA displayed improved growth under low iron conditions (BPS, without iron supplementation) compared to ΔleuB (Figs 3B and S3A). The suppression of the low-iron growth defect of ΔleuB by lack of SreA can be explained by the fact that lack of SreA derepresses iron acquisition and thereby increases iron uptake resembling supplementation with iron.
Taken together, these data collectively suggest that LeuB links iron homeostasis maintenance and leucine biosynthesis. Since LeuB is a putative transcription factor, we next analyzed its cellular localization via C-terminal tagging of LeuB with green fluorescent protein (LeuB-GFP protein) expressed under control of the native leuB promoter. Epi-fluorescence microscopy studies revealed predominant nuclear localization of LeuB-GFP during both iron starvation and sufficiency as well as with and without leucine supplementation (Fig 3C), suggesting posttranslational activation of LeuB independent of its intracellular localization.
Analysis of whole-cell protein extracts, using a non-denaturing extraction buffer, via sodium dodecylsulfate polyacrylamide gel electrophoresis (SDS-PAGE) followed by Coomassie blue-protein staining, revealed increased proteolysis in ΔleuB compared to WT and leuBc strains, i.e., large-size proteins displayed significantly decreased abundance in mycelia cultured under iron starvation (Fig 4A). In agreement, Western blot analysis revealed significantly decreased abundance of the house-keeping protein actin in ΔleuB compared to the control strains (Fig 4A). Increased proteolysis in the ΔleuB whole-cell lysates from iron starved mycelia was further confirmed by a biochemical protease activity assay (Fig 4C).
In contrast, analysis of whole-cell lysates using a denaturing lysis buffer (alkaline lysis: 0.2 M NaOH, 0.2% β-mercaptoethanol) did not result in significant differences between ΔleuB, WT and leuBc strains in SDS-PAGE analysis and actin did not show significant differences in abundance (Fig 4B). Taken together, these data suggest that the increased proteolysis in ΔleuB did not occur in vivo but during protein extraction due to the increased protease content of the cell extract. Increased protease activity is often coupled with increased autophagy. To analyze whether lack of LeuB induces autophagy, we tagged the autophagy marker protein Atg8 N-terminally with GFP in ΔleuB and WT genetic backgrounds. As shown in Fig 4D, Western blot analysis indicated that the WT displayed slightly increased cleavage of GFP from GFP-ATG8 during iron starvation compared to iron sufficiency, which indicates slightly increased autophagy [24,25]. In the ΔleuB genetic background, GFP-cleavage was WT-like during iron sufficiency but significantly increased during iron starvation compared to the WT genetic background. Consistently, epi-fluorescence microscopy studies displayed increased accumulation of GFP-ATG8 in vacuoles in ΔleuB compared to WT under iron starvation (S4 Fig). These data collectively demonstrate that lack of LeuB increases not only protease activity but also autophagy.
As supplementation with iron or leucine rescued the growth phenotype of ΔleuB, we investigated if these supplementations affect the protease activity in ΔleuB extracts. As shown in Fig 4E, supplementation with either iron (50 μM FeCl3), leucine (2 mM), or both decreased the protease activity, as illustrated by Western blot analysis of actin. In contrast to ΔleuB, ΔsreA and ΔhapX displayed WT-like protein stability of actin in cell extracts during iron starvation (Fig 4F), demonstrating that these iron regulators are not involved in control of the protease activity found in ΔleuB extracts. The ΔleuBΔhapX mutant strain showed ΔleuB-like proteolysis, while lack of SreA suppressed protease activity caused by lack of LeuB (ΔleuBΔsreA) (Fig 4F). The latter can be explained by the fact that lack of SreA derepresses iron acquisition and thereby increases iron uptake, which resembles the effect of iron supplementation [16]. In agreement, lack of SreA also cured the growth defect of ΔleuB (Fig 3B). Hence, the growth defects largely match with the increased protease activity in cell extracts.
As shown above, supplementation with iron or leucine rescues the ΔleuB growth phenotype and protease activity (Fig 3B and 4E). As iron starvation did not result in increased protease activity in the WT (Fig 4E), we hypothesized that the increased protease activity of ΔleuB extracts is caused by leucine shortage, particularly, as LeuB orthologs were reported to be involved in control of BCAA biosynthesis [19]. To test this hypothesis, we first analyzed expression of leuA, an essential gene for leucine biosynthesis, which contains a CCGN4CGG promoter motif that is bound in vitro with high affinity by LeuB. Transcript levels of leuA were significantly higher in WT during iron sufficiency compared to iron starvation (Fig 4G). This is in line with previous studies showing that expression of leuA is repressed during iron starvation by HapX [18]. Lack of LeuB caused a significant decrease of the leuA transcript levels during iron starvation but did not significantly affect leuA transcript levels during iron sufficiency (Fig 4G), suggesting that LeuB is a transcriptional activator of leuA expression particularly during iron starvation. Similarly, the leuA ortholog leu1 has previously been shown to be regulated by the LeuB ortholog Leu3 in S. cerevisiae [26]. The decreased leuA transcript levels indicate a reduced leucine content in ΔleuB. To further analyze if decreased leucine content increases protease activity, we generated a leuA deletion mutant, ΔleuA. Western blot analysis of actin demonstrated that cellular extracts of ΔleuA mycelia cultured during iron starvation and supplemented with 1 mM leucine, which was the lowest leucine concentration supporting ΔleuA growth, contain high protease activity (degradation of actin), while supplementation with 3–10 mM leucine decreases protease activity (Fig 4H). In contrast, iron supplementation did not rescue the increased protease activity with 1 mM leucine supplementation (Fig 4I). These data indicate that leucine starvation is the major trigger for increased protease activity in ΔleuB during iron starvation. The fact that iron supplementation decreased protease activity in ΔleuB but not ΔleuA extracts indicates that leucine levels are significantly higher in ΔleuB cultured during iron sufficiency compared to iron starvation (Fig 4E and 4I). This is also in agreement with the positive influence of iron on leuA transcript levels in both WT and ΔleuB strains (Fig 4G).
Taken together, previous reports demonstrating control of leuA expression by iron via HapX [18] and the current study characterizing regulation of leuA expression by LeuB and its connection to protease activity during iron starvation compared to iron sufficiency reveal a regulatory network of HapX and LeuB controlling leucine biosynthesis.
In S. cerevisiae, the C-terminus of the LeuB homolog Leu3 was shown to be crucial for function [19]. Based on these data, a series of C-terminal truncations of LeuB was generated to functionally characterize LeuB protein domains (Fig 5A). Notably, all mutated or truncated LeuB gene variants displayed considerable expression at the transcript level based on semi-quantitative RT-PCR analyses (Fig 5B). Truncation of the C-terminal 54 amino acid residues of LeuB (LeuB866) did not significantly affect growth, while truncation of the C-terminal 105 as well as 258 amino acid residues caused a growth phenotype similar to the ΔleuB mutant strain (Fig 5C). These data demonstrated that the C-terminus of LeuB is essential for its function. Replacement of cysteine240 by alanine (LeuBC240A), which is expected to block coordination of zinc in the DNA-binding Zn2Cys6 domain and consequently DNA-binding, caused a growth defect similar to that caused by lack of LeuB. These data underline the importance of DNA-binding for LeuB function, which is in agreement with its function as transcription factor. In contrast, mutations of some randomly chosen amino acid residues (LeuBL717A, LeuBP823A, S833A and LeuBS846A, D854A) did not affect the growth pattern. The mutants carrying leuB gene variants with truncations or site-directed mutations resulting in growth defects also displayed significantly increased proteolytic activity in cell extracts when grown under iron starvation (Fig 5D and 5E). These data underline the link between LeuB dysfunction and the increased cellular protease activity.
To further understand the role of LeuB during iron starvation, chromatin immunoprecipitation followed by massively parallel DNA sequencing (ChIP-seq) was performed to identify genes under direct transcriptional control of LeuB. Therefore, LeuB was C-terminally labeled with the FLAG-tag. The two biological ChIP-seq replicates performed with mycelia cultured during iron starvation identified 1092 and 809 potential target genes, respectively, sharing 773 common targets (Fig 6A and 6B and S1 Dataset). As expected, the ChIP-seq gene set included several genes encoding enzymes involved in BCAA biosynthesis such as leuA, leu2A, leu2B, bat2, ilv2, ilv3, and ilv5 (the BCAA biosynthetic pathway is shown in S1 Fig.) but also genes involved in biosynthesis of amino acids other than BCAA such as lysine (lys1, lysF), aromatic amino acids (aro8, trpB) and proline (pro3). Moreover, this analysis indicated that crucial genes of nitrogen and sulfur metabolism are direct LeuB targets, e.g. nitrogen regulatory protein-encoding areA, meaB and nmrA as well as nitrogen metabolic enzyme-encoding gdhA and glnA as well as sulfur regulator-encoding metR. In agreement with the growth defect of ΔleuB during iron starvation, the ChIP-seq analysis also indicated that several genes involved in iron metabolism are direct LeuB targets, including hapX, genes involved in siderophore biosynthesis (sidA, sidC, sidI, hmg1, argEF), siderophore uptake (mirB, enb1), intracellular siderophore hydrolysis (sidJ, estB), and mitochondrial iron import (mrsA).
In agreement with the ChIP-seq analysis, EMSA and SPR analyses shown above already demonstrated in vitro binding of LeuB to the promoters of leuA, gdhA and hapX (Fig 2). Exemplary EMSA analyses further confirmed in vitro binding of LeuB to CCGN4CGG motif-containing promoter fragments of leu2A, bat2, lysF, areA, metR, mrsA, estB, and sidJ (Fig 6C), which showed in vivo LeuB promoter occupation in the ChIP-seq analysis (Fig 6B).
To further corroborate cross-regulation of BCAA biosynthesis, nitrogen metabolism and iron metabolism by LeuB, we analyzed transcript levels of selected genes in WT and ΔleuB strains grown during iron starvation with and without leucine supplementation by Northern analysis (Fig 6D). Transcripts of leuB were clearly detectable in WT but not ΔleuB confirming the deletion of the leuB gene in this strain. In WT, transcript levels of the BCAA genes leuA, leu2A and ilv5 were clearly decreased by leucine supplementation indicating transcriptional feed-back inhibition of BCAA biosynthesis by leucine. The transcript levels of these three BCAA genes were significantly decreased in ΔleuB compared to WT under growth conditions without leucine supplementation. In agreement with the ChIP data, these results underline that leuA, leu2A and ilv5 are directly transcriptionally activated by LeuB during leucine shortage. The decreased leuA transcript level in ΔleuB compared to WT during iron starvation also matches the Northern blot analysis data shown in Fig 4G. The gdhA transcript level was responsive to leucine supplementation and significantly decreased in ΔleuB (Fig 6D), again confirming gdhA as a direct target of LeuB as shown previously in A. nidulans and S. cerevisiae [20,27]. In contrast to BCAA biosynthetic genes and gdhA, leucine supplementation did not affect transcript levels of hapX, sidA (encoding ornithine monooxygenase, which is essential for biosynthesis of extra-and intracellular siderophores) and mirB (encoding a transporter for uptake of TAFC) in WT (Fig 6D). However, lack of LeuB significantly decreased expression of these genes, which was partly rescued by leucine supplementation. These data strongly suggest that genes important for regulatory adaptation to iron starvation (hapX) and genes involved in SIA (e.g., sidA, mirB) are under direct control of LeuB, which explains the growth defect caused by lack of LeuB. In agreement with these data, production of extracellular siderophores (TAFC) by ΔleuB reached only 43% of the WT (Fig 6E).
Defective adaptation to iron starvation, e.g., by lack of HapX or siderophore biosynthesis, was previously shown to attenuate virulence of A. fumigatus [14,15,17]. Due to the described growth defects, we analyzed the effect of lack of LeuB in the wax moth Galleria mellonella. In this insect model, lack of LeuB resulted in a significant higher survival rate of G. mellonella larvae compared to the WT (p = 0.007) and the complemented strain leuBC (p = 0.029) over a period of 6 days (Fig 6F). WT and the complemented strain exhibited a similar virulence potential, indicated by survival rates with no statistical difference (p = 0.319). Taken together, these data suggest that LeuB is involved in adaptation to the insect host niche.
Here we functionally characterized the transcription factor LeuB in A. fumigatus. Orthologs in S. cerevisiae and A. nidulans were previously found to transcriptionally regulate BCAA biosynthesis and nitrogen metabolism via controlling expression of gdhA [20,27]. In agreement with a similar function in A. fumigatus, lack of LeuB was partially cured by leucine supplementation (Fig 3A and 3B). Moreover, in this study we provide several lines of evidence demonstrating that LeuB cross-regulates BCAA biosynthesis, nitrogen metabolism and iron metabolism: (i) the growth defect caused by lack of LeuB was rescued by supplementation with either leucine or iron (Fig 3B); (ii) similarly, the increased protease activity caused by lack of LeuB was rescued by supplementation with either leucine or iron (Fig 4H and 4E); (iii) genetic interaction studies demonstrated that combining lack of LeuB with lack of the iron-regulatory transcription factor HapX (which is important for adaptation to iron starvation) aggravates the growth defect caused by lack of one of these regulators during iron starvation, while combining lack of LeuB with lack of the iron-regulatory transcription factors SreA (which is important for adaptation to iron excess by repressing iron acquisition) attenuated the growth defect (Fig 3B); (iv) ChIP-seq analysis demonstrated in vivo binding to regulatory sequences of genes involved in BCAA biosynthesis, biosynthesis of amino acids apart from BCAA, nitrogen metabolism, and iron metabolism including iron regulation (HapX) and SIA (Fig 6B); (v) EMSA and SPR analyses confirmed high-affinity in-vitro binding to CCGN4CGG motifs in promoters of genes involved in BCAA biosynthesis and iron metabolism (Figs 2 and 6C), which are phylogenetically conserved in most Aspergillus species (Fig 1 and S1 Table); (vi) Northern analysis demonstrated that lack of LeuB causes downregulation of genes involved in BCAA biosynthesis, nitrogen metabolism and iron metabolism (Fig 6D); and, in agreement, (vii) lack of LeuB resulted in decreased production of the siderophore TAFC (Fig 6E).
A model summarizing the findings of this study, in the light of previous reports, displaying several regulatory feedback loops in BCAA biosynthesis is depicted in Fig 7. The rational for the LeuB-mediated cross-regulation of BCAA biosynthesis, nitrogen metabolism and iron metabolism is most likely based on the requirement of nitrogen and iron for biosynthesis of BCAA. For example, apart from the general nitrogen-requirement for amino acid biosynthesis, glutamate generated by GdhA is required for the last step in BCAA biosynthesis catalyzed by the transaminases Bat1/2 [19]. The iron dependence of BCAA biosynthesis is based on the requirement of iron-sulfur clusters as prosthetic group for two enzymes, mitochondrial dihydroxyacid dehydratase (Ilv3; in contrast to S. cerevisiae, which contains only one dihydroxyacid dehydratase-encoding gene, A. fumigatus encodes three homologs, termed Ilv3A-C [28]) and cytosolic 3-isopropylmalate dehydratase (termed Leu1 in S. cerevisiae and LeuA in A. fumigatus) [19]. Consequently, upregulation of leucine biosynthesis increases the need for iron, which we hypothesize to be the rational for the cross-regulation of BCAA biosynthesis and iron acquisition. In other words, the fact that iron starvation hampers leucine production might be employed to indirectly sense iron starvation via leucine shortage to upregulate both BCAA biosynthesis and iron acquisition. As iron starvation represses BCAA biosynthesis at the enzyme activity level (lack of prosthetic iron sulfur clusters) and via HapX at the regulatory level (HapX represses transcription of leuA and ilv3 during iron starvation [17]), and as BCAA limitation activates LeuB, activation of iron acquisition by LeuB represents a regulatory feedback loop. Remarkably, LeuB transcriptionally activates iron acquisition by direct transcriptional activation of SIA genes (e.g. sidA and mirB) as well as indirectly via transcriptional activation of the iron regulator-encoding hapX. Previous studies revealed another interrelation between BCAA biosynthesis and iron metabolism: during iron starvation, HapX (HapX-Fe) represses transcription of ilv3 and leuA together with numerous other “iron-dependent genes” [17], while iron sufficiency converts HapX into a transcriptional activator (HapX+Fe) of these genes, most likely via HapX sensing of iron-sulfur clusters [18]. Forming another feed-back, activity of the S. cerevisiae LeuB homolog Leu3 has previously been shown to be posttranslationally activated by leucine shortage via the BCAA biosynthesis intermediate α-isoproylmalate (αIPM) [19]. To foster this regulation, the enzyme producing α-isoproylmalate, α-isoproylmalate synthase (termed Leu4/Leu9 in S. cerevisiae and LeuC in A. fumigatus, respectively) is inhibited by leucine at the enzyme activity level. However, the intracellular nuclear localization of Leu3 is not affected by leucine or αIPM levels, suggesting that Leu3 binds constitutively to its regulatory sites in S. cerevisiae [29]. Similarly, the current study revealed that neither leucine nor iron levels affect nuclear localization of LeuB in A. fumigatus (Fig 3C), suggesting that activation of LeuB may be not affected by trafficking of LeuB probably required for the conformation change of LeuB. In addition, studies of mutants in A. nidulans with perturbed levels of αIPM indicated that αIPM regulates LeuB [20,27], which is similar to S. cerevisiae, suggesting evolutionary conservation of this feed-back regulation also existed in Aspergillus spp. Previous studies indicated that αIPM-sensing by S. cerevisiae Leu3 requires the region between amino acid residues 174–773 [19]. This region shows significant similarity in Leu3/LeuB homologs of S. cerevisiae, A. nidulans and A. fumigatus (25% identity/ 43% similarity between A. fumigatus and S. cerevisiae; 79% identity/87% similarity between A. fumigatus and A. nidulans); an alignment of fungal Leu3/LeuB homologs is shown in S5 Fig). These data strongly suggest that A. fumigatus LeuB might be also regulated by αIPM.
Furthermore, the current study revealed that lack of LeuB increases cellular proteolytic activity and autophagy via leucine shortage during iron starvation but not iron sufficiency. Several lines of evidence suggest that this might be mediated by the “target of rapamycin complex 1 (TORC1)”. TORC1 activates cell proliferation and growth during nutrient availability, while lack of nutrient availability (starvation) blocks TORC1 activity resulting in induction of proteasome activity and autophagy to recycle nutrients [30]. Notably, leucine is one of the major signals activating TORC1, thereby blocking activation of the proteasome and autophagy. These recycling pathways then increase the availability of iron and BCAA. In agreement, autophagy was found to be important for fitness of A. fumigatus during metal depletion [31].
The fact that lack of LeuB resulted in a similar growth pattern in A. fumigatus and A. nidulans indicates that this regulatory circuit is conserved at least in A. nidulans (S3C Fig). Moreover, the evolutionary conservation of the CCGN4CGG motifs in most Aspergillus species predicts conservation of this regulatory circuit within this genus. Site-directed mutagenesis and C-terminal truncations of LeuB demonstrated that DNA-binding and the C-terminus are essential for the LeuB functions confirming that LeuB acts indeed as a classical transcription factor (Fig 5A and 5C). Defects in adaptation to iron starvation, e.g., by lack of HapX or siderophore biosynthesis, have previously been shown to attenuate virulence of A. fumigatus [14,15,17]. In agreement, lack of LeuB caused attenuation of virulence in the wax moth model (Fig 6F). Notably, inactivation of leucine biosynthesis by deleting the gene encoding Leu1 (the homolog of A. fumigatus LeuA) was previously shown to attenuate virulence of the basidiomycete yeast Cryptococcus neoformans [32]. Moreover, this study demonstrated that leucine shortage leads to increased abundance of two mitochondrial iron-sulfur cluster proteins (aconitase and the iron-sulfur cluster biosynthetic enzyme Nfu1) during iron starvation but not iron sufficiency, while a cytosolic iron-sulfur cluster protein (Fra1), was not affected. However, this study did neither provide a mechanistic explanation for the link between leucine shortage and regulation of mitochondrial iron-sulfur cluster proteins, nor did it show any link to iron acquisition.
Taken together, our work underline that the BCAA biosynthetic pathway is not only structurally important but also represents a central road junction involved in cross-regulation of amino acid biosynthesis, nitrogen metabolism, iron homeostasis and cellular proliferation.
A. fumigatus strains, used in this study, are summarized in S2 Table. Generally, A. fumigatus strains were grown on minimal medium (MM)[33] containing 1% (w/v) glucose and 70 mM NaNO3 as sole carbon and nitrogen sources, respectively. For iron starvation, iron was omitted in the trace element solution; for increased iron starvation, the iron-specific chelator bathophenanthroline disulfonate (BPS) was added in iron depleted media. Supplementation with iron (FeCl3) and/or leucine was carried out as described in the Figures. Transformants were screened on media containing 200 μg/ml hygromycin B (Shanghai Sangon Co., China). To analyze the phenotype of the mutants, 2 x 103 conidia were point-inoculated on plates. All plates were incubated at 37°C for two days.
All primers used in this study are shown in S3 Table. For the generation of the leuB deletion cassette, the fusion PCR technique was used as described previously [34]. Briefly, approximately 1 kb of the upstream and downstream flanking sequences of the leuB gene were amplified using the primer pairs LeuB P1/P3 and LeuB P4/P6, respectively. The gene pyr4 from plasmid pAL5 was amplified with the primers pyr4 F/R and used to restore pyrG function in the A1160 WT strain. Next, the three aforementioned PCR products were combined and used as template to generate the leuB deletion cassette using the primer pair LeuB P2/P5. This fragment was then used to transform the recipient strain A1160.
To construct the leuA, sreA, hapX and AnleuB deletion cassettes, the same strategy was employed except the use of different selection markers. sreA and hapX were deleted with the hygromycin resistance cassette (hph) and AnleuB was deleted with AfpyrG. For the construction of the AnleuB null mutant, the recipient strain used was TN02A7. To generate ΔleuBΔhapX and ΔleuBΔsreA double mutants in A. fumigatus, leuB was disrupted in the ΔhapX or ΔsreA background respectively.
To reconstitute ΔleuB with a functional copy of the leuB or AnleuB gene, the following strategy was used. First, the fragment of leuB or AnleuB, which includes the native promoter, 5’UTR, gene sequence and 3’UTR, was amplified with the primer pairs ComLeuB F/R and ComAnleuB F/R respectively and then subcloned into the pEASY-Blunt Zero Cloning Vector (TransGen Biotech) according to the manufacturer’s directions, yielding the plasmids pEASY-leuB and pEASY-AnleuB. In a next step, the hph cassette was amplified using the primer pairs hph-SpeI F/R. pEASY-leuB or pEASY-AnleuB and the resistance cassette were digested with SpeI and ligated to integrate the hph gene into the plasmids as selection marker. The plasmids were then used to transform the ΔleuB strain. Transformation of A. fumigatus was performed as described previously [34]. Gene deletions were confirmed by diagnostic PCR and Southern blot, as shown exemplary in S2 Fig.
To generate the GFP-labeled LeuB strain (LN09), approximately 1.5 kb upstream sequence of leuB referred as fragment 1 (except the termination codon) and downstream sequence of leuB referred as fragment 2 (including the termination codon) were amplified using LeuB-gfp P1/P3 and LeuB-gfp P4/P6, respectively. Fragment 3, containing a 5×GA linker, the eGFP sequence and the selection marker AfpyrG was amplified from plasmid pFNO3 using the primer pair gfp-pyrG F/R. These three fragments were mixed and employed as template to generate the leuB-gfp cassette using the fusion PCR technique with the primers LeuB-gfp P2/P5. After the purification of the leuB-gfp cassette, this fragment was used to transform the A1160 strain by homologous replacing original copy of the leuB gene to generate the leuB(p)::LeuB-GFP strain (LN09) which only had one copy of LeuB. To visualize the cell nucleus of the GFP-labeled LeuB strain (LN09), a nuclear localization sequence labeled RFP plasmid were transformed into GFP-labeled LeuB strain to generate the GFP-labeled LeuB and RFP-labeled nucleus strain (LN10).
For the generation of the GFP-labeled Atg8 strain (LN20), Atg8 was labeled with GFP at the N-terminus under the control of the A. nidulans gpdA (AngpdA) promoter. Briefly, the GFP and Atg8 (without ATG) fragments were amplified with the primer pairs GFP F/R and Atg8 F/R, respectively. The resulting fragments were purified and fused by PCR with the primers GFP F and Atg8 R. This GFP-atg8 cassette was then subcloned into the ClaI site of the pBARGPE-1 vector [35], containing the constitutive AngpdA promoter, resulting in the plasmid gpdA(p)-GFP-Atg8. Subsequently, the gpdA(p)-GFP-Atg8 plasmid was ectopically integrated in the genome of A1160 WT strain to generate the strain expressing GFP-labeled Atg8. In this strain, leuB was deleted with the aforementioned construct to receive the ΔleuB strain expressing GFP-labeled Atg8 (LN21).
To constitutively express LeuB with a FLAG-tag, the ectopic integration method was used. Briefly, 5×flag sequence was amplified using primers Flag F/R from plasmid pFA6a-5×FLAG-kanMX6 and the DNA sequence of leuB without stop codon was amplified with LeuB-Flag F/R. These two fragments were then mixed and employed as template to generate the leuB-flag cassette using the primer pair LeuB-Flag F and Flag R. After purification of the PCR products, the fused leuB-flag cassette was subcloned into the ClaI site of pBARGPE-1, yielding the plasmid OE::LeuB-FLAG, and was then used to transform the A1160 strain to generate the FLAG-labeled LeuB strain (LN11).
To visualize the localization of LeuB-GFP, the LN10 strain (LeuB::GFP, RFP-NLS) was grown on coverslips in 3 ml liquid minimal media with or without iron at 37°C for 18 hours. For iron shift experiments, FeCl3 was added to a final concentration of 50 μM and incubated for 1–3 hours. To visualize the localization of LeuB-GFP under the condition with or without leucine, the similar strategy was employed. Images were captured using a Zeiss Axio imager A1 microscope (Zeiss, Jena, Germany) and managed with Adobe Photoshop.
For site-directed mutagenesis, complementary primers harboring the desired mutation in the center position were designed and synthesized. The plasmid pEASY-leuB used for the complementation of ΔleuB was employed as template and amplified with the respective primers, including the desired mutations. The resulting PCR products were digested with DpnI and then transformed into Escherichia coli. All plasmids used for site-directed mutagenesis were sequenced to verify the mutation. For C-terminal truncation, reverse primers, which comprise a stop codon used for truncation, were designed and synthesized. The plasmid pEASY-leuB was used as template and the fragments were amplified using the primer pair LeuB-trunc F and the respective primer with the desired truncation. Purified PCR products were then co-transformed into the ΔleuB strain.
For Semi-qRT-PCR analysis, total RNA was isolated from the frozen mycelium using TRIzol (Roche) as described in the manufacturer’s manual. The digestion of genomic DNA and synthesis of cDNA was performed using HiScript II Q RT SuperMix for qPCR (+gDNA wiper) kit (Vazyme) as described by the supplier. Primer used for semi-qRT-PCR analysis as labelled in S3 Table.
To analyze the production of TAFC, the ΔleuB mutant and WT strains were cultured under iron starvation and the supernatant was separated from the mycelia. Subsequently, the TAFC content of the supernatant was determined by reversed-phase HPLC as described previously [36,37].
Briefly, two exons of LeuB, which encode the Zn2Cys6 domain (exon 1 with 126 and exon 2 with 334 amino acid residues), were amplified with the primers Ex-LeuB P1/P2 and Ex-LeuB P3/P4 respectively and then fused using the fusion PCR technique with the primers Ex-LeuB P1/P4. The fused cassette, which contains a 6×His tag at the 5’ end, was cloned into the NdeI and EcoRI site of pET-30a(+). The resulting plasmid was then used for transformation of E. coli BL21(DE3). The E. coli cells were cultured in lysogeny broth (LB) medium to an optical density of 0.8 at 37°C measured at 600 nm (OD600) and subsequently induced by 1 mM IPTG at 16°C for 12h. Protein purification was performed as previously described using Ni-NTA agarose [38].
The electrophoretic mobility shift assay was performed as described previously [39]. Cy5 labeled DNA probes were prepared as followed. A DNA fragment of the promoter region of different genes containing the conserved CCGN4CGG motif was amplified by PCR using the respective primer pairs (EMSA-gene name F/R). Forward and reverse primers were labeled with an oligonucleotide, refers as primer pEMSA. The purified PCR product was then employed as template to generate the Cy5 labeled DNA probe using the Cy5 labeled primer pEMSA. For site-directed mutagenesis of the hapX probe, extra complementary primers including the desired mutation were designed and named EMSA-MuhapX F/R. Fragments that contain half of the probe sequence were amplified using MuEMSA-hapX F/EMSA-hapX R and MuEMSA-hapX R and EMSA-hapX F, respectively. The purified fragments were then fused using the primers EMSA-hapX F/R to generate the template for the hapX DNA probe with site-directed mutagenesis. For nonspecific competitor or cold probe, 1 μg salmon sperm DNA or a 100-fold non-labeled DNA probe was added. The Cy5-labeled probes were detected with Odyssey machine.
To reduce unspecific binding of the LeuB protein to the SPR-matrix, a cDNA fragment encoding only the DNA binding domain (DBD) of A. fumigatus LeuB was subcloned into the pET-29a vector (Novagen, Germany). The LeuBM50-125YS protein was produced by autoinduction in E. coli BL21 (DE3) cells grown at 25°C in 1 l Overnight Express Instant TB Medium (Novagen, Germany) in the presence of 1 mM Zn(OAc)2 and kanamycin. 25.8 grams wet cells were collected by centrifugation, resuspended in 200 ml lysis buffer (20 mM HEPES, 150 mM NaCl, 10 μM Zn(OAc)2, 5 mM β-mercaptoethanol, 1 mM AEBSF, pH 7.5) and disrupted using an Emulsiflex C5 high pressure homogenizer (Avestin, Germany). The cleared cellular extract was adjusted to pH 7.5, loaded on a SP Sepharose HP column (GE Healthcare, Germany) and eluted with a salt gradient up to 1 M NaCl. The LeuB DBD containing fraction was adjusted to 150 mM NaCl and loaded on a Cellufine Sulfate column (Millipore, Germany), which was previously equilibrated with 20 mM HEPES, 150 mM NaCl, 10 μM Zn(OAc)2, 5 mM β-mercaptoethanol, pH 7.5, followed by elution with a salt gradient up to 1 M NaCl. The peak fractions were concentrated by ammonium sulfate precipitation and redissolved in 20 mM HEPES, 150 mM NaCl, 10 μM Zn(OAc)2, 5 mM β-mercaptoethanol, pH 7.5. The LeuB DBD was then purified by size exclusion chromatography on a Superdex 75 prep grade column (GE Healthcare) using a 20 mM HEPES, 150 mM NaCl, 10 μM Zn(OAc)2, pH 7.5 containing running buffer. The protein was stored in 50% v/v glycerol at -20°C. The purification of LeuB for SPR analysis is shown in S6 Fig.
Real-time analyses were performed on a Biacore T200 system (GE Healthcare) at 25°C. DNA duplexes were produced by annealing complementary 16 bp oligonucleotides using a 5-fold molar excess of the non-biotinylated oligonucleotide. The dsDNA was injected on flow cells of a streptavidin (Sigma)-coated CM3 sensor chip at a flow rate of 10 μl/min until the calculated amount of DNA had been bound that gives a 100 RU maximum LeuB DBD binding capacity. LeuB DBD samples containing 30 μg/ml poly(dAdT) were injected in running buffer (10 mM HEPES pH 7.4, containing 150 mM NaCl, 0.005% (v/v) surfactant P20, 5 mM β-mercaptoethanol and 10 μM ZnCl2) at concentrations from 6.25 to 400 nM. Sample injection and dissociation times were set to 100 and 200 seconds at a flow rate of 30 μl/min. Refractive index errors due to bulk solvent effects were corrected with responses from DNA-free flow cell 1 as well as subtracting blank injections. Kinetic raw data were processed and globally fitted with Scrubber 2.0c (BioLogic Software) using a 1:1 interaction model including a mass transport term.
For whole-cell lysate extraction, mycelia were ground with liquid nitrogen and two buffers, a mild, non-denaturing lysis buffer (50 mM HEPES pH 7.4, 137 mM KCl, 10% glycerol containing, 1% Triton X-100, 1 mM EDTA, 1 μg/ml pepstatin A, 1 μg/ml leupeptin and 1 mM PMSF) and the alkaline lysis buffer (0.2 M NaOH, 0.2% β-mercaptoethanol). For mild, non-denaturing lysis buffer mediated protein isolation, samples were incubated on ice and vortexed for 30 s every 5 min for three times. Cell debris was removed by centrifugation at 13, 000×g and 4°C for 10 min. The protein concentration in the supernatant was measured by Bio-Rad protein assay kit. For alkaline lysis buffer mediated protein isolation, the following strategy was used. Briefly, 20 mg of powdered mycelium were re-suspended in 1 ml lysis buffer. 75 μl of trichloroacetic acid (TCA) were added, the samples were vortexed and incubated on ice for 10 min. After centrifugation at 13,000×g, for 5 min at 4°C, the supernatants were removed and the pellets were heated up to 95°C and vortexed in 100 μl of 1 M Tris and 100 μl of 2×SDS protein sample buffer until complete dissolution. For Western blot analysis, GFP and actin were detected with the anti-GFP mouse monoclonal antibody (Roche, Cat. No. 11 814 460 001), anti-actin antibody (ICN Biomedicals Inc., clone C4), respectively. The detailed Western blotting procedure was described previously [40].
Virulence assays in G. mellonella were carried out according to Fallon et al. [41]. G. mellonella larvae (SAGIP, Italy) were kept in the dark at 18°C before use. Larvae, in groups of 20, were injected through one of the hind pro-legs with 20 μl of IPS (“Insect Physiological Saline”: 150 mM NaCl, 5 mM KCl, 10 mM EDTA, and 30 mM sodium citrate in 0.1 M Tris–HCl, pH 6.9) containing 1 x 107 conidia of the respective strain. Untreated larvae and larvae injected with 20 μl of IPS served as controls. Larvae were incubated at 30°C in the dark and monitored daily up to 6 days. Significance of survival data was evaluated by using Kaplan-Meier survival curves, analyzed with the log-rank (Mantel Cox) test utilizing GraphPad Prism software. Differences were considered significant at p values < 0.05.
For quantification of chymotrypsin-like protease activity in cell extracts, proteins of the WT and ΔleuB strains were isolated with the aforementioned non-denaturing lysis buffer from cultures grown under iron starvation. Protein concentration was measured with the Bio-Rad protein assay kit. To determinate the chymotrypsin-like activity, the degradation of the fluorogenic peptide succinyl-Leu-Leu-Val-Try-7-amido-4- methylcoumarin (Suc LLVY-AMC; 0.167 mg/ml in 100 mM Tris–HCl, pH7.4; excitation 360 nm, emission 460 nm) was detected as described previously [42,43].
For ChIP-seq, the FLAG-labeled LeuB strain (LN11) was cultured under iron starvation for 24 h and cross-linked by addition of 1% formaldehyde for 10 min under shaking (100 rpm) at 37°C. Crosslinking was stopped by adding glycine to a final concentration of 0.125 M and incubated at room temperature for 5 minutes under shaking. Mycelia were washed with pre-cold PBS and collected using vacuum filtration. Subsequently the collected mycelia were frozen with liquid nitrogen. DNA sonication, chromatin immunoprecipitation, DNA purification and ChIP-seq were performed by Bio-tech & Consult (Shanghai) Co. LTD. Peaks of the ChIP-Seq were called using Model-based Analysis for ChIP-Sequencing (MACS2, version 2.1.1.20160309). Peaks calling were done with the ChIP-seq samples and input samples with a q-value cutoff of 5.00e-02. The obtained data were further analyzed to screen the putative target genes that contain the CCGN4CGG motif in the predicted promoter or 5’UTR.
The WT strain with and without leucine supplementation (5 mM) and the ΔleuB strain with leucine supplementation (5 mM) were cultured for 16 hours during iron starvation at 37°C. To compensate for the reduced growth rate and to yield the same biomass formation, the ΔleuB strain was cultured for 18 hours without leucine supplementation. RNA was isolated from the harvested mycelia using TRI Reagent (Sigma-Aldrich, Vienna, Austria) according to the manufacturer's description. For Northern blot analysis, 10 μg of total RNA (2.5 μg for the detection of gdhA RNA levels) were loaded on a 2.2 M formaldehyde agarose gel for electrophoresis and subsequently blotted onto an Amersham Hybond N membrane (GE Healthcare, Vienna, Austria). RNA levels were detected with PCR amplified DIG-labeled probes. Primers used for amplification of the hybridization probes are listed in S3 Table.
|
10.1371/journal.pgen.1003969 | Deletion of a Conserved cis-Element in the Ifng Locus Highlights the Role of Acute Histone Acetylation in Modulating Inducible Gene Transcription | Differentiation-dependent regulation of the Ifng cytokine gene locus in T helper (Th) cells has emerged as an excellent model for functional study of distal elements that control lineage-specific gene expression. We previously identified a cis-regulatory element located 22 kb upstream of the Ifng gene (Conserved Non-coding Sequence -22, or CNS-22) that is a site for recruitment of the transcription factors T-bet, Runx3, NF-κB and STAT4, which act to regulate transcription of the Ifng gene in Th1 cells. Here, we report the generation of mice with a conditional deletion of CNS-22 that has enabled us to define the epigenetic and functional consequences of its absence. Deletion of CNS-22 led to a defect in induction of Ifng by the cytokines IL-12 and IL-18, with a more modest effect on induction via T-cell receptor activation. To better understand how CNS-22 and other Ifng CNSs regulated Ifng transcription in response to these distinct stimuli, we examined activation-dependent changes in epigenetic modifications across the extended Ifng locus in CNS-22-deficient T cells. We demonstrate that in response to both cytokine and TCR driven activation signals, CNS-22 and other Ifng CNSs recruit increased activity of histone acetyl transferases (HATs) that transiently enhance levels of histones H3 and H4 acetylation across the extended Ifng locus. We also demonstrate that activation-responsive increases in histone acetylation levels are directly linked to the ability of Ifng CNSs to acutely enhance Pol II recruitment to the Ifng promoter. Finally, we show that impairment in IL-12+IL-18 dependent induction of Ifng stems from the importance of CNS-22 in coordinating locus-wide levels of histone acetylation in response to these cytokines. These findings identify a role for acute histone acetylation in the enhancer function of distal conserved cis-elements that regulate of Ifng gene expression.
| Differentiation of multipotent naïve T cell precursors into functionally mature effector cells that control different types of immune responses is an excellent model to study lineage-specific regulation of gene expression. A number of cis-regulatory elements have been reported to control expression of the gene that encodes the cytokine IFN-γ which is a signature product of effector T cells of the Th1 lineage. However, none of these elements has been analyzed for effects on gene expression and chromatin remodeling through deletional analysis in the native Ifng gene locus. Here we have generated mice in which a key element previously implicated in control of Ifng gene expression (CNS-22) was conditionally deleted from the genome. Th1 cells in which CNS-22 was deleted had activation-specific deficits in Ifng expression and demonstrated defects in epigenetic changes across the Ifng locus. Mapping epigenetic consequences of CNS-22 deletion led to identification of acute hyperacetylation of histones immediately flanking this and other cis-regulatory elements that was associated with Ifng gene transcription, as well as more global defects in histone acetylation. These findings support a mechanism whereby regulatory sites that have acquired baseline histone acetylation marks during lineage specification undergo acute, activation-dependent increases in histone acetyl transferase activity that enhance transcription of inducible genes.
| Distal regulatory elements including locus control regions, enhancers, silencers and boundary elements play important roles in regulating cell lineage-specific activation and repression of genes [1], [2], [3], [4], [5], [6]. In addition to genome-wide studies to document and classify putative distal regulatory sites, studies on individual gene loci have been instrumental in shaping our understanding of cis element function [7], [8], [9]. Although genes expressed in several cell types including embryonic stem cells (Hox genes), B-lineage cells (immunoglobulin genes) and erythroid cells (globin genes) have emerged as important models to understand eukaryotic transcription, cytokine genes expressed in T-helper cells are particularly attractive models to study lineage specific regulation. Primary human and murine naïve Th cells can be readily isolated in large numbers and be differentiated into functionally and transcriptionally distinct Th cells as exemplified by Th1, Th2, Th17, and T-regulatory (Treg) cell subsets [10], [11], [12]. In particular, genes that encode Th2 cytokines, comprised of the Il4, Il13 and Il5 genes and the Ifng gene transcribed in Th1 cells have emerged as key models to the study lineage-appropriate gene expression [8], [12] [13], [14].
The importance of distal elements in regulating expression of human and mouse genes that encode IFN-γ was first recognized in mice transgenic for a bacterial artificial chromosome (BAC) that encompassed ∼190 kb flanking the human IFNG gene, which, unlike transgenes that contained more limited flanking sequence, conferred lineage-specific expression of human IFN-γ in mouse Th1 cells [15], [16]. Subsequently, we reported a murine Ifng-Thy1.1 BAC reporter transgene that spanned ∼160 kb surrounding Ifng, which also demonstrated lineage- and activation-specific expression [17], [18], suggesting that distal elements required for lineage specific expression of Ifng were contained in this region. Based on recruitment of CTCF and Rad21 (a cohesin), the IFNG and Ifng loci are predicted to extend from −63 to +119 kb [19] and −70 kb to +66 kb [20], respectively. Within these boundary elements, at least nine conserved non-coding sequences (CNS) have been identified based on the high degree of sequence conservation at these sites in multiple mammalian species [2], [3].
Using ChIP-qPCR and promoter-reporter assays, a subset of these CNSs was probed for trans-factor binding and histone modifications in early studies [18], [21], [22], [23], [24]. More recently, DNase-chip [25] and DNase-seq [20] have been employed to map chromatin conformation of the extended Ifng locus in multiple T cell lineages. In parallel, analyses of trans factor recruitment to these cis elements have facilitated their further functional mapping. T-bet [18], [22], [23], STAT4 [26], [27], Runx3 [28] and members of the NF-κB [29] and NFAT [23] families of transcription factors have been demonstrated to interact with cis elements across the Ifng locus [13].
To date, the functions of four Ifng/IFNG CNSs have been examined by deletional analyses in the context of Ifng or IFNG BAC transgenes [18], [30]. Deletion of the human homolog of Ifng CNS-34 (IFNG CNS-30) from a 190 kb human IFNG transgene resulted in impaired expression of IFNG in T cells but not NK cells [30] suggesting that one or more of these regulatory sequences may have lineage-specific functions. In our own studies, deletion of CNS-22 from the Ifng-Thy1.1 reporter transgene led to nearly complete ablation of Thy1.1 reporter expression in both T cells and NK cells [18]. In more recent studies using IFNG BAC transgenes, human homologs of CNSs −6 and +17–19 were also shown to be important enhancers of IFNG transcription [31]. CNS-22 is of particular interest, since it is hypersensitive not only in Th1 cells, but also in naïve precursors, suggesting that CNS-22 might act as a principal site for recruitment of “pioneer” trans factors that initiate remodeling of the Ifng locus [25]. Further, CNS-22 remains hypersensitive to DNase I in Th2 and Th17 cells despite repression of Ifng expression in these lineages, suggesting that CNS-22 may be involved in Ifng transcriptional silencing in addition to its role in transcriptional activation [25], [29]. Taken together with the profound effect of selective CNS-22 deletion on BAC reporter expression, these findings led us to speculate that this element could be an important node for directing chromatin remodeling of the Ifng locus [18].
Therefore, we generated mice targeted for conditional deletion of CNS-22 in the endogenous Ifng loci (Ifng.CNS-22fl/fl) to enable mapping of epigenetic modifications not previously possible using BAC transgenic mice. Here, we report the functional consequences of the deletion of CNS-22 on epigenetic remodeling and gene expression of the Ifng locus in naïve and differentiated T cells. We find that deletion of CNS-22 in Th1 cells results in greater impairment of Ifng transcription in response to IL-12 plus IL-18 than that induced by TCR dependent signaling. This is associated with a defect in the deposition of histone acetylation marks on nucleosomes immediately flanking CNS-22 as well as those distributed distally across the Ifng locus. These findings identify a previously unappreciated role for activation-induced modulation of HAT activity in driving cytokine gene transcription.
Ifng.CNS-22fl/fl mice were generated using a targeting construct in which 391 bp corresponding to CNS-22 in the Ifng locus was flanked by loxp sites to enable Cre-mediated excision (Fig. S1A, B) [18], [32], [33]. CNS-22 resides in a broad DNase I hypersensitive (HS) site that encompasses CNS-22 at the 3′ end (Fig. S1C). Cre-mediated excision of this element deletes several evolutionarily conserved trans-factor binding sequences, including sites that recruit T-bet, STAT4, RelA and Runx3 (Fig. 1B) [18], [28], [29]. Ifng.CNS-22 was deleted in the germline by crosses with EIIa.Cre mice, such that all cells, including T cells, were CNS-22–deficient (henceforth referred to as CNS-22−/− mice). Phenotypically, CNS-22−/− mice were indistinguishable from littermate controls. Numbers of CD4+ and CD8+ T cells in the periphery were comparable to wildtype controls (unpublished observations).
To examine the impact of CNS-22 deletion on Ifng gene expression, naïve CD4+ T cells from OT-II transgenic WT and CNS-22−/− mice were differentiated ex vivo in the presence of a low or high concentration of IL-12 and the expression of IFN-γ induced by restimulation with IL-12+IL-18 or TCR signaling was examined (Fig. S2A). Irrespective of the concentration of IL-12 used, Th1 cells generated from CNS-22−/− mice were significantly impaired in their expression of IFN-γ in response to IL-12+IL-18 restimulation, whereas a deficit in IFN-γ expression in response to TCR stimulation was only apparent for cells differentiated with the low concentration of IL-12 (Figs. 2A and S2A). In contrast to cells restimulated with IL-12+IL-18, which showed a similar deficit in IFN-γ expression whether activated on day 3 or day 5 of differentiation, the impact of CNS-22 deletion on impaired TCR-driven induction of IFN-γ was more pronounced in Th1 cells restimulated on day 3 (Figs. 2A and S2A, and data not shown). The impairment of Ifng transcription was not due to alterations in the expression of key trans factors, as there were no significant differences in expression of Tbx21 or Runx3 in CNS-22-deficient T cells (Fig. S2B). IL-12+IL-18 driven induction of Ifng was also considerably impaired in Tc1 cells and in NK cells from in CNS-22−/− mice (Fig. 2B). Impairment of Ifng expression was also observed in vivo for CNS-22-deficient T cells responding to infection with Listeria monocytogenes, an intracellular bacterial pathogen that induces a type 1 immune response (Figs. S3A–C).
By comparing the kinetics of induction of Ifng transcripts in Th1 cells, we found that early transcription of Ifng in response to IL-12+IL-18 was significantly compromised in the absence of CNS-22, while there was no significant impairment in response to TCR signaling (Fig. 2C), reinforcing the observation that induction of transcription via the TCR signaling pathway was less dependent on CNS-22 function. Finally, Th1 cells derived from CNS-22+/− mice showed an intermediate, copy number–dependent impairment in IL-12+IL-18 driven Ifng gene transcription (unpublished observations), demonstrating that CNS-22 plays an obligatory role in integrating activating signals downstream of IL-12 and IL-18 receptors [29]. Taken together with our previous findings that CNS-22 recruits STAT4 and NF-κB downstream of the IL-12 and IL-18 receptors, respectively [13], whereas TCR signaling does not activate STAT4, these results suggest that CNS-22 is more important as a STAT4-dependent enhancer for acute activation of Ifng transcription, although they do not exclude a critical contribution for CNS-22 as node for epigenetic modifications of the Ifng locus during the development of immune cells.
Our previous studies demonstrated that deletion of CNS-22 in the context of the Ifng-Thy1.1 reporter transgenic locus resulted in nearly complete ablation of Thy1.1 reporter expression in Th1, Tc1 and NK cells [13]. This led us to hypothesize that CNS-22 was essential for orchestrating remodeling of the Ifng locus in this transgene in addition to its role as a key enhancer (ref. [18], and see Fig. 1A). While we confirm here the importance of CNS-22 as a positive modulator of Ifng transcription (Fig. 2), the less pronounced impairment of Ifng transcription found in CNS-22−/− Th1 cells suggested that deletion of CNS-22 from endogenous Ifng loci might not have as fundamental a role in regulating chromatin accessibility during Th1 cell development as speculated (see below). Nevertheless, IL-12 dependent acquisition of Ifng competence was considerably delayed in CNS-22−/− CD4+ T cells (Fig. 2A). Since we had previously documented that CNS-22 was permissive in naïve CD4+ T cells (ref. [13], and see also Fig. 3A), we hypothesized that CNS-22 might play an obligatory role in epigenetic remodeling the Ifng locus prior to Th1 differentiation.
To examine this, we performed DNase-chip and ChIP-qPCR-based analysis of histone 3 lysine 4 methylation (H3K4) to assess the epigenetic status of the extended Ifng locus in CNS-22-deficient naïve CD4+ T cells and Th1 cells (Fig. 3). In agreement with our previous study [25], we found that several key distal elements in the Ifng locus are hypersensitive (HS) to DNase I in naïve WT CD4+ T cells (Fig. 3A). Notably, naïve cells from CNS-22−/− mice showed marked reduction or a lack of hypersensitivity at most sites identified in WT cells, including those at CNSs +17–19, +30, +46, +54 and +66 (Fig. 3A). In contrast, DNase I HS sites at the upstream CTCF binding insulator element (−70) and CNS+40 arise in a CNS-22–independent fashion, indicating that even prior to exposure to lineage-specifying signals, multiple nodes appear poised to initiate reorganization of the extended Ifng locus. We also observed a prominent HS peak immediately upstream of CNS-22 in naïve CNS-22−/− CD4+ T cells (Fig. 3A), although this was much less pronounced following Th1 differentiation. While the basis for this is unclear, it is plausible that some functions of CNS-22 are not compromised by deletion of the core regulatory element.
In accord with the functional analyses of Ifng expression in CNS-22−/− Th1 cells, defects in locus-wide remodeling apparent in naïve CNS-22–deficient CD4+ T cells were largely reversed upon Th1 differentiation (Fig. 3A). Specifically, the pattern of DNAse HS at regulatory elements downstream of the Ifng gene was nearly indistinguishable from WT Th1 cells, indicating that differentiation-dependent remodeling of the Ifng locus is largely independent of CNS-22.
To complement DNase-chip analyses, we also carried out ChIP-qPCR to evaluate H3K4 mono-, di- and tri methylation status is Th1 cells derived from either WT or CNS-22−/− mice (Fig. 3B). In parallel to the DNase I HS results, deposition of permissive H3K4 methylation marks in naïve CD4+ T cells was significantly impaired in the absence of CNS-22, particularly at CNS-34 and non-conserved site −28 (Fig. 3B). However, levels of H3K4 methylation was comparable at both these sites in CNS-22–deficient and WT Th1 cells (Fig. 3C), indicating that CNS-22 was dispensable for differentiation-dependent remodeling of the Ifng locus in Th1 cells. Thus, the effect of CNS-22 deletion in the context of the endogenous Ifng was less pronounced than in the context of an Ifng BAC transgenic locus we reported previously (Figs. 1A and 4A).
We hypothesized that the observed differences in the contribution of CNS-22 to Ifng transcription from the endogenous or transgenic loci might be attributed to the absence of key distal regulatory sequences in the BAC-transgene. The BAC-transgenic reporter included 60 kb upstream and 100 kb downstream of the Ifng gene and consequently lacked a more recently defined insulator/chromatin looping sequence that is located 70 kb upstream of Ifng (Fig. 4A) [19], [20]. In addition, it was also possible that some distal elements that regulate Ifng lie greater than 100 kb downstream of Ifng. The first clue that this was likely came from ChIP-chip analyses carried out in our previous studies to identify STAT4 and RelA binding enhancer sequences [29]. We discovered at least one prominent STAT4 binding site that was 159 kb downstream of the Ifng gene (Fig. 4B). This site was hypersensitive to DNase I in Th1 cells, but not in Th2 or Th17 cells suggesting that it functioned as a STAT4 responsive module in Th1 cells.
To further explore the possibility that potential regulatory elements may reside greater than 100 kb downstream of Ifng and therefore would be excluded from the transgene, we carried out ChIP-chip analyses to map recruitment of CTCF and a cohesion family member Smc3. We identified at least three prominent CTCF and Smc3 binding sites +105, +151 and +342 kb downstream of Ifng (Fig. 4B). At all three sites, we observed Th1-specific acquisition of histone H4 Lysine-12 acetylation (H4K12ac) marks indicating that these elements are functional in Th1 cells (Fig. 4B). Therefore, we speculate that exclusion of these downstream regulatory sequences and the CTCF binding site located −70 kb upstream of Ifng in our previous BAC-transgenic studies may have magnified the role of CNS-22 in coordinating Ifng expression. Nonetheless, by deleting CNS-22 at the endogenous locus, we demonstrate here that CNS-22 is an obligate enhancer that is primarily important in regulating IL-12+IL-18 dependent induction of Ifng in Th1, Tc1 and NK cells.
Th1/Tc1 specific transcriptional competence of the Ifng gene is marked not only by acquisition of H3K4 mono, di and tri- methylation marks, but also acetylation of several lysine residues on histones H3 and H4 across the extended Ifng locus [18]. Acetylation of histone residues H4K16, H3K4 and H4K12 is present not only at promoters, but also throughout the extended loci of transcribed genes [34]. More recently, acetylation of H3K27 has been correlated with lineage-specific activity of enhancers [35]. We initially examined these modifications and others (unpublished observations), and chose to focus on H4K12 acetylation due to the greater efficiency, specificity and dynamic range of chromatin immunoprecipitation observed for this particular histone modification. Acquisition of H4K12ac marks at CNS-22 and globally across the extended Ifng locus was specific to differentiated Th1 cells (Fig. 4) [18], [26]. Similarly, acquisition of lineage-specific H4K12ac marks correlated with transcriptional competence at other T lineage-specific gene loci, including the Il17a/Il17f and Il4-Il13-Il5 gene clusters in differentiated Th17 and Th2 cells, respectively (Fig. S4).
In a previous study we noted that H4 acetylation at multiple distal sites in the Ifng locus, including CNS-22, was increased by signals that led to acute induction of Ifng transcription in both Th1 and Tc1 cells [18]. Notably, the activation dependent change in H4 acetylation at these non-conserved sites was much greater than changes observed at Ifng CNSs [18]. We therefore speculated that inducible acquisition of histone acetylation marks at these distal non-conserved sites might be linked to activation-induced transcription of Ifng, and that CNS-22 and other CNSs in the Ifng locus might act as nucleation sites for enhanced recruitment or activation of HATs that decorate the histones of neighboring nucleosomes. If true, induction of Ifng transcription in Th1 and Tc1 cells would be predicted to correlate with acute increases in histone acetylation at sites immediately flanking CNSs across the Ifng locus. Accordingly, deletion of CNS-22 would be predicted to impair acquisition of these inducible acetylation marks.
To test this, we evaluated levels of histone H4 acetylation in the immediate vicinity of CNS-22 in resting and activated Th1 cells. In response to both IL-12+IL-18 and TCR driven activation signals, a prominent increase in levels of H4 acetylation at sites within ∼1 kb flanking CNS-22 (−22.9, −22.4, −21.7) was observed (Fig. S5). We also observed similar activation-induced acetylation at Ifng CNSs −34 and +46 (see Fig. 5, below, and data not shown). We subsequently performed ChIP-qPCR experiments in Th1 cells derived from CNS-22−/− mice to examine whether activation-driven acquisition of acetylation marks at −22.9, −22.4 and −21.7 was dependent on CNS-22. While resting levels of H4 acetylation at −22.9, −22.4 and −21.7 were unaffected by deletion of CNS-22, acute hyperacetylation at these sites in response to both IL-12+IL-18 and TCR signaling was significantly impaired (Figs. S5B, C). These results led us to speculate that the ability of CNS-22 to acutely recruit HATs in response to external stimuli is linked to its ability to enhance Ifng transcription. This also led us to ask two further questions. Firstly, is activation-driven acquisition of H4 acetylation marks unique to CNS-22, or is this a common mechanism employed by multiple Ifng enhancers? Secondly, since Th1 cells generated from CNS-22−/− mice show a more prominent defect in IL-12+IL-18 driven activation signals, but acquisition of H4 acetylation near CNS-22 was significantly impaired in response to both IL-12+IL-18 and TCR reactivation signals, does CNS-22 regulate histone acetylation only within its immediate vicinity or does it influence histone acetylation levels at more distal sites as well? To address these two questions, we have carried out ChIP-chip experiments to examine H4K12 acetylation in resting and activated Th1 cells generated from WT and CNS-22−/− mice.
In resting Th1 cells, H4K12ac marks were confined to the regions in immediate proximity to the CNSs themselves and deletion of CNS-22 did not significantly alter acquisition of these marks (Fig. 5A). Upon activation with either IL-12+IL-18 or with anti-CD3+anti-CD28, there was robust, extensive deposition of acute acetylation marks that extended beyond the CNSs to more distal, non-conserved sequences (Fig. 5A). As predicted by conventional ChIP-qPCR, acquisition of acetylation marks in the immediate vicinity of CNS-22 was absolutely dependent upon CNS-22 irrespective of the mode of activation (Fig. S5). However, normalization and comparison on H4K12ac levels between WT and CNS-22−/− Th1 cells revealed a much more prominent defect in IL-12+IL-18 dependent acquisition of H4K12ac marks extending across the Ifng locus (Fig. 5B). Specifically, in response to TCR driven acquisition of H4K12ac marks, deletion of CNS-22 impaired acquisition of H4K12ac in proximity to CNS-22 as well as in enhancers upstream of CNS-22 (CNSs −54 and −34), but had little to no effect on deposition of acetylation marks at CNSs downstream of the Ifng gene (Fig. 5B). In contrast, in CNS-22–deficient Th1 cells IL-12+IL-18 dependent acquisition of H4K12ac was globally altered across the Ifng locus such that defects in acquisition of this epigenetic mark were evident at multiple CNSs downstream of the Ifng gene (Fig. 5B). Therefore, the greater defect in acute acquisition of H4K12ac marks in response to IL-12+IL-18 might account for the greater impairment of IL-12+IL-18 dependent induction of Ifng compared to that observed following TCR driven induction.
The positive correlation between accumulation of histone acetylation marks that accompany lineage specification and induction of gene transcription is well established. More recently, the recruitment of bromodomain-containing HATs, particularly p300 and CBP, to distal regulatory sequences has been used to identify lineage-specific enhancers [36], [37]. In view of the role played by CNS-22 in the HAT-mediated modification of the Ifng locus, we interrogated its association with p300 (Fig. S6). Binding of p300 to CNS-22 and other CNSs across the Ifng locus was evident in resting and activated Th1 cells, and with few exceptions the level of p300 binding — whether to CNS-22 or other CNSs across the locus — was not substantially altered by activation. This finding, coupled with the marked impairment in acquisition of H4K12 acetylation in CNS-22–deficient Th1 cells following IL-12+IL-18 stimulation, led us to speculate that activation-induced enhancement of HAT activity at CNS-22 resulted in increased recruitment of RNA Pol II to the Ifng promoter. To examine this, we compared recruitment of RNA Pol II to the Ifng gene in resting Th1 cells and Th1 cells activated by IL-12+IL-18 or TCR stimulation (Fig. 5C). In response to IL-12+IL-18, recruitment of RNA Pol II to the Ifng promoter, first exon and first intron was significantly impaired in the absence of CNS-22, suggesting that the ability of CNS-22 to enhance Ifng transcription is linked to its ability to up-regulate HAT activity. In contrast, while CNS-22–deficient Th1 cells showed a modest decrement in Pol II recruitment to the promoter and first exon of Ifng in response to TCR signaling, this did not achieve statistical significance (Fig. 5C). Together, these results suggest that trans-factor recruitment to CNS-22 downstream of IL-12+IL-18 signaling induced increased acetylation that was largely independent of enhanced p300 recruitment. These results further suggest that inducible acetylation at distal enhancers that regulate Ifng transcription is directly linked to their ability to modulate recruitment of RNA Pol II to the Ifng gene in response to external stimuli.
CNS-22 was previously identified as an important cis-regulatory element hypothesized to have a central role in epigenetic remodeling of the Ifng locus during lineage-specific Th1, Tc1 and NK cell differentiation [13], [18]. In the current study, CNS-22–deficient mice were generated to enable study of the consequences of deletion of this element for epigenetic remodeling of the endogenous Ifng locus. Here we identify CNS-22 as critical element for early remodeling of the Ifng locus in naïve T cells and establish its importance as an enhancer for optimal Ifng transcription in Th1, Tc1 and NK cells. However, we find that the epigenetic consequences of the deletion of CNS-22 during lineage specification are limited to circumscribed effects on critical upstream regulatory elements surrounding CNS-22, such that remodeling of regulatory elements downstream of the Ifng gene occurs largely independently of CNS-22 in Th1 or Tc1 cells. Unexpectedly, fine mapping of the epigenetic consequences of CNS-22 deletion led to the finding that distal cis-acting elements activate inducible gene transcription through hyperacetylation of nucleosomes that flank trans-factor binding core elements of enhancers. This supports a model of eukaryotic enhancer function wherein the activity and/or composition of HAT complexes loaded onto core enhancer elements during lineage-specific differentiation are rapidly modulated in concert with activation-induced trans-factor recruitment to effect increased gene transcription.
Speculation that CNS-22 is a central node for permissive remodeling of the extended Ifng locus arose from our previous studies wherein deletion of CNS-22 from an Ifng-Thy1.1 BAC reporter transgene caused nearly complete ablation of reporter expression in Th1, Tc1 and NK cells [13], [18]. In the present study, deletion of CNS-22 from the endogenous Ifng locus led to a substantial defect in induction of Ifng, albeit less pronounced than the same deletion from the BAC transgene. Although several factors could account for the disparities observed, the exclusion from our Ifng BAC transgene of boundary elements located −70 kb upstream and the new potential boundary element identified herein located 342 kb downstream of the Ifng gene leads us to speculate that absence of these architectural elements might have compromised the efficiency of approximation of the CNS-22 enhancer to the core promoter mediated by a CTCF-cohesin–dependent mechanism. This might well have resulted in an exaggerated loss of Ifng transcription upon deletion of CNS-22 from the BAC transgene [17], [18]. Also missing from the BAC transgene were hypersensitive sites identified herein located +105, +151 and +159 kb downstream of the transcriptional state site of the Ifng gene. All three of these elements are likely to be functional, as the first two co-recruit CTCF and Smc3 while the site at +159 recruits STAT4. Any or all of these could contribute to the exaggerated expression defect observed on deletion of CNS-22 in from the BAC-transgene. Finally, it should be noted that Tmevp3, a gene that encodes a long intergenic non-coding RNA (lincRNA) that maps to ∼+60 kb to +120 kb downstream of the Ifng gene was excluded in the Ifng-Thy1.1 BAC transgene [38]. Although two recent studies document that the lincRNA encoded by Tmevp3 acts in trans to recruit H3K4 methyltransferases to the Ifng locus, it is possible, although we think it unlikely, that exclusion of Tmevp3 in the Ifng-Thy1.1 transgene might have compromised transcriptional activation upon deletion of CNS-22 in this context [39], [40]. Suffice it to say that further studies will be required to precisely define the basis for the observed expression disparities and should be informative.
Importantly, deletion of CNS-22 from endogenous Ifng alleles enabled analyses of developmental stage-specific effects on chromatin organization not previously possible using the BAC transgenic model. Thus, while CNS-22−/− naïve CD4+ T cells demonstrated compromised development of several HS sites present in WT cells, including those at CNSs +17–19 and +46, as well as H3K4 methylation of CNS-34, Th1 differentiation largely overrode these defects. This indicates that developmentally regulated HS sites that dictate low-level Ifng expression competence characteristic of naïve CD4+ T cells are dependent upon the presence CNS-22, exposing an important role for CNS-22 in early remodeling of the locus. However, differentiation that confers high-level Ifng transcriptional activity in Th1 and Tc1 cells proceeds such that a number of key HS sites, primarily those downstream of the Ifng gene, can be remodeled independently of CNS-22. Thus, although CNS-22 remains an important element for differentiation-dependent remodeling of the extended Ifng locus, its influence is not global, rather it is more local, consistent with the previously proposed organization of the Ifng locus into upstream and downstream regulatory domains coordinated by recruitment of CTCF boundary elements to the intronic CTCF element within the Ifng gene [19], [20].
CNS-22 recruits at least four key transcription factors to activate Ifng transcription: STAT4 [29], T-bet [18], Runx3 [28] and RelA [29]. We predict that defects in Ifng induction in CNS-22−/− T and NK cells stems from deletion of the STAT4 binding site in CNS-22. STAT4 is activated downstream of the IL-12 receptor and plays an essential role in IL-12 dependent polarization of naïve CD4+ T cells to IFN-γ-competent Th1 cells [41], [42]. IL-12+IL-18 dependent induction of Ifng is also absolutely dependent on activation of STAT4 [43]. In addition to the impaired IL-12+IL-18 driven induction of Ifng in CNS-22−/− Th1 cells, we also document that acquisition of competency of the Ifng locus is considerably delayed in the absence of CNS-22. At least three STAT4 recruiting modules, CNSs −22, +40 and +46 are accessible in naïve CD4+ T cells [25]. During the course of Th1 differentiation, additional STAT4 recruiting modules at the Ifng promoter, CNSs −34, +30 and +54 also become accessible [29]. It is likely that sustained IL-12 signaling combined with CNS+40- and +46-dependent remodeling of other STAT4 binding modules compensate for the absence of CNS-22. Nonetheless, subsequent deficiency in IL-12+IL-18 dependent induction of Ifng highlights the importance of the STAT4 binding site within CNS-22. Further studies to mutate and evaluate individual transcription factor binding sites within CNS-22 would provide new insights into the roles of individual trans factor binding sites within CNS-22.
An important finding of this study is the identification of a link between activation-dependent recruitment of HAT activity and enhancer function. Several previous studies have linked acetylation of histones to differentiation-dependent activation of gene transcription, including the recently discovered association between H3K27ac and enhancer actions [35]. In addition, at least two bromodomain-containing HATs — p300 [36], [37] and CBP [36] — have been demonstrated to regulate lineage-specific activation of enhancers. Here, we establish for the first time a clear relationship between enhancer actions and acute increases in HAT activity by demonstrating that CNS-22 not only dictates activation-induced acquisition of H4K12ac marks in its immediate vicinity, but also regulates acquisition of H4K12ac marks at multiple distal sites. Together, these results suggest that CNSs across the Ifng locus interact with each other and coordinate activation-driven recruitment of HATs via long-range interactions. Moreover, by evaluating IL-12+IL-18 dependent RNA Pol II recruitment in CNS-22−/− Th1 cells, we demonstrate that impaired hyperacetylation compromises the ability of CNS-22 deficient Th1 cells to recruit RNA Pol II to the proximal Ifng promoter in response to IL-12+IL-18. Notably, the marked CNS-22–dependent increases in histone acetylation that followed IL-12+IL-18 signaling, and to a lesser extent TCR-mediated signaling, proceeded without substantive changes in the levels of p300 binding at these elements. This is consistent with a role for acute trans-factor-dependent recruitment of other HATs and/or activation of pre-loaded p300 HAT complexes to increase histone acetylation and Pol II loading on the promoter, and will require further study.
In summary, studies herein have delineated a role for CNS-22 in regulating epigenetic changes that control chromatin remodeling and transcriptional competence of the Ifng locus prior to, during and subsequent to lineage specification. CNS-22, while essential for locus remodeling in naïve T cells and optimal Ifng transcription in mature effector T cells, has a more restricted function in developmentally driven remodeling of the Ifng locus than previously thought, with its principal effects being exerted on regulatory elements bounded by the upstream and intronic CTCF elements thought to be important in approximating these upstream elements to the proximal promoter. Further, analysis of T cells from CNS-22−/− mice has uncovered a previously unappreciated role of HATs in modulating actions of eukaryotic enhancers for induction of high-level transcription that is characteristic of cytokine genes. Our findings suggest that recruitment and activation of HATs at CNS-22 and other distal enhancers in the Ifng locus occurs in two stages: the first, an initial wave to induce basal acetylation of enhancer-associated histones that confers receptiveness to a second, subsequent wave of trans-factor–induced hyperacetylation that is linked to high-level Ifng transcription. Although further studies will be necessary, these findings support a model in which a conserved regulatory module, such as CNS-22, contains a core enhancer that includes sites for trans-factor binding contingent upon differentiation-dependent nucleosomal clearing or remodeling, whereas the flanking nucleosomes constitute part of an extended enhanceosome that is less evolutionarily constrained. While whole genome analyses of the epigenome have rapidly advanced our understanding of differentiation-dependent alterations in chromatin remodeling and identified important epigenetic markers that have enabled identification of potential enhancers, detailed locus-specific analyses of the type exemplified herein for the Ifng locus will be important to complement our understanding of cis-element function moving forward.
All animal studies were conducted in accordance with guidelines and oversight of the institutional animal use and care committee of University of Alabama at Birmingham.
C57BL/6 and OT-II TCR transgenic mice were purchased from Jackson Laboratory and/or bred at the University of Alabama at Birmingham. Generation of CNS-22−/− mice is described in the supplement. Primers used for PCR-based screening have been previously described [18]. Antibodies against H3K4me1,2,3 (04-791), H3K4me2 (17-677), H3K4me3 (17-678), Pan H4 (08-858), H4 acetyl (06-598), H3K9ac (17-658) and H4K12ac (07-595) and RNA polymerase II (05-952) were purchased from Millipore (Billlerica, MA). Antibodies against p300 (sc-584, sc-585) were purchased from Santa Cruz Biotechnology (Santa Cruz, CA). All primers and probes were synthesized by IDT (Coralville, IA).
FSClo, CD62Lhi, CD44lo cells were purified from CD4+ T cells isolated by positive selection from spleen and pooled lymph nodes. For generation of Th cells, CD4+ T cells were isolated by positive selection from spleen and pooled lymph nodes. Differentiation of Th1, Th2, Th17 and Tc1 cells was performed as previously described [25], [29], [44], [45]. Positive selection was carried out using CD4-DYNAL beads (Invitrogen) or anti-PE microbeads (Miltenyi). Reactivation of cells for ChIP has been previously described [29]. Briefly, Th1, Tc1 or NK cells were activated with 10 ng/ml rIL-12 and 25 ng/ml rIL-18. For TCR restimulation, anti-CD3 antibody was diluted to 10 µg/ml in phosphate buffered saline (PBS, 150 mM NaCl, 0.02M Phosphate) and coated overnight at 4°C. The following day the plates were was washed twice with PBS and the media was supplemented with 5 µg/ml of anti-CD28 antibody.
Cells were reactivated for 4 hours in the presence of GolgiStop (BD Biosciences; San Jose, CA) as per manufacturer's recommendations and stained with fluorescent-labeled antibodies against CD4, CD8, IL-4 and IFN-γ using the Cytofix/Cytperm kit (BD Biosciences). Dead cells were excluded by staining with LIVE/DEAD fixable stain kits (Invitrogen; Carlsbad, CA). Samples were acquired on an LSRII flow cytometer and analyzed using FlowJo software (Treestar Inc.; Ashland, OR).
Protocols employed for ChIP, ChIP-chip DNase-chip have been previously described [25], [29]. Primer sequences that were not previously published [25] are available upon request. We employed a slightly modified ChIP protocol for p300 ChIP. Cells were dounced in 25 mM Hepes (pH 7.8), 1.5 mM MgCl2, 10 mM KCl, 0.1% NP-40, 1× complete protease inhibitor cocktail (Roche). Nuclei were isolated, resuspended in 0.1× SDS lysis buffer (Millipore) diluted in 1× ChIP dilution buffer (Millipore). Samples were sonicated in and subject to ChIP as previously described using ChIP assay kit (Millipore). For ChIP-chip, samples and inputs were amplified using WGA2 kit (Sigma), labeled and hybridized to custom-designed microarrays (Roche-Nimblegen). A previously described algorithm, ACME was employed for peak calling and identifying enriched regions in the ChIP-chip datasets [46].
RNA was isolated using RNeasy kit (Qiagen), subject to DNAfree treatment (Applied Biosystems) to remove any contaminating DNA. RNA was then reverse-transcribed using Superscript III cDNA synthesis kit (Invitrogen) and transcript levels were normalized against housekeeping gene β2-microglobulin.
|
10.1371/journal.pcbi.1002571 | Virtual Patients and Sensitivity Analysis of the Guyton Model of Blood Pressure Regulation: Towards Individualized Models of Whole-Body Physiology | Mathematical models that integrate multi-scale physiological data can offer insight into physiological and pathophysiological function, and may eventually assist in individualized predictive medicine. We present a methodology for performing systematic analyses of multi-parameter interactions in such complex, multi-scale models. Human physiology models are often based on or inspired by Arthur Guyton's whole-body circulatory regulation model. Despite the significance of this model, it has not been the subject of a systematic and comprehensive sensitivity study. Therefore, we use this model as a case study for our methodology. Our analysis of the Guyton model reveals how the multitude of model parameters combine to affect the model dynamics, and how interesting combinations of parameters may be identified. It also includes a “virtual population” from which “virtual individuals” can be chosen, on the basis of exhibiting conditions similar to those of a real-world patient. This lays the groundwork for using the Guyton model for in silico exploration of pathophysiological states and treatment strategies. The results presented here illustrate several potential uses for the entire dataset of sensitivity results and the “virtual individuals” that we have generated, which are included in the supplementary material. More generally, the presented methodology is applicable to modern, more complex multi-scale physiological models.
| As our understanding of the human body at all scales increases, the construction of a “Virtual Physiological Human” is becoming more feasible and will be an important step towards individualized diagnosis and treatment. As computational models increase in complexity to reflect this growth in understanding, analysis of these models becomes ever more complex. We present a methodology for systematically analysing the interactions between parameters and outputs of such complicated models, using the Guyton model of circulatory regulation as a case study. This model remains a landmark achievement that contributed to the development of our current understanding of blood pressure control, and we present the first comprehensive sensitivity analysis of this model. Effects of varying each parameter are explored over randomized simulations; our analysis demonstrates how to use these results to infer relationships between model parameters and the predicted physiological behaviour. Understanding these relationships is of the utmost importance for developing an optimal treatment strategy for individual patients. These results provide new insight into the multi-level interactions in the cardiovascular-renal system and will be useful to researchers wishing to use the model in pathophysiological or pharmacological settings. This methodology is applicable to current and future physiological models of arbitrary complexity.
| Global initiatives such as the IUPS Physiome project [1], [2] and the Virtual Physiological Human (VPH) project [3], [4] aim to quantitatively understand human physiology at all levels from gene to organism through the use of mathematical modelling. Beyond a certain degree of complexity, the combinatorial number of interactions between the parts of a system can defy intuition and present severe challenges [5]. Mathematical models are appropriate tools for developing our understanding of human physiology, since they can be used to represent and analyse the combinatorial number of interactions between parameters in a rigorous and systematic manner [6].
In short, computational models that integrate physiological data from multiple scales (both physical and temporal) provide a framework for understanding the maintenance of biological entities under physiological and pathological conditions. One significant application for such models is individualized predictive medicine; i. e., tailoring models to the characteristics of an individual patient and predicting the outcomes of different treatment strategies, to help select the best strategy for that patient [3].
Many challenges must be overcome before a truly integrative model of human physiology can be constructed [6], [7]. Gaining a real quantitative understanding of the phenotypic variation in humans as a function of genes and environment in a mechanistic sense (i. e., understanding the genotype-phenotype map in both the explanatory and predictive sense [8]–[10]) is a tremendous challenge that awaits technological, conceptual and methodological breakthroughs [11].
A number of models have already been used to develop insight into aspects of human physiology [12]–[22], many of which have their origin in the control-theory model of whole-body circulatory regulation introduced by Guyton et al. in 1972 [23], [24]. Although it was published over 30 years ago, the Guyton model remains a landmark achievement, and with the rise in the last 10 years of systems physiology, it has attracted renewed attention [18], [25]–[27] and even generated some recent controversy [24], [28]–[30]. It was the first “whole-body”, integrated mathematical model of a physiological system; it was particularly instrumental in identifying and exploring the relationship between blood pressure and sodium balance, and in demonstrating the key role of the kidney in long-term regulation of blood pressure. It allows for the dynamic simulation of systemic circulation, arterial pressure, and body fluid regulation, including short- and long-term regulations.
In previous work, the Guyton models were modularized and re-implemented in Fortran, C++ (M2SL [31]), and Simulink [14]. Furthermore, since one of the main limitations of the early Guyton models is the low-resolution description of most of their constituting modules, a framework was built to allow replacement of the original sub-modules by new versions at a higher temporal or spatial resolution [32]; e. g., a pulsatile heart was introduced to treat systolic and diastolic blood pressures instead of only mean blood pressure [33], and a detailed model of the renin-angiotensin-aldosterone system (RAAS) has also been integrated [34]. That work was also linked to efforts in the European VPH via two Exemplar Projects, one of which used our modular reimplementation of the Guyton model as the basic set of “bricks” for a collaborative core-modeling environment for multi-organ physiology modeling [13], [14], and the other uses the Guyton model as a demonstrator for the tagging of parameters and variables with a set of reference ontologies common to databases of high-throughput genomic and proteomic data [35]. Collaborators in the Physiome/VPH community have also carried out XML markup of the individual modules of the Guyton model in CellML (http://models.cellml.org/workspace/guyton_2008), thus providing precious documentation of its structure and content.
The analysis and results presented here arose naturally from this body of work. Our motivation was to develop a methodology for systematically exploring the ramified implications of multi-parameter interactions in multi-scale physiological models. We present such a methodology, which incorporates the elementary effects technique introduced by Morris [36]. As a case study, we present a sensitivity analysis of the 1992 version of the Guyton model [24], [30], [37], with a focus on the multiple interactions involved in blood pressure regulation. This version was never published, but represents a more complete and modern understanding of the cardiovascular system [24], [30] (e. g., the inclusion of ANP [37]), and it is the version that members of the Guyton group have continued to use. Indeed, such a model, grounded in decades of hands-on experimental work and built with an engineer's approach to control processes, should serve as a rigourous platform for discovery of non-intuitively obvious relationships. However, despite the significance of the Guyton model, the dynamics of the model have not yet been analysed in a systematic and comprehensive study.
The results provide valuable information about the inter-dependencies of parameter effects on the model outputs, thus providing direction for future physiologically-applicable sensitivity studies of the effects of changes to multiple parameters. These results also lay the groundwork for the use of multi-parameter models such as the Guyton model in systematic in silico exploration of possible new drug effects, hypotheses about multiple perturbations leading to disease states, and alternative treatment strategies.
An additional outcome is the production of a virtual population, where each virtual individual is characterized by its set of parameter values (loosely analogous to genotypes) and the associated outputs (“phenotypes”). Note that the parameters of the Guyton model are in fact lower-level phenotypes, but as models continue to span larger physical and temporal scales, model parameters will approach the genotype level [38], [39]. A given real-world patient can be associated with one or more of these virtual individuals on the basis of clinically identifiable parameters or dynamics (e. g., mean arterial pressure, serum total protein, cardiac output, heart rate). Searching an existing collection of simulations in this manner avoids the inherent pitfalls in solving the inverse problem of (uniquely) identifying unknown model parameters and states from clinical observations [40]. Thus, the construction of a comprehensive virtual population could prove a useful tool in future efforts to provide efficient, individualized health-care.
Note that beyond the methodology itself, the results presented in this manuscript also serve to demonstrate some of the uses to which the complete set of elementary effects and virtual individuals may be applied. We provide tables of all of the resulting output in the supplementary material (Dataset S1), which we hope will be of use in physiological, pathophysiological and clinical settings.
The Guyton model comprises parameters and output variables. We restricted our analysis to parameters {} and output variables {} (as indicated in Equation 1, Table 1, and documented in Tables S1 and S2), focusing on those parameters with direct physiological relevance and ignoring parameters with no clear physiological interpretation (such as curve-fitting coefficients). The distribution of these 96 parameters was: 32 cardiac, 21 renal, 16 autoregulation, 16 hormonal, 11 local circulation, and 4 thirst-related. To determine which parameters have significant effects on each of the model outputs, we computed the elementary effects of each parameter using a modification of the formula defined by Morris [36], which we now detail.
The influence of the parameter on some output is defined by Equation 2. Assuming that each parameter is normalized to the unit interval that , the region of experimentation––the portion of the parameter space that will be explored––is a regular -dimensional -level grid , where each parameter may take on values from (Equation 3). For each parameter in turn, a perturbation is chosen (Equation 4). For positive perturbations , we restrict (Equation 5), and for negative perturbations , we restrict (Equation 6), so that .
For any point (where or ), Morris defined the elementary effect of as per Equation 7. In our analysis of the Guyton model, we chose to normalize the elementary effects with respect to (Equation 8) rather than by , which is always a fixed percentage of the range of .
Each elementary effect was calculated times, where each of the simulations was performed with randomized values for all parameters , in order to obtain a representative sample of the magnitude of the effect.
Given a set of values for a single elementary effect , it is important to note that the mean and variance of this set provide different insights into the nature of the relationship between the parameter and the output . The mean indicates the sensitivity of to , while the variance indicates the influence of other parameters on this relationship or the non-linearity of the effect.
For each random input vector , a simulation was started with the default initial state () and progressed for four weeks of simulation (), at which time a pseudo-steady state had either been reached, or a new random input vector was chosen and the simulation was restarted.
The parameter under investigation () was then incremented (or decremented) by and the simulation continued for another four weeks of simulation time, after which either a new pseudo-steady state had been reached, or a new random input vector was generated and the simulation was restarted.
Throughout the simulations, a number of output variables were monitored to ensure that they remained within physiological bounds (i. e., that the virtual individuals remained “alive”, see Table 2). If these bounds were violated during a simulation, the simulation was discarded and a new input vector was chosen.
Since the system is highly non-linear, the effects of a perturbation in the parameter on the output variables vary over time, so elementary effects were calculated at times (Equation 9) and the state of the model () was recorded at times (Equation 10). The parameters for this mass-simulation process are given in Table 3.
This entailed simulations to obtain estimates ( with positive perturbations and with negative perturbations) of the elementary effect of each parameter on each output. In each simulation, two distinct points in parameter space ( before and after the perturbation) resulted in two steady states. Each input vector and steady state can be viewed as a virtual individual; that is, a virtual human whose “genotype” is described by the input vector and whose “phenotype” is described by the resulting steady-state outputs. Thus, the sensitivity analysis simulations also produced a virtual population of virtual individuals. We detail how this virtual population may be of use for diagnosis and exploration of treatment strategies for real-world patients in our discussion.
The results presented here are intended as a demonstration of the analyses that are possible with the complete set of simulation results, which are given in the supplementary material, namely: means and deviations of each elementary effect at each time ; correlations between each parameter and each variable at each time and at time (steady-state) for both the normotensive and hypertensive sub-populations; and correlations between each elementary effect and each variable at all times .
The distribution of mean arterial pressure (MAP) in the virtual population is shown in Figure 1. Given the Common Terminology Criteria for Adverse Events v4.03 (CTCAE) [41] definition of Stage 1 hypertension (systolic BP 140–159 mmHg or diastolic BP 90–99 mmHg) and the formula for estimating mean arterial pressure from systolic and diastolic pressures ( [42]), we define hypertensive individuals as those with . Approximately one third of the virtual individuals were normotensive and two-thirds were hypertensive (see Table 4; using an older definition of hypertension (160/95) leads to 41% of virtual individuals being classified as hypertensive). These proportions differ by less than in the pre-perturbation and post-perturbation steady states, and near-identical proportions were also observed in earlier sets of simulations (not presented here). This demonstrates that the prevalence of hypertension in the virtual population is robust and not dependent on the choice of random input vectors .
Also shown in Figure 1 are gamma and chi-squared distributions that have been fitted to the probability density. The chi-square distribution is a special case of the gamma distribution where the scale parameter is . While the distributions provide reasonable fits, they both underestimate the density for and overestimate the density for .
The analysis of the simulations investigated several aspects of the resulting data. First, we present the sensitivity analysis of the elementary effects on key output variables. The purpose was to determine which parameters induced consistent effects when perturbed, and how these effects are influenced by other parameters. Second, the correlations between parameters and key variables were considered, to identify relationships between the outputs and fixed parameter values. Note that while the elementary effects are shown to vary over time, the correlations remained essentially constant. These correlations were then compared across the normotensive and hypertensive sub-populations, to detect any differences in these relationships between these two populations. Finally, several generalized linear models (GLMs) [43], [44] were evaluated for their predictive power of identifying hypertensive individuals based on a select number of parameters.
Definitions of all model 96 parameters and 276 variables are tabulated Tables S1 and S2. More complete results are tabulated in Dataset S1.
Given our interest in the development of hypertension, we focus the discussion here on variables directly related to blood pressure. For example, Figure 2a shows the most significant elementary effects (at each time ) on three such variables: the mean arterial pressure (MAP), the cardiac output (QAO), and the rate of urine production (VUD).
The single largest effect on all three variables is that of HYL (the quantity of interstitial hyaluronic acid), which affects the tissue hydrostatic and osmotic pressures. This effect is only observed one hour after the perturbation is made. That is, a change in hyaluronic acid takes more than one minute to have an effect, and the effect is no longer evident after 24 hours. The large deviations (significantly larger than those of any other parameter) demonstrate that the effects of HYL are highly non-linear. We will demonstrate how to identify interesting multi-parameter effects, using HYL as an example. To clearly depict the other elementary effects, they are shown in Figure 2b without the effects of HYL. The largest steady-state elementary effects at are shown in Figure 3. The complete table of elementary effects is available in the supplementary material.
Correlations were calculated between each parameter and each output variable at each time , using the Spearman rank-correlation [62]. A rank-correlation method was chosen because such methods are sensitive to any near-monotonic relationship and do not assume that the data is normally distributed. The correlations showed negligible variance () over these times, in contrast to the elementary effects presented earlier. This is because the correlations are sensitive to the absolute value of the parameter, while the elementary effects are sensitive to the influence of a perturbation and not the absolute value. Significant correlations are shown in Figure 5 for the same three variables (MAP, QAO and VUD) whose elementary effects were presented in Figure 2.
Consider the correlations with MAP; the most-highly correlated parameters () are CPR, AARK, EARK, GFLC and HM6, all of which also exhibit significant elementary effects on MAP. As noted earlier, all of these parameters affect glomerular filtration: AARK, EARK and GFLC are all related to physical properties of the glomerulus, while CPR and HM6 affect the driving pressure gradient for ultrafiltration.
In contrast, the parameters most-highly correlated with QAO () are HM6, OMM, CPR, EARK, NID, O2M and RTPPR (the effect of glomerular oncotic pressure on renal tissue oncotic pressure). RTPPR was not seen to exert a significant elementary effect on QAO, but it shows a higher correlation with QAO than do AARK, ANUM, GFLC and LPPR, all of which exerted significant steady-state effects on QAO. Three of these parameters––HM6, OMM and O2M––are directly related to oxygen supply and utilization in the body, whilst CPR and NID affect both the plasma volume and renal filtration, EARK also affects renal filtration, and RTPPR affects tubular reabsorption.
The parameters most-highly correlated with VUD () are NID, CPR, RTPPR, POR, AARK and KID. As was the case for QAO, RTPPR does not exert a significant elementary effect on VUD, but demonstrates higher correlation with VUD than do ANCSNS, ANUM, EARK and GFLC, all of which exhibit significant steady-state effects on VUD. All of these parameters, except for POR, are directly related to renal filtration and reabsorption, while POR modulates the vasoconstrictor effect on blood-flow autoregulation across rapid, intermediate and long-term timescales.
One parameter, CPR, is notable for being highly correlated with all three output variables MAP, QAO and VUD. In particular, CPR has a correlation of with MAP; the only other correlation greater than is that between HM6 and QAO (). This parameter is the critical plasma protein concentration for protein destruction in the liver, which affects the colloid oncotic pressure in the vasculature. The direct effects of this parameter include the rate of glomerular filtration and the rate of capillary leakage. These observations demonstrate that the Guyton model reflects the importance of renal filtration and colloid oncotic pressure to overall haemodynamic regulation [45], [46], [54], [55].
The virtual individuals were divided into normotensive and hypertensive sub-populations based on their mean arterial pressure, as illustrated in Table 4. The probability densities of each parameter and variable were compared across these populations, as were the correlations between the model parameters and the output variables. The probability densities revealed observable differences between the populations (Figure 6), both in the model parameters (e. g., CPR) and output variables (e. g., AAR). Note that the two probability densities shown here for CPR are markedly more distinct than when CPR was classified based on the elementary effect of HYL (not shown).
However, obvious differences were observed for very few parameters, all of which had already been highlighted in the sensitivity and correlation analyses.
Correlations between parameters and variables were then compared between the two populations; some results are shown in Figure 7. The colour-coded regions of each graph represent different relationships between the correlations: green indicates a decreased correlation in the hypertensives; blue indicates an increased correlation in the hypertensives; and red indicates that the correlation has switched sign between the two populations.
The correlations with MAP in the hypertensive population are systematically larger than those in the normotensive population (Figure 7a), which supports the notion that arterial pressure regulation has been reduced in the hypertensive population. However, the correlations with QAO show no such relationship (Figure 7b) with the sole exception of EARK. This suggests that the regulation of cardiac output has not been reduced in the hypertensive population, and that a change in cardiac output is neither a cause nor symptom of the hypertension that is observed in the virtual population, which reflects Guyton's explanation of arterial hypertension being fundamentally a renal pathology [23], [24], [56].
When correlations with blood volume are considered (Figure 7c), the parameters with the largest increases in correlation (ANCSNS, ANUM, ANY) are all related to the effects of angiotensin on arterial resistance and venous volume. Parameters with decreased correlation in the hypertensive population include NID, VV9 and CV (venous compliance). The logical inference is that angiotensin is playing a more significant role in regulating the blood volume in the hypertensive individuals than in the normotensive individuals. Angiotensin plays a role in the activation of the RAAS [56], [63], [64], which increases salt and water retention in the kidney [65]–[67] and raises the “set-point” arterial pressure that the kidney will maintain [50], and these effects are incorporated into the Guyton model. More recent studies have also revealed angiotensin's roles in hypertension via oxidative stress [68]–[70] and inflammatory vascular injury [71], [72], but these phenomena are not included in the Guyton model.
The correlations with urine production (Figure 7d) reveal changes in only a few parameters. The decreased correlation with RTPPR indicates that glomerular oncotic pressure has a smaller effect on tubular reabsorption in the hypertensive population. Of the parameters with increased correlations, AARK and POR are directly related to blood-flow autoregulation and vasoconstriction, and CPR affects the plasma colloid oncotic pressure, which affects the plasma volume and the driving pressure gradient for glomerular filtration. This leads us to conclude that the urine production in the hypertensive population is more sensitive to blood-flow autoregulation and plasma colloid oncotic pressure.
The large virtual population that has been assembled here () can be used not just to analyse relationships between model parameters and outputs, but also to derive and evaluate classifiers for predicting particular phenotypes in virtual individuals. Since hypertension places a heavy burden on health-care systems around the world, and blood pressure regulation is the chief focus of the Guyton model, the most obvious phenotype to predict is hypertension.
The virtual population was divided in two: a randomly-chosen training set of the population size, and the remainder of the population served as an evaluation set. A generalized linear model (GLM) [43], [44] with a binomial distribution function was fitted to the training set to predict whether or not each individual was hypertensive (i. e., ). A minimal GLM was then selected by step-wise reduction of the original GLM with Akaike's information criterion (AIC) [73], resulting in a 30-parameter classifier.
This classifier was then evaluated on the evaluation set (i. e., the rest of the virtual population), shown in Figure 8a, and demonstrated a high degree of accuracy. The sensitivity of the classifier to each of the 30 parameters is shown in Figure 8b. This list of parameters closely resembles those parameters most-highly correlated with mean arterial pressure (Figure 5).
But no matter how accurately this classifier can predict hypertension in the virtual population, one should not conclude that it will be of practical use for predicting hypertension in real-world patients. The classifier is a function of model parameters, many of which are not physiologically derived or measurable. In order to feasibly use such a classifier with real-world patients, the model parameters must be restricted to those that are readily identifiable and measurable in human beings.
Of the parameters listed in Figure 8b, we assume that CPR and LPPR can be estimated from blood tests and that the values of the renal filtration parameters AARK, EARK and GFLC could possibly be estimated from whole-body glomerular filtration rate (GFR) (or, more invasively, from a biopsy). NID can be estimated from the person's diet. The resulting classifier (“Renal+Liver” in Figure 8a, coefficients given in Table 5) predicts hypertension on the basis of these parameters (see Table 6) and suffers from a modest loss of predictive power in comparison to the optimal classifier. It can correctly identify of the hypertensive virtual individuals with a false-positive rate, in comparison to the optimal false-positive rate of . Further restricting the parameters to either solely liver-related or kidney-related (Table 6) significantly reduces the predictive power of the classifiers.
The Guyton model was constructed and refined over many years, and has been validated against a large amount of experimental data [23], [24]. However, many simplifications were necessary in order to permit simulated experiments under the computational resources that were available at the time [24], and the model does not incorporate recent advances in our understanding of the cardiovascular system. Thus, our results will tend to highlight the underlying assumptions and limitations of the Guyton model, rather than physiological phenomena. Indeed, one of the goals of this study was to provide sufficient data (in the supplementary material) to allow researchers to identify whether the Guyton model is sufficiently detailed for specific physiological applications. More recent models have incorporated greater levels of detail for individual organs [12], [74] or for the whole body [16], [19], and a comparison between the Guyton model and these newer models can illustrate the suitability of the Guyton model for clinical applications. Of course, the methodology we employed can be applied to these modern, more detailed models.
Here we present a brief comparison of the Guyton model to the human renal/body fluid model of Uttamsingh et al. [74], which was validated against several sets of experimental data. The result of ingestion of either hypotonic and hypertonic fluid in the Guyton model (shown in Figure 9) produces similar effects on the urine flow rate to that seen in the model of Uttamsingh et al. However, in response to the infusion of hypertonic saline (0.91 g of sodium chloride per kg of body weight, over a period of 65 minutes for a “normal human of 70 kg”) urine flow in the Guyton model increases at a slower rate, plateaus at a lower rate and eventually returns to the baseline level, while urine flow in [74] plateaus at twice the baseline and better matches the experimental data [75].
Larger variation between the two models is observed when aldosterone is increased four-fold, in order to simulate the administration of deoxycorticosterone acetate (DOCA), a mineralocorticoid with similar effects to those of aldosterone [74]. The model of Uttamsingh et al. demonstrates gradual increases in extra-cellular fluid volume (1 L) and mean arterial pressure (10 mmHg), and a rapid drop in sodium excretion in response to the elevated aldosterone level, followed by a slow rise to match the rate of intake. The Guyton model, as shown in Figure 10, produces different behaviour. The extra-cellular fluid volume rises briefly and then gradually decreases until it is 0.1 L below the baseline (Figure 10a) and mean arterial pressure rapidly rises by 10 mmHg and then gradually increases by a further 2 mmHg (Figure 10b). Sodium excretion (Figure 10d) drops rapidly in the first 2 hours, then rises rapidly and overshoots in the following 6 hours, before equilibrating after 24 hours have elapsed. The Uttamsingh et al. model again matches the experimental data [76] better than the Guyton model (e. g., it reproduces the “escape” phenomenon, where the rate of sodium excretion eventually rises to match the increased rate of intake). However, the limited time-resolution (at most one data point every 24 hours) makes a precise comparison impossible. Indeed, with the exception of the extra-cellular fluid volume, the behaviour of the Guyton model also provides a reasonable fit to the data.
The differences highlighted here between the Guyton model and the model of Uttamsingh et al. are certainly due in part to the lower level of detail in the renal portion of the Guyton model, but the Guyton model also includes a more complete cardiovascular model, which would necessarily alter the dynamics produced in response to a chronic increase in aldosterone load. Thus, these observations may indicate a shortcoming in the Guyton model, but further analysis is required before a definitive statement can be made. These results highlight, however, the need to identify portions of the Guyton model that must be refined to replicate experimental data more recent than those used to originally validate the model. We discuss refinement of the Guyton model in the following section.
In our analysis we perturbed a single parameter in each simulaton (although each parameter was perturbed 1000 times, each simulation with a different set of randomly-selected parameter values). Perturbation of multiple parameters would yield a wealth of additional information, but without any guidance the only recourse would be to exhaustively search every combination of parameters, for perturbations. Instead, with the results presented here one can select one parameter () for perturbation and additionally perturb only those parameters that are significantly correlated with the effect of (as per our brief example: “Multi-parameter effects: accounting for the variance in HYL”).
Given the population of virtual individuals that was presented here, an obvious and desirable application is to draw comparisons between subsets of this population and a given real-world patient. That is, given some observations of a real-world patient, we can select those virtual individuals who best match these observations and see whether one can draw conclusions about the condition of the real-world patient based on the long-term dynamics of the selected virtual individuals.
Beyond using virtual populations merely as a reference for the current and ongoing condition of real-world patients who receive no intervention, ongoing refinements of the Guyton model may ultimately support individualized health-care and individualized medicine. The application of mathematical models to individualized medicine would necessarily involve integrating detailed models of physiology, pharmacokinetics and pharmacodynamics. Current efforts on this front include the BIMBO project [77].
Development of chronic diseases such as cardiovascular disease is a complex process that involves environmental and cultural factors shared by the individuals living in the same geographical area, as well as ageing, genetic and disease determinants. Hunter et al. [3] have emphasized the need for diagnostic workflows on the prediction of risk that integrates the influence of both population and patient-specific information in support of tailored interventions aiming at optimizing diagnosis and treatment planning and monitoring.
Researchers of the BIMBO project have defined a modeling approach to estimate the public health impact, in terms of the reduction in the number of cardiovascular deaths (CVD), of administering blood pressure lowering drugs to a virtual population of patients [77]. That virtual population [77] (distinct from the virtual population presented here) reproduces the demographic composition as well as the cardiovascular risk factor profiles of a country population, each virtual individual being characterized by a number of features allowing estimation of CVD risk and treatment efficacy. The individuals eligible for treatment could be selected from their computed CVD risk over a fixed threshold and by having blood pressure in excess of 140/90 mmHg. The authors used a simplified approach where treatment effect was represented by the relative risk, which was assumed to be constant over time and among different individuals, to estimate the public health impact of BP lowering drugs [77].
The work presented here illustrates the value of using population information to predict the success of treatment strategies, whilst also moving towards a more ambitious goal: taking into account the individual genetic backgrounds and pathophysiological profiles. This would contribute to the delivery of individualized healthcare, by optimizing the impact of treatments for both the individual patient and at the population level. Future challenges include the development of more sophisticated effect models [78], such that relative-risks and odds ratios depend on individual characteristics which affect the pharmacokinetic and/or pharmacodynamic parts of the model [79]. Realization of these goals would represent a significant step towards personalizing anti-hypertensive treatment.
The implications of pharmacogenetic parameters on drug efficacy have been explored in the context of diuretic treatment for blood pressure [80]–[82]. One candidate for the identification of responders to thiazide diuretics is the polymorphic gene coding the cytoskeleton protein -adducin, whose mutant form has been associated with an increased rate of sodium reabsorption [83], [84], elevated blood pressure [85], [86], salt-sensitivity [87] and increased risk of cardiovascular events [88]. The same associations first documented in Caucasian populations [84], [87] have not been reported in all other populations, with contradictory evidence from studies in Chinese, African American and Japanese populations [89], suggesting the role of additional factors in mediating the effects attributed to the -adducin polymorphism. But before rejecting the hypothesis of a pharmacogenetic effect of the -adducin variant, a number of epistatic interactions and environmental influences contained in the virtual population characteristics (e. g., different degrees of RAAS activation in response to salt consumption) could be explored through physiological modeling.
With regard to the diagnosis and treatment of hypertension, a practical model would predict the effects of the various diuretics and other drugs that are commonly administered to ameliorate hypertension. This would allow the model predictions to be directly compared to clinical studies such as INDANA [90]. To this end, refinements are being incorporated into the original Guyton model [32] as part of the SAPHIR project [13], such as a detailed model of the RAAS [34]. The culmination of these efforts will result in a richly-detailed and more accurate model of renal autoregulation being incorporated into the Guyton model, providing a platform for pharmacological predictions that may assist in the diagnosis and treatment of hypertension [77].
We have presented a sensitivity analysis of the Guyton model of human physiology (1992 version), which examined the elementary effects of each parameter over a range of timescales and the correlations between model parameters and key output variables. We also demonstrated how interesting multi-parameter combinations can be identified, and how this can highlight shortcomings in the model.
A pool of simulations with randomized parameters (analagous to genetic variants) was generated for this analysis, forming a diverse virtual population of virtual individuals from which representative subsets can be drawn to match characteristics of individual real-world patients. The population was divided into normotensive and hypertensive sub-populations, and a 6-parameter linear classifier was shown to have good predictive power for identifying hypertensive virtual individuals, based on parameters that are feasible to estimate in vivo.
Work is currently underway on comparing these results to real-world patient data from clinical studies of the effect of Avastin on hypertension in cancer patients [91], [92]. About half of the patients develop hypertension in response to Avastin, and are also the most likely to experience a remission. The analysis will aim to identify whether any of the elementary effects or correlations presented in this manuscript are evident in real-world patients, and to evaluate the use of the virtual population in selecting regions of the parameter space of the Guyton model that correspond to the characteristics of a real-world patient. This exploratory project is at a preliminary stage and no results can be presented at this time.
The methodology we have presented here and applied to the Guyton model is generic in that it can be applied to any mathematical model of sufficient complexity. As physiological models encompass larger and larger scales, both spatially and temporally, this methodology should prove beneficial in elucidating the subtle interactions between model parameters in these complex models.
Such an effort is required to evaluate the clinical suitability of using the Guyton model to assist in providing individualized predictive medicine, as per the goals of both the IUPS Physiome and the Virtual Physiological Human projects.
|
10.1371/journal.pntd.0002312 | Exploring Gender Dimensions of Treatment Programmes for Neglected Tropical Diseases in Uganda | Gender remains a recognized but relatively unexamined aspect of the potential challenges for treatment programmes for Neglected Tropical Diseases (NTDs). We sought to explore the role of gender in access to treatment in the Uganda National Neglected Tropical Disease Control Programme.
Quantitative and qualitative data was collected in eight villages in Buyende and Kamuli districts, Eastern Uganda. Quantitative data on the number of persons treated by age and gender was identified from treatment registers in each village. Qualitative data was collected through semi-structured interviews with sub-county supervisors, participant observation and from focus group discussions with community leaders, community medicine distributors (CMDs), men, women who were pregnant or breastfeeding at the time of mass-treatment, and adolescent males and females. Findings include the following: (i) treatment registers are often incomplete making it difficult to obtain accurate estimates of the number of persons treated; (ii) males face more barriers to accessing treatment than women due to occupational roles which keep them away from households or villages for long periods, and males may be more distrustful of treatment; (iii) CMDs may be unaware of which medicines are safe for pregnant and breastfeeding women, resulting in women missing beneficial treatments.
Findings highlight the need to improve community-level training in drug distribution which should include gender-specific issues and guidelines for treating pregnant and breastfeeding women. Accurate age and sex disaggregated measures of the number of community members who swallow the medicines are also needed to ensure proper monitoring and evaluation of treatment programmes.
| This study explored gender-related factors that may influence community member access and adherence to treatment programmes for NTDs in Uganda. A large number of previous studies have identified community-based mass-treatment programmes as an effective strategy to treat affected populations. However, limited evidence is available to discuss challenges to treatment access, adherence, delivery and monitoring at community level. Quantitative data from treatment registers suggested that men were less likely to access treatment than women in at least two villages. It also revealed difficulties in community-based monitoring of the programmes, creating challenges in ascertaining how many persons are able to access the programme. Qualitative data collected from district health workers, community leaders, community medicine distributors and community members suggested that socio-behavioural and structural barriers to treatment access may be present for both genders. Results of the study identify gender-based challenges to treatment access that should be considered in planning, implementing and evaluating national treatment programmes for NTDs.
| Neglected Tropical Diseases (NTDs) are a group of parasitic, viral and bacterial diseases that affect at least a billion people worldwide [1], [2]. Predominantly seen in rural and underserved communities in Africa, the Middle East and Southeast Asia, NTDs can pose significant health implications for both male and female populations. While social and occupational roles may place men at an increased risk of acquiring certain NTDs, such as schistosomiasis [3], NTDs also have significant implications for female populations [4]. Schistosomiasis can cause pregnancy complications [4], [5], [6] and an increased risk of HIV transmission in women especially due to the presence of genital lesions [7], [8]. In addition, stigma from NTDs such as onchocerciasis can significantly affect female populations by causing disfigurement, which may in turn affect marriage prospects [9].
A recent increase in international advocacy and subsequent funding has led to the establishment of national programmes to treat NTDs in Africa, Asia and Latin America. The current strategies to treat NTDs through these national programmes have largely focused on mass drug administration (MDA), either through school-based treatment of children between the ages of 5–14 or through community-based treatment programmes [10], [11]. In the latter, community medicine distributors (CMDs) from Village Health Teams (VHTs) are selected by community members, who are then trained to distribute the drugs to the endemic community in conjunction with health education [10], [11]. Community treatment usually involves either a house-to-house distribution strategy, where CMDs visit each household in the community to distribute medicines, or centralized distribution, where community members gather in a central location and receive treatment from the CMD on specific treatment days [10].
Although these strategies are considered successful in treating affected populations, there is currently very little information on how gender may influence knowledge, perceptions and access to NTD treatment programs. Three studies [12], [13], [14] were identified which provided quantitative gender-stratified data on adherence to a community MDA programme. Clemmons et al [12] and Brieger et al [13] focussed on community directed treatment with ivermectin (CDTI) for onchocerciasis control, which follows a slightly different and more inclusive model of community participation than Uganda's main current MDA strategies of implementing either in April or October as directed by the Ministry of Health. In CDTI, community members select their own drug distributors and choose the time of year in which the distribution takes place. This requires coordination with the centralized distribution programme staff in order to ensure the availability of drug treatments at different times of the year based on community preference [15]. The integrated MDA programmes that have been implemented in Uganda and other countries may or may not necessarily involve community-selected drug distributors, and the time of distribution was set to coincide with other community health initiatives, such as Child Health Days, in order to take advantage of the existing immunisation and bednet distribution programmes [16]. Kamara et al [14]' s study of the Sierra Leone NTD programme found that the majority of districts did not have a gender imbalance in treatment coverage, but that males were significantly more likely to have received treatment in three out of the 13 surveyed districts. Additionally, although the World Health Organization (WHO) has recently created a database with information on MDA coverage rates by disease and country, the data sets are not disaggregated by gender, nor do they separately identify the adult and adolescent populations [17]. Coverage validation surveys of NTD programmes conducted by the US Centre for Disease Control (CDC) collect gender specific information on reported treatment uptake, but data presented in published reports does not show coverage rates by gender [18].
The majority of existing literature on gender and NTDs focuses on female populations. In a study of community perceptions of intestinal schistosomiasis, Anguzu et al. [19], note that the disadvantaged socio-economic status of women within rural communities in Uganda can prevent them from actively participating in health programmes, and from accessing information on control or preventative measures for the disease. However, Clemmons et al [12] note that although women are often less involved in the decision-making processes in community treatment programmes for onchocerciasis, they tend to identify themselves as more susceptible to the disease and express stronger feeling of the benefits of the programme, which may impact their long-term adherence with the programme.
Very little information exists on the participation of pregnant and breastfeeding women in treatment programmes. According to international guidelines established by WHO, praziquantel, used for treatment of schistosomiasis, is safe for pregnant and breastfeeding women; albendazole, used for treatment of soil transmitted helminths (STH) and, when combined with ivermectin, lymphatic filariasis, is safe for pregnant women after the first trimester and for breastfeeding women; ivermectin, which is used to treat the aforementioned lymphatic filariasis and onchocerciasis, should not be provided to pregnant women but is safe for breastfeeding women once the child is a month old; and Zithromax, used to treat trachoma, should not be provided to pregnant or breastfeeding women until the child is one year old [20]. Previous studies have noted that pregnancy may leave large numbers of women untreated if mass treatment takes place during the time of their pregnancy or lactation, and later treatment opportunities are unavailable or unknown [21]. However, no studies have been identified that discuss how integrated treatment programmes, which contain multiple drugs and more complex eligible and ineligible requirements, may further affect access and adherence to treatment among pregnant and breastfeeding women.
Very minimal published research is available to assess male participation in NTD programmes. Including male perspectives in gendered analyses of MDA programmes is particularly significant given that a growing body of literature suggests that men have lower adherence and higher default rates with other types of vertical treatment programmes such as for HIV [22] and tuberculosis [23]. Finally, information on adolescent access and adherence to treatment is scarce. One study examining CDTI for onchocerciasis in Cameroon, Nigeria and Tanzania, briefly mentions that young women were the least likely to be aware of how to prevent and treat onchocerciasis in comparison to older men and women [12].
The specific objectives of this study were (i) to explore gender-related factors which may influence participation in treatment programmes for NTDs in Uganda in adolescent and adult populations; (ii) to examine factors which may contribute to a gender bias in the treatment programme; and specifically, (iii) to identify whether pregnancy and breastfeeding presents a barrier to the participation of women in MDA.
Ethical clearance for the study was obtained from the Uganda National Council of Science and Technology and the University of Toronto Research Ethics Board. District health authorities, sub-county health officers, and community leaders were informed of the study and provided verbal consent prior to the study's commencement. Community leaders who participated in a focus group discussion (FGDs), as well as sub-county supervisors and CMDs, gave written consent. As literacy rates tended to be low within the districts, all other adult and adolescent focus group participants provided verbal consent prior to the commencement of the FGDs. The names of all participants were recorded in a register book prior to the commencement of each FGD, and verbal consent was recorded by the researcher beside the participant's name. Written or verbal consent was also obtained from the parents of adolescents under 18 years of age. Both the Uganda National Council of Science and Technology and the University of Toronto Research Ethics Board approved the use of verbal consent for community members due to potential literacy issues that might prevent written consent from being obtained.
This study employed a mixed methods approach, using quantitative data from community registers on the number of persons treated, as well as qualitative data in the form of FGDs and key informant interviews. Initially, the study proposed to use both quantitative and qualitative data to triangulate findings. However, significant challenges were experienced in collecting quantitative data: in five of the eight villages sampled, community registers were incomplete, leaving only three villages with data that had been recorded in a useable way. As a result, data analysis was carried out using what Creswell [24] describes as a “Concurrent Embedded Strategy”. In this approach, analysis of the available quantitative data provided additional insights into the qualitative findings.
This study was carried out in Wankole and Kidera sub-counties of Kamuli and Buyende Districts respectively, in Eastern Uganda, which were both part of Kamuli district prior to the creation of Buyende as a separate district in 2010. The NTD control programme in both districts is implemented through the coordination of Kamuli District Vector Control Office. These sub-counties were purposively selected to capture a diversity of occupations, village sizes, and accessibility to town and city centres: Kidera, a remote area approximately 50 km north of Kamuli town centre which borders Lake Kyoga, is composed primarily of fishing villages, and Wankole, which lies approximately 35 km between Kamuli town and the city of Jinja, borders swampland and is primarily agro-based, with rice, maize, and sugarcane as the major crops. Two parishes were randomly selected from each sub-county, and from each parish, two villages were randomly selected, creating a total sample of eight villages.
Quantitative data were extracted in each village from the national NTD control programme registers of the number of persons treated in each household, their sex and age, and the number of tablets received. Additionally, population data from Village Household Register Books were gathered and matched to the NTD treatment registers by name, sex, and age to assess under or overestimation of those treated.
Qualitative methods included FGDs with adult men and women, adolescent males and females, community medicine distributors and community leaders in each village. Key informant interviews were also held with sub-country programme staff (Table 1). Focus group and interview guides were informed by Rathgeber and Vlassoff's [25] “Gender Framework for Tropical Diseases Research” and included questions on knowledge and beliefs about the diseases and the treatment programme, perceptions of risk, utilization of health services and the mass drug administration programme specifically, decision-making power and challenges faced in accessing the programme. The guides were developed in English and translated into Lusoga by a Ugandan research assistant, and subsequently back translated into English by a staff member at the Uganda Ministry of Health. The guides were then pretested in Jinja District, Uganda, and revised, according to pilot feedback, for use in the field.
FGDs were moderated in Lusoga by a Ugandan research assistant. Notes were taken of participants' verbal and non-verbal communication. Convenience sampling was used to select participants, with community leaders assisting in identifying 6–12 participants for each focus group (Table 1).
A total of two key informant interviews were conducted at sub-county level, one with each supervisor of the NTD programmes in Kidera and Wankole, respectively, to gain insights into the successes and challenge of the programme, any noted gender biases, and whether training in gender issues has been incorporated into the programme. Participant observation of the MDA was carried out in two communities in Wankole which were carrying out mass-treatment at the time of data collection. Field notes were recorded to document these visits. Data collection was carried out from June–August 2011, with approximately four weeks spent in each of the two districts.
Quantitative data were cleaned, but the planned analysis of gender and age relative to treatment status was restricted to three villages due to concerns with data validity. Using treatment registers kept by the CMDs in each community, the proportion of eligible males and females treated in each village was calculated. To validate the accuracy of the treatment registers, Village Household Register books, which provide a census of persons living in each village, were obtained from the Local Council Chairperson of each village. Proportions of persons treated were recalculated using the number of eligible males and females from the Village Household Register Books as the denominators.
FGD and interview data were recorded and transcribed into English. Qualitative data were analyzed with NVivo 8.0 software using thematic content analysis. Codes and sub-codes were developed through repeated scanning of the transcripts, which were then grouped into three thematic areas described in the Results below.
Each of the eight villages was actively administering MDA, however, most had not completed distribution of all the drugs for the treatment year. While most villages were expected to complete distribution in May 2011, by late June and early July only four villages had done so. Typically, a schedule of three courses of medication is followed, with separation of at least two weeks in between each medication to prevent interactions: first albendazole and ivermectin are distributed together, followed by praziquantel (if available), and finally Zithromax after several more weeks. In Wankole, the CMDs in each village reported distributing ivermectin, albendazole and Zithromax. However, the drugs that the CMDs reported distributing in Kidera varied: CMDs in one village reported distributing ivermectin, albendazole and Zithromax; CMDs in a second village reported distributing praziquantel, ivermectin, albendazole and Zithromax; in a third one, praziquantel, albendazole and Zithromax; and in a fourth village, only Zithromax. After conversations with the sub-county supervisor and CMDs in Kidera and an examination of the community treatment register books, it was unclear why certain communities may not have distributed ivermectin or albendazole. Annual shipments of praziquantel had been delayed in Kampala in 2011, and thus communities who distributed praziquantel were relying on remaining tablets from previous years MDA campaigns.
Accessing valid quantitative data from each of the selected villages was challenging. Although all CMDs had a written record of who had received treatment in a treatment register, record keeping varied in quality. Of the four villages which had completed the annual mass treatment, the CMDs had not registered community members prior to mass treatment, and it was therefore difficult to ascertain if the entire community had been treated. In one of the four villages, very little information was recorded, and thus this village, as well as three other villages which had not completed distribution, were removed from the quantitative analysis. The treatment registers also did not contain important information such as the reasons for treatment refusal. The findings in Table 2 present results for the three villages where treatment had been completed, and thus where data on the numbers treated were also presumed to be complete. However significant discrepancies were noted when the total numbers eligible for treatment from the treatment registers were compared with the total numbers registered in the official Village Household Register Books (Table 2). Despite these challenges, two of the three villages consistently showed a higher proportion of females treated than males treated. Given that some females were also not eligible for treatment due to pregnancy, the gendered differences in eligible numbers treated were likely underestimated.
Our findings reflect some of the challenges for MDA programmes in treating NTDs with regards to monitoring drug coverage and ensuring that all eligible persons are able and willing to access treatment. Our quantitative findings provide some suggestion that men are less likely to receive treatment than eligible women. Because the quantitative denominator data was taken from both NTD programme treatment registers and household register books from the villages, it is most likely that this reflects the number of men who are actual residents of the village, rather than temporary migrants, as has been found in other studies [26]. Findings also indicate programmatic challenges and illustrate the difficulties in assessing treatment coverage rates at a community level. Similar findings were identified by Parker and Allen [26] who observed significant problems with the community NTD control programme registers in some areas, which were often missing or incomplete, creating difficulties in estimating the number of persons treated each year. This is in contrast, however, to a larger coverage study by Kabatereine et al. [27] who found that community medicine distributors were at least moderately accurate with record keeping, with approximately 61% of communities having entered data correctly in their registers. Although the study by Kabatereine and colleagues [27] focused on mass treatment for schistosomiasis and STH only, it nonetheless highlights several areas of concern regarding the community-based strategy for drug distribution, including inadequate registration of community members in certain areas, which may result in shortages of medicines and low treatment coverage.
Our qualitative findings were able to shed light on specific social, behavioural and programmatic issues which are creating and reinforcing gender differences in access to treatment. Unlike Anguzu et al. [19] and Clemmons et al. [12], our findings do not suggest a bias towards males in terms of access to information or decision-making power regarding treatment access. This may be due to social and cultural differences between Anguzu et al. (carried out in Busia District, East Uganda) and Clemmons et al. (carried out in Cameroon, Tanzania and Nigeria) study populations and the present study setting, or to differences in family structure and hierarchy. Alternately, this may simply be a consequence of male absence when the CMDs arrive to distribute treatment, which leaves females with more decision-making power and provides them with an opportunity to ask CMDs questions about the treatment. However, our findings suggest socio-behavioural and structural barriers are present for both men and women, but tend to differ between the genders. The majority of villages in this study relied almost exclusively on a house-to-house distribution strategy which appeared to be particularly unsuitable for accessing males. Whether such gender-specific barriers are also present in a centralized distribution strategy, where community members gather in a central place on specified days to receive treatment, should be explored in future research.
Findings of this study also support previously published work suggesting that access to treatment is particularly difficult for pregnant and breastfeeding women [21]. This is significant given the potential for pregnant women to suffer negative consequences of NTDs during pregnancy. Several studies have identified anthelminthic treatments during pregnancy as having positive outcomes on reducing maternal anemia [28], [29], [30], although this benefit may be realised only in women with moderate to high intensity helminth infections [31], [32]. The potential for additional health gains by reducing low birth weights and infant mortality through anthelminthic treatments have also been noted in observational studies [30]. In addition, treating pregnant women who are co-infected with malaria and HIV for the STH can reduce the intensity of malarial infection [33], and may benefit those with a higher viral load [34]. Finally, it is estimated that women living in schistosomiasis endemic areas may spend up to 25% of their reproductive years pregnant and another 60% of this time lactating [35]. Thus, as indicated by our results, women who miss treatment due to pregnancy and breastfeeding are likely to repeatedly miss treatment, and may be more susceptible to organ damage and cancer due to chronic schistosomiasis infection [35].
The lack of access to treatment by pregnant and breastfeeding women and the health risks resulting highlight the need to improve community knowledge that medicines are available at health centres following the community-based treatment period, a point that was previously raised in a process evaluation carried out in Uganda in 2006 [16]. Greater training for CMDs on guidelines for treating pregnant and breastfeeding women, and encouragement of health workers to provide medicines for women attending postnatal clinics may also help improve access for this population. However, further research and monitoring of adverse outcomes is also necessary as recent studies have suggested that albendazole and praziquantel use during pregnancy may increase risk of infantile eczema [31] and may also have positive or negative effects on the development of the fetal immune system and disease susceptibility later in life [32].
The use of convenience sampling and the limited number of villages included in the sampling frame, while appropriate for the qualitative methods used, does not allow for generalizability to other areas of Uganda. Thus, the extent and type of gender bias present in the sampled villages cannot be compared to other villages or regions in Uganda. Instead, the combined results of the quantitative and qualitative findings provide insights into some of the broad issues that may influence gendered difference in access and adherence to treatment. Figure 1 provides a framework which synthesizes the information from qualitative and quantitative findings as follows: personal circumstances and social and behavioural aspects, which are often gender specific, create and reinforce personal and community attitudes towards the treatment programme. Attitudes towards the treatment programme may create problems with the distribution strategy if community members are unable or unwilling to benefit from the programme. Additionally, problems with the distribution strategy, such as insufficiently-trained CMDs and lack of health education, reinforce negative attitudes towards the programme, as CMDs are unable to refute rumours about the treatment or convince community members of the importance of the programme. Moreover, specific programmatic decisions, such as the annual house-to-house distribution strategy which tends to miss adult males and pregnant women, also contribute to the large number of persons who are apparently missed during mass-treatment.
Rathgeber and Vlassoff's [25] “Gender Framework for Tropical Diseases Research”, which was used to inform data collection tools, provides a comprehensive and integrated synthesis of social, economic and personal factors which are contributing to gendered differences in tropical diseases. This includes social and behavioural factors at the personal level and the community level as well as at the policy level that affect access and use of health services. However, since this framework was developed prior to the initiation of national NTD programmes, its authors did not consider critical programmatic factors such as challenges with the distribution strategy in accessing men and pregnant women. This framework is also limited by its inability to elucidate the direction of causality of these social, economic and personal factors, which would be helpful for refining future NTD programming.
Figure 1 attempts to build on Rathgeber and Vlasoff's framework by placing community attitudes and problems with the distribution strategy at the centre of the issue, arguing that a more comprehensive understanding of the nuances and challenges of community-based treatment programmes is desperately needed to address gender-related challenges and ensure future success of the programmes.
Our results have several implications for community-level programming. In communities where coverage rates are low, there is likely a need to augment the training of community personnel. Rumours about the treatment, the inability of CMDs to answer community member's questions, as well as negative experiences with a previous community immunization programme, can all affect attitudes towards the treatment, and can have important consequences for adherence to the NTD programme. Thus, the capacity of CMDs to facilitate health education, answer community members' questions about treatment, understand the causes of diseases, and manage side-effects caused by the medicines must be strengthened. Brieger et al. [13] have suggested holding ongoing refresher workshops for the CMDs to support them in addressing problems with adherence within their communities. Such workshops could emphasize the importance of the CMDs as motivators of others to take the treatment, and provide in-depth information on side effects of medications and the causes of the diseases. To address problems with the distribution strategy, CMDs should also be encouraged to consult with groups who spend long periods outside of the village and determine an optimal time and strategy for reaching these groups when they are home. They should also be encouraged to discuss these issues with sub-county program staff so strategies for reaching groups with low adherence rates can be developed, and programmes can be modified accordingly.
Finally, there is a need to work with communities to improve self-monitoring and evaluation strategies. However, it should be recognized that it is time consuming for CMDs to travel to each individual household prior to mass-distribution to register village members, and that record-keeping also adds to the workload. In communities where household registration is not completed, or is completed inconsistently, it may make the most sense to encourage CMDs to ensure, at minimum, that treatment numbers are correctly recorded. Concurrently, district, national or international stakeholders should invest increased effort into coverage surveys and qualitative follow-up activities to provide more general validation and insight. Moreover, given the demands of distributing and monitoring drug coverage, the challenges with low coverage rates in some communities, the training and skills required of the CMDs on knowledge of diseases and treatments, and the considerable time that must be spent to troubleshoot methods of improving adherence, there may be a need to reconsider the voluntary status of the CMDs. We recognize that this is a controversial point, given that other work has noted that compensation does not necessarily contribute to higher coverage rates or lower attrition among CMDs [36]. While the results of our study are context-sensitive and cannot be generalized for the rest of Uganda, other research has also raised the point that the growing number of community-based programmes reliant on CMDs may take them away from their own income-generating activities, and thus create a need for monetary compensation [37]. It is therefore suggested that national programme evaluations examine CMD workload and attrition as a potential factor in communities with low coverage rates.
Our results also highlight the critical importance of CMDs in influencing community member adherence to the programme. Many of the challenges described are unlikely to be addressed unless NTD programme management are able to commit greater resources to mitigating community-level challenges. Resources required are those for ensuring quality training of CMDs, NTD-related health education and mobilisation for community members, and motivation for CMDs to carry out their duties efficiently and effectively.
Greater attention to community-level challenges is particularly important given the limitations in monitoring and evaluating coverage of the programme and its operations. According to a recent article by Research Triangle Institute (RTI) International, the main international agency overseeing the current integrated NTD programme, at least 98 million persons have been treated under the programme in the past three years [38]. Evidence of these findings is reported through self-report from drug distributors and their supervisors, which was then validated with subsequent coverage studies [38]. Our findings illustrate the important gender-related social, behavioural and personal factors that may affect community members' ability to access treatment. In addition to quantifying the number of persons by gender and age group who are taking the treatment, there is also a need for a greater understanding of why persons are choosing or not choosing to participate in mass treatment, and the context in which challenges with distribution or adherence are occurring. As Kabatereine et al. [39] note, funding for in-depth monitoring and evaluation is limited given that the majority of resources are currently directed towards intervention delivery which precludes the acquisition of the data needed to determine whether modifications to programmatic activities are needed to increase their effectiveness. Evaluations conducted at the national or international level could compliment community-level data collection, and should include gender and age disaggregated data, as well as additional qualitative or other research methods to understand in a more nuanced way the complex social, behavioural and structural issues that can affect coverage rates.
Over the past decade, significant progress has been made in controlling or eliminating NTDs with billions of dollars devoted to treating the diseases through national programmes in sub-Saharan Africa and elsewhere. However, more resources must be devoted to understanding implementation at the community level to address the nuances and challenges of drug distribution within communities. Our study focused on gender issues and identified several key areas where modifications and increased monitoring at the community level may significantly improve access to treatment for both males and, especially pregnant females. Monitoring and evaluation strategies should include social and behavioural data in addition to accurate sex- and age-disaggregated data on the number of persons who are adhering and not adhering to the NTD programme. These challenges must be addressed if long-term goals of controlling NTDs are to be achieved.
|
10.1371/journal.pgen.1004333 | Mutations in the Cholesterol Transporter Gene ABCA5 Are Associated with Excessive Hair Overgrowth | Inherited hypertrichoses are rare syndromes characterized by excessive hair growth that does not result from androgen stimulation, and are often associated with additional congenital abnormalities. In this study, we investigated the genetic defect in a case of autosomal recessive congenital generalized hypertrichosis terminalis (CGHT) (OMIM135400) using whole-exome sequencing. We identified a single base pair substitution in the 5′ donor splice site of intron 32 in the ABC lipid transporter gene ABCA5 that leads to aberrant splicing of the transcript and a decrease in protein levels throughout patient hair follicles. The homozygous recessive disruption of ABCA5 leads to reduced lysosome function, which results in an accumulation of autophagosomes, autophagosomal cargos as well as increased endolysosomal cholesterol in CGHT keratinocytes. In an unrelated sporadic case of CGHT, we identified a 1.3 Mb cryptic deletion of chr17q24.2-q24.3 encompassing ABCA5 and found that ABCA5 levels are dramatically reduced throughout patient hair follicles. Collectively, our findings support ABCA5 as a gene underlying the CGHT phenotype and suggest a novel, previously unrecognized role for this gene in regulating hair growth.
| Inherited hypertrichoses represent a group of hair overgrowth syndromes that are extremely rare in humans and have remained an area of great interest to evolutionary geneticists since they are considered to be recurrences of an ancestral phenotype. These syndromes often present with additional congenital abnormalities including bone, heart and dental defects; thus, it is crucial to identify the mechanisms and genes underlying the pathology. Copy number variants (CNVs) have previously been reported in several cases of congenital generalized hypertrichosis terminalis (CGHT) with a minimal overlapping region of 555 kb encompassing four genes. However, no point mutations in these or any other single genes have been described to underlie the CGHT phenotype. In this study, we report the first loss-of-function mutation in an ABC transporter, ABCA5 and identified an additional copy number variant in a separate case that lies within the minimal common region. We found high levels of ABCA5 expression in both epithelial and mesenchymal compartments of human and mouse hair follicles, and in CGHT patients, this expression is significantly reduced or completely lost. ABCA5 is a lysosomal protein, and its loss-of-function compromises the integrity of lysosomes and leads to an intra-endolysosomal accumulation of cholesterol. Importantly, our findings support a novel role for ABCA5 in regulating hair growth.
| Inherited hypertrichosis, first described in the sixteenth century, is characterized by hair growth that is excessive for the body site and age of an individual and is independent of androgen stimulation [1], [2]. These syndromes are categorized as ectodermal dysplasias and are often associated with additional congenital defects, including cardiomyopathy, gingival hyperplasia, and craniofacial malformations [3]. The genetic basis of these syndromes remained largely elusive until recently, when our group and others reported several chromosomal rearrangements, copy number variants (CNVs) and position effects involving genes associated with autosomal dominant, recessive, sporadic, and X-linked forms of hypertrichosis [4]–[10].
We previously reported a position effect on the zinc finger transcription factor TRPS1 in Ambras syndrome hypertrichosis in humans and the Koala hypertrichosis phenotype in mice, where Trps1 expression was decreased at the sites of pathology for the phenotype [5]. More recently, we and others elucidated the genetic basis of X-linked hypertrichosis [9],[10], resulting from large interchromosomal insertions on the X chromosome. We found that a position effect occurs on a distant gene, FGF13, whose expression was markedly and selectively reduced in patient hair follicles, suggesting a novel role for this growth factor in hair follicle growth and cycling [10].
In the autosomal dominant form of CGHT, we identified a series of duplications of chromosome 17q24.2-q24.3 and reported a position effect on the SOX9 gene, situated ∼1 Mb downstream of these variants, and found SOX9 expression was markedly reduced throughout the follicular epithelium of patient hair follicles [4]. In a separate report of CNVs in the same region of chromosome 17q24.2-q24.3, four CNVs were identified in several cases of CGHT [6], with an overlapping minimal region of 555 kb encompassing four genes: ABCA6, ABCA10, ABCA5, and MAP2K6, suggesting that disruption of one of these genes may contribute to the CGHT phenotype. Despite the identification of CNVs and/or position effects in this region of chromosome 17q24.2-q24.3, no point mutations in these or any other single genes have been described to underlie the CGHT phenotype.
In this study, we investigated the genetic basis of autosomal recessive CGHT (OMIM135400) in a consanguineous family. Whole-exome sequencing revealed a novel, rare variant in the 5′ donor splice site of intron 32 of ABCA5 that segregates with the phenotype in a homozygous recessive manner. ABCA5 is highly expressed in human skin and hair follicles, and its expression pattern is conserved in mouse tissues as well. Importantly, the ABCA5 loss-of-function mutation not only leads to a decrease in protein levels in both mesenchymal and epithelial compartments of CGHT patient hair follicles, but also to lysosomal dysfunction, which results in a defective clearance of autophagosomes under basal conditions and an overall accumulation of endolysosomal cholesterol in patient keratinocytes. We also identified an unrelated case of sporadic CGHT with a t3;17 translocation and cryptic 1.3 Mb deletion spanning ABCA5, and found that ABCA5 levels were dramatically reduced in patient cells as well as throughout the hair follicle epithelium. Our findings implicate ABCA5 as a gene with an essential role in hair growth.
We ascertained a proband from Yemen with CGHT segregating with gingival hyperplasia as well as epilepsy. Excessive hair growth was observed on the face, including the forehead, cheeks, and upper cutaneous lip, arms, upper and lower back and legs (Figure 1A). No other members of the patient's family are affected, however, the parents of the proband were consanguineous, suggesting an autosomal recessive mode of inheritance (Figure 1B).
Analysis of patient hair follicles obtained from a whole skin biopsy of the forearm by hematoxylin and eosin staining demonstrated that the hairs are of the terminal type since they were medullated, pigmented and penetrated deep into the dermis (Figure S1). Patient hair follicles were thicker than those of controls, and anagen hairs were present in the skin biopsy, whereas no anagen hairs were detected in control skin biopsies. We measured the length of the hair shafts from plucked patient and control forearm hair follicles and found that the patient hair follicles were significantly longer (87%), with an average length of 29.6 mm (±0.9 mm) compared to 15.9 mm (±0.6 mm) for control follicles (P<0.0001) (Figure 1C).
To investigate the genetic basis of CGHT in this family, whole-exome sequencing was performed (Ambry Genetics) on genomic DNA obtained from the patient as well as both parents. Following sequencing, bioinformatics analysis (see Materials and Methods) and filtering on mode of inheritance ultimately lead to the identification of variants in 26 candidate genes (33 alterations) that were homozygous in the proband and heterozygous in both parents (Figure 1D). Further interpretative filtering based on literature searches focused on genotype-phenotype correlation revealed three candidate genes with homozygous mutations, ABCA5, DGKZ, and ZNF253, all of which are not currently associated with a Mendelian disease, and thus are considered novel. The nature of the homozygous mutations identified in these genes includes one splice site mutation (in ABCA5: c.4320+1G>C), one missense mutation (in DGKZ: c.1678C>T(p.P560S)), and one deletion (in ZNF253: c.429delA). Segregation analysis of these mutations revealed that both parents are heterozygous for all three mutations, whereas neither unaffected sister carries the ABCA5 c.4320+1G>C and DGKZ c.1678C>T(p.P560S) mutations, and only one unaffected sister carries the ZNF253 c.429delA mutation.
To determine whether these candidate genes are expressed in the skin and hair follicle, RT-PCR analysis was performed on RNA isolated from whole skin, which revealed that the ABCA5 and DGKZ genes are abundantly expressed, suggesting a potential role for these two genes in the pathogenesis of the CGHT phenotype, whereas ZNF253 is expressed at lower levels. The function of the ZNF253 gene is unknown. The DGKZ gene encodes diacylglycerol (DAG) kinase zeta, a member of the DAG kinase family which phosphorylates DAG to phosphatidic acid and plays important roles in lipid signaling implicated in neurological diseases, including epilepsy, depression and Alzheimer's disease [11] – [13]. Moreover, mice deficient in the gene encoding DAG kinase, epsilon (Dgke) a member of the same gene family, exhibit features associated with epilepsy [14], suggesting the DGKZ substitution mutation may contribute to the pathogenesis of the patient's seizures. ABCA5, an ATP-binding cassette (ABC) protein, is a lipid transporter involved in the efflux of cellular cholesterol levels, and Abca5-deficient mice develop symptoms similar to several lysosomal diseases of the heart [15], [16]. Interestingly, ABCA5 lies in the minimal common region to four reported familial cases and one sporadic case of autosomal dominant CGHT, both with and without gingival hyperplasia [4] – [6], suggesting that mutations in this gene may be associated with CGHT.
The ABCA5 mutation results from a G-to-C substitution in the first nucleotide of intron 32 (Figure 1E). Sanger sequencing was performed on genomic DNA from the proband as well as the unaffected father, which confirmed homozygosity in the proband (II-2) and heterozygosity in the father (I-2) for the c.4320+1G>C mutation (Figure 1 E, F). Importantly, this mutation was not present in control individuals, determined by searching various genome databases and sequencing the genomic DNA of 10 control individuals. Moreover, a query for genetic variants that lie within the ABCA5 locus using the UCSC Human Genome (hg19) and Ensemble Genome Browsers verified that this variant is not a SNP or common variant associated with any human genetic disease. Thus, the c.4320+1G>C mutation in ABCA5 is novel.
In light of the cosegregation of the mutation with the disease phenotype in the family, association with the CGHT phenotype in previously reported cases, and reported expression in the human hair follicle with ABCA5 mRNA levels being the highest out of the four genes within the minimal common region [6], we further investigated the consequence of this mutation on ABCA5 mRNA splicing and the potential role for this protein in hair follicle growth.
In silico analysis of splicing events using the computational algorithms Berkley Drosophila Genome Project (BDGP) [17] and ESEfinder [18], [19] predicted that the c.4320+1G>C mutation results in complete loss of the donor splice site. To determine the consequence of the mutation at the transcript level, we amplified by RT-PCR the ∼200 bp region flanking the mutation from patient, carrier, and control mRNA. Sanger sequencing revealed that the mutation leads to aberrant skipping of exon 32 in the proband (Figure 2A). As a result of joining exon 31 to exon 33, the remainder of the transcript downstream of the mutation is out-of-frame, leading to a premature termination 14 bp downstream of the mutation (Figure 2A). RT-PCR analysis of the exon 31–33 amplicon using RNA isolated from whole skin revealed the complete absence of the wild-type transcript in the proband. In contrast, we found the presence of the wild-type transcript at high levels and the mutant transcript at very low levels in the father (Figure 2A). The mutant mRNA most likely undergoes nonsense-mediated decay, since the mutation resides near the 3′ end of the transcript and the aberrant splicing event is predicted to affect the overall stability of the mRNA.
To investigate this possibility, we compared the relative ABCA5 mRNA levels between the proband and unaffected carrier father by qRT-PCR using primers located at the 5′ end of the mRNA, and found that transcript levels were significantly reduced in patient whole skin (2.7-fold; p<0.0001), cultured keratinocytes (2.8-fold; p = 0.0016), as well as fibroblasts (4.9-fold; p<0.0001), demonstrating that the mutant mRNA is unstable and undergoes nonsense-mediated decay (Figure 2B).
While ABCA5 expression was previously identified in plucked human hair follicles by RT-PCR analysis [6], the specific cell type(s) synthesizing ABCA5 in the hair follicle and potential expression in the surrounding dermal tissue remained unknown. Using in situ hybridization on human hair follicles in the growth phase of the hair cycle (anagen), we observed ABCA5 transcript expression in both the epithelial and mesenchymal compartments, present within the outer root sheath (ORS) of the hair follicle as well as dermal sheath (Figure 2C). To determine the localization pattern of ABCA5 protein in human skin and hair follicles, immunohistochemistry (IHC) was performed on paraffin-embedded skin sections and expression was most evident in the dermal sheath, perifollicular dermis, ORS, and IRS of hair follicles (Figure 2 D–F). Since the ABCA5 antibody is polyclonal, we validated endogenous ABCA5 expression in the hair follicle as well as in the surrounding perifollicular dermis using RT-PCR on these tissues and observed strong ABCA5 expression in plucked hair follicles (HF), microdissected ORS, as well as perifollicular dermis (FD) including the dermal sheath (Figure 2G).
To assess whether the human ABCA5 expression pattern was conserved in mouse skin and hair follicles, we first determined whether we could detect Abca5 immunoreactivity in a site of known Abca5 expression using the same polyclonal antibody. In mice and rats, Abca5 mRNA levels are abundant in the testis by Northern blotting and in situ hybridization [20], [21]. Using immunohistochemistry and immunofluorescence staining on adult mouse testis sections, we observed strong localization of Abca5 to the basal cells of the seminiferous tubules, interstitial cells consisting of Leydig cells (as previously reported in [21]), as well as the tunica albuginea (Figure S2 A–B, G). In the epididymis, we found very strong and specific localization of Abca5 to the connective tissue outlining the cylindrical epithelium in the corpus and cauda regions, including fibrocytes and smooth muscle cells, as well as within the basal and tall columnar cells of the corpus cylindrical epithelium (Figure S2 D–E, H–I). The control testis and epididymis sections (no primary antibody) yielded no signal (Figure S2 C, F). Importantly, we observed the same localization pattern of Abca5 in these tissues using two different fixatives; an organic solvent (methanol/acetone) and cross-linking agent (formalin), and our data are consistent with previous reports on Abca5 mRNA expression [20], [21].
We next investigated the localization pattern of Abca5 in the mouse anagen hair follicle using immunofluorescence staining and immunohistochemistry, and observed high levels of Abca5 localization to the ORS and IRS of hair follicles (Figure S3). Expression was also observed in the dermal sheath and perifollicular dermis by the immunohistochemistry method (Figure S3 E–F), similar to what we observed in human hair follicles (Figure 2D). Importantly, ABCA5 localization in the skin and hair follicle appears to be conserved between human and mouse, and its broad expression pattern spans multiple cell lineages, both within and surrounding the hair follicle. Collectively, these data suggest a prominent, evolutionarily conserved role for this transporter in regulating hair growth.
To determine the consequence of the ABCA5 c.4320+1G>C mutation at the protein level, we performed immunofluorescence staining on control and CGHT keratinocytes cultured from whole skin biopsies, and observed a striking reduction in ABCA5 localization (Figure 3A). Immunoblotting revealed the loss of a 215 kDa band corresponding to the full-length transporter in its glycosylated form, as well as a 187 kDa band, representing the unglycosylated form [15] (Figure 3B). While the full-length transporter was predominantly detected in keratinocytes, in fibroblasts, we observed the expression of a previously reported truncated variant [20] that produces a half-transporter and a ∼100 kDa band. Importantly, the band representing this truncated variant was not detectable in CGHT patient fibroblasts (Figure S4). Since ABCA5 is a reported glycoprotein possessing multiple sites for N-glycosylation, we investigated whether the c.4320+1G>C mutation abolished the glycosylated form of the protein in patient fibroblasts by incubating total protein in the presence or absence of the PNGaseF enzyme that removes all N-glycosyl modifications. Following immunoblotting, we observed that the band corresponding to the ∼100 kDa isoform that is absent in the proband represents the glycosylated form of the protein (Figure S4). Since glycosylation is an important post-translational modification crucial to the proper folding, stability, subcellular localization and/or even function of many lysosomal proteins, including ABC transporters [22], this finding suggests a loss-of-function of both the full- and half-transporters encoded by ABCA5 in CGHT.
Lastly, to evaluate the consequence of the c.4320+1G>C mutation on ABCA5 protein localization at the tissue level, we performed immunofluorescence staining on patient and control hair follicles and observed a striking reduction of ABCA5 protein throughout the outer and inner root sheaths of patient hair follicles in catagen and anagen (Figure 3C). Importantly, loss of ABCA5 expression at the site of pathology for the phenotype further demonstrates that the c.4320+1G>C mutation results in a complete loss-of-function allele.
Lysosomes have been reported to play a role in hair cycling, where mice deficient for lysosomal proteases, cathepsin L (Ctsl) and lysosomal acid phosphatase 2 (Acp2), have delayed progression through the hair cycle resulting in periodic hair loss that is characterized by hyperproliferation of epithelial cells in the hair follicle and basal layer of the epidermis [23] – [26]. In mice, Abca5-deficient cells have aberrant processing of autolysosomes and autophagosomes [15], suggesting a role in lysosome integrity and/or autophagy, a catabolic process of intracellular digestion and recycling of organelles [27]. To gain insight into whether CGHT patient cells possessed intrinsic autolysosomal and/or autophagic defects, we visualized key markers of these organelles at the cytological level. LC3, a well-established marker for autophagosomes, is converted from its cytosolic form (LC3-I) to a lipidated form (LC3-II) which is able to re-localize and bind specifically to autophagosomal membranes [27]. A large pool of LC3-II is degraded upon lysosome-autophagosome fusion, when the internal content of autophagosomes is destroyed by lysosomal hydrolases. Immunofluorescence staining against endogenous LC3 revealed an increased number of LC3-positive particles (i.e., autophagosome-like structures) in the affected keratinocytes compared with control, suggesting defects in the autophagy pathway (Figure 4A).
In order to discriminate between higher levels of basal autophagy and a defective clearance of autophagosomes, bafilomycin A1 (BAF), a proton pump inhibitor that blocks the acidification of lysosomes and thus the clearance of autophagosomes, was added to growing cells 2 hours prior to fixation. BAF treatment caused a 2-fold increase in the number of LC3-positive puncta representing autophagosomes and autophagolysosomes in control keratinocytes, compared to untreated keratinocytes (Figure 4A, Figure S5A). In contrast, BAF treatment failed to significantly increase the number of LC3-positive particles in the patient keratinocytes compared to untreated cells (Figure 4A, Figure S5A). Furthermore, immunofluorescence staining against p62, a polyubiquitin binding protein as well as an autophagic cargo, revealed a strong accumulation of puncta in patient keratinocytes, further suggesting autophagy defects (Figure 4A, Figure S5B). Overall, these results indicate that ABCA5 loss-of-function in CGHT causes defects in the autophagy pathway, and more specifically, point to an impairment in the clearance of autophagosomes and autophagic flux under basal conditions. These results also suggest that the mechanism underlying the autophagy defects caused by the ABCA5 mutation is a decrease in lysosomal function.
Because a previous report has shown a role for ABCA5 in the efflux of cholesterol in macrophages [16], we next tested whether patient cells exhibit defects in the metabolism and/or transport of free cholesterol. Additionally, because ABCA5 is localized to the lysosomal compartment, its mutation may affect the handling of lipoprotein-derived cholesterol in the endolysosomal system, perhaps contributing to the dysfunction of these organelles in patient cells. In order to visualize cholesterol, cells were stained with filipin, a polyene macrolide antibiotic and antifungal that fluoresces and detects unesterified free cholesterol [28]. Remarkably, the mutation in ABCA5 produced an increase in intracellular filipin staining compared to controls (Figure 4B, top panel). More specifically, filipin-positive puncta were observed mostly inside Lamp1-positive structures (Figure 4B, lower panel) suggesting an accumulation of free cholesterol in the lumen of endolysosomal organelles, likely in their intraluminal vesicles. These data suggest that ABCA5 controls the fate of lipoprotein-derived cholesterol and that its mutation alters the intracellular traffic of free cholesterol, somewhat reminiscent of phenotypes observed in lysosome storage disorders, such as Niemann Pick disease Type C [29].
At around the same time we identified the ABCA5 c.4320+1G>C mutation in homozygous recessive CGHT, we independently studied this candidate region of chromosome 17 in a sporadic case of hypertrichosis. We ascertained a patient from Mexico with hypertrichosis universalis congenita (OMIM145700), whose parents (nonconsanguineous union) and siblings were unaffected. The patient exhibited excessive overgrowth of terminal hairs on the extremities, back, chest, and the face. Moreover, histological analysis of patient hair follicles obtained from a skin biopsy of the lower back revealed the presence of large anagen hair follicles that penetrate deep within the dermis (Figure S6).
We initially performed karyotype analysis using G-banding of metaphase chromosomes that revealed a translocation between chromosomes 3q12 and 17q25, while the other chromosomes appeared cytogenetically normal. Chromosomal paint with two FISH probes against chromosomes 3 and 17 verified that no other chromosomes were involved in this rearrangement (Figure 5A), and telomere FISH (Figure S7A) confirmed that no subtelomeric sequences were lost as a result of the chromosome 17 breakpoint near the q telomere, suggesting an apparently balanced translocation event. To test whether this rearrangement segregated in the family, karyotype analysis was performed on both parents and siblings of the patient, but no abnormalities were detected, suggesting a de novo chromosomal rearrangement.
We next performed FISH analysis using BAC clones, which revealed a cryptic deletion at the breakpoint of chromosome 17, spanning BAC clones RP11-387O17 (chr17:66,354,485-66,568,819) and RP11-293K20 (chr17:67,076,187-67,261,438) (Figure S7B, C). To fine-map the deleted region, we utilized the Affymetrix 2.7M SNP array, which defined an 849.4 kb cryptic breakpoint deletion on chromosome 3p12.2 which did not contain any annotated genes, as well as a 1.3 Mb deletion on chromosome 17q24.2-q24.3 (Figure 5B) that encompassed 7 Ref Seq genes: members of the superfamily of ATP-binding cassette (ABC) transporters (ABCA8, ABCA9, ABCA6, ABCA10, ABCA5), mitogen-activated protein kinase kinase 6 (MAP2K6) and potassium inwardly-rectifying channel J16 (KCNJ16). We confirmed the 1.3 Mb deletion with quantitative PCR (qPCR) on patient and control genomic DNA using primers specifically designed to amplify sequences flanking and within the deleted region (Figure 5C). To determine whether the translocation event resulted in the disruption of a gene, we searched for transcripts that mapped to the vicinity of the breakpoints using the Genome Browser available at the UCSC database (hg19) and subsequently cloned the breakpoints, which did not reveal any evidence for gene disruption.
Several cases of CGHT have been reported in which CNVs on chr17q24.2-24.3 that lie within a 2.4 Mb region (Figure S8) have been identified and hypothesized to cause a position effect on the SOX9 gene, a well-defined regulator of hair follicle stem cells [30], [31], which resides 1-2 Mb downstream of these variants. In our previous report of an autosomal dominant CGHT case, we found SOX9 expression was dramatically reduced throughout the hair follicles of patients who possess duplications within the chr17q24.2-24.3 region [4]. To investigate the possibility that the 1.3 Mb deletion identified in the present study led to altered SOX9 expression, we performed qRT-PCR on RNA isolated from patient and control keratinocytes and observed decreased SOX9 expression in the patient keratinocytes by approximately 2.5-fold (P>0.001) (Figure 5D). This result is consistent with our previous findings, suggesting a contribution of decreased SOX9 expression levels to the excessive hair overgrowth phenotype of CGHT. Given the association of SOX9 with CGHT and its known role in regulating hair follicle stem cells and outer root sheath differentiation [30], [31], we investigated the consequence of ABCA5 loss-of-function on SOX9 transcripts in the autosomal recessive CGHT case. Using qRT-PCR on patient and control keratinocytes, we did not detect a significant difference in SOX9 gene expression (data not shown). While it remains possible that SOX9 acts upstream of ABCA5, these genes may reside in separate pathways and utilize distinct mechanisms to contribute to the CGHT pathology.
Since the 1.3 Mb deletion we identified encompasses the ABCA5 locus, and both sporadic and autosomal recessive CGHT patients possess excessive overgrowth of terminal hairs, we next investigated the possibility that the second allele in the sporadic CGHT case may harbor a mutation in the ABCA5 gene. Therefore, we sequenced all the exons of ABCA5 and ∼2 kb upstream of the gene, but did not find any mutations (data not shown). To test ABCA5 mRNA expression levels in the sporadic CGHT case, we performed qRT-PCR in keratinocytes and fibroblasts cultured from a patient skin biopsy compared to control, and observed a 3.7-fold decrease in ABCA5 transcript levels in patient keratinocytes (p<0.0001), and a 2.4-fold decrease in patient fibroblasts (p<0.0001) (Figure 6A). Immunoblotting in fibroblasts cultured from patient and control skin biopsies revealed a striking decrease in ABCA5 protein levels, and treatment with the N-glycosylase PNGaseF revealed loss of the glycosylated form of the protein (Figure 6B). Moreover, we found using immunofluorescence staining on patient and control hair follicles in catagen and anagen that ABCA5 protein levels were dramatically reduced throughout the outer and inner root sheath of the patient hair follicles (Figure 6C). Collectively, these results suggest that loss of one genomic copy of ABCA5 and its surrounding regulatory elements severely disrupts expression from the other allele, and significantly reduces expression levels, suggesting haploinsufficiency of ABCA5 in the sporadic case of CGHT.
In this study, we investigated the genetic basis of a case of CGHT using whole-exome sequencing and identified a homozygous recessive loss-of-function mutation in a donor splice site of ABCA5 that cosegregates with the phenotype. The c.4320+1G>C mutation leads to loss of ABCA5 expression and localization within patient keratinocytes, fibroblasts and hair follicles compared to controls. We found that endogenous ABCA5 is highly expressed in both epithelial and mesenchymal compartments of the hair follicle, and that its homozygous recessive loss-of-function is associated with defective lysosomes, as well as an overall accumulation of autophagosomes and endolysosomal cholesterol. In a case of sporadic CGHT, we performed detailed cytogenetic analyses, breakpoint mapping, CNV analysis and expression studies, which revealed a 1.3 Mb deletion encompassing the ABCA5 locus and markedly reduced ABCA5 levels in patient cells and hair follicles.
In inherited hypertrichoses, the mechanism(s) that lead to an excessive hair overgrowth phenotype observed include an increased anagen duration, increased hair follicle density, as well as a vellus-to-terminal transformation, where fine, unpigmented and unmedullated hairs are stimulated to grow deeper into the dermis, widen, and become medullated as well as pigmented [1], [2]. In the case of CGHT, a condition that is present at birth, terminal hairs develop at sites of the body where vellus hairs should be, a developmental event that most likely occurs during late hair follicle morphogenesis or immediately following. Lanugo hairs shed in utero and then replaced by vellus hairs are instead replaced with terminal hairs. In addition to a vellus-to-terminal transformation, it is likely that the excessive hair overgrowth phenotype in CGHT is also contributed by an increased duration of anagen, as the hairs produced across the body are longer than what is considered normal for the age, race, and gender of the individual. In the autosomal recessive CGHT case in this study, we found that the length of hair follicles plucked from the forearm was significantly greater than that of control hair follicles. Moreover, histological analysis of patient skin biopsies from extensor skin of the extremities revealed the presence of large anagen hair follicles situated deep in the dermis, whereas hair follicles in anagen were not present in control skin biopsies.
Using several methods of detection, we found that ABCA5 is strongly expressed in both mesenchymal and epithelial lineages of the human hair follicle. In mouse hair follicles, we observed the same expression pattern as in human follicles, and using immunostaining with a polyclonal ABCA5 antibody, detected Abca5 localization in the testis and epididymis—positive sites of Abca5 mRNA expression. In the skin, the mesenchymal expression (perifollicular dermis, dermal sheath) appears more intense using formalin-fixed paraffin embedded (FFPE) samples, and the epithelial outer and inner root sheath expression is more intense using frozen sections. These differences are most likely artifacts of antigen accessibility and tissue preservation (organic solvent vs. cross-linking agent) between the two methods (i.e. dermal and subcutaneous tissue is better preserved in the FFPE sections). However, RT-PCR analysis and in situ hybridization demonstrated strong ABCA5 expression in both epithelial and mesenchymal lineages, and negative controls for both mRNA and protein studies yielded no signal.
High levels of ABCA5 localization to the hair follicle IRS is intriguing, as several genes involving lipid metabolism or transport possess strong inner root sheath expression and have been described to underlie human genetic conditions with sparse hair. P2RY5, encoding an orphan G protein-coupled receptor, is mutated in patients with autosomal recessive wooly hair syndrome [32], [33]. Likewise, LIPH encodes the phospholipase A1 family member that converts phosphatidic acid into lysophosphatidic acid, and is mutated in patients with autosomal recessive wooly hair and hypotrichosis [32] – [34]. While the precise mechanism by which these genes control hair growth remains elusive, we postulate that the loss of a functional ABCA5 transporter in the IRS of patient hair follicles may contribute to excessive hair overgrowth through blocking cholesterol efflux.
Human ABCA5 is also highly expressed in perifollicular dermal tissue, including the dermal sheath of the hair follicle, and the number of dermal papilla cells has been shown to be important in determining hair follicle size in mice [35]. Since patient hair follicles are larger and longer, we postulate that they may contain an increased number of dermal papilla cells, but were unable to directly test this using the skin biopsies we received. Additionally, the control of hair follicle size is linked to BMP levels in the skin, where mice overexpressing the BMP antagonist, Noggin, have enlarged anagen hair follicles [36]. Therefore, ABCA5 transport of cholesterol may normally limit hair growth and intersect with the BMP signaling pathway, such that in ABCA5-deficient hair follicles, the excess buildup of intra-endolysosomal cholesterol attenuates BMP signaling and recapitulates the enlarged hair follicle phenotype of the Noggin overexpression mouse model. Functional studies using Abca5-/- mice would be required to directly address this.
ABCA5 has not previously been described to be associated with a human condition, particularly one manifesting a hair or skin phenotype. In humans, ABCA5 is a reported biomarker for tumor stem cells in osteosarcoma based on its overexpression, where it is also highly expressed in melanoma and undifferentiated colon and ovarian carcinomas [37] – [39]. Consistent with the reported role of ABC transporters in tumor biology conferring resistance to drugs and chemotherapy/cancer-related substrates (i.e. phospholipids and cholesterol) [40], ABCA5 upregulation may protect undifferentiated tumor cells from an intracellular accumulation of cholesterol and other sterols.
Mutations in ABC transporters have been reported in several other genetic skin diseases, including ABCA12 in lamellar ichthyosis type 2 as well as the severe and lethal ichthyosis, Harlequin ichthyosis, ABCC9 in Cantú syndrome, and ABCC6 in the connective tissue disorder pseudoxanthoma elasticum (PXE) [41] – [46]. ABCA12 is expressed in the skin and its deficiency leads to abnormal lipid transport in keratinocytes, effectively compromising the skin barrier in a cell-autonomous manner [46]. Recently, dominant missense mutations have been reported in the ABC transporter gene, ABCC9 in several cases of Cantú syndrome, a form of congenital hypertrichosis segregating with osteochondrodysplasia, distinctive facial features, as well as cardiac defects [44], [45]. The ABCC9 gene encodes a transmembrane protein that is a constituent of a potassium channel complex, and electrophysiology experiments revealed that the missense mutations have gain-of-function or activating effects on the protein, as they reduce ATP-dependent inhibition of the channel [44]. Interestingly, the ABCC6 gene encodes the multidrug resistance associated protein 6 (MRP6), and is exclusively expressed in the liver and kidney, suggesting that mutations lead to ectopic mineralization within the skin of PXE patients via a non cell-autonomous mechanism [47]. Importantly, this suggests that blocking the allocrite, or transported substrate, is the key determinant of the phenotype, because ABCC6 itself is not expressed in the skin, yet its loss-of-function manifests with a cutaneous phenotype.
The allocrite of ABCA5 has remained unknown, however, we have observed a cholesterol transport defect in human CGHT keratinocytes and mouse Abca5 has previously been reported to play a role in cholesterol efflux in macrophages [16]. In rats and now mice, Abca5 localizes to the Leydig cells of the testis, a primary site for cholesterol processing and sterol hormone synthesis [21], suggesting cholesterol is a potential allocrite for this transporter. Importantly, these reports as well as the subcellular localization pattern of Abca5 are consistent with our observation that ABCA5 deficiency in CGHT patient cells causes a redistribution of free cholesterol and accumulation of this sterol in the endolysosomal compartment. The resulting sequestration of cholesterol within these organelles may prevent the delivery of this essential lipid to other cellular compartments, such as the plasma membrane, altering their physical and signaling properties. Alternatively, accumulation of endolysosomal cholesterol may decrease the overall fitness of these organelles, leading to a decrease in the degradative capacity of lysosomes and causing secondary defects in autophagy.
Cholesterol as a covalent modification is required for the biological activity of several signaling proteins in the Wnt and Shh pathways that regulate hair follicle morphogenesis and cycling [48]. Overexpression and knockout mouse models for such molecules have demonstrated critical roles in hair follicle patterning and possess phenotypes reminiscent of hypertrichosis. For example, mice deficient for the Smoothened (Smo) gene in Keratin-14-expressing epithelia develop de novo and abnormally large hair follicles from increased mesenchymal SHH signaling [49], and the overexpression of a stabilized beta-catenin (Ctnnb1) induces early placode formation as well as an increased density of hair follicles [50]. We postulate that decreased free cholesterol in ABCA5 deficient CGHT cells may modulate the range and/or morphogenetic activity of such proteins either in the epithelium or mesenchyme during hair follicle patterning. Future experiments employing lineage-specific deletions of ABCA transporters (i.e. Abca1, Abca5) will be useful in testing this, where free cholesterol levels can be measured biochemically and by filipin immunostaining, followed by a thorough analysis of hair follicle morphology and density. Similarly, a pharmacological approach to deplete sterols (i.e. cyclodextrin treatment) or block ABCA transporter activity (i.e. glyburide treatment) in pregnant dams during hair follicle morphogenesis may provide insight into the role of cholesterol transport in hair patterning and growth.
In mice, the biogenesis of cholesterol and other lipids has been shown to play a role in regulating hair growth, as mice deficient for Stearyl-CoA Desaturase I (SCD1), the enzyme required for the biosynthesis of monounsaturated fatty acids that compose cholesterol esters as well as membrane phospholipids and triglycerides, develop alopecia and possess a sparse hair coat with dry, scaly skin [51] – [53]. Furthermore, topical treatment of mouse skin with a cholesterol biosynthesis inhibitor of a sterol precursor leads to hair loss through the induction of catagen [54]. It remains to be determined whether Abca5-/- mice possess a hair phenotype, since this has not been directly examined and hair phenotypes in mutant mice deficient for several factors can be subtle in the absence of detailed histological analysis [55]. If hair patterning is aberrant in Abca5-/- mice, lineage-specific deletions of Abca5 followed by morphometric and densitometric analysis of hair follicles may provide insight into the target cell type underlying the pathology in CGHT.
In the sporadic CGHT case, we found that ABCA5 transcripts were significantly reduced. One possible explanation for this is that a trans mechanism for regulating ABCA5 gene expression via an enhancer or some other regulatory element is in place, such that deletion of one copy of ABCA5 and its regulatory elements abrogates gene expression on the other allele. Conversely, ABCA5 may be imprinted or expressed preferentially from one allele, in which case, deleting the transcriptionally-active allele would result in loss of gene expression, thereby recapitulating the phenotype of the loss-of-function mutation. Alternatively, ABCA5 may be a dosage-sensitive gene, where altering the copy number is sufficient to produce a phenotype, but perhaps in a distinct manner from the loss-of-function mutation identified in the present study, since carriers are unaffected. Since we observed a decrease in SOX9 expression in sporadic CGHT patient keratinocytes, consistent with our previous findings in an autosomal dominant CGHT case [4] and considering the known role of SOX9 in hair follicle stem cells [30], [31], it is likely that the excessive hair overgrowth phenotype observed in sporadic CGHT may be the result of reduced levels of both ABCA5 and SOX9 transcripts. We did not detect altered SOX9 expression in the autosomal recessive CGHT case, suggesting either that SOX9 acts upstream of ABCA5 or that the two genes reside in separate pathways, employing distinct mechanisms to contribute to the CGHT pathology.
Importantly, this is the first report of the ABCA5 gene to be associated with a human genetic condition, and the first reported case of CGHT in which a point mutation in a single gene contributes to the excessive hair overgrowth phenotype. Collectively, our findings provide insight into the mechanisms by which ABCA5 may directly or indirectly act to control hair growth at the subcellular level. While there is evidence to support a role for ABCA5 in both epithelial and mesenchymal compartments of the hair follicle, the precise cell type(s) deficient for ABCA5 that is associated with excessive hair overgrowth remains unclear. In light of our findings, we postulate that the loss-of-function of ABCA5 during morphogenesis abrogates lysosome function and cholesterol efflux in hair follicle cells, leading to the aberrant activity of one or more downstream signaling pathways with crucial, defined roles in hair follicle induction, development and cycling.
Informed consent was obtained from all subjects and approval for this study was provided by the Institutional Review Board of Columbia University in accordance with the Declaration of Helsinki Principles.
In the autosomal recessive CGHT case, the proband is an eleven-year-old girl with congenital generalized hypertrichosis segregating with gingival hyperplasia and epilepsy. She was the product of a full-term uncomplicated pregnancy to a then 18-year old Yemeni mother whose husband is her first cousin. The patient's initial genetic evaluation was at 6 months of age; chromosome analysis revealed 46, XX normal complement and SNP arrays revealed multiple regions of homozygosity consistent with parental consanguinity. Severe hypertrichosis and a low anterior hairline that merged with the eyebrows were observed, and excessive body hair was present on the patient's back and extremities, as well as on the upper lip, genitalia, and axillary regions. Hair overgrowth progressed on her face trunk, eyebrows and eyelashes, and scalp as well as body hair is long and coarse. The gingival hyperplasia became evident at 18 months of age and progressed to the extent that it interfered with tooth eruption and feeding. Partial gum surgical resection was initially performed at 3 years of age and has been performed several additional times over the past eight years due to excessive gum overgrowth. Extensive endocrine and metabolic workups have been performed over the years and all results have been normal.
In the sporadic CGHT case, the patient initially presented to the clinic at the age of three months. On physical exam, she had universal hypertrichosis that was accentuated on the extremities, the back, chest, and the face. As expected, regions that are devoid of hair follicles including the palms, soles and distal phalanges were spared. In addition, the patient had unusual facial features, including a broad base of the nose, widely set eyes, and bulbous tip of the nose. The parents reported that the excessive body hair was shed during the first month of life, but then grew back with increased length. The mother did not report any drug intake, major illness or trauma during pregnancy, and the girl was born at full term. There is no history of hypertrichosis in the family. Extensive physical examination, radiological studies including X-rays (skull, long bones and hands), laboratory evaluations (complete blood count, blood chemistry, general urine examinations) and abdominal ultrasonography did not reveal any abnormalities.
In the autosomal recessive CGHT case, two small punch biopsies from extensor skin of the extremities, 3–5 cc of blood, in addition to plucked hairs from the forearm were received from the proband (II-1) and the unaffected father (I-1). For each individual, one biopsy was divided into two pieces: one flash frozen for RNA extraction and the other embedded in OCT for histological and immunological analyses. The other piece of whole skin was used for the culturing of keratinocytes and fibroblasts. In the sporadic CGHT case, a skin biopsy from the back of the affected individual was obtained; 3–5 cc of blood for DNA and RNA extraction was also obtained from the patient and her unaffected parents. Control hair follicles used in expression studies were obtained either from forearm skin biopsies or from occipital scalp biopsies, designated as non-human subject research under 45 CFR Part 46 and therefore IRB exempted for research.
Full exome sequencing, bioinformatics analysis (consisting of initial data processing, base calls, alignments, variant calls, nucleotide and amino acid conservation, biochemical nature of amino acid substitution, population frequency, and predicted functional impact), and filtering based on autosomal and X-linked dominant and recessive inheritance models was performed on the autosomal recessive CGHT case using genomic DNA from the proband (II-1), the unaffected mother (I-2), and the unaffected father (I-1) through Ambry Genetics (Allso Viejo, CA). Evaluation of relationships between the proband and unaffected parents was performed using Short Tandem Repeat markers and samples were prepared using the SureSelect Target Enrichment System (Agilent Technologies). In brief, genomic DNA was sheared, adaptor ligated, and PCR amplified, followed by incubation with the exome baits, elution, and then PCR amplification. Libraries were generated, quantified, and hybridized to the Illumina HiSeq 2000 flow cell for paired-end sequencing. Bioinformatics analysis was used for initial data processing, base calls, alignments as well as variant calls, and the Ambry Genetics Variant Analyzer tool (AVA) was used to determine conservation of nucleotide and amino acids as well as biochemical nature of substitutions, population frequency, and predicted functional impact. Various genome databases (Human Genome Mutation Database (HGMD), HapMap data, Single Nucleotide Polymorphism database (dbSNP), 1000 genomes) were used to search for previously described mutations and/or polymorphisms, and co-segregation studies were performed for candidate gene mutations. Full-exome sequencing revealed a total of 53 genes (85 alterations), and filtering based on the following criteria was used: deleterious nature of the alteration (34 genes, 5 unique alterations), removal of alterations clearly unrelated to patient's phenotype (26 genes, 33 alterations), and further interpretive analysis based on literature searches for genotype-phenotype correlation resulted in the selection of 3 genes (3 alterations) for further investigation.
Genomic DNA was isolated from whole blood of the proband (I-1) and father (II-1) from the autosomal recessive CGHT case as well as from the patient with sporadic CGHT. Total RNA was extracted from forearm skin biopsies as well as from cultured keratinocytes and fibroblasts from family members I-1 and II-1 using the Qiagen RNeasy RNA extraction kit and standard methods.
PCR amplification of the exon 32-intron 32 boundary was performed in the autosomal recessive CGHT case using standard conditions and the following primers: F: 5′-GAACATCTTCAGAAGACTGTAAAG-3′ and R: 5′- GTAATCTGAGGATTCCCTAGCATAC -3′. To amplify and sequence the mutant ABCA5 transcript, the following primers were used: F (in Exon 28): 5′-GCTGATGGGTTGCCAGTGTTGTGAAG-3′, and R (in Exon 33): 5′-CACATGTGCTGTTTGGCTTTGGGATC-3′. For sequencing of ABCA5 in the sporadic CGHT case, 100–200 ng gDNA was used and primers were designed to flank the intron-exon boundaries 100–150 bp from each exon.
Quantitative PCR was performed on genomic DNA from the sporadic CGHT case to confirm the 1.3 Mb deletion identified; three amplicons on chromosome 17 were tested using Relative Quantification and the following primers: Amplicon 1 (chr17: 66639201-66639359, 159 bp): F: 5′-TAGATCATTCTCCTAAATGCTCTTCC -3′, R: 5′-GATGCAGCAAAGTTCTCAGGTG -3′; Amplicon 2 (chr17: 66958719- 66958932, 214 bp): F: 5′- GCTGAGCCTCTCCTGAAAACTGGACAAC -3′, R: 5′- GACTCAACTGACATAGGCCATGACAG -3′; Amplicon 3 (chr17: 67249799- 67249985, 213 bp): F: 5′- GAACATCTTCAGAAGACTGTAAAG -3′, R: 5′- CTGAGGATTCCCTAGCATACTTAGAGC -3′. Amplicon 4 (chr17: 67716661- 67716904, 244 bp): F: 5′- ACCATGTAAACAAGGAAAACAAC -3′, R: 5′- CTGAGGATTCCCTAGCATACTTAGAGC -3′. Amplicon 5 (chr17: 68403395- 68403416, 244 bp): F: 5′- CATTTATCCATATGGGAGGTAG -3′, R: 5′- AACAGATGTCCAAGAGAGTCAAATC -3′. Values were normalized to the β2M amplicon (161 bp) using the following primers: F: 5′- CACCTATCCCTGTTGTATTTTATCG -3′, R: 5′-CTCTTTATTTCTGCTGAGGTTTT-3′. All coordinates reference UCSC human reference genome build hg19.
Quantitative RT-PCR was performed on RNA isolated from whole skin, cultured keratinocytes and fibroblasts as previously described [5], [10] and as per the manufacturer's instructions. Relative quantification using the ddCT method [56] was performed with the β2M gene as the housekeeping control. The following primers were used for qRT-PCR assays: ABCA5: F: 5′- GAACCAACTTCAGGCCAGGTATT-3′, R: 5′-CACATGTGCTGTTTGGCTTTGGGATC -3′; β2M: F: 5′-GAGGCTATCCAGCGTACTCCA -3′, R: 5′- CGGCAGGCATACTCATCTTTT-3′; SOX9: F: 5′- AGTACCCGCACTTGCACAA-3′, R: 5′- CCGTTCTTCACCGACTTCCT-3′. GAPDH: F: 5′-GGAGCGAGATCCCTCCAAAAT -3′, R: 5′- GGCTGTTGTCATACTTCTCATGG-3′
All experiments were performed in triplicates using materials from affected, carrier and control individuals. Experiments were repeated in triplicates, and a Student unpaired t-test was used to determine statistical significance in qPCR and qRT-PCR experiments with a p value of 0.05 as the cutoff value for significance. For autophagy experiments, quantification statistical analysis was performed using a two-tailed, equal variance Student's t-test. P-values of <0.05 (*), <0.01 (**), <0.001 (***) were determined to be statistically significant. The number of cells analyzed per condition is as follows: control, no bafilomycin: 61 cells; control, bafilomycin: 63 cells; affected, no bafilomycin: 37 cells; control, bafilomycin: 43 cells.
Keratinocytes and fibroblasts were isolated and cultured from whole skin biopsies for the proband (I-1) and unaffected father (II-1) from the first case, and the patient from the second sporadic CGHT case using the method described in [10].
The DIG labeling system (Roche) was used to construct the sense and antisense riboprobes for hABCA5 sequences, amplified using the following primers: hABCA5 (Exon 13–19, 812 bp): F: 5′- GTGCAGAAGGTTTTACTAGATTTAGACA-3′, R: 5′- GTCTGGAACAAGTTTGATGGGAACCAC-3′; hABCA5 (Exon 28–33, 550 bp): F: 5′-GCTGATGGGTTGCCAGTGTTGTGAAG-3′, R: 5′-CACATGTGCTGTTTGGCTTTGGGATC -3′. For human expression studies, in situ hybridization was performed on 10 µM skin and hair follicle sections from control, carrier and affected individuals using the methods described in [10].
Immunohistochemistry was performed on formalin-fixed paraffin-embedded (FFPE) sections of whole skin and hair follicles. In brief, slides were deparaffinized and rehydrated with a series of ethanol washes and then with 1X TBS. Antigen retrieval was performed for 10 minutes in a 1M sodium citrate, pH 6.0 solution heated to 95°C, and then slides were cooled and washed three times in 1X TBS. Tissues were blocked in either 10% normal donkey serum (Jackson ImmunoResearch, PA, USA) or 2% fish skin gelatin (Sigma Aldrich, MO, USA) and incubated with primary antibody in 1X TBS at 4°C overnight. The anti -rabbit ABCA5 antibody (ab99953, Abcam) was used at a concentration of 1:200 and the anti -rabbit IgG isotype primary antibody (Santa Cruz Biotechnologies, CA, USA) control was used at the same concentration. Slides were then washed with 1X TBS, incubated with the goat anti -rabbit biotin-conjugated antibody (1:800 in 1X TBS) for an hour at room temperature, washed again with 1X TBS and then incubated with the streptavidin-alkaline phosphatase (AP) tertiary antibody (Invitrogen; 1:300 in 1X TBS) for 30 minutes at room temperature. The SIGMA FAST Fast Red TR/Naphthol AS-MX Tablets (Sigma Aldrich, MO, USA) were used to develop the slides, which were then mounted with Dako Glycergel Mounting Medium (Dako, CA, USA).
Immunofluorescence staining was performed on whole skin 10 µm sections from both human and mouse embedded in Optimal Cutting Temperature (O.C.T.). Slides were fixed with 50% MeOH/50% Acetone for 10 minutes at −20°C, washed with 1X PBS, and then blocked with 2% fish skin gelatin (Sigma Aldrich, MO, USA). The ABCA5 (Abcam, ab99953) and anti-rabbit IgG (Santa Cruz Biotechnologies) primary antibodies were used at a concentration of 1:200. Slides were then washed, incubated with the Alexa Fluor 488 donkey anti-rabbit IgG (Molecular Probes, Invitrogen) secondary antibody (1:800 in 1X PBS), mounted with VECTASHIELD mounting medium with DAPI (Vector Laboratories, Burlingame, CA, USA), and imaged using a LSM 5 laser-scanning Axio Observer Z1 confocal microscope (Carl Zeiss).
For autophagy analysis, human keratinocytes were grown to confluence, seeded on 12 mm coverslips, and then fixed with 4% paraformaldehyde for 20 min at room temperature. After permeabilization with 200 µg/ml digitonin (Invitrogen) in PBS for 10 min, cells were incubated with the specified primary antibodies for 1 hr at room temperature. Subsequently, cells were incubated with the appropriate Alexa-Fluor-conjugated secondary antibodies for 1 hr at room temperature, for cholesterol staining cells were additionally incubated with Filipin complex (Sigma) for 1 hr at room temperature. Images were acquired by confocal laser scanning microscopy (Zeiss LSM-700) and analyzed with Zeiss Zen and ImageJ Software (NIH). The number of LC3-positive compartments and their surface areas (expressed as number of pixels per field) were normalized to the number of cells in each field. The average size was obtained by dividing the surface area of the LC3-positive compartment (in pixel2) by the number of LC3 puncta. Similar measurements were made for Lamp1 and p62 compartments. Primary antibodies used for immunofluorescence: mouse anti-LC3 (MBL), guinea pig anti-p62 (Progen), rabbit anti-Lamp1 (Abcam) mouse anti-Lamp2 (Santa Cruz). Bafilomycin A1 (50 nM, Wako) was added to the media 2 hours prior to cell fixation.
Immunoblotting was performed using 10 µg total protein extracted with RIPA buffer/proteinase inhibitor cocktail from cultured keratinocytes and fibroblasts and SDS-PAGE was used to separate proteins, followed by a wet transfer to a hybond ECL nitrocellulose (Amersham, NJ, USA) or PVDF membrane (BioRad, CA, USA) membrane. All membranes were blocked with 5% milk for one hour at room temperature and then incubated with the primary antibodies, ABCA5 (1:500) and beta-actin (1:1000; Santa Cruz Biotechnologies, CA, USA) diluted in Washing Buffer (1X PBS 0.1% Tween-20). Membranes were then washed several times in 15-minute intervals with 1X PBST, incubated with goat anti-rabbit or mouse- HRP conjugated secondary antibody (Invitrogen; 1:1000), and then developed with the SuperSignal West Dura Extended Duration Substrate (Thermo Scientific, IL, USA).
G-banding analysis was performed using standard techniques, and FISH using chromosomal paint was performed on metaphase chromosomes obtained from peripheral blood leukocytes (PBLs), as per the manufacturer's directions (VYSIS). Sub-telomeric FISH was performed using a mixture of probes for chromosomes 17q, 17centromere, 9p, and 9q obtained from VYSIS in accordance with the manufacturer's instructions. Chromosomes were counterstained with DAPI (VYSIS) and hybridized metaphase chromosomes were viewed using a Nikon microscope fitted with a filter wheel and Cytovision Applied Imaging software.
Breakpoint mapping was performed using FISH with BAC clones in the chromosome 17q24 region. To fine map the deleted region on chromosome 17q, genome scanning was initially performed using the Affymetrix 500k whole-genome mapping array, and then the 2.7M array (Affymetrix) to further examine CNVs in the region of the breakpoint.
|
10.1371/journal.pgen.1003260 | Single Transmembrane Peptide DinQ Modulates Membrane-Dependent Activities | The functions of several SOS regulated genes in Escherichia coli are still unknown, including dinQ. In this work we characterize dinQ and two small RNAs, agrA and agrB, with antisense complementarity to dinQ. Northern analysis revealed five dinQ transcripts, but only one transcript (+44) is actively translated. The +44 dinQ transcript translates into a toxic single transmembrane peptide localized in the inner membrane. AgrB regulates dinQ RNA by RNA interference to counteract DinQ toxicity. Thus the dinQ-agr locus shows the classical features of a type I TA system and has many similarities to the tisB-istR locus. DinQ overexpression depolarizes the cell membrane and decreases the intracellular ATP concentration, demonstrating that DinQ can modulate membrane-dependent processes. Augmented DinQ strongly inhibits marker transfer by Hfr conjugation, indicating a role in recombination. Furthermore, DinQ affects transformation of nucleoid morphology in response to UV damage. We hypothesize that DinQ is a transmembrane peptide that modulates membrane-dependent activities such as nucleoid compaction and recombination.
| Exposure of the bacterium Escherichia coli to DNA damaging agents induces the SOS response, which up-regulates gene functions involved in numerous cellular processes such as DNA repair, cell division, and replication. Most of the SOS regulated genes in E. coli have been characterized, but still there are several genes of unknown function. One of these uncharacterized genes is dinQ. In this work we characterize dinQ and two novel small RNAs, agrA and agrB, that regulate expression of dinQ. The DinQ peptide is localized in the inner membrane as a single transmembrane peptide of 27 amino acids. Small proteins of less than 50 amino acids are important in cellular processes such as regulation, signalling, and antibacterial action. Here we demonstrate that DinQ modulates recombination and transformation of nucleoid morphology in response to UV damage. Our results provide new insights into small hydrophobic peptides that could regulate important DNA metabolic processes dependent on the inner membrane of the cell.
| Exposure of E. coli to DNA damaging agents induces the SOS response, which is under control of the RecA and LexA regulatory proteins. The SOS response upregulates gene functions involved in numerous cellular processes such as nucleotide excision repair (NER), UV induced mutagenesis, recombination, inhibition of cell division and replication. The LexA repressor downregulates more than 50 SOS genes by binding to the operator sequence in their promoter regions [1], [2]. SOS inducers (e.g. UV) cause replication blocks and generate RecA/ssDNA nucleoprotein filaments that mediate auto-proteolysis of the LexA repressor.
Both NER and recombination are required to maintain DNA integrity. NER repairs numerous lesions introducing helical distortions, in which UvrA, UvrB and UvrC work in sequential steps to recognize and remove the lesion. The RecBCD complex is the major component for initiation of recombinational repair (RR) of DNA double strand breaks (DSBs) by processing a blunt dsDNA end into a dsDNA molecule possessing a 3′-terminated ssDNA tail. As part of this process RecBCD mediates RecA filamentation required for presynaptic processing of dsDNA ends.
Most of the characterized LexA regulated genes play important roles in the physiology of E. coli, but there are still several genes of unknown function. One of these uncharacterized genes is dinQ, which is located in the 823 bp region between arsR and gor, 78.58 min on the E. coli chromosome. DinQ is predicted to encode an open reading frame (ORF) of 49 aa 139 nt downstream from a LexA operator sequence [1]. Small proteins of less than 50 amino acids are important in cellular processes such as regulation, signalling and antibacterial action [3]–[5]. More than 50 chromosomally encoded small proteins with a validated expression of less than 50 aa have been identified so far in E. coli [6]–[10]. Several of the newly discovered peptides are hydrophobic single transmembrane helices belonging to toxin-antitoxin systems.
In this work we characterize the arsR-gor intergenic region in which an endonucleolytic product of dinQ is translated into a small hydrophobic peptide of 27 aa. DinQ is under LexA control and antisense regulation by a novel small RNA, agrB. DinQ is localized in the inner membrane as a single transmembrane peptide that modulates nucleoid compaction and conjugal recombination.
The SOS inducible dinQ gene in E. coli was identified in a search for new LexA regulated genes [1]. The dinQ gene was found in the 823 bp arsR-gor intergenic region (Figure 1A), encoding a ∼330 nt transcript with a putative LexA operator sequence (heterology index (HI) = 4.83) in the promoter region and two putative ORFs of 18 and 49 amino acids. No biological function, phenotype or significant homologies to proteins with known function were associated with DinQ [1]. Except for the dinQ gene, no other genes have been reported in the arsR-gor intergenic region. We performed a search for promoter and transcriptional terminator sequences in the arsR-gor intergenic region. As expected this search identified the dinQ LexA operator sequence identified earlier in a screen for LexA regulated genes in E. coli [1]. However, a second operator sequence for LexA (HI = 13.82) in close proximity to the first was identified (Figure 1A and 1B). Further, we identified putative −10 and −35 sequences corresponding to the dinQ promoter which overlaps both operator sequences, and a putative dinQ terminator sequence a few nucleotides downstream of the translational stop codon of the gor gene (Figure 1B). Finally, the sequence search identified two new small noncoding RNAs, termed agrA and agrB (arsR-gor region gene A and B, respectively), containing consensus like −10 and −35 sequences and rho independent terminator sequences (Figure 1A and 1B). AgrA and agrB are transcribed in the opposite direction of dinQ but encode no putative ORFs. Thirty-one nucleotides at the 5′ end of agrA and agrB show antisense complementarity to dinQ (Figure 1B and 1C). Twenty-five out of 31 nucleotides show complementarity to agrA while 30 out of 31 nucleotides show complementarity to agrB. This putative base pairing is indicative of possible RNA interference with the dinQ transcript. It thus appears that the arsR-gor region contain one protein coding gene, dinQ, and two small non-coding RNAs, agrA and agrB, with antisense complementarity to dinQ.
To examine a potential role for the small non-coding RNAs agrA and agrB in regulating dinQ, three single mutants (dinQ, agrA and agrB), one double mutant (agrAB) and one triple mutant (dinQ agrAB) were generated (Table S1). To estimate the approximate size of the dinQ, agrA and agrB transcripts, northern blots with total RNA isolated from UV exposed (and unexposed) wild type and mutant strains (dinQ, agrA and agrB) were hybridized with radiolabeled riboprobes against the respective genes (Figure 2A). The dinQ probe generates five specific signals (a–e), in which the main transcript (a) migrates according to the expected size of full-length dinQ, ∼330 nt. DinQ-b, -c, -d and -e migrates as transcripts of about 290 nt, 250 nt, 200 nt and 130 nt, respectively, according to the size marker. All signals are absent in the dinQ mutant demonstrating that all five transcripts are derived from dinQ. The full-length dinQ product is 3- and 4.6-fold upregulated in the agrA and agrB mutants respectively under normal growth (without UV exposure). Notably, the dinQ-b signal is 4.8- and 3-fold upregulated in the agrB mutant as compared to the wild type and agrA mutant respectively. The dinQ-c product is not detectable in the agrB mutant. Further, the dinQ transcripts are induced in response to UV in wild type and agrA but not in agrB. These data indicate a regulatory mechanism by RNA interference, in which the agrA and agrB interfere differently with the dinQ transcript. Further, primer extension and 3′ mapping of dinQ RNA revealed transcript starts at 0, +44 and +125 corresponding to the estimated size of dinQ-a, -b and –d, respectively (Figure 2B). In agreement with the northern analysis we find that the +44 primer extension product (dinQ-b) is upregulated in the agrB mutant as compared to wild type and agrA mutant. The primer extension could not identify any products corresponding to dinQ-c or -e. The agrAB probes showed that the agrB transcript migrates slightly slower than the agrA transcript, and none of the transcripts were regulated in response to UV irradiation (Figure 2A, middle panel). The agrA transcript is upregulated 9.5 times in the unexposed agrB mutant as compared to wild type, indicating that the absence of agrB somehow promotes agrA RNA stability or transcription of agrA. Further, primer extension revealed that the sequence of transcription start was identical for the agrA and agrB genes (Figure 2C). 3′ mapping of agrA and agrB showed different transcription stops around the rho terminator (data not shown), indicating that the transcripts are processed/terminated differently at the 3′end. In summary, these data demonstrate that both agrA and agrB downregulate the level of dinQ full-length transcript whereas agrB is particularly important for down regulation of +44 dinQ (dinQ-b).
The 31 nt antisense region in agrB gene (Figure 1C) indicate a function in antisense regulation of dinQ via RNA interference. This antisense sequence is partially complementary in agrA (Figure 1C) suggesting that both the agrA and agrB transcripts could base pair with the dinQ transcript. To ensure the function of dinQ when generating mutants, the deletions of agrA and agrB were made without destroying the dinQ promoter. DinQ belongs to the LexA regulon in E. coli [1] which regulates the SOS response. Several mutants of the SOS response, which play a direct role in DNA repair, display UV sensitivity. To examine the role of the dinQ-agrAB locus in the SOS response we tested the UV sensitivity of the various mutants (Figure 3A). The agrB single mutant and agrAB double mutant showed a significant increase in UV sensitivity compared to the isogenic wild type. In contrast, the agrA and dinQ single mutants and the dinQ agrAB triple mutant showed no UV sensitivity. Further we examined UV survival of the agrB mutant carrying a plasmid expressing the agrB gene (Figure S1), demonstrating that the mutant recovered completely. These data indicate a role for agrB in protection against UV exposure, in which the agrB transcript modifies the dinQ transcript. According to the northern analysis (Figure 2A) agrB represses accumulation of the +44 dinQ/dinQ-b transcript. It thus appears that the +44 dinQ product mediates the UV sensitivity of the agrB mutant.
To further investigate the role of the arsR-gor intergenic region in UV protection, we cloned agrA, agrB and dinQ separately or the entire region containing both small RNAs and the dinQ gene into the cloning vector pKK232-8. Wild type cells transformed with pKK232-8-dinQ or pKK232-8 (control plasmid) showed the same viability during normal growth conditions (data not shown). In contrast, it was not possible to transform pKK232-8-dinQ into the agrB mutant (Table S2), indicating that the endogenous level of agrB in wild type cells is sufficient to inhibit dinQ expression from the pKK232-8-dinQ construct during normal growth. However, wild type transformed with pKK232-8-dinQ showed increased sensitivity to UV as compared to wild type transformed with the control plasmid (pKK232-8) (Figure 3B). When the wild type was transformed with constructs expressing agrA or agrB they did not increase sensitivity of the cells to UV (data not shown). Interestingly, wild type cells transformed with the plasmid carrying the entire arsR-gor locus, expressing all three genes, showed no UV sensitivity, demonstrating that agrAB can neutralize the UV sensitizing effect of dinQ. The agrB mutant (Figure 3A) and the wild type cells transformed with pKK232-8-dinQ (Figure 3B) displayed similar sensitivity to UV. These results suggest that the agrB transcript counteracts the UV sensitivity induced by dinQ expression.
During construction of the agrB and agrAB mutants we observed that they form small colonies when plated on LB agar. To further investigate this growth phenotype we compared the growth rate in LB medium of the agrB, agrA and dinQ mutants and their isogenic wild type. OD600 was measured during growth and a sample was diluted and plated for viable counts. This experiment showed that only the agrB single mutant and agrAB double mutant grow more slowly than the wild type cells (Figure S2A). Also in glucose-CAA medium agrB mutant cells grew more slowly than wild type cells (Figure S2B and S2C). In another set of experiments we utilized flow cytometry to analyze whether DNA replication was affected in the growth impaired agrB mutant. We found that cells were smaller than normal with a reduced DNA concentration (Figure S2D). The total time for replication from origin to terminus was shorter in the mutant and the number of origins and replication forks per cell were fewer compared to the wild type. There was also a considerable heterogeneity in the observed reduction in cellular DNA concentration. This heterogeneity could be due to cell-to-cell differences in expression of the DinQ peptide.
The mechanism underlying the UV sensitive phenotype of dinQ in a multicopy situation or under constitutive upregulation in an agrB mutant is not clear. The dinQ gene contains two putative ORFs in which the second ORF contains three putative start codons (Figure 3C). In this work the corresponding peptides are termed DinQ I-IV. None of the putative DinQ peptides show homology to known peptides. To examine if any of the ORFs mediate the UV sensitivity shown by dinQ, each of the putative DinQ peptides (I–IV) were cloned into the expression vector pET28b(+) and expressed under control of IPTG. DinQ I displayed no increased sensitivity in absence or presence of UV, suggesting that the putative peptide translated from ORF I (Figure 3C) does not induce DinQ toxicity (Figure 3D and data not shown). In contrast, we observed that DinQ peptides II, III and IV showed a strong toxic/growth inhibitory effect even in absence of UV, demonstrating that the C-terminal amino acid sequence translated from start codon IV of the second ORF is sufficient to induce DinQ toxicity (Figure 3D). Next, expression of DinQ II was titrated with increasing concentrations of IPTG in absence or presence of UV (Figure 3E), showing that DinQ is highly toxic to the cells at very low doses of IPTG induction and the toxicity was UV independent.
In another set of experiments we used a coupled in vitro transcription/translation E. coli T7 S30 extract to examine translation from pET28b(+) constructs encoding the putative peptides predicted from DinQ I–IV. Expression of DinQ I could not be detected whereas DinQ II–IV were highly expressed (Figure 4A, lanes 2–5). Notably, extracts with the DinQ IV construct produced two peptides of approximately 7.0 and 5.0 kDa, in which the smallest peptide indicates a fifth start codon. A closer inspection of the dinQ sequence uncovered a Shine Dalgarno motif within the DinQ IV sequence in optimal position to initiate translation at a GTG (termed codon V in Figure 3C), which encodes a putative peptide of 27 aa, termed DinQ V. To examine if the DinQ V peptide induces toxicity we cloned the sequence into pET28b(+), transformed the construct into wild type cells and monitored cell survival under the control of IPTG induction in presence or absence of UV exposure. DinQ V expression induced cell killing independently of UV treatment, suggesting that the sequence of peptide V is sufficient to mediate DinQ toxicity (Figure 3C and 3D). In vitro transcription/translation assays with the pET28b(+) DinQ V construct produced a peptide corresponding to the predicted molecular weight of peptide V (Figure 4A, lane 6).
Northern analysis identified five dinQ transcripts (Figure 2A; dinQ a–e), in which the start site were determined for transcript a, b and d. To assess the translational activity of the in vivo dinQ transcripts a, b and d we synthesized the corresponding PCR products carrying the T7 RNA polymerase promoter and added E. coli T7 S30 extract. Only the dinQ-b/+44 transcript was translationally active, generating a peptide with a molecular weight similar to DinQ V, whereas the other transcripts were translationally inert (Figure 4A, lanes 7–10). Thus, it appears in vivo that the biologically active DinQ peptide (peptide V) is translated from the post transcriptionally modified +44 dinQ RNA.
To examine endogenous expression of DinQ in vivo, a 3×FLAG tag was inserted chromosomally in frame with the C-terminal of the dinQ gene in the wild type and agrB mutant. Western analysis revealed a faint band for the FLAG tagged DinQ peptide in the agrB mutant while the peptide was barely detectable in wild type (Figure 4B, lanes 5 and 6). In UV treated cells the DinQ level was about two fold higher in the agrB mutant as compared to wild type (Figure 4B, lanes 3 and 4). It thus appears that the phenotypes of the agrB mutant are not due to polar effects of extensive overexpression of DinQ.
Further, we introduced a chromosomal stop codon in the Lys4 position of dinQ in the agrB mutant and wild type (Figure 3C, base labeled in red). Survival experiments showed that UV resistance was restored to wild type level in the agrB mutant (Figure 4C), indicating that the UV sensitivity of the agrB mutant is caused by translation of a functional DinQ peptide. Next, we introduced three chromosomal point mutations in the dinQ up-stream sequence predicted to be involved in base pairing with agrB (Figure 3C, bases labeled in red). Exposure of these strains to UV in the agrB mutant background showed wild type levels of survival (Figure 4C). All together these results suggest that DinQ is translated into a peptide of 27 aa and the agrB-dinQ RNA interference is important for correct regulation of DinQ translation.
To determine the intracellular localization of DinQ, western analysis was performed on extracts after subcellular fractionation. As antibodies against the native DinQ peptides could not be obtained, we introduced a 3×FLAG epitope at the N-terminal of DinQ (peptide V). Spot assays on LB agar containing IPTG showed that the FLAG tagged peptide induced the same toxicity as the native peptide, demonstrating that the N-terminal FLAG tag had no effect on DinQ toxicity (data not shown). Cells were harvested at several time points after IPTG induction to test the level of expression by western analysis of whole cell extracts. The FLAG-DinQ peptide could not be detected before induction, but showed strong signals 5 to 40 min after induction (data not shown). To examine subcellular fractionation, antibodies against Lep and TolC were used as positive markers for inner and outer membrane fractions, respectively. The western blot showed that the inner and outer membranes are completely separated whereas the cytoplasmic fraction contains some contamination from the inner membrane (Figure 4D, middle panel). DinQ localized to the inner membrane but could not be detected in the outer membrane (Figure 4D, lower panel). The faint signal of the FLAG epitope in the cytoplasmic fraction is possibly due to cross contamination from the inner membrane. These data suggest that DinQ localization is confined to the inner membrane of E. coli.
Analysis of the DinQ amino acid sequence using the consensus secondary structure prediction tool Jpred3 [11] revealed that DinQ has high propensity to form a single α-helix. All residues except a few on each flanking terminal are predicted with high confidence to belong to the predicted α-helix (Figure 4E). With 20–22 residues in a single α-helix, the DinQ peptide could straightforwardly form a transmembrane helix of 6 full turns spanning more than 30 Å, as shown by modelling of DinQ using a regular α-helical template (Figure 4F). The two positively charged lysine residues (Lys4 and Lys9) are close to the phospholipid head groups, while particularly the charged Glu17, but also Arg20 and Gln24 may form a polar patch that can interact with other membrane embedded proteins (Figure 4F). The predicted single transmembrane peptide supports the localization of DinQ in the inner membrane (Figure 4D).
Previously, we showed that overexpressing another SOS inducible peptide, TisB, which encodes a small toxic inner membrane peptide, inhibits several SOS functions in wild type E. coli [12]. To determine whether DinQ affected induction of the SOS response we measured the level of recA and lexA mRNA in a mutant which constitutively overexpressed dinQ, agrB mutant or a dinQ deletion mutant (Table S1). Exponentially growing cells were exposed to UV and the amount of recA and lexA mRNA were determined prior to, and 20 min after irradiation by RT-qPCR. The expression levels of recA and lexA were similar in both mutants and wild type indicating that DinQ in contrast to TisB is not affecting regulation of the SOS response (Figure S3).
To investigate a potential role of dinQ in filamentation, we stained cell samples with acridine orange prior to and after UV exposure. All strains showed the same filamentation pattern, in which cells displayed short filaments 1 h after irradiation and long filaments after 2.5 h, indicating that DinQ is not involved in the filamentation process of E. coli (Figure S4). Next, we measured spontaneous and UV induced mutagenesis as the frequency of rifampicin resistant colonies in wild type and the dinQ and agrB single mutants. The results showed no significant differences in mutation frequency in the mutant strains as compared to wild type suggesting that DinQ is not altering spontaneous and SOS induced mutagenesis (data not shown).
To examine whether high levels of DinQ induce changes in membrane potential we tested the ability of E. coli cells overexpressing DinQ to take up the dye DiBAC4(3) [bis-(1,3-dibarbituric acid)-trimethine oxanol]. The quantity of dye entering cells is proportional to membrane polarization, the less polarised the membrane the greater the quantity entering the cells and so increased fluorescence intensity due to binding to the membrane and intracellular components [13]. Cells were analyzed 5 and 20 min after IPTG induction of DinQ expression, incubated with DiBAC4(3) for 20 min and analyzed by flow cytometry (Figure 5A). No changes were observed for the plasmid control pET28b(+). However, IPTG induction of DinQ (peptide V) showed a rapid increase in DiBAC4(3) uptake (Figure 5A), suggesting that elevated levels of DinQ depolarize the cell membrane. This data indicates that DinQ overexpression interferes with membrane polarity and could therefore lead to a loss of viability.
Subcellular fractionation of E. coli showed that DinQ is localized to the inner membrane (Figure 4D). DinQ is predicted to be a hydrophobic single transmembrane peptide that might compromise inner membrane integrity (Figure 4F). We speculated that if DinQ affected the proton motive force, it would affect ATP production and intracellular ATP concentration. The intracellular ATP concentration was measured in wild type cells and in the agrB mutant, using a quantitative luciferase-based assay. This experiment showed that the concentration of ATP in the agrB mutant was about 50% of the concentration measured in wild type cells (Figure 5B). Further, UV exposure increased the ATP concentration 0.6 fold in both cell types. Thus, it appears that insertion of the DinQ peptide into the inner membrane of E. coli impairs the energy supply in the form of ATP.
The agrB mutant displayed sensitivity to UV suggesting that DinQ could have a role in the repair of UV induced DNA damage. Both nucleotide excision repair (NER) and recombinational repair (RR) counteract the genotoxic effects of UV irradiation. NER is required for the repair of UV induced photoproducts such as thymine dimers and cyclopyrimidine dimers, while RR is required for the repair of strand gaps and double strand breaks. To examine if DinQ is involved in NER we analysed UV sensitivity of the uvrA agrB and uvrA dinQ double mutants as compared to the single mutants. The survival analysis showed an additive effect between agrB and uvrA (Figure 6A), whereas uvrA and dinQ showed no additional effect (Figure S5). These data indicate that elevated levels of DinQ in the agrB mutant sensitize the cell to UV via a pathway which is independent of NER. To examine the role of DinQ in recombination, mutant strains dinQ, agrB, recB, (recB agrB) and (dinQ recB) were exposed to UV irradiation. The double mutant (recB agrB) was slightly more sensitive to UV than the agrB single mutant (Figure 6A), whereas recB and dinQ showed no additional effect (Figure S5).
To further examine the role of DinQ in recombination we performed Hfr conjugation assays with a donor strain containing Tn10, which carries the tetracycline resistance gene integrated in its chromosome and agrB, dinQ and uvrA single mutants as recipient strains. We used the uvrA mutant as control strain since UvrA is not involved in recombination (and carry the kanamycin resistance gene required to detect the recipient). Hfr conjugation of Tn10 was at least 400-fold more efficient in dinQ and uvrA mutants as compared to agrB (Figure 6B), suggesting that elevated levels of DinQ inhibit recombination. To examine if the conjugational process itself is affected in an agrB recipient we performed plasmid conjugation assays with a donor strain carrying an F′-plasmid with tetracycline resistance and the same recipient strains as in the Hfr conjugation experiment. Hfr conjugation differs from F′-plasmid conjugation in that transfer of genes after Hfr requires recombination whereas the F′-plasmid does not recombine in the recipient. In these experiments we find no differences in plasmid conjugation frequencies between the dinQ, uvrA and agrB recipient strains (data not shown). In sum, these results suggest that recombination is inhibited in the agrB mutant during Hfr conjugation, but not in the transfer and uptake of DNA or survival of the agrB recipient. It thus appears that elevated levels of DinQ affect the recombination process.
In dividing cells, replication forks are stalled by DNA lesions that impair DNA unwinding or block synthesis by the DNA polymerase subunits. In E. coli, UV lesions cause a delay in DNA synthesis for a period of time while stalled forks undergo repair. Fluorescence microscopy of Hoechst stained cells has demonstrated that the DNA often forms a compact structure during this phase, and suggests that the nucleoids undergo a major reorganization after UV exposure [14]. To investigate whether DinQ affects nucleoid organization, we used this technique to examine the shape and size of the nucleoids at different time points after UV exposure. In undamaged cells the nucleoids have characteristic shapes and numbers depending on the growth medium. When grown in glucose-CAA medium most cells have two nucleoids and some (the largest cells) have four (Figure 7A, 0 min). Microscopy of cells 15 min after UV irradiation shows that all the wild type cells had lost the normal nucleoid morphology. In approximately 45% of the cells the nucleoids had been rearranged into a highly compact structure, whereas in the rest of the cells the nucleoids were found to be extended throughout the cells (Figure 7A and 7B). Sixty minutes after UV irradiation all cells were found to contain extended nucleoids. After 90 min approximately 30% of the cells had divided and contained nucleoids with normal morphology. In the agrB mutant the degree of nucleoid compaction was similar to that of wild type cells at 15 min after UV exposure (Figure 7A and 7B). However, in the period from 30 to 90 min 25–30% of the nucleoids of the agrB mutant cells were still locked in a compact state whereas a decreasing number of the wild type cells contained compact nucleoids. The result indicates that the transition from compact to extended nucleoid was inhibited in the agrB mutant.
We also investigated the dinQ mutant with respect to nucleoid morphology after UV irradiation. At the 15 min time point approximately 35% of dinQ mutant cells contained a compact nucleoid compared to about 45% of wild type cells (Figure 7B). This indicates that the compaction process might be affected in cells without DinQ. At 30, 60 and 90 min similar numbers of cells with compact and extended nucleoids were found in the dinQ mutant compared to in wild type cells (Figure 7B). The results indicate that cells lacking DinQ have an impaired ability to form a compact nucleoid structure after UV irradiation. Taken together the data reveals that the presence of DinQ is required in order to execute a transformation of nucleoid morphology in response to UV damage, and that overexpression of DinQ leads to a delay in decompaction and extension of the nucleoid during the later stages of the response.
In conclusion, DinQ is under the control of the SOS response and the agrB antisense RNA, and expresses a single transmembrane peptide that has an effect on nucleoid compaction and when overexpressed on conjugal recombination (summarized in Figure 7C).
Recently, a search for small proteins in E. coli could not detect any translation of the DinQ ORF [6], [7]. In this paper we characterize the arsR-gor intergenic region of E. coli, which contain the SOS inducible dinQ gene and two constitutively expressed small RNAs, agrA and agrB, with antisense complementarity to the dinQ gene. We show that DinQ is tightly regulated at both the transcriptional and translational level. Five different dinQ transcripts were identified in which only the endonucleolytic +44 transcript (dinQ-b) is translationally active. Further, agrB appears to repress accumulation of dinQ-b by RNA interference. Unexpectedly, DinQ is not translated from any of the three ATG codons within the ORF but from an alternative GTG start codon, encoding a 27 aa peptide which is localized in the inner membrane. The agrB mutant, which expresses elevated levels of dinQ-b, displays increased sensitivity to UV induced DNA damage and an impaired frequency of conjugal recombination. It thus appears that DinQ could be involved in the modulation of homologous recombination. Small single transmembrane peptides such as DinQ may be key regulators of processes at the inner membrane, in which their expression are strictly regulated to avoid toxicity. In summary, the experiments presented in this paper provide insights into the complex regulation of dinQ and suggest a mode of action within the bacterial inner membrane (Summarized in Figure 7C).
Previous attempts to identify a translation product for dinQ have been unsuccessful, presumably because transcription and translation of dinQ are strictly regulated. First, the promoter region of the dinQ gene contains two LexA operators with different HI, which may suggest differential expression of the transcript early and late in the SOS response. Second, we identified two novel small non coding RNAs agrA and agrB with sequence complementarity to dinQ in the arsR-gor region that regulate dinQ by RNA interference. Notably, only the agrB RNA repressed the translational active +44 dinQ transcript (dinQ-b) whereas both sRNA repressed the primary translationally inactive dinQ transcript. Further, only the agrB antisense RNA counteracts DinQ toxicity. As such the dinQ-agrAB system appears to conform to the definition of a classical type I toxin-antitoxin (TA) system. It thus appears that the dinQ/agrB complex inhibit the endonucleolytic cleavage producing the active mRNA (dinQ-b). Presumably, these antisense sRNAs have been tandemly duplicated in the genome, in which agrA has partly degenerated and appears to be non-functional due to less antisense sequence complementarity with the dinQ sequence as compared to agrB.
The genomic organization and mode of antisense regulation of dinQ in the arsR-gor region resembles regulation of another SOS induced TA system, tisAB [15], [16]. Similar to the dinQ RNA, endonucleolytic processing of the primary tisAB transcript is required to generate an active mRNA producing the toxic TisB peptide. Further, the tisAB locus contains an antisense RNA, IstR-1 that inactivates the translationally active mRNA by RNaseIII dependent cleavage. It appears that agrB may have a similar role in RNase dependent cleavage of the translationally active dinQ mRNA (dinQ-b). In addition, Darfeuille et al revealed that the antisense RNA IstR-1 inhibits translation of the TisB toxin by competing with standby ribosomes binding upstream of the translation initiation region (TIR). It is proposed that binding to the “standby” site is required for initiation of protein synthesis at the highly structured tisB TIR by ribosome sliding to the transiently open TIR [16]. In a similar manner, we speculate that the agrB antisense RNA regulates/inhibits translation from the active +44 transcript by binding a potential “standby” site upstream of the dinQ TIR.
In order to investigate DinQ biochemically, we have attempted to purify DinQ, including fusion peptides. However, all attempts to purify DinQ as well as chemical synthesis of the peptide failed because of the hydrophobic nature of the peptide. Further, a general feature of small hydrophobic peptides including DinQ is the lack of obvious phenotypes associated with their inactivation [17]–[20]. As an alternative strategy we characterized the phenotype of augmented DinQ in an agrB mutant. The DinQ concentration in the agrB mutant after UV treatment is elevated only two fold as compared to the wild type, indicating that the DinQ levels in the agrB mutant is physiologically relevant. Of particular interest was the 400-fold reduction in the recombination frequency in the agrB mutant as compared to wild type cells, suggesting that augmented DinQ inhibits recombination in the agrB mutant. However, genetic data suggests that DinQ may also play a role in UV protective mechanisms independent of recombination.
It appears that the large, ordered hyperstructures involved in homologous recombination are associated with the cell membrane [21]. The hyperstructures are dynamic and their size is dependent on the extent of the initial or ongoing DNA damage. The DinQ peptide is localized in the inner membrane of the cell and it is tempting to speculate about a role for DinQ in regulating DNA repair hyperstructures at the inner membrane. In addition, the prolonged period of nucleoid condensation in the agrB mutant may contribute to the impairment of DNA repair processes. Although such a direct role for DinQ is speculative, several small hydrophobic peptides have been demonstrated to modulate membrane dependent processes. The B1500 protein (65 aa) interacts with the PhoQ sensor [18], the 30 aa protein MgtR (30 aa) interacts with MgtC [19], the KdpF protein (29 aa) is part of the Kdp complex [9] and the SidA protein (29 aa) interacts directly with FtsW and FtsN [22].
The intracellular concentration of ATP was reduced in the agrB mutant compared to the wild type both before and after UV exposure. In wild type cells UV irradiation induces a two fold increase in ATP concentration during the first 20 to 30 min after exposure, and the increase is RecBC dependent [23]. These data suggest that loss of agrB and thereby excess of DinQ limit the cellular energy supply and may also explain some of the observed phenotypes. Our FLAG tag experiments revealed that DinQ increases only two fold in an agrB mutant and this apparently modest increase is sufficient to mediate dramatic effects on conjugal recombination rates, membrane depolarization, ATP levels and nucleoid reorganization. In a wild type cell population the level of DinQ translation is kept strictly under control by the LexA repressor, antisense agrB RNA and dinQ RNA processing, so for the majority of cells DinQ may never reach a level high enough to mediate the effects observed in an agrB mutant. Heterogeneity in the expression of LexA repressed genes has been observed by studying SOS promoter fusions in combination with imaging techniques and a subpopulation of cells clearly have a stronger SOS induction [24]–[26]. It is tempting to speculate that a higher level of DinQ is reached only in a subpopulation of cells where SOS induction is particularly strong or long lasting leading to a permanent or temporary agrB phenotype. Such an effect has been proposed for some toxin/antitoxin pairs in promoting formation of persister cells [27], [28]. To gain a more detailed knowledge about the biological function of DinQ the agrB mutant could be an excellent model for studying the effects of DinQ and similar hydrophobic peptides in bacterial subpopulations.
The experiments were carried out in an AB1157 background [29]. Except for chromosomal point mutations and chromosomal 3×FLAG tags all mutants were made in strain BW25113-pKD46 [30] and introduced into AB1157 via T4GT7 transduction [31]. The agrA (BK4042), agrB (BK4043) and dinQ (BK4040) single mutants were made by deleting each of the genes and introducing a kanr cassette. Next, the agrAB double mutant (BK4041) was generated by deleting both genes and introducing a kanr cassette. To construct a triple mutant the entire arsR-gor intergenic region containing dinQ, agrA and agrB was deleted (BK4044) and replaced with the kanr cassette. Table S1 summarizes all strains used and generated in this work. Vector pKK232-8 (10–25 copies pr cell in E. coli) contains a promoter less cat gene allowing selection of DNA fragments containing promoter activity [32]. pBK440 (dinQ-agrAB)/pBK444 (dinQ) is based on the vector pKK232-8 (Pharmacia) with a 2065/415 bp insert respectively from the intergenic region between arsR-gor in MCS, resulting in a plasmid that expresses E. coli dinQ from its own SOS inducible promoter. Cloning primers are listed in Table S3. Expression plasmids pET28b(+)-DinQ I, pET28b(+)-DinQ II, pET28b(+)-DinQ III, pET28b(+)-DinQ IV and pET28b(+)-DinQ V contain the DinQ I to V ORFs inserted in the NcoI-BamHI restriction sites of the pET28b(+) vector (Novagen). Chromosomal point mutations in dinQ to either introduce a premature translational stop codon in DinQ ORFV (K4stop) or introduce three point mutations in the agrB antisense region of dinQ (A108T, C112G, A115G) or to introduce a chromosomal DinQ C-terminal 3×FLAG tag were made by splicing PCR products with overlap extension (SOEing PCR) and recombine the final SOEing PCR product into a MG1655 background as described [30]. All SOEing products contained a flanking kanr cassette close to the arsR gene to facilitate selection of recombinants. To avoid unwanted recombination between the kanr cassette and the point mutations or the 3×FLAG tag during strain construction the SOEing products were transformed into strain BK5444-pKD46 which lacks the chromosomal dinQ-agrAB locus and where insertion/recombination of the SOEing products is possible only in the flanking homologous DNA sequences. The final PCR products were transformed into MG1655 containing pKD46. Cells were cured for pKD46 and insertions verified by PCR and sequencing. Details of strain construction and oligos used are listed in Tables S1 and S3, respectively. GenScript Corp. gene service constructed DinQ II and V with an N-terminal 3×FLAG tag that was inserted in the NcoI-BamHI restriction sites of the pET28b(+) vector (Novagen). Cells were grown in LB- or K-medium [33] with appropriate antibiotics (100 µg/ml ampicillin and 50 µg/ml kanamycin). For the nucleoid compaction studies cells were grown in AB minimal medium [34] supplemented with 1 µg/ml thiamine, 0.2% glucose and 0.5% casamino acids.
In vitro transcription/translation on circular pET28b(+) templates or linear PCR products were performed according to Promegas protocols E. coli T7 S30 Extract System for Circular DNA and E. coli S30 Extract System for Linear Templates, respectively, with [14C]-Leucine as radiolabeled amino acid. The translation products were analysed by SDS-PAGE and visualized on Typhoon 9410 (Amersham).
Aliquots of exponentially growing ER2566/pET28b(+)-DinQ V were harvested by centrifugation 20 min after IPTG induction (1 mM). Cells were resuspended in 4 ml 50 mM phosphate buffer pH 7.2 and sonicated three times for 15 sec. Further fractionation was performed as described by [20]. Proteins from all fractions were acetone precipitated 1∶1 overnight at -20°C, pellets after centrifugation was resuspended in 4× NuPAGE sample loading buffer (Invitrogen) and loaded onto 10% NuPAGE Novex Bis-Tris gels (Invitrogen).
Cells were grown to OD600≈0.4 in LB and induced with IPTG (1 mM). At 0, 5 and 20 min culture samples were diluted 1∶10 in filtered AB minimal medium [34] +10 µg/ml DiBAC4(3) (Sigma-Aldrich). After 20 min incubation in the dark at room temperature, cells were analysed in a Flowcytometry LSRII (Becton Dickinson) equipped with an argon ion laser and a krypton laser (both Spectra Physics). DiBAC4(3) was detected using 488 nm laser. The distribution of DiBAC4(3) fluorescence was plotted on a logarithmic scale. The data obtained was analyzed by winMDI software.
Cell aliquots were harvested before and 20 min after induction with IPTG (1 mM) and washed once in 50 mM Tris-acetate pH 7.75. ATP was extracted from washed cells by 1% trichloroacetic acid (TCA) in 50 mM Tris-acetate pH 7.75 for 10 min. Tris-acetate pH 7.75 was added 1∶10 to obtain optimal pH of 7.75 before mixing with rL/L reagent (ENLITEN ATP assay, Promega) at room temperature. The amount of ATP extracted (RLU value) was measured with 20/20 Luminometer (Turner Designs) and related to the OD600 for each sample.
Aliquots of exponentially growing recipient strains dinQ (BK4040), agrB (BK4043) and uvrA (BK4180) were mixed in equal volumes with donor strains BW7623 (with the tetracycline resistance gene, Tn10, integrated in its chromosome) or ER2738 (carrying a tetracycline resistance conjugative plasmid) and incubated at 37°C for 30 minutes. BW7623 was used to examine chromosomal transfer to the recipient strains (recombination dependent) whereas ER2738 was used for plasmid conjugation. Cells were vortexed thoroughly and spread on selective LB plates. Hfr recombination rate and plasmid conjugation rate was calculated as number of recombinants/conjugated cells pr 106 cells.
Exponentially growing wt (AB1157), dinQ (BK4040) and agrB (BK4043) cells were UV irradiated with 3 J/m2 while stirring. 1.5 ml samples were taken at 0, 15, 30, 60 and 90 min after irradiation. Washed once and resuspended in 100 µl cold, filtered TE buffer. Then 1 ml of cold, filtered 77% ethanol was added for fixation. Fixed cells were mounted on a poly-L-lysine coated microscope slide and the DNA was stained with Hoechst 33258 (5 µg/ml, Sigma) in mounting medium (40% glycerol in PBS pH 7.5). Visualization of stained cells was performed using a Leica DM6000B phase-contrast/fluorescence microscope equipped with a 63× objective and a BP340-380 excitation filter. Pictures were taken using a Leica DFC350 FX digital camera that was connected to a computerized image analysis system (LAS AF software, version 2.0.0, Leica). The fluorescent image was merged with the phase-contrast image.
|
10.1371/journal.pntd.0004036 | In Vivo MRI Assessment of Hepatic and Splenic Disease in a Murine Model of Schistosmiasis | Schistosomiasis (or bilharzia), a major parasitic disease, affects more than 260 million people worldwide. In chronic cases of intestinal schistosomiasis caused by trematodes of the Schistosoma genus, hepatic fibrosis develops as a host immune response to the helminth eggs, followed by potentially lethal portal hypertension. In this study, we characterized hepatic and splenic features of a murine model of intestinal schistosomiasis using in vivo magnetic resonance imaging (MRI) and evaluated the transverse relaxation time T2 as a non-invasive imaging biomarker for monitoring hepatic fibrogenesis.
CBA/J mice were imaged at 11.75T two, six and ten weeks after percutaneous infection with Schistosoma mansoni. In vivo imaging studies were completed with histology at the last two time points. Anatomical MRI allowed detection of typical manifestations of the intestinal disease such as significant hepato- and splenomegaly, and dilation of the portal vein as early as six weeks, with further aggravation at 10 weeks after infection. Liver multifocal lesions observed by MRI in infected animals at 10 weeks post infection corresponded to granulomatous inflammation and intergranulomatous fibrosis with METAVIR scores up to A2F2. While most healthy hepatic tissue showed T2 values below 14 ms, these lesions were characterized by a T2 greater than 16 ms. The area fraction of increased T2 correlated (rS = 0.83) with the area fraction of Sirius Red stained collagen in histological sections. A continuous liver T2* decrease was also measured while brown pigments in macrophages were detected at histology. These findings suggest accumulation of hematin in infected livers.
Our multiparametric MRI approach confirms that this murine model replicates hepatic and splenic manifestations of human intestinal schistosomiasis. Quantitative T2 mapping proved sensitive to assess liver fibrogenesis non-invasively and may therefore constitute an objective imaging biomarker for treatment monitoring in diseases involving hepatic fibrosis.
| Schistosomiasis (or bilharzia), a major helminth disease, affects more than 260 million people worldwide. While the adult worms survive for years within veins of the gastrointestinal system, symptoms are due to inflammatory reactions to their eggs in several organs. Hepatic fibrosis may develop in chronic cases of infection with Schistosoma mansoni and lead to portal hypertension with potentially lethal complications. In this study, we aimed at establishing a non-invasive quantitative and readily available magnetic resonance imaging (MRI) technique to monitor in vivo the development of hepatic fibrosis and portal hypertension in Schistosoma mansoni infected mice. We evaluated the transverse relaxation time T2, an easily measurable MRI parameter, as an early and quantitative imaging biomarker for hepatic fibrogenesis and validated it with histological techniques for fibrosis detection and quantification. In addition, we confirmed that this mouse model of schistosomiasis replicates the human pathology closely. The quantitative imaging biomarkers validated in this study will aid in the preclinical and clinical evaluation of new therapeutic strategies against hepatic fibrogenesis.
| Present in many tropical and subtropical countries, schistosomiasis (or bilharzia), the second most prevalent parasitic disease in the world after malaria, affects more than 260 million people and leads to 200 000 deaths per year [1]. This helminthic disease is caused by trematodes of the Schistosoma genus. S. haematobium, S. mansoni, and S. japonicum are the main species infecting humans. These parasites have aquatic gastropods as intermediate hosts and a final vertebrate host. S. mansoni is the principal agent of digestive forms of the human disease. The gastropod is infected with miracidia released from the S. mansoni eggs that transform into sporocysts. These sporocysts shed cercariae in the water that can penetrate the skin of the mammalian hosts. After maturation, male and female worms reproduce in the mesenteric venous plexus and produce eggs that are discharged with the stool into the environment [2,3,4].
However, many eggs are not eliminated and disseminate in the intestines and the liver where they obstruct presinusoidal capillary venules [2,4,5]. The host immune response elicited by these eggs leads to the formation of periovular granulomas and tissue damage. An imbalance between scarring and regeneration causes an accumulation of extracellular matrix rich in collagen (particularly types I and III) leading to hepatic fibrosis [6]. Hepatomegaly occurs early in the disease as a consequence of the granulomatous inflammation [2,3]. In ca 10% of patients, 5–20 years after infection, serious complications occur such as splenomegaly, pulmonary arterial hypertension accompanied by right heart failure, periportal fibrosis resulting in portal hypertension and esophageal varices, that can lead to ascites and gastrointestinal bleeding with high mortality [3,4,5,7,8]. Chronic schistosomiasis is also associated with an increased incidence of hepatocellular carcinoma [4,9]. The standard treatment is the trematodicide Praziquantel, which kills adult worms, but is ineffective on juvenile mammalian-stage schistosomes. The mechanism of action and molecular targets of Praziquantel are unknown. However many studies have reported vacuolation and blebbing of worm teguments and suggested a direct disruptive effect on Ca2+ channels, whereas a recent work has described distinct effects on male and female worms [10,11]. Although effective even at a single dose, reinfection is frequent and in 2013 only 13% of people necessitating treatment could benefit from it [1]. Despite extensive research, an anti-schistosomiasis vaccine is not yet available [12,13].
The mouse model of schistosomiasis is well suited for pharmaceutical and basic research purposes since mice are infected with the parasite species that are pathogenic for humans, and are therefore expected to accurately recapitulate the pathological features of the human disease. The objective of this study was to provide the first characterization of hepatic and splenic features of murine intestinal schistosomiasis using in vivo magnetic resonance imaging (MRI). Another aim was to search for non-invasive biomarkers allowing future evaluation of new therapeutic strategies against the intestinal form of the disease in murine models.
Ultrasound is the leading medical imaging examination for the diagnosis of human intestinal schistosomiasis, and allows the assessment of liver involvement and portal hypertension [14,15]. Liver fibrosis can be monitored using ultrasound transient elastography (FibroScan), a technique allowing the assessment of liver stiffness [16]. Elastography is based on the measurement of tissue elasticity following the propagation of a mechanical shear wave through the liver. However, ultrasound transient elastography has the disadvantage of being highly operator dependent, less reliable for deep organs and not readily available for the exploration of rodent models [17]. Analysis of texture features from computed tomography (CT) images enables staging of fibrosis throughout the liver, but is less accurate in case of heterogeneous fibrosis distribution and is considered inferior to ultrasound transient elastography [18]. Whole abdominal coverage can also be achieved with MRI. In addition to the detection of morphological indicators of liver fibrosis such as splenomegaly and portal hypertension on conventional images [19], various MRI methods can be used for the study of (schistosomiasis-induced) liver disease. Magnetic resonance elastography (MRE) measures viscoelastic properties of the liver tissue, namely its capacity to return with time to its original shape after the application of deforming forces, by quantifying the propagation of the shear waves. This technique can be implemented on standard MRI systems but requires a mechanical vibrator device for the generation of shear waves. Liver stiffness has been shown to correlate with fibrosis stage in patients [20] and animal models [21]. A recent study suggests that MRE more accurately discriminates between early fibrosis stages than ultrasound transient elastography [22]. However, increased liver stiffness is not specific for fibrosis [23], and MRE can be hampered in subjects with severe iron overload [22]. Qualitative and quantitative scores of liver texture features on double contrast-enhanced MRI, an MRI technique based on the injection of two different types of contrast agents (super paramagnetic iron oxides and gadolinium chelates) have been shown to distinguish between mild and severe fibrosis [24]. In rodent models, MRI using a collagen type I targeted gadolinium-based contrast agent has shown differential uptake and washout in fibrotic livers with a good correlation to histological quantification of collagen [25,26]. Diffusion-weighted MRI, a technique sensitive to water diffusivity generally used to study tissue microstructure, has also been performed since fibrosis is expected to restrict water motion [27]. However, due to lack of standardization and confounding factors such as altered perfusion or accompanying inflammation this technique is not sensitive to mild stages of liver fibrosis [28,29]. Another MRI technique, known as intravoxel incoherent motion imaging [30] can distinguish between microscopic motion of water molecules in intra- and extracellular compartments and the microcirculation of blood. Consequently, the derived parameters correlate better with fibrosis stage than conventional diffusion-weighted imaging in patients [31] and animal models [32]. However, this technique has not proved superior to MRE in distinguishing mild to intermediate fibrosis stages in patients [33]. Relaxometry has been proposed for the staging of hepatic fibrosis and evaluated in a number of studies. Relaxometry studies magnetic relaxation, a process by which magnetization of magnetic nuclei returns to its equilibrium state (parallel to the static magnetic field, B0) after it was disturbed from equilibrium and tipped into the plane orthogonal to B0 (transverse plane) by a radiofrequency pulse. Two relaxation time constants can be measured: the longitudinal relaxation time-constant T1 corresponding to the recovery of the magnetization along B0, and the transverse relaxation time-constant T2 describing the decay of the magnetization in the transverse plane. Another relaxation time, called T2*, is a measure of T2 taking into account the effects of static magnetic field inhomogenities on relaxation. Relaxation time constants depend on tissue structure and composition, which may vary with physiological or pathological processes, and strongly influence contrast in MRI. Maps derived from MRI images representing a spatial distribution of quantitative values of a selected relaxation time-constant (T2, T2* or T1) can be generated. Although relaxation time-constant values may vary with magnetic field strength, the tendency to increase or decrease upon a physiological or pathological process is independent of the magnetic field strength. An increase of the longitudinal relaxation time T1 has been observed with increasing severity of hepatic fibrosis. However, alterations of the transverse relaxation time T2 during the process of hepatic fibrogenesis were generally more sensitive [34,35,36]. In particular, in pre-clinical MRI studies using MRI systems equipped with high or ultra high magnetic field strength (3T < B0 < 21T) to increase the spatial resolution or the signal-to-noise ratio, T1 relaxation times tend to differ less between tissues than transverse relaxation times. In addition, T2 mapping is less challenging than T1 mapping in conjunction with respiratory gating (synchronization of MRI sequences with respiration) since all echoes required for T2 analysis are acquired within a few hundred milliseconds, while T1 mapping requires repeated sampling over a couple of seconds. The T2 and T2* relaxation times are also sensitive markers of iron deposition and hemorrhage [37,38].
In this study, multiparametric MRI including quantitative T2 and T2* mapping was performed in an attempt to establish a quantitative measure of the liver damage and related complications caused by schistosomiasis, and in particular to assess early stages of liver fibrosis. A mouse model of the intestinal form of the disease obtained with S. mansoni was explored. In vivo anatomical and relaxometry findings were compared to histological staging of liver disease and fibrosis progression.
Animal studies were in agreement with the French guidelines for animal care from the French Ministry for Agriculture (Animal Rights Division), the directive 2010/63/EU of the European Parliament and of the Council of 22 September 2010, and approved by our institutional committee on Ethics in animal research (Comité d’Ethique de Marseille n°14, project authorization n°: 02157.02).
Twenty-four female CBA/J mice (6-week old) from Charles River Laboratories (l’Arbresle, France) were used. Mice were maintained at 22.5°C with a 12h light/12h dark cycle in an enriched environment with free access to food and water. Twelve mice were infected percutaneously at the age of seven weeks with 30 cercariae of the Venezuelan strain of S. mansoni under intraperitoneal anesthesia (ketamine 100 mg/kg, xylazine 4 mg/kg). Cercariae, which are maintained in our laboratory by passage through Biomphalaria glabrata snails, were counted under binocular microscope, diluted in 500 μl of water, and placed for 60 minutes on the sheared abdomen of mice to replicate the natural route of infection.
Mice were weighed before each MRI session. In vivo MRI was performed on two animal cohorts. The first cohort included a group of 6 infected mice and a group of 6 uninfected mice, which were both imaged at 2 weeks and 6 weeks post infection. MRI at 6 weeks was followed by histology. The second cohort consisting of a group of 6 infected mice and a group of 6 uninfected mice was explored only once at 10 weeks post infection and the animals were sacrificed after MRI for histology.
MRI experiments were performed on a Bruker AVANCE 500 WB MR system (Bruker, Ettlingen, Germany) operating at very high magnetic field (11.75 T), equipped with actively shielded gradients (1 T/m maximum gradient strength and 9 kT/m/s maximum slew rate) and a 30 mm-diameter transmitter/receiver volume birdcage coil. A catheter was inserted into the intraperitoneal cavity of the mice for contrast agent delivery. The animals were positioned in a cradle, and a pneumatic pressure probe was placed under their chest for respiration monitoring. Anesthesia was maintained with isoflurane in air using 1.3–1.8% via a face mask and body temperature was maintained using the water circulation of the gradient cooling system set to 42°C.
All sequences were prospectively gated with respiration using an MRI compatible monitoring and gating system (PC-SAM, Small Animal Instruments Inc., Stony Brook, NY). Images were acquired in the transverse plane with a field of view of 24×24 mm2 and a slice thickness of 0.5 mm. Structural imaging at high in plane resolution (matrix 240×240, in-plane resolution 100 μm) was performed using a 2D spin-echo sequence (repetition time TR ≥ 448 ms; echo time TE = 14 ms) with 20 contiguous slices and repeated on adjacent 20 slices to cover the liver and spleen entirely. Images were acquired before and 15 minutes after intraperitoneal injection of a paramagnetic contrast agent (50 μl of 0.5 M gadoteric acid, DOTAREM, Guerbet, Villepinte, France) (number of accumulations for each acquisition: 2 and 4 respectively). Prior to contrast agent injection, T2 and T2* maps (TR ≥ 9 s; matrix 64×64, 2 accumulations) were acquired in a single slice positioned 0.5 mm caudal of the bifurcation of the portal vein, using a multi-spin echo sequence (12 equally spaced echoes at TE = 7.5 to 120 ms) and a multi-gradient echo sequence (8 equally spaced echoes at TE = 1.6 to 13.5 ms), respectively. Using the bifurcation of the portal vein as landmark this axial slice position was reproducible between animals and covered sufficient liver tissue. Respiratory rate was kept between 60 and 70 breaths per minute by adjusting the isoflurane percentage leading to a total acquisition time of approximately 15 min for the anatomical imaging and 20 min for each map. For quality control, an external reference with a known T2 of ca 21 ms, consisting of a capillary filled with the paramagnetic contrast agent diluted in saline was placed in the image field of view.
Relaxometry studies were performed at 11.75T on water and solutions of collagen at four different concentrations (1.66 g/L, 3.33 g/L, 6.25 g/L and 12.5 g/L). Collagen solutions were prepared with Type I collagen from rat tail (Sigma-Aldrich, St Quentin Fallavier, France) solubilized under magnetic stirring in 0.1 M acetic acid at 40°C. T1 and T2 relaxation times were measured using an inversion recovery gradient echo sequence with 7 inversion times (Tinv 15 ms to 15 s) and a multi-spin echo sequence with 80 TE (10 to 800 ms). TR was 20 s.
Images were analyzed under ImageJ (Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, Maryland, USA, http://imagej.nih.gov/ij/, 1997–2014, last access January 2015) for liver and spleen volumetry, portal vein diameter, and T2-map generation.
Liver and spleen were delineated on each slice (i) to measure the cross-section (Ai) of the organ and estimate its volume in mm3 by
V=0.5∑iAi
. To assess portal hypertension, the portal vein cross-section was measured 0.5 mm caudal of the portal bifurcation, and its diameter was estimated as the Feret’s diameter. The Feret’s diameter is the maximum caliper length namely the longest distance between any two points along the section boundary. Relaxation time maps were computed by fitting the signal intensity S to
S(TE)=S0exp(TE/T2(*))
or S(Tinv) = S0|1 − 2E exp(−Tinv/T1)| using the simplex algorithm in ImageJ. Here, S0 is the signal at thermal equilibrium, and E the efficiency of the inversion pulse. Histograms of quantitative T2 and T2* were obtained from a large region of interest (ROI) covering the liver but excluding the hepatic hilus and large vessels as well as the gall bladder, bowels or stomach when present in the slice.
After euthanasia, livers were fixed in 10% buffered formalin for a minimum of 48 hours, sampled according to standardized procedures [39], paraffin-embedded and routinely processed for histology. Three 3-μm paraffin serial sections per mouse were stained with Hematoxylin-Eosin (HE), with Sirius Red for collagen, and Perls’ stain for iron. Grading of inflammatory activity and staging of fibrosis was performed according to the METAVIR scoring system, a histological scale used to quantify the degree of inflammation and fibrosis of a liver biopsy. “A” refers to the intensity of necrosis and inflammation and may vary from A0 to A3 (A0 = no activity, A1 = mild activity, A2 = moderate activity, and A3 = severe activity). “F” refers to the extent of fibrosis and may vary from F0 to F4 (F0 = no fibrosis, F1 = portal fibrosis without septa, F2 = portal fibrosis with rare septa, F3 = numerous septa without cirrhosis, and F4 = cirrhosis) [40]. Sirius Red stained sections were examined by semiautomatic computer-based morphometry using the NIKON DS-Fi2 camera and NIS Elements imaging software (Nikon, Japan). Morphometric analysis was made on one entire lobe section, measuring semi-automatically delineated Sirius Red stained zones (S1A Fig) and subtracting when present the egg area or unstained central granuloma area deprived of collagen deposition. All analyses were performed in a blinded manner.
Statistical analyses were performed using either GraphPad Prism version 5.00 (San Diego, CA) or JMP 9.0 (SAS, Cary, NC). Values are reported as means ± standard deviation. The non-parametric Mann-Whitney test was used to compare uninfected and infected mice at each time point. The Kruskal-Wallis test was used to compare the METAVIR score or the percentage of Sirius Red stained zones among controls and diseases mice at 6 and 10 weeks after infection. Correlations between T2 values and METAVIR score or T2 values and fibrosis quantification by Sirius Red were performed using the Spearman test. The Smirnov test was used to compare distributions from two independent samples [41]. Values of P<0.05 were considered significant.
Weight control performed before each MRI session showed a continuous increase without significant difference between infected and control groups at any time point. No signs of general health alteration, such as cachexia, reduced mobility or behavioral changes were observed, suggesting animals were in good health condition during the study. All the animals of the first cohort exposed to the cercariae of S. mansoni were successfully infected as confirmed by MRI and histology at 6 weeks after infection, whereas only 4 out of 6 mice of the second cohort were successfully infected and developed hepatosplenic signs as shown by MRI and histology at 10 weeks after infection.
All the images acquired were of good quality, except one T2* map obtained from a control animal of the second cohort, which showed motion artefacts and therefore was not included in the analysis. The two mice of the second cohort in which the infection had failed were also discarded from the analysis.
While no differences were detected between infected and control groups at 2 weeks post infection by anatomical MRI (Fig 1A) and relaxometry, anatomical MRI was able to detect signs related to murine schistosomiasis (Fig 1B–1D) as early as 6 weeks post infection. Indeed, hepatomegaly (Fig 1B), splenomegaly (Fig 1C), and portal hypertension (Fig 1D) as assessed by MRI were significant at 6 weeks: +19% (P = 0.0043), +52% (P = 0.0087) and +60% (P = 0.0087) and progressed to +72% (P = 0.0095), +170% (P = 0.0095) and +139% (P = 0.0139) 10 weeks post infection, respectively. The evolution of the portal vein diameter was similar to that of the cross-section (control: 1.19±0.12 mm, infected: 1.25±0.13 mm, P = 0.393 at 2 weeks, control: 1.17±0.19 mm, infected: 1.43±0.16 mm, P = 0.027 at 6 weeks, and control: 1.16±0.10 mm, infected: 1.68±0.13 mm, P = 0.004 at 10 weeks). However, the most remarkable findings were multifocal hyperintensities (disseminated “white spots”) of the liver in 4 infected mice at 10 weeks post infection (Fig 1A). These hyperintensities were visible before contrast agent injection (Fig 2A), and were contrast-enhanced after injection.
The average T2 value in hepatic ROIs (Fig 2A) at 10 weeks post infection was not significantly increased compared to the average T2 value in control livers (control: 10.4 ± 1.3 ms, infected: 12.2 ± 1.5 ms, P = 0.1714). However, the distributions of hepatic T2 values in infected and control mice were different (Fig 2B). The area fraction with 16 ms < T2 < 26 ms was 13.4 ± 6.9% in infected animals at 10 weeks, while it was 1.1 ± 1.0% in control mice (Kruskal-Wallis test P = 0.0048) (Fig 2C). Liver T2 values > 16 ms on the T2 maps were co-localized with the hyperintensities on anatomical images (Fig 2D). The distributions of T2* values in the livers of infected mice were shifted to lower values at 6 weeks (average T2* = 5.4 ± 1.5 ms in infected mice versus T2* = 5.8 ± 1.6 ms in control mice) and at 10 weeks (average T2* = 4.6 ± 1.5 ms in infected mice versus T2* = 6.1 ± 1.5 ms in control mice) post infection (Fig 2E).
Relaxometry studies performed on collagen solutions did not reveal any significant change of either T1 or T2 relaxation time with increasing collagen concentration (Fig 3).
Histology showed that adult parasites (Fig 4A) and eggs began to lodge in the liver with minimal inflammation and occasionally with isolated periovular granulomas (S1B Fig) by 6 weeks post infection, corresponding to METAVIR scores ≤ A1F0. All infected animals showed discrete periportal and portal inflammation at 6 weeks post infection but only one of them was considered A1F0. The number of eggs trapped in liver tissue dramatically increased thereafter. By 10 weeks post infection, 4 mice presented with severe portal fibrosis with granulomatous chronic inflammation, corresponding to METAVIR scores of A1F1 to A2F2 (Fig 4B–4H). Eggs were surrounded by a dense population of immune cells, with mild to marked extracellular matrix deposition leading to intergranulomatous fibrosis and fusion of several granulomas replacing some portal spaces (Fig 4E–4H).
In a large number of advanced stage granulomas, pigment accumulation in macrophages (S1C Fig), which appeared negative on the Perls’ stain (S1D Fig), was present. This is highly suggestive of hematin, a degradation product of hemoglobin excreted by the worm [42]. Histology confirmed the absence of lesions in the two remaining mice in which the infection had failed; consequently these mice were excluded from analysis.
METAVIR score and area fraction of fibrosis stained with Sirius Red increased simultaneously from 6 weeks post infection on (Fig 5A and 5B) and were both significantly correlated with the area fraction of T2 values comprised between 16 and 26 ms (Fig 5C and 5D) at 10 weeks after infection. Since control values obtained at 6 and 10 weeks for the METAVIR score or the Sirius Red staining were identical (A0F0, or 0% respectively), they were pooled in the statistical analysis (Fig 5A and 5B). The correlations obtained for the values measured at 6 and 10 weeks are shown in S2A and S2B Fig.
To our knowledge, this is the first time a murine model of schistosomiasis is evaluated by quantitative MRI. Anatomical MR images of S. mansoni infected mouse liver had already been published, but a systematic follow-up during disease development was not reported [43]. These images were obtained from a mouse infected with subcutaneous injection of 150 cercariae, while our model relies on the percutaneous administration of only 30 cercariae, replicating the natural route of infection and leading to a similar disease burden obtained with the injection of 150 cercariae. The pathological findings obtained with in vivo MRI in this model are in accordance with known features of the human disease, confirming its relevance to human disease. Volumetric analyses revealed hepato- and splenomegaly as well as portal vein cross section increase as early as 6 weeks post infection, and reliably detected aggravation of these signs during disease progression. These symptoms are unlikely to be a consequence of hepatic fibrosis, which was undetectable at 6 weeks post infection, but rather of vascular damage and obstruction caused by the schistosome eggs [7,44]. Hepatic tissue alteration and iron accumulation were detected 10 weeks post infection by MRI as T2 and T2* changes, respectively, and were confirmed by histology.
From a technical point of view, the multiparametric MRI protocol used in this study was sensitive enough to monitor disease severity and identify infection failure in two out of 6 mice. In particular, the correlation with the area fraction of collagen stained with Sirius Red in liver histology shows that the transverse relaxation time T2 can be used as a quantitative non-invasive biomarker of hepatic fibrosis.
Few techniques are sensitive to potentially reversible early stages of liver fibrosis. All existing imaging approaches rely on the measurement of parameters that are only indirectly related to liver fibrosis (tissue stiffness, restricted water diffusion). Likewise, relaxation time constants depend on physicochemical properties that are altered during fibrogenesis. In ex vivo studies, correlation between transverse relaxation times and fibrosis has been controversial. For example, at 11.75T a T2 decrease has been observed in a 3,5-diethoxycarbonyl-1,4-dihydrocollidine induced mouse model of hepatic fibrosis [45]. But in another study at 0.94T a positive correlation between liver T2 and degree of fibrosis was observed in a thioacetamide induced mouse model of liver cirrhosis [34]. It is known that the T2 is a rather unspecific parameter, influenced by inflammation [34,46], edema, perfusion changes or steatosis [47]. Indeed, free water is characterized by a long T2 (hundreds of ms) whereas lipids have much shorter T2 values (tens of ms). In the vast majority of cases, severe hepatic fibrosis in human schistosomiasis is not associated with ultrasound detectable steatosis namely the accumulation of triglycerides in hepatocytes, except in patients with obesity or diabetes. Most subjects with severe disease are rather underweight. Moreover, liver metabolic profiling by 1H nuclear magnetic resonance spectroscopy revealed reduced levels of lipids in S. mansoni infected mice [48]. In agreement with previous results [49], our in vitro experiments show that collagen at concentrations up to 12.5 g/L does not significantly modify the relaxation times. The range of collagen concentrations tested in our phantom studies corresponds to those measured by biochemical techniques in whole-liver extracts of S. mansoni-infected mice at 8 weeks after infection with 50 cercariae (S. mansoni infected mice: 10.3 ±1.60 mg/g of wet liver, control mice: 0.591±1.144 mg/g of wet liver) [50]. However, deposition of collagen (Fig 4D) in the extracellular matrix in living tissue is accompanied by pathophysiological reactions that have a measurable effect on relaxation times [19]. We cannot discard the possibility that other components of the extra-cellular matrix, which accumulate during fibrogenesis, (fibronectin, elastin, proteoglycan, hyaluronan etc…) may contribute to T2 changes. Along the same line, it is not excluded that the observed T2 increase in the liver of S. mansoni infected mice is due to the decreased content of hepatic lipids [48], but further investigations are required.
In agreement with other in vivo MRI studies in animal models of hepatic fibrosis [35,36] and in humans [51], at 10 weeks post infection we observed a T2 increase superior to 14% compared to normal liver tissue. Although the T2 increase might be related to inflammation and edema, the hepatic T2 values were not yet increased 6 weeks after infection (10.4 ± 2.4 versus 10.7 ± 2.5 ms in infected versus control mice) although histology revealed a discrete (A0 –A1) portal inflammatory reaction at this stage of the disease, without any sign of fibrosis.
In contrast to previous works that assessed the alteration of the average T2 value in the presence of liver fibrosis [34,35,36,45], we quantified the area fraction of altered T2 on T2 maps of liver tissue. This measure was not performed in previous studies, despite a non-uniform distribution of fibrosis in most cases and attempts to correlate quantitative MRI parameters to the area fraction of fibrosis observed in histology. Our results demonstrate that despite the lack of specificity of the T2 relaxation time, the area fraction of increased T2 in the liver is significantly correlated with the area fraction affected by fibrosis. The area fraction of increased T2 enabled us to differentiate between early and intermediate fibrosis stages (i.e. stages ≤ F2).
This exploratory study on a small number of animals aimed at evaluating a non-invasive MRI method for the assessment of liver fibrosis progression. A limitation of this study is that different animal groups were used at the last two imaging time points, due to the fact that a validation by histology was mandatory. A longitudinal study on the same group of animals up to a time point beyond 10 weeks post infection would be desirable.
For disease monitoring and treatment evaluation in small animals T2 mapping is easy to implement and less challenging than transient elastography or MRE. Indeed, this technique does not require the use of additional equipment such as vibrators and sequences for relaxometry are readily available on standard pre-clinical MRI systems.
Magnetic resonance relaxometry has the advantage of being quantitative without any contrast agent injection that could interfere with biological parameters and alter disease development. However, administration of a commercially available nonspecific extracellular contrast agent such as Gd-DOTA enhanced the lesions, demonstrating increased vessel permeability.
The quantitative T2 and T2* maps performed in this study allow further optimization of image weighting parameters for future studies on schistosomiasis.
T2 mapping is a quantitative and non-invasive marker of non-steatotic liver fibrosis with sensitivity close to that of histology even at early stages.
This multiparametric and quantitative MRI approach can monitor hepatic and splenic disease progression and assess liver fibrosis non-invasively. In preclinical studies and in settings were MRI facilities are readily available, this objective quantitative imaging biomarker may be used to monitor response to therapy in diseases involving hepatic fibrosis.
|
10.1371/journal.pgen.1003656 | Yeast Pol4 Promotes Tel1-Regulated Chromosomal Translocations | DNA double-strand breaks (DSBs) are one of the most dangerous DNA lesions, since their erroneous repair by nonhomologous end-joining (NHEJ) can generate harmful chromosomal rearrangements. PolX DNA polymerases are well suited to extend DSB ends that cannot be directly ligated due to their particular ability to bind to and insert nucleotides at the imperfect template-primer structures formed during NHEJ. Herein, we have devised genetic assays in yeast to induce simultaneous DSBs in different chromosomes in vivo. The repair of these breaks in trans could result in reciprocal chromosomal translocations that were dependent on classical Ku-dependent NHEJ. End-joining events leading to translocations were mainly based on the formation of short base pairing between 3′-overhanging DNA ends coupled to gap-filling DNA synthesis. A major proportion of these events were specifically dependent on yeast DNA polymerase Pol4 activity. In addition, we have discovered that Pol4-Thr540 amino acid residue can be phosphorylated by Tel1/ATM kinase, which could modulate Pol4 activity during NHEJ. Our data suggest that the role of Tel1 in preventing break-induced chromosomal translocations can, to some extent, be due to its stimulating effect on gap-filling activity of Pol4 to repair DSBs in cis. Overall, this work provides further insight to the molecular mechanisms of DSB repair by NHEJ and presents a new perspective to the understanding of how chromosomal translocations are formed in eukaryotic cells.
| Chromosomal translocations are one of the most common types of genomic rearrangements, which may have a relevant impact on cell development. They are often generated from DNA double-strand breaks that are inaccurately repaired by DNA repair machinery. In this study, we have developed genetic assays in yeast to analyze the molecular mechanisms by which these translocations can arise. We found evidence showing that the classical nonhomologous end-joining repair pathway can be a source of chromosomal translocations, with a relevant role for yeast DNA polymerase Pol4 in such processes. The involvement of Pol4 is based on its efficient gap-filling DNA synthesis activity during the joining of overhanging DNA ends with short sequence complementarity. In addition, we discovered that DNA polymerase Pol4 can be modified during the repair of the breaks via phosphorylation by Tel1 kinase. This phosphorylation seems to have important structural and functional implications in the action of Pol4, which can finally influence the formation of translocations. This work provides a useful tool for deciphering factors and mechanisms involved in DNA double-strand break repair and identifying the molecular pathways leading to chromosomal translocations in eukaryotic cells.
| DNA double-strand breaks (DSBs) are one of the most cytotoxic lesions. They can originate during cellular metabolism or upon exposure to DNA damaging agents such as radiation or chemicals. DSBs can be repaired by two main mechanisms, homologous recombination (HR) or nonhomologous end-joining (NHEJ) [1]. In the absence of DNA homology, NHEJ is the main source of chromosomal translocations in both yeast [2] and mammalian cells [3], [4]. In the latter, those translocations generated as byproducts of V(D)J and class switch recombination in B cells are particularly relevant, since they can promote cancer, especially leukemia and lymphoma [5], [6]. Despite the ability of NHEJ to join breaks directly, most DSBs occurring in vivo are not fully complementary or have chemical modifications at their ends, and cannot be directly ligated. In these cases, additional processing, such as DNA end trimming or gap-filling DNA synthesis, may be required in order to optimize base pairing before ligaton [7]. The extent of DSB end processing influences the speed of repair and defines the existence of two forms of NHEJ. Classical NHEJ (c-NHEJ) is the fastest and most conservative form, as it relies on a limited degradation of DNA ends. On the other hand, the alternative NHEJ pathway (alt-NHEJ) relies on an extensive end resection that exposes hidden sequence microhomologies surrounding DNA ends to be rejoined. Core components of c-NHEJ are the Ku70/80 and XRCC4/DNA Ligase IV complexes (YKu70/80 and Lif1/Dnl4 in yeast, respectively) [7], [8]. In vertebrates, Ku is part of a larger complex called DNA-dependent protein kinase (DNA-PK), whose catalytic subunit is DNA-PKcs kinase. The Ku complex initially mediates the synapsis between the two broken DNA ends, protecting them from extensive degradation. Thereafter, it also recruits other components, such as the XRCC4/DNA Ligase IV complex. In the absence of Ku, or due to its departure from DSB ends, the occurrence of alt-NHEJ increases relative to the extent of DSB resection, as it allows uncovering larger microhomologies to be used for end-joining [9].
NHEJ also involves accessory factors such as DNA polymerases belonging to the PolX family [10]. Among mammalian PolX polymerases, Polλ and Polμ are specialized DNA polymerases with a large capacity to use imperfect template-primer DNA substrates. Thus, they are able to extend DNA ends that cannot be directly ligated by NHEJ, as demonstrated in vitro with human whole-cell extracts [11]. This is mainly due to their capability of simultaneously binding both the 5′ and 3′ ends of small DNA gaps, which permits an efficient gap-filling [12], [13]. Based on such DNA binding properties, these polymerases can efficiently search for sequence microhomologies and utilize DNA substrates with unpaired bases at or near the 3′-terminus [14]–[16]. These scenarios are frequent in NHEJ when DNA ends have extremely low sequence complementarity. PolX polymerases are specifically recruited to DSBs during NHEJ by interacting with Ku and XRCC4/DNA Ligase IV through their BRCT domains [17], [18]. This interaction allows gap-filling during end-joining reactions, as demonstrated both in vitro [18]–[20] and in vivo [21]–[24]. Whereas mammalian cells have four PolX polymerases (Polλ, Polμ Polβ, and TdT), in yeast there is a unique member, Pol4. Yeast Pol4 combines most of the structural and biochemical features of its mammalian counterparts Polλ and Polμ [25], [26], including the BRCT-mediated interaction with core NHEJ factors [27]. It has been shown that Pol4 is required to recircularize linear plasmids having terminal microhomology, as an example of NHEJ reactions performed in vivo [28]–[31]. In addition, Pol4 is involved in NHEJ-mediated repair of chromosomal DSBs induced in cis [32]–[34], and in NHEJ reactions where no base complementarity between DSB ends is available [29].
Here we have devised intron-based assays in yeast to generate two simultaneous DSBs in different chromosomes in vivo, whose repair by NHEJ could generate reciprocal chromosomal translocations. End joining events leading to translocations were mainly based on the formation of short base pairing between 3′-overhanging ends coupled to gap-filling. A major proportion of these events were specifically dependent on yeast DNA polymerase Pol4, as the DNA synthesis-mediated repair signature disappeared in pol4Δ cells. Other results, suggesting that Tel1-mediated suppression of translocations can be in part due to Pol4 regulation to promote DNA synthesis-dependent NHEJ, will be also discussed.
We have modified a previously reported yeast genetic assay [35] to analyze the repair mechanism through which two induced DSBs can be joined by NHEJ to form chromosomal translocations. The system is mainly based on two nonhomologous halves of the LEU2 gene (leu2Δ5′ and leu2Δ3′), each one fused to either an HO or I-SceI endonuclease cleavage site and integrated into a different chromosome (Figure 1A). In the experimental conditions used, DSBs were induced by continuous expression of both endonucleases in cells accumulated in the G1 phase of the cell cycle, when NHEJ is the predominant DSB repair pathway. NHEJ-mediated repair of DSBs can generate reciprocal translocations that restore a functional LEU2 gene and can be selected as Leu+ colonies in selective plates. Within the LEU2 gene, translocation breakpoints are embedded in a functional intronic sequence that can tolerate the variability produced during NHEJ (Figure 1A). Breakpoints can be further analyzed by PCR amplification and DNA sequencing, and the repair events can then be deduced. After DSB induction, Leu+ translocants were obtained at a frequency of 0.27×10−3 in a wild-type strain (Figure 2 and Table S1). The electrophoretic karyotyping of wild-type Leu+ translocants, as determined by pulsed-field gel electrophoresis (PFGE), verified the expected molecular nature of translocations. Thus, ethidium bromide staining of gels and Southern analysis with both LEU2 and HYG specific probes showed two new 596- and 811-kb long chromosomes resulting from reciprocal translocations (Figure 1 and Figure S1). LEU2 signal was specifically detected in the smaller translocated chromosome, which carried the joined LEU2 halves (Figure 1C). Simultaneously, an HYG signal was specifically detected in the larger translocated chromosome (Figure S1). No Leu+ translocants were recovered in the absence of Yku70 (Figure 2), demonstrating that translocations were mediated by c-NHEJ. These results validated our assay to analyze the genetic requirements and mechanisms leading to chromosomal translocations via c-NHEJ.
After the induction of endonucleases cleavage, 4-nt long 3′-protruding DSB ends with partial complementarity were generated (Figure 1A). To unravel the molecular events leading to NHEJ-mediated translocations, we analyzed the breakpoints of 24 independent wild-type Leu+ translocants by sequencing ACT1 intron within the reconstituted LEU2 gene (all sequencing data are available in Figure S2). This analysis showed a major proportion of repair events based on the formation of either 1-nt or 2-nt base pairing between the 3′-protruding DSB ends, which generated 2-nt gaps on both strands (Type I, 67% of the events; Figure 3 and Table 1). These small gaps should necessarily be filled-in through a templated insertion (+CA/+AT), as occurs in NHEJ-mediated repair of DSBs induced in cis [36]. The second more represented repair event in wild-type cells (Type II, 21%) involved the use of short (4-nt) microhomologies between one 3′-protruding DNA end and adjacent sequences in the other DSB end for base pairing (Figure 3). A third type of repair in wild-type cells (Type III, 8%) implied the formation of a 3-nt base pairing between the two 3′-protruding DSB ends and the exonucleolytic removal of the terminal nucleotides (Figure 3). These DSBs could then be directly ligated without the need of gap-filling. Type III events would involve the formation of a T:G mismatch, which should be processed later by mismatch repair machinery (Figure 3). Finally, a less frequent repair type (Type IV, 4%) implied the degradation of one 3′-protruding end to generate a blunt end. This could be utilized as a primer in a DNA synthesis reaction that used the other intact 3′-protruding end as a template in an end-bridging-like reaction (Figure 3) [37]. These results indicated a major role for gap-filling-mediated repair of induced DSBs leading to translocations in our experimental system.
DNA polymerase Pol4, the only member of PolX family in yeast, synthesizes DNA efficiently from 3′-protruding ends that are annealed to form small gaps during classical NHEJ. As shown in Figure 2, we observed a significant decrease in the frequency of translocations in our assays when Pol4 was absent (0.27 vs. 0.01, 27-fold decrease, p<0.001). This suggested a relevant role for Pol4 in NHEJ-mediated repair leading to translocations. In agreement, pol4Δ cells completely lost gap-filling-mediated repair events (Type I; Table 1). Intriguingly, these cells did not lose type IV events, which also implied DNA synthesis for repair. Ectopic overexpression of POL4 gene restored wild-type translocation frequency (Figure 2 and Table S1). Importantly, cells overexpressing wild-type Pol4 repaired induced DSBs mainly by gap-filling-mediated repair, as wild-type cells did (Table 1). This result validated the use of this overexpression system for the analysis of Pol4 mutants in vivo. Translocation frequency was partially dependent of Pol4 DNA polymerase activity, as it was reduced (0.40 vs. 0.18, 2-fold decrease, p<0.001) when we overexpressed a catalytically inactive Pol4 mutated at two of the three aspartic residues required for polymerization (pol4-D367A,D369A mutant; Figure 2). This reduction was even higher (4-fold, p<0.001) under more physiological conditions in pol4Δ cells expressing a catalytically inactive Pol4 from the POL4 endogenous promoter (Figure S3). Notably, pol4Δ [pol4-D367A,D369A] cells did not show gap-filling-mediated repair events (Type I), thus confirming the role of the Pol4 polymerization activity during translocations formation (Table 1). It has been shown that a functional BRCT domain is strictly required for the recruitment of Pol4 to DSBs in vivo to catalyze gap-filling during NHEJ [27], [28], [32]. Accordingly, the overexpression of a Pol4ΔBRCT mutant protein in pol4Δ cells strongly inactivated Pol4 function during NHEJ-mediated repair of induced DSBs in our assays. These cells showed a similar translocation frequency level to pol4Δ cells and no gap-filling-mediated repair events (Type I; Figure 2 and Table 1). It is worth noting that the overexpression of POL4 alleles in pol4Δ cells induced a strong increase of direct ligation repair events, which did not imply gap-filling (Type III, see Table 1). Altogether, these results suggested that Pol4 played a major role in the joining of DSBs with partial complementarity by filling the small DNA gaps present on both strands during NHEJ.
Yeast Tel1 (homolog of mammalian ATM) is a serine/threonine protein kinase that is recruited and activated by DSBs. It has been reported that the absence of Tel1/ATM increases break-induced chromosomal translocations, likely due to a defect in DSB end tethering and resection [38], [39]. This finding was confirmed in our experimental system, as the frequency of translocations in tel1Δ cells significantly increased over wild-type level (2.99 vs. 0.27, 11-fold increase, p<0.001; Figure 2). Interestingly, the analysis of repair types in tel1Δ translocants showed a different repair pattern compared to wild-type, which included a significant decrease in gap-filling-mediated repair reactions (Type I) (from 67% to 33%, p<0.005; Table 1). Concomitantly, end-bridging reactions and those reactions that did not involve gap-filling increased in tel1Δ cells (Table 1). Thus, we asked whether Pol4 could be a target of Tel1/ATM during NHEJ-mediated DSB repair. We searched for potential Tel1 phosphorylation sites in the amino acid sequence of Pol4, and we found two threonine residues (Thr64 and Thr540) within [S/T]Q consensus sites, which have been defined for all PIIK-kinases, including Tel1 (Figure 4A). The carboxy-terminal T540Q consensus motif is highly conserved in different Saccharomyces species, probably reflecting its functional relevance (Figure 4A). To know whether Tel1 phosphorylates any of these threonine residues we partially purified His-tagged wild-type and mutant Pol4 proteins where the Thr64 and Thr540 amino acids were mutated to non-phosphorylatable alanines (Figure S4A). We analyzed their phosphorylation in vitro using HA-Tel1-enriched immunoprecipitates obtained as previously described [40] (Figure 4B and Figure S4B). Control immunoprecipitates from cells that were not transformed with the HA-Tel1-encoding plasmid were also used to detect the possible activities of other kinases (Figure 4B). We observed that in vitro phosphorylation of Pol4 was clearly higher when using Tel1-enriched immunoprecipitates than with those obtained from non-transformed cells (Figure 4B). As deduced from quantification of phosphorylation signals, wild-type Pol4 and mutant Pol4-T64A proteins were similarly phosphorylated by Tel1 (Figure 4C). However, a significant decrease of Pol4 phosphorylation was observed in the Pol4-T540A mutant, which was even higher in the Pol4-T64A,T540A double mutant (Figure 4C). These results indicated that Pol4-Thr540 residue is the most efficiently phosphorylated by Tel1 in vitro.
Next, we sought to determine if Pol4 phosphorylation also occurred in response to DSBs in vivo. For this purpose, Flag-tagged wild-type and T540A Pol4 proteins were overexpressed in pol4Δ cells in which we simultaneously induced DSBs with zeocin (Figure 4D). To promote NHEJ processing, DSBs were induced in G1-arrested cells. Flag-tagged Pol4 proteins were immunoprecipitated with anti-Flag antibodies and subsequently immunodetected using both anti-Flag antibodies and antibodies that specifically recognize phosphorylated SQ/TQ motifs. As shown in Figure 4D, a damage-induced SQ/TQ phosphorylation signal was specifically observed in pol4Δ [POL4] cells, which was detected as a slower migrating protein with respect to Pol4 molecular mass. Importantly, such a phosphorylation was barely detected when the Pol4-T540A phosphomutant was overexpressed in the same experimental conditions (Figure 4D). To further verify that the observed phosphorylation signal was dependent on Tel1, wild-type Pol4 was overexpressed in a tel1Δ pol4Δ double mutant. As expected, damage-induced SQ/TQ phosphorylation was again much weaker than that obtained in pol4Δ [POL4] cells, confirming its dependence on Tel1 (Figure 4D). As deduced from the quantification of phosphorylation signals, the decrease of damage-induced Pol4 phosphorylation either in the pol4Δ [pol4-T540A] mutant or in the absence of Tel1 kinase was statistically significant (Figure 4E). Altogether, our data suggested that Pol4 can be phosphorylated on Thr540 residue by Tel1 in response to DSBs.
To determine the relevance of Tel1-mediated Pol4 phosphorylation in vivo, we analyzed the effect of overexpressing the different non-phosphorylatable Pol4 proteins in our system. Both translocation frequency and repair events observed in pol4Δ [pol4-T64A] mutants were similar to those observed in pol4Δ [POL4] cells (Figure 2 and Table 1). Interestingly, pol4Δ [pol4-T540A] mutants showed a significant reduction in the frequency of translocations compared to control cells (0.40 vs. 0.13, 3-fold decrease, p<0.001; Figure 2 and Table S1). This reduction was even stronger (7-fold, p<0.001) under a more physiological situation by expressing the Pol4-T540A phosphomutant from the POL4 endogenous promoter (Figure S3). Overexpression of a double phosphomutant (pol4-T64A,T540A) generated a translocation frequency similar to that obtained in pol4Δ [pol4-T540A] single mutant, confirming that Pol4-Thr64 residue is not involved in the regulation of Pol4 activity (Figure 2 and Table S1). The molecular analysis of repair events in pol4Δ [pol4-T540A] mutants showed that the repair of the induced DSBs mainly occurred as in tel1Δ cells (Table 1). Notably, this included a 2-fold decrease in gap-filling-mediated repair (Type I) events compared to control conditions (from 68% to 36%, p<0.005; Table 1). Simultaneously, an increase in microhomology-mediated repair (Type II) and end-bridging repair (Type IV) was observed (Table 1). To further investigate the genetic interaction between Tel1 and Pol4 phosphorylation in our assays, we analyzed tel1Δ pol4Δ double mutants. According to the involvement of Pol4 in the formation of translocations, we observed a significant decrease of translocation frequency in tel1Δ pol4Δ cells compared with tel1Δ single mutants (0.71 vs. 2.99, 4-fold decrease, p<0.001, Figure 2 and Table S1). This decrease was lower than that observed in pol4Δ cells compared to wild type (0.01 vs. 0.27, 27-fold decrease, p<0.001), consistent with the presence of a basal level of gap-filling-mediated repair in the tel1Δ pol4Δ double mutants (Table 1). The overexpression of wild-type Pol4 complemented the absence of Pol4 in tel1Δ pol4Δ cells, as deduced by comparing tel1Δ pol4Δ [POL4] cells and tel1Δ cells carrying an empty vector (0.88 vs. 1.29; Table S1). The analysis of repair types in tel1Δ pol4Δ [POL4] cells showed a significant decrease of type I events (from 68% to 43%; p<0.005; Table 1) and a concomitant increase in type IV events (from 0% to 10%; p<0.005; Table 1) when compared to pol4Δ [POL4], similar to what occurred in tel1Δ cells. Finally, both the translocation frequencies and the types of repair in tel1Δ pol4Δ [POL4] and tel1Δ pol4Δ [pol4-T540A] were similar (Figure 2 and Table 1), which demonstrated the epistatic relationship between tel1Δ and pol4-T540A mutations. Together, these results indicated that the phosphorylation of Pol4-Thr540 by Tel1 stimulated Pol4-mediated gap-filling synthesis during NHEJ repair of DSBs with partial complementarity.
Next we sought to examine the role of Pol4 in the formation of translocations in the absence of nucleotide complementarity between DNA ends to be repaired. For this purpose, we devised another system in which we introduced the I-SceI endonuclease cleavage site in inverse orientation with respect to the previous assay (Figure 5). Thus, concomitant DSBs produced by HO and I-SceI endonucleases generated 3′-protruding DNA ends that were totally non-complementary (Figure 5). In agreement with the greater difficulty of repairing such DSBs, cells carrying this new system showed lower survival frequencies compared to the previous assay (two orders of magnitude, Tables S1 and S2). Despite this, we found the same dependence on Yku70 to repair the induced DSBs (Figure 6A and Table S2). The main repair type in wild-type cells carrying this system was not mediated through partial annealing of 3′-overhanging ends and gap-filling on both strands (Type I; Table 2 and Figure 6B), as expected by the non-complementary nature of DSB ends. Instead of this, DSB repair in this new assay was favored by the use of short microhomologies around end sequences and gap-filling reactions on only one strand (Type II; Table 2 and Figure 6B). Nevertheless, the absence of Pol4 in this new assay resulted in a stronger decrease in translocation frequency with respect to wild-type (1.49 vs. 0.005, 300-fold decrease; Figure 6A and Table S2), as compared to the previous assay. In agreement, repair events involving gap-filling on both strands (Type I) completely disappeared in pol4Δ. Concomitantly, a new class of events (Type III), which were mediated by microhomology searching and did not require gap-filling, appeared in these cells (Table 2). Pol4 overexpression in pol4Δ cells restored translocation frequency levels (Figure 6A and Table S2) and increased type I repair events over levels found in wild-type cells (Table 2). The overexpression of Pol4 phosphomutant proteins in this new system generated the same effects observed in the previous assay. Thus, whereas pol4Δ [pol4-T64A] mutant behaved like pol4Δ [POL4] cells, both translocation frequency and repair events using 2-strand gap-filling were significantly decreased in pol4Δ [pol4-T540A] mutant cells (from 28% to 16%, p<0.005; Table 2 and Figure 6). Overall, these results indicated that the phosphorylation of Pol4-Thr540 by Tel1 stimulated Pol4-mediated gap-filling synthesis also during NHEJ repair of non-complementary DSBs.
Finally, we asked whether phosphorylation of Pol4-Thr540 also affected DNA synthesis-mediated NHEJ of DSBs formed simultaneously in the same chromosome (in cis). To address this question, we used a previously described yeast assay [34], in which two I-SceI sites are integrated with opposing orientation on each side of the URA3 gene in chromosome V (Figure S5). Upon continuous expression of the I-SceI endonuclease, almost all survivors repaired the induced DSBs by joining the two distal non-complementary DSB ends and lost the intervening URA3 gene. This repair occurs via Pol4-mediated NHEJ [34]. Thus, we analyzed the effect of the pol4-T540 mutant allele in the repair of these two DSBs generated in cis (Figure S5). As expected, DSB repair frequency decreased significantly in pol4Δ cells compared to wild-type (13-fold decrease, p<0.001, Figure S5). Whereas the expression of wild-type Pol4 in pol4Δ cells efficiently restored wild-type repair frequency, the expression of a catalytically inactive Pol4 did not (Figure S5). Of our particular interest, DSB repair frequency in pol4-T540A mutants decreased significantly with respect to pol4Δ cells expressing wild-type Pol4 (8-fold decrease, Figure S5). These results indicate that the phosphorylation of Pol4-Thr540 influenced gap-filling DNA synthesis during NHEJ repair independently of DSBs location.
In this work, we have devised yeast assays to understand the mechanisms by which DSBs generated in vivo in different chromosomes can be joined by NHEJ to form chromosomal translocations. These assays allow the formation of two site-specific DSBs with 3′-overhangs having either partially- or non-complementary end sequences. Breakpoint sequence analysis of translocations showed that end-joining events were mainly based on short base pairing between overhanging ends coupled to efficient Pol4-dependent gap-filling. In addition, we discovered a relevant role for Tel1 kinase in the modulation of Pol4 activity during NHEJ through the phosphorylation of Thr540 amino acid residue. Indeed, the phosphorylation state of this residue might have relevant structural and functional implications in the action of Pol4, promoting gap-filling DNA synthesis during NHEJ repair.
Eukaryotic cells have two different types of NHEJ, which essentially differ in their dependence on Ku proteins [7]. Our assays rely on the classical Ku-dependent NHEJ (c-NHEJ) pathway, which mainly operates on both blunt and fully complementary DSBs that can be directly ligated. Moreover, it is also able to utilize DSBs with 3′-overhanging single-stranded ends that can partially anneal. However, in these cases an additional processing of DNA ends is needed. Most of end-joining events that we recovered in our assays relied on base pairing between overhanging sequences coupled to an efficient DNA end processing. This processing frequently implied gap-filling DNA synthesis prior to ligation, and occasionally DNA end trimming. In cells carrying our systems, we also observed some NHEJ events that used short microhomologies present in sequences adjacent to DSB ends for base pairing before ligation. Nevertheless, in all these events, the extent of microhomology used for base pairing did not exceed 5-nt. Therefore, they cannot be considered as alternative (Ku-independent) NHEJ-mediated events [9]. Our assays do not permit very long DNA end resections, since an extensive degradation of intronic sequences used would impede the recovery of selectable funcional LEU2 genes. This is in agreement with the high dependence of translocations on the presence of Yku70 that we observe.
Among the repair types analyzed in our assays, those end-joining reactions that required the filling of short gaps formed on both DNA strands showed a complete dependence on Pol4. This demonstrates the relevance of Pol4-mediated DNA synthesis in NHEJ, in agreement with previous data [28]–[32], which is a result of the special ability of Pol4 to stabilize base pairing via protein-DNA interactions when continuity of both strands is disrupted [31]. We still found NHEJ events involving gap-filling DNA synthesis on only one strand in pol4Δ translocants with our second system. This is probably due to the fact that base stacking interactions across broken strands can occasionally stabilize template continuity, allowing other polymerases to substitute for Pol4, as previously reported [31]–[33], [41]. The involvement of other polymerases in NHEJ when Pol4 is not present is also demonstrated by the existence of residual gap-filling repair events in tel1Δ pol4Δ double mutants in our assays. In fact, although we do not know how the lack of Tel1 could affect the action of these other polymerases during NHEJ, it is tempting to speculate that it could facilitate their activity. This would explain why the decrease of NHEJ repair generated by the absence of Pol4 is much higher in wild-type cells than in tel1Δ mutants. It is worth noting that Pol4 overexpression in our assays also increased the occurrence of NHEJ reactions by direct ligation. This is especially noticeable when overexpressing a dominant negative Pol4 (pol4Δ [pol4-D367A,D369A] mutant) and suggests that Pol4 might also act as a scaffold in some circumstances, in agreement with previous results [32]. In these cases, it could protect DNA ends from extensive resection and favor direct ligation, as has been also suggested for other polymerases [41]. Similarly, the presence of Polμ (a Pol4 orthologue) limits the resection of DNA ends at Ig genes in vivo during VDJ recombination in murine B cells [42].
One of the initial events in c-NHEJ is the binding of Ku proteins to DSBs. Once Ku binds to DNA ends, they are protected from degradation and other NHEJ components can now be recruited with a high flexibility [43]. This recruitment could be directed by the complexity of DNA ends, that is, depending on their base complementarity extent. In this scenario, phosphorylation of downstream proteins emerges as a relevant mechanism to coordinate the repair process [44]. Tel1/ATM is the main kinase initially recruited to DSBs, where it phosphorylates a number of downstream effector proteins. Through the phosphorylation of some of these proteins, Tel1/ATM promotes the accurate DNA end utilization during c-NHEJ [39] and avoid formation of dangerous chromosomal rearrangements [38], [45], [46]. Our results confirm Tel1 involvement in preventing translocations and identify Pol4 as a novel target of Tel1 after DSBs generation. Interestingly, mammalian Polλ (a Pol4 orthologue) is phosphorylated by ATM in response to DNA damage [47], although the physiological significance of this phosphorylation remains to be elucidated. As shown here, Pol4 phosphorylation specifically occurs at C-terminal Thr540 residue. This modification may have relevant structural implications, as expected from its location in the thumb subdomain. Since Pol4 amino acid sequence is relatively well conserved (i.e. up to 25% amino acid identity with Polλ catalytic core), it is possible to model yeast Pol4 using the crystal structure of human Polλ forming a ternary complex with a 1-nt gapped DNA substrate and the incoming nucleotide (Figure 7) [48]. According to this model, Pol4-Thr540 residue would be part of a short hairpin comprising residues 540 to 543 (TQHG) that is located quite near the DNA template (Figure 7). Interestingly, an equivalent motif in human Polμ has been implicated in the correct positioning of its Loop1 structural motif and the template strand, two critical features for an efficient DNA synthesis-mediated NHEJ reaction in vitro (unpublished data).
From our structural model, it can be predicted that phosphorylation of Pol4-Thr540 by Tel1 could affect the interaction with the DNA template (Figure 7). As a consequence, this would modify the ability of Pol4 to use 3′-ended NHEJ substrates stabilized by extremely short terminal base pairing. Our data suggest that the phosphorylation of Pol4 by Tel1 may optimize Pol4 to handle DNA ends as a function of the base complementarity extent. This would enhance Pol4-mediated gap-filling activity during NHEJ repair. Supporting this hypothesis, we found that preventing Pol4 phosphorylation at Thr540 residue (pol4Δ [pol4-T540A] mutant) produced a significant decrease in the occurrence of translocations in our systems, mainly due to a reduced gap-filling-mediated repair of both partially- and non-complementary DSBs. Remarkably, end-bridging reactions, which involve DNA synthesis from an unpaired template to join the DSB ends, increased in the pol4Δ [pol4-T540A] mutant. This type of repair events, rare in wild-type cells, also became more visible in tel1Δ cells, in which Pol4-Thr540 residue cannot be phosphorylated. Thus, it is tempting to speculate that the increase of translocations observed in the absence of Tel1 could be, in part, a consequence of the absence of phosphorylation at Pol4-Thr540, which would impede an efficient gap-filling-mediated repair and favor end-bridging reactions. The combination of tel1Δ and pol4Δ mutations allowed us to get a more detailed analysis of the genetic interaction between Tel1 and Pol4. First, we observed that the overexpression of wild-type Pol4 or Pol4-T540A mutant in tel1Δ pol4Δ double mutant cells resulted in similar translocation frequency levels. This ruled out a possible negative effect of T540A mutation on the catalytic activity of Pol4, since Pol4-T540A mutant complemented tel1Δ pol4Δ as wild-type Pol4 did. In addition, the analysis of pol4Δ [pol4-T540A] mutants confirmed the epistasis between tel1Δ and pol4-T540A mutations, as repair types observed in double mutants were similar to those in single mutants. Again, this included a significant decrease in gap-filling-mediated repair and a concomitant increase in end-bridging repair. Finally, we also present evidence that the pol4-T540A mutation equally affects repair of DSBs generated both in cis and in trans. Together, our results show that phosphorylation of Pol4 by Tel1 promotes gap-filling-dependent NHEJ repair independently of the location of the DSBs. Thus, in spite of the decrease in translocations observed in the absence of Pol4 phosphorylation, we believe that such modification is, at the same time, essential to prevent NHEJ repair of DSBs in trans, since it also stimulates efficient gap-filling-mediated NHEJ repair of DSBs in cis. Indeed, in the absence of Tel1, defective DSB end tethering and resection, together with a less efficient Pol4-mediated NHEJ repair in cis, would lead to an increased DSB persistence and, ultimately, to an increased occurrence of chromosomal translocations.
In summary, this work uncovers a new insight during DSB repair by NHEJ, showing Pol4 to be a double-edged sword: although it primarily would contribute to repair DSBs in cis, it may occasionally promote the repair in trans generating chromosomal translocations. The finding that classical NHEJ can be another source of chromosomal rearrangements is particularly important in yeast, where it is known that simultaneous DSBs are recruited to centralized repair centers to make the repair more efficient [49]. In this process PolX polymerases could have a relevant role, as recently suggested [50]. Interestingly, the molecular features of the yeast translocations described here resemble some translocation junctions from human cancer cells, often characterized by the presence of short nucleotide deletions and/or additions as a result of NHEJ-mediated processing [51]. Therefore, this work provides further insight to the molecular mechanisms of NHEJ, and presents a new perspective to understand how chromosomal translocations are formed in cancer cells.
Yeast strains used in this study are listed in Table S3. All yeast strains were isogenic to W303 and contained both HO and I-SCEI genes under the GAL1 promoter. Strains also had deleted the endogenous LEU2 gene and ACT1 intron. To obtain the DSB repair assay with partially-complementary ends (Figure 1) complementary oligos SacII-ISceI-SmaI-F and SacII-ISceI-SmaI-R were used (all primers used are listed in Table S4). They were annealed to generate the I-SceI cleavage site. This fragment was digested with SacII and SmaI and cloned in canonical 5′-3′ orientation at the same sites of plasmid pGLB-ACT1i-U [52] (plasmids used are listed in Table S5). The resulting plasmid (GLB-ACT1i-U-pce) was used as a template to amplify the GAL1p::leu2Δ3′::ACT1-iΔ3′::I-SceI::URA3 fragment by PCR. This fragment was then integrated in chromosome III of J00 strain as previously described [52]. To obtain a non-complementary ends system (Figure 5), complementary oligos SacII-IecSI-SmaI-F and SacII-IecSI-SmaI-R were used along with the same strategy as described above to introduce the I-SceI cleavage site in a reverse orientation in plasmid pGLB-ACT1i-U. The corresponding GAL1p::leu2Δ3′::ACT1-iΔ3′::IecS-I::URA3 fragment was then amplified by PCR using the oligos ADH4int-GAL1-F and ADH4int-URA3-R for its integration in chromosome VII of J00 strain. Chromosome integrations were confirmed by PCR and Southern analysis. Single- and double-deletion mutants (pol4Δ, yku70Δ, tel1Δ, tel1Δ pol4Δ) were generated by PCR-based gene replacement and were confirmed by PCR and Southern analysis following standard procedures.
Full-length POL4 DNA coding sequences were obtained by PCR amplification with primers CT-P4s and CT-P4as, which had ClaI and NotI cleavage sites, respectively. POL4ΔBRCT DNA sequence was obtained by PCR amplification with primers CT-P4ΔB and CT-P4as. Yeast POL4 and POL4ΔBRCT overexpression plasmids were obtained by cloning the corresponding ClaI-NotI PCR fragments under the Tet-promoter into pCM184 plasmid. POL4 single (T64A, T540A) and double (T64A,T540A and D367A,D369A) mutations in pCM184 plasmid were obtained by site-directed mutagenesis using the corresponding mutated primers. All mutated overexpression plasmids were verified by DNA sequencing. Wild-type and point mutant T540A versions were also tagged with Flag epitope by PCR amplification using primers CT-P4s and p4FLAGnot-as, together with the corresponding pCM184-[POL4] plasmids as a template. The different PCR products were digested with ClaI and NotI and then cloned into pCM184. Wild-type and mutant (T64A, T540A, T64A-T540A) POL4 versions were also fused to 6×His-tag epitope by subcloning the corresponding BamHI-NotI fragments from pCM184 into pET28c(+) vector (Novagen).
Determination of recombination frequencies was performed as described previously [52] with some modifications. Briefly, at least four independent colonies were grown until reaching the logarithmic phase in glucose-containing synthetic complete medium (SC) and then switched to glycerol-lactate (SC-3% glycerol/2% lactate). Cells in glycerol-lactate were allowed to complete one cell cycle. In such conditions, they naturally accumulate in the G1 phase of the cell cycle, allowing the DSB induction to take place when NHEJ is predominant. Appropriate dilutions were then plated on SC (glu) to determine the total cell number before DSB induction by the addition of 2% galactose to liquid cultures. After galactose addition, yeast cultures were incubated for 4 h in order to quickly induce the DSBs in G1-accumulated cells. After this incubation time, appropriate dilutions were plated onto complete galactose-containing media with (SGal) or without (SGal-Leu) leucine. Cell survival was determined by dividing the number of colonies growing on SGal after DSB repair by the number of colonies growing on SC before DSB induction. The frequency of translocations was determined by dividing the number of colonies growing on SGal-Leu by the number of colonies growing on SC (total cells). This parameter was used as a reference value to compare different strains.
To determine recombination frequencies in the repair of DSBs generated in cis we used a previously reported yeast genetic assay [34]. Briefly, appropriate dilutions of cells from overnight cultures in glycerol-lactate without uracil were spread on glucose- and galactose-containing plates. Survivor colonies on galactose-containing plates were replica-plated on SC plates containing 5-FOA (USBiological), to discriminate between Ura− and Ura+ cells. The frequency of DSB repair involving the loss of the URA3 gene was determined by dividing the number of colonies growing on SC+5-FOA by the number of colonies growing on SC.
Statistical significance of translocation frequencies in mutant strains was evaluated with the Mann-Whitney test compared to wild-type cells (in mutant strains yku70Δ, pol4Δ, tel1Δ and tel1Δ pol4Δ), or compared to pol4Δ [POL4] cells (in pol4Δ cells overexpressing mutant Pol4 versions). The distribution of repair events obtained in the different mutant strains was compared to that of wild-type strain using the Chi-square test. The distribution of repair events obtained in pol4Δ cells overexpressing mutant Pol4 versions was compared to that of pol4 [POL4] strain using the same test.
Cells were grown up to the exponential phase and were then synchronized at G1 by addition of α-factor. DSBs were induced by addition of 100 µg/ml Zeocin (Invitrogen). After 1 h incubation, cells were broken using glass-beads in lysis buffer (20 mM Hepes-KOH pH 7.5, 150 mM NaCl, 10% glycerol, 0.1% Tween-20, 1 mM phenylmethylculphonyl fluoride, Complete protease inhibitor cocktail (Roche), PhosSTOP phosphatase inhibitor cocktail (Roche)) for 20 min at 4°C. Extracts were clarified twice by centrifugation. Flag-Pol4 proteins were immunoprecipitated from supernatants with anti-Flag M2 antibody (Sigma) coupled to Protein G Sepharose 4 Fast Flow (GE Healthcare) in lysis buffer overnight at 4°C on a rotating wheel. Sepharose-bound proteins were centrifugated, washed extensively with lysis buffer and eluted in Laemmli buffer. Anti-Flag M2 antibody (Sigma) and anti-phospho [S/T]Q ATM/ATR Substrate Antibody (Cell Signaling) were used in immunoblotting experiments following standard procedures. For in vitro kinase assays, we partially purified recombinant His-tagged Pol4 proteins using Ni-NTA agarose (Qiagen) following manufacturer's instructions. Tel1-HA was immunoprecipitated from cells previously transformed with plasmid pKR5, which encodes an HA-tagged TEL1 gene [40]. Control non-transformed cells were assayed in parallel to obtain HA-immunoprecipitates without HA-Tel1 enrichment that were used as a negative control in kinase assays. Both transformed and non-transformed cells were grown to exponential phase and broken using glass-beads in lysis buffer (25 mM MOPS pH 7.2, 15 mM EGTA, 0.1% NP-40, 150 mM KCl, 1 mM DTT, protease inhibitor cocktail (Sigma), 1 mM phenylmethylsulfonyl fluoride). Extracts were clarified by centrifugation and HA-tagged Tel1 was immunoprecipitated from soluble fractions with anti-HA antibodies (Roche). Immunocomplexes were collected with Protein G-coupled DynaBeads (Life Technologies) and used in kinase assays as described previously [40].
Multiple alignment of the three Saccharomyces Pol4 DNA polymerases was done using MULTALIN (http://multalin.toulouse.inra.fr/multalin). Pol4 amino acid sequence was modeled using human Polλ PDB coordinates and Swiss-Model software (http://swissmodel.expasy.org). For tridimensional structure extrapolations, we compared this Pol4 model with crystal structure of human Polλ in a ternary complex with a 1-nt gapped DNA substrate and the incoming nucleotide (PDB code:1XSN) [48]. This was obtained from the Protein Data Bank (http://www.rcsb.org/pdb). Pol4-Thr540 residue and the corresponding point mutation was identified by using PyMol software (http://pymol.org/).
Chromosomal breakpoint analysis by PCR and DNA sequencing, and molecular karyotyping of Leu+ translocants by pulsed-field gel electrophoresis were performed as previously described [35], [52]. Breakpoint sequences from all sequenced Leu+ translocants are shown in Figure S2.
|
10.1371/journal.pgen.1006544 | Age-Dependent Neuroendocrine Signaling from Sensory Neurons Modulates the Effect of Dietary Restriction on Longevity of Caenorhabditis elegans | Dietary restriction extends lifespan in evolutionarily diverse animals. A role for the sensory nervous system in dietary restriction has been established in Drosophila and Caenorhabditis elegans, but little is known about how neuroendocrine signals influence the effects of dietary restriction on longevity. Here, we show that DAF-7/TGFβ, which is secreted from the C. elegans amphid, promotes lifespan extension in response to dietary restriction in C. elegans. DAF-7 produced by the ASI pair of sensory neurons acts on DAF-1/TGFβ receptors expressed on interneurons to inhibit the co-SMAD DAF-3. We find that increased activity of DAF-3 in the presence of diminished or deleted DAF-7 activity abrogates lifespan extension conferred by dietary restriction. We also observe that DAF-7 expression is dynamic during the lifespan of C. elegans, with a marked decrease in DAF-7 levels as animals age during adulthood. We show that this age-dependent diminished expression contributes to the reduced sensitivity of aging animals to the effects of dietary restriction. DAF-7 signaling is a pivotal regulator of metabolism and food-dependent behavior, and our studies establish a molecular link between the neuroendocrine physiology of C. elegans and the process by which dietary restriction can extend lifespan.
| Reductions in food intake have long been observed to improve longevity, extending lifespan in many evolutionarily divergent organisms. While great progress has been made in identifying the mechanisms by which nutritional interventions act to delay the aging process, much remains unclear. Particularly, while work in multiple species has found evidence that the sensation of food availability by the nervous system contributes to lifespan extension in response to reduced food levels, little is known about how these contributions are executed. Here, we have characterized how a specific neuroendocrine peptide, expressed in a set of sensory neurons, responds to changes in food conditions to modulate lifespan effects of dietary restriction at the organismal level. We further find that age-related changes in expression of this neuroendocrine signal contribute to the declining efficacy of nutritional interventions as animals get older. This work highlights the importance of neuroendocrine regulation in both the aging process and in treatments aimed at increasing longevity.
| Adult reduction in caloric intake and restriction of feeding periods have been shown to substantially increase lifespan across evolutionarily diverse organisms [1,2]. Collectively, such treatments have been referred to as dietary restriction (DR). DR has been shown to be effective even when initiated in later phases of adult life, although the efficacy of the treatment has been observed to diminish with advancing age in Caenorhabditis elegans [3,4]. Genetic studies in C. elegans have defined roles for mediators of stress response pathways, such as DAF-16/FoxO, PHA-4/FoxA and SKN-1/Nrf2, as well as the intracellular energy sensors TOR and AMPK in mediating the effects of DR on longevity [5–8]. Other studies have suggested that external cues are also critical in eliciting a DR response that extends lifespan in C. elegans [9]. In both C. elegans and Drosophila, the efficacy of DR treatment can be abrogated by the addition of food odors, and longevity in Drosophila can be extended by reduction of olfactory function [10,11]. Similarly, studies in C. elegans have shown that mutation of genes implicated in sensory systems or ablation of chemosensory neurons results in extended lifespan [12–14]. Specifically, a pair of gustatory neurons in C. elegans, the ASI neuron pair, have been shown to be required for lifespan extension in response to dietary restriction [6].
In the present study, we sought to explore the signaling mechanisms by which perceptions in the nervous system of food availability contribute to the DR response in peripheral tissues. We have focused our attention on the gene daf-7, which encodes a TGFβ ligand that is secreted from the ASI neurons to control diverse behaviors of C. elegans [15–18]. DAF-7 has previously been implicated in longevity and food sensing; daf-7 mutant animals are reported to be long-lived in a manner that is dependent on food levels and also exhibit defects in adjusting feeding behaviors in response to periods of starvation [19–21]. However, the role of DAF-7 in lifespan extension in response to DR has not been fully investigated.
Here, we have focused on expanding understanding of the role that the DAF-7 signaling pathway has in lifespan extension in response to limited nutrient availability. We have determined that DAF-7 is a key neuroendocrine signal required in the ASI neurons for response to dietary restriction. Moreover, we find that age-related changes in daf-7 expression contribute to the reduced sensitivity that older animals have to DR treatment, suggesting that the efficacy of DR interventions that delay aging can be modulated by neuroendocrine signaling.
We investigated the role of the DAF-7/TGFβ pathway in lifespan extension in response to dietary restriction using the bacterial deprivation (BD) method, where animals are moved to solid media completely lacking a bacterial food source during adulthood [3,4]. Using this protocol (see Methods for details) at 25°C and initiating BD treatment at day 3 of adulthood, we observed an average 19.5% extension of mean lifespan in wild-type animals, comparable to what has been reported previously when taking into account changes in experimental temperature (Fig 1B and 1E; S2 Table). Using multiple loss-of-function alleles, we observed that mutations in the daf-7 gene, encoding a TGFβ family ligand, or in the daf-1 gene, encoding the Type I TGFβ receptor, abrogated the lifespan extension conferred by BD (Fig 1C and 1E; S2 Table). This is consistent with a prior report which found that daf-7 mutant animals are resistant to longevity fluctuations due to altered food levels [21]. We observed that the strong dependence of lifespan extension conferred by BD on DAF-7 was temperature dependent, as daf-7 mutant animals retained lifespan extension, albeit reduced relative to wild type, when propagated 20°C (S2 Table), as reported previously [22].
Different regimens of dietary restriction have been found to extend lifespan in C. elegans through separate genetic pathways [23]. To ensure the effects we observed were not an outcome specific to the BD method of DR, we also tested daf-7 pathway mutants in a second, distinct protocol for dietary restriction, referred to as solid dietary restriction (sDR), in which adult animals are exposed to a diluted bacterial food source that is refreshed every other day [8]. Using the sDR method, we observed results consistent with our BD data, where mutants in either daf-7 or daf-1 have diminished lifespan extension in response to sDR (S1 Fig; S3 Table). DAF-7 signaling through DAF-1 has been shown to act through inhibition of the co-SMAD DAF-3 (Fig 1A) [24,25]. We found that daf-3 mutation could suppress the loss of sensitivity to dietary restriction observed in daf-7 and daf-1 mutants (Fig 1D and 1E, S1 Fig; S2 and S3 Tables).
Mutations in daf-7 have previously been observed to result in phenotypes such as diminished pumping, increased dauer entry, and increased fat storage [17]. Genetic analysis of the individual phenotypes of daf-7 mutant animals has identified distinct downstream genetic pathways that act to mediate each of these DAF-7-dependent phenotypes [17], enabling us to determine if any of these pleiotropies might be associated with the diminished ability of DAF-7 pathway mutants to respond to DR. The pumping defect of daf-7 mutants is small in magnitude compared to the decrease in pharyngeal pumping observed in feeding-defective eat mutants that are used as genetic models of DR [26,27]. Nonetheless, to test this possibility, we determined the effects of combining a daf-1 mutation with mutations in tbh-1 and tdc-1, which have been shown to suppress the feeding rate changes in daf-1 and daf-7 mutants [17]. To determine if signaling through pathways promoting dauer formation might be involved in the DR phenotype, we examined a daf-1;daf-12 double mutant. To determine if fat storage might be contributing to the DR defects we observed, we constructed daf-1 mgl-3;mgl-1 mutants, in which fat storage increases arising from diminished DAF-7 signaling are specifically suppressed [17]. None of these secondary mutations were able to suppress the BD defect of daf-1 mutant animals, decoupling these three phenotypes from the DR response that is dependent on DAF-7 signaling (S2 Fig; S2 Table).
Prior studies established that daf-7 is expressed principally in the ASI neuron pair, but also in additional sensory neurons when C. elegans is propagated on E. coli bacterial food, and that daf-7 expression is induced in the ASJ neuron pair upon exposure to metabolites of Pseudomonas aeruginosa [15,16,18]. We found that reintroducing wild-type daf-7 into daf-7(ok3125) mutants rescued the BD defect of these animals (Fig 2A and 2B). Additionally, daf-7(+) driven by ASI or ASJ specific promoters was also sufficient to rescue the BD defect of daf-7 mutant animals, consistent with the secretory nature of the DAF-7 ligand (Fig 2C). Unlike the expression of the DAF-7 ligand, the DAF-1 receptor is broadly expressed in the C. elegans nervous system [25,28]. To determine the functional targets receiving DAF-7 signal, we examined the ability of daf-1(m40) animals to respond to DR when a wild-type daf-1 transgene had been expressed in different subsets of cell types under heterologous promoters [17]. daf-1 expression in the nervous system was sufficient to restore lifespan extension in response to BD. Furthermore, as has been demonstrated for other daf-7 regulated phenotypes [17], we observed that the RIM/RIC interneurons are the specific sites of action for the daf-1 receptor for lifespan extension in response to BD treatment (Fig 2D–2F).
Given the results of our genetic analysis of the DAF-7 signaling pathway in dietary restriction, we sought to examine how daf-7 expression might change in response to DR intervention. We were unable to detect a change in expression using quantification of the transcriptional reporter, ksIs2[daf-7p::GFP], in fed versus BD treated animals (Fig 3A). We have previously observed that fluorescent in situ hybridization (FISH) provides more precise kinetic resolution of the dynamics of daf-7 transcription than does the ksIs2 GFP reporter [18]. By performing FISH on animals subjected to BD, we were able to detect a slight but consistent upregulation of daf-7 mRNA transcription. Worms exhibited an increase in daf-7 mRNA in ASI neurons in animals fixed 24 hours after BD treatment was initiated, but no detectable difference was found after a period of 5 days had passed (Fig 3B). Of note, we observed that aging adult animals began to exhibit low-level expression in the ASJ neurons, but we did not observe any changes in daf-7 mRNA in the ASJ neurons in response to BD (Fig 3A and 3C). These data suggest that in response to food deprivation, daf-7 transcription is acutely activated in the ASI neuron pair, which promotes lifespan extension mediated by DR (Fig 3D).
In response to food cues, neuroendocrine signals originating from chemosensory neurons can influence the activity of DAF-16/FoxO in the intestine [29,30]. To determine if DAF-7 signaling contributes to the DR response via DAF-16/FoxO activation, we monitored the localization of the zIs356[daf-16p::daf-16::GFP] transgene in wild-type and daf-7 mutant backgrounds. In response to food deprivation, wild-type animals shift from mostly cytosolic to nuclear localized DAF-16::GFP [29]. A daf-7 loss-of-function mutation abrogated this intestinal DAF-16::GFP translocation in BD conditions compared to wild-type animals (Fig 4). These data were surprising particularly considering that DAF-16 activation has been implicated in the setting of daf-7 loss-of-function [31]. However, we note that consistent with reports by others [19], we did observe an increase in nuclear DAF-16::GFP in the daf-7(e1372) background in other tissues such as the muscle and hypodermis in both fed and BD conditions (S3 Fig). This observation suggests that specifically in response to BD, an increase in daf-7 expression stimulates activation of DAF-16 in the intestine, which helps to promote longevity. This model is fitting with prior reports that have implicated a role for DAF-16/FoxO in mediating lifespan extension in response to various forms of DR [8,23] and in food sensing mutants [12].
We measured daf-7 expression as animals aged during adulthood using the ksIs2[daf-7p::GFP] reporter strain. We observed that daf-7 expression is maintained throughout the life of adult animals in the ASI neurons. As noted above, we also observed daf-7 expression in the ASJ neuron pair as animals age, with all animals exhibiting ASJ expression by day 3 of adulthood (Fig 5A). In contrast to the marked induction of daf-7 expression in both ASI and ASJ neurons in response to P. aerugionsa [18], in aging animals, daf-7 expression in ASJ remained relatively low (Fig 5B). Moreover, we observed that daf-7 expression in the ASI neuron pair significantly decreased with age (Fig 5B). We confirmed these findings by FISH using probes targeted to endogenous daf-7 mRNA to eliminate the possibility that these observations were an artifact of using a transgenic reporter. Our FISH results support our observations of the ksIs2 GFP reporter strain. ASI neurons from aged animals show decreased daf-7 expression; and while there is no detectable daf-7 mRNA in ASJ neurons of young animals, we were able to observe daf-7 mRNA in older adults (S4 Fig).
We sought to corroborate these changes in daf-7 expression in these sensory neurons with a measure of how much functional DAF-7 was secreted, so we utilized the cuIs5[C183::GFP] reporter of DAF-3 activity. DAF-3 negatively regulates C183 enhancer activity in vivo, resulting in low GFP fluorescence when DAF-3 is active [32]. The transgenic cuIs5[C183::GFP] reporter provides a measure of DAF-7 signal production by examining the downstream effects on DAF-3 in a neighboring tissue. We found that GFP fluorescence was diminished in an age-related, DAF-7-dependent manner, consistent with less overall DAF-7 signaling in aging worms (Fig 5C).
In addition to experiencing declines in healthspan indicators such as feeding rate and mobility, aging worms also become diminished in their ability to respond to dietary restriction treatment to extend lifespan [3,33]. We wondered if part of the insensitivity older animals have to dietary restriction treatment could be attributed to diminished levels of DAF-7 that cause an increased amount of DAF-3 activity that blocks responses leading to lifespan extension in response to DR in aging animals. To test this hypothesis, we conducted BD experiments where BD treatment was initiated at multiple time points, beginning on days 1, 3, 5 or 7 of adulthood, in wild-type or daf-3 mutant animals. We found that wild-type animals experience a robust lifespan extension when BD was began on days 1 or 3, but were unable to respond when BD was started on days 5 or 7 (Fig 5D, S4 Table), consistent with prior studies [3]. By contrast, daf-3 mutant animals were able to maintain the ability to respond to BD on day 5 (Fig 5D), suggesting that age related decline in the ability to respond to dietary restriction can be attributed, in part, to increased DAF-3 activation as a result in diminished daf-7 expression. Additionally, animals overexpressing daf-7 retain the ability to respond to BD and extend lifespan late in life at a time when wild type animals no longer exhibit lifespan extension in response to BD (S5 Fig).
DAF-7 is at the nexus of feeding behaviors and fat metabolism [17,20], suggestive of neuroendocrine links between the nervous system and secondary tissues. We have described how neuroendocrine signaling through the DAF-7/TGFβ pathway is required for lifespan extension in response to DR in C. elegans. Whereas canonical energy sensing pathways, such as AMPK and TOR, have been shown to be involved in lifespan extension in response to DR, the role of neural regulation by sensory systems of the DR response is less understood [1,10,11]. Prior studies have established the ASI neuron pair as a cell non-autonomous regulator of the DR response, identifying the insulin-like peptide INS-6 and the SKN-1/Nrf2 transcription factor as relevant agents in initiating communication to downstream cells and tissues [6,29]. We have shown that in response to DR, the ASI pair also secretes the neuroendocrine ligand, DAF-7, which signals to the RIM/RIC interneurons to suppress the co-SMAD DAF-3. In the absence of negative regulation by DAF-7, increased DAF-3 activity blocks the lifespan extension caused by DR (Fig 6).
In the developing animal, the DAF-7 ligand is produced in favorable conditions that promote entry into reproductive development, specifically in the presence rather than the absence of bacterial food [15,16]. Our data are suggestive of an acute increase in daf-7 expression in the ASI neuron pair in response to the withdrawal of bacterial food, indicating that the dynamic expression of daf-7 of developing larvae may differ from that of adult animals in response to changing environmental conditions such as DR treatments. Indeed, while bacterial deprivation extends the lifespan of adult animals, the introduction of DR-like treatments in young larvae either prompts entry into the dauer state or has detrimental effects on developing animals that have already surpassed the dauer decision checkpoint [34,35].
Whereas a recent study showed that adult animals exposed to diminishing amounts of bacterial food exhibit decreased daf-7 expression in the ASI neurons after a period of four days [21], our data, recording levels of daf-7 mRNA using FISH-based detection at multiple time points after the complete withdrawal of food, reveal a complex relationship pattern of dynamic daf-7 expression in the ASI neurons of adult animals in response to the withdrawal of bacterial food. We observe an initial increase in daf-7 expression in animals subjected to BD conditions, consistent with our genetic data implicating a requirement for DAF-3 inhibition for lifespan extension in response to BD. We observe that at later times following the withdrawal of bacterial food, daf-7 expression is maintained relative to initial levels of expression, in marked contrast to what has been observed when developing larvae are subjected to starvation conditions [15].
Our study builds upon previous observations that have linked the daf-7 gene with aging and the influence changing food levels has on longevity [19,21]. Together, our genetic findings and expression analyses support a model where active DAF-3 is sufficient to disrupt the animals’ sensory abilities and prevent lifespan extension in response to DR (Fig 6). Because daf-3 mutant worms are capable of responding to DR, the DAF-7 signaling pathway does not seem to have a direct role in altering metabolism in other tissues to extend lifespan in response to limited food levels. Rather, DAF-7 secreted by the chemosensory neurons seems to be a key neuroendocrine signal that allows animals to properly sense reductions in nutrient availability, which eventually results in activation of DAF-16/FoxO in the intestine under food deprivation. Moreover, our data suggest that an age-dependent decline in neuronal daf-7 expression also underlies the diminished sensitivity of aging animals to the lifespan effects of DR, linking a decline in neuroendocrine function to the loss of DR efficacy with advancing age.
In human aging, decline in olfactory function is one of the largest predictors of mortality- a stronger independent risk factor for death than causes such as cancer or heart failure [36]. Our study suggests that the modulation of a specific neuroendocrine signaling pathway active in chemosensory neurons involved in the sensation of bacterial food may alter the sensitivity of C. elegans to the effects of DR. We speculate that therapeutic strategies targeting analogous neuroendocrine pathways in mammals may be able to function in concert with dietary modifications to promote longevity.
C. elegans were maintained at 16°C on E. coli OP50 as previously described [37]. For a list of all strains used in this study, see S1 Table.
Due to the egg-laying defect of daf-7 pathway mutant animals, synchronized populations were prepared by egg-prep of gravid adult worms in bleach followed by L1 arrest overnight in M9 buffer. L1s were placed on OP50 seeded Nematode Growth Media (NGM) plates and raised to the L4 larval stage at 16°C. Upon reaching L4, worms were transferred onto NGM plates containing 12 μM FUDR (to prevent matricidal effects of daf-7 pathway mutants as well as progeny production) and 0.01 mg ampicillin seeded with 10X concentrated OP50 from an overnight culture and moved to 25°C (to avoid AVID [38] as well as enhance daf-7 mutant phenotypes [15,39]). Unless otherwise noted, on day 3 of adulthood (where day 0 is defined as L4 stage), worms were transferred to either fed or DR conditions on NGM plates made without peptone to prevent bacterial growth and rimmed with 150 μL of 10 mg/mL palmitic acid to prevent worms from crawling off the plates. For BD experiments, fed plates were seeded with 200 μL of 10X concentrated OP50 from an overnight culture and BD plates were unseeded. For sDR experiments, fed plates were seeded with 200 μL of OP50 at a concentration of 2x1010 bacteria/mL and sDR plates with 200 μL of OP50 at 5x108 bacteria/mL. At least 2 plates per condition were used in all experiments. Worms were scored for death (defined as failure to respond to prodding with a platinum wire) every 1–3 days beginning around day 4 of adulthood. Animals exhibiting vulval rupture were censored. Worms that crawled off the plate were never considered. Representative experiments are presented here. For lifespan statistics of individual experiments, see S2–S4 Tables.
Synchronized populations were prepared as above and treated in the same manner as worms subjected to lifespan analysis (raised to L4 16°C, then shifted to ampicillin/FUDR plates and placed at 25°C). Animals were examined for GFP fluorescence on the indicated days. All images were acquired with an Axioimager Z1 microscope using animals mounted on glass slides, anesthetized by 100mM sodium azide. Quantification of daf-7p::GFP was performed by taking the maximum intensity by FIJI software [40] within the ASI or ASJ neuron at 40X magnification. Quantification of C183::GFP was done by taking the average intensity by FIJI software [40] within the entire pharynx at 20X magnification. All quantifications were normalized by exposure time and background fluorescence (measured individually for each image). Day 3 adult zIs356[daf-16p::daf-16a/b::GFP] strains were examined on a fluorescent dissecting microscope after 4 hours of bacterial deprivation. Representative images were taken at 20X magnification. Two to four replicates were performed for all experiments presented.
Synchronized populations were established as above. FISH was performed as previously described [41]. At the indicated times and treatments, animals were washed twice with M9 buffer before fixation with 4% formaldehyde at room temperature, followed by PBS washes and suspension in 70% RNase free ethanol and stored at 4°C. To image, all samples from an individual experiment were incubated overnight with FISH probes designed against daf-7 mRNA (coupled to Cy5 dye) [18] in hybridization solution at 30°C. The next day, animals were imaged with a Nikon Eclipse Ti Inverted Microscope outfitted with a Princeton Instruments PIXIS 1024 camera. A GFP marker was used to focus on the neuron of interest and obtain a single image using a Cy5 filter. This method of image acquisition does not allow resolution of single mRNA molecules, thus quantification of daf-7 was done using FIJI software [40] to outline either ASI or ASJ and obtaining the mean intensity and subtracting background fluorescence (measured by obtaining the mean intensity of a small space immediately adjacent to the neuron being quantified). A minimum of 2 replicates was performed for all experiments presented.
The log-rank statistical test was used to determine p-values for lifespans. Using Graphpad Prism, an unpaired t-test, one-sample t-test, or one-way ANOVA was used to determine significance in quantification of expression experiments.
|
10.1371/journal.pgen.1004737 | White Cells Facilitate Opposite- and Same-Sex Mating of Opaque Cells in Candida albicans | Modes of sexual reproduction in eukaryotic organisms are extremely diverse. The human fungal pathogen Candida albicans undergoes a phenotypic switch from the white to the opaque phase in order to become mating-competent. In this study, we report that functionally- and morphologically-differentiated white and opaque cells show a coordinated behavior during mating. Although white cells are mating-incompetent, they can produce sexual pheromones when treated with pheromones of the opposite mating type or by physically interacting with opaque cells of the opposite mating type. In a co-culture system, pheromones released by white cells induce opaque cells to form mating projections, and facilitate both opposite- and same-sex mating of opaque cells. Deletion of genes encoding the pheromone precursor proteins and inactivation of the pheromone response signaling pathway (Ste2-MAPK-Cph1) impair the promoting role of white cells (MTLa) in the sexual mating of opaque cells. White and opaque cells communicate via a paracrine pheromone signaling system, creating an environment conducive to sexual mating. This coordination between the two different cell types may be a trade-off strategy between sexual and asexual lifestyles in C. albicans.
| In eukaryotic organisms, cells often undergo differentiation into distinct cell types in order to fulfill specialized roles. To achieve a certain function, different cell types may behave coordinately to complete a task that they may otherwise be incapable of completing independently. The human fungal pathogen Candida albicans can exist as two functionally and morphologically distinct cell types: white and opaque. The white cell type is thought to be the default state and may be the majority cell population in nature. However, only the minority opaque cells are mating-competent. In this study, we report that white and opaque cells show a coordinated behavior in the process of mating. When in the presence of opaque cells with an opposite mating type, white cells release sexual pheromones, and thus create an environment conducive for both opposite- and same-sex mating of opaque cells. The two cell types communicate via a paracrine pheromone signaling system. We propose that this communal coordination between white and opaque cells may not only support the fungus to be a successful commensal and pathogen in the host, but may also increase the fitness of the fungus during evolution over time.
| Sexual reproduction is pervasive in eukaryotic organisms due to its propensity to permit genetic exchange, eliminate harmful mutations, and produce adaptive progeny to changing environments [1], [2]. It has been demonstrated to be critical for environmental adaptation, morphological transitioning, and virulence of human fungal pathogens [3], [4]. However, the evolutionary advantages of sexual over asexual reproduction in single-celled organisms are extremely complex when it comes to deconvoluting the interactions between host and pathogen [5]–[7]. For example, the three most frequently isolated human fungal pathogens – Cryptococccus neoformans, Aspergillus fumigatus and Candida albicans – have all maintained their mating machinery and are capable of undergoing sexual and/or parasexual reproduction, and yet their population structures appear to be largely clonal with little or no observable recombination [5]–[7]. It has been proposed that a balance between asexual and sexual reproduction may allow pathogenic fungi to generate clonal populations to thrive in their well-adapted environmental niches and to reproduce sexually and produce genetically diverse offspring in response to novel environmental pressures [6]. C. albicans has recently been shown to undergo opposite- and same-sex mating [8]–[10]. In this study, we demonstrate that morphological transitions play an important role in the control of sexual mating, and function to balance sexual and asexual lifestyles in C. albicans. This unique mode of sexual reproduction not only confers the fungus the ability to quickly adapt to the environment as a short-term strategy, but also provides a means to generate genetic diversity in response to unforeseen challenges during evolution over time.
There are three configurations of the mating type locus (MTLa/α, a/a and α/α) in C. albicans. The majority of natural isolates are a/α at the mating type locus [11]. C. albicans can frequently undergo a transition between two distinct cell types: white and opaque [12].To mate, C. albicans must first undergo a homozygosis at the mating type locus to become a/a and α/α, and then switch from the white to the opaque cell type [13]; only opaque cells can mate efficiently. Aside from mating-competency, white and opaque cells also differ in a number of other aspects, including global gene expression patterns, metabolic profiles, cellular appearances, and virulence properties in the host [12], [14]–. The white cell type is thought to be the default state since white cells are more stable than opaque cells at the host physiological temperature (37°C) and are also less vulnerable to stresses, antifungals and host immune system attacks [16]–[18].
Given that the white cell type is the default state and that the minority population of the opaque cell type is the only mating-competent form, one would hypothesize that mating in natural conditions would be rare. If this is the case, the many advantages of sexual reproduction over asexual reproduction in C. albicans would be very limited. This also raises an interesting question, that is, why does C. albicans undergo white-opaque switching, while still retaining such a costly sexual reproduction system? The discovery by Daniels et al. (2006) of the ability of opaque cells to signal white cells to form biofilms provides a clue to answer this question [19]. White and opaque cells may coordinate to regulate pathogenesis and resistance to environmental stresses through the development of biofilms. Recently, Park et al. (2013) reported that biofilms formed by white cells facilitate opaque cell chemotropism and thus increase mating efficiency in C. albicans [20]. In addition, pheromone has been shown to up-regulate the expression of a number of mating-associated genes in mating-incompetent white cells [19], [21]. In opaque cells (MTLa/a), α-pheromone induces the expression of both MFA1 and MFα1 genes, which encode a- and α-pheromone precursors, respectively [21], [22]. Alby et al. (2009) further demonstrated that the addition of extracellular pheromone released by α opaque cells can induce same-sex mating in opaque a cells of C. albicans [10]. They found that the Bar1 protease plays a critical role in unisexual reproduction by controlling the autocrine pheromone signaling pathway [10]. Interestingly, species of the basidiomycete fungus Cryptococcus can also undergo opposite- and same-sex mating [23], [24]. Given that the population structures of these fungi are primarily clonal, unisexual reproduction may provide a long-term survival advantage, potentially increasing their ability to adapt to environmental changes.
Here we demonstrate that the interaction of white and opaque cells activates a paracrine pheromone signaling pathway in C. albicans. We further show that white cells facilitate both opposite- and same-sex mating of opaque cells. Given that the white phenotype is the default state and that opaque cells are less stable and more vulnerable in the host, our study provides additional clues to understanding how sexual mating in this organism is regulated. We suggest that the two cell types of C. albicans coordinate in order to balance the organism's commensal, pathogenic and sexual lifestyles.
In several negative controls (e.g. the “WT, wh a×WT, op α” cross) performed in a mating assay, we observed that a very high proportion of opaque α cells formed mating projections, while the mating efficiency of the cross was extremely low (Table S1). One possible explanation for this low mating efficiency is that a small proportion of white a cells spontaneously switched to the opaque state to induce the formation of mating projections. To test this possibility, a mating assay of the cross of “wor1Δ/Δ, wh a×WT, op α” was performed and mating response was examined. The wor1Δ/Δ mutant is locked in the white phase because the white-opaque master regulator Wor1 is essential for opaque cell formation [25]–[27]. As shown in Table S1, similar to the cross of “WT, wh a×WT, op α”, the mating efficiency of the cross of “wor1Δ/Δ, wh a×WT, op α” was also very low, while the proportion of opaque α cells with mating projections was over 75%. These results indicate that white a cells induce opaque α cells to form mating projections in C. albicans, but do not increase the mating efficiency of the cross between white cells and opaque cells.
To further confirm this phenomenon, we tested the effect of white a cells of four strains with different genetic backgrounds on the induction of mating projection formation of opaque α cells. The assay was performed on nutrient solid agar (Lee's glucose medium). As shown in Figure 1, over 75% of opaque α cells formed mating projections in all of the mixed cultures containing white a cells of the wild type strains. Consistently, cells of the wor1Δ/Δ and wor2Δ/Δ mutants, which are “locked” in the white phase under this culture condition, also induced mating projection formation in opaque α cells (Figure 1B). Opaque a cells served as a positive control, and white a/α cells and white α cells served as negative controls. As expected, opaque a cells induced mating projection formation in opaque α cells, while white a/α cells and white α cells did not. The images of single strain cultures and the ratios of opaque cells with mating projections are shown in Figure 1A and 1B, respectively. Consistently, white cells of another clinically independent WT a strain (SZ306a, a/Δ) and the wor1Δ/Δ mutant (GH1248, a/a) also induced mating projection formation in opaque α cells when cultured in liquid medium (Figure S1). These results indicate that the induction of mating projections of opaque cells by white cells is a general feature of clinical isolates of C. albicans.
We next examined the effect of the ratio of opaque α cells to white a cells in the mixed cultures on the formation of projections in opaque α cells. White a cells of the WT (SZ306a), wor1Δ/Δ and wor2Δ/Δ mutants were tested. As shown in Figure S2, the percentages of projections in opaque α cells were inversely related to the ratio of initial cell numbers of opaque α cells to white a cells added to the mixture. To ensure that the observed projections of the opaque cells were indeed mating projections, 4′-6-diamidino-2-phenylindole (DAPI)-DNA staining assays were performed. A single nucleus was observed by fluorescence microscopy in cells with newly formed projections (Figure S3). These results provide additional evidence that white a cells can induce mating projection formation in opaque α cells.
MFα1 is constitutively expressed in opaque α cells examined over a 48-hour growth period (Figure S4) [28]–[30], while MFA1 is poorly expressed in opaque a cells [22]. When treated with α-pheromone, MFA1 is induced in opaque a cells [22]. We hypothesized that MFA1 may also be induced in white a cells upon addition of α-pheromone to the medium or through production by opaque α cells. If true, in a mixed culture of white a cells and opaque α cells, the α cells should form mating projections as a result of exposure to a-pheromone produced by white a cells. To test this hypothesis, two reporter strains (SZ306MFA1p-GFP and wor1Δ/ΔMFA1p-GFP), in which a GFP coding sequence was integrated at the MFA1 locus and controlled by the MFA1 promoter, were constructed. As shown in Figure 2A, α-pheromone clearly induced MFA1 expression in a proportion of white cells of the two reporter strains as indicated by the GFP fluorescence. Opaque cells of the SZ306a-MFA1p-GFP strain treated with α-pheromone served as a positive control. In the mixed cultures of white a cells and opaque α cells (Figure 2B), the expression of GFP in the two reporter strains was also clear. However, the GFP fluorescence was not observed in the single strain cultures. These results indicate that the presence of α-pheromone, either from its addition to the medium or produced by opaque α cells, is able to induce the expression of MFA1 in white a cells.
MFα1 is constitutively expressed in opaque α cells, but not in white α cells (Figure S4 and S5). We next tested whether the expression of MFα1 could be induced in white α cells by a-pheromone. A MFαp-GFP reporter strain was constructed as described in the Materials and Methods section. Considering that the expression level of MFA1 is extremely low in opaque a cells in the absence of α-pheromone, opaque α cells were added to one of the mixed cultures to provide α-pheromone to the mixture. As shown in Figure S5, the expression of MFα1 in white α cells was induced as indicated by the GFP fluorescence in the mixed cultures of both “white α cells + opaque a cells” and “white α cells + opaque a cells + opaque α cells”. We were surprised that MFα1 was induced in the former culture since the expression level of MFA1 in opaque a cells is extremely low in the absence of opaque α cells. There are two possible explanations for this finding. First, low levels of a-pheromone secreted by opaque a cells may have induced MFα1 expression in white α cells. Alternatively, a small proportion of white α cells may have spontaneously converted to the opaque form and induced the expression of MFA1 in opaque a cells, which in turn induced MFα1 expression in white α cells. Overall, these results suggest that white α cells can produce pheromone and may play a similar role as white a cells in promoting an environment conducive to mating.
We hypothesized that α-pheromone produced by opaque α cells could induce MFA1 expression in white a cells. We, therefore, deleted the MFα1 gene, encoding the precursor protein of α-pheromone, in a WT α strain (α/α, GH1617). We found that opaque cells of the mfα1Δ/Δ mutant could not induce MFA1 expression in white a cells (Figure S6). Consistently, white a cells were unable to induce mating projection formation in opaque cells of the mfα1Δ/Δ mutant (Figure S6). To test whether a-pheromone is essential for the communication between white and opaque cells, we next deleted MFA1 in a WT a strain (a/a, GH1609). As shown in Figure 3, white cells of the mfa1Δ/Δ mutant failed to induce mating projection formation in opaque α cells. These results indicate that pheromones act as signaling molecules in the interaction between white and opaque cells.
The pheromone receptors (Ste2 and Ste3) and the downstream MAPK pathway are highly conserved in the regulation of sexual mating in fungi [5], [31], [32]. In Saccharomyces cerevisiae, Cryptococcus neoformans, and C. albicans, the MAPK pathway governs both mating and filamentation [33]–[36]. The downstream transcription factor Cph1 in C. albicans, a homolog of Ste12 in Saccharomyces cerevisiae, is also essential for mating [37], [38]. The Ste2/3-MAPK-Cph1 pathway is involved in pheromone response in both white and opaque cells of C. albicans [39], [40]. Given this information, we examined the role of this pathway in the interactions between white and opaque cells. As shown in Figure 3, deletion of genes (STE2, STE11, HST7 and CPH1) of the α-pheromone response pathway in white a cells blocked the induction of mating projection formation by opaque α cells. The morphological images of single strain cultures and the ratios of projected opaque α cells are presented in Figure 3A and 3B, respectively. These results suggest that both the α-pheromone response pathway and a-pheromone are required for white a cells to induce mating projection formation in opaque α cells.
The Ste3-MAPK-Cph1 pathway is required for mating of opaque cells [22], [38], [41] and is essential for pheromone-induced biofilm formation [39]. We next tested the ability of the ste3Δ/Δ, cph1Δ/Δ and cek1Δ/Δ cek2Δ/Δ opaque α cell mutants to form mating projections when co-cultured with white a cells. As shown in Figure 4, opaque cells of these three mutants failed to form mating projections, while over 80% of opaque cells of the WT control formed mating projections.
Sexual pheromones are essential for mating in C. albicans. We next tested whether white a cells could facilitate mating of opaque cells by producing a-pheromone, causing an increase in pheromone concentration to the level required to activate the mating signaling process. We designed an opposite-sex mating system, which contained 1×104 opaque a cells, 3.2×106 opaque α cells and 4.8×106 “helper” white cells. The system was so designed for the following reason. Mating between opaque a and α cells in the presence of high cell densities may increase mating efficiency. This system should amplify the promoting function of white cells by using less opaque a cells in order to reduce the high mating efficiency between opaque cells. White cells of the a/α strain (BWP17), wor1Δ/Δ (a/a), mfa1Δ/Δ (a/a), and wor1Δ/Δ mfa1Δ/Δ (a/a) mutants served as the “helper” white cells in the mating system. The wor1Δ/Δ mutant was used for this experiment because cells of this strain are “locked” in the white phase under all conditions tested [25]–[27]. As shown in Table 1, the mating efficiency of the cross when the wor1Δ/Δ (a/a) mutant served as the “helper” was about six-fold higher than that of the other three crosses with the a/α strain, mfa1Δ/Δ, and wor1Δ/Δ mfa1Δ/Δ mutants as the “helpers”. To further verify these results, we tested the roles of white cells of the wor1Δ/Δ mutants in facilitating opposite-sex mating of opaque cells in two additional genetic backgrounds (derivatives of SZ306 and SN152). Consistently, compared with the a/α strains (the WT and wor1Δ/Δ mutant), white a cells of the wor1Δ/Δ mutant (a/Δ) increased mating efficiencies by six- to nine-fold (Table 1). Consistently, deletion of MFA1 in white cells blocked this facilitation role in opposite-sex mating of opaque cells (Table 1, mfa1Δ/Δ and wor1Δ/Δ mfa1Δ/Δ mutants).
To further demonstrate that white a cells are able to facilitate opposite-sex mating of opaque cells by producing a-pheromone, we designed another mating system. We deleted the MFA1 gene in opaque cells of the “MTLa” mating partner (mfa1Δ/Δ, a/a), resulting in the failure to produce a-pheromone. White cells of the a/α strains (the WT and wor1Δ/Δ mutant) and the a strain (wor1Δ/Δ, a/Δ) served as the “helpers” in the mating cross of the “a-op (mfa1Δ/Δ, a/a)×α-op (WT, GH1349, α/α)”. Strains of two different genetic backgrounds (SZ306 and SN152) were also used as “helpers”. As shown in Table 2, mating of the “a-op (mfa1Δ/Δ, a/a)×α-op (WT, GH1349, α/α)” cross only occurred in the presence of white a cells. These results indicate that white a cells facilitate opposite-sex mating of opaque cells by producing a-pheromone.
Opaque cells of C. albicans can undergo same-sex mating in the presence of the opposite mating pheromone [10]. We, therefore, predicted that white a cells could facilitate same-sex mating of opaque α cells by producing a-pheromone in a co-cultured mating system. As shown in Table 2, opaque α cells of SZ306α and GH1349 (α/α) were unable to mate when white cells of the WTa/α (mating efficiency <5×10−9) or wor1Δ/Δ (a/α) mutant (mating efficiency <4×10−9) served as the “helper” strains. However, the mating efficiency was increased to (2.7±0.5)×10−7 (over 54 fold) when white a cells of the wor1Δ/Δ mutant (a/Δ, a derivative of SZ306) served as the “helper” strain. To further validate these results in another genetic background, we performed the same mating assays using white cells of the SN152 background strain as the “helpers”. Consistently, compared with the controls (when a/α cells severed as “helpers”), the mating efficiency was dramatically increased when white a cells of SN152a (over 450-fold increase) or the wor1Δ/Δ mutant (a/Δ, a derivative of SN152, over 220-fold increase) served as the “helpers” (Table 2). Consistently, in the absence of white α cells, same-sex mating was unable to occur in opaque a cells (Figure S5), suggesting that white α cells are also capable of promoting same-sex mating of opaque a cells.
To evaluate the in vivo relevance of our findings, we next tested whether white cells could facilitate opaque cell mating in a mammalian host. As shown in Figure 5, white a cells of the WT strains or the wor1Δ/Δ mutant strain induced the development of mating projections in opaque α cells. However, in the absence of white a cells or in the presence of white a/α cells, opaque α cells were unable to develop mating projections on the mouse skin. Quantitative mating assays also demonstrated that white a cells promoted opaque cell mating in this mouse skin infection model (Table S2). These results demonstrate that white cells are capable of facilitating opaque cell mating in a natural environmental niche.
Although white cells can be induced to produce pheromone, they are unable to mate (Table S1), suggesting that the pheromone response pathway of white cells is different from that of opaque cells. To explore how white cells respond to pheromone and how white cells create an environment conducive for opaque cell mating, we performed RNA-Seq analysis to investigate the global gene expression profile in white a cells in response to α-pheromone. Although gene expression profiling of white cells in response to pheromone have been previously published [21], [28], [42], these studies used white cells of wild type strains, which are opaque-competent (able to spontaneously switch from the white to the opaque phase). Because these strains were switching competent, a small proportion of opaque cells in the population could confound the results. In order to mitigate the effects of switching on the gene expression profile in response to pheromone treatments, we used the wor1Δ/Δ mutant (GH1602), which is “locked” in the white phase [25]–[27], for our RNA-seq experiment We believe that this dataset improves and strengthens the already published datasets, and more accurately reflects the effects of pheromone treatment on white cells exclusively. As shown in Table 3, 75 genes involved in a number of biological aspects were up-regulated and 124 genes were down-regulated in the presence of α-pheromone (using a two-fold cutoff). Our key findings are summarized below: (i) We observed differential expression in a subset of the mating-related or pheromone receptor-MAPK signaling pathway genes, including MFA1, HST6, FAV1, and STE2. Consistent with our quantitative real-time PCR (Q-RT-PCR) and MFA1p-GFP reporter results (Figure 2 and S7), the expression of the MFA1 gene was up-regulated hundreds of fold when treated with α-pheromone. Q-RT-PCR assays were performed to verify the expression levels of MFA1, STE2, and STE3, in white cells of the WT a/α, WT a, and wor1Δ/Δ mutant strains, as well as opaque a cells (Figure S7 and S8). (ii) A number of mating-related genes up-regulated by pheromone in opaque cells were not up-regulated in white cells [21], [28], [42]. These genes include FIG1, FUS1, CEK1, CEK2, FAR1, CPH1, and HST6. This result suggests that the mating and cell fusion pathways are not fully activated by pheromone in white cells as is the case in opaque cells, and provides an explanation as to why white cells are unable to mate. (iii) We observed a reduction in metabolism-related genes in the presence of α-pheromone, especially for nucleotide, lipid and fatty acid metabolism as well as for genes encoding ribosomal proteins and transporters. (iv) We also found that genes encoding cell surface proteins, which are involved in cohesion, adhesion, and biofilm formation, were differentially regulated by α-pheromone. We validated the expression levels of eight genes using quantitative RT-PCR assays (Figure S8C). Some pheromone-regulated genes observed in our study were also identified in previous studies performed in different strain backgrounds [19], [21], [42]. A detailed functional categorization and description of the differentially expressed genes in response to pheromone are presented in Figure S8 and Table S3.
White and opaque cells of C. albicans are two distinct cell types differing in a number of biological aspects [14], [16], [18]. Given that only opaque cells are mating-competent [13] and that white cells are the majority population in nature [18], the relationship between white-opaque transitions and sexual mating in C. albicans is extremely complex. These facts also raise several intriguing questions. Why is the white-opaque switch required for mating in C. albicans? What roles do white cells play in the process of sexual reproduction? How do white and opaque cells communicate?
The discovery of pheromone-induced biofilm formation in white cells of C. albicans [19], provides some intriguing clues to address these questions. It was suggested that biofilm formation by MTL-homozygous white cells in turn facilitate opaque cell mating [19], [20]. The white cell biofilm (or “sexual biofilm”) formed by MTL-homozygous white cells is distinct from that formed by MTL-heterozygous (a/α) cells. For example, the former was reported to be more permeable than the latter and to form gradients of pheromone for chemotropism [40].
In this study, we provide additional evidence for the evolution of coordination between white and opaque cells during sexual mating in C. albicans. We demonstrate that opaque cells can induce mating-incompetent white cells to secrete pheromone (Figure 2 and S5). Consistent with our data, Lin et al. recently reported that the expression level of MFA1 in white a cells was increased ∼475 fold upon treatment with α-pheromone [42]. We note that the studies by Yi et al. [39] and Sanhi et al. [43] demonstrate that the expression of MFA1 in white a cells remains unchanged in response to α-pheromone. As discussed in a recent review article [44], this discrepancy may be due to differences in laboratory growth conditions. In a system where white and opaque cells co-exist, pheromone signaling leads to the formation of a positive feedback loop, promoting the occurrence of opposite- and same-sex mating. Example scenarios of white and opaque cells co-existing, and the functional consequences of these interactions are summarized in Figure 6A and 6B, respectively. As shown in Figure 6B, opaque α cells constitutively secrete α-pheromone, which activates the pheromone response signaling pathway (Ste2-MAPK-Cph1) of white a cells. “Activated” white a cells are then induced to produce a-pheromone, which in turn activates the pheromone response signaling pathway (Ste3-MAPK-Cph1) and induces mating projection formation of opaque α cells. Of note, the expression of MFA1 is extremely low, even in opaque a cells, although it can be enhanced by treatment of the opaque a cells with α-pheromone (Figure 2 and [22]). This positive feedback loop for pheromone response is widely conserved in other yeasts. It is known that α cells can induce a-pheromone secretion of a cells in Saccharomyces cerevisiae [45]. Nielsen and coworkers reported that mating pheromone also triggers a positive feedback response in the fission yeast Schizosaccharomyces pombe [46]. This positive feedback loop for pheromone response is, therefore, a general feature in yeast species.
Since sexual mating in C. albicans is directed by the pheromone-mediated signaling pathway, it is perhaps not surprising that pheromone released by white cells is able to facilitate opaque cell mating by increasing the levels of extracellular pheromone. This is the case for both opposite- and same-sex mating of opaque cells (Tables 1 and 2). Given that the white phase is the default state, opaque cells are likely to be the minority in a natural population. In such a situation, mating between opaque cells would be rare because the concentration of pheromone produced by opaque cells would not reach the threshold required for activating the mating signaling process. Moreover, low pheromone levels do not arrest opaque cells in the G1 phase of the cell cycle [21], [47], which is a prerequisite for mating in C. albicans. In the presence of pheromone-secreting white a or α cells, the general pheromone level of the population may be increased and thus opaque cell mating could become possible. In the absence of opposite MTL type cells, same-sex mating is unable to occur due to the absence of the opposite mating type pheromone. In Figure 7, we propose a model depicting how white cells could facilitate same-sex mating of opaque cells under natural conditions. In response to α-pheromone released by opaque α cells, white a cells secrete a-pheromone and thus promote same-sex mating of opaque α cells (Figure 7A). In the absence of white a cells, same-sex mating of opaque cells could not occur (Figure 7B and 7C). Our experiments were performed on colonies on plates and on planktonic liquid cultures. We believe that both of these culture conditions are relevant for commensal and pathogenic lifestyles in C. albicans. Colonies represent an architecturally structured community, where cells exist in close proximity to one another, while the planktonic state is the state that cells exist in during a disseminated bloodstream infection. It was suggested that C. albicans can use different strategies to increase mating efficiency [20]. Alby and Bennett (2011) recently reported that interspecies pheromone signaling can promote same-sex mating in C. albicans [48]. This is interesting because C. albicans is often present with other microbiome members within the host, including other fungal, bacterial, and archaeal species. Therefore, to mate efficiently, opaque cells likely take advantage of a number of different strategies and may utilize a multitude of environmental signals to communicate in natural environments.
Sexual reproduction has many adaptive benefits over asexual reproduction in eukaryotic organisms. However, sexual reproduction is also an extremely costly process in terms of energy expenditure. How does C. albicans balance these reproductive strategies to better adapt to the changing host micro-environments, and increase its fitness during evolution over time in the host? The discovery of phenotypic switching may provide some clues to address this question. Differentiated white and opaque cells of C. albicans play specialized roles in these processes. Mating machinery can be simply shut down or activated through phenotypic transitions. Inducing expression of pheromone in mating-incompetent white cells and opaque a cells, may not only serve to save energy, but could also serve to promote sexual mating when there is a need for it. We believe that the existence of this cell type heterogeneity, creating, in a sense, a “labor division,” amongst the population, in addition to the multicellular coordination between the white and opaque cell populations, may be the primary reasons as to why this fungus is so successful at surviving and thriving in the human host as both a commensal and pathogen.
The strains used in this study are listed in Table S4. All strains used are derivatives of the following independent clinical isolates: SC5314, WO-1, SZ306, and P37005. All strains used in this study are diploid. In Figures and Tables, MTLa (or “a”) and MTLα (or “α”) indicate the mating type locus is a/a (or a/Δ) and α/α (or “Δ/α”), respectively. Modified Lee's glucose medium [49] was used for routine culture of C. albicans cells and for mating projection formation and mating assays. Construction of strains:
A 14-mer α-pheromone peptide (GFRLTNFGYFEPGK) of C. albicans was chemically synthesized. Cells of C. albicans were first grown in liquid media at 25°C for 36 hours to stationary phase and then inoculated into fresh Lee's glucose medium (1×107 cells/ml) for pheromone treatment assays. α-Pheromone peptide was added to the cultures every two hours after inoculation over an eight-hour period of growth. The final concentration of α-pheromone in the cultures was 8×10−6 M.
4×106 opaque cells were mixed with equal number of white cells of different background. The mixtures were spotted onto Lee's glucose medium plates and cultured at 25°C in air. After a 24-hour incubation period, cells were examined under a microscope and the ratio of “projected” opaque cells in each mixture was calculated.
Opposite-sex mating assays were performed according to our previous publication with modifications [50]. Briefly, 4.8×106 of “helper” white cells of different background were added to the mating mixture (1×104 opaque a cells plus 3.2×106 opaque α cells). The mating mixtures were spotted onto Lee's glucose medium plates and cultured at 25°C for two to five days as indicated in the table legends. The mating mixtures were replated onto SD-histidine-uridine, SD-arginine-uridine and SD-uridine-arginine-histidine media for prototrophic selection growth. All the “helper” strains are ura3Δ/Δ mutants and could not grow on media without uridine.
The opposite-sex mating assay of “a-op (mfa1Δ/Δ)×α-op” cross: 9.6×107 white cells of “helper” strains (60%) were mixed with 3.2×107 opaque a cells of GH1013 (a/a, ura3Δ/Δ, mfa1Δ/Δ, 20%) and 3.2×107 opaque cells of GH1349 (α/α, arg4Δ/Δ, 20%). The mating mixtures were spotted onto Lee's glucose medium plates and cultured at 25°C for five days. Then, mixed cells were replated onto SD-uridine and SD-arginine-uridine media for prototrophic selection growth.
The same-sex mating assay for the “α-op×α-op” cross: 9.6×107 white cells of “helper” strains (60%) were mixed with 3.2×107 opaque cells of SZ306α (Δ/α, ura3Δ/Δ, 20%) and 3.2×107 opaque cells of GH1349 (α/α, arg4Δ/Δ, 20%). The mating mixtures were spotted onto Lee's glucose medium plates and cultured at 25°C for five days. Then, mixed cells were replated onto SD-uridine and SD-arginine-uridine media for prototrophic selection growth. PCR of the MTLa1 and α2 was used to confirm the tetraploid colonies of “α-op×α-op” fusion and to exclude possible tetraploid α/a colonies due to the low-frequency of the “α-op×a-wh” fusion.
The same-sex mating assay for the “a-op×a-op” cross (Figure S5C). Opaque cells of GH1013h (a/a, his1Δ/Δ) were first grown in liquid Lee's glucose medium at 25°C for 24 h. Cells were then harvested and resuspended in fresh Lee's glucose medium (2×108 cells/ml) containing 10−4 M of α-pheromone peptide and incubated at 25°C for an eight-hour period of growth. 9.6×107 white cells of “helper” strains (60%) were mixed with 3.2×107 α-pheromone-treated opaque cells of GH1013h (a/a, his1Δ/Δ, 20%) and 3.2×107 opaque cells of SZ306u-a (a/Δ, ura3Δ/Δ, 20%). The mixture of opaque “a” cells (GH1013h) and opaque “a” cells (SZ306u-a) served as a negative control. The mating mixtures were spotted onto Lee's glucose medium plates and cultured at 25°C for four days. Then, mixed cells were replated onto SD-uridine and SD-histidine-uridine media for prototrophic selection growth. PCR of the MTLa1 and α2 was used to verify the tetraploid colonies of “a-op×a-op” fusion and to exclude possible tetraploid a/α colonies due to the low-frequency of “a-op×α-wh” mating.
Cells were first grown in Lee's glucose liquid medium at 25°C for 24 hours and inoculated into fresh Lee's glucose medium (1×107 cells/ml). α-Pheromone peptide was added to the cultures every 8 hours over a 24 hours period. The final concentration of α-pheromone in the culture was 1.6×10−5 M. Cells were then collected and total RNA was extracted for RNA-Seq analysis and quantitative PCR assays. RNA-Seq analysis was performed by the company BGI-Shenzhen according to the company's protocol [55]. Approximately 10 million (M) reads were obtained by sequencing each library. The library products were sequenced using the Illumina HiSeq 2000. Illumina OLB_1.9.4 software was used for base-calling. The raw reads were filtered by removing the adapter and low quality reads (the percentage of low quality bases with a quality value ≤5,>50% in a read). Clean reads were mapped to the genome of C. albicans SC5314 using SOAP aligner/soap2 software (version 2.21) [56]. The more complete and detailed RNA-seq dataset has been deposited into the NCBI Gene Expression Omnibus (GEO) portal (Accession number: GSE56039). Q-RT-PCR assays were performed to verify the relative gene expression levels of pheromone-treated and untreated samples. Quantitative PCR was performed according to our previous publication with modifications [57]. Briefly, 0.6 µg of total RNA per sample were used to synthesize cDNA with RevertAid H Minus Reverse Transcriptase (Thermo Scientific). Quantification of transcripts was performed in Bio-Rad CFX96 real-time PCR detection system using SYBR green. The signal from each experimental sample was normalized to expression of the ACT1 gene.
Skin infection assays were performed as described previously [57]. Newborn ICR mice (2 to 4 days old) were used. In vivo skin mating assay of the “a-op×α-op” cross: 1.2×108 white cells of “helper” strains (∼60%) were mixed with 2.5×105 opaque cells of GH1013h (a/a, his1Δ/Δ) and 8×107 opaque cells of GH1349 (α/α, arg4Δ/Δ, ∼40%). The mixture of opaque “a” cells (GH1013h) and opaque “α” cells (GH1349) served as a control. The mating mixtures were spotted onto the skin on the back of a newborn mouse. After water evaporated, a small sterile filter paper with First Aid tape was used to cover the area of the fungal spot. After two days of infection, C. albicans cells colonized on mouse skin were washed with PBS and plated onto SD-arginine-histidine-uridine and SD-histidine-uridine media for prototrophic selection growth.
SEM assays. 1×107 white “helper” cells were mixed with 1×107 opaque “α” cells. The mixtures were used for the skin infection. The infection method was similar to that of the quantitative mating assay. After 24 h of infection, the infected skin areas with C. albicans cells were excised for SEM assays according to our previous protocols [57].
All animal experiments were performed according to the guidelines approved by the Animal Care and Use Committee of the Institute of Microbiology, Chinese Academy of Sciences. The present study was approved by the Committee.
|
10.1371/journal.pbio.1000035 | Genome-Wide Control of the Distribution of Meiotic Recombination | Meiotic recombination events are not randomly distributed in the genome but occur in specific regions called recombination hotspots. Hotspots are predicted to be preferred sites for the initiation of meiotic recombination and their positions and activities are regulated by yet-unknown controls. The activity of the Psmb9 hotspot on mouse Chromosome 17 (Chr 17) varies according to genetic background. It is active in strains carrying a recombinant Chr 17 where the proximal third is derived from Mus musculus molossinus. We have identified the genetic locus required for Psmb9 activity, named Dsbc1 for Double-strand break control 1, and mapped this locus within a 6.7-Mb region on Chr 17. Based on cytological analysis of meiotic DNA double-strand breaks (DSB) and crossovers (COs), we show that Dsbc1 influences DSB and CO, not only at Psmb9, but in several other regions of Chr 17. We further show that CO distribution is also influenced by Dsbc1 on Chrs 15 and 18. Finally, we provide direct molecular evidence for the regulation in trans mediated by Dsbc1, by showing that it controls the CO activity at the Hlx1 hotspot on Chr 1. We thus propose that Dsbc1 encodes for a trans-acting factor involved in the specification of initiation sites of meiotic recombination genome wide in mice.
| In many organisms, an essential feature of meiosis is genetic recombination, which creates diversity in the gametes by mixing the genetic information from each parent into new combinations. Reciprocal recombination, or crossovers, also play a mechanical role during meiosis and are required for the proper segregation of homologous chromosomes to the daughter cells. Crossovers do not occur randomly in the genome but rather are clustered in small regions called hotspots. The factors that determine hotspot locations are poorly understood. We have analyzed a particular recombination hotspot in the mouse genome, called Psmb9, and showed that its activity is induced by a specific allele of a locus that we have mapped and named Dsbc1, for Double-strand break control 1. We have analyzed the properties of Dsbc1 both by the direct detection of recombinant DNA molecules in specific regions and by chromosome-wide cytological detection of proteins involved in recombination. Our results show that Dsbc1 acts genome wide and regulates the distribution of crossovers in several regions on different chromosomes, at least in part by regulating the initiation step of meiotic recombination characterized by the formation of DNA double-strand breaks.
| During the first meiotic prophase, chromosomes undergo a series of unique events that lead to the formation of stable connections between homologous chromosomes (homologs), which are required for the reductional segregation at the first meiotic division. In most species, these connections result from the formation of at least one reciprocal recombination event or crossover (CO) per chromosome arm between chromatids from homologs, stabilized by the maintenance of cohesion between sister chromatids [1]. In addition to this mechanical role, COs increase genetic diversity by reshuffling alleles in the genome and thus are thought to increase the efficiency of selection [2].
The main lines of the mechanism of CO formation have been uncovered in Saccharomyces cerevisiae and Schizosaccharomyces pombe and are conserved in other eukaryotes [3]. Meiotic recombination is induced by the formation of localized DNA double-strand breaks (DSBs), catalyzed by the evolutionary conserved Spo11 protein at the leptotene stage, corresponding to the beginning of the meiotic prophase [4]. These DSBs are repaired by homologous recombination preferentially with a chromatid from the homologous chromosome. This homologous repair leads to gene conversion with CO or to gene conversion without CO (noncrossover or NCO). One important implication of this mechanism is that the number and distribution of COs depend both on the number and distribution of DSBs and on the proportion of DSBs repaired towards COs or NCOs [5,6].
Direct analysis of DSB events in yeasts has shown that initiation occurs nonrandomly in the genome and preferentially in small regions known as recombination hotspots. In S. cerevisiae, DSBs occur almost exclusively in intergenic regions adjacent to transcription promoters [7,8]. When tested, the transcription activity of the adjacent promoter was found not to be required, although binding of transcription factors can stimulate DSB formation [9]. At several hotspots, DSBs were found to occur in accessible regions of the chromatin [10,11]. In addition to these local constraints, another control appears to act at the level of chromosomal domains. In fact, an active initiation site loses its activity when inserted in a region with no or low DSB activity, suggesting a role for higher order chromosome structure [12]. In S. pombe, the ade6-M26 initiation site is located next to the binding site of the Atf1/Pcr1 transcription factor, and DSB formation is enhanced upon binding of Atf1/Pcr1, which induces a reorganization and modification of the chromatin [13–15]. Open chromatin configurations were also detected at several other DSB sites [16]. Recently, the global analysis of DSB formation in S. pombe has shown that DSBs are clustered in small intervals separated by large regions with low or no DSB activity. DSBs do not necessarily occur next to transcription promoters; they appear to preferentially occur in large intergenic regions [17]. Thus, in both yeasts, the parameters that define an initiation site are not defined by simple features of the primary DNA sequence; they are somehow related to a control of accessibility. This notion is indeed supported by experiments in S. cerevisiae that show that DSBs can be induced by targeting Spo11 to a Gal4 binding site through the expression of a Gal4-Spo11 fusion protein [18]. In such situations, DSBs occur adjacent to the Gal4 binding site. Interestingly, constraints imposed by chromosomal domains are still observed in this case since Gal4-Spo11-induced DSBs do not form in domains with low natural DSB activity [19].
In other organisms, DSBs have not been detected directly and the distribution of recombination is derived from mapping COs, cytologically, genetically or by population diversity analysis [20,21]. In humans, CO distribution, which has been analyzed throughout the genome at high resolution, is not random and shows a specific clustering in small intervals (up to 1–2 kb long) separated by larger regions (50–100 kb) with no or low CO activity [22,23]. Interestingly, a common sequence motif has been identified in a substantial fraction of human recombination hotspots [24]. Moreover, when measured over several Mb-long domains, CO activity varies, some domains showing high (jungle) or low (desert) CO rates [25,26]. In mice, low-resolution analysis of CO also suggests similar large-scale constraints [25,27]. Several hotspots have been localized and shown to have similar properties to those described in the human genome, in particular with a tight clustering of COs in 1–2-kb intervals [21,28,29]. A recent high-resolution CO map of mouse Chromosome 1 (Chr 1) has shown that this pattern of distribution of recombination is indeed a general feature in the genome and has led to the identification of several new hotspots in the mouse genome [30].
Interestingly, variations of CO activities between individuals, sexes, and populations in human, or strains and sexes in mice have provided additional clues to identify factors that contribute to hotspot localization and activity [31]. The difference in CO activity between sexes reveals a DNA sequence-independent control of CO sites for which the mechanism, potentially related to epigenetic modifications, remains to be understood. Variations in genetic maps in different mouse hybrids have been observed [32] and could be the consequence of heterozygosity and/or of genetic factors. The contribution of genetic determinants on CO distribution and hotspot activity has been recently deduced from the analysis of interindividual variations in humans [33]. Variation of hotspot activity according to genotypes has also been shown at several sites in the mouse genome [29,34–36].
In order to analyze the molecular basis of the control of hotspot activity, we have taken advantage of the unique properties of the Psmb9 hotspot located on Chr 17 in the mouse genome. Psmb9 hotspot activity is influenced by several factors such as sex and genotype [37]. We have shown that both cis- and trans-acting elements control the initiation activity of Psmb9. The regulation in trans was deduced from the observation that a specific haplotype on Chr 17 on one homolog could activate initiation at Psmb9 on the other homolog [38]. In the present study, by generating new recombinant chromosomes, we have been able to locate the genetic element responsible for Psmb9 activity within 6.7 Mb on Chr 17. We show that this element actually acts by regulating the distribution of COs genome wide. Specific analyses on Chr 17 suggest that the regulation acts at the level of initiation. These results uncover a novel genetic control of the distribution of meiotic recombination events, predicted to influence where initiation occurs without altering the total CO activity along chromosomes. A similar control depending on a locus mapped in an overlapping region is described in the accompanying paper (Parvanov et al. [39]).
The activity of the Psmb9 hotspot has been shown to depend on the presence of the wm7 haplotype derived from an isolate of Mus musculus molossinus [36,40]. Both Psmb9 and the wm7 haplotype are located on Chr 17 and the wm7 fragment required for Psmb9 activity extends from the centromere of Chr 17 to the center of the hotspot [37,38]. The wm7 haplotype needs to be present in only one parental chromosome and leads to a 2,000-fold increase in CO activity at Psmb9 in hybrids in which the homologous chromosome can be from different genetic origins [41]. The molecular analysis of CO and NCO events at Psmb9 in hybrids with or without a chromosome carrying the wm7 haplotype showed that the variation of CO frequencies were most likely due to variations in initiation frequencies, given the parallel variations of NCO events. Furthermore, these analyses revealed that the presence of the wm7 haplotype on one homolog was sufficient to induce initiation at Psmb9 on both homologs, indicative of a regulation acting in trans [38].
To determine more precisely the position of the genetic element responsible for the high level of recombination at the Psmb9 hotspot, we generated recombinant chromosomes carrying shorter wm7 fragments by genetic crosses. The starting strain was B10.A(R209), named R209. In this strain, the wm7 haplotype extends from the centromeric region of Chr 17 to Psmb9. R209 was crossed with C57BL/10 (B10), and progeny carrying recombined Chr 17 were screened. The extent of the wm7 fragments in the progeny were determined by genotyping using microsatellite markers, and the derived strains carrying recombined Chr 17 were mated with either B10 or B10.A in order to generate hybrids in which sperm DNA could be assayed for COs and NCOs at Psmb9. Six new strains carrying different segments of the wm7 haplotype (R5, RJ3, R115, RB2, RJ2, and RK2) were thus generated (Figure 1).
Unexpectedly, in the hybrids B10xR115 and B10xRJ3, neither CO nor NCO products could be detected at Psmb9, indicating that in these hybrids, initiation activity had been lost on both homologs and that the sequences of wm7 surrounding the hotspot are not sufficient for inducing its activity. On the R115 chromosome, the wm7 fragment extends 10.2–14.1 Mb to the left of Psmb9. In contrast, B10xR5 hybrid had high CO and NCO frequencies at Psmb9 indicating that a remote element was required for Psmb9 activity. Indeed, in the RB2xB10.A hybrid in which the RB2 chromosome contains a wm7 fragment from the centromeric region up to 9.3–10.2 Mb from Psmb9, high frequencies of COs and NCOs, similar to the ones of B10xR209, were detected, demonstrating that an element from the wm7 haplotype located more than 10.2 Mb away from the hotspot is necessary and sufficient for its recombinogenic activity. Two additional recombinant Chr 17, derived from RB2 (named RJ2 and RK2), and carrying shorter wm7 fragments, retain full Psmb9 recombinogenic activity. Therefore, the genetic element necessary and sufficient for the activation of the Psmb9 recombination hotspot is located in a 6.7-Mb region of the wm7 fragment between positions 10.1 and 16.8 Mb of Chr 17.
In order to analyze the mechanism involved in Psmb9 activity, we designed an approach to determine whether the effect of the wm7 haplotype was specific to Psmb9 or if it could affect other regions on Chr 17. In order to address this question, we monitored CO activity by the detection of MLH1 foci on Chr 17 synaptonemal complex (SC) by immunofluorescent in situ hybridization (immuno-FISH) (Figure 2A). MLH1 foci are detected from middle to late pachytene, localize to future sites of chiasmata [42], and have the same global distribution as COs obtained from the genetic map [43]. Given the high CO frequency at Psmb9, which is 1.9 ± 0.6% in sperm DNA from the RB2xB10.A hybrid (Figure 1), one predicts that about 3.8 ± 1.2% of pachytene spermatocytes should have a MLH1 focus colocalizing with Psmb9 in this hybrid. We thus measured the positions of MLH1 and Psmb9 signals in B10xB10.A and RB2xB10.A hybrids. The signals were considered to colocalize when the centers of at least one Psmb9 focus and of the MLH1 focus on Chr 17 were separated by less than 300 nm (Figure 2B–2D). RB2xB10.A hybrid had a higher MLH1/Psmb9 colocalization frequency compared to B10xB10.A (4.7 ± 1.2% vs. 1.4 ± 1.2%; Fisher exact test, p-value = 0.021). This suggests that 3.3 ± 1.5% colocalization is due to the activity of Psmb9, which is in good agreement with the predicted value based on CO frequency. This analysis also indicates that the frequency of colocalization by chance is around 1.4%. The number of detectable Psmb9 foci per nucleus varied, as expected from one to four, depending on the relative positions of the four chromatids. RB2xB10.A mice showed a lower proportion of nuclei with two, three, or four foci as compared to B10xB10.A mice (Figure S1), indicating that the higher MLH1/Psmb9 colocalization frequency observed in the RB2xB10.A hybrid compared to B10xB10.A could not be explained by a higher average number of Psmb9 foci per nucleus. Taken together, these results validate our immuno-FISH approach and provide further evidence that MLH1 foci are faithful markers of CO events at our level of resolution.
We then tested whether the increased activity at Psmb9 could reflect a global increase of CO frequencies on Chr 17 by comparing the average number of MLH1 foci on Chr 17 in B10xB10.A and RB2xB10.A hybrids. These were found to be identical (Table 1). SC length was also identical in the two hybrids (Table 1). In order to identify potential differences in SC compaction, we determined the positions of three different bacterial artificial chromosome (BAC) probes along the SC of Chr 17. In both hybrids, all three BAC probes localized to the same positions (Figure S2, Table 1), indicating that chromosome compaction along the SC is not significantly different between B10xB10.A and RB2xB10.A hybrids.
In order to test whether the Psmb9 control element could affect CO frequencies elsewhere on Chr 17, we mapped MLH1 foci along SC of Chr 17 in B10xB10.A and RB2xB10.A hybrids (Figure 3A). Surprisingly, the global distribution of MLH1 foci on Chr 17 was different in the two hybrids (Figure 3B). We evaluated the differences in MLH1 focus distributions, using two independent statistical tests. In the first test, the data (MLH1 focus positions) were grouped in 20 intervals, each 5% of SC length (about 350 nm for this chromosome), corresponding to the resolution limit of the analysis. After pooling intervals when sample sizes were too low, chi-square tests were applied to compare these distributions (see Materials and Methods). In the second test, Kolmogorov-Smirnov, we compared directly the distributions of the focus positions along the SC. Both analyses revealed a statistically significant difference of MLH1 distributions between B10xB10.A and RB2xB10.A hybrids (Table 2), indicating that the Psmb9 control element affects CO distribution in several regions of Chr 17. In some intervals, the number of MLH1 foci was similar in both hybrids (for instance, intervals from 85% to 95% of SC length). In contrast, in other intervals, the RB2xB10.A hybrid displayed either a higher (region from 40% to 60% of SC length) or a lower density of MLH1 foci (region from 70% to 85% of SC length) as compared to B10xB10.A. At least two interpretations could be considered to account for these variations: either the unusually high recombination frequency at Psmb9 leads to compensatory changes elsewhere on the chromosome or the Psmb9 control element has independent effects in different regions of Chr 17.
In order to distinguish between these possibilities, we mapped MLH1 foci along Chr 17 in SGRxSGR mice. In fact, these mice contain the genetic element–activating Psmb9; however, the Psmb9 hotspot has low or no CO activity due to the presence of a repressive element located at the distal side of Psmb9 on the SGR chromosome [38]. This was deduced from the analysis of B10xSGR and B10.AxSGR hybrids, and leads to the prediction that the same control should operate in the SGRxSGR mice analyzed here. In SGRxSGR mice, using the Kolmogorov-Smirnov test, we found that the MLH1 distribution was significantly different from that of B10xB10.A, but not from RB2xB10.A (Figure 3C, Table 2). In particular, the region 70%–85% showed a significant reduction of MLH1 foci in SGRxSGR compared to B10xB10.A (Fisher exact test, p = 0.00037). These results suggest that the redistribution of CO events along Chr 17 observed in RB2xB10A is not a consequence of the high activity at Psmb9, is not specific to the hybrid RB2xB10.A, and is rather due to the presence of the wm7 regulatory element having multiple independent effects in several regions of Chr 17.
The analysis of cis and trans regulations of Psmb9 led to the conclusion that the high recombination activity at Psmb9 is due to a regulation of the initiation rate [28,38]. In order to test whether the wm7 regulatory element in RB2xB10.A modifies the frequency of initiation events along Chr 17 or only alters the choice of DSB repair towards COs or NCOs, we mapped γH2AX foci, the phosphorylated form of H2AX, along this chromosome. Among the markers that have been used to mark all meiotic DSB repair events, such as RAD51, DMC1, RPA, MSH4, and γH2AX, γH2AX gave the most reliable signals with our immuno-FISH protocol. γH2AX is detected as large chromosomal domains at leptotene and zygotene [44] and forms small foci on the SC at the beginning of pachytene [45,46]. The numbers of γH2AX foci, being in excess as compared to the number of MLH1 foci, are predicted to mark both CO- and NCO-designated events. This analysis shows an average number of 2.76 ± 0.17 and 2.67 ± 0.18 γH2AX foci per chromosome in B10xB10.A and RB2xB10.A hybrids, respectively. Interestingly, in both hybrids, the overall γH2AX distributions were significantly different from that of MLH1, with a larger fraction of foci on the proximal half of the chromosome (Figures 3–4, Table 2). This observation is similar to that described from the comparison of MSH4 or RPA and MLH1 foci and is thought to reflect specific CO regulations such as interference and crossover assurance [47]. In addition, the overall γH2AX distributions were significantly different in B10xB10.A and RB2xB10.A hybrids as determined by chi-square analysis (Table 2). Furthermore, regression analysis showed a significant correlation (R2 = 0.528; p < 0.01) between the variations in MLH1 and γH2AX foci distributions in these two hybrids (Figure 5). These results strongly suggest that the wm7 regulatory element is not simply inducing a redistribution of CO events on Chr 17 by regulating pathways of DSB repair but is influencing the formation of meiotic DSBs in several regions of this chromosome. We therefore named this genetic element, Dsbc1 for Double-strand break control 1 and the two alleles Dsbc1wm7 (present in the wm7 haplotype) and Dsbc1b (B10 strain).
As the presence of the wm7 fragment leads to a redistribution of DSB and CO events over long chromosomal distances, we addressed the question of whether this regulation was specific to Chr 17 by monitoring CO distribution on other chromosomes. We thus analyzed MLH1 foci along Chrs 2, 15, and 18 in B10xB10.A and RB2xB10.A hybrids. For all three chromosomes, the total CO activity estimated by the average MLH1 number was identical in the two hybrids (Table S1). Using both chi-square and Kolmogorov-Smirnov tests, MLH1 distributions were found to be significantly different in B10xB10.A and RB2xB10.A on Chrs 15 and 18, showing that Dsbc1 acts in trans on different chromosomes. We did not detect significant differences on Chr 2, however (Figure 6, Table 2).
In order to show at the molecular level and at high resolution that Dsbc1 regulates recombination activity on another chromosome, we tested the recombination activity at the Hlx1 hotspot located on Chr 1 [30] in hybrids carrying the wm7 haplotype on Chr 17. Parvanov et al. [39] have shown that the activity of this hotspot requires an allele from Mus musculus castaneus at a genetic locus located on Chr 17 and named Rcr1 (see Discussion). Using an allele-specific PCR assay, we found a high CO activity at Hlx1 in the presence of the Dsbc1wm7 allele (for one type of recombinant molecule, f(CO) = 0.1 ± 0.07%) and no detectable CO in the presence of Dsbc1b (f(CO) < 0.015%; Figure 7). The frequencies of CO were found similar in the presence of Dsbc1wm7 and in hybrids containing a Chr 17 from M. m. castaneus.
We report the identification of a novel regulatory element of the distribution of meiotic recombination events located between positions 10.1 and 16.8 Mb on mouse Chr 17. Comparative analysis between several hybrids suggests that a single locus, named Dsbc1, is regulating the distribution of crossing over in several regions of Chrs 15, 17, and 18. Two different alleles of Dsbc1 appear to be present in B10 and mice derived from M. m. molossinus. Direct analysis of recombinant molecules shows that the M. m. molossinus allele of Dsbc1, Dsbc1wm7, stimulates the CO activity of two recombination hotspots, Psmb9 on Chr 17 and Hlx1 on Chr 1. We also demonstrate that Dsbc1 controls the distribution of γH2AX foci along Chr 17. We thus propose that the Dsbc1 locus controls the distribution of initiation of meiotic recombination genome wide in mice.
The Psmb9 hotspot was known to be active in hybrids in which one parent carries the wm7 haplotype from M. m. molossinus, the region of the wm7 haplotype required for this activity extending from the centromere to the center of the hotspot. By genetic analysis, we show that the region necessary and sufficient for Psmb9 activity is located between positions 10.1 and 16.8 Mb of Chr 17, at least 17.5 Mb away from Psmb9, indicating a long-distance control. Interestingly, the Psmb9 hotspot was also found to be active in hybrids in which one parent carries the cas3 MHC haplotype derived from M. m. castaneus [48]. This haplotype extends from the MHC to at least position 18.6 Mb towards the centromere (unpublished data) and therefore potentially overlaps with the regulatory element from the wm7 haplotype. It is important to note that in most hybrids tested, Dsbc1 is heterozygous, indicating a dominant or semidominant effect of the Dsbc1wm7 allele over the Dsbc1ballele from the B10 strain. Heterozygosity is not required, however, since we have shown high Psmb9 activity in hybrids homozygous at Dsbc1wm7 [38].
Based on measurements of CO and NCO frequencies, we infer that the enhanced recombination activity at Psmb9 results from an enhancement of initiation frequency. The only alternative hypothesis would be to assume that B10 strain has a high initiation frequency at Psmb9, but that all DSBs are repaired using the sister chromatid, an unlikely hypothesis given the repression of sister-chromatid recombination during meiosis. Given that the presence of Dsbc1wm7 on one homolog is sufficient to promote initiation on both [38], this locus acts, not only at long distance, but also in trans. A complementary approach to follow initiation events during meiosis is through the detection of γH2AX foci which mark meiotic DSBs [44]. Using this approach, we have shown that the distribution of γH2AX on Chr 17 is significantly different in B10xB10.A and RB2xB10.A hybrids, therefore in the absence or presence of the Dsbc1wm7 allele. Although indirect, this observation is in agreement with the molecular analysis at Psmb9, and thus suggests that the Dsbc1 locus controls the initiation of meiotic recombination in multiple genomic regions on Chr 17. Through the analysis of MLH1 focus distribution on Chr 17, we detect significant differences between the hybrids tested and found that the variations of γH2AX distribution are strongly correlated with those of MLH1. We therefore consider that these γH2AX variations are not likely due to modifications of the pathway leading to H2AX Ser129 phosphorylation that would be independent from DSB formation. We thus propose that Dsbc1 regulates initiation activity and, therefore, CO rates. We do not exclude, however, that Dsbc1 also affects other aspects of the recombination pathway after initiation.
In order to test whether the Dsbc1 locus acts on other chromosomes, we determined by immuno-FISH the localization of MLH1 foci on Chrs 2, 15, and 18. In all cases, the total CO activity as measured by the average number of MLH1 foci per chromosome was not changed. However, excepted for Chr 2, we found significant differences in the distribution of MLH1 foci along Chrs 15 and 18, depending on the presence or absence of the Dsbc1wm7allele. These differences are validated both by chi-square analysis after subdividing the data in intervals and by a comparison of the focus position distributions by the Kolmogorov-Smirnov test. In some regions, CO is increased, in others decreased, or not affected. The detection of these effects in single intervals is obviously limited by our assay in two respects: The spatial resolution and the statistical significance when sample size is small. The spatial resolution of MLH1 foci is close to 300–400 nm, corresponding to about 5 Mb or 1/20th of the chromosome length (Chrs 15, 17, and 18 are of similar lengths). If Dsbc1 controls initiation as proposed above, it might do so independently at single sites or over domains, or both, and the net result over 5 Mb intervals is the sum of these effects. In the case of Chr 2, the number of foci measured per unit of physical distance was lower than on other chromosomes, potentially weakening the statistical significance of the measures. Pooling data to normalize for focus number per interval did not reveal significant differences, possibly due to decrease of spatial resolution. Within these constrains, it is not possible to determine which fraction of the hotspots are influenced by Dsbc1.
As a direct molecular demonstration that Dsbc1 regulates, not only the Psmb9 hotspot, but also hotspots on other chromosomes, we tested the Hlx1 hotspot on Chr 1. We found that indeed the presence of Dsbc1wm7 leads to a high CO frequency at Hlx1. This result not only demonstrates that Dsbc1 acts on other chromosomes, but also that it might correspond to the same locus as one identified by Parvanov et al. [39]. In fact, the work by Parvanov et al. reports the identification of a locus, named Rcr1, whose position overlaps with the one of Dsbc1 and which influences initiation activity at Hlx1. More specifically, the Rcr1 allele from the Cast/Eij strain (M. m. castaneus) leads to a high CO frequency at Hlx1 as well as variations of CO rates at several hotspots in the neighboring region of Chr 1. Genetically, although the mouse strains used in both approaches were different, it is interesting to note that M. m. molossinus is a hybrid between M. m. domesticus and M. m. castaneus. It is therefore possible that the Dsbc1wm7allele is derived from M. m. castaneus. Alternatively, the two loci Rcr1 and Dsbc1 might be distinct.
Our analysis reveals that the strains from B10 background and from M. m. molossinus wm7 haplotype have two different alleles at the Dsbc1 locus, Dsbc1b and Dsbc1wm7. Although the size of the genomic fragment from wm7 origin is large and differs in several hybrids, the consistency of the effects leads us to assume that these are under the control of this single locus. The presence of one Dsbc1wm7 allele is sufficient to induce hotspot activity at some sites (Psmb9 and Hlx1, for instance). This allele thus behaves as a dominant trait specifying initiation at new locations. However, given that the total amount of recombination is constant (estimated by γH2AX and MLH1 foci), one expects the activity at other hotspots to decrease in the presence of Dsbc1wm7 (heterozygous or homozygous). This could be due either to a regulation of the frequency of initiation events such as the one observed in S. cerevisiae in which DSBs can be suppressed by insertion of strong DSB sites nearby [19,49] or to a competition between Dsbc1wm7 and Dsbc1b alleles. These observations raise several questions: Is the Dsbc1wm7 allele an inducer of new initiation sites? Does the Dsbc1b play a role in initiation site specification? If so, which fractions of sites are controlled by the Dsbc1 locus? At the molecular level, Dsbc1 is expected to be a diffusible factor, RNA, or protein, that could act either on the recombination machinery involved in initiation or at the level of the substrate. One could, for instance, envision that Dsbc1 controls directly or indirectly some aspects of chromatin accessibility that are known to be important for initiation site specification as shown by studies in yeast [50]. We have indeed obtained evidence that the structure of the chromatin at Psmb9 differs in the context of Dsbc1wm7 or Dsbc1b (J. Buard, P. Barthès, C. Grey, and B. de Massy, personal correspondence).
Modifiers of recombination have been described in fungi. In particular, in Neurospora crassa, three loci, rec-1, rec-2, and rec-3, have been found to influence meiotic recombination in a region-specific way and were proposed to modify patterns of initiation [51]. Recently, in Caenorhabditis elegans, mutations in a subunit of a condensing complex, DPY-28, were found to alter CO distribution by influencing DSB activity. Differently from the effect mediated by Dsbc1, dpy-28 mutants show an overall increase of CO frequencies [52]. However, such observations indicate that modifications in structural components of chromosomes can affect recombination initiation, and Dsbc1 could therefore interact directly or indirectly with these components.
Genetic control of the frequency of meiotic recombination has been suggested from analysis of several hotspots in the mouse genome. For instance, a very active hotspot was identified in the Eβ gene, for which activity was suppressed in the presence of the p haplotype [53,54]. Although the Eβ hotspot is active in many different genetic contexts, its activity is highly variable in different hybrids [29,54]. The activity of the Eβ hotspot was suggested to depend on distal elements or to be influenced by genetic background [34]. Among other less-characterized hotspots some are haplotype-specific such as the Pb hotspot, active in the presence of the cas4 haplotype derived from M. m. castaneus [55,56], whereas others, such as G7c, have detectable activity in a variety of hybrids [57]. Whether these variations depend on local or distal controls is not known. In humans, the regulation of CO activity by remote control has been proposed from the analysis of the MSTM1a hotspot showing variable CO frequencies in men with identical haplotypes in the 100-kb region around the hotspot [58]. In addition, genome-wide mapping has shown a heritable variation in hotspot activity between individuals [33]; it would be interesting to know whether specific loci are responsible for such effects and whether any locates to the region of the human genome synthenic to that of Dsbc1.
If a single locus can modify initiation activity in such a dramatic way as we have uncovered, it obviously has consequences, not only on genetic maps as mentioned above, but also on their evolution. In particular, it provides an answer to the hotspot paradox. In fact, due to the directionality of mismatch repair during DSB repair in meiosis, the initiating chromatid is the recipient of genetic information. Consequently, in a population, if an allele appears that lowers initiation frequency in cis, and is located in the region subject to gene conversion, its frequency will tend to increase, and it will eventually be fixed. This phenomenon has been called the hotspot paradox because it is expected to lead to loss of hotspot activity in the genome and to strongly limit the appearance of new hotspots [59]. In fact, we have already reported that at the Psmb9 hotspot, a suppressive mutation, as observed in the SGR strain, might have spread rapidly through the population due to the disparity of gene conversion and the very high efficiency of the Psmb9 site in this context [38]. The forces that maintain hotspot activity remain unknown. If the activity of a hotspot can be influenced by elements outside the region that is subject to gene conversion, alleles that lower initiation frequency might not necessarily be driven to fixation, and new hotspots could appear independently from mutations in the region of the hotspot itself [60]. We propose that elements such as Dsbc1 that regulate in trans the distribution of recombination sites can counteract the loss of activity at some sites. If true, this suggests that the Dsbc1 locus itself might be under specific selection to maintain the recombination activity at an appropriate level in the genome.
The mouse strains used in this study were C57BL/10JCrl (B10), B10.A, B10.MOL-SGR (SGR), B10.A(R209) (R209), and CAST/EiJ (CAST). The strains R5, RJ3, and RB2 have been obtained by back-crossing B10 x R209 F1 with B10 and screening the progeny for CO events between D17Mit164 and the Psmb9 hotspot. The recombinant Chr 17 of strains RJ2 and RK2 have been generated in a RB2xB10.A F1 and a RB2xB10 F1 hybrid, respectively. The recombinant Chr 17 of strain R115 has been generated in a Mlh1tm1Lisk × R209 F1 hybrid, backcrossed with R209. For the molecular analysis at the Hlx1 hotspot, mice heterozygous for the B10 and CAST alleles at the chromosome 1 Hlx1 hotspot and carrying the appropriate genotype on Chr 17 have been generated as follows: CASTxR209 F1 hybrids were generated and crossed with B10. Males heterozygous at the Hlx1 hotspot and containing either the haplotypes from B10 and CAST (b/c) or the haplotypes from B10 and R209 (b/wm7) on Chr 17, were analyzed. In addition, b/c females were crossed with B10 in order to generate males heterozygous at the Hlx1 hotspot and homozygous b/b on Chr 17. The region of Chr 17 taken into account in this analysis is the proximal third, from D17Mit164 (position 3.9 Mb) to D17Mit21 (position 34.4 Mb).
The markers used for mapping the fragments derived from R209 in the recombinant lines are listed in Table S2. Their position on Chr 17 sequence is based on the National Center for Biotechnology Information (NCBI) m37 mouse assembly. The sequences of the primers used for amplifying the microsatellite markers have been found on the Mouse Genome Informatics site (http://www.informatics.jax.org/). The PCR cycling conditions were 94 °C for 10 s, 55 °C for 30 s, and 72 °C for 30 s for 36 cycles.
The allele-specific PCR protocol for the direct molecular detection and analysis of recombination products at Psmb9 and at Hlx1 in sperm or testis DNA was as described [38,61]. For Hlx1, the cycling conditions of the first PCR using the primer pair BF1/CR1 were: denaturation at 94 °C for 2 min followed by 25 cycles at 94 °C for 15 s, 64 °C for 30 s, 68 °C for 210 s, and 68 °C for 5 min as the final extension step. The cycling conditions of the second PCR using the primer pair BF2/CR2 were: Denaturation at 94 °C for 2 min followed by 28 cycles at 94 °C for 15 s, 61 °C for 30 s, 68 °C for 210 s, and 68 °C for 5 min as the final extension step. The primers used for the detection of COs and NCOs are listed in Table S3.
Spermatocyte nuclei spreads were performed by the dry-down technique as described [62].
Chromosome-specific BAC probes were labeled with FISH Tag DNA Kit (Invitrogen) according to the manufacturer's manual. BAC probes were selected to be between 120 and 220 kb in size and poor in repetitive elements. On Chr 17, for colocalization assays, analysis of focus distribution, and SC length measurements, we used two BACs (RP24-67L15 and RP23-10B20) that locate at 86.3 Mb and 87.6 Mb, respectively, on Chr 17 and were mixed together. To determine the position of Psmb9 on SCs, one BAC probe that covers the Psmb9 sequence was used (RP23-95J18). To determine the global SC structure, we used two different BAC probes (RP23-268M4 and RP24-279K24) that locate at 46.2 Mb and 73.4 Mb, respectively. For Chrs 2, 15, and 18, we used single BAC probes (RP23-101G16, RP23-408L3 and RP24-266A12, respectively). To identify the Psmb9 hotspot, six DNA fragments of 800 to 1,200 bp covering a region of 10 kb around Psmb9 were amplified by PCR (P1U: 5′-CCCCTTCCTGTAGACAT-3′, P1L: 5′-ACAAATAAGCATATACCACG-3′, P2U: 5′-GCCATGTTATTTCTTGACATG-3′, P2L: 5′-CCACACAGGATAAATAATGCT-3′, P3U: 5′-AAATTAAAAAGTCAACCC-3′, P3L: 5′-AGCTGGAGTTACAGGTGTT-3′, P4U: 5′-CTCTGGACCACAAAGCTAGAA-3′, P4L: 5′-CAGAGCCAAGCACATCTAACT-3′, P5U: 5′-GCAACGGTGGTTGTATGG-3′, P5L: 5′-GAAGGTGTGGGGGAAGTAGAT-3′, P6U: 5′-GAATGAGCTTCCCAAGTTGAC-3′, and P6L: 5′-CCCTGGCCTGTCGTGTT-3′). An equimolar mix of all PCR fragments was biotinylated with the BioNick Labeling System (Invitrogen) according to the manufacturer's manual.
For the FISH, we adapted the protocol of H. Scherthan [63] to our purpose. Briefly, spermatocyte spreads were treated with NaSCN (1 M) at 70 °C for 30 min; washed in 1× sodium saline citrate (SSC) solution, and digested with RNAse A (100 μg/ml) at 37 °C for 1 h. The slides were then washed with 1× SSC solution, covered with a solution of 70% formamide/2× SSC, and the DNA was denatured by heating at 85 °C for 5 min. After denaturation, the slides were immediately washed in ice-cold 2× SSC. Probes (300 ng per slide for Psmb9 PCR probe or 80 ng per slide for BAC probes) were denatured in 10–15 μl of hybridization solution (50% formamide, 2× SSC solution, 10% dextran, and mouse Cot1-DNA 10 ng/μl) for 3 min at 93 °C. Hybridization solution was applied to slides, covered with a cover slip, sealed with rubber cement, and incubated in a humid chamber at 37 °C for 24 h. The slides were then washed in 0.05× SSC at 37 °C and treated for immunostaining.
Immunostaining was performed as described [64], using a milk-based blocking buffer (5% milk, 5% donkey serum in 1× PBS, phosphate buffer saline). Antibodies were guinea pig anti-SYCP3 at 1:500 dilution, mouse monoclonal anti-MLH1 (Pharmigen) at 1:50 dilution, and mouse monoclonal anti–phospho-H2A.X (Upstate) at 1:20,000. All incubations with primary antibodies were performed overnight at room temperature. Secondary antibodies were goat anti–guinea pig Alexa Fluor 488 (Molecular Probes) and donkey Cy3-conjugated anti-mouse. Incubations with secondary antibodies were performed at 37 °C for 1.5 h. Psmb9 PCR probe was revealed with Cy5-conjugated Streptavidin (Jackson Immuno Research) at 1:750 dilution, incubation was performed at 37 °C for 1 h. Incubations with secondary antibodies were performed simultaneously, whereas incubation with Cy5-conjugated Streptavidin was always preformed last. Nuclei were stained with DAPI (4′-6-Diamidino-2-phenylindole, 2 μg/ml) during the final washing step.
Digital images were obtained by using a cooled charge-coupled device (CCD) camera, Coolsnap HQ (Photometrics) coupled to a Leica DM 6000B microscope. Each color signal was acquired as a black-and-white image using appropriate filter sets and was merged with Metamorph Imaging software.
The identity of Chr 17 was determined by the use of a mix of two Chr 17–specific BAC probes (RP24-67L15 and RP23-10B20). To asses the colocalization frequency, only nuclei that were MLH1 positive on Chr 17 were counted. Distances between the center of MLH1 and Psmb9 foci were measured with Metamorph Imaging software. Colocalization was considered when the distance between the centers of both foci (MLH1 and Psmb9) was less than 300 nm (4.6 pixels with a 100× lens).
The chromosome to be analyzed was identified using a BAC probe that was revealed with a different color than the SC. SC length was measured with Metamorph Imaging software. To determine the distribution of MLH1 (or γH2AX) foci, the position of each focus on the chromosome was recorded as a relative distance (percentage of total SC length) from the centromere. The centromeric end was identified by surrounding DAPI-bright heterochromatin. MLH1 (or γH2AX) foci were mapped when MLH1 (or γH2AX) and SYCP3 signals overlapped.
Distributions were compared both by chi-square analysis and by the nonparametric Kolmogorov-Smirnov test. For chi-square analysis, all focus positions were grouped in 20 intervals of 5% SC length (about 5 Mb) for Chrs 15, 17, and 18 and in 35 intervals of 2.85% SC length (about 5 Mb) for Chr 2 (Tables S4, S5, and S6). Interval contents were systematically examined starting from the centromere proximal end for each chromosome. If necessary, intervals were pooled such as to have expected sample sizes of five foci minimum. For each chromosome and each hybrid, data were obtained from two different mice (except for Chr 2, one mouse, and for MLH1 analysis on Chr 17 in RB2xB10.A, four mice). In each case, the homogeneity of the two sets of data was validated by chi-square and Kolmogorov-Smirnov analysis showing the reproducibility of distributions for a given genotype on a given chromosome (Table S7). The Kolmogorov-Smirnov test compares the distribution of foci positions on the SC and quantifies the distance between two distributions. For the analysis of variations of γH2AX and MLH1 focus densities on Chr 17, the ratios between focus percentages per 5% SC length intervals in RB2xB10.A and in B10xB10.A were calculated (values of intervals 0%–25%, 25%–35%, and 35%–45% were pooled due to the small number of MLH1 foci in these regions), for each type of foci (γH2AX and MLH1). A linear regression analysis was performed to compare the variations of γH2AX and MLH1 focus densities.
|
10.1371/journal.pntd.0002159 | Virus-Specific Differences in Rates of Disease during the 2010 Dengue Epidemic in Puerto Rico | Dengue is a potentially fatal acute febrile illness (AFI) caused by four mosquito-transmitted dengue viruses (DENV-1–4) that are endemic in Puerto Rico. In January 2010, the number of suspected dengue cases reported to the passive dengue surveillance system exceeded the epidemic threshold and an epidemic was declared soon after.
To characterize the epidemic, surveillance and laboratory diagnostic data were compiled. A suspected case was a dengue-like AFI in a person reported by a health care provider with or without a specimen submitted for diagnostic testing. Laboratory-positive cases had: (i) DENV nucleic acid detected by reverse transcriptase-polymerase chain reaction (RT-PCR) in an acute serum specimen; (ii) anti-DENV IgM antibody detected by ELISA in any serum specimen; or (iii) DENV antigen or nucleic acid detected in an autopsy-tissue specimen. In 2010, a total of 26,766 suspected dengue cases (7.2 per 1,000 residents) were identified, of which 46.6% were laboratory-positive. Of 7,426 RT-PCR-positive specimens, DENV-1 (69.0%) and DENV-4 (23.6%) were detected more frequently than DENV-2 (7.3%) and DENV-3 (<0.1%). Nearly half (47.1%) of all laboratory-positive cases were adults, 49.7% had dengue with warning signs, 11.1% had severe dengue, and 40 died. Approximately 21% of cases were primary DENV infections, and 1–4 year olds were the only age group for which primary infection was more common than secondary. Individuals infected with DENV-1 were 4.2 (95% confidence interval [CI]: 1.7–9.8) and 4.0 (95% CI: 2.4–6.5) times more likely to have primary infection than those infected with DENV-2 or -4, respectively.
This epidemic was long in duration and yielded the highest incidence of reported dengue cases and deaths since surveillance began in Puerto Rico in the late 1960's. This epidemic re-emphasizes the need for more effective primary prevention interventions to reduce the morbidity and mortality of dengue.
| Dengue is a potentially fatal acute febrile illness that is endemic throughout the tropics and sub-tropics. Dengue has been endemic in Puerto Rico for several decades and recent epidemics occurred in 1994–5, 1998 and 2007. In January 2010, dengue surveillance indicated that an epidemic had begun. The epidemic peaked in early August and ended in December with a total of 26,766 suspected dengue cases identified, of which 128 were fatal. The 2010 epidemic was one of the longest in Puerto Rico history and resulted in the greatest number of cases and deaths ever detected. We analyzed the epidemiologic and immunologic characteristics of laboratory-confirmed dengue cases and age group-specific attack rates, and determined the frequency of first DENV infection and DENV-types among persons experiencing their first infection. This analysis indicated that 10–19 year-olds were most affected during the epidemic, and that DENV-1 was roughly four times more likely to be associated with clinically apparent illness upon first DENV infection than were DENV-2 or -4. The 2010 dengue epidemic demonstrated the heavy burden of illness due to dengue in Puerto Rico, re-emphasizing the critical need for effective primary prevention tools to reduce the morbidity and mortality due to dengue worldwide.
| Dengue virus (DENV) transmission is endemic throughout most of the tropics and sub-tropics and is estimated to result in ∼50 million symptomatic infections and ∼20,000 deaths each year [1], [2]. Infection with any DENV can result in dengue, an illness characterized by fever, headache, retro-orbital eye pain, myalgia and rash [2]. In some cases, dengue can progress to severe dengue [2], which includes dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) [3] and is characterized by thrombocytopenia, increased vascular permeability with plasma leakage, severe organ involvement, and/or clinically significant bleeding [2]. Supportive care with appropriate intravascular volume repletion has been shown to lower mortality associated with severe dengue [2].
The four related but serotypically distinct DENV-types, DENV-1, -2, -3 and -4, are transmitted by Aedes aegypti or Ae. albopictus mosquitoes [4], [5]. Following infection, individuals develop short-lived, heterotypic immunity and long-lived, type-specific immunity [6]. Primary infection is an individual's first DENV infection, and secondary infection is any subsequent infection with a DENV-type different from the first. Severe dengue is more common upon secondary infection [2], [7] and may be affected by the order in which an individual is infected with the respective DENV-types [2], [8]. Thus, increases in the force of DENV infection can result in a decrease in the age of primary and secondary infection [2]. Both local patterns of circulation of the four DENV-types and force of infection can influence the age groups most affected by dengue and severe dengue.
The unincorporated United States territory of Puerto Rico is composed of 78 municipalities, an area of 3,515 square miles, and a population of 3,725,789 [9]. The demographics of Puerto Rico are similar to the United States as median age is 36 years and 78.6% are white, although 99% are self-described Hispanic [9]. Since the mid-1990's the health care system in Puerto Rico has included both public and private health care services, and dengue has been a reportable condition for several decades. Ae. aegypti is the predominant DENV vector on the island.
Dengue was first described in Puerto Rico in 1915 [10] and outbreaks have been recognized since 1963 [11], [12]. DHF was first reported in 1975 [13], [14], all four DENV-types have been identified on the island since 1982 [15], [16], and the first confirmed dengue-related death was reported in 1986 [17]. Recent epidemics were detected in 1994–1995, 1998 and 2007, with 24,700 [18], 17,000 [19] and 10,508 [20] reported suspect cases, respectively (Table S1). During both epidemic and non-epidemic periods, 10–19 year olds have been the most affected age group for several decades.
In the present investigation, we describe a dengue epidemic that occurred in 2010, including differences in the epidemiology of cases infected with different DENV-types with respect to primary versus secondary infection.
A retrospective analysis of suspected dengue cases reported to surveillance systems was performed to: 1) describe the epidemiology of the 2010 dengue epidemic, including disease severity; 2) determine the proportion of primary and secondary DENV infections, and the molecular epidemiology of the DENVs responsible for the epidemic; and 3) describe relationships between demographic variables (e.g. age, sex, municipality of residence) and characteristics of illness (e.g. infecting DENV-type, severity of illness). This investigation underwent institutional review at CDC and was determined to be public health practice and not research, including the post-hoc determinations of DENV molecular epidemiology and primary/secondary infection rates in reported cases; as such, Institutional Review Board approval was not required.
Surveillance data from five sources were used to identify cases. First, Centers for Disease Control and Prevention Dengue Branch (CDC-DB) and Puerto Rico Department of Health (PRDH) jointly operate the island-wide Passive Dengue Surveillance System (PDSS) that requires an acute serum specimen and completion of a Dengue Case Investigation Form (DCIF) (cdc.gov/dengue/resources/dengueCaseReports/DCIF_English.pdf) for case reporting and diagnostic testing. Second, the Enhanced Dengue Surveillance System (EDSS) operates solely in the municipalities of Patillas and Guayama and utilizes an on-site nurse epidemiologist to encourage case reporting and patient follow-up to obtain a convalescent serum specimen [21]. Third, identification of fatal dengue cases is conducted via PDSS and EDSS [22], and enhanced fatal case surveillance was initiated in January 2010 in collaboration with the Instituto de Ciencias Forenses de Puerto Rico, which obtains blood and tissue specimens at autopsy from suspected dengue-related deaths. Fourth, PRDH operates the Notifiable Diseases Surveillance System (NDSS) wherein suspected dengue cases are reported without diagnostic testing at CDC-DB. Last, in addition to dengue diagnostic testing performed at CDC-DB for PDSS and EDSS, testing is performed by two private diagnostic laboratories outside of Puerto Rico according to their internal protocols [23]. Diagnostic test results from these laboratories and patient data, including sex, age, and date of illness onset (if unavailable, specimen collection date was used instead), were entered into an independent database. Deduplication of individuals reported to more than one data source was achieved by matching records on name and date of birth and consolidation into a single case if two or more reports from any data source had symptom onset dates within 14 days of each other. As case-patients' travel history is not well captured via the surveillance systems used in this investigation, reported cases may represent both locally-acquired as well as travel-associated cases.
All diagnostic testing was performed at CDC-DB for serum specimens received through PDSS or EDSS using the following algorithm: acute specimens (collected ≤5 days after symptom onset) were tested by DENV-type-specific real-time reverse-transcriptase-polymerase chain reaction (RT-PCR) [24] adapted for high throughput using MDX-10 Universal and M48 systems (Qiagen, Valencia, CA); acute specimens collected 5 days after symptom onset and all convalescent specimens (collected ≥6 days after symptom onset) were tested for the presence of anti-DENV immunoglobulin M (IgM) antibody with an antibody-capture enzyme-linked immunosorbent assay (MAC ELISA) and a cut-off value of the OD450 of the specimen versus that of the negative control (ie. P/N ratio ) ≥2.0 [25], [26]. All serum specimens from fatal cases were tested by both RT-PCR and MAC ELISA. Tissue specimens were tested at CDC Infectious Diseases Pathology Branch in Atlanta, GA by immunohistochemistry (IHC) [27] and flavivirus-specific RT-PCR [28] followed by sequencing.
A suspected dengue case was a dengue-like illness in a person in Puerto Rico whose health care provider: 1) submitted a DCIF and serum or tissue specimen to CDC-DB for dengue diagnostic testing; 2) submitted a serum specimen to a private laboratory for dengue diagnostic testing; or 3) reported the case via NDSS.
A laboratory-positive case was a suspected dengue case that met the following criteria for current (criteria 1 and 2) or recent (criterion 3) DENV infection: 1) detection of DENV nucleic acid in a serum or tissue specimen; 2) detection of DENV antigen in a tissue specimen; or 3) detection of anti-DENV IgM antibody in a serum specimen.
A laboratory-negative case was a suspected dengue case with: 1) no anti-DENV IgM antibody detected in a convalescent specimen; or 2) no DENV nucleic acid or antigen detected in a fatal case with only a tissue specimen submitted.
A laboratory-indeterminate case was a suspected dengue case with no DENV nucleic acid or anti-DENV IgM antibody detected in an acute specimen with no convalescent specimen available for testing.
Dengue with warning signs and severe dengue were defined according to 2009 WHO clinical guidelines [2]; dengue, DHF and DSS were defined according to 1997 WHO clinical guidelines [3].
A representative sample of all RT-PCR-positive cases reported to PDSS or EDSS with illness onset between January 1 and December 31, 2010 was selected to determine the rates of primary and secondary DENV infection. Cases were stratified by age group with optimal allocation to allow for comparison between age groups, and further allocated to reflect the proportion of DENV-types that occurred during 2010 to allow for comparison between DENV-types and age groups. Sample size was calculated using an estimate of the proportion of secondary infections by age group based on data from the 2007 dengue epidemic [20], an error of 20%, 95% significance, and an expected 20% of specimens having insufficient specimen volume remaining for testing to be completed. Of the 1,000 selected cases, 818 had sufficient specimen volume and were tested at a dilution of 1∶100 for the presence of anti-DENV IgG antibody by ELISA using DENV-1–4 antigen and a cut-off value of OD450≥0.15 [29], [30]. A secondary DENV infection was defined by detection of anti-DENV IgG antibody in an acute specimen, and a primary DENV infection by lack of anti-DENV IgG antibody detection in an acute specimen.
Serum specimens with DENV-1 (n = 7), DENV-2 (n = 2) or DENV-4 (n = 4) detected by RT-PCR were randomly selected from municipalities with the highest incidence of the respective DENV-type and inoculated into cultured C6/36 cells; the presence of virus was confirmed by RT-PCR and indirect immunofluorescence [31]. Isolates were further propagated and viral RNA was extracted from culture supernatants using the M48 BioRobot System (Qiagen; Valencia, CA). The envelope glycoprotein (E) gene was amplified and sequenced; sequence data were restricted to the E gene open reading frame (1,485 basepairs). Multiple sequence alignment was performed using MUSCLE available in MEGA 5 (megasoftware.net) and GTR+Γ+I4 was selected as the best nucleotide substitution model as determined by MODELTEST v3.7. Genetic relatedness was inferred and represented with phylogenetic trees using the maximum likelihood method in MEGA 5. MCMC was run in BEAST v1.6.1 (beast.bio.ed.ac.uk) under Bayesian skyline prior, constructed in TreeAnnotator found in the same BEAST package, and visualized in FigTree v1.3. Both trees rendered almost identical tree topologies, therefore confirming genetic relatedness. Evolutionary distances were corroborated by pairwise alignment in BioEdit v7.1.3 and E gene sequences from GenBank were included in the phylogenetic tree to support tree topology by currently circulating genotype. Tree topology was supported by bootstrapping with 1,000 replicates. Genotypes were referred to by previously described nomenclature [32], [33].
The frequencies of clinical, demographic and laboratory data were calculated by performing descriptive analyses of all suspected dengue cases identified in 2010. Rates of suspected dengue and laboratory-positive cases were calculated using population denominators obtained from the 2010 United States Census [9]. Statistical differences in proportions were tested by applying the Chi-squared test and Fisher's exact test when applicable. Unless otherwise noted, relative risk ratios were used to calculate all differences between effect sizes. All data analyses were conducted using SAS 9.2 (SAS Institute Inc., Cary, NC), graphs were produced in Microsoft Excel (Microsoft Corp., Redmond, WA), and maps were created using ArcView (ESRI, Redlands, CA).
We identified 26,766 suspected dengue cases with illness onset between January 1 and December 31, 2010 (7.2 suspected dengue cases per 1,000 residents). Of these, 22,496 (84.0%) were reported to PDSS, 1,846 (6.9%) were identified though diagnostic testing at a private laboratory, 1,304 (4.9%) were reported to NDSS, and 1,120 (4.2%) were reported to EDSS (Fig. S1). Suspected dengue cases exceeded the PDSS epidemic threshold in the first week of 2010, increased steeply in week 20 (May 14–20), and peaked at 1,157 in week 32 (August 6–12) (Fig. 1). Suspected dengue cases slowly declined thereafter and returned to below the historic average in mid-December.
Of all suspected dengue cases, 25,852 (96.6%) had a specimen tested for evidence of DENV infection, of which 25,246 (97.7%) were tested by CDC-DB and the remainder by a private laboratory; paired specimens were available for 1,996 (7.5%) cases. Of all cases with a specimen tested, 3,664 (14.2%) were laboratory-negative, 10,140 (39.2%) were laboratory-indeterminate, and 12,048 (46.6%) were laboratory-positive (3.2 laboratory-positive cases per 1,000 residents). The median weekly proportion of cases that tested laboratory-positive was 48.3%, and was highest (64.5%) in week 24 (June 11–17) and lowest (11.1%) in week 53 (December 31).
Laboratory-positive case-patients resided in all 78 municipalities of Puerto Rico (Fig. 2A), and the median rate of laboratory-positive cases by municipality was 2.68 per 1,000 residents. Rates were the highest in the municipality of Patillas (16.34 cases per 1,000 residents), the southeastern municipality where the EDSS site is located [21], and lowest in Aibonito (0.12 cases per 1,000 residents) in the mountainous center of Puerto Rico. Of 7,426 RT-PCR-positive cases, DENV-1 was detected in 5,126 (69.0%) and incidence was highest in the southeast (Fig. 2B). DENV-2 was detected in 545 (7.3%) cases primarily in the west (Fig. 2C), whereas DENV-4 was detected in 1,757 (23.7%) cases and incidence was highest in south-central and northwestern Puerto Rico (Fig. 2D). DENV-3 was detected in just two (<0.1%) cases in early 2010.
The age distribution of laboratory-positive cases was significantly different from suspected dengue cases only for case-patients between 30 and 69 years of age (Fisher's exact, p≤0.04). The median age of laboratory-positive case-patients was 18 years (Table 1). The most affected age group was 10–14 year olds (7.8 cases per 1,000 individuals), followed by 15–19 year olds (7.4 cases per 1,000 individuals) (Fig. 3A). Five-to-nine year olds were the next most affected age group followed by individuals <1 year of age (4.6 and 4.1 cases per 1,000 individuals, respectively). Individuals 50–59 years of age were the least affected age group (1.7 cases per 1,000 individuals).
The distribution of RT-PCR-positives cases among age groups was not significantly different from that of laboratory-positive cases (Fisher's exact, p>0.05) except for the 50–59 year-old age group, for which serum specimens were collected later (median: 6 days post-illness onset [DPO]) than all other age groups (median: 4 DPO) (Fisher's exact, p = 0.04) and thus tested less frequently by RT-PCR. Despite this, the distribution of DENV-types was not consistent among age groups (Fig. 3B). The strong majority (89.3%) of RT-PCR-positive cases in individuals 1–4 years of age were due to infection with DENV-1, whereas 8.1% and 2.6% were due to infection with DENV-4 and -2, respectively. The percent of infections due to DENV-1 decreased and those due to DENV-4 increased with age until a plateau of approximately 60% DENV-1, 30% DENV-4 and 10% DENV-2 was reached in the 20–29 year old age group.
From the sample of 818 RT-PCR-positive specimens tested for primary versus secondary DENV infection, 169 (20.7%) were primary and 649 (79.3%) were secondary. The median age of individuals experiencing primary infection was 14 years, compared to 23 years for individuals experiencing secondary infection. Eighty-one percent of individuals 1–4 years of age had primary infection and were the only age group for which primary infection was significantly more common than secondary (p = 0.003) (Figure 3C). More than 89% of infections in all adult age groups (i.e. age ≥20 years) were secondary. The frequency with which anti-DENV IgG antibody was detected in specimens taken from infants was likely due to the presence of maternal antibody [2].
Whereas 28.5% of all DENV-1 infections were primary, significantly fewer DENV-2 (6.8%) and DENV-4 (7.1%) cases were primary infections (p<0.0001) (Table 2). Calculation of relative risk ratios (RR) indicated that individuals infected with DENV-1 were 4.2 and 4.0 times more likely to be experiencing primary infection than were individuals infected with DENV-2 or -4, respectively (Table 2).
Sequencing and phylogenetic analyses of randomly selected DENV isolates showed that DENV-1 belonged to the American-African genotype (genotype V [34]), but to a clade distinct from virus isolated during the 1998 Puerto Rico epidemic (Fig. 4A). Available sequence data suggest that close ascendants of the 2010 DENV-1 clade had been circulating in Puerto Rico and the Caribbean since at least 2006 (Fig. 4A). DENV-2 sequencing indicated that the virus belongs to clade 1B of the American-Asian genotype (genotype IIIb [35]) (Fig. 4B), which is composed of DENV strains endemic to Puerto Rico [36]. DENV-4 belonged to the Indonesian genotype (genotype II [37]), but was distinct from virus isolated in 1998 (Fig. 4C). Viruses closely-related to the DENV-4 isolated in 2010 were first detected in Puerto Rico in 2004 (Fig. 4C).
Of 12,048 laboratory-positive cases, 31.5% had at least one hemorrhagic manifestation and sufficient clinical data was provided to classify 74.0% as dengue and 2.4% as DHF (Table 1). Nearly half (49.7%) of all laboratory-positive cases had dengue with at least one warning sign, and 11.1% had severe dengue. Of 128 suspected dengue deaths, 40 (31.3%) were laboratory-positive cases. While adults represented nearly half of laboratory-positive cases with dengue (47.1%), dengue with warning signs (44.6%), and severe dengue (49.7%), they accounted for nearly all (92.5%) fatal dengue cases. Laboratory-positive severe and fatal dengue occurred at a rate of 0.36 and 0.01 cases per 1,000 residents, respectively; laboratory-positive fatal dengue cases occurred at a rate of 30.0 per 1,000 severe dengue cases. From the sample of cases for which primary and secondary DENV infection status was determined, secondary infection was identified in 102 (87.9%) case-patients with severe dengue and 547 (77.9%) case-patients without severe dengue (RR = 1.2; 95% CI = 1.1–1.2).
Case-patients with DHF were more likely to have been infected with DENV-4 than DENV-1, and those with severe dengue were more likely to have been infected with DENV-4 than DENV-1 or -2 (Table 2). There was no significant difference between infection with DENV-1, -2 or -4 and likelihood of being a fatal case.
In 2010, Puerto Rico experienced the largest and longest dengue epidemic ever documented on the island. In total, more than 12,000 individuals had laboratory-confirmed dengue, of which more than 1,300 experienced severe dengue and 40 died. The most common DENV identified was DENV-1, and 1–4 years old were the only age group more frequently experiencing a primary versus secondary DENV infection. Individuals infected with DENV-1 were four times more likely to have a primary infection than were those infected with DENV-2 or -4. A strength of this investigation was utilization of multiple surveillance systems to identify all reported suspect dengue cases. However, a minor weakness was that data obtained from each system may not be directly comparable due to different diagnostic algorithms used by CDC-DB and private laboratories, and we were not able to determine status of primary versus secondary infection or perform sequencing on specimens from private laboratories. Because private laboratories contributed <5% of all laboratory-positive dengue cases, this likely did not affect the conclusions of this investigation.
The 2010 dengue epidemic was similar in several respects to the 1998 epidemic: both began in January during El Niño events accompanied by above average temperatures, which while not a determinant of epidemics in Puerto Rico [38] may contribute to increased DENV transmission [39]; and both epidemics peaked in week 32 of the calendar year and were predominated by transmission of DENV-1 and -4 [19]. A notable difference was that DENV-3 was essentially absent in 2010, whereas it accounted for ∼6% of cases during the 1998 epidemic [19]. DENV-3 was re-introduced into Puerto Rico in 1998 following a 20-year absence and was the predominant virus-type in the 2007 dengue epidemic [20]. Thus, susceptibility to DENV-3 infection was likely high in 1998 and low in 2010, which likely explains these observations.
The American-African and Indonesian genotypes of DENV-1 and -4 have been circulating in Puerto Rico since introduced in 1978 and 1981, respectively [16], [40]. However, the DENV-1 isolated in 2010 was distinct from the DENV-1 isolated during the 1998 epidemic (Fig. 4A and [41]) and was more closely related to the DENV-1 isolated during the 2007 epidemic (Fig. 4A). Similarly, the DENV-4 isolated during the 2010 epidemic was distinct from the DENV-4 isolated in 1998 and was more closely related to viruses circulating since 2004 (Fig. 4B). These findings suggest that DENV-1 and -4 may have both experienced clade replacements at some point after 1998 but prior to 2007. After the re-introduction of DENV-3 into Puerto Rico in 1998, DENV-1 was not detected between 2001 and 2006 and DENV-4 was not detected between 2000 and 2005 [42]. Nonetheless, apparent re-introductions of DENV-1 in 2007 and DENV-4 in 2006 were soon followed by the disappearance of DENV-3 in 2010 (this paper and [42]). In place of the convenience sample used in this investigation to describe the DENVs responsible for the epidemic, sequencing of a representative sample of specimens and longitudinal sequence analysis will be necessary to both confirm apparent clade replacements and determine if other DENV clades contributed to the 2010 epidemic.
Similar to previous epidemics in Puerto Rico (Table S1), 10–19 year olds were most affected during the 2010 epidemic; however, unlike previous epidemics, 5–9 year olds were the next most affected age group. The median age of individuals experiencing secondary DENV infection declined from 27 years in 2007 [20] to 23 years in 2010, likely due to the relative proximity of the periods of high infection pressure. Taken together, these observations indicate an increase in incidence of dengue and a decrease in the age of secondary infection, suggesting that the overall force of DENV transmission may have been higher in 2010 than in previous epidemic years.
The observation that DENV-2 and -4 cause relatively infrequent clinical apparent illness upon primary DENV infection is consistent with previous studies [43]–[48]. Similarly, our finding that DENV-1 was a more frequent cause of clinically apparent illness upon primary infection has also been previously reported [43], [49], including the observation of increased disease severity during primary infection with DENV-1 compared to other DENV-types [44], [50], [51]. Nonetheless, of 545 DENV-2 and 1,755 DENV-4 infections, roughly 7% were primary, indicating that primary infection with these DENVs can cause clinically apparent illness, contrary to previous assertions [46], [47]. The relative abundance of DENV-1 compared to DENV-2 and -4 is unlikely to be responsible for the observed differences in likelihood of causing clinically apparent illness upon primary infection, as relative risk ratios compare the proportion of exposed individuals experiencing the outcome of interest. This is supported by the findings in the 1–4 year-old age group, of which ∼80% experienced a primary infection with DENV-1. Alternative explanations for these observations include potential variations in the sensitivity of detection of DENV-type-specific anti-DENV IgG antibody and differences in force of infection between the DENV-types circulating in 2010.
We also saw that DENV-1 and -2 were less frequently a cause of severe dengue than DENV-4. This is in contrast to previous studies where DENV-1 was a more frequent cause of DHF than DENV-4 [52], and a study where DENV-2 was twice as likely to result in DHF as DENV-4 [43]. Possible explanations for these differences include: the comparatively small number of DENV-4 infections observed in previous studies; differences in clade and/or viral fitness leading to differential pathogenicity [33], [53], [54]; and/or the DENV-type(s) and sequence to which individuals were previously exposed, which may affect the likelihood of developing severe dengue [44], [55], [56].
This investigation had several limitations. First, because individuals experiencing secondary infection may have a diminished anti-DENV IgM antibody response [57], suspected dengue cases tested solely for anti-DENV IgM antibody may have been misclassified. Second, although DENV is the sole flavivirus known to cause clinically apparent illness in humans in Puerto Rico (CDC, unpublished data), some proportion of anti-DENV IgM or IgG positive results could have been due to infection with or vaccination against another flavivirus [58], resulting in misclassification. Third, because clinical data was provided for >90% of case-patients on only one occasion and some data variables were incompletely reported (e.g. only 56% of suspected cases had a reported status of hospitalization), severity of disease and the rates of dengue with warning signs and severe dengue reported here were likely underestimated. Finally, the description of the epidemiology and molecular characteristics of dengue reported here is only representative of reported, clinically apparent DENV infections and may not be reflective of asymptomatic and sub-clinical DENV infections.
The 2010 dengue epidemic in Puerto Rico demonstrated that dengue continues to be a public health concern for Puerto Rico residents and visitors, and surveillance systems and control initiatives should continue to be supported and strengthened. This epidemic also highlights the need for effective primary prevention tools such as a dengue vaccine to reduce disease morbidity and mortality.
|
10.1371/journal.pbio.2006025 | Import volumes and biosecurity interventions shape the arrival rate of fungal pathogens | Global trade and the movement of people accelerate biological invasions by spreading species worldwide. Biosecurity measures seek to allow trade and passenger movements while preventing incursions that could lead to the establishment of unwanted pests, pathogens, and weeds. However, few data exist to evaluate whether changes in trade volumes, passenger arrivals, and biosecurity measures have altered rates of establishment of nonnative species over time. This is particularly true for pathogens, which pose significant risks to animal and plant health and are consequently a major focus of biosecurity efforts but are difficult to detect. Here, we use a database of all known plant pathogen associations recorded in New Zealand to estimate the rate at which new fungal pathogens arrived and established on 131 economically important plant species over the last 133 years. We show that the annual arrival rate of new fungal pathogens increased from 1880 to about 1980 in parallel with increasing import trade volume but subsequently stabilised despite continued rapid growth in import trade and recent rapid increases in international passenger arrivals. Nevertheless, while pathogen arrival rates for crop and pasture species have declined in recent decades, arrival rates have increased for forestry and fruit tree species. These contrasting trends between production sectors reflect differences in biosecurity effort and suggest that targeted biosecurity can slow pathogen arrival and establishment despite increasing trade and international movement of people.
| When people and goods move around the world, they spread nonnative species—including pathogens that can cause disease—leading to huge economic impacts. Many countries try to limit pathogen arrivals by screening goods and people before they enter. But are these biosecurity measures effective? Pathogens are hard to detect, and we rarely have data on key metrics such as the volume of goods imported, number of people arriving, and new nonnative pathogens establishing over time. Our study uses a database of all known New Zealand plant pathogen records to estimate how many fungal pathogens arrived and established on 131 economically important plant species each year over the last 133 years. Pathogen arrivals increased exponentially for 100 years starting in 1880, paralleling an increasing volume of goods imported. Since about 1980, the rate of new pathogen arrivals has stopped increasing, despite imports and the arrival of people continuing to accelerate. However, these recent trends differ among plants from different economic sectors. Pathogen arrivals on crop and forage plants have declined but continue to increase on forestry and fruit trees. This trend reflects differences in the biosecurity measures imposed, suggesting that targeted biosecurity can reduce the establishment of nonnative pathogens even while global trade and travel continue to increase.
| International movements of goods and people are major pathways for transporting species to new regions and can result in harmful biological invasions [1,2]. Over the last half century, international trade and travel have risen dramatically [3,4] in parallel with large increases in the arrival and establishment of nonnative species [5–9]. Worldwide, the number of nonnative species in different regions of the world correlates with the magnitude of trade imports in a range of taxa [10–12], and within regions, trade measures are closely linked to new species’ arrival and establishment rates [10,13]. International travellers also transport nonnative species, including plants, pathogens, and invertebrates, some of which establish as biological invaders [14–16]. Forecasts predict continued increases in international trade and travel and more links among countries [17,18]. Based on historical patterns, these increases have the potential to accelerate the arrival and establishment of nonnative species in new regions [19,20], with consequent economic and ecological impacts.
To counter the threat posed by the arrival of unwanted species through trade and transport pathways, many developed countries have invested heavily in border biosecurity surveillance [21], phytosanitary inspection, and quarantine. Biosecurity measures are designed to prevent unwanted or unknown species entering trade or transport pathways, to detect species arriving in trade shipments or with passengers, and to prevent the release of species into the wild [22–26]. Effective biosecurity is particularly important to countries that rely heavily on primary production, because new pests and diseases that threaten plant or animal health can have major economic consequences [7,23]. Developed countries have invested more in biosecurity than less developed nations [27–30], but even developed countries have difficulty assessing the value of their biosecurity investment because costs are often spread across multiple agencies, and the benefits of such interventions are often unclear [31]. New Zealand is a major exporter of primary produce and one of the few countries that provide a nationwide accounting of biosecurity investment, spending more than US$137 million in 2014 [32], slightly more than 0.3% of its gross domestic product (GDP). Justifying this substantial expense requires that biosecurity measures cost less than the economic and ecological costs of the pest, pathogen, and weed incursions that are prevented by such interventions [33,34].
Pathogens are key biosecurity targets because they can readily enter transport pathways and pose significant threats to animal and plant health [7,11,35]. Plant fungal pathogens are responsible for crop yield losses that cost individual economies billions of dollars annually [7,36,37], with impacts across a wide range of production sectors, including agriculture, forestry, horticulture, and livestock [7,38,39]. Despite substantial investment by countries in biosecurity and global initiatives to coordinate these efforts [40,41], it has proven difficult to evaluate the effectiveness of biosecurity measures. Quantifying how pathogen arrival and establishment rates have changed over time is especially problematic because pathogens are difficult to detect in the early stages of invasion. Extensive host surveys are often necessary for initial detection and require expert pathologists to isolate and identify pathogens in symptomatic hosts. These difficulties make detecting new pathogens particularly sensitive to variation in survey effort [42,43].
Here, we use a long-term, comprehensive database of all known associations between nonnative plants and fungal pathogens in New Zealand [44] to examine trends in fungal pathogen arrival rates over time in relation to changing trade and transport patterns while accounting for variation in sampling effort. We use these data to evaluate whether biosecurity investment has been effective in reducing pathogen establishment. We focus on nonnative host plants in 4 primary production sectors that are major targets for biosecurity in New Zealand: crops (46 species, including wheat, tomatoes, and onions), fruit trees (30 species, including grapes, apples, and kiwifruit), commercial forestry (42 species, including pines and eucalypts), and pastures (13 species of forage grasses and legumes). Because these are the major primary production sectors, plant species in these groups comprise the majority of host–pathogen records for well-sampled species in New Zealand. We use these data to estimate how annual rates of pathogen arrival have changed over time while accounting for variation in survey effort and then to address three questions: (1) Is the overall rate of nonnative fungal pathogen establishment in New Zealand more strongly linked to changes in import trade volume or passenger arrivals? (2) Do changes in pathogen arrival rates differ among the primary production sectors, and are changes related to variation in sector-specific imports? (3) Do changes in pathogen arrival rates over time coincide with the implementation of specific biosecurity measures?
The data comprised 6,691 host–pathogen records from 131 nonnative host plant species spanning the years 1881–2012. We restricted our analysis to the 466 pathogen species whose first New Zealand record was on one of the 131 focal host plants, which identified these hosts as the source of the new pathogen arrivals. Time series plots revealed substantial variation in the number of host–pathogen records per year (Fig 1), indicating substantial variation in annual survey effort, which we accounted for in our analyses (Materials and methods). The estimated annual rate at which new fungal pathogens arrived and established on the focal host plants increased from the 1880s until about 1980, after which the annual arrival rate slowed, albeit with wide uncertainty around recent arrival rates (Fig 2). Since 2000, we estimate an average of 5.9 new species of fungal pathogens per year have established on the focal host plant species.
In contrast to the slowdown in pathogen arrivals, both import trade volume and passenger arrivals to New Zealand have increased dramatically in recent decades, with import volume starting to accelerate in the 1940s and international passenger arrivals in the 1970s (Fig 3A and 3B). To directly compare changes in trade volume and number of passengers with changes in pathogen arrival rates, we plotted the mean values for these variables in each year (Fig 3C and 3D). These plots indicate that pathogen arrival rates were more strongly linked to import volume than to passenger arrivals, with pathogen arrival increasing in concert with increasing import volume until about 1980, when import volume declined briefly and then increased rapidly while pathogen arrival rates slowed (Fig 3C). In contrast, passenger arrivals changed little between 1920 and 1980, during which time pathogen arrival increased, while the substantial and rapid rise in passenger arrivals since about 1980 coincides with slowing of pathogen arrival rates (Fig 3D). These trends suggest that pathogen arrival into New Zealand was most strongly linked historically to increasing import trade volumes, but this relationship has weakened significantly since about 1980.
Trends in overall pathogen arrival rates, however, obscure substantial variation among the four production sectors. Pathogen arrival rates have declined in recent decades for both pasture and crop species, with declines beginning around the 1970s for crops and slightly earlier for pasture species (Fig 4A and 4D). In contrast, pathogen arrival rates have continued to accelerate for forestry and fruit tree species, especially in recent decades (Fig 4B and 4C). These trends were not consistently associated with changes in sector-specific import volumes since 1960, the period for which sector-specific trade data were available (Fig 4E and 4H; see Materials and methods). Pathogen arrival rates on crop and pasture species have declined since 1960, while import volumes have increased in these sectors. Pathogen arrival rates on forestry species have increased despite declining import volume, while pathogen arrival rates on fruit trees have been relatively steady or increased slightly while trade volume has risen steadily (Fig 4). Consequently, the pathogen arrival rate per host species per million tonnes of import trade has declined for pasture and crop species since about 1980 (Fig 5A and 5D). In contrast, pathogen arrival rate per host species per million tonnes of import trade has remained steady for fruit trees but has increased markedly for forestry species since 1960 (Fig 5B and 5C). Thus, the recent stabilisation in overall pathogen arrival rates (Fig 1) is the sum of contrasting trends among the different production sectors.
Two processes could explain the recent declines in pathogen arrival rates for crop and pasture species despite continued increases in import trade. First, as fungal pathogens arrive, there will be progressively fewer pathogens remaining elsewhere to be introduced [13,20], and pathogens not yet introduced are more likely to be those with a lower probability of transport or establishment [9]. This would imply the pool of readily transported and highly invasive fungal pathogens associated with crop and pasture species is being exhausted, leading to a decline in arrival and establishment rates. While this is a possibility, a recent review found that only about one-third of global pest and pathogen species associated with crops grown in New Zealand were currently present in the country [45]. Moreover, of the ten host plant species with the most fungal pathogens in our data, seven still have fewer than 40% of the fungal pathogens recorded for these species globally [46, See S1 Table]. This implies that a substantial fraction of pathogens have yet to arrive in New Zealand and that saturation is unlikely to explain declining rates of pathogen arrival. This is consistent with models using the distribution of agricultural pathogens coupled with trade patterns to evaluate the risk of new pathogen arrivals, which indicate New Zealand has a moderate to high risk of future plant pathogen invasions [30].
A second explanation for the decline in pathogen arrival rates is increased biosecurity. New Zealand has a long history of plant biosecurity [47]. Yet, consolidated data on government biosecurity spending are only available beginning in the 1990s, and our results show declines in pathogen arrival rates for crop and pasture species commenced much earlier, in the 1960–1970s (Fig 4). Historical developments in plant biosecurity in New Zealand, however, are consistent with the timing of declines in pathogen arrival rates for crop and pasture species. Agricultural biosecurity ramped up in the 1950s, marked by the establishment of the Plant Quarantine Service in 1952 soon after New Zealand signed the International Plant Protection Convention. This is evident in our data, with a substantial increase in pathogen survey effort commencing in the 1950s (Fig 1). A more unified border protection service with a strong legal mandate emerged in 1962 as the Port Agriculture Inspection Service, which evolved further to manage cargo, air, and passenger pathways as the Agriculture Quarantine Service in 1981 [47]. This increase in capacity and effort in agricultural border biosecurity coincides with a weakening relationship between trade and pathogen arrival rates and suggests that biosecurity efforts played a role in limiting new pathogen arrivals.
Biosecurity initiatives targeting pathways specific to pastures and crops are also consistent with the timing of declines in pathogen arrival rates in these sectors. Most pasture and crop species are imported as seed. Voluntary, industry-backed seed certification for agricultural species began as early as the 1920s [48]. However, New Zealand’s entry into the Organisation for Economic Co-operation and Development’s (OECD) seed certification scheme in 1967 likely led to significant improvements in the management of seed-borne diseases, particularly those from overseas [48]. The combination of government investment in agricultural quarantine coupled with an industry-based seed certification scheme targeted key pathways by which pathogens of crop and pasture species entered the country, which could explain the decline in pathogen arrival rates in these sectors from the 1960s onward (Figs 4 and 5).
In contrast, the forestry and fruit tree sectors do not appear to have placed as much emphasis on preborder biosecurity, which could account for the ongoing increase in pathogen arrival in these sectors (Figs 3 and 5). The New Zealand seed certification scheme did not include horticultural species [48], and although phytosanitary inspections of timber imports began in 1949, they focussed primarily on invertebrate pests [47], while broader forestry biosecurity efforts focussed on treating existing tree diseases rather than preventing new arrivals [49,50]. Our data on pathogen survey efforts reinforce these differences among sectors (Table 1): Relative to pasture and crop species, fruit tree and forestry species had, on average, fewer records per species, individual species surveys began later, and peak survey effort occurred several decades later (1960s for pasture and crop species; 1980 and 2000 for fruit tree and forestry species, respectively).
Pathogens of forestry and fruit tree species have additional potential vectors, including soil and live plant material (e.g., rootstock) and untreated wood products (e.g., wood pallets), that may facilitate further pathogen arrival [51,52]. Postentry quarantine of live plant material, implemented in the 1990s [53], should have slowed arrival rates via this pathway, but no corresponding decrease is evident in pathogen arrival rates (Fig 4B and 4C). This may be because wood packaging, which is used extensively in transporting goods, is potentially a significant pathogen source, and wood packaging volume is likely to have increased in concert with rapidly increasing import trade volumes, potentially contributing to the continued rise in pathogen arrivals for woody species [51,52,54]. International phytosanitary standards for wood products are relatively recent (2002) and are not used for all transport methods, and even treated wood packaging material can still harbour pathogens [54]. Since the relevant trade standard (ISPM-15 dealing with the treatment of wood packaging in international trade; [55]) was only recently revised to include more stringent treatment guidelines, it is likely too early to assess whether this might reduce pathogen arrival rates for woody species.
International travellers can be vectors for nonnative species, and New Zealand has invested heavily in preventing incursions via this pathway using, for example, soft-tissue X-ray machines and detector dogs at international airports since 1996 [56]. Prior to that time, the Ministry for Agriculture and Forestry estimated that it was detecting only 55% of risk goods brought in by passengers, with detection levels rising to 95%–100% after 2001 [56]. These initiatives, however, occurred at least a decade after the observed decline in pathogen arrival rates for crop and pasture species, suggesting that for plant pathogens, other measures were responsible for slowing arrivals.
Postborder pathogen survey efforts to detect new incursions have declined since about 2000 despite the increase in pathogen arrival rates for forestry and fruit tree species (Fig 1). This decline makes it more difficult to evaluate trends in arrival rates, as revealed by the wide uncertainty intervals associated with arrival rate estimates in recent years (Figs 2 and 4). Moreover, there is a time lag between the arrival of new pathogens and their discovery, the length of which will depend on survey effort. We statistically controlled for this in our analysis by explicitly modelling the processes of pathogen arrival and discovery (Materials and methods). This provided an estimate of the number of pathogen species that had arrived and established on the focal host plants but had not yet been detected. In addition to the 466 known pathogen species, we estimated a further 55 species (95% credible interval 30–85) were present but undetected, highlighting the need for ongoing postborder surveillance to detect new incursions. We cannot ascribe these undetected species to a specific introduction period or sector, but our results indicate that about 90% of pathogens have been detected, meaning our overall findings should be robust.
In conclusion, we provide the first detailed analysis of plant pathogen arrival rates through time, accounting for variation in survey effort in a country that invests heavily in border biosecurity. Our analysis revealed that for the first half of the 20th century, the rate at which plant pathogens arrived and established on economically important plant species in New Zealand increased in concert with increasing import trade volume but was not linked to passenger arrivals. For crop and pasture species, pathogen arrival rates started diverging from imports around the 1960–1970s, coinciding with a greater biosecurity effort designed to limit pest and pathogen arrivals in the agricultural sector. Biosecurity measures appear to have been less effective in preventing pest and pathogen arrivals in the forestry and fruit tree sectors until recently, which may explain why pathogen arrival rates for woody species have continued to increase in recent decades. Our findings provide the first evidence, to our knowledge, that targeted investment in biosecurity may be effective in reducing pathogen arrival, despite increasing trade, and limiting the establishment of microorganisms but highlight the importance of sustained surveillance due to the significant risk, posed by increasing levels of trade, for unwanted introductions in the absence of effective biosecurity measures.
We compiled a database of observed host–fungal (senso lato) associations in New Zealand recorded between 1847 and 2012. Each record comprised an observation of a fungus and its associated host plant and the year of observation. The data are stored in the NZFungi2 database (Landcare Research; http://nzfungi2.landcareresearch.co.nz/; [44]) and comprise essentially all known host–fungal records from New Zealand. The New Zealand economy’s historical reliance on primary production has meant there have been repeated systematic surveys of the diseases associated with agriculture, horticulture, and forestry [57–59] carried out by government agencies tasked with the diagnosis and surveillance of plant diseases [60]. Consequently, while the database includes native host plants, most records are of fungal taxa associated with introduced, economically important hosts. For well-surveyed host plants, there are typically multiple records of a given host–fungal association (mean of 4.9 records per association for hosts with more than 50 records), reflecting observations at different times in different parts of the country as part of surveillance efforts.
We standardised fungal and plant taxonomic names and removed duplicate entries and invalid names from the database [61]. We pooled all records at the species level and excluded border intercepts and hybrids, with the exception of well-sampled commercial hybrid plants (i.e., Fragaria × ananassa, Cupressus × leylandii, Malus × domestica). Database processing was performed in R [62].
We filtered the database to include only introduced host plants associated with four primary production sectors—crops (46 species), fruit trees (30 species), commercial forestry (42 species), and pastures (13 species sensu [63])—and included only species with at least 10 records (See S2 Table). We included all nonnative pathogens (including fungi, oomycetes, and plasmodiophorids) for which the first record for the pathogen in New Zealand was on one of the selected host species. Pathogen status was determined by expert opinion (A. Stewart, P. Johnston), and the nonnative status of pathogens in New Zealand was drawn from NZFungi2 [44,61].
The records of host–pathogen associations in our data allow us to identify the year in which a particular pathogen was first discovered on an introduced host plant in New Zealand and thus to document the rate at which pathogens accumulated on host plants over time. The observed rate of pathogen accumulation, however, results from both an arrival and discovery process [42,43], and we developed a statistical model to separate these two processes, drawing on the approach described by Belmaker and colleagues in 2009 [64]. To ensure we had adequate sample sizes to quantify changes in pathogen arrival rates over time, we pooled host plant species and examined the accumulation of new pathogens across all species and across species in each of the four production sectors (crop, forestry, fruit tree, and pasture). Because our interest was in the effectiveness of border biosecurity measures, we did not examine the spread of pathogens from one host to another postborder.
For each group of host plant species in each year t, we have data on the number of new pathogen species discovered on those hosts, Dt. We would like to know the number of pathogens that actually arrived and colonised those hosts in each year, At, and the mean arrival rate, μt, but we cannot observe these outcomes directly, and we have to estimate them from data on the number of discoveries in each year, Dt, and sampling effort, measured here as the total number of host–pathogen records recorded in each year on a host plant group, Nt (see Fig 1).
To do this, we assume that for each host–pathogen record in a given year, there is a probability, pt, that the record is the discovery of a new pathogen species. If there are Nt host–pathogen records in year t, then the number of new pathogens discovered in that year can be modelled as a draw from a binomial distribution with probability pt:
Dt~Binomial(pt,Nt)
The actual number of pathogen species arriving and colonising a group of host species in a given year, At, is unknown, and we model it as a random variable drawn from a negative binomial distribution with mean arrival rate μt and dispersion parameter r. We specified a negative binomial distribution to allow for the possibility that the number of pathogens arriving in each year might exhibit greater variation than would be expected under a Poisson distribution, which is a common distribution for modelling number of events per unit time:
At~NegativeBinomial(μt,r)
The first year (t = 1) was set to the first year a host–pathogen association was recorded for a given group of host plants. Many host plant species, however, would have been present in New Zealand prior to the first record in the database and may have had pathogens with them when they arrived and been accumulating new pathogens since arrival. To allow for this, we included a term A0 to represent the number of pathogens already present on host species when the first host–pathogen association was recorded in the database. The number of pathogen species available to be discovered in year t is then equal to the total number of pathogen species that had arrived by the end of that year (A0+∑t=1tAt) minus the total number of pathogen species that had been discovered at the start of that year (∑t=1t−1Dt). We can estimate the probability, pt, that a host–pathogen observation in year t was a newly discovered pathogen as the number of undiscovered pathogen species in year t divided by the total number of pathogen species that have arrived:
pt=(A0+∑t=1tAt−∑t=1t−1DtA0+∑t=1tAt)
Finally, we modelled the mean rate of pathogen arrival, μt, as a function of time t. We fitted a semiparametric regression model using penalised splines to allow the shape of the curve describing arrival rate through time to be determined by the data. We followed the method from Crainceanu and colleagues in 2005 [65] and fitted a low-rank thin-plate spline of the form:
μt=β0+β1t+∑k=1Kbk|t−κk|3,
where t is time; β0, β1, and bk are regression coefficients; and κ1 < κ2 …< κK are fixed knots that determine the flexibility of the spline. We chose K = 5 knots and specified the location of each knot at the quantiles of t corresponding to probability k / (K + 1), thus ensuring the knots were evenly spaced over time. The above spline can be expressed in the form of a linear mixed-effects model and can therefore be fitted using software for mixed or hierarchical models [65].
We fitted the above model to the data in a Bayesian framework using Markov Chain Monte Carlo (MCMC) simulation implemented in JAGS called from R through the jagsUI package [66]. For A0, we specified a uniform prior in the range 0 to the total number of pathogens recorded; for r, the dispersion parameter of the negative binomial distribution, we specified a uniform prior in the range 0–500; for the parameters β0 and β1, we specified a flat normal prior with mean 0 and standard deviation 10,000. To avoid overfitting, we penalised the parameters bk by modelling these as drawn from a normal distribution with mean 0 and variance estimated from the data [65]. We ran our models with 3 chains using the function autojags to ensure the chains converged. We first ran the chains for 10,000 iterations with a 5,000-iteration burn-in. At the end of this run, the autojags function assessed the chains for convergence, defined as the Gelman-Rubin statistic being less than 1.1 for all sampled parameters [67]. If the Gelman-Rubin statistic was greater than 1.1, a further 5,000 iterations were run, and this was repeated until the chains had converged.
Data on overall trade volume came from the tonnage of international cargo unloaded at New Zealand ports, obtained from combining Overseas Trade Statistics data 1923–1988 [68] with more recent data from StatsNZ Infoshare (http://www.stats.govt.nz/infoshare), in which data were only available from 1989–2017. We also evaluated data on the value of trade imports to New Zealand for the period 1914–2011, which were available from StatsNZ Infoshare as well. Raw data on import values were available for the period 1841–2011, but we used the Consumer Price Index (CPI) to inflation adjust these to NZ$ in 2012, and CPI data were only available from 1914 onward (StatsNZ Infoshare: http://www.stats.govt.nz/infoshare). Cargo tonnage and the value of trade imports were highly correlated (r2 = 0.92; S1 Fig). We used volume as it is more likely to reflect the ‘size’ of the potential pathogen pathway by measuring the quantity rather than value of material entering the country. We used international passenger arrival count data from 1900 to present (includes both international visitors and returning residents), also available through StatsNZ Infoshare. For import trade and passenger arrivals, we used a loess smoothing function to capture the trend in arrival rate over time.
Sector-specific trade data for the period 1960–2012 were obtained from the UN Food and Agriculture Organization (FAO). The FAOSTAT dataset (http://www.fao.org/faostat/en/#data/TP) catalogues imports of plant-based commodities in different categories. We selected those categories likely to be potential sources of plant pathogens (See S3 Table): Seeds, dry or fresh, hulled or unhulled, were included as long as they have not been roasted or milled, and only fresh, unprocessed fruit was included, with all dry or preserved fruit excluded. In addition, we determined the plant families associated with each plant commodity category (based on the species included in the commodity category) and excluded any commodity categories where the plant family was not present in our host plant data, since taxonomic affiliation of host plants is a key indicator of susceptibility and spread of pathogens among plants [61]. Import volumes for pasture products were very low, but several crops are in the same family as most pasture species (grasses and legumes). As such, pasture imports were coupled with these selected crop species as a measure of potential sources of imported pathogens for this sector. Forestry import volume data were obtained from the FAO forestry site (http://www.fao.org/faostat/en/#data/FO). Commodity item classes for forestry encompass species with broad taxonomic affiliations, so we excluded only nonconiferous tropical wood products, as there are no economically important nonconiferous tropical species grown in New Zealand, and this category made up less than 1% of forestry imports.
|
10.1371/journal.pntd.0004715 | Characterization of Neutrophil Function in Human Cutaneous Leishmaniasis Caused by Leishmania braziliensis | Infection with different Leishmania spp. protozoa can lead to a variety of clinical syndromes associated in many cases with inflammatory responses in the skin. Although macrophages harbor the majority of parasites throughout chronic infection, neutrophils are the first inflammatory cells to migrate to the site of infection. Whether neutrophils promote parasite clearance or exacerbate disease in murine models varies depending on the susceptible or resistant status of the host. Based on the hypothesis that neutrophils contribute to a systemic inflammatory state in humans with symptomatic L. braziliensis infection, we evaluated the phenotype of neutrophils from patients with cutaneous leishmaniasis (CL) during the course of L. braziliensis infection. After in vitro infection with L. braziliensis, CL patient neutrophils produced more reactive oxygen species (ROS) and higher levels of CXCL8 and CXCL9, chemokines associated with recruitment of neutrophils and Th1-type cells, than neutrophils from control healthy subjects (HS). Despite this, CL patient and HS neutrophils were equally capable of phagocytosis of L. braziliensis. There was no difference between the degree of activation of neutrophils from CL versus healthy subjects, assessed by CD66b and CD62L expression using flow cytometry. Of interest, these studies revealed that both parasite-infected and bystander neutrophils became activated during incubation with L. braziliensis. The enhanced ROS and chemokine production in neutrophils from CL patients reverted to baseline after treatment of disease. These data suggest that the circulating neutrophils during CL are not necessarily more microbicidal, but they have a more pro-inflammatory profile after parasite restimulation than neutrophils from healthy subjects.
| Leishmania spp. are protozoan parasites that cause a spectrum of human diseases, and L. braziliensis causes chronic inflammatory skin lesions in residents of endemic regions of Latin America. Leishmania are obligate intracellular parasites in mammalian hosts, found in macrophages throughout infection. Nonetheless, other cell types including neutrophils also take up the parasite, but the role of neutrophils throughout chronic leishmaniasis remains unclear. We analyzed circulating neutrophils from patients in northeast Brazil with cutaneous leishmaniasis (CL) caused by L. braziliensis, compared to healthy controls from the same region. Our data revealed that neutrophils from both infected and healthy hosts took up comparable numbers of parasites, and parasite phagocytosis induced similar degrees of neutrophil activation. However, CL patient neutrophils produced more reactive oxidants than control neutrophils, and increased amounts of the chemokines CXCL8 and CXCL9 after parasite exposure. Interestingly, according to surface markers of PMN activation (CD62L, CD66b), we found that L. braziliensis activates both infected and uninfected “bystander” neutrophils from both patients and controls. Importantly, repeated measures showed the production of reactive oxidants and chemokine release were significantly decreased after therapeutic cure of infection. These data suggest that CL promotes a heightened inflammatory state in circulating neutrophils during active infection.
| Cutaneous leishmaniasis (CL) is the most common form of human leishmaniasis, a group of diseases caused by the Leishmania spp. protozoa. CL is widely distributed, but Brazil is among the countries with the highest estimated disease prevalence [1]. Within Latin America, Leishmania braziliensis is the most common cause of CL and other related tegumentary forms of leishmaniasis, including mucosal and disseminated leishmaniasis. Patients with CL due L. braziliensis develop a strong Th1-type adaptive immune response with high levels of IFN-γ and TNF-α produced primarily by CD4+ T cells [2–4]. These responses facilitate the control of parasite proliferation within by macrophages, but they also contribute to the pathologic changes that characterize disease [4–6].
The Leishmania spp. are obligate intracellular parasites in their mammalian hosts, and most are found within macrophages of infected tissues. In addition to macrophages, other cell types such as dendritic cells and neutrophils participate in the pathogenesis of Leishmania infection. Studies of mouse models show that neutrophils migrate to the site of infection soon after the sand fly bite, and are the first infiltrating cells to encounter L. major [7–9]. Migration of neutrophils to sites of infection is mediated by the interactions between endothelial cells and adhesion molecules expressed on neutrophil surfaces, which allow for binding and “rolling” prior to extravasation from vasculature [10]. Neutrophils may influence adaptive immune responses by producing chemokines, which recruit others cell types that in turn participate in the response to infection [11,12]. A partial list of neutrophil microbicidal responses includes assembly of the multi-protein NADPH oxidase complex with resultant production of reactive oxygen species, release of granule contents into intracellular microbial compartments, and release of defensins [13,14].
The role of neutrophils in Leishmania spp. infection has been predominantly studied in murine models, and findings have varied depending on both the species of Leishmania used and the resistance or susceptible genetic background of the mouse [15–17]. Confusing the picture, some methods for depletion of neutrophils in mice also deplete other critical cell subsets (e.g., dendritic cells, monocytes and macrophages), depending on the choice and dose of depleting antibody [18,19]. There is evidence that a subset of L. donovani survive intracellularly in murine neutrophils, raising the question whether neutrophils represent a “safe haven” facilitating parasite survival prior to delivery to its permanent host cell, the macrophage [20]. Neutrophils are also found to kill parasites, documented in experimental model of L. braziliensis infection, in which infection trigger neutrophil activation, increased ROS production and this leads to parasite clearance [21,22].
Studies of human neutrophils suggest there could also be an important role for these cells in human leishmaniasis. Neutrophils from healthy donors infected with L. major produce a strong oxidative response that eliminates internalized parasites [23]. Infection with L. amazonensis promotes neutrophils activation, degranulation and production of leukotriene B4 which promotes parasite killing [24]. Additionally, interactions between healthy human neutrophils and Leishmania-infected macrophages modulate the intracellular replication of both L. amazonensis [25] and L. braziliensis [26]. Based upon the hypothesis that neutrophils contribute to the inflammatory environment observed during infection, the current study was initiated to evaluate the phenotype of neutrophils from patients with CL due to L. braziliensis infection. Our data showed that neutrophils from CL patients and from healthy controls display both distinct and common characteristics. The phenotype of circulating neutrophils in CL subjects suggested that they behaved more like primed neutrophils, poised for rapid activation, in contrast to resting neutrophils from healthy subjects [27,28].
All samples were obtained specifically for this study. The study was approved by Institutional Review Boards (IRBs) of the Federal University of Bahia (Ethical Committee), the University of Iowa and the NIH. Written informed consent was obtained from all participants. The UFBA IRB is registered with the NIH.
Corte de Pedra is a village belonging to the municipality of Presidente Tancredo Neves, located in the southern region of the state of Bahia, Brazil. This endemic area is the most prevalent area for L. braziliensis transmission in Brazil, and more than a thousand cases of CL are seen in the Health Post in Corte de Pedra annually [29].
Participants in this study included 21 CL patients diagnosed at the Corte de Pedra Health Post. Diagnosis was based on the presence of clinical manifestations characteristic of CL, confirmed with at least one of three methods: parasite isolation, identification of amastigotes by histopathologic examination studies of biopsies, or a positive quantitative polymerase chain reaction test (qPCR) specific for parasite DNA derived from parasite tissue samples [30]. A control group was composed of 17 healthy Brazilian subjects (HS) who resided in a non-endemic area of northeast Brazil. As a part of their medical care, after confirmed diagnosis, patients underwent treatment with pentavalent antimony (Sbv), which is standard therapy for leishmaniasis in Brazil. CL patients received intravenous Sbv at a dosage of 20 mg per kg of body weight per day over 20 days. Subjects were observed throughout the course of therapy and evaluated after completion of therapy for cure. Patients were considered cured of there was complete healing of lesions 90 days after initiation of therapy, with skin re-epithelialization and the absence of raised borders [31].
A single L. braziliensis isolate (MHOM/BR/LTCP11245) was used in all experiments. This isolate was obtained from a skin lesion of a CL patient from Corte de Pedra and was characterized as L. brazilensis using both a qPCR assay and the standard isoenzyme electrophoretic mobility assay [32]. Parasites were cryopreserved after isolation in biphasic medium (NNN) after isolation. Before use, parasites were grown in Schneider’s medium (Aldrich Sigma, St. Louis, MO, USA) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Gibco-Life Technologies, Grand Island, NY, USA) and 2% sterile urine [33] obtained from healthy volunteers after informed consent.
Parasites were labeled with carboxyfluorescein succinimidyl ester (CFSE) (Invitrogen) as previously described [34]. Briefly L. braziliensis promastigotes were washed in saline and resuspended at 6x107 in 10 ml of saline with 5 μM of CFSE, and incubated at 37°C for 5 minutes. Then parasites were washed twice saline containing FBS and resuspended in RPMI.
Neutrophils were obtained from heparinized venous blood by density gradient centrifugation using Ficoll Hypaque (LSM; Organon, Durham, NC, USA). The PBMC monolayer was collected from above the Ficoll layer, and erythrocytes were removed from the layer below Ficoll by Dextran sedimentation (Pharmacosmos A/S, Denmark) leaving a population of predominantly neutrophils [35]. The purity of neutrophils isolated using this technique was 95–99%, determined by microscopy using May-Gruenwald-Giemsa staining of cytocentrifuged slides. The cell concentration was adjusted to 1x106/ml in complete culture media, consisting of RPMI 1640 (Gibco-Life Technologies, Grand Island, NY, USA) supplemented with 100U penicillin/ml, 100 μg streptomycin/ml and 10% heat-inactivated fetal bovine serum (FBS) or 10% autologous serum.
One x 106 neutrophils were co-incubated with L. braziliensis promastigotes at a parasite to PMN ratio of 5:1, 37°C, 5% CO2, in 1 ml of complete medium with 10% autologous serum. Neutrophils were stimulated with 10 ng/ml Phorbol 12-myristate 13-acetate (PMA) as a positive control.
After 10, 90 or 180 minutes of incubation, cytocentrifuge slides were prepared and stained with Giemsa, and the numbers of infected cells and intracellular L. braziliensis per 100 neutrophils were quantified by optical microscopy.
After incubation at 37°C, 5% CO2, cells were incubated at 4°C for 15 minutes, stained with fluorochrome conjugated monoclonal antibodies, and suspended in saline. Antibodies were: CD16 PE, CD62L-PECy7 and CD66b-PerCPCy5.5 (BD Pharmingen, San Diego, CA, USA). Flow cytometry data (at least 50,000 events per sample) were acquired using either a FACSVerse Flowcytometer (BD Bioscience) or a FACS CantoII (BD Bioscience). Data were analyzed using FlowJo software (Tree Star Inc., Ashland, OR, USA).
Neutrophils were identified by forward- and side-scatter characteristics and, in some cases, CD16+ expression. Infected or bystander population of neutrophils were identified as CFSE+ or CFSE- cells, respectively (S1 Fig).
As the experiments with healthy controls subjects were performed at the laboratory at Complexo Hospitalar Universitário Professor Edgard Santos in Salvador city, we used a different flow cytometry than we have at the laboratory in the endemic area.
The production of reactive oxygen species (ROS) was evaluated by flow cytometry using the fluorogenic substrate dihydrorhodamine 123 (DHR 123, Cayman Chemical Company, Ann Arbor, MI, USA) as an indicator. Briefly, 1x106 neutrophils were incubated with 10 ng/ml of DHR123 after which either L. braziliensis promastigotes, PMA or buffer (control) was added. Control samples containing no stimulus were run in parallel. After 15 minutes incubation at 37°C, 5% CO2, cells were washed in PBS and analyzed by flow cytometry. The neutrophil population was gated on the basis of forward and side scatter followed by DHR123 fluorescence. Separate controls verified that this population corresponded to neutrophils according to CD16+ surface stain.
In a separate experiment, neutrophils were treated with 10 mM of NADPH oxidase inhibitor, Diphenyleneiodonium chloride (DPI) prior to parasite exposure. The frequency of infected cells and the parasite burden were evaluated by microscopy.
Parasite viability was evaluated by enumeration after recovery in culture as previously described [24]. Briefly, after 180 minutes of infection with L. braziliensis, neutrophils were incubated in Schneider’s medium at 24°C for an additional 24 hours. L. braziliensis viability was measured by assessing the number of extracellular motile promastigotes.
After incubation with L. braziliensis, PMA or medium, neutrophil culture supernatants were collected. Chemokine levels (CCL4, CXCL9, CXCL8, CXCL10) were measured by sandwich ELISA according to the manufacturers’ instructions (R&D Systems, Minneapolis, MN, USA).
A nonparametric Wilcoxon Signed-Rank Test was used to compare the results obtained with cells in different conditions from the same subject. A nonparametric Mann-Whitney U-test was used to evaluate differences among the groups. Statistical analyses were performed using GraphPad Prism 4.0 (GraphPad Software, Inc., San Diego, CA, USA). An alpha of P<0.05 was considered statistically significant.
Comparison of L. braziliensis-infection of neutrophils from CL patients versus healthy subjects showed that neutrophils from both groups were similarly infected (Fig 1), according to both the percent infection (panel A) and the parasite load (panel B) at all-time points evaluated. No difference was observed between parasite loads in neutrophils from CL patients versus healthy subjects: 187[131–249], 227[151–271], 235[126–321] versus 205[79–224], 201[136–270], 203[170–289] parasites/100 neutrophils, after 10, 90 and 180 minutes respectively. Despite this, the number of parasites per neutrophils from CL patients increased over time (panel B). These data suggest that L. braziliensis is taken up by both CL and HS neutrophils at similar rates.
We evaluated surface expression of neutrophil activation markers to investigate whether infection with L. braziliensis stimulates neutrophils to assume an activated phenotype. Activation markers were CD62L, an integrin shed from neutrophil surfaces upon activation, and CD66b, a granule marker that increases upon neutrophil degranulation. These surface markers were measured on both Leishmania-infected and uninfected “bystander” neutrophils. As the abundance of surface activation markers on CL versus healthy subject PMNs was assessed on different flow cytometers, these results are shown in different graphs. Thus the patterns of surface expression can be compared, but the absolute values of fluorescence intensity cannot.
Flow cytometry was sufficient to detect CFSE-labeled L. braziliensis in infected and uninfected populations of neutrophils (Fig 2A). The surface L. braziliensis infection led to a significant reduction in surface CD62L on infected neutrophils (CFSE+) compared to basal state, unexposed (CFSE-) neutrophils (Fig 2C). A similar reduction was also observed on neutrophils stimulated with PMA. Additionally, CD62L decreased significantly on the surface of uninfected bystander neutrophils from healthy subjects, although the decrease did not reach statistical significance in subjects with CL. During this short incubation time (90 minutes) it seems likely that any internalized parasites would remain morphologically intact even if killed by phagocytosis (S2 Fig). Thus it seems likely that bystander cells were likely truly uninfected.
Surface CD66b on neutrophils from CL subjects also increased significantly on neutrophils from CL or healthy subjects after incubation with L. braziliensis or with PMA (Fig 2C). Similar to above CD62L results, changes were observed both in infected and bystander neutrophils, although at to a lower magnitude in bystander cells. Together, these data suggest that L. braziliensis or PMA can trigger neutrophil activation. Furthermore, these responses did not differ between subjects with active CL or healthy control subjects.
As neutrophils from both CL patients and controls presented a similar phenotype, the functional profiles of these cells were investigated. We first evaluated the capacity of Leishmania parasites to trigger oxidant production in neutrophils. Fluorescence of DHR-123, an indicator of the abundance of cellular reactive oxidants, was measured by flow cytometry (Fig 3 panel A).
Following exposure to either L. braziliensis or to PMA, neutrophils from both groups of subjects released significantly greater amounts of oxidants than neutrophils incubated in basal conditions (Fig 3B). However, the abundance of reactive oxidants produced by neutrophils from CL patients was significantly greater than that generated by healthy controls in response to either stimulus (Fig 3B). Thus, despite similar levels of infection (Fig 1), these data suggest that neutrophils form CL patients are capable of producing significantly more reactive oxidants than neutrophils from healthy controls after in vitro exposure to L. braziliensis.
To investigate whether high levels of reactive oxidants produced by neutrophils from CL patients could control the growth of intracellular parasites, NADPH oxidase was inhibited with DPI and the number of infected cells and the parasite burden were evaluated. The data showed there was no difference between the frequency of infected neutrophils or the total parasite burdens in neutrophils incubated in the presence or absence of DPI (89.6±5.4 versus 82±6.3 percent infected, respectively), or (242.2±48 versus 187.6±34 parasites per 100 neutrophils, respectively). These data suggest that high levels of ROS produced by neutrophils from CL patients do not result in parasite killing.
In order to confirm that neutrophils did not participate in the parasite killing we assessed parasite viability. There was no difference between the number of live promastigotes recovered from L. braziliensis infected neutrophils from CL patients compared to those from healthy subjects after 24 hours of culture (5.3x 106 parasites/ml ±1.8 versus 5x106 parasites/ml ±0.7). This demonstrates that, in this time frame, neutrophils did not participate in the control of intracellular L. braziliensis proliferation.
The production of chemokines was measured as an additional measure of neutrophils function. Thus CCL4, CXCL8, CXCL9 and CXCL10 were measured in supernatants from unstimulated or L. braziliensis-infected neutrophils from CL patients or healthy subjects using ELISA (Fig 4).
The basal production of CXCL8 was significantly lower in CL patient neutrophils than neutrophils from healthy subjects (P<0.001). Leishmania braziliensis exposure significantly enhanced CXCL8 production by both groups, although the level reached a higher average in CL subjects than healthy controls (Fig 4). Additionally, the production of CXCL9 by neutrophils from CL versus healthy subjects induced by L. braziliensis exposure was significantly higher than neutrophils from healthy subjects. The abundance of CCL4 was low or below the assay detection level in samples from most subjects, and there was no difference in CXCL10 production between neutrophils from CL compared to healthy subjects.
To evaluate whether neutrophil characteristics were modified by successful treatment of patients with leishmaniasis, we compared the oxidative responses of neutrophils from the same CL patients before therapy, and after treatment-induced cure. The abundance of both the spontaneously produced reactive oxidants, and oxidants produced in neutrophils incubated with L. braziliensis, were significantly lower in subjects after treatment than before (Fig 5). There was no difference observed in CD62L expression on neutrophils from CL patients before and after cure (Fig 6A). However, CD66b expression was significantly lower on both unstimulated (medium) and L. braziliensis-stimulated neutrophils from subjects after cure (Fig 6B).
A comparison of CXCL8 and CXCL9 released by neutrophils from CL patients before or after successful treatment is shown in Fig 7. CXCL8 produced by L. braziliensis-exposed neutrophils from CL subjects after treatment was significantly lower than before treatment (Fig 7A). Analysis revealed no significant differences between CXCL9 levels before or after treatment (Fig 7B), although careful examination of the data revealed all but two subjects exhibited a drop in CXCL9.
Studies of the pathogenesis of human CL have revealed a fine balance between type 1 adaptive immune responses leading to parasite clearance and exaggerated inflammatory responses leading to tissue damage [2,5]. As an illustration, IFN-γ and TNF, which are required for cure of infection in mouse models, do not prevent ulceration in humans and actually correlate with the development of disease [5]. Levels of IFN-γ and TNF directly correlate with lesion size [5] and the levels fall after successful therapy [4]. Moreover, immunomodulators that downmodulate the immune response and decrease TNF production, such as GM-CSF or pentoxyfilline, are more effective than antimony alone at reducing the time to healing and promoting cure of patients who are refractory to treatment with antimony alone [36,37]. Furthermore, peripheral blood cells from individuals with subclinical infection, detected by a positive delayed type hypersensitivity test (DTH) to soluble leishmanial antigen (SLA) with no history of symptomatic disease, produce lower levels of these cytokines than CL patients [6].
Although neutrophils have been observed in CL lesions [38], a role for these cells in the pathogenesis of L. braziliensis disease pathogenesis has not been defined. Neutrophils are generally thought to be short-lived hematopoietic cells that migrate quickly to sites of infection. In mice, neutrophils migrate in large numbers to tissues infected with L.braziliensis [21,39]. Neutrophils are also found in tissues of CL patients albeit usually in small numbers [40,41]. In contrast, macrophages and lymphocytes are the main hematopoietic cells at the site of inflammation in patients with CL, after several weeks to months of infection when biopsies are usually performed [38,42]. The current study was based on the hypothesis that neutrophils contribute to the inflammation observed in human CL. Neutrophils can migrate to the site of infection and may produce inflammatory mediators in response to L. braziliensis infection triggering adaptive immune response, and thus could have an impact on the outcome of disease.
Our data show that the frequency of infected cells in neutrophils from both CL patients and healthy subjects remained unchanged over a 180 minutes course of in vitro infection, showing that neutrophils from both groups were similarly susceptible to L. braziliensis infection. However, we observed increased parasite loads in CL patient neutrophils during increased lengths of parasite exposure. Previous studies have been demonstrated that blocking neutrophil CR3 reduces the uptake of L. braziliensis [23] and TLR2 expression increases after L. braziliensis infection [24]. It is possible that neutrophils from CL patients may increase their expression of these receptors associated with parasite uptake, and this may influences parasite burden.
Following infection, neutrophils from both CL patients and healthy subjects presented a similar pattern of activation characterized by increased CD66b and decreased CD62L expression. CD66b is endogenous in specific granules and its increased appearance on the PMN surface indicates exocytosis from specific granules [43]. CD62L, also called L-selectin, is a homing receptor that is cleaved from the neutrophil surface upon activation, and its loss facilitates migration out of the circulation [44]. The combined changes in both surface markers is indicative of activated phenotype [23,45]. Similarly, activated neutrophils were observed in a murine model of L. braziliensis infection [21] and studies of L. amazonensis-infected human neutrophils [24] showed that neutrophils from patients with CL due to a different organism a decrease in CD62L after exposure to the parasite. We also observed that, like infected cells, bystander neutrophils also presented an activated phenotype. This could have occurred due to exposure to infected neutrophils, and/or to transient contact with parasites. Alternatively, it has been demonstrated that exosomes, released from Leishmania spp. parasites have proinflammatory properties [46] and can activate resting neutrophils [47] or dendritic cells [48]. Furthermore, bystander dendritic cells express high levels of class II, CD80 and CD86 after exposure to L. braziliensis, and their activation has been shown to require both, and host TNF [48].
Innate anti-microbial mechanisms of neutrophils include generation of reactive oxygen species (ROS), release of granule contents [14], and production of neutrophil extracellular traps (NETs) [49,50]. Phagocytosis can activate neutrophil NADPH oxidase, generating reactive oxygen species that can contribute to the elimination of internalized microorganisms [23]. Data shown in the current report document an increase in ROS generation upon L. braziliensis infection of neutrophils from CL patients compared to controls. This result agrees with studies showing that L. braziliensis triggers ROS production by murine neutrophils [21,22]. Monocytes from patients with CL also produce ROS after exposure to L. braziliensis, and in this cell the ROS may contribute to control of parasite replication [51,52]. In contrast to monocytes, we did not detect evidence that the excess ROS generated by infected CL neutrophils contributed to control of intracellular parasite level. As further evidence for its lack of effect, inhibiting ROS generation by inhibition of the NADPH oxidase in neutrophils did not alter either the number of infected cells or the number of internalized parasites.
Recently, roles for neutrophils in the pathogenesis of leishmaniasis have been explored both in vitro and using mouse models. Phagocytosis of apoptotic leishmania-infected neutrophils by macrophages results in transfer of live parasites to macrophages, while changing the macrophage phenotype to an anti-inflammatory state characterized by production of TGF-β [53]. This has raised the hypotheses that neutrophils harboring intracellular leishmania may act as a “Trojan Horse”, serving to both pass live parasites to macrophages and inhibit macrophage microbicidal activity. Neutrophils have opposing effects in vivo depending on the genetic background of the host mouse [54]. Thus, neutrophil depletion from genetically susceptible BALB/c mice infected with L. major decreased parasite burden, whereas neutrophil depletion did not affect the development of a protective type 1 response in genetically resistant C57BL/6 mice [15]. Similarly, neutrophil depletion from BALB/c mice infected with L. amazonensis increased both parasite burden and lesion size, whereas neutrophil depletion did not modify the course of L. amazonensis infection in resistant C57BL/6 mice [16]. This may be in part due to phenotypic differences between neutrophils from susceptible and resistance mice; BALB/c neutrophils express lower levels of TLR2, TLR7 and TLR9 and secrete lower amounts of IL-12p70 after L. major infection than C57BL/6 neutrophils [17]. Both results suggest either a protective or an indifferent role for neutrophils in disease pathogenesis. Our data suggest neutrophils may be indifferent to control of parasite loads, but might contribute to the inflammatory state of the host.
The above reports in murine models of leishmaniasis raise the hypothesis that infiltrating neutrophils may influence the development of adaptive immune responses to L. braziliensis in humans. Although we cannot directly test this hypothesis, the release of chemokines and cytokines from infected neutrophils suggests they may influence cellular responses. Other reports have documented neutrophils producing chemokines and cytokines including CXCL8, CXCL9, CXCL10, IFN-γ, IL-12, CCL3, CCL4, IL-17 and IL-23 [55,56]. Of particular interest to us in this study were CXCL8 which induces neutrophil migration, and CXCL9 and CXCL10 which participate in recruitment of Th1-type lymphocytes [2,5]. Also of interest is CCL4, which recruits monocytes and NK among others [57]. Although CCL4 and CXCL10 levels did not differ between CL and healthy control neutrophils, the chemokines CXCL8 and CXCL9 were augmented in neutrophil supernatants from subjects with CL. These chemokines may participate in the recruitment of neutrophils and T cells to the site of L. braziliensis infection, thus contributing to the overall inflammatory state. Our findings do not suggest that the previous report of L. major inhibition of neutrophil CXCL10 can be to generalized to L. braziliensis [58], but our data do suggest that neutrophils augment CXCL9 similar to the reported increase in macrophages from CL patients [59]. It remains to be seen whether differences or similarities can be attributed to different host responses of human neutrophils to distinct Leishmania species.
Our data show that circulating peripheral blood neutrophils from patients with CL were more activated, they produced higher levels of reactive oxidants and they generated higher amounts of the proinflammatory chemokines CXCL8 and CXCL9 than neutrophils from healthy subjects. These neutrophil changes were largely reversed after successful therapy of CL. Surprisingly, the heightened activation state and greater ROS production by neutrophils from CL subjects did not result in a greater capacity to control intracellular parasites. These data suggest that neutrophils contribute to the inflammatory environment observed in cutaneous leishmaniasis, primarily through the production of inflammatory mediators responsible for the recruitment of T cells and by ROS production, but that they may not contribute to parasite clearance.
Although CL is a localized disease, it is well known that proinflammatory cytokines are increased in plasma [60] and they are generated by peripheral blood lymphocytes stimulation with Leishmania antigen [61]. Neutrophils are short lived cells, they can become primed or activated by cytokines produced by T cells including TNF and IFN-γ, resulting in enhanced ROS and chemokine release [62,63]. Thus, circulating CL neutrophils behaved more like primed neutrophils, poised for rapid activation, than resting neutrophils. The reversal of the neutrophil function after therapeutic cure of localized CL supports this hypothesis.
The dissociation between the inflammatory profile and the ability of neutrophils to kill intracellular Leishmania killing has been shown in monocytes and macrophages infected with L. braziliensis [51,59]. This lack of microbicidal activity differed from healthy control neutrophils, suggesting that the circulating inflammatory neutrophil phenotype does not help to clear infection. Additional studies will be required to determine whether this altered circulating neutrophil phenotype is responsible for maintenance of the inflammatory response observed in tegumentary leishmaniasis due to L. braziliensis. These observations highlight the importance of correlating phenotypic changes with function in circulating and local tissue cellular responses, in order to understand the extent of inflammatory dysregulation that occurs during tegumentary leishmaniasis and other chronic infections.
|
10.1371/journal.pgen.1004687 | BMP-FGF Signaling Axis Mediates Wnt-Induced Epidermal Stratification in Developing Mammalian Skin | Epidermal stratification of the mammalian skin requires proliferative basal progenitors to generate intermediate cells that separate from the basal layer and are replaced by post-mitotic cells. Although Wnt signaling has been implicated in this developmental process, the mechanism underlying Wnt-mediated regulation of basal progenitors remains elusive. Here we show that Wnt secreted from proliferative basal cells is not required for their differentiation. However, epidermal production of Wnts is essential for the formation of the spinous layer through modulation of a BMP-FGF signaling cascade in the dermis. The spinous layer defects caused by disruption of Wnt secretion can be restored by transgenically expressed Bmp4. Non-cell autonomous BMP4 promotes activation of FGF7 and FGF10 signaling, leading to an increase in proliferative basal cell population. Our findings identify an essential BMP-FGF signaling axis in the dermis that responds to the epidermal Wnts and feedbacks to regulate basal progenitors during epidermal stratification.
| Epidermis, a thin layer of stratified epithelium forming the outmost surface of the skin, provides the crucial function to protect animals from environmental insults, such as bacterial pathogens and water loss. This barrier function is established in embryogenesis, during which single layered epithelial cells differentiate into distinct layers of keratinocytes. Many human genetic diseases are featured with epidermal disruption, affecting at least one in five patients. Skin regeneration and future therapeutics require a thorough understanding of the molecular mechanisms underlying epidermal stratification. Wnt ligands have been implicated in hair follicle induction during skin development and self-renewal of stem cells in the interfollicular epidermis of adult skin; however, little is known about the mechanism of how Wnt signaling controls epidermal stratification during embryogenesis. In this study, by using a genetic mouse model to disrupt Wnt production in skin development, we found that signaling of epidermal Wnt in the dermis initiate mesenchymal responses by activating a Bone Morphogenetic Protein (BMP) and Fibroblast growth factor (FGF) signaling cascade, and this activation is required for feedback regulations in the embryonic epidermis to control stratification. Our findings identify a genetic hierarchy of signaling essential for epidermal-mesenchymal interactions, and promote our understanding of mammalian skin development.
| Vertebrate epidermis, the outermost layer of skin, functions as a barrier for protection against environmental insult and dehydration. At approximately embryonic day 8.5 (E8.5) during mouse embryogenesis, the single-layered surface ectoderm adopts an epidermal developmental fate by turning off the expression of keratins 8 and 18 (K8/K18) and switching on the expression of K5/K14, leading to the replacement of the unspecified ectoderm by the embryonic basal layer [1], [2]. Subsequently, the change of cell proliferation from symmetric to asymmetric division becomes evident at E12.5 to 14.5 [3]. The proliferative basal layer periodically produces intermediate suprabasal cells positive for K1/K10, programmed for terminal differentiation of keratinocytes [2]. The transient intermediate keratinocytes then exit the cell cycle, followed by detachment from the basal layer and migration outward to form the spinous layer, characterized by the expression of K1 and K10. Subsequent developmental events engage the expression of differentiation genes, including loricrin and filaggrin, as spinous keratinocytes further develop into the granular and cornified layers contributing to barrier establishment at late embryonic stages ([2].
The tumor-suppressor p53-related transcription factor, p63, encodes regulators required for initiating epithelial stratification during development and maintaining proliferative potential of the basal layer keratinocytes [4], [5], [6], [7]. Two different classes of protein are encoded by p63: the first contains the amino terminal transactivation domain (TA isoforms) and the second lacks this domain (ΔN isoforms) [8]. ΔNp63 is expressed predominantly in the basal layer keratinocytes but its expression is down-regulated in the post-mitotic suprabasal keratinocytes, suggesting that p63 plays a crucial role in proliferative capacity of the epidermal progenitors [9], [10].
Several families of secreted signaling molecules, including bone morphogenetic protein (BMP), fibroblast growth factor (FGF), Hedgehog (Hh), and Wnt, have been implicated in embryonic epidermal morphogenesis. Among them, Wnt appears to be the earliest signal known to promote epidermal development [11], [12], [13]. Our previous studies have demonstrated that embryonic epidermis is the source of Wnts essential for establishing and orchestrating signaling communication between the epidermis and the dermis in hair follicle initiation [14]. Overexpression of Dkk1, a Wnt antagonist, in the epidermis also results in the absence of hair follicles [11], whereas expression of a constitutively active form of β-catenin in the epithelium leads to premature development of the hair follicle placode [15]. In chicks, high levels of Wnt are able to activate BMP signaling through repression of FGF signaling, leading to a switch of neural cell fate into epidermal cell fate [16], [17]. In addition, BMP signals have also been suggested to control p63 expression during ectodermal development [18]. In an embryonic stem cell (ESC) model recapitulating the stepwise appearance of the epidermal stratification in vitro, BMP4 treatment activates the expression of ΔNp63 isoforms, promoting an induction of the proliferative basal keratinocyte makers, K5 and K14, and a progressive enhancement of the terminal differentiation markers, K1, K10, involucrin and filaggrins [19]. In addition, BMP signals have also been suggested to control p63 expression during ectodermal development. Moreover, BMP signaling is also active in the interfollicular epidermis where it may act as a morphogen by promoting epidermal development through inhibition of the hair follicle fate during skin morphogenesis [1], [11], [20], [21]. It has been suggested that FGF7 (KGF) and FGF10 function in concert via FGFR-2 (IIIb) to stimulate keratinocyte proliferation in the epidermis [22], [23], [24], [25], [26], despite the fact that targeted loss of Fgf7 has no effect on skin development in the mouse [27]. Interestingly, FGF ligands appear to be expressed in the dermis while the receptor is present in the epidermis during skin development [22], [24], [28]. However, how these developmental signals are integrated and interplayed across the epithelium and mesenchyme to control epidermal stratification remains to be elucidated.
In this study, we investigated the genetic regulation of these signaling pathways during epidermal stratification and elucidated the mechanism underlying this developmental process orchestrated by the Wnt, BMP, and FGF signaling pathways. Using a mouse model with epithelial ablation of Gpr177 (also known as Wls/Evi/Srt in Drosophila), a regulator essential for intracellular Wnt trafficking, to disrupt Wnt secretion in skin development [29], [30], [31], [32], we identified a crucial role of Wnt signaling in orchestrating epidermal stratification. We demonstrate that signaling of epidermal Wnt to the dermis initiates mesenchymal responses by activating a BMP-FGF signaling cascade. This activation is required for feedback regulations in the epidermis to control the stratification process. Our findings thus decipher a hierarchy of signaling loop essential for epithelial-mesenchymal interactions in the mammalian skin development.
Gpr177 is expressed in the skin of the developing limb bud as early as E11.5 (Figure S1A, B). Similar to our previous observations in dorsal body skin [14], Gpr177 protein can be found predominantly in the epidermis and weakly in the underlying dermis (Figure 1A–C) at E11.5–13.5. To assess the requirement of epidermal Wnts in the development of skin, we generated Gpr177K14 mice in which Gpr177 is inactivated by the K14-Cre transgenic allele to disrupt the secretion of Wnt proteins [32]. Using a R26R reporter line, we examined the Cre-mediated deletion, which occurs only in the epidermis (Figure S1C, D). The loss of Gpr177 was clearly evident in the epidermis but not the dermis of Gpr177K14 (Figure 1A′–C′), indicating a targeted removal of Gpr177 in the mutants. We noted that the Cre recombination is uniformly detected in the limb skin (Figure S1C, D) but not in the dorsal body skin (Figure S1 E, F–G, F′–G′) using the K14-Cre line. Compared to the Gpr177K5 mice that exhibited a uniform expression pattern of Cre and consistent phenotypes associated the Gpr177 deletion described previously [14], the Gpr177K14 mice are not suitable for the study of the body skin due to inconsistent results on skin thickness (Figure S1 F–H, F′–H′). However, the Gpr177K14 model is ideally suited for studies on epidermal development of the limb.
The Gpr177K14 autopods displayed severe deformities including loss of nail formation (Figure 1D, D′). The interdigital and dorsal soft tissues appeared to be edematous (Figure 1E, E′), but skeletal staining revealed comparable structures between controls and mutants (Figure 1F, F′), suggesting that the dysmorphic features of the Gpr177K14 autopods is likely due to impairments in the skin tissue. Histological analysis of autopods showed a reduction in skin thickness as well as in cell proliferation rate, indicating the ablation of skin stratification in Gpr177K14 (Figure 1G–H′ and Figure S2A, B and E). To further investigate the edematous skin abnormalities, we characterized epidermal stratification of the limb skin using markers specific for basal, spinous, and granular epidermal layers. The deletion of Gpr177 diminished the number of basal cells expressing KRT5 (Figure 1I, I′). Significant reduction of the spinous layer positive for KRT1 and KRT10 was also identified in the longitudinal sections along the dorsal skin of the mutant autopods (Figure 1J–K, J′–K′). However, the granular layer positive for loricrin and the basal membrane protein, laminin 1, did not show significant alterations (Figure 1L–M, L′–M′). The results were consistent with alterations of the limb skin thickness caused by the Cre-mediated deletion of Gpr177 (Figure S2). Besides, an uneven decrease in skin thickness also occurred in the dorsal body of Gpr177K14, as shown by histology (Figure S1H, H′) and immunohistochemistry specific for the spinous and basal layers (Figure S1I–J, I′–J′). TUNEL assay did not reveal significant changes in apoptosis, indicating that defects in the spinous layer were not caused by abnormal cell death (Figure S3). Thus, the spinous hypoplasia is likely attributed to defects in the epithelial vertical expansion of Gpr177K14 mice.
The deletion of Gpr177 has been shown to affect Wnt signaling during the development of other organs [14], [32], [33]. This is also true during the morphogenesis of the limb skin, as the the expression of several downstream mediator critical for Wnt signal transduction including Axin2, Dkk1, Fzd1, Lef1, and TCF4 was significantly reduced in the skin of Gpr177K14 autopods (Figure 2A), and the activity of Wnt/β-catenin signaling in the mesenchyme underlying the interfollicular epithelium was almost completely eliminated, evidenced by the lack of TopGal reporter activity (Figure 2B–C, B′–C′). In situ hybridization analysis further confirmed that epidermal ablation of Gpr177 affects the expression of Lef1 and Axin2 in both the epithelium and mesenchyme (Figure 2D–G, D′–G′). These observations are consistent with our observations in dorsal body skin (Figure S4A–D, E–F, E′–F′) [14], indicating a requirement of epidermal Wnt for signaling activation in both epidermal and dermal layers. Consistent with this finding, Dermo1-Cre mediated deletion of Gpr177 in the dermis did not alter the skin thickness (Figure S5), suggesting a dispensable role of dermal Wnt in epidermal stratification.
To decipher the effects of the alteration in Wnt signaling during autopod skin morphogenesis, we performed RNA expression profiling analysis using microarray to identify genes that are differentially expressed in the E15.5 distal limbs (Figure S6 and Table S1 and Table S2). Among those altered genes, members of BMP family were significantly affected in Gpr177K14. In response to β-catenin/Wnt signaling, BMP signaling in the dermal mesenchyme plays critical role in hair follicle induction [14]. Thus, we hypothesized that BMPs are downstream targets of Wnt signaling and regulate epidermal stratification. Real time RT-PCR analysis validated that Bmp2, 4, and 7 expression was decreased in the mutants (Figure 3A). During normal development of the autopod skin, Bmp2 and Bmp7 were found in both the epidermis and dermis while Bmp4 appeared to be exclusively expressed in the dermis (Figure 3B–G). However, epidermal deletion of Gpr177 caused profound reduction of Bmp2, Bmp4, and Bmp7 in the developing skin (Figure 3B′–G′ and Figure S7A–G, A′–G′), suggesting that BMP signaling, regulated by Wnt signaling, is likely to be involved in epidermal stratification.
To test the functional requirement of BMP signaling in the Gpr177-mediated skin morphogenesis, we used a conditional Bmp4 transgenic allele. The Tg-pmes-Bmp4 transgenic mouse was crossed onto the Gpr177K14 background to generate Gpr177K14/Tg-pmes-Bmp4 mice. The transgenic Bmp4 expression from this transgenic allele was tightly controlled by a transcription and translation STOP cassette flanked by two loxP sites, permitting the Cre-mediated activation (Figure 3H–I) [34], [35].
The transgenic expression of Bmp4 was able to alleviate the dysmorphic phenotype caused by the deletion of Gpr177 (Figure 3J–L). The Gpr177K14/Tg-pmes-Bmp4 autopods displayed five separated digits without skin edema (Figure 3L), suggesting that BMP4 acts downstream of Wnt signaling in skin stratification. To determine if this epidermal expression of transgenic Bmp4 could substitute for mesenchymal Bmp4 to rescue spinous layer defect, we examined the spinous layer of Gpr177K14/Tg-pmes-Bmp4 autopods. Immunostaining of KRT5 and KRT1/10 revealed a significant enhancement in their expression (Figure 3N–P, R–T and V–X). Histological (Figure S2 C–F) and ultrastructural analyses (Figure S2 F–H) further showed that the thickness of the spinous layer was obviously increased in the E18.5 Gpr177K14/Tg-pmes-Bmp4 epidermis, as compared to that in Gpr177K14 epidermis.
The transgenic expression of Bmp4 in the epidermis (Figure 3H–I) may exert its signaling effects in a cell autonomous or non-cell autonomous manner. For non-cell autonomous signaling, it requires the diffusion of BMP4 through an inter-tissue signal transduction mechanism. It has been shown that BMPR1A is responsible for mediating BMP signaling in epidermal development [20], [36], [37]. If the transgenic Bmp4 indeed acts in a cell autonomous manner, we assumed that activation of BMPR1A-mediated signaling in the epidermis would also alleviate the stratification defects in Gpr177K14 autopods. Accordingly, we compounded a conditional transgenic allele that expresses a constitutively active form of BMPR1A receptor (caBMPR1A) with Gpr177K14 mice (Figure 3H–I) [34]. However, ectopic activation of BMPR1A signaling neither rescued the autopod defects at the morphological (Figure 3M) and histological (Figure S2D) levels nor restored the expression of the basal and spinous layer makers, KRT5 (Figure 3Q), KRT1 (Figure 3U), and KRT10 (Figure 3V), as compared to that in Gpr177K14 mice (Figure 3O, S, W). These results thus suggest a non-cell autonomous BMP signaling across tissue layers to alleviate the epidermal defects of Gpr177K14, and the BMP4 activity in the dermal mesenchyme, but not in the epidermis, is required for proper stratification of the mammalian skin.
Maturation of the spinous layer first requires the mitotic suprabasal intermediate cells to be replaced by the post-mitotic cells [2]. The hypoplasia developed in the Gpr177K14 spinous layer might be attributed to failure in this replacement. To test this possibility, we performed a BrdU labeling experiment to identify the KRT1 positive keratinocytes undergoing active proliferation between E13.5 and 16.5. Double labeling was able to detect cells positive for BrdU and KRT1 in the E13.5 and 14.5 wild type epidermis (Figure 4A, B). No double positive cells were found at E15.5 and 16.5 (Figure 4C, D). In addition, this replacement process did not seem to be affected by Gpr177 deletion or transgenic Bmp4 expression (Figure 4E–L and Y). Thus, the initial programming of intermediate cells to become spinous keratinocytes is independent of the Gpr177 mediated regulation and BMP signaling.
As skin stratification requires proper proliferation of the basal cells [9], [10], we further examined if defects in basal cell division contribute to the epidermal abnormalities caused by Gpr177 deficiency. Double labeling of BrdU and KRT5 permits quantification of the ratio of basal cells proliferation. Closer examinations revealed that the number of KRT5-positive basal cells labeled with BrdU (Figure 4M–P) is significantly reduced by Gpr177 ablation (Figure 4Q–T). However, this hypoplastic feature was alleviated in the Gpr177K14/Tg-pmes-Bmp4 mutants (Figure 4U–X), where the ratio of BrdU labelled basal cells arises between E14.5 and 16.5 to the levels close to controls (Figure 4Z). These observations suggest that the Gpr177-mediated regulation of BMP signaling maintains the high proliferative potential of the basal cells essential for epidermal stratification.
It has been shown that p63 transcription factor is critical for the proliferative potential of epidermal stem cells in the stratified epithelium [9], [10], [18], [38]. We therefore tested if p63 is involved in the epidermal stratification mediated by the Wnt/BMP regulatory axis. In situ hybridization analysis showed that the expression of p63 in the epidermis was affected by Gpr177 deletion at E13.5 and 14.5 (Figure 5A–B, A′–B′). The loss of p63 transcripts in the mutants suggests a role of Wnt signaling in the maintenance of its expression in the basal cells (Figure 5A′–B′ and Figure S8A). We next examined the alteration of p63 at the protein level using antibodies against total p63 and its specific isoforms, TA-p63 and ΔNp63. The percentage of the total p63 and ΔNp63 positive basal cells was significantly decreased in Gpr177K14 mutants (Figure 5C–D, F–G, I–J and Figure S8B–J). Consistent with the previous reports [4], [5]. TA-p63 was not involved in epidermal development at these stages (Figure S8 K–P). In addition, transgenic BMP4 was able to elevate the percentage of the total p63 and ΔNp63 positive cells in the basal layer similar to that of wild type control at E13.5, 14.5 and 16.5 (Figure 5C–K and U). To further determine the role of p63 in basal cell proliferation, we performed double labeling of BrdU and p63. The number of the p63-expressing mitotic keratinocytes was reduced in the Gpr177K14 basal layer (Figure 5L–M, O–P and R–S and V), but this reduction was restored by transgenically expressed BMP4 (Figure 5N, Q, T and V), suggesting an involvement of p63 in maintaining the high proliferative potential of basal cells mediated by the Wnt/BMP regulatory axis during epidermal stratification.
To further elucidate the mechanism underlying epidermal stratification mediated by BMP signaling, we examined the activation of Smad1/5/8 mediators that transduce the BMP canonical pathway. Immunostaining of phosphorylated Smad1/5/8 revealed that their activations were significantly affected in the dermis, but not in the epidermis of Gpr177K14 mice (Figure 6A, B, G and D, E, H). The dermal-specific effect was restored by transgenically expressed Bmp4 in Gpr177K14/Tg-pmes-Bmp4 mutants (Figure 3H–I, Figure 6A, B, C, G and Figure S9). In contrast, activation of BMPR1A-mediated signaling failed to restore dermal activation of Smad1/5/8 in the Gpr177K14/Tg-pmes-caBmpr1a mutants (Figure 6D, E, F, H and Figure S9), consistent with non-cell autonomous effects of BMP signaling on the spinous layer (Figure 3). These findings strongly suggest that BMP signaling functions primarily in the dermis, through the canonical pathway, to regulate downstream signaling molecules that act back on the epidermis to control epidermal stratification.
We next sought to identify the downstream mediators of BMP signaling on epidermal stratification. FGF signaling came to our attention because several FGF ligands are known to be expressed exclusively in the dermal cells [22], [39], and knockout of Fgf10 or its receptor FGFR2-IIIb leads to epidermal hypoplastic defects [23], similar to that seen in Gpr177K14 mutants (Figure 1). Using real time RT-PCR analysis, we found that Gpr177 deficiency significantly diminishes the expression of Fgf7 (KGF) and Fgf10 (Figure 7A), both working in concert to activate downstream signaling via FGFR2-IIIb [24], [26], [28], [40]. Furthermore, the reduced expression of Fgf7 and Fgf10 in Gpr177K14 mutants was restored by transgenic Bmp4 expression (Figure 7A and Figure S10A–C). Interestingly, a decrease in the expression of epidermal-specific Fgfr2IIIb was not significantly detected in the Gpr177K14 mutant at the early stage, but was observed at E14.5 (Figure 7A), suggesting an indirect consequence of activation. This reduction of Fgfr2IIIb expression was restored in Gpr177K14Tg-pmes-Bmp4 mice (Figure 7A). In vitro beads implantation assays further demonstrated that exogenously applied BMP2 or BMP4 was able to induce the expression of Fgf7 (17/20 in BMP2 implants and 15/21 in BMP4 implants) and Fgf10 (15/19 in BMP2 implants and 22/25 in BMP4 implants) in the dermal explants of Gpr177K14 mice (Figure 7B), supporting our hypothesis that FGF signaling acts downstream of the Wnt/BMP regulatory axis. To further determine if both Fgf7 and Fgf10 are transcription targets of pSmad1/5/8 signaling, we tested potential binding of pSmad1/5/8 to the regulatory region of Fgf7 and Fgf10 by in vivo chromatin immunoprecipitation (ChIP) assays using embryonic limb skin samples. We utilized five sets of oligos pairs (see Methods and Materials) that amplify five potential binding sites of Smad1/5/8 [41], [42] in the regulatory regions of Fgf7 (Figure 7C) and two sets of oligo pairs for the binding sites in that of Fgf10 (Figure 7C). Quantitative PCR showed that after immunoprecipitation of linked chromatin there was specific enrichment of Smad to a DNA fragment that corresponds to one of potential sites with antibodies against either pSmad1/5/8 or Smad1/5/8 compared to IgG controls (Figure 7D). Thus, ChIP results strongly support the notion that in embryonic limb skin of mouse in vivo, activated Smad1/5/8 is present in the regulatory regions of Fgf7 and Fgf10 loci. To further demonstrate the involvement of FGF signaling in epidermal stratification, organ culture analysis was performed. The wild type and Gpr177K14 skin explants were supplemented with BSA as controls, or with exogenous FGF7 and FGF10. Immunostaining of keratinocyte markers was carried out 48 hours in organ culture. Although the wild type explants exhibited minimal response to the exogenous FGF7 and FGF10, the mutant explants exhibited increased thickness of the spinous layer, elevated number of KRT5-expressing mitotic cells, as well as enhanced expression of p63 in the presence of FGF7 and FGF10 (Figure 8A and Figure S10). Our results thus uncover a functional requirement of the Wnt/BMP/FGF signaling axis as well as their signaling interplay across the epidermis and dermis to orchestrate epidermis stratification.
The Wnt, BMP, and FGF signaling pathways play critical roles in the embryonic development of the skin [11], [23], [24], [25], [43], [44]. Recent studies using mouse models with Wls/Gpr177 deletion have shown that Wnt secreted from the epidermis is essential for the dermal activation of the canonical Wnt pathway and activation of BMP signaling during hair follicle induction [14], [33]. However, how Wnt, BMP, and FGF pathways interact in the genetic networking that regulates the epidermal stratification during embryogenesis remains unclear. Here we used a transgenic Bmp4 mouse line to successfully rescue the defective epidermal stratification of Gpr177K14 mice. We dissect the sequential relationship and signaling crosstalk by which these key pathways interact and mediate epidermal stratification. Based on our results, we propose a genetic hierarchy model that integrates Wnt, BMP, and FGF signaling in the regulation of epidermal stratification (Figure 8B). In this model, a BMP/Smad1/5/8/FGF7/10 signaling cascade in the dermis is activated by epidermal Wnts and feedbacks to regulate basal cell proliferation and the subsequent epidermal stratification. Although the specificity of the Cre mouse line used in this study allows us to present this molecular circuit based on data from the limb skin, our observations from the dorsal skin suggest that the molecular responses involved in this model do not bias the body regions (Figure S1, S4, S6, S8A).
Our in vivo results showed that the proliferating basal cells expressing ΔNp63 were targets of the epidermal Wnt signal, and failed expression of ΔNp63 accounts for the hypoproliferation of these basal cells in the absence of epidermal Wnt. It is consistent with the functional importance of p63 in controlling basal cell proliferation of epidermal development and homeostasis [5], [10], [18], [45], suggesting that sustained expression of Wnt pathway regulated ΔNp63 is critical in maintaining the capability of basal keratinocytes to form the stratified epidermis in the developing mouse embryo.
ΔNp63 has been implicated in the developmental program of epidermal stratification through several mechanisms, including aymmetric division of basal cells and cell cycle exit of intermediate suprabasal cells [3], [5], [46], [47]. Although the basal layer lacking epidermal Wnt failed to maintain the proliferative capability of ΔNp63-expressing cells to form a normal spinous layer, the developmental events of epidermal stratification do take place normally, as evidenced by the occurrence of the asymmetric basal cell division to form intermediate mitotic keratinocytes and the replacement of these cells by post-mitotic keratinocytes in spite of a thinned spinous layer. Hence, our studies suggest that the mechanism by which epidermal production of Wnt affects the vertical expansion of the epidermis underlying the ΔNp63-governed basal keratinocytes is independent of both initiation of the intermediate keratinocytes and cell cycle exit for epidermal differentiation.
Notably and interestingly, unlike the effects of autocrine Wnt signaling on the interfollicular epidermal stem cells (IFESCs) of adult skin [48], loss of epidermal Wnt production in the embryonic skin in our study is not associated with premature differentiation of basal cells. Given the evidence of the embryonic epidermis as a tissue source for activation of β-catenin/Wnt signaling in the dermis of the developing skin [14], [33], there appears to exist a functional requirement for paracrine Wnt signaling in the maintenance of proliferative basal cells in epidermal stratification of embryonic skin.
Epidermal deletion of Gpr177 disrupts the canonical Wnt signaling in the dermis [14], [33] at E13.5, prior to the formation of the intermediate keratinocytic layer and maturation of the spinous layer [3], [5], [9]. Subsequently, expression of Bmp2, Bmp4, Bmp7, Fgf7, and Fgf10, critical for epidermal development [21], [22], [39], [44], [49], is specifically disrupted in the dermis [14], indicating that Wnt signaling functions upstream of these signals. BMP signaling appears to act downstream of Wnt signaling to mediate Wnt function, because activation of BMP signaling (Smad1/5/8 signaling) in the dermis of Gpr177K14 mutants successfully rescues the development of epidermal stratification and underlying molecular events. Irrespective of the contribution of BMPR1A and BMPR1B [36], [50], canonical BMP signaling is activated both in the epithelium and in the dermal mesenchyme of developing skin [51], [52], [53]. Our findings show that while the expression of transgenic Bmp4 is activated in the epidermis of Gpr177K14 mice, the activation of canonical BMP signaling in the dermis enable it to rescue epidermal stratification, suggesting that BMP/Smad1/5/8 signaling in the dermis mediates Wnt signaling to control basal cell proliferation, consistent with the recognized role of balanced BMP signaling in the maintenance of epidermal stem cells, progenitor cell differentiation, and hair follicle induction [1], [21], [36], [44], [54]. Based on the specific activation of Smad1/5/8 pathway by non-cell autonomous transgenic BMP4 seen in the dermis of Gpr177K14 mutants, we suggest that the downstream signaling feedback mechanism is required for the regulation of epidermal basal cells.
Given that loss of epidermal Wnt production at least partially phenocopies the epidermal defects in mice lacking Fgfr2-IIIb [23], the expression of Fgf7 and Fgf10 in the dermis is directly dependent on the presence of BMP/Smad1/5/8 signaling in the dermis in response to Wnt signaling. This implicates FGF7/10 as the downstream mediator for canonical BMP signaling in the dermis for the maintenance of basal cell proliferation. This hypothesis is supported by our skin organ culture experiments where exogenously applied FGF7/FGF10 are sufficient to functionally attenuate the reduction of proliferative basal cells and to rescue the hypoplastic spinous layer of the Gpr177K14 skin, consistent with the function of FGF7 and FGF10 in epidermal development [25], [28], [55]. It would be interesting to see if other keratinocyte mitogens such as EGF can exert similar rescue functions as the FGFs in future investigations. Nevertheless, we propose that in normal stratification of embryonic epidermis, FGF7 and FGF10 secreted from the dermis diffuse to the epidermis to mediate feedback regulation of Wnt and BMP/Smad1/5/8 signaling, which is required for the maintenance of proliferative keratinocytes in the basal layer through modulation of ΔNp63 [56], [57]. Consistent with previous studies that showed FGFR2 is a transcription target of p63 in the epidermis [56], [58], our quantitative RT-PCR results showing the down-regulation of Fgfr2-IIIb at the late stages of epidermal development further support a role of Fgfr2 signaling acting downstream of p63 in epidermal development. Nonetheless, our data suggest that the FGF7/FGF10 function as feedback factors to epidermis, but cannot rule out the possibility of involvement of additional feedback mechanisms [58], [59] between FGF7/10, Fgfr2, and p63 in the epidermis. However, the mechanism of how FGF7/10 signaling feedbacks to the epidermis and positively regulates ΔNp63 to maintain the proliferative basal cells remains unknown and warrants future studies.
In the adult skin, interfollicular epidermal basal cells, unlike hair follicles, proliferate throughout animal life. Recent studies on subtle genetic deletions by Millar and colleagues [60] have distinguished that Wnt/β-catenin signaling contribute to the mechanism controlling interfollicular epidermal cell (IFE) proliferation in the postnatal skin rather than the long-term maintenance of IFE stem cells. In embryonic skin development, our current study supports the notion that the epidermal Wnt initiates mesenchymal responses in the dermis by activating a BMP-FGF signaling cascade. This activation is crucial for the feedback regulations that control the stratification processes in the interfolliclular epidermis, indicating a profound effect of Wnt on signaling interplays across the epithelium and the mesenchyme in orchestrating the basal cell proliferation during epidermal stratification.
Mice carrying Gpr177 floxed allele [30] was crossed with K14-Cre transgenic mice [61] to generate mice with epidermal loss-of-function of Gpr177 (Gpr177K14). A Dermo1-Cre mouse was crossed to Gpr177 floxed allele to delete Gpr177 in dermal compartment of the skin [14]. TOPOGAL reporter [62], BATGAL reporter [63], R26R reporter, Dermo1-Cre mice, and transgenic K14-Cre mice were purchased from The Jackson Laboratory, Maine. Generation of transgenic Tg-pmes-Bmp4 and Tg-pmes-caBmpr1a mice has been described previously, in which the transgenic allele expresses Bmp4 (or caBmpr1a) and Gfp (Green fluorescent protein) simultaneously via an IRES (Internal Ribosome Entry Site) [34], [35]. Animal experimental protocols were approved by The Animal Committee of Hangzhou Normal University, China.
Embryo collection, histology, and in situ hybridization for whole-mount and on sections were performed as previously described [32].
For real-time RT-PCR, embryonic autopods were dissected and treated with 0.1% collegenase to separate the dermal and epidermal compartments. RNA extraction using RNA isolation kit (ambion, RNAqueous-4RNA) and real-time RT-PCR analysis for RNA expression were performed as previously described [32]. The primers: QAxin2: 5′-ACGCAC- TGACCGACGATT-3′ and 5-AAGGCAGCA- GGTTCCACA-3′; QFzd1: 5′-GAGTTCTGGACCAGTAATCCGC-3′ and 5′- ATGAGCCCGT- AAACCTTGGTG-3′; QLef1: 5′- AACGAGTCCGAAATCATCCCA-3′ and 5′- GCCAGAGTA- ACTGGAGTAGGA-3′; QTcf4: 5′-GATGGGACTCCCTATGACCAC-3′ and 5′- GAAAGGGTT- CCTGGATTGCCC-3′; QBmp2: 5′- TCTTCCGGGAACAGATACAGG-3′ and 5′- TGGTGTCC- AATAGTCTGGTCA-3′; QBmp4: 5′-GACTTCGAGGCGACACTTCTA-3′ and 5′- GAATGA- CGGCGCTCTTGCTA-3′; QBmp7: 5′-AGGGCTTCTCCTACCCCTAC-3′ and 5′- GGTGGTAT- CGAGGGTGGAAGA-3′; Q18S: 5′- GAAACGGCTACCACATCC-3′ and 5′- ACCAGAC- TTGCCCTCCA-3′; QDkk1: 5′- GACCTGCTACGAGACCTGGA-3′ and 5′- CTGGAGAGGG- TATGGTTGCC-3′; QFgf7: 5′-CAGAACAAAAGTCAAGGAGCAACCG-3′ and 5′- GTCGCTCGGGGCTGGAACAG-3′; QFgf10: 5′- TCAGCGGGACCAAGAATGAAG-3′ and 5′-CGGCA- ACAACTCCGATTTCC-3′; QFgfr-IIIb: 5′- CCTCGATGTCGTTGAACGGTC-3′ and 5′- CAGCATCCATCTCCGTCACA-3′. QTg-Bmp4: 5′- GGGCTGGCCATTGAGGTGAC-3′ and 5′-ATGGCGACGGCAGTTCTTATTCTT-3′. QTg-caBmpr1a: 5′- TAATAACACATGCATAACTAAT-3′ and 5′-GCTTTTGGTGAATCCTTGCA -3′.
Cell proliferation rate was measured by BrdU incorporation as previously described [32]. Briefly, timed pregnant mice were injected intraperitoneally with BrdU solution at a dosage of 3 mg/100 g of body weight using BrdU Labeling and Detection Kit (Roch Applied Science) 30 minutes prior to embryo collection. Cell apoptosis was detected with TUNEL assay kit (Roche Applied Science). At least 4 embryonic limbs for each genotype were fixed in 4% paraformaldehyde and processed for at 5–7 µm paraffin sections for immunofluorescence analysis according to manufacturer's instructions.
Embryonic limb were fixed in 4% PFA for 30 minutes, washed several times in PBS, and then processed for either paraffin sections or cryostat sections. For cryostat sections, samples were treated for in 5% sucrose and 15% sucrose, 2 hours each, in 30% sucrose. For 2–3 days. Samples were embedded in OCT and sectioned at 20 µm. To conduct immunohistochemical staining, sections were washed 3 times in PBST (0.1%Triton X-100/PBS), then blocked in 5% BSA for 30 minutes, and incubated with primary antibodies diluted with 5% BSA at 4°C overnight in a humid chamber. Sections were subsequently washed in PBST, 3 times for 10 minutes each. Secondary antibodies (1∶1000) and DAPI (1∶500) diluted in 5% BSA were applied for 30 minutes in the dark. Following application of secondary antibodies, the sections were washed several times with PBST, for 10 minutes for each, mounted with Mowiol (Sigma) and stored at 4°C. Primary antibodies used in this study were commercially purchased from Abcam, as detailed below: Cytokeratin 5 (ab24647), Cytokeratin 10 (ab9025), Cytokeratin 1 (ab24643), Filaggrin (ab24584), Loricrin (Ab24722), Anti-laminin (ab14055), p63 (ab53039). Antibody against ΔN-p63 was purchased from Santa Cruz (sc-8609), Antibody against BrdU was purchased from Roche (19691800) and antibody against pSmad1/5/8 purchased from Cell Signaling. Antibody against FGF10 was purchased from Santa Cruz (sc-7375).
For quantification of proliferation, BrdU-positive cells were counted (n = 3–7 limb samples, ≥15 consecutive fields at 40× magnification) and calculated as a percentage of antibody labeled cells and total nuclear stained cells (DAPI positive) otherwise within a defined arbitrary area. For quantification of pSmd1/5/8-positive cells in either the epidermis or the underlying dermis in Figure 6G–H, the numbers of pSmad1/5/8 positive cells in every 300 DAPI positives were counted and calculated as a percentage (n = 3–5 limb samples, ≥15 fields at 40× magnification for each genotype). For quantification of epidermal p63-positive cells in Figure 5, p63-posive cells were counted and calculated in similar way as described above (n = 3 limb samples for each geneotype). Statistical significance was determined using Student's t-test.
Embryonic limbs were dissected from embryos at E13.5 and dorsal skin was separated manually using fined forceps and placed dorsal upward onto a Nucleopore membrane in a culture plate with a central well. Protein beads were soaked with BMP2 (100 ng/µl, R&D), BMP4 (100 ng/µl, R&D), BSA (l00 ng/µl). Explants were cultured at 37°C for 24 hours after implantation of beads onto explants.
Skin organ culture of the dorsal-autopod was conducted using a modification of a previously published procedure [24]. Briefly, dorsal skin portions were dissected from embryonic autopods (hands/feet) at late E13.5 with the assistance of 0.1% collagenase treatment. Skin explants were placed epidermal side up onto a Nucleopore filter (Whitman, pore-size 0.7 µm) that was coated with rat tail collagen type 1 (Sigma) in an organ culture plate with a central well, and cultured in DMEM without serum in 5% CO2 for 72 hours. Protein mixtures of recombinant FGF7 (R&D) and FGF10 (R&D) were applied onto DMEM medium at a final concentration of 250 ng/µl each, and the protein-containing media were replaced every 12 hours. In parallel experiments, BSA was applied onto DEME medium at the same concentration of proteins as control. Organ-cultured skin samples were fixed with 4% PFA and processed for paraffin sections for either immunohistochemistry or H&E staining.
β-Gal staining for both whole-mount and cryostat sections were performed with commercial purchased Kit (Roche) according to manufacturer's instructions. For electronic microscopic analyses, embryonic limbs were fixed in 2.5% glutaraldehyde and dehydrate through graded ethanol and acetone. Samples were processed according to standard protocols
Limb skin tissues from E13.5 mouse embryos were cut into small pieces, and then rinsed in 1% formaldehyde/PBS for 30 min on ice for cross-linking. The cross-linking reaction was stopped by adding glycine to a final concentration of 0.125 M and rotating for 5 min. The crosslinked tissues were ground by Dounce tissue grinder in tissue lysis buffer from Magna ChIP G Tissue Kit. Lysed cells were collected by spin at 10,000× g for 5 min. The pelleted cells were resuspended in 200 µl of Micrococcal nuclease buffer per 30 mg of the pelleted cells. The resuspended cells were digested with 1 µl of Micrococcal nuclease (New England Biolabs) at 37°C for 20 min. Then the reaction was stopped by adding EDTA to a final concentration of 50 mM and followed by sonication on ice at 30 W for 12 pulses of 1 second on, 3 seconds off to further disrupt and release chromatins. Chromatin immunoprecipitation was performed with antibody against Smad1/5/8 (Santa Cruz, sc-6031), pSmad1/5/8 (Cell signaling technology, 9511) or normal rabbit IgG (Beyotime, A7016) using Magna ChIP G Tissue Kit (Millipore) according to the user manual. For the detection of the immunoprecipitated Fgf7 and Fgf10 promoter region, eluted DNA was used as template for quantitative real time PCR analysis with primers specific for Smad-binding sites [41], [42]. Real-time PCR was performed in triplicate using SsoFast EvaGreen Supermix with CFX96 Real-Time PCR Detection System (Bio-Rad Laboratories).
Primers: Fgf7-L1:5′-CTCCATCCTGGTTTTCCTCC-3′ and 5′-GAATAGGACACAGGAAGACAG-3′; Fgf7-L2:5′-AACCTGCTCAGTGACATTCC-3′ and 5′-ACTACAGAATGCCCAGTCTC-3′; Fgf7-L3:5′-TTAGGGTGGTGATACGATGG-3′ and 5′-CTTTCCAGCCTGAGCTTGTG-3′; Fgf7-L4:5′-AGCTGAGCCATGGGGAAGTA-3′ and 5′-GGCTGAGAAGACCTAGTTTC-3′; Fgf7-L5:5′-TTGCTTCCAATGAGGTCAGC-3′ and 5′-GATTTTCTCCGTGTGTGAGC-3′; Fgf10-L1:5′-GGCCATAGAAACAGAGCATG-3′ and 5′-GCTTCAGATTAGAATGGTACC-3′; Fgf10-L2,3:5′-GCAATTAGCAGGAGCTGCAG-3′ and 5′-GATGCCTTTG- CTCTGAGCTG-3′.
|
10.1371/journal.pbio.1000564 | Genomic DNA Sequences from Mastodon and Woolly Mammoth Reveal Deep Speciation of Forest and Savanna Elephants | To elucidate the history of living and extinct elephantids, we generated 39,763 bp of aligned nuclear DNA sequence across 375 loci for African savanna elephant, African forest elephant, Asian elephant, the extinct American mastodon, and the woolly mammoth. Our data establish that the Asian elephant is the closest living relative of the extinct mammoth in the nuclear genome, extending previous findings from mitochondrial DNA analyses. We also find that savanna and forest elephants, which some have argued are the same species, are as or more divergent in the nuclear genome as mammoths and Asian elephants, which are considered to be distinct genera, thus resolving a long-standing debate about the appropriate taxonomic classification of the African elephants. Finally, we document a much larger effective population size in forest elephants compared with the other elephantid taxa, likely reflecting species differences in ancient geographic structure and range and differences in life history traits such as variance in male reproductive success.
| The living elephants are the last survivors of a once highly successful mammalian order, the Proboscidea, which includes extinct species such as the iconic woolly mammoth (Mammuthus primigenius) and the American mastodon (Mammut americanum). Despite numerous studies, the phylogenetic relationships of the modern elephants to the woolly mammoth, as well as the taxonomic status of the African elephants of the genus Loxodonta, remain controversial. This is in large part due to the fact that both the woolly mammoth and the American mastodon (the closest outgroup to elephants and mammoths available for genetic studies) are extinct, posing considerable technical hurdles for comparative genetic analysis. We have used a combination of modern DNA sequencing and targeted PCR amplification to obtain a large data set for comparing American mastodon, woolly mammoth, Asian elephant, African savanna elephant, and African forest elephant. We unequivocally establish that the Asian elephant is the sister species to the woolly mammoth. A surprising finding from our study is that the divergence of African savanna and forest elephants—which some have argued to be two populations of the same species—is about as ancient as the divergence of Asian elephants and mammoths. Given their ancient divergence, we conclude that African savanna and forest elephants should be classified as two distinct species.
| The technology for sequencing DNA from extinct species such as mastodons (genus Mammut) and mammoths (genus Mammuthus) provides a powerful tool for elucidating the phylogeny of the Elephantidae, a family that originated in the Miocene and that includes Asian elephants (genus Elephas), African elephants (genus Loxodonta), and extinct mammoths [1]–[8]. In the highest resolution study to date, complete mitochondrial DNA (mtDNA) genomes from three elephantid genera were compared to the mastodon outgroup. The mtDNA analysis suggested that mammoths and Asian elephants form a clade with an estimated genetic divergence time of 5.8–7.8 million years ago (Mya), while African elephants diverged from an earlier common ancestor 6.6–8.8 Mya [8]. However, mtDNA represents just a single locus in the genome and need not represent the true species phylogeny since a single gene tree can differ from the consensus species tree of the taxa in question [9]–[11]. Generalizing about species relationships based on mtDNA alone is especially problematic for the Elephantidae because their core social groups (“herds”) are matrilocal, with females rarely, if ever, dispersing across groups [12]. This results in mtDNA genealogies in both African [13],[14] and Asian elephants [15] that exhibit deeper divergence and/or different phylogeographic patterns than the nuclear genome.
These observed discrepancies between the phylogeographic patterns of nuclear and mtDNA sequences have led to a debate about the appropriate taxonomic status of African elephants. Most researchers have argued, based on morphology and nuclear DNA markers, that forest (Loxodonta cyclotis) and savanna (Loxodonta africana) elephants should be considered separate species [13],[16]–[19]. However, this notion has been contested [20] based on mtDNA patterns, which reveal some haplogroups with coalescent times of less than half a million years [21] that are shared across forest and savanna elephants, indicating relatively recent gene flow among the ancestors of these taxa. Taxonomies for African elephants based on mtDNA phylogeographic patterns have suggested anywhere from one to four species [20],[22],[23], whereas analysis of morphology and nuclear data sets has suggested two species [13],[16]–[19].
The study of large amounts of nuclear DNA sequences has the potential to resolve elephantid phylogeny, but due to technical challenges associated with obtaining homologous data sets from fossil DNA, no sufficiently large nuclear DNA data set has been published to date. Although a draft genome is available for woolly mammoth (Mammuthus primigenius) [5] and savanna elephant (loxAfr; http://www.broadinstitute.org/ftp/pub/assemblies/mammals/elephant/), comparative sequence data are lacking for Asian (Elephas maximus) and forest elephant, as well as for a suitable outgroup like the American mastodon (Mammut americanum). Using a combination of next generation sequencing and targeted multiplex PCR, we obtained the first substantial nuclear data set for comparing these species.
We carried out shotgun sequencing of DNA from an American mastodon with a Roche 454 Genome Sequencer (GS), using the same DNA extract from a 50,000–130,000-yr-old tooth that we previously used to generate a complete mtDNA genome sequence from the mastodon [8]. After comparing the 45 Mb of shotgun DNA data that we obtained to the Genbank database, and only retaining reads for which the best match was to sequences of the savanna elephant draft sequence (loxAfr1), we were left with 1.76 Mb of mastodon sequence (Figure 1 and Figure S1).
To amplify the same set of loci across all species, we designed PCR primers flanking the regions of mastodon-elephant alignment, using the loxAfr1 savanna elephant sequence as a template (Figure 1) (a full list of the primers is presented in Dataset S1). We used these primers in a multiplexed protocol [24] to amplify one or two Asian elephants, one African forest elephant, one woolly mammoth, and one African savanna elephant unrelated to the individual used for the reference sequence (Figure 1 and Table S1). We then sequenced the products on a Roche 454 GS to a median coverage of 41-fold and assembled a consensus sequence for each individual by restricting to nucleotides with at least 3-fold coverage. After four rounds of amplification and sequencing, we obtained 39,763 base pairs across 375 loci with data from all five taxa (Text S1; Figure S2; Table S2, Table S3). We identified 1,797 nucleotides in this data set in which two different alleles were observed and used these sites for the majority of our analyses (the genotypes are provided in Dataset S2). A total of 549 of these biallelic sites were polymorphic among the elephantids, while the remaining sites were fixed differences compared to the mastodon sequence.
To assess the utility of the data for molecular dating and inference about demographic history, we carried out a series of relative rate tests, searching for an excess of divergent sites in one taxon compared to another since their split, which could reflect sequencing errors or changes in the molecular clock [25]. None of the pairs of taxa showed a significant excess of divergent sites compared with any other (Table 1). When we compared the data within taxa, we found that the savanna reference genome loxAfr1 had a significantly higher number of lineage-specific substitutions than the savanna elephant we sequenced (nominal P = 0.03 from a two-sided test without correcting for multiple hypothesis testing). This is consistent with our data being of higher quality than the loxAfr1 reference sequence, presumably due to our high read coverage.
In contrast to our elephantid data, our mastodon data had a high error rate, as expected given that it was derived from shotgun sequencing data providing only 1-fold coverage at each position. To better understand the effect of errors in the mastodon sequence, we PCR-amplified a subset of loci in the mastodon, obtaining high-quality mastodon data at 1,726 bases (Text S2). Of the n = 23 sites overlapping these bases that we knew were polymorphic among the elephantids, the mastodon allele call always agreed between the PCR and shotgun data, indicating that our mastodon data are reliable for the purpose of determining an ancestral allele (the main purpose for which we use the mastodon data). However, only 38% of mastodon-elephantid divergent sites validated, which we ascribe to mastodon-specific errors, since almost all the discrepancies were consistent with C/G-to-T/A misincorporations (the most prominent error in ancient DNA) [26]–[28], or mismapping of some of the short mastodon reads (2). Thus, our raw estimate of mastodon-elephantid divergence is too high, making it inappropriate to use mastodon for calibrating genetic divergences among the elephantids, as we previously did for mtDNA where we had high-quality mastodon data [8].
We estimated the relative genetic diversity across elephantids by counting the total number of heterozygous genotypes in each taxon, and normalizing by the total number of sites differing between (S)avanna and (A)sian elephants (tSA). Within-species genetic diversity as a fraction of savanna-Asian divergence is estimated to be similar for savanna elephants (8±2%) and mammoths (9±2%), higher for Asian elephants (15±3%), and much higher for forest elephants (30±4%) (standard errors from a Weighted Jackknife; Methods). This supports previous findings of a higher average time to the most recent common genetic ancestor in forest compared to savanna elephants (Table 1) [13],[17]. We caution that these diversity estimates are based on analyzing only a single individual from each taxon, which could produce a too-low estimate of diversity in the context of recent inbreeding. Encouragingly, however, in Asian elephants where two individuals were sequenced for some loci, genetic diversity estimates are consistent whether measured across (18±5%) or within samples (15±3%). A further potential concern is “allele specific PCR”, whereby one allele is preferentially amplified causing truly heterozygous sites to go undetected [29]. However, we do not believe that this is a concern since we preformed an experiment in which we re-amplified about 5% of our loci using different primers and obtained identical genotypes at all sites where we had overlapping data (Text S2).
We next inferred a nuclear phylogeny for the elephantids using the Neighbor Joining method (Methods and Figure S3). This analysis suggests that mammoths and Asian elephants are sister taxa, consistent with the mtDNA phylogeny [8], and that forest and savanna elephants are also sister taxa. We estimate that forest-savanna genetic divergence normalized by savanna-Asian is tFS/tSA = 74±6%, while Asian-mammoth genetic divergence normalized by savanna-Asian tAM/tSA = 65±5% (Table 1). These numbers are all significantly lower than savanna-mammoth (tSM/tSA = 92±5%), forest-Asian (tFA/tSA = 103±5%), and forest-mammoth (tFM/tSA = 96±7%) normalized by savanna-Asian genetic divergence, which are all consistent with 100% as expected if they reflect the same comparison across sister groups (Table 1).
An intriguing observation is that the ratio of forest-savanna elephant genetic divergence to Asian-mammoth divergence tFS/tAM is consistent with unity (90% credible interval 90%–138%), which is interesting given that forest and savanna elephants are sometimes classified as the same species, whereas Asian elephants and mammoth are classified as different genera [20],[30]. To further explore this issue, we focused on regions of the genome where the genealogical tree is inconsistent with the species phylogeny, a phenomenon known as “incomplete lineage sorting” (ILS) [8],[11],[31]. Information about the rate of ILS can be gleaned from the rate at which alleles are observed that cluster taxa that are not most closely related according to the overall phylogeny. For example, in a four-taxon alignment of (S)avanna, (F)orest, (E)urasian, and mastodon, “SE” and “FE” alleles that cluster savanna-Eurasian or forest-Eurasian, to the exclusion of the other taxa, are likely to be at loci with ILS (in what follows, we use the term “Eurasian elephants” to refer to woolly mammoths and Asian elephants, while recognizing that the range of the lineage ancestral to each species included Africa as well). Similarly, in a four-taxon alignment of (A)sian, (M)ammoth, (L)oxodonta (forest plus savanna), and mastodon, “AL” or “ML” sites reveal probable ILS events. We find a higher rate of inferred ILS in forest and savanna elephants than in Asian elephants and mammoths: (FE+SE)/(AL+ML) = 3.1 (P = 4×10−8 for exceeding unity; Table 2), indicating that there are more lineages where savanna and forest elephants are unrelated back to the African-Eurasian speciation than is the case for Asian elephants and mammoths (Table 2). This could reflect a history in which the savanna-forest population divergence time TFS is older than the Asian-mammoth divergence time TAM, a larger population size ancestral to the African than to the Eurasian elephants, or a long period of gene flow between two incipient taxa. (We use upper case “T” to indicate population divergence time and lower case “t” to indicate average genetic divergence time (t≥T)).
To further understand the history of the elephantids, we fit a population genetic model to the data (input file—Dataset S3) using the MCMCcoal (Markov Chain Monte Carlo coalescent) method of Yang and Rannala [32]. We fit a model in which the populations split instantaneously at times ΤFS (forest-savanna), ΤAM (Asian-mammoth), ΤLox-Eur (African-Eurasian), and ΤElephantid-Mastodon, with constant population sizes ancestral to these speciation events of ΝFS, ΝAM, ΝLox-Eur, and ΝElephantid-Mastodon, and (after the final divergences) of ΝF, ΝS, ΝA, and ΝM (Figure 2). We recognize that elephantid population sizes likely varied within these time intervals, given recurrent glacial cycles [33], changes in geographic ranges documented in the fossil record [15],[30],[34],[35], and mtDNA patterns suggesting ancient population substructure [13],[15]. Nevertheless, the constant population size assumption is useful for inferring average diversity and obtaining an initial picture of elephantid history. MCMCcoal then makes the further simplifying assumptions that our short (average 106 bp) loci experienced no recombination and that they are unlinked (the latter assumption is justified by the fact that when we mapped the loci to scaffolds from the loxAfr3 genome sequence, all but one pair were at least 100 kilobases apart; Text S3). MCMCcoal then infers the joint distribution of the “T” and “N” parameters that is consistent with the data, as well as the associated credible intervals (Table 3; Text S4).
The MCMCcoal analysis infers that the initial divergence of forest and savanna elephant ancestors occurred at least a couple of Mya. The first line of evidence for this is that forest-savanna elephant population divergence time is estimated to be comparable to that of Asian elephants and mammoths: ΤAM/ΤFS = 0.96 (0.69−1.36) (Table 4). Secondly, MCMCcoal infers that the ratio of forest-savanna to African-Eurasian elephant population divergence is at least 45%: ΤFS/ΤLox-Eur = 0.62 (0.45−0.79) (Table 4). Given that African-Eurasian genetic divergence (TLox-Eur) can be inferred from the fossil record to have occurred 4.2–9.0 Mya (Text S5), this allows us to conclude that forest-savanna divergence occurred at least 1.9 Mya (4.2 Mya × 0.45). We caution that because MCMCcoal fits a model of instantaneous population divergence, our results do not rule out some forest-savanna gene flow having occurred more recently, as indeed must have occurred based on the mtDNA haplogroup that is shared among some forest and savanna elephants. However, such gene flow would mean that the initial population divergence must have been even older to explain the patterns we observe.
We also used the MCMCcoal results to learn more about the timing of the divergences among the elephantids (Figure 2). To be conservative, we quote intervals that take into account the full range of uncertainty from both the fossil calibration of African-Eurasian population divergence (TLox-Eur = 4.2–9.0 Mya; Text S5), and the 90% credible intervals from MCMCcoal (TFS/TLox-Eur = 45%–79% and TAM/TLox-Eur = 46%–74%; Table 4). Thus, we conservatively estimate TFS = 1.9–7.1 Mya and TAM = 1.9–6.7 Mya. Our inference of TAM is somewhat less than the mtDNA estimate of genetic divergence of 5.8–7.8 Mya [8]. However, this is expected, since genetic divergence time is guaranteed to be at least as old as population divergence but may be much older, especially as deep-rooting mtDNA lineages are empirically observed to occur in matrilocal elephantid species.
Our study of the extant elephantids provides support for the proposed classification of the Elephantidae by Shoshani and Tassy, which divides them into the tribe Elephantini (including Elephas—the Asian elephant and fossil relatives—and the extinct mammoths Mammuthus) and the tribe Loxodontini (consisting of Loxodonta: African forest and savanna elephants and extinct relatives) [36]. This classification is at odds with previous suggestions that the extinct mammoths may have been more closely related to African than to Asian elephants [37].
Our study also infers a strikingly deep population divergence time between forest and savanna elephant, supporting morphological and genetic studies that have classified forest and savanna elephants as distinct species [13],[16]–. The finding of deep nuclear divergence is important in light of findings from mtDNA, which indicate that the F-haplogroup is shared between some forest and savanna elephants, implying a common maternal ancestor within the last half million years [21]. The incongruent patterns between the nuclear genome and mtDNA (“cytonuclear dissociation”) have been hypothesized to be related to the matrilocal behavior of elephantids, whereby males disperse from core social groups (“herds”) but females do not [13],[38]. If forest elephant female herds experienced repeated waves of migration from dominant savanna bulls, displacing more and more of the nuclear gene pool in each wave, this could explain why today there are some savanna herds that have mtDNA that is characteristic of forest elephants but little or no trace of forest DNA in the nuclear genome [13],[14],[39],[40]. In the future, it may be possible to distinguish between models of a single ancient population split between forest and savanna elephants, or an even older split with longer drawn out gene flow, by applying methods like Isolation and Migration (IM) models to data sets including more individuals [41]. Our present data do not permit such analysis, however, as IM requires multiple samples from each taxon to have statistical power, and we only have 1–2 samples from each taxon.
Our study also documents the highly variable population sizes across recent elephantid taxa and in particular indicates that the recent effective population size of forest elephants in the nuclear genome (NF) has been significantly larger than those of the other elephantids (NS, NA, and NM) (Table 5) [13],[17],[19]. This is not likely due to the “out of Africa” migration of the ancestors of mammoths and Asian elephants as these events occurred several Mya [35], and any loss of diversity due to founder effects would have been expected to be offset by subsequent accumulation of new mutations in the populations. The high effective population size in forest elephants could reflect a history of separation of populations into distinct isolated tropical forest refugia during glacial cycles [33], which would have been a mechanism by which ancestral genetic diversity could have been preserved before the population subsequently remixed [1],[2],[23]. A Pleistocene isolation followed by remixing would also be consistent with the patterns observed in Asian elephants, which carry two deep mtDNA clades and where there is intermediate nuclear diversity. Intriguingly, our estimate of recent forest effective population size is on the same order as the ancestral population sizes (NFS, NAM, and NLox-Eur) (Table 5), providing some support for the hypothesis that forest elephant population parameters today may be typical of the ancestral populations (a caveat, however, is that MCMCcoal may overestimate ancestral population sizes since unmodeled sources of variation across loci may inflate estimates of ancestral population size). An alternative hypothesis that seems plausible is that the large differences in intra-species genetic diversity across taxa could reflect differences in the variance of male reproductive success [42] (more male competition in mammoth and savanna elephant than among forest elephants, with the Asian elephant being intermediate [43]).
The results of this study are finally intriguing in light of fossil evidence that forest and savanna lineages of Loxodonta may have been geographically isolated until recently. The predominant elephant species in the fossil record of the African savannas for most of the Pliocene and Pleistocene belonged to the genus Elephas [30],[34],[35]. Some authors have suggested that the geographic range of Loxodonta in the African savannas may have been circumscribed by Elephas, until the latter disappeared from Africa towards the Late Pleistocene [30],[34],[35]. We hypothesize that the widespread distribution of Elephas in Africa may have created an isolation barrier that separated savanna and forest elephants, so that gene flow became common only much later, contributing to the patterns observed in mtDNA. Further insight into the dynamics of forest-savanna elephant interaction will be possible once more samples are analyzed from all the taxa, and high-quality whole genome sequences of forest and savanna elephants are available and can be compared with sequences of Asian elephants, mammoths, and mastodons.
For our sequencing of mastodon, we used the same DNA extract that was previously used to generate the complete mitochondrial genome of a mastodon [8]. We sequenced the extract on a Roche 454 GS, resulting in 45 Mb of sequences that we deposited in the NCBI short read archive (accession: SRA010805). By comparing these reads to the African savanna elephant genome (loxAfr1) using MEGABLAST, we identified 1.76 Mb of mastodon sequences with a best hit to loxAfr1 that we then used in downstream analyses.
To re-sequence a subset of these loci in the living elephants and the woolly mammoth, we used Primer3 to design primers surrounding the longest mastodon-African elephant alignments. A two-step multiplex PCR approach [24] was used to attempt to sequence 746 loci in 1 mammoth, 1 African savanna elephant, 1 African forest elephant, and 1–2 Asian elephants. After the simplex reactions for each sample, the PCR products were pooled in equimolar amounts for each sample and then sequenced on a Roche 454 GS, resulting in an average read coverage of 41× per nucleotide (Text S1). We carried out four rounds of PCR in an attempt to obtain data from as many loci as possible and to fill in data from loci that failed or gave too few sequences in previous rounds (Text S1).
To analyze the data, we sorted the sequences from each sample according to the PCR primers (746 primer pairs in total) and then aligned the reads to the reference genome (loxAfr1), disregarding sequences below 80% identity. Consensus sequences for each locus and each individual were called with the settings described by Stiller and colleagues [44], with a minimum of three sequences required in order to call a nucleotide and a maximum of three polymorphic positions allowed per locus (to filter out false-positive divergent sites due to paralogous sequences that occur in multiple loci in the genome). We finally generated multiple sequence alignments for each locus and called divergent sites when at least one allele per species was available. In the first experimental round we were not able to call consensus sequences for more than half of the loci, a problem that we found was correlated with primer pairs that had multiple BLAST matches to loxAfr1, suggesting alignment to genomic repeats. Primer pairs for subsequent experimental rounds were excluded if in silico PCR (http://genome.ucsc.edu/cgi-bin/hgPcr) suggested that they could anneal at too many loci in the savanna elephant genome.
Of the 1,797 biallelic divergent sites that were identified, we removed 22 to produce Tables 1 and 2. The justification for removing these sites is that derived alleles were seen in both African and Eurasian elephants, which is unlikely to be observed in the absence of sequencing errors or recurrent mutation. For the MCMCcoal analysis we did not remove these divergent sites, since the method explicitly models recurrent mutation.
To obtain standard errors, we omitted each of the 375 loci in turn and recomputed the statistic of interest. To compute a normally distributed standard error, we measured the variability of each statistic of interest over all 375 dropped loci, weighted by the number of divergent sites at the locus that had been dropped in order to take account of the variable amount of data across loci. This can be converted into a standard error using the theory of the Weighted Jackknife as described in [45].
For our relative rate tests, we compute the difference in the number of divergent sites between two taxa since they split, normalized by the total number of divergent sites. The number of standard errors (computed from a Weighted Jackknife) by which this differs from zero represents a z score that should be normally distributed under the null hypothesis and thus can be converted into a p value for consistency of the data with equal substitution rates on either lineage.
To construct a Neighboring Joining tree relating the proboscideans in Figure S3, we used MEGA4 [46] with default settings (10,000 bootstrap replicates).
To prepare a data set for MCMCcoal, we used input files containing the alignments in PHYLIP format (Dataset S3) [47], restricting analysis to the loci for which we had diploid data from at least one individual from each of the elephantids we resequenced (we did not use data from the loxAfr1 draft savanna genome, or from the second Asian elephant we sequenced at only a small fraction of loci). The diploid data for each taxon were used to create two sequences from each of the elephantids, allowing us to make inferences about effective population size in each taxon since its divergence from the others.
We ran MCMCcoal with the phylogeny ((((Forest1,Forest2), (Savanna1,Savanna2)), ((Asian1,Asian2), (Mammoth1,Mammoth2))) Mastodon). Since MCMCcoal is a Bayesian method, it requires specifying a prior distribution for each parameter; that is, a hypothesis about the range of values that are consistent with previously reported information (such as the fossil record). For the effective population sizes in each taxa (NF, NS, NA, NM, NFS, NAM, NLox-Eur, and NElephantid-Mastodon) we used prior distributions that had their 5th percentile point corresponding to the lowest diversity seen in present-day elephants (savanna) and their 95th percentile point corresponding to the highest diversity seen in elephantids (forest). For the mastodon-elephantid population divergence time TElephantid-Mastodon we used 24–30 Mya [30],[35],[48]–[50]. For the African-Eurasian population divergence time ΤLox-Eur we used 4.2–9 Mya [30],[35],[51]. For the Asian-mammoth population divergence time ΤAM we used 3.0–8.5 Mya [30],[35],[52]. The taxonomic status of forest and savanna elephants is contentious. To allow us to test the hypotheses of both recent and ancient divergence while being minimally affected by the prior distribution, we use an uninformative prior distribution of TFS = 0.5–9 Mya. This prior distribution has substantial density at <1 million years, allowing us to test for recent divergence of forest and savanna elephants. A full justification for the prior distributions is given in Text S5.
MCMCcoal also requires an assumption about the mutation rate, which is poorly measured for the elephantids. We thus ran MCMCcoal under varying assumptions for the mutation rate, to ensure that our key results were stable in the face of uncertainty about this parameter. For each of the three mutation rates that we tested, MCMCcoal was run three times starting from different random number seeds with 4,000 burn-in and 100,000 follow-on iterations. Estimates of all parameters that were important to our inferences were consistent across runs suggesting stability of the inferences despite starting at different random number seeds (we did observe instability for the parameters corresponding to mastodon-elephantid divergence, but this was expected because of the high rate of mastodon errors and is not a problem for our analysis as this divergence is not the focus of this study). We computed the autocorrelation of each sampled parameter over MCMC iterations to assess the stickiness of the MCMC. Parameters appear to be effectively uncorrelated after a lag of 200 iterations. Given that we ran each chain over 100,000 iterations, we expect to have at least 500 independent points from which to sample, which is sufficient to compute 90% credible intervals. The detailed parameter settings and results are presented in Text S4.
|
10.1371/journal.pbio.1000057 | Cargo and Dynamin Regulate Clathrin-Coated Pit Maturation | Total internal reflection fluorescence microscopy (TIR-FM) has become a powerful tool for studying clathrin-mediated endocytosis. However, due to difficulties in tracking and quantifying their heterogeneous dynamic behavior, detailed analyses have been restricted to a limited number of selected clathrin-coated pits (CCPs). To identify intermediates in the formation of clathrin-coated vesicles and factors that regulate progression through these stages, we used particle-tracking software and statistical methods to establish an unbiased and complete inventory of all visible CCP trajectories. We identified three dynamically distinct CCP subpopulations: two short-lived subpopulations corresponding to aborted intermediates, and one longer-lived productive subpopulation. In a manner dependent on AP2 adaptor complexes, increasing cargo concentration significantly enhances the maturation efficiency of productive CCPs, but has only minor effects on their lifetimes. In contrast, small interfering RNA (siRNA) depletion of dynamin-2 GTPase and reintroduction of wild-type or mutant dynamin-1 revealed dynamin's role in controlling the turnover of abortive intermediates and the rate of CCP maturation. From these data, we infer the existence of an endocytic restriction or checkpoint, responsive to cargo and regulated by dynamin.
| Clathrin-mediated endocytosis is the major pathway for the uptake of molecules into eukaryotic cells and is regulated by the GTPase dynamin. Adaptor proteins recruit clathrin to the plasma membrane, where clathrin-coated pits capture transmembrane cargo molecules, again via adaptors. The pits invaginate and pinch off to form clathrin-coated vesicles that carry the cargo into the cell. Live cell imaging has revealed striking heterogeneity in the dynamic behavior of clathrin-coated pits associated with the plasma membrane, yet the nature of this heterogeneity and its functional implications are unknown. We used particle-tracking software to establish an unbiased and complete inventory of the trajectories of clathrin-coated pits visible by total internal reflection fluorescence microscopy. Through statistical analyses, we identified three dynamically distinct subpopulations of coated pits: two short-lived subpopulations corresponding to aborted intermediates, and one longer-lived productive subpopulation. The proportion of each subpopulation and their lifetimes respond independently to molecular perturbations. As a result of systematic modulation of cargo concentration, adaptor levels, and analysis of dynamin mutants, we postulate the existence of an endocytic restriction or checkpoint that governs the rate of clathrin-mediated endocytosis by gating the maturation of clathrin-coated pits.
| Clathrin-mediated endocytosis (CME) is the major endocytic pathway in eukaryotic cells. It occurs via clathrin-coated pits (CCPs) that are assembled from cytosolic coat proteins. CCPs capture transmembrane cargo molecules, invaginate, and then pinch off to form clathrin-coated vesicles (CCVs). CME is a constitutive, yet highly regulated process. Biochemical assays of endocytosis score ligand uptake and measure only the ensemble average of successful internalization events, thereby obscuring critical, rate-limiting early stages of maturation and alternative outcomes that might cause variability in individual CCP dynamics. Indeed, live cell imaging has revealed striking heterogeneity in the dynamic behavior of plasma membrane–associated CCPs [1–5].
An important parameter for analyzing CCP heterogeneity is their lifetimes. The lifetime of an individual CCP at the plasma membrane, i.e., the time required for (1) coat initiation, (2) coat propagation, (3) neck constriction, and (4) vesicle budding, is critical for understanding CME. Changes in lifetimes caused by specific molecular perturbations can reveal mechanisms that regulate each of these steps. However, selective probing of all stages of CCP maturation is only possible by mild perturbation of the underlying molecular processes. Detection and interpretation of these necessarily milder phenotypes requires sensitive and comprehensive analysis of individual CCP lifetimes and behavior. To this end, we have employed total internal reflection fluorescence microscopy (TIR-FM), the premier assay to detect early intermediates in CCV formation and visualize the dynamics of CCPs in living cells [1,3–9]. By selectively exciting fluorophores associated with molecular components of CCPs at the ventral plasma membrane, TIR-FM provides excellent signal-to-background ratio and high time resolution. In spite of these strengths, it has remained a challenge to extract reliable measurements of CCP lifetimes from TIR-FM videos. Lifetime measurements are notoriously susceptible to tracking errors, which typically break CCP trajectories into two or more subtrajectories, leading to systematic bias of lifetimes towards shorter values. As a result, tracking has previously been accomplished either manually for a low number of well-discernable, high-intensity CCPs [1,6], or using semiautomated tracking restricted to isolated CCPs, for which no close neighbors are likely to confuse the tracking algorithm [2,4]. Both approaches sample the behavior of arbitrary and typically small subpopulations with relatively uniform properties. To solve these problems and to better exploit the heterogeneity of CCP dynamics as a source of mechanistic information, we have employed particle-tracking software [10] capable of detecting and tracking all CCPs visualized by TIR-FM in an unbiased fashion. Automated detection and tracking enabled analysis of several tens of thousands of trajectories per condition, 100 times more than previous studies, thus providing a comprehensive and accurate measurement of CCP lifetime distributions.
We used TIR-FM and our automated tracking assay [10] (see Materials and Methods, Figure S1, and Videos S1, S2, and S4) to obtain large and unbiased datasets of CCP dynamics in well-characterized BSC1 cells expressing a fully functional enhanced green fluorescent protein (EGFP)-tagged clathrin light chain a (LCa-EGFP; Figure 1A and 1B) [2]. To capture both fast events at the timescale of seconds and slower events at the timescale of minutes, we combined data from time-lapse sequences taken at a frame rate of 0.4 s for ≥3 min with data from sequences taken at a frame rate of 2 s for 10–15 min (see Materials and Methods). Objects not detected for at least five consecutive frames were not counted, to exclude transient, highly motile structures. Nonetheless, CCPs displayed a nearly exponential decay of lifetimes (Figure 1C), indicating that a large number appear and disappear on the timescale of a few seconds. To further analyze CCP lifetimes, the raw data were fitted to a series of models that differed in the number and types of subdistributions. Our goal was to identify the minimal number of kinetically distinct subpopulations that could account for the overall lifetime distribution observed (see Figure S2A). Model selection was achieved by three strategies. The first two involved minimization of the Bayesian Information Criterion (BIC) [11,12], which defines the optimal tradeoff between the goodness of fit of the model and the number of free model parameters. The application of the BIC requires a priori knowledge of the distribution of experimental errors, which is unknown for lifetime data. Thus, we performed BIC minimization, first, in a fit of the cumulative lifetime histogram, i.e., with the lifetime and the cumulative frequency as the independent and dependent, error-perturbed variables, respectively; and second, in a fit of the inverse of the cumulative histogram, i.e., with the percentage rank as the independent and the lifetime as the dependent error-perturbed variables. In both cases, we approximated the distribution of the fitting errors as normal. The third strategy for model selection involved a nonparametric test of the distribution of the fitting residuals, which did not require a priori assumptions (see Figure S2).
All three strategies identified three statistically significant subpopulations (Figure 1D) with distinct time constants, but broad and overlapping lifetime distributions (Figure 1F). Importantly, at the noise level of our TIR-FM images, accurate assignment of these subpopulations required analysis of >5,000 trajectories (see Figure S2C), and hence, our results strongly relied on the accurate and automatic tracking of all CCPs in multiple videos per experimental condition (see Materials and Methods). Our data contained two short-lived subpopulations with time constants of 5.2 ± 0.1 s and 15.9 ± 1 s, respectively (±jackknifed cell-to-cell error, see Table S1) that were best fit with Rayleigh distributions, i.e., the shortest and longest lifetimes within the population occur less frequently than the intermediate ones. This suggests that our time sampling of 0.4 s per frame was sufficient to capture all events of significant clathrin coat accumulation. A single longer-lived subpopulation (time constant 86.9 ± 5.8 s) was best approximated by an exponentially decaying distribution. Given its long time constant, accurate measurements of the mean lifetime of this population required imaging for ≥10 min. The longer-lived subpopulation was designated the “productive population,” because (1) its kinetics match those of surface-bound transferrin internalization measured biochemically (t1/2 ≈ 104 s; Figure S3A), and (2) manual tracking of 450 CCPs (for which internalization was confirmed by sequential disappearance from the TIR-FM and the epifluorescence microscopy (EPI-FM) field [5,6]) also yielded a mean lifetime of approximately 100 s (Figure S3B). Accordingly, we hypothesized that the two shorter-lived species correspond to transient, nonproductive events and therefore termed them “early abortive” and “late abortive” CCPs, respectively.
In BSC1 cells, the productive population constituted only 38.6 ± 3.4% of total CCPs at the plasma membrane, with early and late abortive CCPs representing 38.1 ± 3.1% and 21.9 ± 1.4%, respectively (±jackknifed cell-to-cell error; Figure 1E, Table S2). Taking into account the different mean lifetimes and relative contributions of the three individual subpopulations, the mean lifetime of all CCPs is 39 s, much shorter than the half-time of transferrin (Tfn) uptake. This further supports the hypothesis that the two short-lived subpopulations are abortive and do not contribute to Tfn uptake.
To determine whether the existence of these three subpopulations was affected by the nature of the fluorescent tag or the cell type used, we performed lifetime analyses on BSC1 and HeLa cells transiently expressing LCa-tomato and NIH3T3 cells stably expressing LCa-DsRed. We consistently observed one long-lived and two short-lived populations (Table S1B), suggesting that this categorization into kinetically distinct CCP populations is a universal phenomenon of CME. As described by others [4,13], however, we also observed in both HeLa cells and NIH3T3 cells, a higher proportion of larger clathrin-coated structures (CCSs) from which multiple CCSs emerge and disappear. These so-called “nonterminal” events are rarely detected in BSC1 cells, consistent with previous findings [2].
We also analyzed CCP lifetimes by tracking the adaptor protein, AP2, in BSC1 cells expressing the EGFP-tagged σ2 subunit (σ2-EGFP), shown not to interfere with AP2 function [2] (Videos S3 and S5). Our model selection again identified three kinetically distinct populations of σ2-containing CCPs (Figure 2C and 2D, Tables S1 and S2); the preference for three versus two subpopulations was weaker but still highly significant (p < 10−4 as compared to p < 10−10 in LCa-EGFP–expressing cells). This further supports our conclusion that each of these subpopulations represents bona fide plasma membrane–associated CCPs rather than clathrin-bearing endosomal structures transiently approaching the cell surface. The percentage of productive CCPs with σ2-EGFP labeling was higher than those labeled with LCa-EGFP (56.3 ± 10.1% as compared to 38.6 ± 3.4%; compare Figure 2A and 2C, Table S2), confirming previous suggestions [2] that adaptors enhance the maturation efficiency of CCPs. The characteristic lifetimes of early abortive CCPs labeled with σ2-EGFP (4.8 ± 0.4 s) were similar to those observed for LCa-EGFP (5.2 ± 0.1 s; compare Figure 2B and 2D, Table S2), suggesting that this very short-lived subpopulation represents stochastic coated pit nucleation events, perhaps triggered by low-affinity interactions between AP2 and phosphatidylinositol-4,5-bisphosphate at the plasma membrane. In contrast, the characteristic lifetimes of both late abortive and productive subpopulations of σ2-EGFP–labeled CCPs (8.4 ± 1.8 s and 71.5 ± 6 s, respectively) were significantly shorter than their LCa-EGFP–labeled counterparts (15.9 ± 1 s and 86.9 ± 5.8 s, respectively). This observation is consistent with findings of others, reporting a general shift of AP2-containing structures toward shorter lifetimes [2,8], although interpretations have varied. It may reflect dissociation of AP2 complexes from CCPs prior to clathrin [8] and/or a nonuniform distribution of AP2 complexes in the clathrin lattice resulting in their differential illumination by the TIRF field [9]. The former interpretation would be consistent with recent findings that Sar1p and the Sec23/24 adaptors can dissociate from budding vesicles prior to the Sec13/31p-containing outer shell of the COPII coat [14,15] and the notion that multivalent clathrin interactions dominate over AP2 interactions at later stages of CCV formation [16].
Using semiautomated analysis, Ehrlich et al. [2] previously detected a single, short-lived subpopulation of CCPs in BSC1 cells, which displayed a mean lifetime and relative contribution similar to our late abortive population. In addition, they reported that cargo-associated CCPs rarely failed to proceed to completion, leading to the suggestion that cargo might stabilize abortive CCPs. However, a direct link between CCP maturation and cargo load has not been established. To test the hypothesis that CCP maturation is responsive to the concentration of cargo, we infected BSC1 cells with an adenovirus coding for the human transferrin receptor (TfnR) in a tetracycline (tet)-repressible system. The TfnR is constitutively internalized even in the absence of ligand [17] and thus serves as a model transmembrane cargo molecule. Removal of tet induced TfnR overexpression by >30-fold in nearly all cells (Figure 3A–3C). There have been conflicting observations regarding the effect of TfnR overexpression on surface density of CCPs [18–20]. We confirmed previous biochemical measurements [20] showing that at high levels of TfnR overexpression, the endocytic machinery becomes saturated, i.e., the bulk endocytic efficiency of TfnR declines (Figure S3A), thus obscuring any effect that cargo might have on CCP dynamics. In addition, we did not observe an increase in membrane recruitment of either clathrin or the TfnR adaptor protein AP2, as measured by subcellular fractionation and western blot analysis (Figure S3C and S3D) or by TIR-FM (Figure 3D and 3E).
In cells overexpressing TfnRs, we could no longer detect a significant (p > 0.05) late abortive subpopulation labeled with LCa-EGFP, and this population was completely undetectable in σ2-EGFP labeled cells. The contribution of the productive population increased to 67.9 ± 7.9% of LCa-EGFP–labeled CCPs (Figure 2A) and 76.8 ± 5.6% of σ2-EGFP–containing CCPs (Figure 2C). From this, we conclude that CCPs mature with their highest efficiency when a threshold amount of AP2 and cargo are incorporated into the nascent clathrin lattice. In contrast, the relative contribution of early abortive CCPs is unchanged by cargo overexpression further supporting the notion that these are transient structures assembled in a cargo-independent manner.
The mean lifetime of the productive population was only slightly decreased upon cargo overexpression when compared to control conditions (Kolmogorov-Smirnov test [KS-test] p = 0.03, Table S1) in LCa-EGFP–expressing cells (Figure 2B) and showed no decrease in σ2-EGFP–expressing cells (Figure 2D). Thus, we conclude that elevated TfnR concentrations result in more efficient CCP maturation, but that TfnR concentration is not rate-limiting for CCV formation. The approximately 2-fold increase in efficiency of CCP maturation in the presence of an approximately 40-fold increase in cargo concentration points to the existence of other limiting factors and is consistent with the saturation of endocytic efficiency observed biochemically (Figure S3A).
To further probe the role of AP2 adaptors in pit maturation, we decreased cellular AP2 levels by approximately 50% through short-term small interfering RNA (siRNA)-mediated knockdown of the μ2-subunit (unpublished data) [21]. AP2 depletion reduced the densities of all classes of pits (from 0.477 ± 0.012 μm2 in control to 0.276 ± 0.069 μm2 in the knockdown), although the relative contributions of the three populations and their lifetimes remained unchanged (Figure 2A and 2B). This suggests that AP2-dependent nucleation events lead proportionally to both short-lived abortive and productive pits, implying a precursor/product relationship. In contrast to control cells, TfnR overexpression in AP2-depleted cells did not lead to an increase in the relative contribution of the productive population (Figure 2A). We conclude that the increase in CCP maturation efficiency with increased cargo concentration requires AP2 and/or is limited by AP2 concentration.
Given that TfnR/cargo concentration only marginally affects lifetimes of CCPs, we next investigated which factor(s) might regulate this aspect of CCP dynamics. The self-assembling GTPase dynamin has been suggested to play a dual role in CME, both as a regulator and/or fidelity monitor during early, rate-limiting steps in endocytosis, and as a well-documented component of the fission apparatus late in CCV formation [22–24]. Dynamin is recruited along with clathrin and AP2 [2], and the early activities of dynamin presumably occur while it is associated with coated pits in its unassembled state, utilizing its basal GTP binding and hydrolysis activities [22,23]. In contrast, dynamin function in membrane fission requires its self-assembly and assembly-stimulated GTPase activities [22,23,25] and occurs subsequent to a burst of recruitment at late stages of CCP formation [5–7]. We sought direct evidence for dynamin's dual role in CME by examining the effects of well-characterized dynamin mutants on CCP dynamics by siRNA-mediated knockdown of dynamin-2 and reintroduction of siRNA-resistant wild-type (WT) or mutant dynamin-1. Because dominant-negative dynamin mutants block endocytosis and lead to clustering of nonproductive CCPs (unpublished data), we focused our analysis on three well-characterized hypoactive dynamin mutants: (1) Dyn1K694A is impaired in self-assembly and hence specifically in assembly-stimulated GTPase activity [22]. Overexpression of dyn1K694A was shown to increase rates of CME [22]. (2) Dyn1S61D exhibits WT GTP binding, but reduced basal and assembly-stimulated GTP hydrolysis rates [26]. Overexpression of dyn1S61D was shown to reduce rates of CME [26]. (3) Dyn1T141A exhibits reduced GTP binding in the unassembled, basal state, but increased basal GTP hydrolysis rates. It also exhibits a slight increase in assembly-stimulated GTPase activity and overexpression of dyn1T141A slightly increases the rate of CME [26].
The effects of dynamin-2 knockdown and expression of these mutants on the relative contributions of the different subpopulations were minor compared to the effect of cargo overexpression (Figure 4A); however, after dynamin-2 knockdown we now detect a substantial increase in the fraction of long-lived (>10 min), “persistent” CCPs that were negligible in control BSC1 cells (Figures 4B, S1E, and S1F). Reintroduction of WT dynamin-1 (dyn1WT) or dyn1K694A reduced the number of persistent CCPs, whereas reintroduction of dyn1S61D or dyn1T141A mutants increased their numbers.
In contrast to cargo overexpression, perturbations of dynamin function had dramatic effects on the lifetimes of CCP subpopulations. Knockdown of dynamin-2 to approximately 17% of endogenous levels (see Figure S3E and S3F) significantly (KS-test, p < 10−10) increased the characteristic lifetime of productive CCPs (Figure 4C and Table S1), confirming a role for dynamin as a rate-limiting factor in CCV formation. Dynamin-2 knockdown also increased the characteristic lifetime of late abortive CCPs (Figure 4C). This observation is consistent with its proposed role during early stages in CCV formation.
After knockdown of dynamin-2, overexpression of dyn1WT significantly decreased the characteristic lifetime of the productive population (KS-test, p < 10−7, Figure 4C, Table S1), again consistent with dynamin controlling rate-limiting steps in CME. In addition, overexpression of dynWT decreased the lifetime of late abortive coated pits, suggesting a role for dynamin as a fidelity monitor that initiates rejection and disassembly of nonviable CCPs. Overexpression of the self-assembly–impaired dyn1K694A mutant also significantly decreased the characteristic lifetime of both the productive and late abortive subpopulations (KS-test, p < 10−16, p < 10−2, respectively; Figure 4C). These data suggest that unassembled dynamin functions early, that this assembly-impaired mutant stimulates CME by enhancing the rate of CCP maturation, and that dynamin self-assembly and subsequent assembly-stimulated GTPase activities per se are not rate-limiting for CCV formation [22]. Overexpression of the GTPase-defective dyn1S61D was unable to fully restore the rate of productive CCV formation in dynamin-2 siRNA-treated cells (Figure 4C), and indeed increased the lifetimes of late abortive and productive CCPs even when overexpressed in the presence of endogenous dynamin-2 (unpublished data). Lastly, overexpression of dyn1T141A was unique in that it differentially affected the abortive and productive lifetimes (Figure 4C): the lifetimes of abortive CCPs remain slightly increased relative to control cells (KS-test, p = 0.08), whereas the lifetime of productive CCPs became shorter (KS-test, p = 0.008). Together with the dyn1S61D findings, these results demonstrate that GTP binding and hydrolysis in the basal state are required for an early surveillance function of dynamin and that the basal rate of GTP hydrolysis might be rate-limiting for maturation towards the productive CCP subpopulation. These data provide direct evidence that dynamin plays a dual role in CCP maturation and vesicle budding.
To provide further evidence for dynamin's role in CCP maturation, we extracted the intensity time courses of the LCa-EGFP signal during CCP maturation. For this purpose, we focused our analysis on a subset of long-lived, isolated CCPs, which are highly likely (>99%) to represent productive events. The typical intensity time course is skewed (see example in Figure 5A) and can be divided into three distinct phases: (1) an “assembly phase” corresponding to the initial fast increase of signal intensity, which occurred during the first approximately 20 s of detectable clathrin lattice assembly; followed by (2) a “maturation phase,” during which LCa-EGFP signal intensity plateaus or increases only moderately; followed by (3) a “departure phase,” characterized by a sudden final drop of signal intensity. To measure the duration of these phases, the intensity time courses of individual CCP trajectories were fitted with a smoothing spline (Figure 5A) to identify the points where the approximated slope drops below a set threshold. siRNA-mediated knockdown of dynamin-2 markedly increased the duration of the maturation phase (t-test, p < 10−10), without significantly altering assembly or departure phases (see Table S3). These data are consistent with a role for dynamin in regulating rate-limiting steps in CCP maturation and with the fact that dynamin-mediated membrane fission is not rate-limiting for CME [22]. The average length of the assembly and departure phases (Figure 5B) were largely unaffected by reintroduction of the various dynamin mutants. In contrast, the length of the maturation phase increased and decreased depending on the GTP binding and hydrolysis activities of the reintroduced dynamin. Specifically, re-expression of dyn1WT or dyn1K694A mutants decreased, whereas re-expression of GTPase-defective dyn1S61D or dyn1T141A increased the duration of the maturation phase. These findings support a role for the basal GTP binding and hydrolysis activities of dynamin in CCP maturation.
We report a comprehensive and unbiased analysis of CCP dynamics in living cells. This was accomplished using a new particle tracking algorithm that defines the correspondence between CCP images in consecutive frames based on spatial and temporal global optimization [10], which allowed us to reliably follow the fate of CCPs in areas of both low and high pit density. The algorithm incorporates a gap closing scheme that permitted tracking of faint and temporarily unstable CCPs. The performance of this algorithm was extensively validated for its application to CME analysis [10]. To capture the sub-second scale events of CCP formation and the much slower events of CCP maturation, we merged lifetime data from high frequency, shorter time-lapse videos with lower frequency, longer time-lapse videos. Thus, we tracked tens of thousands of both short-lived and long-lived species for each experimental condition without biasing the selection of CCPs to a subpopulation with a specific characteristic, e.g., only isolated or bright pits. The large sample number enables application of statistical model-selection methods to determine the minimum number of subpopulations necessary to accurately fit the measured lifetime distribution. Indeed, application of these statistical methods requires n > 5,000, a criterion met in each of our analyses, but which greatly exceeds the 100–600 selected events previously assessed in other studies [1–5]. Importantly, subsequent molecular perturbations identified certain conditions in which the contribution of subpopulations to the overall lifetime distribution changed while their time constants were unaffected, whereas other conditions left contributions unchanged while significant shifts occurred in the time constants. This indicates the orthogonality of the two parameters we extracted and establishes that they can be independently determined to probe distinct mechanistic aspects of CCP maturation.
In control BSC1 cells expressing LCa-EGFP, we detected three CCP subpopulations (early abortive, late abortive, and productive), with distinct time constants (∼5 s, ∼15 s, and ∼90 s, respectively). A previous analysis of CCP dynamics in the same cells suggested an average lifetime of approximately 46 s, assuming a single population of CCPs [2]. Taking into consideration the different contributions of these subpopulations to the entire ensemble of CCPs, we obtain a value of 39 s, consistent with this previous data. We suspect that the slightly lower ensemble lifetime in our measurements may be associated with a more systematic exclusion of the very faint, short-lived early abortive CCPs in the previous study [2]. The longer CCP lifetimes (60–90 s) consistently reported by others [1,6] reflect selection and analysis of a single subset of typically productive CCPs.
The functional assignment of productive CCPs rested on the agreement of the lifetimes of the long-lived subpopulation with biochemical rates of TfnR uptake, as well as with the lifetimes of LCa-EGFP structures tracked manually in quasi-simultaneous TIRF- and epifluorescent images showing CCV internalization. Therefore, we interpret the short-lived CCPs to represent abortive events, but it is also conceivable that they could represent clathrin-coated intracellular structures, e.g., endosomes, that transiently move as visitors through the evanescent field of the TIRF microscope. However, for the following reasons, we think this possibility is unlikely: (1) Although we occasionally observe fast-moving visitors in LCa-EGFP–labeled cells, their displacement between consecutive frames is so much above average that particle correspondences are generally outside the tracking algorithm's self-adaptive search range [10]. Thus, the trajectories of visitors are typically fragmented into short sub-trajectories (less than three frames), similar to trajectories associated with false-positive detections. To exclude both types of false structures from the lifetime analysis, trajectories shorter than five frames were removed from the dataset (see Materials and Methods). (2) Early and late abortive events detected with LCa-EGFP are also detected by our statistical model selection following σ2-EGFP, a selective marker of plasma-membrane–associated CCPs. In addition, the early abortive σ2-labeled CCPs have virtually the same lifetime as early abortive LCa-labeled CCPs, giving us confidence that these are bona fide plasma membrane–associated structures. (3) The relative contributions of both abortive and productive CCPs are affected by transferrin receptor overexpression, and their lifetimes are affected by dynamin. There is no reason why these parameters should be affected for visitors unrelated to CCPs. (4) The strongest indication that the vast majority of the structures we have studied are plasma membrane–associated CCPs comes from the AP2 depletion experiments in which the numbers of all three subpopulations are proportionally decreased. This would not be expected if the shorter-lived species were derived from internal membranes. Furthermore, AP2 depletion prevents the shift to productive CCPs induced by TfnR overexpression. Thus, although it is possible that there is a minor contamination of CCPs by clathrin-coated internal membrane vesicles or clathrin-coated nonendocytic structures, their contribution appears not to be significant enough to affect our findings.
Having identified three kinetically distinct subpopulations of CCPs, we next showed that their relative distributions and lifetimes could be affected by systematically manipulating cargo concentration, adaptor protein levels and the level and activity of the GTPase dynamin. Based on the shifts in contribution and lifetime of the three subpopulations we propose that CME might be governed by an endocytosis checkpoint or restriction point, which is regulated, in part, by dynamin. The following observations support the existence of this checkpoint: (1) the identification of abortive and productive CCPs (see also [2]), (2) the finding that AP2-containing (and presumably cargo-enriched) CCPs are more likely to be productive, (3) the finding that cargo load enhances the efficiency of CCV formation leading to more productive CCPs at the expense of abortive ones, and (4) the finding that this effect of cargo concentration is dependent on or limited by AP2 adaptor concentrations.
Progression through a restriction or checkpoint requires the tight interaction of a monitor and an activator system. As the major GTPase involved in CME, dynamin was a prime candidate to regulate the endocytosis checkpoint. Two models have been proposed for dynamin function in endocytosis: as a regulatory molecule [27] and as a component of the fission machinery [28–30]. However, these are not mutually exclusive, and recent data support aspects of both [23–26]. Our data on the effects of dynamin depletion and dynamin mutants on CCP dynamics provide several lines of evidence that unassembled Dyn•GTP acts early and controls progression through the endocytosis checkpoint: (1) siRNA reduction of dynamin decreases the rate of both productive CCV formation and turnover of abortive CCPs, whereas overexpression of WT dynamin accelerates each of these rates; (2) a mutant defective in self-assembly (K694A) that is predicted to increase cellular levels of unassembled Dyn•GTP further increases the rates of abortive CCP turnover; and (3) dynamin GTPase domain mutants predicted to be defective in basal GTP binding (T141A) or hydrolysis (S61D) selectively reduce the rates of abortive CCP turnover. Importantly, these conclusions rely on the use of well-characterized, hypomorphic dynamin mutants that mildly alter the kinetics of CME, yet have robust and readily detectable effects when assessed by large-scale image analysis. Strong dominant-negative dynamin mutants that stop CME lead to the accumulation of aggregated CCPs, thus limiting their usefulness for mechanistic interpretation.
Dynamin is also positioned to act as a monitor of factors that satisfy restriction point requirements through its many SH3 domain–containing binding partners. These have additional domains that recognize coat proteins (e.g., amphiphysin, SNX9,) membrane curvature (e.g., amphiphysin, endophilin, SNX9), and/or cargo molecules (e.g., SNX9, grb2, TTP). It is known that these proteins can differentially affect dynamin's basal GTPase activity and assembly properties [31–33]. Hence, they provide a potential mechanism for regulating dynamin function in response to these parameters of CCP maturation. Dynamin has also been shown to interact with auxilin and hsc70 [34], thus providing a potential mechanism for the dynamin-dependent turnover of abortive CCPs that we have observed.
A model describing these results is illustrated in Figure 6. In this model, productive CCP formation is a stochastic event initiated by the cargo-independent association of AP2 at the plasma membrane, which nucleates clathrin assembly. If a critical mass of the additional factors required for CCP stabilization is not reached during this assembly phase, these structures, which correspond to early and late abortive CCPs, fail to pass through the restriction point and disassemble. We propose that dynamin regulates the checkpoint and controls entry into and progression through the CCP maturation phase. The basal GTP binding and hydrolysis activities may enable unassembled dynamin to function either as a sensor or timer of CCP assembly, through its SH3 domain–containing partners, and thus directly or indirectly control the termination or progression of early endocytic intermediates. Further work will be needed to test this hypothesis and to determine the mechanistic details of how dynamin may monitor CCP assembly and maturation.
In sum, we propose that the presence of sufficient cargo, a threshold concentration of AP2 adaptors and perhaps other parameters such as the recruitment of endocytic accessory factors, acquisition of membrane curvature, etc., are detected by dynamin to signal progression beyond the endocytosis checkpoint. Whereas three kinetically distinct subpopulations are detected with statistical significance in our analyses, the lifetime distributions—particularly of the productive population—remain very broad. Thus, we expect that there are other aspects of functional heterogeneity and other factors regulating the endocytosis checkpoint masked within this subpopulation. Future studies involving mild perturbation of other endocytic accessory factors together with comprehensive quantitative analysis of CCP dynamics should provide further mechanistic insight into this functional heterogeneity.
TIR-FM was performed on BSC1 monkey kidney epithelial cells stably expressing rat brain clathrin LCa-EGFP or the AP2 rat brain σ2-adaptin fused to EGFP (provided by Dr. T. Kirchhausen, Harvard Medical School, and described in [2]) using a 100 × 1.45 NA objective (Nikon) mounted on a Nikon TE2000U inverted microscope (Nikon). CCP lifetimes range from a few seconds to several minutes. To fully capture this range of dynamics would require image sampling over minutes at a high frame rate (<1 s). Such exposure leads to significant photobleaching and also substantially impairs cell viability, both affecting the accuracy of lifetime measurements. To avoid these problems, we applied a multi-timescale imaging approach. For each experimental condition, three to nine videos with a frame rate of 0.4 s/frame were acquired over >200 s, and five to 21 videos with a frame rate 2 s/frame, each from different cells, were acquired over approximately 10 min, using a 14-bit mode operated Hamamatsu Orca II-ERG camera.
CCPs were detected using à-trous wavelet transform decomposition of the image [35]. Tracking of CCP was accomplished using spatially and temporally global particle assignment described in detail elsewhere [10]. The histograms of CCP lifetimes extracted from the two TIR-FM video categories were merged for the final lifetime analysis by normalizing the relative contribution of the CCP population with a lifetime in the range 30 to 50 s. Thus, the integrals of the measured lifetime probability density functions gi,[0.4] from all N[0.4] videos sampled at 0.4 s/frame and the integrals of the measured lifetime functions fj,[2] from all M[2] videos sampled at 2 s/frame were set to equal values:
From the merged lifetime histograms, the underlying distributions of multiple CCP populations with different lifetime dynamics were extracted using both parametric and nonparametric model selection as described in detail in Text S1.
For the intensity analysis, we extracted the trajectories of CCPs that were long-lived (>60 s) and isolated (no nearest neighbors within six pixels); this criterion ensured that the chosen CCPs were in fact “productive,” i.e., that they underwent full maturation and internalization, and that there was no crossover to neighbors, both in terms of the physical material of the CCP and in terms of the tracked CCP trajectories, both of which can lead to artifacts in the intensity time course.
Text S1 contains a more detailed description of methods including particle tracking, lifetime analysis, cell culture, adenoviral infection, cell fractionation, western blotting, immunofluorescence, siRNA transfection, and biochemical measurement of endocytic uptake. Tables S1 and S2 summarize the mean lifetimes and relative contributions of CCP subpopulations in each experimental condition. Table S3 summarizes the intensity phase data.
Figure S1 shows data relating to the validation of tracking and lifetime analysis. Figure S2 shows three statistical methods used to identify CCP subpopulations. Figure S3 shows the effect of TfnR overexpression on (1) endocytosis efficiency and (2) subcellular distribution of AP2 and clathrin, and a western blot of dynamin knock-down and corresponding quantification. Videos S1–S6 show examples of particle detection and tracking in simulated videos (Video S6) and those obtained imaging live cells (Videos S1–S5).
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.